Generalizations in Practice: Investigating Generality
and Specificity in Developmental Biology
A Dissertation
SUBMITTED TO THE FACULTY OF
THE UNIVERSITY OF MINNESOTA BY
Yoshinari Yoshida
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS
FOR THE DEGREE OF DOCTOR OF PHILOSOPHY
Alan C. Love
2023
© Yoshinari Yoshida 2023
Acknowledgments
First and foremost, I would like to express my deepest gratitude to my advisor, Alan Love. I first
encountered Alan’s work as an undergraduate student at Kyoto University. As a young student
interested in philosophy and history of biology, I was impressed by his nuanced discussions based
on a close examination of scientific practice. At a conference of the International Society for the
History, Philosophy, and Social Studies of Biology (ISHPSSB) in Montpellier in 2013, I had the
opportunity to meet Alan. He read and commented on a manuscript I was drafting with Hisashi
Nakao, which was later published in Biological Theory. When I decided to pursue a Ph.D. abroad, I
immediately knew that I wanted to work with Alan. Although I was fortunate enough to be admitted
to a few graduate schools, choosing the University of Minnesota was an easy decision. Working
with Alan has been the best decision I’ve made in the last ten years. He has been a constant source
of support and encouragement, and I have learned so much from him.
I am also grateful to other members of my committee. Samuel Fletcher has provided me with valu-
able feedback that allowed me to reinterpret my work from different perspectives. I have greatly
benefited from interacting with Bennett McNulty, who is not only a thoughtful scholar, but also an
excellent teacher. Michael Travisano’s critical approach has helped me to deepen my thoughts and
refine my work. Jamie Davies is an excellent developmental biologist with an interest in conceptual
issues in science. I thank him for joining the committee and bringing a unique perspective as a
developmental biologist.
Philosophy is a social enterprise. This dissertation was made possible because I have had many
people around me willing to discuss my works. The “Love Lab” has been one of the most crucial
communities. I am so fortunate to have had Max Dresow as both a friend and peer during my
Ph.D. program. His encyclopedic knowledge and exceptional productivity have consistently pro-
vided me with valuable assistance and inspiration. Nathan Lackey has not only been a valuable
member of the lab, but also an amazing roommate over the past few years. Nathan and I have had
countless discussions about research, teaching, career development, and life in general. Lauren
Wilson possesses a remarkable skill for asking insightful questions, which has assisted me in re-
vising my works and has set a fine example of a supportive conversationalist. Many thanks to the
past postdocs and visiting graduate students—Janella Baxter, Joseph Madison, Kelle Dhein, and
Amanda Corris—for bringing their expertises to the community. Special thanks to Katherine Liu,
not only for her contribution to improving some of my chapters, but also for convincing me that
i
Minnesota is a great place to pursue my Ph.D. back in March 2016, when I was still waitlisted. I
am also thankful to the members of the Biological Interest Group (BIG), which introduced me to a
wide variety of issues in philosophy, history, and methodology of biology. I have loved discussing
with people in the History of Science, Technology, and Medicine (HSTM) program. I am grateful
especially to Mark Borrello, Luis Felipe Eguiarte Souza, Kele Cable, and Miaofeng Yao for their
friendship and all of the intellectual interactions. Thanks to Gabriela Huelgas Morales and Jack
Powers, who were not only members of the community of the Minnesota Center for Philosophy of
Science (MCPS), but also excellent roommates.
A number of other people have given me useful feedback on different parts of this dissertation:
William Bausman, Sara Green, Rachel Ankeny, Nathan Crowe, Devin Gouvêa, David Colaço,
Marta Halina, Rose Novick, Cory Wright, Ingo Brigandt, Sabina Leonelli, Jun Otsuka, Ryosuke
Igarashi, Masahide Asano, Takuto Enomoto, Jim Griesemer, Celso Neto, and Melinda Bonnie
Fagan. Thank you for all of your criticisms and suggestions. Additionally, several workshops and
summer schools allowed me to broaden and deepen my philosophical perspectives and thereby
contributed directly or indirectly to this dissertation: the Marine Biological Laboratory (MBL)
summer seminars (“A Century of Engineering Life: Cells and Organisms”); MBL workshop (“The
Life Cycles of Microscopic Imaging in Biology”); Banff summer institute (“Practices of Individ-
uation and Classification in Science”); St Andrews workshop (“The Conceptual Legacy of On
Growth and Form: Interdisciplinary Perspectives”); and Institute of Vienna Circle summer school
(“Representation in Art and Science”). I thank the organizers and everyone with whom I interacted
in these events.
I would like to express my appreciation to the cooperative environment of the Department of Phi-
losophy. I am grateful for the friendship and support of Michael Calasso, Chris Nagel, Chris Small,
Qiannan Li, Aaron Vesey, Dongwoo Kim, Justin Ivory, Rachel Pedersen, Tucker Marks, Taylor
Smith, Kylie Shahar, Sara Parhizgari, Michelle Hirschboeck, Codi Stevens, Becca Kosten, and
Manon Andre De St Amant. As somebody who had known little philosophy outside philosophy of
science, I have greatly benefited from course work and working with professors in the department.
Thanks to Jos Uffink, David Taylor, Peter Hanks, Valerie Tiberius, Michelle Mason Bizri, Jessica
Gordon-Roth, Tamara Fakhoury, Joseph Owens, Roy Cook, Cat Saint-Croix, and Sarah Holtman.
Outside the department, I took two graduate seminars provided by the Department of Genetics,
Cell Biology, and Development (GCD). I am grateful especially to Laura Gammill and David
Greenstein for the excellent seminars. My research and learning also have depended crucially
on administrative support. Many thanks to Anita Wallace, Judy Grandbois, Pam Groscost, Aleks
Zarnitsyn, and Janet McKernan.
My transition from a Japanese university to an American university would have been much more
challenging without the fellowship provided by the Japan Student Services Organization (JASSO).
I appreciate that the fellowship allowed me to focus on course work without teaching responsibility
for the first two years. I am also thankful to William J. M. Hrushesky, whose generous donation
to the Minnesota Center for Philosophy of Science allowed me to travel to Oslo for an ISHPSSB
conference in 2019. Thanks to Douglas Lewis, whose publication fellowship provided me with
ii
financial support.
My experience in the Department of Philosophy and History of Science at Kyoto University in
Japan shaped the foundation of me as a scholar of history and philosophy of science. I am grate-
ful to my previous advisors, Tetsuji Iseda and Kazuyuki Ito, who introduced me to history and
philosophy of science and helped me develop academic interest and skills. Hisashi Nakao was
virtually another advisor of mine. I cannot thank him enough for all of the support he provided.
Many thanks to Kohei Morita for the coutless long, critical discussions that we had, as well as
the friendship that continues until today. I am also grateful to a number of people for supporting
me: Akihisa Setoguchi, Shunkichi Matsumoto, Toma Kawanishi, Kotaro Namura, Yuki Sugawara,
Fumiaki Hirashimizu, Mai Sugimoto, Yukinori Onishi, Hajime Inaba, Yuichi Amitani, Katsuya
Takao, Daichi Suzuki, Yoshihiro Maruyama, Naoyuki Kajimoto, Ryuma Shineha, and Gyo Nakao.
In addition, the time I spent as a visiting student in Yoshiko Takahashi’s laboratory at Kyoto Uni-
versity familiarized me with practices of experimental biology and made me choose to focus on
developmental biology (in particular, morphogenesis and organogenesis) in my Ph.D. research. In
particular, I thank Yoshiko Takahashi for accepting me as a member of her laboratory for a year,
and Yuji Atsuta for guiding and supporting my laboratory experience.
Fictional stories have always been an indispensable part of my life. They have provided me with
refreshment, enjoyment, and sometimes opportunities to rethink how I want to live. Taikaisyu (au-
thored by Ken-ichi Ogomori) and A Place Further than the Universe (created by Madhouse) both
reminded me of the value of attempting something without a guaranteed success and encouraged
me to take on challenges. Wonder Egg Priority (created by CloverWorks) made me experience a
lot of emotions—excitement, sadness, surprise, fear, and confusion—which gave me the energy to
keep going.
Finally, I am very grateful to my parents, Toyohiko and Hiroko Yoshida, for continuously acknowl-
edging my interest and having faith in me. My interest in science and love of reading were nurtured
by them; they made me who I am today on a deep level. Many thanks also to my sister, Tomo Mori
for her generous support.
Some chapters of this dissertation are based on already published articles. I am grateful for
permission to use this material here.
• University of Chicago Press
Yoshida Y (2021) Multiple-models juxtaposition and trade-offs among modeling desiderata.
Philosophy of Science 88(1):103–123.
• Elsevier
Yoshida Y (2023) Joint representation: Modeling a phenomenon with multiple biological
systems. Studies in History and Philosophy of Science 99:67–76.
iii
Abstract
Although there is a consensus that pursuits of general knowledge are crucial in almost all fields of
science, themajority of philosophical analyses of generalizations have focused narrowly on universal
generalizations or laws of nature and what role generalizations play in scientific explanations. This
narrow focus has limited the scope of philosophical discussions about scientific generalizations.
This dissertation proposes and exemplifies a broader inquiry into scientific generalizations that
is motivated by the question: how do scientists pursue, formulate, reason about, utilize, and
communicate generalizations? In other words, how are generalizations practiced in science?
To address this broad set of questions, I focus on a particular field—developmental biology—
and examine investigative and representational practices surrounding generalizations. Like many
other fields, developmental biology seeks both widely shared regularities and the details of causal
processes peculiar to specific systems. My analyses show how this dual interest in generality and
specific details is interconnected and mutually contribute to each other.
This dissertation is organized as follows. Chapter 1 provides a brief overview of how philoso-
phers have discussed generalizations. I point out that the interests in laws and explanation have
dominated the past discussions. In contrast, my approach focuses on investigative and representa-
tional practices of generalization, which have received very little philosophical attention. Chapter 2
analyzes two approaches to generalizations in developmental biology: mechanisms and principles.
iv
These are distinguished based on the relevance of abstraction. I show that the two approaches
are associated with different investigative practices. This analysis illustrates what forms of non-
universal generalizations developmental biologists seek and formulate, which serves as a basis for
discussions in the following chapters. Chapter 3 explores generalizations from the perspective of
modeling desiderata. I offer a characterization of what I call multiple-models juxtaposition (MMJ),
a strategy for managing a trade-off between generality and detail in scientific models. MMJ dis-
plays models of distinct processes together and fulfills different desiderata both in the individual
models and by a comparison of those models. I also clarify the distinction between MMJ and
multiple-models idealization (MMI), which also uses multiple models to manage trade-offs among
desiderata. Chapter 4 focuses on the use of model systems. Biologists often study particular bio-
logical systems as models of a phenomenon of interest, even if they know that the phenomenon is
produced by diverse mechanisms and hence none of those systems alone can sufficiently represent
it. I argue that even if generalizability of results from a single model system is significantly lim-
ited, generalizations concerning specific aspects of mechanisms often hold across certain ranges of
biological systems. This enables multiple model systems to jointly represent such a phenomenon.
Chapter 5 considers the question “how and why do scientists generalize?” by challenging three
influential assumptions: (1) generalizations are expressed linguistically; (2) scientists generalize
by formulating a single representation with wide applicability; and (3) generalizations are valuable
because they enable scientific explanations. My analysis of a concrete example illustrates roles
that visual representations play in generalizations. It also shows that formulating a single, unified
representation is not the only way to generalize; scientists often generalize by configuring multiple
representations. Finally, I argue that generalizations serve to facilitate cross-fertilization among
studies of different target systems, which complements the explanation-centered view.
v
Contents
Acknowledgments i
Abstract iv
List of Tables ix
List of Figures xi
1 Introduction: Broadening Philosophical Inquiries into Scientific Generalizations 1
1.1 Philosophizing Generalizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Laws of Nature, Explanations, and Scientific Status . . . . . . . . . . . . . . . . . 2
1.3 A Practice-Based Approach to Scientific Generalizations . . . . . . . . . . . . . . 6
1.4 The Structure of the Dissertation . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2 Mechanisms and Principles: Different Approaches to Scientific Generalizations 14
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2 Generalizations in Developmental Biology . . . . . . . . . . . . . . . . . . . . . . 16
2.3 Mechanisms versus Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4 Consequences for the Structure of Inquiry . . . . . . . . . . . . . . . . . . . . . . 25
vi
2.5 Consequences for Analyses of Scientific Generalizations . . . . . . . . . . . . . . 28
2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3 Multiple-Models Juxtaposition and Trade-Offs among Modeling Desiderata 32
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.2 Generality and Detail . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.3 Case Study: Research on Branching Morphogenesis . . . . . . . . . . . . . . . . . 39
3.3.1 Comparison 1: Shared Molecular Signaling . . . . . . . . . . . . . . . . . 41
3.3.2 Comparison 2: Shared Cellular Activities and Organization . . . . . . . . 44
3.4 Modeling Desiderata and Presentational Choice . . . . . . . . . . . . . . . . . . . 47
3.5 Multiple-Models Idealization and Multiple-Models Juxtaposition . . . . . . . . . . 54
3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4 Joint Representation: Modeling a Phenomenon with Multiple Biological Systems 59
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.2 Model Systems in the Life Sciences . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.3 Collective Cell Migration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.3.1 Fruit Fly Border Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.3.2 Zebrafish Lateral Line Primordium . . . . . . . . . . . . . . . . . . . . . 69
4.3.3 Mouse Mammary Gland . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.3.4 Collective Cell Migration as a Phenomenon . . . . . . . . . . . . . . . . . 71
4.4 Modeling a Phenomenon with Multiple Biological Systems . . . . . . . . . . . . . 75
4.4.1 Insufficiency of the Basic Accounts . . . . . . . . . . . . . . . . . . . . . 76
4.4.2 Local Generalizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
vii
4.4.3 Utility for Characterizing and Investigating Individual Mechanisms . . . . 85
4.5 Joint Representation and Integration-Based Accounts . . . . . . . . . . . . . . . . 87
4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5 Generalization Reconfigured: An Inquiry into Representational and Investigative
Practices of Scientific Generalization 92
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.2 Generalizing with Different Types of Visualizations . . . . . . . . . . . . . . . . . 98
5.2.1 Generalizing with Single Representations . . . . . . . . . . . . . . . . . . 100
5.2.2 Generalizing through Configuring Multiple Representations . . . . . . . . 103
5.2.3 Complementary Relations between Different Types of Visualizations . . . . 110
5.2.4 Responding to Possible Objections . . . . . . . . . . . . . . . . . . . . . . 114
5.3 Generalizations Facilitate Cross-Fertilization . . . . . . . . . . . . . . . . . . . . 117
5.3.1 Promoting Better Characterizations of Individual Mechanisms . . . . . . . 118
5.3.2 Guiding Investigations into Less-Understood Systems . . . . . . . . . . . . 119
5.3.3 Promoting Comparisons between Different Kinds of Systems . . . . . . . . 121
5.3.4 Multiplicity of Generalizations as a Resource . . . . . . . . . . . . . . . . 124
5.4 Philosophical Implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
5.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
6 Concluding Remarks 131
References 134
viii
List of Tables
3.1 Comparison between MMI and MMJ . . . . . . . . . . . . . . . . . . . . . . . . . 57
ix
List of Figures
2.1 BMP signaling mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2 The principle of reaction-diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3 Biological patterns that might be produced by reaction-diffusion processes . . . . . 24
3.1 Examples of branched organs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.2 Fly trachea formation and mammalian lung formation . . . . . . . . . . . . . . . . 43
3.3 Fly trachea formation and vertebrate angiogenesis . . . . . . . . . . . . . . . . . . 46
4.1 Collective cell migration of fruit fly border cells . . . . . . . . . . . . . . . . . . . 68
4.2 Collective cell migration of lebrafish lateral line primordium . . . . . . . . . . . . 69
4.3 Collective cell migration in mouse mammary gland . . . . . . . . . . . . . . . . . 71
4.4 Diagrams of five distinct mechanisms of collective cell migration . . . . . . . . . . 73
4.5 A table that characterizes several mechanisms of collective cell migration . . . . . 81
4.6 There are multiple ways to group mechanisms according to similarity . . . . . . . 83
5.1 A principle diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
5.2 A component mechanism diagram . . . . . . . . . . . . . . . . . . . . . . . . . . 102
5.3 Juxtaposed mechanism diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
x
5.4 Juxtaposed diagrams presenting an abstract principle . . . . . . . . . . . . . . . . 107
5.5 A spectrum presentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
5.6 A table that characterizes several mechanisms of collective cell migration . . . . . 111
5.7 A figure illustrating that collective cell migration in mouse mammary gland . . . . 120
5.8 Spectrum presentations of developmental and cancer systems . . . . . . . . . . . . 122
xi
Chapter 1: Introduction: Broadening Philosophical
Inquiries into Scientific Generalizations
1.1 Philosophizing Generalizations
Science seeks general knowledge. Generality has been treated as one of basic epistemic values
that defines science (Friedman, 1974; Kitcher, 1989; Kuhn, 1977). Many important scientific
achievements, from Newton’s theory of mechanics to the discovery of the broad conservation of
homeobox genes, are celebrated because they revealed regularities that hold across wide ranges
of systems or phenomena. Of course, different fields of science exhibit different degrees of
generalizability. But this fact does not undermine the importance of general knowledge for the
sciences. Even if a field lacks laws of nature or overarching theories, it still seeks and formulates
generalizations that apply broadly to the extent that its subject matter allows. Generalizations are
crucial and prevalent in most fields of science.
This dissertation concerns scientific generalizations. Attention to generalizations is not unique
in philosophy of science. Philosophers have long been interested in generalizations, and this dis-
sertation is one among many inquiries into the topic. However, past philosophical discussions have
focused on a few, very specific issues, such as the nature of natural laws and the role of general-
1
izations in scientific explanations. Although previous studies have promoted our understanding of
these issues, a flip side of this narrow focus is that philosophers have not paid sufficient attention to
diverse aspects of scientific generalizations that concern neither laws of nature nor their indispens-
ability to explanations. This dissertation proposes and exemplifies a broader inquiry into scientific
generalizations. It is motivated by the question: how do scientists pursue, formulate, reason about,
utilize, and communicate generalizations? In other words, how are generalizations practiced in
science?
In this introductory chapter, I situate the dissertation within a context of existing literature and
characterize its investigative approach. I start by providing a brief overview of how philosophers
have discussed generalizations over the last several decades (Section 1.2). This illustrates that much
effort has been devoted to articulating the nature of universal laws and the role of generalizations
in scientific explanations. Then, I contrast the approach that I adopt in this dissertation, which is
inspired by Andrea Woody’s (2014) characterization of practice-based philosophy of science, with
the standard, law- and explanation-centered approach (Section 1.3). Finally, I provide an outline of
this dissertation (Section 1.4).
1.2 Laws of Nature, Explanations, and Scientific Status
Philosophers of science have been interested in scientific generalizations for many years, but in
a few, very specific ways. Much philosophical effort has been devoted to answer the following
questions: What is the nature of laws of nature? Do special sciences have laws? How do
generalizations enable scientific explanations? These questions are closely related—or so many
philosophers have believed. This is illustrated most clearly by the covering law model of scientific
2
explanation (Hempel, 1965). This model regards explanation as derivation of the explanandum
(i.e., the phenomenon to be explained) from a law of nature, where a law is characterized as an
empirically necessary generalization (Hempel, 1965).1 What makes science special is the power
to explain phenomena. Since explanation consists in deriving a phenomenon from a law, having a
law is a necessary condition for a field to genuinely explain phenomena. Therefore, having a law
is crucial for a field to be a genuine (or mature) science. Attempts to define scientific explanation
have been one of the central themes of the “analytic project” of philosophy of science over the last
half century (Barker and Kitcher, 2014, Chapter 2).2 Generalization has attracted philosophers’
attention primarily because it has been regarded as playing a key role in explanation (e.g., Friedman,
1974; Kitcher, 1981; Woodward, 2001).
The association among generalizations, explanations, and scientific status has led to debates
over whether specific non-physical sciences such as biology have laws of nature. Some authors
have argued that biology has no laws. Smart (1963) holds that laws in the strict sense must be
applicable everywhere in space and time and can be expressedwithout referring to any proper names.
According to this view, biology has no genuine laws because we do not expect generalizations
formulated by biologists to apply in all life forms that (may) exist across space and time in the
universe. Any such biological generalizations concern the life forms that have evolved on Earth
and hence has to implicitly refer to the particular planet. Another argument against the existence
of biological laws is provided by Beatty (1995), according to which biological generalizations
1 An example of a covering-law type explanation is a situation where one explains the volume of a gas by
appealing to the ideal gas law. Here, the volume (V ) can be deductively derived from the law (pV = nRT) if
the pressure (p), temperature (T), and amount of substance (n) are known.
2 The analytic project has sought general accounts of what make scientific knowledge particularly reliable
and important. This project has consisted of three major themes: confirmation, theories, and explanation
(Barker and Kitcher, 2014, Chapter 2).
3
describe facts that are merely contingently true as a result of evolutionary history and hence do not
have the necessity required for them to qualify as laws.3
In response to these claims, some philosophers have defended the scientific status of biology.4
One approach is to expand or revise the notion of laws such that generalizations in biology qualify as
laws (e.g., Mitchell, 1997, 2000). Mitchell (2000) proposed a non-dichotomous view of laws, which
characterizes different scientific generalizations in terms of three parameters (stability, strength, and
abstraction). Although biological laws (such as Mendel’s law) might apply much more narrowly
(i.e., less stable) and allow more exceptions (i.e., less strong) than laws of theoretical physics (such
as the law ofmass/energy conservation), the difference is quantitative, not qualitative. They are both
scientific laws; they just exhibit the three qualities to different degrees. Another approach is to accept
that generalizations that are not laws still have explanatory power. For example, Brandon (1997)
argues that evolutionary biology formulates generalizations concerning contingent regularities and
that such generalizations have explanatory power, though it is more limited compared with the
explanatory power of universal laws.
Arguably the most influential response is new developments in the causal approach to explana-
tion. Some authors distinguish generalizations that provide information about causation and those
that do not, and attribute explanatory power only to the former and not to the latter. Waters (1998)
is an example. He distinguishes what he calls distributions and causal regularities; the former
concerns contingent facts about where in the biological world certain properties are distributed,
while the latter concerns regular causal structures that can be used to explain phenomena. Even
3 Gould (1989) proposes a famous thought experiment, “replaying life’s tape,” to illustrate the idea of
contingency of evolutionary history. He argues that if life on Earth evolved again from the same initial
condition, it would result in very different life forms.
4 Braillard and Malaterre (2015) provide a useful review of how philosophers have addressed the issue
of the mismatch between nomological (i.e., law-based) view of explanation and biological explanations.
4
though distributions of particular traits are the result of contingent evolution, biology explains phe-
nomena as long as it elucidates causal regularities that produce the phenomena. Woodward (2001,
2003) makes a similar distinction. He argues that spatiotemporal applicability and distributions
are irrelevant to scientific explanation. What matters for explanation is invariance, which is the
stability of a causal relationship under a range of interventions. Thus, biology does not have to
formulate generalizations that apply broadly across space and time or across species. As long as
biology articulates causal relationships that hold across certain ranges of interventions, it is able to
explain and hence qualifies as a science.
Another version of the causal approach is the mechanistic account of explanation, which
has rapidly developed since the end of the 20th century. This account holds that to explain a
phenomenon is to describe a mechanism for it. Describing a mechanism consists in articulating
how specific types of entities and their activities are organized to produce the phenomenon (Bechtel
and Richardson, 1993; Glennan, 1996; Machamer et al, 2000). Unlike many previous accounts of
scientific explanation, the mechanistic account denies that laws or law-like generalizations play a
crucial role in scientific explanations. Hence, for those who accept that causal-mechanistic accounts
are appropriate models of explanation in (at least many areas of) biology, the lack of laws or law-like
generalizations is no longer a serious problem; biology explains phenomena by articulating causal
interactions that produce them, whether or not law-like generalizations are available. However,
these developments have not ended the debate on laws and explanations in biology. Discussions
continue about whether biology has laws, (if it does) what their nature is, and whether and how
laws play roles in biological explanations (e.g., Lange, 2013; Press, 2015; Raerinne, 2013, 2015;
Reutlinger, 2014a,b).
5
1.3 A Practice-Based Approach to Scientific Generalizations
The above overview illustrates that philosophical analyses of generalizations have addressed a few,
very specific questions: What is the nature of natural laws? Do special sciences have laws? How
do generalizations enable scientific explanation? The studies described in the previous section
have no doubt advanced our understanding of these issues. At the same time, the strong focus on
these issues has limited the scope of philosophical discussions about scientific generalizations. The
assumption that generalizations are valuable as long as they enable explanations has resulted in a
neglect of features of generalizations that are not indispensable to explanation. Because the law-
centered picture has been dominant, non-overarching generalizations with limited scope, which still
are an important part of generalization practice, have not received much philosophical attention.
Moreover, many of the previous studies have adopted a generalist approach, which has sought
an account of generalization that is applicable broadly across fields. In particular, analytically-
oriented studies have tended to prioritize formal rigor, which has required them to focus on toy
examples or highly idealized examples of generalizations. A side-effect of this approach is the lack
of fine-grained analyses of scientific activities surrounding generalizations in particular contexts.
This dissertation aims at broadening philosophical inquiries into scientific generalizations. To
do so, it takes a practice-based approach. What does “practice-based approach” mean here? While
there are different interpretations of practice-based philosophy of science (see Soler et al, 2014),
this dissertation is guided largely by AndreaWoody’s (2014) characterization, which consists of the
following four shifts away from traditional, more analytically-oriented approaches (pp. 123–124):
• Conception to representation: Instead of focusing on theories and models as conceptual
objects which are characterized by logical structures, practice-based philosophy of science
6
treats theories and models as artifacts that are rendered as particular representations (e.g.,
linguistic, diagrammatic, or other).
• A priori to empirical: Instead of analyzing the concepts of interest in a generalist, a priori
manner, practice-based philosophy of science takes a more empirical, bottom-up approach,
which examines the reasoning of scientists in particular contexts.
• Ideal agent to human practitioner: Instead of adopting the perspective of an ideal agent,
practice-based philosophy of science emphasizes that scientists are human practitioners who
have limited cognitive capacities, which are constrained or influenced by various factors.
• Knowing subject to social epistemology: Instead of focusing on individual scientists, practice-
based philosophy of science sees science as a social enterprise, where knowledge is generated
and transmitted under the influences of various features of the community and investigations
are dependent crucially on interactions within and between communities.
Therefore, instead of seeking an account that analyzes generalizations in terms of highly abstract
logical structures that are formulated for an ideal reasoner, I pursue an account that, on the basis
of a close examination of specific case studies, characterizes generalizations as existing in concrete
representational contexts, formulated and used by human practitionerswithin a research community.
Some philosophers have adopted practice-based approaches (which can be characterized in
terms of some or all of the above shifts) to investigate scientific generalizations. For example,
Sara Green closely examines how systems biologists and their predecessors have pursued abstract
principles that apply to different types of biological systems and what roles such principles play
in biological reasoning (Green and Wolkenhauer, 2013; Green, 2015; Green et al, 2015). In phi-
losophy of ecology, a number of authors have discussed the nature, status, kinds, and roles of
7
generalizations in modeling practice in ecology (e.g., Cooper, 1998; Elliott-Graves, 2018, 2022;
Linquist et al, 2016; Turchin, 2001). Some studies focus on processes through which scientists
achieve general knowledge. For instance, William Bechtel characterizes a heuristic strategy em-
ployed to generalize mechanistic explanations in research that relies heavily on a limited number
of model organisms (Bechtel and Abrahamsen, 2005; Bechtel, 2009). Mary Morgan examines
explanatory and reasoning practice in social sciences to analyze how locally generated knowledge
is transferred and generalized in case-based research (e.g., Morgan, 2017, 2019). These studies
shed new light on scientific generalizations. However, many aspects of generalization practice still
remain to be investigated.
While this dissertation shares with the above studies the focus on the practice of generalization, it
is characterized and organized particularly by the following set of questions, each ofwhich highlights
an aspect of generalization practice that has been neglected in past philosophical discussions.
(1) The majority of past effort to categorize generalizations has focused on articulating which
generalizations enable scientific explanation. Instead, we can examine: What different forms
of generalizations are pursued within a particular field? How are they associated with other
investigative practices in that field? (Chapter 2 mainly addresses these questions.)
(2) Once we shift our attention away from overarching laws to generalizations with limited
scope, we realize that it is often the case that multiple generalizations are pursued and
formulated within an area of research. This observation leads to questions such as: How
are multiple generalizations related? Do they function together in scientific research? If so,
how? (Chapters 4 and 5)
(3) Although it has been widely assumed that generalizations must be expressed linguistically
8
(typically in the form of a universally quantified proposition),5 generalizations in scientific
research (especially in the biological sciences) do not always take such a form. What
representational formats do scientists employ when they express generalizations? How does
representationmatter for generalizations serving their roles in scientific research? (Chapters 3
and 5)
(4) Although philosophers have focused primarily on whether and how generalizations enable
scientific explanations, we should not presuppose that that is the only possible role, and even
the most important role, that generalizations play in science. What roles do generalizations
play in science beyond explanation? (Chapters 4 and 5)
In order to tackle these questions through fine-grained description of practice, I narrow the
scope of analysis to a particular field of science. This thesis takes all of its examples from
developmental biology. Developmental biology is a field that studies how various kinds of processes
at the molecular, cellular, and tissue levels generate organismal features during an individual’s life
from fertilization to death (Love, 2014, 2022). Limiting the sources of examples to one field of
science is beneficial because there are field-specific features of generalizations that are relevant
to philosophical considerations. Developmental biology also is a good example of a science that
is interested in identifying regularities as well as in elucidating system-specific details. Although
generality is highly valued, developmental biologists are interested not only in widely shared
developmental processes; they also hope to understand details of developmental processes that are
peculiar to specific biological systems. Such a dual interest in the general and the specific seems
5 A universally quantified proposition is a proposition stating that all instances of a set share a certain
characteristic. A famous example is “all uranium spheres are less than a mile in diameter,” which has been
one of philosopher’s favorite examples of a law of nature (Carroll, 2016).
9
to be shared in many fields of life science and even outside of it. More importantly, pursuits
of regularities and specific details are neither opposed nor separate. Rather, they are closely
interconnected in ways that these pursuits mutually contribute to each other. Therefore, although I
have characterized this project as a philosophical inquiry into scientific generalizations, it is actually
an inquiry into how scientists investigate generality and specificity in the world. The chapters of this
dissertation show different ways in which pursuits of generality and specificity are interconnected.
Below, I give an outline of this dissertation, which provides an overview of how the following
four chapters answer questions (1)–(4) formulated above. Before that, however, it is important
to note that I do not intend to argue that my approach is the only or the most preferable way of
philosophically studying scientific generalizations. The law- and explanation-centered approaches
that I summarized in Section 1.2 have their own advantages that my approach lacks. I regard
philosophical accounts of science as idealized representations of scientific activities, just as scien-
tific models are idealized representations of natural phenomena. Scientists idealize phenomena of
interest in various ways by distorting certain features to build models that fulfill specific purposes,
such as description, explanation, prediction, and intervention. Furthermore, scientists often keep
using different idealizations of the same phenomenon because different idealizations can exhibit
different sets of desiderata and hence have different advantages and disadvantages (Weisberg, 2007,
2013). Philosophers also formulate different idealizations of scientific activities, which are suited
for different purposes. We do not have to always debate which philosophical account is the right
one. I believe that in many situations, a more productive thing to consider is which account is more
suitable for a specific purpose. This dissertation is an attempt at formulating a philosophical ide-
alization of generalization practice that helps us recognize certain (previously neglected) features,
such as their representational and social dimensions and thereby promoting further philosophical
10
inquiries into different aspects of generalization practice.
1.4 The Structure of the Dissertation
This dissertation is structured as follows. Chapter 2 (“Mechanisms and Principles: Different
Approaches to Scientific Generalizations”) focuses on two approaches to generalizations in de-
velopmental biology: mechanisms and principles. These are distinguished on the basis of the
relevance of abstraction. While mechanisms generalize by virtue of specific types of biological
entities and their interactions that are evolutionarily conserved, principles generalize because ab-
stract relations or interactions can be instantiated by different biological entities. I point out that
the two approaches are associated with different investigative practices (e.g., research tools, modes
of reasoning, and confirmation strategies) and hence the distinction enables us to better understand
why research in developmental biology is structured in a certain way. This analysis provides a
picture of what forms of non-universal generalizations developmental biologists seek and identify,
which serves as a basis for discussions found in the following chapters.
Chapter 3 (“Multiple-Models Juxtaposition and Trade-Offs among Modeling Desiderata”) dis-
cusses generalizations from the perspective of modeling desiderata. More specifically, it addresses
the question of how generality and detail—desiderata that are often in a trade-off relationship—are
coordinated in scientific modeling. I offer a characterization of what I call multiple-models jux-
taposition (MMJ), a strategy for managing such trade-offs among different modeling desiderata.
MMJ displays models of distinct phenomena together and fulfills different desiderata both in the
individual models and by a comparison of those models. I use research on branching morphogen-
esis as a case study and show that MMJ has been employed to pursue both generality and detail
11
in representations. I also clarify the distinction between MMJ and Michael Weisberg’s (2007,
2013) notion of multiple-models idealization (MMI), which also uses multiple models to manage
trade-offs among desiderata. MMJ and MMI differ in several points, such as the ways they manage
trade-offs and the purposes of using multiple models.
Chapter 4 (“Joint Representation: Modeling a Phenomenon withMultiple Biological Systems”)
focuses on another crucial aspect of generalization practice in developmental biology: the use of
model systems. A model system is a biological system (a type of cell, tissue, organ, organism,
etc.) that is studied as a convenient loci for investigating a phenomenon of interest. Although many
authors have discussed how biologists generalize or extrapolate findings from a model system to
other biological systems (e.g., Ankeny and Leonelli, 2011, 2020; Bolker, 2009; Levy and Currie,
2015), there is a puzzling practice that has not yet received philosophical scrutiny: biologists often
study certain biological systems as models of a phenomenon of interest, even if they already know
that the phenomenon is produced by diverse mechanisms and hence none of those systems alone
can sufficiently represent it. To understand this modeling practice, this chapter provides an account
of how multiple model systems can be used to study a phenomenon that is produced by diverse
mechanisms. Even if generalizability of results from a single model system is significantly limited,
generalizations concerning specific aspects of mechanisms often hold across certain ranges of
biological systems, which enables multiple model systems to jointly represent such a phenomenon.
Comparing mechanisms that operate in different biological systems as examples of the same
phenomenon also facilitates characterization and investigation of individual mechanisms. I also
compare my account with two existing accounts of the use of multiple model systems and argue
that my account is distinct from and complementary to them.
Chapter 5 (“Generalization Reconfigured: An Inquiry into Representational and Investigative
12
Practices of Scientific Generalization”) combines conceptual resources from the previous chapters
to consider the question: “how and why do scientists generalize?” It does so through challenging
three implicit assumptions that have been influential in philosophical discussions about general-
izations: (1) generalizations are expressed linguistically; (2) scientists generalize by formulating
a single representation with wide applicability; and (3) generalizations are valuable because they
enable scientific explanations. The chapter examines the use of different types of visualizations in
review articles concentrating on mechanistic research into collective cell migration. Scientists em-
ploy not only linguistic representations but also visual representations to formulate generalizations.
Importantly, formulating a single, unified representation is not the only way to generalize. Scientists
often display multiple representations in different configurations, which enables them to character-
ize features shared across different biological systems without ignoring important dissimilarities.
Furthermore, I argue that enabling scientific explanations is not the only role generalizations play
in scientific research. Generalizations also serve to facilitate cross-fertilization among studies of
different target systems by promoting cross-system comparisons and “mutual informing” among
researchers or laboratories focusing on different model systems. This chapter exemplifies that philo-
sophical discussions about scientific generalizations can be substantially informed by examining
representational and investigative practices of generalization.
13
Chapter 2: Mechanisms and Principles: Different
Approaches to Scientific Generalizations
2.1 Introduction
Confirmed empirical generalizations are central to the epistemology of science.1 Through most of
the 20th century, philosophers focused their attention on the special case of universal, exceptionless
generalizations—laws of nature—and took these as essential to both scientific theory structure
and explanation (e.g., Hempel, 1965). However, over the past two decades, many philosophers
of science, and especially those interested in biology, have sought to characterize a broader range
of generalizations. For example, Waters (1998) divided biological generalizations into two cate-
gories: descriptive generalizations concerning the distribution of certain biological entities, and
explanatory generalizations concerning causal regularities. Mitchell (2000) proposes a conceptual
framework for characterizing and categorizing scientific generalizations in terms of three distinct
properties: stability (how widely a generalization holds), strength (how probable a conditional re-
lation is), and degree of abstraction (how many details are ignored). Mitchell argues that scientific
1 This chapter is based on a manuscript that I coauthored with Alan Love, which we presented at the
2022 Philosophy of Science Association biennial meeting.
14
generalizations in different areas exhibit these properties to different extents and thereby rejects
the traditional dichotomy between universal laws and accidental generalizations. More recently,
Green (2015) provides a taxonomy of biological generalizations that consists of two broad cate-
gories: generalizations concerning material homogeneity or causal regularities, and generalizations
derived from constraints imposed on possible behaviors of systems. These analyses have yielded
an effective toolkit that facilitates more illuminating comparisons and a better understanding of
generalizations in diverse areas of scientific inquiry.
A complementary analytical strategy is to go beyond categorizing or taxonomizing generaliza-
tions and examine how different forms of generalizations are associated with different strategies of
inquiry. Such an analysis can illuminate the fine-scale epistemic structure of scientific reasoning
because the rationale presumed by scientists for different forms of generalizations would become
less opaque. In particular, since scientists must differentially evaluate distinct forms of generaliza-
tions, the associated standards of adequacy should be more visible. Additionally, differences in the
inferential and explanatory roles they play can be isolated and explicated. We concentrate here on
two forms of generalizations in developmental biology: mechanisms and principles. Mechanism
generalizations (MGs) are descriptions of constituent biomolecules organized into causal relation-
ships, which operate in specific times and places to produce a characteristic phenomenon and are
shared across different biological entities. Principle generalizations (PGs) are abstract descriptions
of relations or interactions, typically susceptible to mathematical representation, that obtain across
different biological entities.
We begin by discussing investigative aims in the context of developmental biology, which
provides a necessary template for identifying the distinct inferential and explanatory roles that
mechanism generalizations and principle generalizations play in scientific reasoning. We then
15
characterize both forms of generalizations and show that they correspond to different investigative
aims. As a result of our characterization, we can comprehend why one form of generalization
with a particular combination of properties is (or is not) sought in a research context and thereby
understand why the practices of inquiry are structured in a certain way. More generally, our
analysis isolates key issues in prior philosophical discussions of the properties of generalizations,
such as ambiguities regarding “scope” (how widely a generalization holds) and a presumption that
abstraction is always correlated positively with generalizations of wide scope.
2.2 Generalizations in Developmental Biology
The field of developmental biology attempts to explain how embryogenesis occurs through inter-
twined processes of cell differentiation, pattern formation, growth, andmorphogenesis (Love, 2014,
2022). As a consequence, developmental biologists seek to account for a wide range of phenomena,
including but not limited to cell movement, the emergence of different cell types, the origin of body
axes, and organ formation. A generalization consists of a description or explanation of a feature or
process of ontogeny and the range of biological systems where the description or explanation holds
(“scope”). Scope can be represented in terms of four distinct but non-exclusive dimensions: across
taxa, across component systems, across developmental stages, and across spatial scales.2
First, and perhaps most familiar, developmental biologists seek to identify generalizations that
hold across many different taxa, which undergirds the rationale for using a small set of model
organisms in experimental practice. “The motivation for their study is not simply to understand
2 Robert (2004) offers a similar classification relevant to generalizations in stem cell biology: develop-
mental stage, system type, species, genetic and epigenetic background, and experimental setting. Waters
(1998) also points out that distributions of biological properties generalize over various domains, such as
taxa, cell lineage, or spatial regions of organisms of a taxon.
16
how that particular animal develops, but to use it as an example of how all animals develop”
(Slack, 2006, p. 61). Second, developmental biologists seek to generalize across component
systems of organism(s), such as organs or tissues. For example, blood vessels share patterns
of organization with the nervous system because their development involves responsiveness to
localized molecular cues that initiate reticulation and guide them to destinations on similar spatial
scales throughout the organism (Carmeliet and Tessier-Lavigne, 2005). Third, developmental
biologists seek generalizations across time in embryogenesis, such as whether adult blood vessels
are made through the same developmental processes as those working earlier in development. This
is because a generalizationmight not hold for all developmental stages, even for the same component
system of the same species. Although the initial formation of vasculature (vasculogenesis) and adult
formation of vasculature (angiogenesis) share a number of molecular features, the two processes
are not identical, which is relevant for treating cardiovascular pathologies (Fancher et al, 2008).
Fourth, developmental biologists seek generalizations that hold across spatial scales. For example,
the relative role of general optimization principles can differ across spatial scales and thereby
account for patterns of asymmetry in vascular networks (Tekin et al, 2016).
For all of these four dimensions, generalizations can be descriptive or explanatory. Descriptive
generalizations are critical for characterizing features of developing embryos and identifyingwhat is
in need of explanation. The claim that lungs, kidneys, and blood vessels of mammals and salivary
glands, trachea, hindgut, and dorsal appendages of fruit flies are formed through budding is an
across-taxa and across-component systems descriptive generalization (Iruela-Arispe and Beitel,
2013).3 It describes a type of morphogenetic process (budding) that is manifested in different
3 Budding is a process where cells form a new tube by invaginating from a sheet of cells or preexisting
tube.
17
systems (e.g., lung, kidney, or vasculature) and across different taxa (mammals and insects). In
contrast, an explanatory generalization for the growth of blood vessels in vertebrates involves the
causal claim that vascular endothelial growth factor (VEGF) secreted by a nearby tissue binds to its
receptor (VEGFR) on a vascular cell, which causes the cell to migrate toward the source of secreted
VEGF and directs a growing blood vessel. This mechanism is common across different vertebrate
taxa (e.g., zebrafish, chicken, and mouse) (Ochoa-Espinosa and Affolter, 2012).
In most cases, descriptive and explanatory generalizations appear in mixed combinations. For
example, an explanatory generalization that depicts specific causal interactions is often presented
together with one or more descriptive generalizations about the relevant entities or processes that
exhibit the causal relationship. Explanatory generalizations usually do not provide every detail
of the relationships they represent; the above claim about vertebrate vascular growth does not
include how VEGF binding to VEGFR triggers changes in a vascular cell’s mobility. Many details
are intentionally abstracted away for the sake of convenience or because they are held fixed in
experiments that establish the causal relationship (see, for example, Strevens, 2011; Woodward,
2003).
Another issue related to characterizing generalizations is the conditions under which a de-
scriptive or explanatory generalization holds. At least two conditions are relevant: material and
conceptual. The former refers to experimental settings (e.g., in vivo or in vitro), and the latter
is concerned with research contexts and the framing of inquiry, such as what research questions
a generalization pertains to (e.g., research questions about morphogenesis vs. research questions
about differentiation). Conceptual conditions are especially relevant for explanatory generalizations
because they make explicit why a generalization is explanatory (i.e., because it answers a research
question from one domain rather than another).
18
Both of these conditions are critical for understanding descriptive and explanatory generaliza-
tions in developmental biology. For example, a generalization in one experimental setting, such as
the pattern of gene expression detected via in situ hybridization in arthropod segments, may not
hold in a different experimental setting where patterns of protein accumulation are detected via
antibody staining in the same segments (Abzhanov and Kaufman, 1999). Similarly, an explanatory
generalization that answers a research question in one conceptual context, such as the mechanism
of anterior-posterior axis formation in animals (Kimelman and Martin, 2012), typically does not
export to another context where different research questions are in view, such as the mechanisms
of neuron formation (Reichert, 2009). And just as descriptive and explanatory generalizations can
appear together in various combinations with different dimensions emphasized, so also the values
for different material and conceptual conditions can be independent of each other and represented
differently in generalizations. Often, aspects of these dimensions and conditions are only implicit
in scientific discourse.
Similar to many other sciences, developmental biologists seek both descriptive and explanatory
generalizations. These are structured in four different dimensions—across taxa, across component
systems, across developmental stages, and across spatial scales—and in terms of two conditions:
material and conceptual. These generalizations often appear in complex combinations within
scientific reasoningwhere different dimensions or conditions are foregrounded (e.g., distributions of
developmental phenomena and causal interactions that underlie them in a specific component system
at a particular stage under specified experimental conditions to answer some subset of research
questions). This reconstruction of the geography of generalizations in developmental biology
positions us to see a crucial distinction between two distinct forms of explanatory generalizations.
19
2.3 Mechanisms versus Principles
Explanatory generalizations in developmental biology account for why developmental processes
operate in a particular fashion and yield specific phenotypic outcomes across taxa, across component
systems, across developmental stages, or across spatial scales for specified material and conceptual
conditions. Two different forms of explanatory generalizations can be distinguished: mechanism
generalizations (MGs) and principle generalizations (PGs). The former concern specific types of
interactions (especially molecular ones) that play roles in different biological systems, whereas the
latter concern relations or interactions that can be instantiated by heterogeneous entities. A key
aspect of the difference between them can be understood in terms of the role of abstraction or the
intentional omission of details in scientific representations. For a PG, the properties of particular
entities, such as a transcription factor or a signaling molecule, are abstracted away and only the
relation or interaction among them is represented. Although a MG can and often does involve
abstraction, the specific molecule types and particular interactions among them are not abstracted
away and play the primary explanatory role.
Consider a common molecular mechanism in developmental biology: bone morphogenetic
protein (BMP) signaling (Fig. 2.1). BMP signaling generally involves secreted BMPs binding to
BMP receptors (BMPRs) on the cell surface, which triggers phosphorylation of Smad proteins
in the cytoplasm. This, in turn, transduces the signal and activates transcription factors that can
translocate to the nucleus and regulate specific genes (Wang et al, 2014). This change in gene
expression makes a difference in the formation of diverse phenotypic features and manifestation
of various developmental processes. Thus, BMP signaling is a mechanism involving cell-cell
interactions that functions to regulate expression of particular genes through a chain of specific
20
Figure 2.1: The bone morphogenetic protein (BMP) signaling mechanism (Wang et al, 2014, Fig. 1).
21
molecular interactions. It is considered a MG when BMP signaling can be shown to operate across
taxa, across component systems, across developmental stages, or across spatial scales to produce
particular features of development under specified conditions.
Many processes in early development are dependent on BMP signaling for cell growth,
apoptosis, and differentiation. BMPs also play important roles in maintaining adult
tissue homeostasis, such as the maintenance of joint integrity, the initiation of fracture
repair, and vascular remodeling. (Wang et al, 2014, p. 88)
This quotation emphasizes the generality of BMP signaling across component systems (joints and
vasculature) and developmental stages (early development and adult tissue), as well as highlighting
several different conceptual conditions (e.g., research questions about growth and differentiation).
A significant PG in developmental biology involves processes of reaction and diffusion produc-
ing distinctive outcomes (Fig. 2.2). Models of reaction-diffusion processes show how biological
patterns (e.g., pigment stripes or dots) can emerge from an initially homogeneous state through
the interaction of entities (typically molecules) that change in concentration and diffuse at different
rates in a spatially contained area (Turing, 1952). In a simple case, an activator molecule promotes
the production of an inhibitor molecule and of the activator itself, whereas the inhibitor retards the
activator’s production. If the inhibitor diffuses faster than the activator, then initially homogeneous
distributions of the activator and inhibitor can change into periodic patterns based on their concen-
trations (Green and Sharpe 2015). This is a PG when the relevant type of reaction-diffusion process
can be shown to operate across taxa, across component systems, across developmental stages, or
across spatial scales to produce periodicity under specified conditions (Fig. 2.3).
[Reaction-diffusion processes] are thought to underlie many different examples of de-
velopmental patterning, including mesendodermal and left-right organization, mam-
malian palatal rugal ridge formation, hair follicle spacing, finger formation and nano-
features of insect cornea. (Davies, 2017, p. 1148)
22
Figure 2.2: The principle of reaction-diffusion (Torii, 2012, Fig. 1). (a) The simplest version of the reaction-
diffusion model involves two factors, an activator (A) and inhibitor (I) that satisfy the following conditions:
A activates its own production as well as the production of the I; I inhibits the production of A; and I diffuses
at a faster rate than A. (b) Reaction-diffusion processes can produce different periodic patterns depending on
different values for their parameters. (c) Differential equations that represent the relations and interactions
between A and I.
23
Figure 2.3: Examples of biological patterns for which the involvement of reaction-diffusion processes is
studied or suggested. A: hair follicle distribution in the mouse embryo (Cheng et al, 2014, Fig. 1A). B:
feather bud distribution in the chicken embryo (Painter et al, 2012, part of Fig. 3a). C: skin pigmentation
of the giant pufferfish (https://commons.wikimedia.org/wiki/File:Giant_Puffer_fish_skin_
pattern.JPG).
This passage focuses on across-taxa (e.g., mammals and insects) and across-component system
(different anatomical features and tissue organization) dimensions of generalization under different
conceptual conditions where the reaction-diffusion model is applicable (i.e., research questions
regarding different aspects of pattern formation, such as left-right asymmetry and epithelial peri-
odicity).
Although both MGs and PGs provide general explanations of developing systems, they require
distinct research strategies and are justified differently. The account of BMP signaling includes
many particular details (BMPs, BMPRs, Smads, phosphorylation, etc.). These details are critical for
understanding how the signaling mechanism operates (e.g., BMPs bind to BMPRs, not something
else). However, the reaction-diffusion model involves more abstract descriptions of its constituents
(“activator” or “inhibitor”) and their interactions (“promote” or “retard”). Different types of entities
instantiate the same reaction-diffusion processes in different developmental contexts (Green and
Sharpe, 2015; Kondo and Miura, 2010). The specific details do not matter as much as the
overarching “principle”: periodicity emerges from types of interactions among entities that only
24
depend on their concentration and diffusion rate in a bounded region. “These principles stand above
the level of the specific details of any particular developmental system” (Davies, 2017, p. 1146).
This is in sharp contrast with MGs that hold precisely because of specific details in particular
developmental systems.
This difference in the relevance of abstraction between MGs and PGs is related to the difference
in what underwrites or justifies their scope, which is characterized in terms of the four dimensions.
The scope of a MG is a consequence of evolutionary conservation; the same type of mechanism
with particular molecular components is common across different biological entities because it has
been conserved through evolutionary history and operates in the same context of development or
has been co-opted into different developmental contexts (spatially or temporally). The scope of a
PG does not have the same basis. Wide applicability of a PG is based on logical, mathematical or
physico-chemical rules (e.g., laws of diffusion and mass action), many of which express abstract
relationships that can be exhibited by heterogeneous entities (molecular and otherwise). These may
or may not correlate with patterns of evolutionary history.
2.4 Consequences for the Structure of Inquiry
An important result of our analysis is that we can better account for why scientific practices are
structured in different ways to isolate different forms of generalizations. An illuminating example
is the coincidence between the identification of MGs and the use of model organisms. We can
understand this coincidence by considering the distinction between conservation and applicability.
Recall that the motivation for studying a model organism: “to use it as an example of how all
animals develop” (Slack, 2006). Developmental biologists study a relatively small number of
25
species as exemplars and extrapolate the results to related taxonomic groups (Ankeny and Leonelli,
2011). This extrapolation is justified empirically by identifying conserved mechanisms (i.e., MGs)
across model organisms (i.e., across taxa) where phylogenetic relationships among species have
been established (Love, 2018). If researchers have found that a mechanism is shared by several
vertebrate model organisms, then this supports the inference that the mechanism is conserved
throughout most vertebrates (i.e., a non-universal generalization). Thus, evolutionary relationships
undergird the conservation that is the basis for MGs across taxa and this is discoverable in concrete
practices of model organism research.
In contrast, the applicability of PGs is not based on evolutionary conservation and instead is
dependent on whether physico-chemical properties or formal principles are instantiated in one or
more of the four dimensions. This illuminates why PGs are often applicable across spatial scales;
in these cases, specific constituents and their interactions differ. MGs do not often generalize across
spatial scales because specific types of activities of specific types of biomolecules are confined
to one spatial scale. The dependence of PGs on whether physico-chemical properties or formal
principles are instantiated also informs why computer simulations are often pursued to identify PGs
but not MGs. Abstract logical and mathematical properties can be modeled in silico but specific
molecular interactions must be experimentally dissected in concrete model organisms. Therefore,
recognizing the distinct bases for MGs and PGs illuminates why developmental biologists use
particular strategies of inquiry, such asmodel organisms to elucidateMGs and computer simulations
to identify PGs.
A related issue is the distinctive kinds of difficulties involved in confirming MGs and PGs. A
PG involves very few entities and a limited set of interactions, all modeled with a high degree
of abstraction. This makes PGs amenable to formal representation and facilitates establishing a
26
how-possibly explanation (Bokulich, 2014). Mathematical tools (such as differential equations)
are available for demonstrating with precision that a principle can account for the phenomenon of
interest in appropriate circumstances. However, empirically confirming a PG—that is, showing that
the principle is actually instantiated by specific entities in a biological system—is often challenging.
It is not easy to confirm whether the principle is causally responsible for the phenomena and, if so,
which entities and activities in the system instantiate the principle. Furthermore, even if entities
that implement the principle are identified in a system, the involvement of those entities is not
necessarily generalizable because the principle can be instantiated with different entities across
taxa, across component systems, across developmental stages, and across spatial scales.
A good example of this difficulty can be observed in studies using reaction-diffusion models
to explain the formation of digit and fin patterns in different vertebrate species. That the pertinent
physico-chemical process can be realized by different entities and activities in different taxa makes
comparisons across taxa difficult. Mouse digits are patterned in part through interactions among
Bmp, Sox9, and Wnt proteins, whereas chicken digits are patterned in part through interactions
between Gal1a and Gal8 (Stewart et al, 2017). Whether Bmp, Sox9, and Wnt interactions are
relevant for patterning chicken digits, or Gal1a and Gal8 interactions are relevant for patterning
mouse digits remains unclear, though the latter is empirically unlikely because mutant mice lacking
Gal1 show no negative phenotypic effects (Georgiadis et al, 2007).
Unlike a PG, a single MG—such as the involvement of BMP signaling in cell growth and dif-
ferentiation across taxa—often involves many entities and their activities organized into a complex
set of causal interactions. Developmental biologists are equipped with various experimental tech-
niques to detect whether specific molecular types and their activities are present. These facilitate
the empirical confirmation of the complex set of causal interactions that compose the mechanism in
27
a studied system. Furthermore, it is empirically established that molecular mechanisms are widely
conserved through evolution, which supports extrapolation of a mechanism from one system to
another. However, the initial characterization of a mechanism is often laborious because of its
complexity. Dissecting the various elements of the relevant causal pathways is experimentally
demanding. And, even after a MG is established, the causal dynamics of an entire mechanism are
typically understood only qualitatively and without mathematical precision. MGs are not as suit-
able for mathematical formalization as PGs because of their dependence on particular biochemical
details.
2.5 Consequences for Analyses of Scientific Generalizations
Although our analysis has focused on MGs and PGs in the context of developmental biology, it
has consequences for more general discussions of scientific generalizations. First, the distinction
between MGs and PGs is pertinent to other sciences. For example, some generalizations in
physiology describe specific metabolic pathways common across a wide range of taxa. The citric
acid cycle is a series of reactions through which acetyl-CoA is oxidized and releases chemical
energy (Pratt and Cornely, 2014). This cycle is widely conserved across taxa, although there are
variations in which enzymes are involved. This is a MG because its generality depends crucially on
reactions that occur among specific substances. Other generalizations in physiology are PGs, which
concern abstract relations between properties of an organism. For instance, during the life history
of pelagic (i.e., living in open water) animals, metabolic rates often increase in a 1:1 proportion to
body mass. This PG applies to diverse pelagic animals across five different phyla (Glazier, 2006).
Our analysis is also applicable to ecology. According to a meta-analytic study (Liao et al, 2008),
28
plant invasion increases carbon and nitrogen stocks in an ecosystem across types of invasive species
(woody and herbaceous; nitrogen-fixing and non-nitrogen-fixing) and across types of ecosystems
(forests, grasslands, and wetlands). This is an example of a MG in ecology that describes the
behavior of specific entities (carbon and nitrogen) across different situations of species movement
and ecosystem composition. On the other hand, consumer-resource oscillation, which is represented
typically by the Lotka-Volterra equations, is a PG in ecology that can be implemented by a variety of
organisms (Turchin, 2001). Importantly, some aspects of our analysis do not apply straightforwardly
to other fields. For example, the four dimensions we characterized are specific to developmental
biology; different fields can have different numbers and kinds of dimensions for generalizations.
Nevertheless, the core ideas of our analysis—dimensions, conditions, and the distinction between
MGs and PGs—illuminate research practices in a wide range of scientific fields.
A second general consequence of our analysis is that it isolates ambiguities in discussions of
“scope” (how widely a generalization holds). The scope of scientific generalizations has often been
represented in terms of the width of application in space and time (e.g., Mitchell, 2000; Smart,
1963). Philosophers of biology also have discussed generalizations distributed across biological
entities (e.g., Robert, 2004; Waters, 1998). We isolated four distinct dimensions related to the
scope of generalizations for both MGs and PGs in developmental biology: across taxa, across
component systems, across developmental stages, and across spatial scales. We also identified both
material and conceptual conditions that calibrateMGs and PGs. (Other sciences, such as physiology
or ecology, will have various dimensions and conditions associated with their generalizations as
well.) Dimensions of scope and different conditions correlate with different explanatory and
inferential roles that distinct types of generalizations play. In developmental biology, MGs provide
explanations based onmechanisms conserved across taxa that are identified through the investigation
29
of concrete model organisms, whereas PGs provide explanations applicable across spatial scales
due to the irrelevance of specific constituents and are modeled in silico. All of this implies that
characterization of the scope of a generalization must take into account various factors specific to
disciplinary and investigative contexts.
A final consequence of our analysis has to do with the relationship between abstraction and
generality. It might appear natural to assume that abstraction is crucial for generalizations of wide
scope, but our analysis of MGs shows otherwise. Unlike PGs, the wide scope of MGs depends
on particular details being conserved widely in different biological systems through evolutionary
time. This does not mean that wide scope is never positively correlated with abstraction in MGs.
For example, BMP falls under a larger category of proteins called the transforming growth factor 
(TGF-) superfamily, which includes other signaling proteins (Wu and Hill, 2009). Consequently,
the TGF- superfamily signaling mechanism is more abstract than BMP signaling mechanism,
and, in some circumstances, the former has wider scope than the latter (for specified dimensions
and conditions). However, a crucial point of our analysis is that MGs in developmental biology—
unlike PGs—can have wide scope (for specified dimensions and conditions) even without such
abstraction because of the evolutionary conservation of specific constituent molecules and their
specific activities.
2.6 Conclusion
Generalizations play central roles in scientific research. In this chapter, we focused on developmen-
tal biology to distinguish two kinds of explanatory generalizations: mechanisms and principles.
MGs describe specific entities organized into causal relationships that operate in particular times
30
and places during ontogeny to produce a characteristic phenomenon, and are shared across different
biological entities by virtue of evolutionary conservation. PGs are abstract descriptions of relations
or interactions that are exemplified in a wide variety of different entities. This distinction, along
with the characterization of associated dimensions and conditions for generalizations, accounts for
how particular research practices are structured, clarifies ambiguities in prior discussions of scope,
and demonstrates that increased abstraction does not always facilitate generalizations of wide scope.
Additionally, our analysis is germane to generalizations across a broad range of scientific fields.
Similar studies within and across diverse areas of science are needed to provide further insights
about the nature and role of scientific generalizations.
31
Chapter 3: Multiple-Models Juxtaposition and
Trade-Offs among Modeling Desiderata
3.1 Introduction
The formulation of scientific models is often constrained by trade-offs among modeling desiderata,
that is, qualities that are valuable to scientists but that may or may not be exhibited by a particular
model.1 Aclassical discussion of this issue is Levins (1966) analysis ofmodel building in population
biology. Levins describes the difficulties involved in formulating a model that is maximally general,
realistic, and precise at the same time. Although generality (applicability of a model to a wide range
of things), realism (faithfulness of amodel as a representation of the target), and precision (regarding
predictions made by a model about the target) are all valuable, population biologists cannot build a
single model that maximizes all three desiderata simultaneously. As a consequence, they typically
sacrifice one desideratum to make a model with one or more of the other desiderata. For instance,
one approach focuses on particular cases of the phenomenon of interest and makes a model based
on accurate measurements; the resulting model can produce precise and realistic predictions, but its
applicability is narrow. Another approach makes an unrealistic model that ignores parameters like
1 This chapter is based on an article published in Philosophy of Science (Yoshida, 2021).
32
time lags and physiological states but is generally applicable and can produce precise predictions
(Levins, 1966).
It is debatable whether the exact trade-off relationship that Levins describes really holds. Orzack
and Sober (1993) take Levins to be discussing a trade-off derived from properties of the formalism
used in population biology and argue that there is no such trade-off relationship among generality,
realism, and precision. Odenbaugh (2003) disagrees with Orzack and Sober and argues that the
trade-off does exist in the sense that pragmatic constraints involved in population biologists’model
building do not allow the simultaneous fulfillment of those desiderata in a single model. Despite
such disagreement over interpretation and evaluation of the detail of Levins’s account, philosophers
nonetheless agree that its core insight is right and useful; that is, in model building and evaluation
in science, some desiderata exist in trade-off relationships and hence cannot be simultaneously
maximized or increased in a single model (Odenbaugh, 2003; Weisberg, 2004; Matthewson and
Weisberg, 2009; Matthewson, 2011; Gelfert, 2013). How do scientists manage such trade-offs? As
I alreadymentioned, a strategy that Levins advocates is to formulate more than onemodel to address
a phenomenon. Weisberg (2007, 2013) develops this idea and calls the strategy multiple-models
idealization (MMI). MMI makes use of multiple models that idealize the phenomenon of interest
differently. Those models prioritize different desiderata and collectively help scientists achieve
their goals (explanation, prediction, understanding, control, etc.). MMI is used most often when
researchers are investigating highly complex phenomena.
Although MMI is an important strategy that is adopted in various contexts in science, it is
not the only strategy available to manage trade-offs among modeling desiderata. This article
discusses another strategy, which also involves multiple models but is distinct from MMI. In
scientific practice, models of multiple related phenomena are sometimes displayed together in a
33
representation. This presentational practice facilitates the fulfillment of certain desiderata through
comparison of those models, while maintaining desiderata that the individual models exemplify.
Consequently, desiderata that a single model cannot exhibit simultaneously can be fulfilled in the
set of juxtaposed models. I call this strategy multiple-models juxtaposition (MMJ).
To illustrate how MMJ works, I conduct a case study that focuses on investigations of branch-
ing morphogenesis in developmental biology. Mechanistic models of the formation of different
branched organs are often displayed together and compared in review articles. Moreover, the par-
ticular mechanistic models that are compared have changed over time in the field. I argue that we
can understand this presentational practice as a case of MMJ that coordinates a trade-off between
the desiderata of generality and detail. Although researchers of branching morphogenesis are
interested in both providing detailed descriptions of individual branching mechanisms and finding
general features of them, these are often in a trade-off relationship for a single model. By display-
ing multiple models together and comparing them, researchers provide generalizations of features
shared across the distinct mechanisms while keeping detailed descriptions of those mechanisms in
the individual models. The shift from one comparison to another over time can also be understood
in terms of modeling desiderata. It reflects the preference for generalizing certain features of the
mechanisms over another. Thus, we can account for this case by adopting the idea of MMJ.
The next section discusses generality and detail asmodeling desiderata and the trade-off between
them, especially as manifested in the context of mechanistic explanation (Section 3.2). Then I turn
to the case study (Section 3.3). As a part of this case study, I describe a shift from one comparison
of mechanisms to another over time, with a special focus on what commonalities were highlighted
in each comparison. Section 3.4 asks three questions about the case: Why do the researchers
of branching morphogenesis often display multiple models together? Why has one comparison
34
become more common than the other? And why do the researchers not make a single unified model
instead of formulating multiple models and displaying them together? Answers to those questions
illustrate how MMJ manages the trade-off between generality and detail. Section 3.5 compares
MMJwithMMI. Despite their apparent similarity, these strategies have several contrasting features.
This comparison leads us to a deeper understanding of MMJ, as well as promoting new inquiry
into MMI.
3.2 Generality and Detail
Generality and detail are often (but not always) in a trade-off relationship. Generality is a desider-
atum shared in many fields and contexts of science. A model is general when it is applicable to
a wide range of things (Matthewson and Weisberg, 2009; Matthewson, 2020). I use an inclusive
expression (“a wide range of things”) because different sciences have different domains of inquiry.
Even within biology, generalizations can hold over various domains, such as geographical regions,
taxa, cell lineages, spatial parts of an organism, and periods of time, depending on which subfield
of biology a generalization belongs to (Waters, 1998). Here I focus on developmental biology,
where generality of a model is often understood in terms of taxa, component systems of an or-
ganism, developmental stages, and spatial scales. “The induction of numerous organs is effected
by a relatively small set of paracrine factors. The embryo inherits a rather compact genetic ‘tool
kit’ and uses many of the same proteins to construct the heart, kidneys, teeth, eyes, and other
organs. Moreover, the same proteins are used throughout the animal kingdom—the factors active
in creating the Drosophila eye or heart are very similar to those used in generating mammalian
organs” (Gilbert, 2014, p. 84). This passage emphasizes that a signaling mechanism often accounts
35
for developmental phenomena across component systems (e.g., “heart, kidneys, teeth, eyes, and
other organs”) as well as across taxa (e.g., “throughout the animal kingdom”).
The desideratum of detail consists in the inclusion of relevant features of the target system in
the model. A detailed model provides substantial information about component features of the
system it represents. Although detail might appear similar to realism, these are distinct desiderata.
Realism is the faithfulness of a model as a representation of a target system, and this faithfulness
depends on the extent to which false assumptions or idealizations are not included in a model. The
more realistic a model is, the fewer false assumptions it will have (i.e., the less idealized it is). But,
detail concerns the extent to which relevant features of the target system are included in a model.
The more detailed a model is, the fewer features of the target system it will ignore (i.e., the less
abstract it is).2
Detail plays an important role in mechanistic explanation. A mechanistic model is often
detailed in a specific way. According to a philosophical formulation called minimal mechanism, a
“mechanism for a phenomenon consists of entities (or parts) whose activities and interactions are
organized so as to be responsible for the phenomenon” (Glennan and Illari, 2017, p. 2). Hence, a
model of a process underlying a phenomenon is considered mechanistic when it provides relevant
details about what entities are involved, what those entities do, and how those activities and
interactions are organized to bring about the phenomenon. These details are often represented in
the form of diagrams, where visual representations facilitate the reasoning of and communication
among researchers (Sheredos et al, 2013; Abrahamsen and Bechtel, 2015; Abrahamsen et al, 2017).
It is important to note that “detail is a desideratum” does not mean that more detailed models
2 Throughout this article, I use the term “abstraction” as a property of representations. It means omission
or ignorance of detail in representations and has nothing to do with abstract entities (see Levy and Bechtel,
2013).
36
are always preferable. What degree of detail is appropriate depends heavily on the communities,
contexts of research, and the questions being asked. Similarly, what types of relevant features must
be included in the model can vary depending on the situation (Levy and Bechtel, 2013; Bechtel,
2021).
Generality and detail are often in a trade-off relationship. This relationship is illustrated by the
following example. A simple model of a cell in which only the cell membrane and nucleus are
depicted is applicable to most eukaryotic cells. This is because characteristic features of different
types of eukaryotic cells are abstracted away from the representation. But if the model includes an
axon and dendrites, then it becomes less general; the model is applicable only to neurons. If more
characteristic features of a specific type of neuron are added to the model, its generality decreases.
The more detailed a representation is, the less general it is likely to be.
This trade-off is an obstacle when both generality and detail are pursued. We can find examples
of such situations in many life sciences, where describing mechanisms in detail is a common way of
explaining biological phenomena, although many researchers also seek widely applicable models.
The simplest response to this trade-off is to seek a balance between generality and detail in a single
model. There are biological processes that can be regarded as occurring in a stereotypical manner
when they are modeled at the level of detail appropriate for the purposes of research. For instance,
a simple model of synaptic transmission in an introductory biology textbook is sufficiently general
as well as sufficiently detailed because the model, which describes the transmission mechanism in
a degree of detail appropriate for a novice, is applicable to a broad range of instances (i.e., different
types of neurons in different species). Such reconciliation of generality and detail in a single model
is a matter of widespread importance in the formulation of mechanistic models, as well as in many
other fields and contexts in science. However, it is not always effective. Phenomena that scientists
37
hope to explain are often patterned while also being variable, in particular in the life sciences. It is
often the case that an interesting feature is shared in a range of processes, whereas there are also
non-negligible dissimilarities among them. In such cases, if one formulates a model with enough
detail for the purposes of research, generality of the model may substantially decrease; that is, detail
and generality of an appropriate degree cannot be reconciled in a single model.
Matthewson (2020) calls attention to a different way to reconcile generality and detail in a
single model. This representational strategy describes a causal interaction in a relatively abstract
way, while inserting in the representation a component mechanism of the interaction as an inset.
This inset provides information about the detail of the mechanism that is a part of the causal
interaction in specific cases. This does not impair the generality of the entire model because one
can recognize the abstract pattern in the whole causal interaction and understand that the pattern
is applicable to various phenomena, even if the inserted component mechanism occurs only in a
subset of them. Another strategy that can coordinate the trade-off between generality and detail
is MMI. One can fulfill both desiderata by constructing an abstract model of a phenomenon that
is generally applicable to many instances and a detailed model of the same phenomenon that has
narrower applicability. For example, whereas chronobiologists build a detailed model of a gene
regulation mechanism underlying certain periodic behavior that includes specific types of genes
and proteins, they also construct an abstract model of the same mechanism that ignores specific
types of entities and focuses on its organization (e.g., the causal connectivity of the mechanism).
The former model provides detailed descriptions of specific cases, while the latter model enables an
explanation that can be applied to a wide range of cases (Levy and Bechtel, 2013; Bechtel, 2021).
MMJ coordinates generality and detail in a way different from any of these strategies. It
involves multiple models about related but distinct phenomena and provides a detailed description
38
of a causal process in each model and a generalization of shared features through a comparison of
the multiple models. Let us move on to a concrete case to see how this strategy works in scientific
practice.
3.3 Case Study: Research on Branching Morphogenesis
My case study is taken from research on branching morphogenesis in developmental biology.
Branching morphogenesis is a set of processes by which branched structures in organs (e.g., blood
vessels, kidneys, lungs, and mammary glands) are formed through various cellular behaviors in
biological development (Fig. 3.1). Developmental biologists have studied cellular and molecular
mechanisms that produce such branched structures. They have been interested not only in how each
branched structure is made but also in how similar the mechanisms are across species and across
organs (and sometimes across developmental stages and across spatial scales). This dual interest
in generality and detail is expressed most typically in review articles, where the authors seek to
provide detailed descriptions of individual branching mechanisms as well as discuss applicability
of and similarity between those mechanisms (e.g., Davies, 2002; Affolter et al, 2003, 2009; Ochoa-
Espinosa and Affolter, 2012; Varner and Nelson, 2014; Spurlin and Nelson, 2017).
I focus on two comparisons of mechanistic models of branching morphogenesis, each of
which is between models of two mechanisms that operate in different biological systems. Around
2000, researchers often compared the mechanism for fly trachea formation with the mechanism
for mammalian lung formation.3 In this comparison, shared molecular types involved in both
mechanisms were highlighted. I call this comparison 1. Over the last decade, however, it has
3 The trachea system of insects consists of ramifying epithelial tubes that directly transport the air to
tissues throughout the body for gas exchange.
39
Figure 3.1: Some examples of branched organs within the mouse (Lu and Werb, 2008, Fig. 1): salivary
gland (A), lung (B), kidney (C), and mammary gland (D).
become more common for review articles to compare the mechanism for fly trachea formation
with the mechanism for vertebrate angiogenesis.4 This new comparison emphasizes shared cellular
activities and overall organization, instead of molecular types that trigger morphogenesis. I call
this comparison 2.
I selected this example because it exemplifies clearly howMMJ functions to manage a trade-off
betweenmodeling desiderata. Importantly, what the case illustrates is not peculiar to developmental
biology. Pursuit of both generality and detail in representations is common in many fields of life
science, such as molecular biology, cell biology, physiology, and neuroscience, and researchers
4 Angiogenesis refers to blood vessel formation through branching from preexisting blood vessels.
40
in these fields often display models of distinct phenomena together to achieve these kinds of
comparisons (e.g., Harmer et al, 2001; Ryan and Grant, 2009; Fontana et al, 2010). Hence, although
I concentrate on a single example, the pattern of practice that I describe and the philosophical
insights derived from it have a wide range of application. Indeed, MMJ may be used even outside
of the life sciences, although I leave this as a topic for future research.
As we saw in the previous section, a mechanistic model provides information about relevant
entities, their activities, and the overall organization of those activities. When mechanistic models
are compared, it becomes important to determine how many of these components are similar and
how similar those components are (Love, 2018). Note that we are here talking about the compar-
ison of mechanistic models in general, not about the judgment of the evolutionary conservation
of mechanisms. Although evolutionary conservation is an important reason why we find similar
mechanisms in different living systems (Bechtel, 2009; Halina, 2018), the comparison of mech-
anistic models precedes the judgment of evolutionary conservation and can be seen as a distinct
activity.
3.3.1 Comparison 1: Shared Molecular Signaling
In 1996, it was reported that a newly found fly homolog of the fibroblast growth factor (FGF),
which was named Branchless (Bnl), is required for trachea formation in fruit flies (Sutherland et al,
1996).5 Bnl is a protein secreted by the tissue surrounding the trachea. It guides branching of
the trachea by attracting cells at the leading tip of the developing tubes. Following this finding,
evidence was provided that FGF-10 is secreted in mesenchymal tissue surrounding the developing
epithelial tubes in the mouse lung and that FGF-10 guides the directional growth and controls
5 FGFs are a family of growth factors known to regulate various developmental processes.
41
branching of the lung (Bellusci et al, 1997; Min et al, 1998; Park et al, 1998; Sekine et al, 1999).
Some of the studies pointed out the similarity between the mechanism for fly trachea formation
and the mechanism for mouse lung formation. “In both Drosophila and mouse the production
of a branched respiratory system involves the directional movement of respiratory cell precursors
towards a localized source of an FGF ligand, either by migration and elongation, or by outgrowth
of epithelial buds” (Bellusci et al, 1997, p. 4876). In this comparison, the most crucial similarity
between the two mechanisms was the involvement of the same kind of molecule (i.e., FGF). FGF
molecules (Bnl in fly trachea and FGF-10 in mouse lungs), which are secreted locally in nearby
tissues, guide the directional outgrowth of the tubes by stimulating their extension. However,
there were also important differences. For example, cellular activities underlying tube extension
appeared to be different. In fly trachea formation, cells that experience a high concentration of FGF
become migratory and move toward the source of FGF. In contrast, it was known that cell migration
is not the driving force of the directional outgrowth during mouse lung development (Nogawa et al,
1998).6
A review article that compared the two mechanisms displayed diagrams of these mechanisms
together, which is shown in Fig. 3.2 (Metzger and Krasnow, 1999). These diagrams highlight
the shared type of signaling molecule (FGF) and its role of guiding the extension of the tubes.
The figure also showed some differences, such as cellular activities and overall organization. The
difference in cellular activities is indicated in the right diagrams; fly trachea cells at the tip of the
tube extend filopodia for migration, whereas mouse lung cells are not migratory and maintain the
6 Cellular activities during mouse lung development is less understood even today. It was suggested that
branching in the lung is based on differential proliferation. The idea was that pulmonary cells that experience
a high concentration of FGF become more proliferative, which causes the directional outgrowth (Bellusci
et al, 1997). However, a later study suggested that the formation of pulmonary outgrowth is not likely to be
based on differential proliferation (Nogawa et al, 1998).
42
Figure 3.2: Mechanistic models of fly trachea formation (top) and mammalian lung formation (bottom) are
displayed together (Metzger and Krasnow, 1999, Fig. 2B, 2C), where the commonality between the two
mechanisms is highlighted by the use of the same colors, which can be seen in the online version of this
figure. Differences (the activity of the epithelial cells and the overall organization) are also indicated visually.
Reprinted with permission of the American Association for the Advancement of Science. Color version
available as an online enhancement.
smooth, sheet-like structure throughout the tube.7 This and the difference in the number of tubular
cells involved in each outgrowth make the overall organization of these mechanisms quite different.
Therefore, comparison 1 accentuated the common type of secreted signaling molecule (FGF).
It also involved the difference in cellular activities (tube extension based on directed migration
vs. nonmigratory extension), which made the overall organization of the mechanisms dissimilar as
well.
7 Filopodia are long and thin cytoplasmic projections that play important roles in cell migration.
43
3.3.2 Comparison 2: Shared Cellular Activities and Organization
An understanding of vertebrate angiogenesis at the molecular level developed several years after
the basic mechanisms for fly trachea formation and mouse lung formation were elucidated. It was
demonstrated that vascular endothelial growth factor A (VEGF-A) secreted from a nearby tissue
guides angiogenic sprouting by promoting filopodia extension from the cells at the tip of angiogenic
vasculature.8 VEGF-A also promotes proliferation of cells located at the stalk of the blood vessels
(the former are called “tip cells,” while the latter are called “stalk cells”; Gerhardt et al. 2003).
Studies published several years later revealed that Notch signaling is involved in the specification
of tip and stalk cells.9 Vascular cells that experience a high concentration of VEGF acquire the tip
cell phenotype and prevent neighboring cells from becoming tip cells by lateral inhibition through
Notch signaling (Hellström et al, 2007; Siekmann and Lawson, 2007). Then, stalk cells push the
tip cells by elongating the tube through proliferation and rearrangement. These studies, along with
further studies of fly trachea formation, revealed interesting similarities between the mechanisms
for fly trachea formation and for vertebrate angiogenesis. For example, it had been shown that
determination of migratory cells and other cells in the fly trachea is also based on lateral inhibition
through Notch signaling (Llimargas, 1999; Ghabrial and Krasnow, 2006). As a result, over the last
decade, researchers of branching morphogenesis have focused more on comparison 2, rather than
comparison 1. It has become common for review articles to display the mechanistic models of fly
trachea formation and vertebrate angiogenesis (instead of fly trachea formation and mammalian
lung formation). Some authors explicitly mention the shift and emphasize the similarity between
8 VEGFs are a family of growth factors that are known for their roles in the development and maintenance
of blood and lymphatic vessels.
9 Notch signaling is a signaling pathway that works between cells in contact and plays crucial roles in
various developmental processes.
44
themechanisms compared in comparison 2: “Formany years, trachea branching has been compared
most often to mammalian lung branching because both organs are involved in oxygen transport and
because branching of both organs is controlled by FGF signalling ... However, recent studies have
unravelled unexpected and stunning similarities in cellular behaviour between trachea branching in
D. melanogaster [fruit fly] and angiogenic sprouting in vertebrates” (Affolter et al, 2009, p. 833).
The mechanisms compared in comparison 2 have features in common different from those
compared in comparison 1. What is salient is the similarity of cellular activities. During both fly
trachea formation and vertebrate angiogenesis, tubular cells that experience high concentrations
of the signaling molecules become tip cells; tip cells migrate toward the source of the signaling
molecules; tip cells prevent neighboring cells from becoming tip cells by lateral inhibition through
Notch signaling; and those neighboring cells become stalk cells, which collectively push the tip
cells. (Tip cells and stalk cells are defined in terms of the functions they play during branching
morphogenesis, and these terms are now used for describing both mechanisms.) Because of these
similarities, the overall organization of the mechanisms is very similar as well. However, there
are also several non-negligible differences. The most significant difference is the types of secreted
signaling molecules that guide tip cell migration. FGF plays this role in the fly trachea, whereas
VEGF accomplishes it during vertebrate angiogenesis, and FGF and VEGF are not very closely
related. There is also a difference in cellular activities. Stalk cells push the tip cells by intercalation
during fly trachea formation, whereas during vertebrate angiogenesis, the pushing force of stalk
cells is based on their proliferation and rearrangement.
Figure 3.3 shows some examples of displaying the two mechanistic models together in review
articles about branching morphogenesis (Ochoa-Espinosa and Affolter, 2012; Spurlin and Nelson,
2017; Wang et al, 2017). They reveal the cellular activities and overall organization shared between
45
Figure 3.3: Examples of figures where the mechanistic models of fly trachea formation and vertebrate
angiogenesis are displayed together. Commonalities (roles of tip and stalk cells and interaction between
them) are highlighted by the use of text, colors (which can be seen in the online version of this figure), and
flat-edged arrows. Differences (types of the diffusing signalingmolecules and activities of stalk cells) are also
indicated visually and textually. A: Ochoa-Espinosa and Affolter (2012), Fig. 2. Left, fly trachea formation;
right, vertebrate angiogenesis. B: Spurlin and Nelson (2017), Fig. 3. Top, vertebrate angiogenesis; bottom,
fly trachea formation. C: Wang et al (2017), Fig. 2B, 2C.
46
the two mechanisms. The functional difference between tip and stalk cells (tip cells migrate toward
the source of the diffusing ligands, whereas stalk cells push the tip cells and extend the outgrowth)
is highlighted by distinct colors or text within the diagrams. The interaction between tip and
stalk cells—lateral inhibition through Notch signaling—is also clearly shown in all figures by flat-
edged arrows and text (a flat-edged arrow is the convention to represent an inhibitory influence).
These mechanism diagrams are displayed with very similar compositions, which emphasizes the
similarity in overall organization of the two mechanisms. Each pair of mechanistic models also
shows differences between the two mechanisms. For example, they indicate by text within the
diagrams the different types of proteins (Bnl/FGF in fly trachea formation; VEGF in vertebrate
angiogenesis) that play the role of attracting tip cells. They also show the difference in how stalk
cells push tip cells (intercalation during fly trachea formation; proliferation and rearrangement
during vertebrate angiogenesis) visually and textually.
Comparison 2 can be summarized as follows. It highlights the commonality of some cellular
activities and overall organization (directedmigration, lateral inhibition, pushing of tip cells), aswell
as signaling thatmediates the tip-stalk interaction (Notch signaling). There are also differences, such
as types of secreted signaling proteins (FGF vs. VEGF) and some cellular activities (intercalation
vs. proliferation and rearrangement).
3.4 Modeling Desiderata and Presentational Choice
This section asks three questions about the cases introduced in the previous section: (a) Why do the
researchers of branchingmorphogenesis—in particular, those writing about comparison 2—display
multiple mechanistic models together? (b) Why has comparison 2 become more common than
47
comparison 1? And (c) why do the researchers of branching morphogenesis not make a single
unified model instead of formulating multiple models and presenting them together? This section
answers these questions by focusing on the two desiderata, generality and detail, and the trade-off
between them.
First, why do the researchers of branching morphogenesis display multiple models together? A
short answer to this question is to highlight features shared across different branching mechanisms.
This reflects the desideratum of generality. Researchers of branching morphogenesis are interested
in which pairs (or groups) of mechanisms are more similar than others. Take comparison 2, for
example. The two mechanistic models are displayed together so that the similarities between them
are easily recognized. The same terms (i.e., “tip cell” and “stalk cell”) are used to characterize
both mechanisms. The similar roles these types of cells play are made evident by the use of distinct
colors, text, or arrows in the diagrams. The similar compositions of the diagrams also facilitate
recognition of the resemblance in overall organization. Through these forms of representation,
those diagrams highlight general features—cellular activities and overall organization—of the two
mechanisms.10 Importantly, the generalization highlighted in comparison 2 has influenced re-
search on branching morphogenesis and related areas. For instance, reflecting on the similarity
between the mechanisms for fly trachea formation and vertebrate angiogenesis, Muñoz-Chápuli
(2011) asks why they are so similar and proposes a hypothesis that they have evolved by adopting a
conserved hypoxia-response mechanism. Another example is the suggestion that fly trachea system
might serve as a useful model of tumor angiogenesis because of the known similarity between the
10 Generality is not the only answer to why researchers display multiple models together because we
can ask further why the strategy of MMJ is adopted instead of another strategy for generalization. A fuller
answer is provided in my response to the last question, which is about why researchers do not produce a
single, unified model.
48
mechanisms for trachea formation and angiogenesis (e.g., Murray, 2015). These examples show
that comparison 2 is more than just a convenient way to summarize different branchingmechanisms;
it sometimes influences directions of research.11
Note that each of the mechanistic models being compared is by itself a generalization. For
example, the mechanism for vertebrate angiogenesis underlies the development of different blood
vessels (e.g., retinal blood vessels and intersegmental arteries) in different species (e.g., mouse
and zebrafish) at different developmental stages (e.g., embryonic and adult Siekmann et al, 2008).
However, by presenting the model of this mechanism together with the model of the mechanism
for fly trachea formation, one can capture features common across these mechanisms. The shared
cellular activities and overall organization highlighted by this comparison are more general than
those of each mechanism.
Next, why has comparison 2 become more common than comparison 1? This shift occurred
in part because the studies showed that the mechanism for vertebrate angiogenesis is similar in
many respects to the mechanism for fly trachea formation. However, this is not enough as an
answer because we cannot say that the mechanism for vertebrate angiogenesis is more similar to
the mechanism for fly trachea formation simpliciter than to the mechanism for mammalian lung
formation. Recall that although comparison 2 highlights cellular activities and overall organization
shared across the two mechanisms, it also involves differences, such as in types of signaling
molecules that guide the directional outgrowth. Recall also that comparison 1 captures exactly the
11 These cases also suggest that there aremultiple reasonswhy scientists pursue generality. Generalizations
sometimes serve to identify objects for further investigation (e.g., the evolutionary basis of the similarity
between the different developmental mechanisms). In other cases, they help in the search for useful models
to study systems of particular interest (e.g., fly trachea as a model of tumor angiogenesis). We should
understand the value of generality for scientists as consisting of both its intrinsic value and usefulness for
other purposes. For more detailed discussion of what roles generalizations play in science, see Chapters 4
and 5.
49
latter kind of feature, that is, the type of signalingmolecule, that is shared across the mechanisms for
fly trachea formation and mammalian lung formation. We cannot judge which pair of mechanisms
are more similar without having decided what kind of features of those mechanisms are of interest.
Therefore, to have a deeper answer to the question of why comparison 2 is now more common than
comparison 1, we must further ask: Given the known facts about these three mechanisms, why
has comparison 2 (which focuses on the shared cellular activities and overall organization) become
more common than comparison 1 (which focuses on the shared type of signaling molecule)?
To answer this question, we have to consider two subclasses of the desideratum of detail: cellular
detail and molecular detail. Cells and molecules are two major types of entities that constitute
developmental mechanisms. A model with more cellular detail provides more information about
cellular features of the target phenomenon, such as what kind of cells are involved and how those
cells act and interact. Similarly, a model with more molecular detail involves more information
about molecular features of the target. Within the field of developmental biology, how these
desiderata are treated varies depending on subcommunities, contexts of research, and questions
being asked. Where only cellular detail is pursued, researchers formulate models that focus on
cellular features and include very little information about molecular features. In other situations,
mechanistic models that focus on molecular detail are formulated (Love, 2018). There are also
situations in which both desiderata are pursued, as is the case for comparisons 1 and 2. Even in the
latter type of situation, however, one desideratum can be prioritized over the other.
I argue that comparison 2 has become more common than comparison 1 because cellular detail
is prioritized in the studies of branching morphogenesis. Branching morphogenesis falls under
the category of morphogenesis. Morphogenesis is characterized as a set of processes by which
three-dimensional biological structures are formed through various kinds of cellular behaviors
50
(Davies, 2013, p. 3). Thus, cellular features and dynamics have been regarded as a key component
of morphogenesis (e.g., Trinkaus, 1969). Consider, for example, the following passage from Scott
Gilbert’s Developmental Biology (a well-known textbook in the field): “During development, cells
divide, migrate, and die; tissues fold and separate. Our fingers are always at the tips of our hands,
never in the middle; our eyes are always in our head, not in our toes or gut. This creation of
ordered form is called morphogenesis, and it involves coordinating cell growth, cell migration, and
cell death” (Gilbert, 2014, p. 2; emphasis added). We can find a similar idea in older texts, for
example: “The word ‘morphogenesis’ is often used in a broad sense to refer to many aspects of
development, but when used strictly it should mean the moulding of cells and tissues into definite
shapes” (Waddington, 1956, p. 433; emphasis added).
To solve problems of branching morphogenesis involves articulating how various kinds of
cellular behaviors (proliferation, migration, death, etc.) produce three-dimensional branched struc-
tures. For those who have a strong interest in morphogenesis, in particular generalizations therein,
comparison 2 is preferable to comparison 1 because the former generalizes cellular activities and
organization. This explains why comparison 2 has become more common over the last decade. In
the late 1990s, when the detailed mechanism of vertebrate angiogenesis was not known, researchers
were focusing on comparison 1, which highlights the shared signaling molecule. However, once
cellular aspects of vertebrate angiogenesis were revealed, attention has shifted to comparison 2
because it fits their interest in (generalizing) cellular features. This does not necessarily mean that
comparison 2 is a better comparison simpliciter. It is possible that in different communities or
contexts where generalizing molecular features is prioritized, comparison 1 becomes more com-
mon. Also, it does not mean that researchers of branching morphogenesis are not interested in
signaling molecules; molecular detail is also important for them (see below). What the claimmeans
51
is that, with respect to the interest in common cellular behaviors, there is a good reason to prefer
comparison 2, since its focus of generalization pertains to cellular features.
Now consider the third question: Why do the researchers not make a single unified model? If
the researchers of branching morphogenesis want to generalize the cellular activities and overall
organization common across the mechanisms for fly trachea formation and vertebrate angiogenesis,
there is a simpler way than displaying multiple models together. A single model that can explain
both phenomena could be formulated by abstracting certain elements away from each model.
Recall the differences between the two mechanisms in comparison 2 (e.g., FGF vs. VEGF). By
ignoring these differences, one could make a single model that is a little more abstract but is more
general than both mechanistic models. Such a model would contain only the information about the
common features between the twomechanisms; it would use generic expressions without specifying
the differences (e.g., “signaling molecule” instead of “FGF” and “VEGF”). Nevertheless, many
authors choose to display the two mechanistic models together instead of formulating such a unified
abstract model. Why do they do this?
My answer is that it is because such an abstract model does not provide enough molecular
detail, and molecular detail is an important desideratum in developmental biology. Identifying
molecules that trigger a developmental process has been regarded as a crucial step in articulating
an explanation since the molecularization of the research on development (Burian and Thieffry,
2000; Hopwood, 2009). Because of this desideratum, the researchers of branching morphogenesis
usually do not abstract away key signaling molecules from their representations. They adopt MMJ
instead.
On the one hand, the researchers seek to formulate the general features of the different mech-
anisms for branching morphogenesis; in particular, they are interested in generalizing cellular
52
activities and overall organization, which reflects the desiderata of generality and of cellular detail.
On the other hand, the molecules that trigger the branching processes—FGF or VEGF in the cur-
rent case—must not be ignored because of another desideratum shared widely in developmental
biology (i.e., molecular detail). MMJ is adopted because it enables satisfying all of these desiderata
simultaneously. The multiple models displayed together can represent cellular activities and the
overall organization shared across the two mechanisms for fly trachea formation and vertebrate
angiogenesis, while keeping the information about the signaling molecules, which differ between
these mechanisms. MMJ serves to coordinate the desiderata that are in the trade-off relationship.
Before proceeding to the next section, I defend my focus on MMJ against a potential objection.
MMJ is a presentational strategy and concerns the summarization and communication of models
in review articles, textbooks, and conference presentations. This might cause the reader to wonder
whether MMJ is worthy of philosophical investigation. While the presentation of models might
have practical import for scientists, does it have any distinct philosophical implications? I respond to
this objection by appealing to a functional approach to philosophy of science articulated by Andrea
Woody (2004, 2014, 2015). For example, (Woody, 2014) uses the example of the explanatory roles
of the periodic table to argue for the philosophical importance of representation. According to
Woody, it is the specific representation (i.e., periodic table), rather than the abstract content being
represented (i.e., periodic law), that has explanatory power in chemistry. To understand how the
periodic law enables particular explanatory activities, we must look at the representational practice.
This approach recognizes that scientists are practitioners with limited cognitive capacities and that
science is a social enterprise (Woody, 2014). What scientists can do with their models depends in
part on how those models are presented and represented (e.g., Sheredos et al, 2013; Abrahamsen
and Bechtel, 2015; Abrahamsen et al, 2017). My case study makes a point analogous to Woody’s
53
argument by focusing on generalization instead of explanation. As I showed, the desideratum of
generality is often fulfilled by comparisons of multiple models, which are facilitated by a specific
form of presentation. If we simply look at the abstract content of individual models and ignore how
they are presented and represented, then we would not understand what and how generalizations
are embraced in the field. This in turn means that we would not understand how the trade-off
between generality and detail is managed. Therefore, presentation and representation are essential
to understand the justification of significant and widespread scientific reasoning activities.
3.5 Multiple-Models Idealization and Multiple-Models Juxta-
position
So far, this article has focused on characterizing MMJ and analyzing how it functions. This section
asks how and in what respects MMJ is different from MMI. MMI is characterized as a strategy to
manage trade-offs by fulfilling different desiderata in different models (Weisberg, 2007, 2013). This
strategy is effective, especially when scientists aim to account for a highly complex phenomenon.
If the phenomenon of interest is highly complex, then no single model that is tractable to human
beings can instantiate all of the modeling desiderata that scientists hope to fulfill. This problem
can be dealt with by formulating multiple models with different desiderata. Weisberg highlights
a few examples where MMI is adopted: ecologists’ constructing multiple models to explain a
phenomenon such as predation, chemists’ reliance on both molecular orbital and valence bond
models, and the use of multiple models of global circulation by the US National Weather Service
to predict weather.
Both MMI and MMJ use multiple models to fulfill desiderata that a single model cannot exhibit
54
simultaneously because of trade-offs among them. However, there are some contrasting features
between the two strategies. The most crucial difference is the ways they manage trade-offs among
modeling desiderata. MMI can fulfill desiderata that are in trade-off relationships because the
multiple models it formulates are of different types (e.g., they are based on different idealization
assumptions), and hence they can exhibit or maximize different desiderata. The point of MMI is
that each model fulfills distinct desiderata. In contrast, the multiple models in MMJ are formulated
and represented in the same way. In comparisons 1 and 2, for example, the models displayed
together are both mechanistic models; they are depicted at the same resolution (i.e., in the same
degree of detail) and adopt the same rules of representation (i.e., colors, arrows, and text are used
consistently in the two models; Figs. 3.2, 3.3). MMJ does not rely on different types of models to
manage trade-offs. Instead, it avoids trade-offs by fulfilling certain desiderata through a comparison
of models that are distinct from the desiderata exhibited by the individual models. The source of
generality is a comparison of models, while detail is instantiated by the individual models.
Model comparison itself is not unique to MMJ. It plays a role also for MMI in what is called
robustness analysis. Robustness analysis is a means to judge the reliability of predictions made
by models (Levins, 1966; Weisberg, 2006, 2013; Wimsatt, 2012). Most models involve some
idealization, and it is sometimes difficult to determine which features of a model reflect properties
of the target phenomenon and which features are attributed to idealization assumptions peculiar
to the model. A comparison of multiple models facilitates this judgment. If those models make
a common prediction, even though they are based on different idealization assumptions, then that
prediction is robust and thus likely to capture a genuine feature of the target phenomenon. There is
a crucial difference between robustness analysis in MMI and model comparison in MMJ. Although
multiple models are compared to identify features common across them in both cases, these
55
are epistemically distinct activities. Robustness analysis is aimed at evaluating how reliable the
predictions made by the individual models are. It compares the models to identify components of
those models that are likely to reflect genuine features of the target phenomenon (Weisberg, 2013).
Model comparison in MMJ is not aimed at such an evaluation of the components of individual
models but at fulfilling certain desiderata. The reliability of aspects of the individual models is not
at stake. Displaying multiple models together in MMJ tells us nothing about which components of
those models are reliable. Instead, it highlights features shared across the target phenomena in a
way that we can easily recognize them.
This difference in the use ofmodel comparison suggests another difference about the purposes of
usingmultiplemodels. InMMI,multiplemodels are formulated to account for a broad phenomenon
from different perspectives. The point of the multiplicity of models in MMI is that they contribute
in different ways to the understanding of the same phenomenon. (Note that these multiple models
need not and often do not represent exactly the same target phenomenon; they are used to address
the same phenomenon in MMI.) In contrast, the purpose of using multiple models in MMJ is
to compare multiple distinct phenomena by comparing models of them. Here the multiplicity
of models reflects the multiplicity of the target phenomena. For example, fly trachea formation
and vertebrate angiogenesis are distinct subclasses of branching morphogenesis. The multiple
mechanistic models are formulated to compare them efficiently.
Table 3.1 is a summary of the contrasting features of MMI and MMJ. MMI is an idealization
strategy to manage trade-offs among modeling desiderata by formulating multiple models on the
basis of different idealization assumptions so that those models exhibit different (combinations of)
desiderata. It uses the multiple models to better account for a single phenomenon. A comparison
of models is not essential to MMI, but it is sometimes conducted for robustness analysis in which
56
Table 3.1: Comparison between MMI and MMJ (Yoshida, 2021, Table 1). MMI: multiple-models idealiza-
tion. MMJ: multiple-models juxtaposition.
models are compared to judge which components of them are likely to reflect genuine features of the
target phenomenon. MMJ is a presentational strategy to manage trade-offs by fulfilling modeling
desiderata both in individual models and by a comparison of those models. Model comparison is
essential to MMJ. It displays multiple models of the same type that represent distinct phenomena
to facilitate a comparison between them. Although I have focused on dissimilarities between MMI
and MMJ, the purpose of this section was to contrast the two strategies in order both to provide a
more detailed characterization of MMJ and to identify potential interesting issues concerning MMI
that have not been analyzed in depth. By comparing the two strategies and trying to articulate how
and in what respects they are different, we can acquire a deeper understanding of both of them.
3.6 Conclusion
This article characterized MMJ as a strategy for managing trade-offs among different modeling
desiderata. MMJ displays models of similar but distinct phenomena together and fulfills desiderata
57
that are in trade-off relationships in the individual models and by a comparison of those models.
This point is illustrated by the case study of branchingmorphogenesis, where mechanistic models of
the formation of different branched structures are frequently displayed together in a representation.
I focused on two desiderata that are important for this area of research, generality and detail, and
showed that MMJ is adopted to manage the trade-off between them. I also compared MMJ with
MMI. Although both strategies use multiple models to fulfill desiderata in trade-off relationships
simultaneously, there are interesting differences, such as the ways they manage trade-offs, the types
of models they involve, how they usemodel comparison, and the purposes of usingmultiple models.
This comparison helps us better understand the scientific practice of MMJ, as well as MMI, and
suggests that further inquiry is needed to understand the diverse ways that scientists fulfill a variety
of modeling desiderata.
58
Chapter 4: Joint Representation: Modeling a
Phenomenon with Multiple Biological Systems
4.1 Introduction
Contemporary biology often studies particular biological systems, such as organisms, asmodels of a
phenomenon of interest, where the models are expected to serve as convenient loci for investigating
the phenomenon (Ankeny and Leonelli, 2011, 2020; Bolker, 2009; Levy and Currie, 2015).1 This
chapter discusses how such a modeling practice works in a specific type of situation: when the
target phenomenon occurs through diverse mechanisms. Some biological phenomena are brought
about by very different mechanisms. For example, studies of developmental biology have shown
that mechanisms underlying certain developmental phenomena differ significantly across taxa and
organs. A consequence of such diversity is a limitation on the degree to which a single biological
system represents the phenomenon of interest. This point is exemplified by the following passage
from a review article on branching morphogenesis,2 where the authors point out that responsible
mechanisms are so diverse that the phenomenon cannot be modeled by a single biological system:
1 This chapter is based on an article published in Studies in History and Philosophy of Science (Yoshida,
2023).
2 Branching morphogenesis is a phenomenon in which a branched biological organ or structure (such as
nerves, blood vessels, lungs, kidneys, and mammary glands) is formed.
59
“the differences between [organ systems] are large enough to suggest that no single branching
epithelium can be considered as representative of the development of all branching systems”
(Varner and Nelson, 2014, pp. 2756–2757). Interestingly, even when it is already known that a
phenomenon is produced by diverse mechanisms, biologists often keep regarding certain biological
systems as models of that phenomenon. In the above example, although the authors recognize the
differences inmechanisms among organ systems, that does not stop them from treating certain organ
systems as models of the phenomenon of interest: “The advent of fluorescent reporter strategies,
including tissue-specific promoter-driven transgenic expression, and of mosaic reporters has begun
to reveal the dynamics and kinematics of branching morphogenesis in a variety of model organs”
(Varner and Nelson, 2014, p. 2750; emphasis added).
This chapter asks how we should understand such a modeling practice in mechanistic research.
In what sense are biological systems regarded as “models” of a phenomenon when there is not a
single mechanism for the phenomenon to be elucidated? A related question is: why do biologists
keep treating such a phenomenon as one thing? When an apparent phenomenon is produced
by diverse mechanisms, this could be taken to suggest that the apparent phenomenon is actually
multiple phenomena, each ofwhich can be represented sufficiently by particular biological system(s)
(Craver, 2004; Craver and Darden, 2013). However, there are cases in which researchers do not
give up a phenomenon and keep regarding certain biological systems as its models, not as models
of its distinct subclasses, even after recognizing the diversity of underlying mechanisms. Why?
I aim to explain this modeling practice by focusing on a concrete example: research on collective
cell migration. Studies in the last few decades have revealed that diverse cellular and molecular
mechanisms bring about the phenomenon of collective cell migration in different organs and taxa
(e.g., Friedl and Gilmour, 2009; Mishra et al, 2019; Rørth, 2009; Scarpa and Mayor, 2016). Yet
60
certain biological systems are still regarded as model systems of collective cell migration and
this phenomenon remains a single, legitimate object of research. I argue that there are good
epistemological and methodological reasons for this practice. To explain this, I focus on how
multiple model systems are used together to study this phenomenon. Even if generalizability from
a single model system is significantly limited due to the diversity of mechanisms, generalizations
concerning specific features of mechanisms still hold in certain ranges of biological systems (which
I call “local generalization”; see Section 4.4). Furthermore, which mechanisms are regarded as
similar to a given mechanism varies depending on which aspect of mechanisms one focuses on.
Consequently, each model system can represent different subclasses of the target phenomenon with
respect to different features of the mechanisms. This makes it possible for multiple biological
systems to jointly represent the target phenomenon. I also argue that regarding different biological
systems as models of a single phenomenon facilitates comparisons between them, which help
characterization and investigation of individual mechanisms. These considerations clarify why
collective cell migration remains a legitimate object of research that is studied by employing
multiple model systems.
This study provides a new contribution to the philosophical literature on how multiple bio-
logical model systems are used together. Philosophers of science are aware that biological model
systems do not function in isolation; model systems very often work in combination with other
models. It has been pointed out that a model organism plays a role in building and maintaining
a research community, where various representational practices (such as mathematical and dia-
grammatic modeling) are integrated through the use of the organism (Ankeny and Leonelli, 2020).
Furthermore, some authors have examined how multiple biological model systems are combined to
fulfill specific research purposes. Their discussions focus on how findings from different biological
61
systems are integrated, either to develop a single, overarchingmechanistic account of a phenomenon
(Baetu, 2014) or to elucidate a single, specific target system by using multiple surrogate models
complementarily to address practical limitations peculiar to the individual models (Fagan, 2016;
Green et al, 2021). I contrast my analysis of research on collective cell migrationwith these previous
accounts. This comparison shows that there is another way in which multiple model systems are
studied together. The type of integration of research findings from different model systems, which
is central in the previous accounts, is not the major goal in research on collective cell migration
as a whole. Instead, my analysis reveals how multiple generalizations (each of which is about a
specific feature of mechanisms and holds in a certain range of biological systems) and cross-system
comparisons facilitate the elucidation of individual mechanisms operating in different biological
systems.
The goal of this chapter is to formulate a representational relationship between a phenomenon
that occurs through diversemechanisms andmultiplemodel systems. To do so, I focus on articulated
mechanisms and researchers’ treatment of them in a concrete example. This is not because I believe
that other aspects of biological model systems are irrelevant or less important. As some authors
emphasize, whether a biological system is a good or plausiblemodel of the target phenomenon relies
on various factors, including the availability of institutional and political resources that facilitate
the use of the system (e.g., Ankeny and Leonelli, 2020; Dietrich et al, 2020). But a comprehensive
discussion of all such relevant factors is beyond the scope of this chapter.
This chapter is structured as follows. Section 4.2 clarifies the notion of biological model system.
Section 4.3 introduces research on collective cell migration with an emphasis on the diversity of
mechanisms underlying it. Section 4.4 analyzes the case. It first shows that accounts of how a
single biological system works as a model cannot fully accommodate the case of collective cell
62
migration. Then it provides a new account, which focuses on how multiple biological systems can
jointly serve as models of a phenomenon that occurs through diverse mechanisms. Section 4.5
compares my account with two existing accounts of the use of multiple model systems. I argue that
my account is distinct from and complementary to them.
4.2 Model Systems in the Life Sciences
In this chapter, “model system” refers to a biological system, such as a type of cell, tissue, organ, or
organism that is studied to learn about a phenomenon of interest. Model systems’ representational
roles can be characterized in terms of representational scope and representational target (Ankeny
and Leonelli, 2011, 2020).3 Representational scope of a model system refers to the range of
biological systems to which findings from the model system might be projected. Representational
target of a model system refers to the specific phenomenon to be explored by employing that model
system. A classic example of a model system is the squid giant axon in neurophysiological research
(e.g., Hodgkin and Huxley, 1952). The squid giant axon (model system) was studied to articulate
the phenomenon of nerve conduction (representational target), and this led to the discovery of the
process of action potential, which turned out to underlie nerve conduction in different nerves of
different species (representational scope).
I use the term “model system” to highlight the idea that not only an organism, but also a
component system of an organism, such as a cell, tissue, organ, etc., can and often do serve
as a model in biological and biomedical research.4 Although this idea is not novel (see, for
3 Although these notions are formulated originally to analyze how organisms, not biological systems
more broadly, function as models, their basic ideas can be applied to analyze how model systems work.
4 Another use of the term “model system” is to refer to a system that “encompasses not only the organism,
but also the techniques and experimental methodologies surrounding the organism itself” (Ankeny, 2007,
63
example, Ankeny and Leonelli, 2020; Bolker, 2009), most philosophical discussions about “living
models” focus on organisms that serve as models, such as the house mouse Mus musculus as a
model of humans. Making it explicit that a component system can serve as a model is important
because extrapolations, generalizations, and inter-model comparisons are made not always across
taxa, but also across component systems. For example, the lung, mammary gland, kidney, and
blood vessels of mice are all regarded as model systems to elucidate how branched organs are
formed during development (Varner and Nelson, 2014). In this context of research, the intended
representational scope of these model systems might include any biological systems with branched
structures. Specific findings from the mouse lung might be extrapolated to the corresponding organ
of other vertebrates, e.g., the human lung (across-taxa extrapolation); but the mouse lung might
also be compared with other branched organs of mice, e.g., retinal blood vessels (a within-species,
across-component systems comparison) or with different branched organs of other species, e.g.,
the fruit fly respiratory system (an across-taxa, across-component systems comparison). Note that
I am not arguing for replacing the idea of organisms as models with the notion of model systems
altogether. Organism-based analyses of biological modeling have their own advantages. However,
when we analyze cases that involve across-component systems extrapolations, generalizations, or
comparisons, the notion of model system serves as a better conceptual tool. The example that I
discuss in the following sections (research on collective cell migration) is one such case.
Two accounts of how a biological system works as a model are worth introducing: exemplary
models and Krogh-principle models.5 Exemplary models are those biological systems that serve as
p. 47). This is not the definition adopted in this chapter.
5 Like the notions of representational scope and representational target, the notions of exemplary models
and Krogh-principle models are typically used to refer to organisms that serve as models. But I apply their
basic ideas to model systems more broadly.
64
models of a larger group of biological systems (Bolker, 2009). The action potential discovered in
the squid giant axon turned out to be shared across different nerves and different animals. Hence,
the squid giant axon served as an exemplary model by representing a larger group of biological
systems to which it belongs.6 Krogh-principle models are those biological systems that are chosen
and studied to articulate a particular biological phenomenon (Love, 2010). They are based on the
idea that there will be a system, or a few systems, on which the phenomenon of interest “can be
most conveniently studied” (Krogh, 1929, p. 202). Convenience here is interpreted typically in
terms of features that make experimental work easier, or that provide useful insights into the target
phenomenon that other biological systems cannot provide. The squid giant axon can be seen as a
Krogh-principle model for the study of nerve conduction because it was particularly convenient for
physiological experimentation in the mid-20th century due to its size. (Green et al (2018) defend
a more sophisticated interpretation of the Krogh principle, which emphasizes its heuristic nature
and the importance of the comparative method. I discuss this in Section 4.4.) As illustrated by the
example of the squid giant axon, these two accounts are not mutually exclusive; the same biological
system might serve as both an exemplary model and Krogh-principle model simultaneously. These
accounts provide basic ideas of how model systems work, which are useful (though not sufficient)
to analyze the case that is introduced in the next section.
6 Other examples of exemplary models include fruit flies as a model of animals in genetics research,
yeasts as a model of eukaryotes in research on gene regulation, and cultured stem cell lineages as an example
of stem cells in general in differentiation research (Bolker, 2009).
65
4.3 Collective Cell Migration
Collective cell migration refers to a set of processes through which cells migrate as a group in a
cohesive manner (Mishra et al, 2019).7 While observational studies existed as early as the mid-20th
century, collective cell migration has become an active area of research in the last few decades. The
phenomenon has been studied by a range of researchers. On the one hand, collective cell migration
is involved in development of different organs. Elucidating causal interactions underlying it is an
important part of explaining biological development, and in particular, the formation of various
biological forms (i.e., morphogenesis, which is one of the major problem agendas of developmental
biology; Love, 2014). On the other hand, collective cell migration plays crucial roles in cancer
invasion, metastasis, andwound healing, and hence has been studied for clinical interests. Therefore,
research on collective cell migration as a whole has multiple related goals, including explaining
development of various biological forms and elucidating pathological and regenerative processes
in humans.
Various biological systems are used to study collective cell migration. Those systems are often
called model systems, or simply models, of the phenomenon: “The molecular and biomechanical
mechanisms underlying collective migration of developing tissues have been investigated in a
variety of models, including border cell migration, tracheal branching, blood vessel sprouting, and
the migration of the lateral line primordium, neural crest cells, or head mesendoderm” (Scarpa and
Mayor, 2016, p. 143; emphasis added). Those biological systems, each of which is identified in
terms of a specific component part of an organism in a specific taxon, are regarded as useful loci
7 I adopt a simple definition of collective cell migration here. More detailed definitions have been
proposed by several authors (e.g., Friedl and Gilmour, 2009; Rørth, 2009; Theveneau and Mayor, 2011).
66
for investigating the phenomenon.
The choice of model systems of collective cell migration has reflected various factors, includ-
ing the availability of technological, institutional, and social resources, as well as anatomical and
developmental characteristics preferable for experimental investigations (just as the choice of exper-
imental organisms does; see Dietrich et al, 2020). Many model systems of collective cell migration
are component systems of standard model organisms, such as the fruit fly, mouse, zebrafish, and
Xenopus (the African clawed frog).8 Such systems were chosen as models of collective cell migra-
tion not only because they undergo this phenomenon and are experimentally tractable, but also they
belong to standard model organisms, which were already widely studied beyond the context of cell
migration research. Detailed information about development, genetics, and genomics, experimen-
tal techniques, and infrastructures that were available in studies of standard model organisms have
often motivated researchers to use component systems of such organisms to articulate collective
cell migration.
A major approach to collective cell migration aims to articulate cellular and molecular mech-
anisms underlying it.9 Researchers ask various questions to articulate those mechanisms, for
example10:
• How is the balance between cohesion between cells and their relocation maintained?
• How is the direction of migration determined?
8 However, not all model systems are component systems of standard model organism. For example,
although the slime mold Dictyostelium discoideum is not a standard model organism, it has been studied as
an important model of collective cell migration.
9 Another approach is to formulate mathematical models of collective cell migration at different scales
(reviewed, for example, by (Buttenschön and Edelstein-Keshet, 2020). Articulating how the mechanistic
approach is related to such a formal approach requires a separate paper.
10 In other words, the problem of collective cell migration consists of the specific questions about the
details of the underlying mechanisms. For a more general discussion of how problems and questions are
organized in developmental biology, see (Love, 2014, 2022).
67
Figure 4.1: Collective cell migration of fruit fly border cells (Scarpa and Mayor, 2016, Fig. 2e–2f). Left:
Several border cells migrate together towards an egg cell (oocyte) among other cells (nurse cells). Right:
An enlarged view of the migrating border cells. One of them is serving as the leader cell, which extends
protrusions to the environment and leads migration towards the source of chemoattractant (PVF/EGF).
• How do the migrating cells interact with one another?
• How do the migrating cells interact with the microenvironment (e.g., surrounding tissues)?
• Is there a functional difference among migrating cells?
Experimental work has shown that answers to these questions vary across biological systems. To
illustrate this diversity, I introduce three mechanisms that operate in different model systems: fruit
fly border cells, zebrafish lateral line primordium, and mouse mammary gland.
4.3.1 Fruit Fly Border Cells
Fruit fly border cell migration is one of the best-studied examples of collective cell migration
(Prasad et al, 2011). Border cells are several epithelial cells that undergo collective migration in
the developing ovary of the fruit fly Drosophila.11 They migrate between other cells (called nurse
cells) towards an egg cell (or oocyte) (Fig. 4.1, left). At any given moment, there is typically
only one border cell that extends protrusions in between the surrounding cells and leads migration,
11 Epithelial cells are tightly connected with each other and constitute a sheet-like structure, while
mesenchymal cells are more loosely associated.
68
Figure 4.2: Zebrafish lateral line primordium (Scarpa and Mayor, 2016, Fig. 2c, 2d). Left: The posterior
lateral line primordium (pLLP) migrates on the sides of the zebrafish embryo from head to tail. Right: An
enlarged view of the migrating lateral line primordium. The migrating cohort consists of leader cells and
follower cells and is guided by chemoattractant (CXCL12/SDF-1), the gradient of which is produced by the
migrating cohort itself (see text).
although border cells are dynamically rearranged and which one plays this leading role can vary
(Fig. 4.1, right). The cell that is playing this leading role (leader cell) at the moment suppresses
protrusions of the other border cells that follow it (follower cells). The leader cell detects and is
guided by graded concentrations of several kinds of chemoattractants, which are secreted near the
destination. Border cells are tightly associated with each other by an adhesion molecule, which
enables them to move coherently as a cluster. The same molecule is used for dynamic interaction
between border cells and the surrounding cells, which provides traction for migration (Mishra et al,
2019).
4.3.2 Zebrafish Lateral Line Primordium
Lateral lines are sensory organs that extend along the sides of aquatic vertebrates to detect changes
in water current and pressure. In zebrafish, they are formed as a result of head-to-tail migration
of posterior lateral line primordia, each of which consists of about 100 cells, during embryonic
development (Fig. 4.2, left; “pLLP” is the abbreviation for “posterior lateral line primordium”).
69
Like fruit fly border cells, there is a distinction between leader and follower cells, but the overall
configuration of the migrating cohort is different (Fig. 4.2, right). There is a group of leader cells
that exhibit mesenchymal character, which extend protrusions and lead the cohort. Follower cells
are epithelial; they form rosette-like structures, which are deposited serially during migration and
will differentiate into mechanosensory structures (Olson and Nechiporuk, 2018). The lateral line
primordium is maintained by two types of adhesion molecules, which mediate homotypic (between
leader cells; between follower cells) as well as heterotypic connections (between leader cells and
follower cells). The lateral line primordiummigrates on a particular tissue, which secretes a protein
that serves as a chemoattractant. Unlike the case of border cell migration, there is no preexisting
gradient of the chemoattractant in the microenvironment that guides migration; the chemoattractant
is uniformly expressed by the tissue. Instead, the lateral line primordium itself produces a gradient.
Follower cells express a specific receptor, which acts as a “sink” of the chemoattractant and reduces
its concentration in the rear side of the migrating cohort, while leader cells do not express that
receptor. This results in a local gradient of the chemoattractant from the front to the rear of the
lateral line primordium. Leader cells express at a high level another receptor, by which they detect
the local gradient and lead directed migration (Mishra et al, 2019).
4.3.3 Mouse Mammary Gland
Mammary glands consist of branched epithelial tubes. Although the rudimentary structure of the
gland is formed during embryonic development, further growth and branching occur during puberty.
The tip at each growing branch forms a structure called the terminal end bud. Each terminal end
bud contains cap cells, which constitute the outer layer of the bud, and body cells, which fill the
interior of the bud (Fig. 4.3, left). Although body cells are categorized as epithelial cells, they
70
Figure 4.3: Mouse mammary gland (taken fromMishra et al, 2019). Left: Mammary gland forms a branched
structure. Collective cell migration occurs at the end of each branch, within the terminal end bud (TEB).
Right: An enlarged view of migrating cells at the end of a branch. The cells filling the interior of the bud
(body cells) compete for the front position of the bud. A secreted protein (FGF) regulates their migration.
exhibit epithelial features only incompletely (Huebner and Ewald, 2014). The migrating body cells
are confined within the terminal end bud, which is a feature distinct from border cell migration and
lateral line primordium migration. Since they are surrounded by the layer of cap cells, they cannot
extend protrusions to the outside tissue. Instead, body cells migrate over one another by using
cell-cell adhesion and compete for the front position of the terminal end bud (Fig. 4.3, right). This
leads to extension and bifurcation of the branch. Unlike fruit fly border cells and zebrafish lateral
line primordium, there is no functional distinction between leader and follower cells. A secreted
protein is known to guide and regulate body cell migration (Mishra et al, 2019).
4.3.4 Collective Cell Migration as a Phenomenon
Although these are just three examples, they show the diversity of mechanisms underlying collective
cell migration. This diversity is illustrated by a variety of answers to the questions that characterize
mechanisms of the phenomenon. For example, different answers are given to the question “How is
the direction of migration determined?” In the case of fruit fly border cell migration, the migrating
cohort is guided by gradients of secreted chemoattractant preexisting in the microenvironment.
The migration of body cells of the mouse mammary gland is also regulated by a secreted signaling
71
molecule, but a different molecule plays the role and body cell migration is heavily restricted
within the terminal end bud. In the case of zebrafish lateral line primordium migration, there is
no preexisting gradient of a signaling molecule; the migrating cohort itself produces a gradient
of the chemoattractant that guides its own migration. Similarly, questions such as “is there a
functional difference among migrating cells?” and “how do the migrating cells interact with
the microenvironment?” are answered differently. The same applies to many other systems that
undergo collective cell migration; collective cell migration is produced by different cellular and
molecular mechanisms in different biological systems.
The existence of different mechanisms has been recognized by researchers for more than a
decade (e.g., Montell, 2008; Rørth, 2009). The diversity of mechanisms also has been reflected
in explanatory and representational practices in this area. Review articles about collective cell
migration rarely present the mechanism of this phenomenon. Instead, they often discuss distinct
mechanisms that operate in several major model systems. How diagrammatic representations are
used in those articles also illustrates this point. Review articles often display multiple mechanism
diagrams together, which operate in different biological systems, to explain collective cellmigration.
Fig. 4.4 is an example; here, the authors display within one figure several diagrams representing
distinct mechanisms of collective cell migration. Different mechanisms are depicted in a way that
highlights certain common features, such as the functional distinction between leader and follower
cells, while differences in cellular configurations and signaling molecules are also indicated. This
form of presentation suggests that whereas the researchers are interested in common features among
thosemechanisms, differences between them are also noteworthy and non-negligible (see Chapter 3;
Yoshida, 2021). One might expect that the diversity of mechanisms can be understood simply in
terms of molecular details differing across mechanisms, and that all mechanisms of collective
72
Figure 4.4: Diagrams of five distinct mechanisms of collective cell migration are juxtaposed within a single
figure in a review article (Scarpa and Mayor, 2016, Fig. 2).
73
cell migration still share the same set of abstract principles. This is not the case, at least for the
current state of knowledge. Whereas researchers often appeal to abstract features to characterize
similarities among certain mechanisms, such abstract features are not universally applicable. For
instance, the leader-follower distinction, an abstract feature mentioned above, is not shared by all
mechanisms of collective cell migration. Shifting our attention from molecular details to abstract
features does not erase the diversity of mechanisms.
Even though it has been recognized that mechanisms of collective cell migration are diverse, bi-
ological systems studied in this area (such as fruit fly border cells, zebrafish lateral line primordium,
and mouse mammary gland) are commonly treated as models of collective cell migration. (See the
quote in the second paragraph of this section.) How should we understand this modeling practice?
How can a biological system be a good model of a phenomenon when the phenomenon is produced
by diverse mechanisms?
A related puzzle concerns the fact that collective cell migration is still treated as a single
phenomenon. Although distinctions are sometimes made between its subtypes (such as epithelial
collective cell migration and mesenchymal collective cell migration), collective cell migration as
a general phenomenon remains the major category in this area. “Collective cell migration is a
widely observed phenomenon during animal development, tissue repair, and cancer metastasis”
(Qin et al, 2021, p. 1267; emphasis added). This contrasts with what some new mechanists claim,
according to which, when multiple distinct mechanisms are identified for a single phenomenon,
scientists recharacterize the phenomenon into multiple phenomena according to the underlying
mechanisms. “If the goal is to provide a mechanistic explanation, the phenomena should be
chunked in such a way that they correspond to distinct underlying mechanisms. [...] For example,
in a lumping error, one might assume that several distinct phenomena are actually one, leading one
74
to seek out a single underlying mechanism when one should in fact be looking for several more
or less distinct mechanisms” (Craver and Darden, 2013, p. 61, emphasis original). This picture
does not accommodate the case of collective cell migration, where the recognition of diverse
mechanisms has not led to splitting the phenomenon. A major criticism of Craver and Darden
(2013)’s view is provided by Colaço (2020). According to Colaço, identifying distinct mechanisms
does not necessarily lead to splitting a phenomenon into multiple phenomena. Rather, such
splitting is warranted when a discovery is made that is inconsistent with the accepted description
or characterization of the phenomenon in question. I agree with Colaço that discovering distinct
mechanisms itself does not warrant splitting a phenomenon. I also agree that evidence about
features of a phenomenon plays an important role in the recharacterization of the phenomenon.
Unlike Colaço, however, I focus on another reason why retaining a phenomenon category is
justified even after distinct mechanisms are identified for it: keeping such a phenomenon category
can facilitate inquiries into and comparisons among individual mechanisms.
In the next section, I present an account of how certain biological systems serve as models of
collective cell migration despite the diversity of the underlying mechanisms. This account also
shows that there are sometimes good epistemological and methodological reasons for not splitting
a phenomenon that is known to occur through different mechanisms.
4.4 Modeling a Phenomenon with Multiple Biological Systems
In this section, I first show that the two basic accounts of model systems that we have seen in
Section 4.2 are not sufficient to analyze the case of collective cell migration. Then I provide my
account, which discusses the use of multiple model systems.
75
4.4.1 Insufficiency of the Basic Accounts
How can we characterize representational roles of model systems of collective cell migration?
Consider the accounts of exemplary models and Krogh-principle models. As I explained in
Section 4.2, exemplary models are those biological systems that are studied for the purpose of
generalization, i.e., to learn about a larger group of biological systems to which they belong,
while Krogh-principle models are biological systems that are the most convenient for elucidating
a particular phenomenon of interest. Although these accounts provide useful, basic conceptual
resources, they both are insufficient for analyzing the use of model systems in research on collective
cell migration.
The account of the exemplary model does not fully capture the situation. It is true that model
systems of collective cell migration are exemplary models since their supposed representational
scope is a larger group of biological systems (i.e., those systems that undergo collective cell migra-
tion). However, exemplary models are typically associated with the idea of wide generalizability
of research findings, which in turn is based on the assumption of broad conservation of traits
and mechanisms across taxa (Bolker, 2009). We cannot rely on this idea of wide generalizabil-
ity because what we are asking here is how, despite the diversity of underlying mechanisms, a
phenomenon can be studied by using model systems.
The famous idea of convenience emphasized by the Krogh principle no doubt plays important
roles in the choice of model systems. For example, zebrafish lateral line primordium has been
studied as a model system in part because of the ease of observation and manipulation (since
zebrafish embryos are transparent and lateral line primordia migrate close to the surface of the
skin), as well as the availability of various resources (such as materials, experimental techniques,
76
and information about zebrafish development, genetics, and genomics) (e.g., Olson andNechiporuk,
2018). However, the fact that a particular biological system is useful for research does not provide a
satisfactory answer to the question of how it serves as a model of a phenomenon, together with other
biological systems, when it is known that the phenomenon’s underlying mechanisms are diverse.
Green et al (2018) argue for a more sophisticated interpretation of the Krogh principle. Ac-
cording to them, the Krogh principle should be understood as a heuristic for studying an organism
with an “extreme” trait to obtain generalizable insights into an underlying mechanism. Importantly,
this heuristic works in combination with the comparative method. How widely and to what species
findings from a “Krogh organism” can be extrapolated is not assumed in advance; rather, it is to be
empirically investigated. Furthermore, even when a finding from the organism turns out not to be
widely generalizable, it can provide various useful insights into mechanisms through comparisons
with other species. This emphasis on the heuristic nature and the importance of the comparative
method is useful to analyze the case of collective cell migration as well (see Section 4.3). However,
representational practices in research on collective cell migration are not limited to this specific
heuristic strategy. Moreover, even though it highlights the importance of the comparative method,
the Krogh principle’s focus is on how a single biological system can be a useful research tool be-
cause of its distinct physiological or morphological feature. In contrast, my goal in the remainder of
this section is to articulate howmultiple biological systems serve as models of a diversely-produced
phenomenon. Thus, although Green et al (2018) and this chapter both focus on how studies of
biological variation produce generalizable findings concerning mechanisms, they make distinct
contributions.
In what follows, I provide an account that focuses on the use of multiple model systems.
Even if no single explanatory account can cover diverse mechanisms, generalizations concerning
77
specific aspects of mechanisms are often formulated across certain ranges of biological systems,
which makes it possible for those systems to jointly represent the phenomenon. Furthermore, there
is utility for characterization and investigation of individual mechanisms in comparing different
biological systems as models of the same phenomenon. We can appeal to these facts to understand
how multiple biological systems can represent collective cell migration.
4.4.2 Local Generalizations
By local generalization, I mean a generalization concerning a particular feature of mechanisms
that holds neither universally nor even nearly universally, but across a certain range of biological
systems that undergo the phenomenon of interest. A local generalization is local in two senses.
First, it does not apply to all or almost all examples of the phenomenon. Second, it is local because
what is generalized is not the entire mechanism description but a specific feature or aspect of it.
Recall that in Section 4.3, I listed several questions that are typically asked to characterize each
mechanism of collective cell migration. No two migration mechanisms that operate in different
model systems are so similar that the same answers are given to all of those questions. However, if
one focuses on a particular question, one can often find that multiple mechanisms are characterized
by the same answer, or similar answers, to that question (for a related discussion, see Bechtel and
Abrahamsen, 2005; Halina, 2018).
Let us consider some examples. An important question for characterizing a migration mech-
anism is what determines the direction of migration. There is some similarity concerning this
question between zebrafish lateral line primordium and Xenopus neural crest cells. The two sys-
tems both use the same type of protein as chemoattractant. Furthermore, they both self-generate
directional guidance, instead of being guided by a gradient of the chemoattractant already existing
78
in the microenvironment. As I described in Subsection 4.3.2, zebrafish lateral line primordium
produces a local gradient of the chemoattractant by reducing its concentration in the rear side of
the migrating cohort. It has been suggested that Xenopus neural crest migration also involves
self-generation of directional guidance, although the way that it is done is not exactly the same
(Theveneau et al, 2013). Therefore, these two mechanisms are similar in this specific feature. This
generalization concerning the chemoattractant and self-generation of directional guidance is local
in the two senses I specified above. It applies only to some examples of collective cell migration;
in other examples, other molecules serve as chemoattractant, migration is guided by preexisting
gradients of chemoattractant in themicroenvironment, or the direction of migration is determined in
a totally different manner. The generalization also concerns only a specific feature of the migration
mechanisms, namely, the kind of chemoattractant and self-generation of directional guidance. Not
all features of the mechanisms of zebrafish lateral line primordium migration and Xenopus neural
crest migration are similar. Despite this locality, these two migration mechanisms are sometimes
discussed together to highlight the similarity between them (Mayor and Etienne-Manneville, 2016;
Scarpa and Mayor, 2016).
Another example of a local generalization concerns a functional difference among the cells
constituting a migrating cohort. Leader cells and follower cells play distinct functions in many
migrating systems, such as fruit fly border cells and zebrafish lateral line primordium. Like
the example of self-generated directional guidance, this generalization about the leader-follower
distinction is local. Although the functional difference between leader and follower cells is observed
in several model systems, it is by no means a universal feature of collective cell migration; there are
migration mechanisms that do not exhibit this functional difference, such as mouse mammary gland
development (Section 4.3.3). This generalization is also about a specific feature of the migration
79
mechanisms and not about the entire mechanisms. But researchers often discuss this functional
distinction and compare those model systems that share it, which suggests the importance of the
generalization (e.g., Mayor and Etienne-Manneville, 2016; Norden and Lecaudey, 2019; Scarpa
and Mayor, 2016).
Some local generalizations are consequences of evolutionary conservation. Mechanisms of
collective cell migration sometimes share homologous components across taxa and/or organs,
even if the entire mechanisms are not likely to be homologous. But evolutionary conservation
is not necessary for local generalizations. Some local generalizations concern specific roles that
certain types of cells play in various migration mechanisms. The above-mentioned generalization
concerning the leader-follower distinction focuses not on a conserved molecular signaling, but on
a specific kind of functional distinction that contributes to the organized migration of a group of
cells. Local generalizations are also sometimes instantiations of highly abstract principles, such
as organizational features or “design principles” (Green, 2015; Levy and Bechtel, 2013). Note
that design principles and local generalizations are distinct categories. While design principles
are characterized and studied as highly abstract principles concerning how systems behave under
a similar set of constraints (Green, 2015), local generalizations are characterized more concretely
as interactions between cells and/or molecules. However, local generalizations can instantiate such
highly abstract principles, and hence can mediate between mechanistic and formal approaches.
Importantly, whichmechanisms are regarded as similar to the givenmechanismvaries depending
on which feature of them one focuses on. This point is illustrated by a table that (Scarpa andMayor,
2016) present (Fig. 4.5). In this table, rows correspond to several model systems, while columns
indicate different questions that characterize mechanisms of collective cell migration. Depending
on which column (i.e., which feature of migration mechanisms) one focuses on, different sets of
80
Figure 4.5: A table that characterizes several mechanisms of collective cell migration (Scarpa and Mayor,
2016, Table 1). It compares important features of migration mechanisms (the seven columns) across different
model systems (the five rows).
81
model systems are grouped together as similar systems. For example, the generalization concerning
the leader-follower distinction applies to fruit fly border cells and sprouting blood vessels of mice,
whereas mesendoderm of zebrafish and Xenopus is excluded from its scope (Fig. 4.5, the second
column from the left).12 However, if one focuses on what types of molecular interactions are used
to exert tractive force, fruit fly border cells and zebrafish mesendoderm can be grouped together,
on the one hand, and sprouting blood vessels of mouse and Xenopusmesendoderm can be grouped
together, on the other (Fig. 4.5, the fourth column from the left). The point I am making here is
this: there are different possible and useful ways to divide the diverse mechanisms into groups
of similarity (Fig. 4.6). This means that each model system can represent different subclasses
of collective cell migration depending on which specific feature of the mechanisms one focuses
on. Therefore, by studying those model systems, the community of researchers can elucidate
diverse mechanisms underlying the phenomenon. The multiple biological systems jointly represent
collective cell migration.
This consideration also highlights an important aspect of the comparativemethod inmechanistic
research. Biologists are often interested in how and why similar morphological or physiological
features result from distinct mechanisms across biological systems. (In research on collective cell
migration, developmental biologists and cell biologists often have this type of interest.) Answering
such a question requires characterizing, comparing, and mapping diverse mechanisms. My account
shows how local generalizations serve as a crucial basis for such inquiries of mechanistic diversity.
Formulating local generalizations is a strategy to identify and study regularities without ignoring
differences in mechanisms across biological systems.
12 Mesendoderm (an embryonic tissue) migrates from the surface to the inside of the embryo during early
embryogenesis of vertebrates.
82
Figure 4.6: There are multiple ways to group mechanisms according to similarity. For the same set of
mechanisms (M1, M2, M3, and M4), different mechanisms are regarded as similar depending on which
feature of them one focuses on.
We can now answer the question of why recharacterization or “splitting” of a phenomenon in
the sense formulated by some new mechanists (Craver, 2004; Craver and Darden, 2013) has not
occurred in the case of collective cell migration. For the purpose of elucidating diversemechanisms,
it is fruitful to treat collective cell migration as a single phenomenon and regard certain biological
systems as models of it, not as models of particular subclasses of it. If the phenomenon were
recharacterized or split in a single, particular way, then researchers would not benefit from local
generalizations that crosscut the recharacterized phenomena. This consideration also suggests that
if, for a given phenomenon, certain biological systems were always grouped together no matter
83
which feature of the mechanisms one focuses on, then the phenomenon would be more likely to
be split into multiple phenomena corresponding to that grouping. Another possibility is that if the
community of researchers were interested in a particular feature of the mechanisms much more
than in other features, then the phenomenon might be split according to a grouping based on that
feature, no matter what other groupings are supported concerning other features. I do not deny the
possibility that either of these might become the case in the future in research on collective cell
migration. Collective cell migration might turn out to be a tentative category that is eventually
replaced by some other categories. However, at least so far, the phenomenon has not experienced
such splitting, and the idea of joint representation formulated above helps us understand why.
Before proceeding to the next subsection, I clarify where I disagree with Craver and Dar-
den (2013). Craver and Darden claim that when distinct mechanisms are identified for a single
phenomenon, scientists should divide it into distinct phenomena, each of which corresponds to a
particular mechanism, and that indeed this is what scientists do. This is both a descriptive and
normative claim: it describes scientific practice, as well as providing a normative guide regarding
what scientists seeking mechanistic explanation should do. In contrast, I have shown that scien-
tists sometimes retain a general phenomenon (rather than split it according to distinct mechanisms
underlying it) and this is epistemologically beneficial. (I discuss more epistemological benefits in
the next subsection.) However, I am not denying that in some (possibly many) cases, scientists do
engage in phenomenon-splitting as described and prescribed by Craver and Darden. My position is
a pluralism about recharacterization of scientific phenomena. Scientists might often recharacterize
phenomena according to underlying mechanisms, but they do not do so when retaining a general
phenomenon has epistemological benefits.13
13 Craver and Darden (2013)’s account also has a metaphysical component: natural classification of
84
4.4.3 Utility for Characterizing and Investigating Individual Mechanisms
The previous subsection focused on how certain biological systems can serve as models of a
diversely produced phenomenon in a narrow sense, namely, how findings from those systems can
be projected to other systems. In this subsection, I discuss broader benefits in regarding certain
biological systems as examples of the same phenomenon. Comparing mechanisms that operate in
different biological systems asmodels of the same phenomenon can promote research by facilitating
characterization and investigation of individual mechanisms.
Let us start with a benefit for characterization. Even when the migration mechanisms being
compared are not similar with respect to the feature one is interested in, contrasting those mecha-
nisms often helps to characterize themmore precisely. This is commonly done in review articles. In
some cases, the purpose of a review article is to characterize a particular mechanism in detail, and
to do so, the authors compare that mechanism with other ones. For instance, Olson and Nechiporuk
(2018) aim at clarifying what is known about the mechanism of collective cell migration of ze-
brafish lateral line primordium. To do so, they compare this mechanism with mechanisms that
operate in several other model systems. In other cases, an article aims at a more comprehensive
review of diverse mechanisms, where comparisons are an effective way of doing it. Scarpa and
Mayor (2016)’s table is a good example, which compares different migration mechanisms in terms
of several variables (Fig. 4.5). Each mechanism is characterized more precisely by recognizing not
only similarities to, but also differences from, other mechanisms. Displaying diagrams of different
phenomena is determined by underlying mechanisms, which provides the basis for the descriptive and
prescriptive aspects of their account. In contrast, this chapter focuses on description and the epistemology of
scientific practice; the metaphysics of phenomena classification is beyond its scope. Thus, I do not discuss
whether collective cell migration is a natural category. I argue that no matter whether it is a natural category
or not, collective cell migration has been treated as a single phenomenon, and there are good epistemological
reasons for doing so.
85
migration mechanisms in one place is another example of characterization through comparisons
(Fig. 4.4).
Comparisons of different model systems also can promote investigations into individual mech-
anisms. For example, Scarpa and Mayor (2016)’s table indicates that some features of migration
mechanisms are “[n]ot yet elucidated” for certain model systems (Fig. 4.5). Features that require
further studies are effectively identified and highlighted by comparing a given mechanism with
what is known about other mechanisms. Comparisons also have heuristic value. When biologists
investigate a less-explored system, they often assume as a working hypothesis that the system em-
ploys a similar mechanism to those that operate in certain other (better-understood) model systems.
For example, molecular signaling that is known to play a crucial role in some model systems
might play the same role in the new system under study. Such a working hypothesis might be
confirmed by experimentation, which leads to the formulation of a new local generalization. Even
if it is disconfirmed, i.e., even if it turns out that the system under study does not employ a similar
mechanism, that discovery itself is an achievement because the researchers learned something new
about the system and can utilize that finding to proceed to the next step (Bechtel, 2009). It is not a
necessary condition for this heuristic that different biological systems are regarded as models of a
single phenomenon. But the heuristic is facilitated by such a situation, because if certain biological
systems are seen as models of the same phenomenon, they are more actively compared with one
another in review articles, collected volumes, and conference sessions.
In summary, there are good epistemological and methodological reasons to keep regarding
collective cell migration as a single phenomenon and certain biological systems as models of it,
not as particular subclasses of it. A crucial point is that doing so facilitates research activities.
Local generalizations about different features of mechanisms that hold across different ranges of
86
biological systemsmake it possible for multiplemodel systems to jointly represent the phenomenon.
Comparisons of different model systems also have benefits for characterization and investigation
of individual mechanisms. In these ways, multiple model systems enable efficient inquiries into
different mechanisms of collective cell migration.
4.5 Joint Representation and Integration-Based Accounts
Some philosophers of science have discussed how multiple biological model systems are combined
to fulfill specific research goals (Baetu, 2014; Fagan, 2016; Green et al, 2021). But the case of
collective cell migration exemplifies a different way multiple model systems are used together. In
this section, I introduce two accounts that discuss the use of multiple model systems (Baetu, 2014;
Fagan, 2016) and contrast my account with them. While these accounts are concerned with how
results from different model systems are integrated, such integration is not a central element of my
account; rather, the focus of my account is on efficient investigations into diverse mechanisms.
Baetu (2014) points out the “mosaic” nature of mechanistic knowledge through his detailed
discussion of immunological research. He argues that in immunology, mechanistic accounts are
often constructed by combining data acquired in studies of different model systems. “Bits of
information about the causal-mechanistic basis of a phenomenon of interest are first gathered from
data generated by several experiments, conducted in the context of distinct experimental models,
each designed to overcome a particular experimental difficulty” (Baetu, 2014, pp. 52–53).14 For
14 Baetu’s notion of “experimental model” is not exactly the same as the notion of “model system” adopted
in this chapter. In his terminology, an experimental model refers to “an experimental setup well suited for
studying a phenomenon,” where the experimental setup is characterized in terms not only of the biological
system (e.g., an organism or cell) but also of an operationalized protocol and information about various
aspects of the system, such as its source and the process of its standardization (Baetu, 2014, p. 50). However,
my interest here is in what Baetu says about the use of multiple models, and the above difference is irrelevant
87
example, a single mechanistic diagram to explain a particular immunological phenomenon is
very often produced by integrating contributions from studies conducted in different experimental
models, such as human primary cells, genetically engineered human cells, and murine models (see
also Baetu, 2016). Fagan (2016)’s focus is on the use of human embryonic stem cells (hESC)
and other kinds of stem cells studied as models in stem cell biology. A central goal of stem
cell biology is to understand early human cell development. This is a taxonomically narrow, but
mechanistically complex target, and this complexity requires researchers to rely on different kinds
of stem cells, including hESC and induced pluripotent stem cells (iPSC). Researchers integrate
pieces of information acquired from different stem cell models in order to develop mechanistic
explanations for this specific target. “This complex phenomenon [early human cell development]
is represented by an ever-expanding family of related models, each narrowly targeting a different
aspect of this complex phenomenon of interest. hESC is one of many stem cell model organisms,
interrelated in their construction and use” (Fagan, 2016, p. 128).
In discussing how multiple model systems are used together, Baetu (2014) and Fagan (2016)
both emphasize the integration of results acquired from different model systems. In Baetu’s case,
integration results in a generalized, “mosaic” mechanistic account that explains the target phe-
nomenon; in Fagan’s case, integration leads to explanations of a single, specific target system (i.e.,
early human cell development). My account of joint representation is not concerned with such inte-
gration. Its point is neither to develop a single, overarching mechanistic account by combining data
from studies of different model systems, nor to utilize information from different model systems in
order to elucidate a single, specific target system. Instead, my account characterizes how individual
mechanisms are investigated through various local generalizations and cross-system comparisons.
given this purpose.
88
I also highlighted that local generalizations allow researchers to identify and study regularities
without ignoring differences in mechanisms among biological systems. By doing so, they serve as
a crucial basis for inquiries into why and how similar morphological or physiological features result
from distinct mechanisms. Here, the diversity of mechanisms is neither problematic complexity to
be abstracted away, nor a mere means to achieve an integrated mechanistic explanation. Rather,
the diversity itself interests researchers and motivates investigations into and comparisons among
different mechanisms.
My account is not a rival of, but rather complementary to, the existing accounts. It characterizes
a different way that multiple model systems are used in combination within an area of research.
Indeed, the accounts of Baetu and Fagan seem helpful to analyze the use of multiple model
systems in some local contexts of research on collective cell migration. Baetu’s idea of mosaic
nature of mechanistic knowledge is useful for understanding how each migration mechanism has
been elucidated. For example, although the mechanism of collective cell migration in sprouting
blood vessels is often treated as one thing, it is informed by studies of different types of blood
vessels, such as mouse retinal blood vessels and zebrafish intersegmental arteries (Gerhardt et al,
2003; Siekmann and Lawson, 2007). However, such integration to construct a single, generalized
mechanistic account is not the dominant approach to the phenomenon of collective cell migration.
This point is illustrated by the common presentational practice of displaying diagrams of multiple
distinct mechanisms together (Fig. 4.4). Fagan’s account also seems effective to analyze certain
aspects of this area. Some researchers studying collective cell migration are interested primarily
in medical application. To them, collective cell migration in a particular biological system (e.g.,
human breast cancer) is the target, and knowledge of other migration mechanisms is a means to
it. Fagan’s account fits such situations, where researchers try to explain a particular target by
89
utilizing pieces of information from studies of different model systems (e.g., Stuelten et al, 2018).
However, no interest in a single, particular biological system dominates research on collective cell
migration. As I explained in Section 4.3, this area involves researchers from different disciplines
and is motivated by a range of interests, including explaining the development of various biological
forms and better understanding pathological and regenerative processes in humans. Therefore, to
understand representational relationships between multiple model systems and the phenomenon of
collective cell migration as a whole, my account is more suitable; it explains how multiple model
systems are studied as loci for investigation and jointly promote elucidation of diverse mechanisms
in order to pursue different goals in this area of research.
4.6 Conclusion
There are biological phenomena whose underlying mechanisms are so diverse that single model
systems cannot sufficiently represent them. Despite such diversity, biologists often keep regarding
certain biological systems as models of those phenomena. I proposed that, to account for this
modeling practice, we should examine how multiple model systems are used together within an
area of research. The case study from research on collective cell migration showed that despite
the mechanistic diversity, local generalizations concerning specific features of the mechanisms
hold across certain ranges of biological systems. Such local generalizations enable multiple
model systems to jointly represent the target phenomenon. Furthermore, comparisons of different
model systems facilitate research in a number of ways: they enable more precise characterization
of individual mechanisms; help to identify and highlight issues that require more studies; and
provide a basis for a heuristic to study less-explored systems. These considerations provide further
90
explanations of the use of multiple model systems in research on a phenomenon that occurs through
diverse mechanisms. Finally, I compared my account of joint representation with two existing
accounts of the use of multiple model systems and argued that they are distinct and complementary.
This comparison suggests that more philosophical inquiry is needed to understand different ways
that multiple biological model systems are combined to fulfill specific research goals.
91
Chapter 5: Generalization Reconfigured: An Inquiry
into Representational and Investigative Practices of
Scientific Generalization
5.1 Introduction
The previous chapters have discussed different aspects of generalization practice in developmental
biology: Chapter 2 characterized two basic forms of generalizations, mechanisms and principles,
which are associated with different investigative practices. Chapter 3 discussed how a specific
presentational strategy—multiple-models juxtaposition—serves to manage a trade-off between the
modeling desiderata of generality and detail. Chapter 4 articulated how local generalizations
provide a crucial basis for inquiries into diverse mechanisms that produce a phenomenon. In this
final chapter, I combine insights from the previous chapters to address the question: how and
why do scientists generalize? I do so by challenging three assumptions that have been influential
in philosophical discussion about scientific generalizations: (1) generalizations are expressed
linguistically (canonically as quantified propositions), (2) scientists generalize by formulating a
single representation with wide applicability, and (3) generalizations are valuable because they
92
enable scientific explanations.
First, philosophers have assumed that generalizations are expressed linguistically. More specifi-
cally, a common assumption has been that generalizations can and should be analyzed as quantified
propositions. This idea is illustrated by how philosophers have discussed laws of nature (universal
generalizations), where generalizations are characterized and analyzed as universally quantified
propositions, such as “all uranium spheres are less than a mile in diameter” and “all gold spheres
are less than a mile in diameter” (The former proposition has often been regarded as a law because
it is necessarily true, while the latter has not because it is merely contingently true; Carroll, 2016).
Philosophers have debated what features such a proposition must possess to qualify as describing
a law. Here, it is a background assumption that generalizations are expressed as propositions.
Although such a proposition-based approach allows us to examine the logical structure of general-
izations and the nature of laws of nature, it tells us very little about how generalizations are practiced.
Scientific generalizations rarely take the form of a universally quantified proposition. Furthermore,
other kinds of representations than linguistic representations play significant roles in generalization
practice (see below). Quantified propositions are highly idealized philosophical reconstructions of
scientific generalizations. Relying just on the proposition-based approach obscures how scientists
pursue, characterize, reason about, and use different kinds of generalizations.
The recently developing area of philosophy of scientific modeling provides a different concep-
tualization of scientific generalizations, which can complement the proposition-based approach. In
this area, generality is regarded as one of the basic desiderata of a scientific model (e.g., Gelfert,
2013; Levins, 1993; Matthewson, 2020; Matthewson andWeisberg, 2009; Odenbaugh, 2003, 2006;
Orzack, 2005; Orzack and Sober, 1993; Weisberg, 2004, 2007, 2013). Here, a generalization is
not assumed to be a law-like statement that takes the form of a quantified proposition; instead, a
93
generalization is characterized as a representation that is applicable to a certain range of systems
or phenomena. This conceptualization allows us to appreciate representational practices surround-
ing scientific generalizations; studies in this literature pay closer attention to different forms of
representations (such as mathematical models and schematic diagrams) that are used to express
generalizations.
However, even this approach maintains a second assumption: scientists generalize by formu-
lating a single representation with wide applicability. According to this view, a generalization
always takes the form of a single representation that applies to a certain range of things, whether
it is a proposition, mathematical equation, diagram, or some other forms of representation. This
assumption is rarely stated explicitly, but its influence is suggested by how philosophers discuss
generalizations. For example, one of the major focuses of the literature on trade-offs among mod-
eling desiderata is whether generality trades off against specific desiderata—such as precision and
realism—in a single mathematical model. Some authors focus on trade-off relationships derived
purely from formal properties of mathematical models, while others consider trade-offs derived in
part from pragmatic constraints (such as computational power available to scientists). But both
assume that the desideratum of generality is attributed to individual models. The same assumption
is shared by the proposition-based approach. A universally quantified proposition—a “paradigm
case” of a generalization in this framework—is regarded as a generalization because it is a statement
that applies to all members of a set.
The first half of this chapter challenges these two assumptions by examining a specific example:
mechanistic research on collective cell migration. In review articles about collective cell migra-
tion, researchers generalize by employing different types of visualizations, including: individual
diagrams that depict abstract principles or component mechanisms; juxtaposition or ordering of
94
multiple diagrams; and tables. This example illustrates that formulating a single, unified represen-
tation is not the only way to generalize. Scientists sometimes generalize by configuring multiple
representations. This is an important strategy, especially given that mechanisms of collective
cell migration exhibit some regularity as well as certain degrees of specificity and diversity. Re-
searchers’ interests in generality and mechanistic details are coordinated by this strategy because
it highlights shared features without ignoring mechanistic details that differ across systems. Fur-
thermore, different types of visualizations have different advantages and disadvantages. They are
often used in a complementary way to facilitate understanding of and reasoning about generality,
specificity, and diversity of mechanisms. These observations indicate that how researchers of col-
lective cell migration generalize can be properly understood only when we pay sufficient attention
to relevant representational practice, which includes the formulation of individual representations
and configurations of those representations. Reconstructing all scientific generalizations as propo-
sitions would make it difficult, if not impossible, to understand how scientists coordinate their
interests in generality and specific details of mechanisms.
In the latter half of this chapter, I challenge the third assumption, which concerns what roles gen-
eralizations play in science. Philosophical discussions about generalizations have been dominated
by the interest in scientific explanation. Much attention has been focused on how generalizations
enable us to explain phenomena by, e.g., providing a crucial premise for deductive reasoning, iden-
tifying a common pattern under which different phenomena are subsumed and unified, or indicating
the invariance of causal relationships (Hempel, 1966; Kitcher, 1981; Woodward, 2001). I do not
deny that generalizations enable certain forms of scientific explanations. Rather, I argue against the
implicit assumption that generalizations are valuable insofar as they generate scientific explana-
tions. A similar criticism is provided by Bogen (2005). He examines the case of neuroscience and
95
argues that generalizations play several important epistemic roles in mechanistic research, such as
to “describe facts to be explained, suggest and sharpen questions about causal mechanisms, suggest
constraints on acceptable explanations, measure or calculate crucial quantities, and support induc-
tive inferences without which mechanisms could not be successfully studied, and the results of their
study could not be applied to new instances of causal productivity” (p. 401). These roles are not
explanatory (e.g., they themselves do not serve to explain phenomena); rather, these are epistemic
roles that facilitate the search for mechanisms in various ways. Waters (1998) also discusses several
non-explanatory roles of generalizations. For example, identification of the distribution of a certain
property in the biological world can provide insights into structure, mechanism, or ecological rela-
tions of interest. He also points out that certain generalizations can serve as tools for various kinds
of investigations. For instance, in classical genetics, knowledge of the regularities concerning how
recessive genes influence adult fly phenotypes helped geneticists to determine relative positions of
genetic factors on the chromosome.
Whereas my view on roles of generalizations builds on those of Bogen (2005) and Waters
(1998), I argue that generalizations also play a kind of role that neither of these authors discusses:
generalizations facilitate cross-fertilization among studies of different target systems. I show
this by examining research on collective cell migration again. Researchers of collective cell
migration study different biological systems (developmental, pathological, or regenerative systems
in different organs of different species) as model systems and identifies generalizations that hold
across certain ranges of those biological systems. Such generalizations provide perspectives from
which different mechanisms are characterized and compared, heuristic hypotheses for studying
less-explored systems, and a basis for comparing different classes of biological systems, such as
developing tissues and invasive cancer. Importantly, there are multiple generalizations that concern
96
different aspects of migration mechanisms and apply to different ranges of biological systems. I
argue that those generalizations together provide an interface where studies of the diverse biological
systems mutually inform, which leads to new discoveries about and better characterizations of
individual mechanisms.
Before proceeding, let me clarify the nature of my challenge to these three assumptions.
Assumptions (1), (2), and (3) are features of the “analytic project” in philosophy of science, which
sought highly general accounts of how scienceworks by employing logical tools, andwhere defining
scientific explanation was one of the major goals (Barker and Kitcher, 2014, p. 22).1 This chapter
provides an alternative, more practice-based account of scientific generalizations by challenging
these assumptions. However, I do not intend to argue that the analytic project is a wrong way
to philosophically study science. The analytic project has its advantages. For example, the use
of highly idealized examples (such as “all uranium spheres are less than a mile in diameter”)
allows philosophers to formulate logically rigorous arguments, while ignoring the complexity and
idiosyncrasies of scientific research. At the same time, the analytic project neglected certain
aspects of generalization practices because of the assumptions associated with its methodological
orientation. By challenging these assumptions, this chapter proposes an approach that facilitates an
inquiry into those previously neglected aspects of generalization practices. But again, this does not
meanmy approach is the right one in any absolute sense; it also has disadvantages. For example, as a
result of examining investigative and representational practices in detail, my discussion necessarily
1 For instance, the deductive-nomological (D-N) model—the pioneering model of scientific
explanation—contends that explanation consists in deriving the phenomenon of interest from a universal
generalization and other assumptions (Hempel, 1965). It was crucial in this framework that generalizations
were expressed as universally quantified propositions, because the D-N model understood explanation as a
deductive inference. (And assumption (2) is a natural consequence of characterizing a generalization as a
proposition.)
97
involves the complexity and idiosyncrasies of a particular example. The point of my approach
is to shed light on aspects of scientific generalization that have received very little philosophical
attention.
This chapter is structured as follows. Section 5.2 challenges assumptions (1) and (2). I describe
different types of visualizations that are employed to express generalizations. The central claim
of this section is that configuring multiple representations is an important form of generalization.
Section 5.3 focuses on assumption (3) and considers the question of what roles generalizations
play in science. I show that generalizations facilitate productive interactions among studies of
different biological systems by promoting cross-system comparisons. In Section 5.4, I discuss
some implications of my account of generalization for several philosophical themes, such as
conceptions of generalization, modeling and representation, and the applicability of my approach
to generalization for other fields of science.
5.2 Generalizing with Different Types of Visualizations
This section examines how generalizations are practiced in research on collective cell migration.
Collective cell migration refers to a phenomenon in which a group of cells migrate together in
a cooperative manner. Cellular and molecular aspects of this phenomenon have been actively
studied in the last few decades. Collective cell migration is known to play crucial roles in normal
development of various organs in different species, cancer invasion and metastasis, and wound
healing. On the one hand, different cellular and molecular mechanisms for this phenomenon have
been described that operate in different biological systems. On the other hand, researchers have
articulated important commonalities shared across some of those mechanisms. In other words,
98
mechanisms of collective cell migration exhibit some generality as well as certain degrees of
specificity and diversity.
This section focuses specifically on how andwhat visual representations are employed to express
generalizations in review articles about collective cell migration. Why visual representations, and
why review articles? There has recently been increased philosophical attention to the use of
diagrams in mechanistic research (e.g., Abrahamsen and Bechtel, 2015; Bechtel and Abrahamsen,
2012; Sheredos and Bechtel, 2017; Tee, 2018). A major role of diagrams in this context is to
present spatial and temporal relations in a two-dimensional space in an easily accessible manner,
which helps scientists characterize, reason about, and explain phenomena (Sheredos et al, 2013).
Despite the increased philosophical attention to the role of visual representations in mechanistic
research, however, there has been little discussion about how visual representations play roles in
generalizations about mechanisms. This chapter takes a step toward filling this gap.
I focus on review articles because they are a major locus for generalization in contemporary
experimental biology. Original research articles often focus on one or a few biological systems and
provide new information about specific mechanisms operating in these systems. It is in review arti-
cles where researchers gather results from different laboratories, compare mechanisms articulated
in different model systems, and discuss general features shared across distinct mechanisms. Many
results obtained by studying specific biological systems are compared, summarized, and processed
into general knowledge in review articles. Furthermore, general knowledge produced in this way
in turn suggests where further research can or should proceed. Therefore, review articles are an
ideal place to examine when we want to understand how generalizations are practiced as part of an
ongoing process of knowledge production.2
2 Textbooks play a role similar to review articles, but there are some differences between them. While
99
5.2.1 Generalizing with Single Representations
One type of visualization depicts a particular feature shared across a range of mechanisms of
collective cell migration. Such a representation expresses a generalization by ignoring features that
differ across systems. This type of visualization can be divided into two subtypes, depending on
what details it ignores. What I call a principle diagram ignores specificity of biological entities (such
as names of particular biomolecules and types of cells); it is an abstract representation that focuses
on a relation or interaction that can be instantiated by different entities in different mechanisms.
In contrast, what I call a component mechanism diagram depicts a specific component mechanism
(such as a mechanism for directional guidance) that is part of an entire mechanism of collective cell
migration. Mechanisms of collective cell migration consist of multiple component mechanisms. A
component mechanism diagram focuses on one of them while ignoring others.
Fig. 5.1 is an example of a principle diagram (Mayor and Etienne-Manneville, 2016). It
represents what is called the leader-follower distinction, a feature shared across many examples
of collective cell migration. It means that there is a functional difference among migrating cells.
Those cells at the leading edge or migrating front (the right-hand side in the diagram) actively
extend protrusions and sense the extracellular environment surrounding them. Their migration is
stimulated by external guidance cues (such as signaling molecules diffusing from certain cells or
tissues). In this way, these cells “lead” migration. Other cells in the migrating cohort (the cells at
the left-hand side in the diagram) do not play such a role and “follow” the leader cells (Mayor and
textbooks usually provide systems of knowledge that are more widely accepted by the community, review
articles often include recent and less-established findings and more directly reflect authors’ specific perspec-
tives. Furthermore, whereas textbooks are usually aimed at educating novices, many review articles are
targeted towards researchers working in a specialized area.
100
Figure 5.1: A principle diagram. It abstractly represents the leader-follower distinction, a widely shared
feature of collective cell migration (Mayor and Etienne-Manneville, 2016, Fig. 1).
Etienne-Manneville, 2016).3 Fig. 5.1 ignores many idiosyncrasies that exist in actual mechanisms:
how leader and follower cells are spatially arranged; what types of cells serve the leader and follower
roles; molecular details of the guidance of leader cells; molecular details of interactions between
leader cells and the extracellular environment; and many others. By representing the relation and
interaction abstractly, Fig. 5.1 expresses a generalization.
Fig. 5.2 is an example of a component mechanism diagram (Zegers and Friedl, 2014). This
diagram depicts how a group of signaling proteins (Cdc42, Rac, and RhoA in the diagram) function
in certain types of collective cellmigration bymediating external signals, cell polarity, and dynamics
3 Theveneau and Linker (2017) argue that “leader cells” is not accurate terminology and suggest different
terms, such as “front cells,” “rear cells,” and “steering cells.” But I use “leader cells” throughout this chapter
because it is still commonly used in research on collective cell migration.
101
Figure 5.2: A component mechanism diagram. It depicts how three proteins (Cdc42, Rac, and Rho) interact
in cell polarization and cytoskeletal dynamics, which are common in some mechanisms of collective cell
migration (Zegers and Friedl, 2014, Fig. 1).
of cytoskeleton.4 Cdc42, Rac, and RhoA are concentrated in different regions within a cell, and
their concentrations control certain subcellular activities. While Rac and Cdc42 regulate protrusion
formation, Rho regulates cell contraction. In certain types of collective cell migration, external
signals (such as signaling molecules and adhesion to the substrate) activate Cdc42 and Rac at the
free edge of leader cells, which promotes protrusion formation (1). Cytoskeletons of migrating cells
are arranged such that they transmit traction forces across cells. This “supracellular” cytoskeletal
organization is mediated by cell-cell junctions (2, 3). Follower cells form protrusions underneath
neighboring cells in the direction of migration, which is regulated by Rac activity (4). Fig. 5.2
expresses a generalization, but unlike Fig. 5.1, it includes many molecular details. Its generality
is based on the specific component mechanism that is evolutionarily conserved and shared across
many examples of collective cell migration.5
4 The cytoskeleton is a network of special kinds of molecules within the cell that influences the shape
and mechanical properties of the cell.
5 The distinction between principle diagrams and component mechanism diagrams is not strict. Rather,
102
5.2.2 Generalizing through Configuring Multiple Representations
Using a single diagram is not the only way to visually express generalizations about mechanisms
of collective cell migration. Another widely adopted style of visualization is to juxtapose multiple
diagrams, where those diagrams represent mechanisms operating in different biological systems.
Examples include: Friedl and Gilmour (2009), Khalil and Friedl (2010), Lu and Lu (2021), Mayor
and Etienne-Manneville (2016), Mishra et al (2019), Olson and Nechiporuk (2018), Saraiva and
Barriga (2021), Scarpa and Mayor (2016). Such a visualization serves as a generalization by
highlighting common or similar features without eliminating details peculiar to different biological
systems.6 By doing so, juxtaposed mechanism diagrams exhibit a different set of representational
desiderata than principle diagrams and component mechanism diagrams.
Fig. 5.3 illustrates how mechanism diagrams are often juxtaposed. The three diagrams depict
different examples of collective cell migration: (a) epidermal regeneration (a healing skin wound),
(b) border cell migration (in fruit fly ovaries), (c) neo-angiogenesis (blood vessel sprouting in
vertebrates). The diagrams are drawn and arranged in a way that guides expert readers to recognize
certain shared features. For example, the reader would notice that collective cell migration in
these examples all exhibit the leader-follower distinction. In each diagram, leader cells (labeled as
“tip cell” in diagrams b and c) extend protrusions to the microenvironment and express receptor
they should be regarded as two prototypes of generalizations. For instance, there are diagrams that represent
abstract principles with a few molecular details, or diagrams that represent a component mechanism but
ignore some of its molecular details.
6 Although the practice of juxtaposition has rarely received a substantial philosophical analysis, some
historical and philosophical studies of diagrammatic practice briefly point out that juxtaposing multiple
diagrams can highlight common features across different target systems (e.g., Abrahamsen et al, 2017;
Steinert and MacCord, 2018). Outside history and philosophy of science, the data visualization researcher
Edward Tufte (1990) provides a classic discussion of what he calls “small multiples,” which is a method of
presenting multiple illustrations of the same format together to facilitate comparative reasoning.
103
Figure 5.3: Juxtaposed mechanism diagrams. The three diagrams represent different mechanisms of collec-
tive cell migration operating in different (types of) biological systems: a healing wound of skin (a), border
cells in a fruit fly ovary (b), and a sprouting blood vessel of vertebrates (c) (Friedl and Gilmour, 2009,
Fig. 2). The three diagrams together express generalizations by highlighting shared features, such as the
leader-follower distinction, while containing information about differences among the mechanisms.
104
proteins (represented as Y-shaped icons), through which they detect guidance molecules and lead
the migration. At the same time, these diagrams provide information about differences between
the represented mechanisms, such as the spatial arrangements of the migrating cells (flat sheet,
detached cluster, or extending tube), substrates on which cells migrate (extracellular matrix or other
cells), and types of guidance molecules (EGF and ROS; EGF and PVF1; or VEGF and FGF).
Juxtaposition is a way to generalize about shared features without abstracting away causal details
that differ across systems (Chapter 3; Yoshida, 2021).
Juxtaposed mechanism diagrams have a distinct set of advantages. An important advantage
is that they convey information about distributions of shared features. Biologists are not only
interested in a causal regularity that is shared across some biological systems; they are also
interested in what biological systems share that causal regularity (Waters, 1998). In research on
collective cell migration, the distribution of a general feature is often characterized in terms of
specific biological systems that share it. Biological systems, in turn, are characterized in terms of a
specific taxon and tissue the system belongs to (i.e., which tissue of which organism is it?) as well
as the condition of the system (i.e., is it a developmental, pathological, or regenerative system?).
Principle diagrams and component mechanism diagrams do not convey information about
distributions. They are often detached from any particular mechanisms of collective cell migration
(Fig. 5.1 and 5.2). In other words, they just represent features shared across some examples of
collective cell migration without indicatingwhich examples. In contrast, juxtaposed diagrams show
how certain features are distributed across different biological systems that undergo collective cell
migration. For example, by looking at Fig. 5.3, an experienced reader can know that the leader-
follower distinction is shared at least across epidermal regeneration, border cell migration (in fruit
105
fly ovaries), and neo-angiogenesis (in vertebrate blood vessels).7
Another advantage that juxtaposed mechanism diagrams exhibit is that they present different
features and components of mechanisms as integrated wholes. Principle diagrams do not convey
this type of information since each of them focuses on an abstract feature and ignores everything else
(Fig. 5.1). (Some component mechanism diagrams convey this type of information.) In contrast,
in juxtaposed mechanism diagrams, shared features are often represented as being embedded in
specific mechanisms. For example, Fig. 5.3 shows how the leader-follower distinction is related
to or causally connected with other features or components (e.g., secretion and sensing of specific
types of guidance molecules; spatial configuration of leader and follower cells; and molecular
interactions between leader and follower cells) in each of the three mechanisms. The general
feature is presented not by itself, but as a part of specific integrated mechanisms.
Juxtaposition is also effective for indicating a range of processes through which an abstract
principle is instantiated. Fig. 5.4 illustrates self-generation of a chemoattractant gradient (Mayor
and Etienne-Manneville, 2016). According to this principle, migrating cells are not guided by a
preexisting gradient of a signaling molecule. Instead, the migrating cell cluster itself generates a
gradient, which then guides directional migration of the cluster. This feature has been studied in at
least three model systems: zebrafish lateral line primordium, melanoma cells, and frog neural crest
cells. But how the principle is instantiated differs between the model systems. In zebrafish lateral
line primordium, cells at the rear-side of themigrating cohort express a “scavenger” receptor protein
7 This is not a complete distribution of the leader-follower distinction; there are other systems that are
known to share this feature. In general, distributions can never be shown thoroughly. There are so many
biological systems that undergo collective cell migration, and many of them have never been studied in detail.
Even if one focuses on several well-studied model systems, including all of them in one figure is not always
appropriate due to limited space and for the sake of readability. Nevertheless, even showing incomplete
distributions is beneficial for the reader to get a rough idea of what biological systems share the feature of
interest.
106
Figure 5.4: Juxtaposed diagrams presenting an abstract principle. The three diagrams illustrate different
ways in which a gradient of a signaling molecule is generated by an activity of the migrating cell cluster
itself (Mayor and Etienne-Manneville, 2016, Fig. 5). a: Zebrafish lateral line primordium. The “scavenger”
receptor (red) binds the signaling molecule (grey) and reduces its concentration around the rear side of the
migrating cluster. This produces a gradient of the signaling molecule. This gradient guides migration of the
cluster (white), which is mediated by another receptor of the signaling molecule (green). b: Melanoma cells.
Melanoma cells (orange) break down the signaling molecule (grey), which reduces its concentration around
the melanoma cell cluster. The gradient produced in this way drives melanoma cells to leave the cluster and
spread into surrounding tissues. c: Frog neural crest cells. Frog neural crest cells (white) are attracted by a
signaling molecule (grey) secreted by placodes (pink). When neural crest cells reach and contact the placode
cells, the latter migrate away.
that binds the signaling molecule that originally exists uniformly in the extracellular environment
(Fig. 5.4a). Because of this receptor, the rear-side of the migrating cluster functions as a “sink”
of the signaling molecule. As a result, a gradient of the signaling molecule is generated and the
migration of the cluster is guided by this gradient. In the case of melanoma cells, the signaling
molecule also originally exists uniformly in the extracellular environment (Fig. 5.4b). Melanoma
cells break down this signalingmolecule, which reduces its concentration around themelanoma cell
cluster. The gradient produced in this way drives melanoma cells to leave the cluster and spread
into surrounding tissues. Finally, frog neural crest cells are attracted by a signaling molecule
secreted by a group of cells called placodes (Fig. 5.4c). When neural crest cells reach and contact
107
Figure 5.5: A spectrum presentation. It orders different cell behaviors according to different degrees of
coordination, cooperation, collectiveness, and supracellularity (Shellard and Mayor, 2019, Fig. 4).
the placode cells, the latter migrate away. This “chase and run” process is repeated. In the case
of Fig. 5.4, there are interesting idiosyncrasies in how the gradients of the signaling molecules are
produced by migrating cell clusters. By displaying the diagrams of the three processes together,
Fig. 5.4 effectively presents the abstract feature shared by them (i.e., that the migrating cell group
itself generates a gradient of the signaling molecule), while illustrating the variability in how this
principle is instantiated.
Another version of juxtaposed diagrams is what I call a spectrum presentation. In this type
of visualization, diagrams are not simply juxtaposed, but ordered. In research on collective cell
migration, spectrum presentations are usually adopted for a very specific purpose: to characterize
different degrees of collectiveness (e.g., Campbell and Casanova, 2016; Ferrari and Giampietro,
2019; Friedl, 2004; Friedl et al, 2012; Gray et al, 2010; Mayor and Carmona-Fontaine, 2010;
Shellard and Mayor, 2019; Theveneau and Mayor, 2011). In Fig. 5.5, diagrams of different
biological systems are ordered according to a set of features that define collective behaviors
of cells: coordination, cooperation, collectiveness, and supracellularity (Shellard and Mayor,
108
2019). Migration of a group of cells is coordinated when individual cells are migrating not in
random directions, but in parallel directions. Migration of a group of cells is cooperative when
there are interactions between migrating cells (such as promotion or inhibition of migration).
When migration of a group of cells is both coordinated and cooperative, it is collective. Finally,
supracellularity refers to a situationwhere a group of cells behave as if a single cell. Supracellularity
consists of several different features, such as polarity, cytoskeletal organization, force transmission,
and flows of cells within a migrating cluster (Shellard and Mayor, 2019).
Like Figs. 5.3 and 5.4, Fig. 5.5 provides a generalization by highlighting shared features. It
clarifies what examples of collective cell migration (such as fruit fly Drosophila’s follicle cells,
Drosophila border cells, epithelial wound healing, and the clawed frog Xenopus’s neural crest;
Fig. 5.5C–F) have in common: coordination, cooperation, and (hence) collectiveness. At the
same time, this figure shows that examples of collective cell migration exhibit different degrees of
supracellularity. Importantly, Fig. 5.5 includes diagrams of individually migrating cells (Fig. 5.5A,
B) and a highly integrated tissue whose morphogenetic movements do not depend on migration
of individual cells (Fig. 5.5G). The former are examples of individual (and hence not collective)
cell migration, while the latter is on the fringe of the extension of the category of collective
cell migration. By including these systems, Fig. 5.5 characterizes typical examples of collective
cell migration (Fig. 5.5C–F) in contrast to individual cell migration and highly integrated tissue
behaviors. Spectrum presentations highlight general characteristics that distinguish collective cell
migration from other cell behaviors, while showing variation in collectiveness and spracellularity
within the category of collective cell migration.
Importantly, the application of the strategy of juxtaposition (and ordering) is not limited within
the context of diagrammatic representation. Textual representations also can be juxtaposed to
109
generalize about mechanisms, although it is much less common compared to juxtaposed diagrams
in research on collective cell migration. Fig. 5.6 presents an example. In this table, rows correspond
to specific model systems of collective cell migration, while columns correspond to variables that
characterize its mechanisms (Scarpa and Mayor, 2016). This table indicates features shared across
different examples of collective cell migration while conveying information about details that differ
across them. In this sense, the table works similarly to juxtaposed diagrams (such as Figs. 5.3,
5.4).8 There are important differences between the two types of representations, of course. For
example, arguably the most important benefit of representing mechanisms diagrammatically—
visual depiction of spatial and temporal relations among mechanism components—is not available
in tables. A flip side of this difference in the formats is that in tables, features of target mechanisms
are displayed discretely with explicit labels (“chemoattractant,” “leader/follower,” etc.). Because
of this discrete displaying, the reader can even more easily and clearly recognize which aspects of
the target mechanisms are significant and how important features are distributed across different
mechanisms. Despite these differences, one crucial advantage of juxtaposition as a generalization
strategy applies to both: generalizations can be formulated by configuring representations (either
diagrams or texts) of multiple target systems, which highlights features shared across them without
eliminating interesting differences.
5.2.3 Complementary Relations between Different Types of Visualizations
I have introduced several types of visualizations that are used to express generalizations about
mechanisms of collective cell migration. How are those different types of visualizations related to
8 Tables are a kind of visual representation because they function by exploiting their visual nature: two
dimensional, ordered display of textual representations (Perini, 2005).
110
Figure 5.6: A table that summarizes different mechanisms of collective cell migration (Scarpa and Mayor,
2016, Table 1). The rows correspond to different (types of) mechanisms, while the columns correspond to
different variables that characterize these mechanisms.
111
one another? One possibility is that juxtaposition of multiple representations is used only at early
stages of research, when researchers have not yet developed a single, unified representation that
captures shared features (e.g., a principle diagram or component mechanism diagram). According
to this view, juxtaposition is a transient generalization practice that is replaced by more unified
representations when research progresses. I doubt that this is the case. In this subsection, I argue
that the different types of visualizations are in a complementary relationship. They have different
advantages in expressing generalizations, and hence work in a complementary way to address the
complexity, diversity, specificity, and regularity of mechanisms of collective cell migration.
I have already suggested that the different types of visualizations have different advantages.
For example, principle diagrams are simple and relatively easy to understand, even to novices.
Component mechanism diagrams are often complex and harder to grasp, but they provide detailed
mechanistic depiction of specific parts or components of mechanisms. Unlike these two types
of diagrams, juxtaposed mechanism diagrams convey information about how certain features are
distributed across different mechanisms. They also indicate how different features and components
are organized together to constitute individual, integrated mechanisms. Spectrum presentations
characterize collective cell migration in contrast to and in relation with other cellular behaviors,
and so on.
The idea that multiple representational approaches work in a complementary way has been ac-
tively discussed by philosophers and scientists, especially in the last two decades (e.g., Green, 2013;
Levins, 1966; Matthewson andWeisberg, 2009; Morrison, 2011; Weisberg, 2007, 2013). Weisberg
(2007, 2013) observes that scientists often formulate multiple models for a single phenomenon,
especially when the target phenomenon is highly complex (such as predator-prey interactions
in ecology and Earth’s global circulation in climate science). Because of the complexity, it is
112
impossible to construct a single tractable model that exhibits or maximizes all desiderata of mod-
eling. Instead, researchers construct multiple models based on different idealization assumptions
that exhibit or maximize different modeling desiderata. Weisberg names this modeling strategy
multiple-models idealization (MMI).
The basic idea of MMI applies to the example of research on collective cell migration. Mech-
anisms of collective cell migration are individually complex. Furthermore, there is diversity in
cellular and molecular details of the mechanisms across different biological systems. Because
of the complexity and diversity, it would be impossible to formulate a single visualization that
effectively represents those mechanisms. Instead, researchers formulate multiple, different types of
visualizations that provide different information about the target mechanisms or provide the same
information in different ways.9
To illustrate the complementary relationship, let us consider Scarpa and Mayor (2016)’s article
as an example. This article reviews findings about mechanisms of collective cell migration in
developmental systems. It has one table and five figures, including juxtaposedmechanism diagrams,
juxtaposed principle diagrams, and component mechanism diagrams. This article first presents a
table and juxtaposed mechanism diagrams that summarize different mechanisms of collective cell
migration in ways that highlight important shared features. (Although the table and the juxtaposed
mechanism diagrams have much in common in terms of their content, they aid different reasoning
because of the difference in representational formats.) Then it proceeds to more focused visual
representations (such as juxtaposed principle diagrams and component mechanism diagrams) to
9 There is an important difference between Weisberg’s view and mine. In Weisberg’s framework, a
modeling desideratum is attributed to a single model or a set of models. In contrast, I assume that certain
desiderata, such as generality, are attributed not always to a single model or a set of models; they can be
attributed to a configuration of models in a physical space (Chapter 3; Yoshida, 2021). I discuss this point
more in Section 5.4.
113
discuss some of those shared features in more detail, such as cell-substrate and cell-cell interactions,
generation of gradients of signaling molecules, and interactions between leader and follower cells.
In these visualizations, the same features of mechanisms of collective cell migration (such as cell-
cell and cell-matrix interactions) are presented repeatedly, but with different degrees of detail. The
reader is supposed to connect and relate these visualizations (and sometimes move back and forth
among them) so that they understand how individual mechanisms are structured, what features
they share, and what interesting differences there are. The different types of visualizations help
understanding of and reasoning about those mechanisms.
While the different types of visualizations introduced in this section all express generalizations,
they have different advantages and exhibit different representational desiderata. Combining them
is an effective way to represent complex target systems that exhibit some regularity and certain
degrees of specificity and diversity. Mechanisms of collective cell migration are a target of this
kind.
5.2.4 Responding to Possible Objections
This section has challenged two assumptions concerning scientific generalizations: (1) generaliza-
tions are expressed linguistically and (2) scientists generalize by formulating a single representation
with wide applicability. The example of research on collective cell migration shows that different
types of visualizations are employed to express generalizations. In particular, I have analyzed
several visual practices that configure multiple representations, such as juxtaposed mechanism
diagrams, juxtaposed principle diagrams, spectrum presentation, and tables. These are effective
generalization strategies that highlight shared features while maintaining information about mecha-
nistic details that differ across systems. We as philosophers would fail to appreciate these strategies
114
if we ignored how mechanisms are represented (e.g., as diagrams) as well as configured (e.g.,
juxtaposed within a figure of a review article).
Before proceeding to the next section, let me respond to potential objections. One might object
that visual representations cannot express generalizations by themselves. Visual representations
must always come with some text, and it is the latter that is crucial for generalizations; images
are merely a tool that supplements text. In response to such an objection, I would first emphasize
that I do not deny the importance of text. Visual representations in review articles are usually
accompanied with textual explanations in captions and main text. Such text no doubt plays an
important role. However, visual representations also play a unique role, which text cannot replace:
presenting spatial and temporal features in a two-dimensional space in an easily accessible manner.
This is particularly important in mechanistic research because understanding a mechanism requires
grasping spatial and temporal relations among multiple entities (e.g., Abrahamsen and Bechtel,
2015; Abrahamsen et al, 2017; Bechtel and Abrahamsen, 2005). Bechtel and Abrahamsen provide
a useful analysis:
[A] diagram has clear advantages over linguistic description. The most obvious
advantage—that all parts and operations are available for inspection simultaneously—
probably is the weakest one. Due to processing limitations, people can only take in one
or a few parts of the diagram at a time. Nonetheless, more so than when reading text,
they have the freedom to move around it in any number of ways; and as the diagram
becomes more familiar, more of it can be taken in at one time. A stronger advantage
is that diagrams offer relatively direct, iconic resources for representation that can be
invaluable. For example, it is immediately apparent in the heart diagram that blood is
being pumped simultaneously from the two atrial chambers to the two ventricles and
that these two parallel operations are in a sequential relationship to two other parallel
operations (pumping from the two ventricular chambers). The value of consulting a
diagram in this way is even more apparent in mechanisms with feedback loops, through
which an operation that is conceptually downstream (closer to producing what is taken
to be the product of the mechanism) has effects that alter the execution of operations
earlier in the stream at subsequent time steps. (Bechtel and Abrahamsen, 2005, p. 428)
115
Thus, visual representations have special advantages that facilitate understanding of and reason-
ing about mechanisms. Indeed, the visual format is often preferred for expressing mechanistic
explanations to text (Bechtel and Abrahamsen, 2005).
When scientists generalize about mechanisms, another layer of complexity is often added to
this task of representing mechanisms. One must understand not only how individual mechanisms
are constituted and operate, but also what features multiple mechanisms share as well as how and
where they differ. Juxtaposition and ordering display multiple distinct representations together,
where the reader can actively compare them by moving around different diagrams. What is crucial
is that this kind of visualization allows the reader to switch among different focuses, such as
those on individual mechanisms, shared features, and significant differences (e.g., Figs. 5.3, 5.4,
5.5, and 5.6). In this way, it bridges between simplified representations of shared features (such as
principle diagrams and component mechanism diagrams) and detailed representations of individual
mechanisms. This promotes more integrated understanding of generality, diversity, and specificity
of target mechanisms. Again, I do not deny that text plays an important role in this process. My
point is that visual representations also play significant and unique roles in generalizations, and
hence analyzing them gives us useful insights into the question of how scientists manage regularity,
specificity, and diversity in the world.
Another potential objection is that although configurations might be important for scientists’
practical purposes, philosophers do not have to pay attention to them. What matters for philosophy
is the content—the common feature that is highlighted through such a configuration. Hence,
philosophical analysis can abstract away visual practice and focus on the content. My response to
this objection is based on the same logic asmy response to the first objection. I have emphasized that
juxtaposition and ordering are often used to express generalizations without eliminating features
116
that differ across mechanisms. Therefore, by analyzing methods of configuration, we can acquire
an insight into how scientists coordinate their interests in generality and details specific to different
systems. Configurations are not mere practicalities irrelevant to philosophical research; rather, they
are methods that are worthy of epistemological analysis (Chapter 3; Yoshida, 2021).
5.3 Generalizations Facilitate Cross-Fertilization
The discussion in the previous section focused on how scientists generalize by employing visual
representations. This section discusses another closely related question: what roles do generaliza-
tions, expressed visually or otherwise, play? My answer to this question is that generalizations serve
to facilitate cross-fertilization among studies of different target systems. By cross-fertilization, I
mean mutual contributions of insights that promote better understanding of and further inquiries
into different target systems. Generalizations facilitate cross-fertilization in a number of ways.
In this section, I discuss the following, partially overlapping functions, by examining research on
collective cell migration again. First, each generalization provides a specific perspective from
which scientists can characterize and compare different target systems. Individual target systems
are characterized and understood more precisely by comparing or contrasting them with other
systems from such a perspective. Second, generalizations established through studying certain
systems guide inquiries into new or less-understood target systems. Here, generalizations provide
default assumptions about how the phenomenon of interest is produced, which serve as heuristic
hypotheses that help new investigations. Finally, generalizations sometimes promote large-scale
comparisons between different classes of target systems, such as developing tissues and invasive
cancer. In such a case, a higher-order generalization concerning similar patterns of regularity
117
and variability provides a basis of comparisons and “mutual informing” across studies of different
classes of systems. These functions are not limited to generalizations expressed in a specific repre-
sentational format. However, in research on collective cell migration, visual representations often
contribute to generalizations playing these functions (as shown below).
5.3.1 Promoting Better Characterizations of Individual Mechanisms
Let us consider some examples of how generalizations facilitate cross-fertilization among studies of
different biological systems. For instance, the generalization about the leader-follower distinction,
which is actively discussed in review articles, invites the reader to view biological systems that
undergo collective cell migration in a specific way. Juxtaposed mechanism diagrams, such as those
in Fig. 5.3, are illustrative. There, different mechanisms are depicted in a way that highlights
common features (such as the leader-follower distinction). Generalizations help us understand
the multiple, complex and diverse mechanisms in an organized manner. This promotes further
comparisons between those biological systems to further articulate aspects of the leader-follower
distinction. Researchers might ask and inquire whether and to what extent molecular details
underlying the leader-follower distinction are similar across those system: “[d]espite [the leader
cells’] crucial role in controlling collective migration, and therefore their involvement in tumour
spreading, the mechanisms leading to the emergence of leader cells and the molecular specificities
of these cells remain unclear” (Mayor and Etienne-Manneville, 2016, p. 106). Or researchers might
examine why a specific generalization has the distribution that it has. For example, vertebrate blood
vessel sprouting and fruit fly tracheal development are both examples of collective cell migration
with the leader-follower distinction, and they are known to be particularly similar; they share a
specific type of molecular interaction between leader and follower cells. Because of this further
118
similarity, the mechanisms of collective cell migration in these systems are often compared by
juxtaposing mechanism diagrams (Chapter 3; Yoshida, 2021). The recognition of this resemblance
has led Muñoz-Chápuli (2011) to an evolutionary hypothesis that the two mechanisms have evolved
by adopting the same, conserved component mechanism for sensing and responding to hypoxia
(i.e., lack of sufficient oxygen). In such a way, a generalization sometimes describes a shared feature
that requires an explanation, which becomes a target of new inquiries.
5.3.2 Guiding Investigations into Less-Understood Systems
Generalizations also guide investigations into less-understood biological systems. Once a gener-
alization is established (that is, once it is confirmed that a feature is shared across a certain range
of target systems), it starts providing a default assumption about how mechanisms of collective
cell migration operate. Let us consider the example of the leader-follower distinction again. The
leader-follower distinction has been observed and studied in many model systems and highlighted
as one of paradigmatic features of collective cell migration since the 2000s (e.g., Friedl, 2004).
Because of this recognition, when researchers study a new or less-explored example of collective
cell migration, they expect that that system exhibits the leader-follower distinction as well. This
can lead to confirming the hypothesis and expanding the scope of the generalization. Or it can lead
to a discovery of an exception for the generalization.10
For example, Ewald et al (2008) examined three-dimensional culture of mouse mammary gland
and reported that collective cell migration in this system does not exhibit the active extension of
10 My argument here is similar to Bechtel (2009)’s. He argues that biologists’ assumptions about shared
or similar mechanisms serve as a heuristic for new discoveries. However, although Bechtel focuses on
evolutionary conservation as the basis of such assumptions, I emphasize that biologists’ assumptions about
shared or similar mechanisms do not have to be based on the idea of evolutionary conservation.
119
Figure 5.7: A figure illustrating that collective cell migration in mouse mammary gland does not involve
the leader-follower distinction (Huebner and Ewald, 2014, Fig. 3). It contrasts photomicrographs of mouse
mammary gland (C, C’, D, D’) with a diagram representing a common image of the leader-follower distinction
(B).
protrusions that is characteristic of leader cells in many other systems. The leader-follower distinc-
tion provided a typical image of collective cell migration against which the authors investigated
and characterized the mechanism operating in mouse mammary gland.
Recent time-lapse imaging studies have establishedmodels for the collectivemovement
of groups of cells, including neuronal precursors in the zebrafish lateral line, epithelial
cells duringDrosophila dorsal closure, and border cellmigration inDrosophila. In each
of these examples, cells at the front of the migrating group extended cellular extensions
or protrusions in the direction of movement. By contrast, cells at the front of elongating
mammary ducts did not have leading cellular extensions or actin-rich protrusions. As
protrusive activity can function to guide cells, how elongating mammary ducts move
directionally remains an open question. (Ewald et al, 2008, p. 577)
As the last sentence suggests, this discovery opened up a new research question of how a group of
cells lacking the leader-follower distinction moves directionally. In review articles, this exceptional
feature is sometimes highlighted by contrasting an image of collective cell migration in mouse
mammary gland with an abstract diagram that depicts the leader-follower distinction (Huebner and
Ewald, 2014; Uechi and Kuranaga, 2017) (Fig. 5.7). In this example, the generalization about the
120
leader-follower distinction, which was established through studies of other model systems that share
the feature, provided an image of what a mechanism of collective cell migration typically looks like.
This image has guided investigation into and characterization of collective cell migration in mouse
mammary gland by providing a heuristic hypothesis and mediating the comparison (or contrast)
between the mechanism operating in this system and those operating in other model systems.
5.3.3 Promoting Comparisons between Different Kinds of Systems
Generalizations also have promoted comparisons between two classes of biological systems that
undergo collective cell migration: developmental and pathological systems. Researchers of col-
lective cell migration have been interested in the similarity between collective cell migration in
development and cancer invasion (e.g., Friedl, 2004). This interest originated in part from the ob-
servation that these two classes of systems exhibit similar morphological variations. Fig. 5.8 shows
two spectrum presentations from two review articles. One orders different forms of developmental
cell migration according to different degrees of collectivity (Fig. 5.8, left; Friedl, 2004). The other
adopts a very similar format to order different forms of cancer cell invasion (Friedl et al, 2012,
Fig. 5.8, right;). Certain characteristic forms of cell behaviors are observed in both development
and cancer invasion, such as chain migration, detached clusters, sheets or strands, and hollow tubes.
This morphological similarity is often explicitly discussed:
Similar to morphogenesis, the phenotypic and junctional organization of moving can-
cer cell groups varies greatly (“collective plasticity”). In experimental live-cell models,
all types of collective movements can be adopted by tumor cells including (1) cohesive
sheets or strands, typically detected in epithelial cancers; (2) isolated clusters de-
tached from the primary/metastatic lesion such as epithelial tumors and melanoma; (3)
neuronal-like networks of connected cells, detected in neuroectodermal tumors, such
as glioblastoma; or (4) as “jammed” collective cohorts induced by spatially narrow
tissue boundaries (confinement) of otherwise transiently/loosely connected (single)
121
Figure 5.8: Spectrum presentations of developmental and cancer systems. Left: A spectrum presentation
that orders developmental cell behaviors according to molecular features associated with different degrees
of collectiveness (Friedl, 2004, Fig. 1). Right: A spectrum presentation that orders invasive behaviors of
cancer cells according to some characteristics associated with degrees of collectiveness (Friedl et al, 2012,
Fig. 2a).
122
cells in experimental melanoma and sarcoma models (Friedl and Mayor, 2017, p. 11;
emphasis added)
A higher-order generalization is going on here. The two spectra capture collectivity and variability
of developmental cell behaviors and cancer cell behaviors, respectively. Then, researchers general-
ize about the two spectra, identifying the similar patterns of collectivity and variability in the two
kinds of biological systems. This generalization provides a basis for comparisons between collec-
tive cell migration in development and cancer invasion: “cross-fertilization between developmental
biology and cancer biology should accelerate progress in both fields” (Mishra et al, 2019, p. 2).
These examples illustrate how generalizations promote cross-fertilization between studies of
different biological systems in a number of ways. Each mechanism of collective cell migration
is efficiently investigated and better characterized by comparing and contrasting it with other
mechanisms operating in different biological systems. By “different biological systems,” I do
not just refer to taxonomic difference. It also refers to different organs (or component parts of
an organism) and different conditions of those systems (i.e., developmental, pathological, and
regenerative). Studies of apparently very different and distantly related biological processes (for
example, fruit fly ovary development, sprouting of zebrafish blood vessels, streaming migration of
the slime mold Dictyostelium, and human breast cancer invasion) inform each other by comparing
and contrasting them from specific perspectives. Indeed, an important rationale for having the
category of collective cell migration is that it promotes researchers to compare different biological
systems, which they would otherwise not compare, in a productive manner (Chapter 4; Yoshida,
2023). This cross-fertilization is mediated by generalizations.
123
5.3.4 Multiplicity of Generalizations as a Resource
It should be noted that multiple generalizations have been formulated about mechanisms of col-
lective cell migration. (This is different from my earlier claim that research on collective cell
migration employs multiple different types of visualizations to express generalizations; here I am
talking about the fact that there are multiple features that are shared across different examples of
collective cell migration about which researchers generalize.) For example, Schumacher (2019)
discusses eight important generalizations concerning collective cell migration that have been for-
mulated. (Schumacher calls them “principles” in the article.)
• Heterogeneity of cell states (equal to what I have called the leader-follower distinction)
• Substrate-free migration
• Contact-inhibition of locomotion
• Confinement and repulsive cues
• Self-generated gradients
• Stochastic group decisions
• Cell migration and substrate mechanics
• Reprogramming
Most of these generalizations are not universally applicable even to known examples of collective
cell migration; they have limited distributions. And their distributions only partially overlap. In
other words, although a single example of collective cell migration can and often does exemplify
multiple “principles” listed above, it is not the case that two principles always coincide. Although
Schumacher’s is not the only possible list of important generalizations in research on collective
cell migration, it provides a useful insight: there are multiple important generalizations about
124
mechanisms of collective cell migration, whose distributions only partially overlap. And different
model systems serve as useful sources of information about different features of mechanisms of
collective cell migration.
The multiplicity of generalizations has an important implication for research on collective cell
migration. That different generalizations have different distributions means that one model system
can contribute to articulating different subsets of examples of collective cell migration, depending
on which feature of the mechanisms one focuses on (Chapter 4; Yoshida, 2023). Furthermore,
contributions are often mutual. Studies of one model system can both inform and be informed
by studies of other model systems. Thus, multiple generalizations together provide a platform on
which studies of different biological systems mutually inform.
Whereas the discussion of this section has focused mostly on how generalizations facilitate
productive interactions among studies of different biological systems, generalizations also mediate
between studies of living model systems and studies of mathematical models of collective cell
migration. Mathematical modeling of collective cell migration has recently become an active area
of research (e.g., Buttenschön and Edelstein-Keshet, 2020; Méhes and Vicsek, 2014; Schumacher
et al, 2016). A model typically focuses on a very specific feature of mechanisms of collective
cell migration. It is often the case that a generalization (especially an abstract principle) is first
established in experimental research, which stimulates mathematicians, physicists, and theoretical
biologists’ attempts to formally represent that feature. Mathematical models developed in this way
can provide unique contributions to the area. They are especially useful in assessing which of the
features that appear in a mechanistic model are causally responsible for central characteristics of
collective cell migration.
These considerations suggest how generalizations facilitate cross-fertilization among different
125
approaches. Collective cell migration is studied by researchers with different interests: develop-
mental biologists who are interested in explaining development of various biological forms; cancer
biologists who are trying to elucidate mechanisms of cancer invasion and metastasis for inventing
better treatments; regeneration biologists who are hoping to better understand regeneration pro-
cesses; and mathematicians, physicists, and theoretical biologists who aim at identifying a few,
abstract and generally applicable principles. None of these interests can be regarded as the goal of
this area. A better characterization is that research on collective cell migration involves multiple
aims and interests. Moreover, these different interests contribute to each other, and suchmutual con-
tributions are (at least in part) mediated by generalizations: developmental biology, cancer biology,
and regeneration biology contribute to each other by exchanging insights into general principles and
conserved component mechanisms of collective cell migration; experimental research contributes
to theoretical approach informal generalizations to be formalized, while theoretical research con-
tributes to experimental research formalized general models that can be used in hypothesizing and
confirmation in experimental research. Generalizations mediate productive interactions between
studies motivated by different interests.
5.4 Philosophical Implications
My account regards the configuration of multiple representations as a form of generalization. The
skepticmight still think that this is toomuch of an expansion of the notion of generalization. Whether
this conceptual move is appropriate or not depends on what kinds of questions we want to answer
by studying generalizations. For example, much of the philosophical debate on laws of nature
has aimed at understanding the nature of empirically necessary regularities about how systems
126
must behave (Carroll, 2016). For this purpose, it seems appropriate to characterize generalizations
as universally quantified propositions since this formulation captures the idea of an exceptionless
regularity. My motivation is different. I am interested in how scientists study and make use
of regularities and diversity in the world by utilizing different investigative and representational
resources. The focus on configuration contributes to this project because it reveals how scientists
understand, explore, reason about, and communicate regularities without ignoring specificity and
diversity of their target systems. Adopting such a broader picture will enable us to analyze new
aspects of scientific generalizations and conduct more practice-based research on generalizations.
This will supplement the existing philosophical literature on scientific generalizations.
The emphasis on configuration also provides the philosophy of modeling with new insights. In
previous philosophical discussion about trade-offs among modeling desiderata, each desideratum
has been attributed to individual models. This is understandable, given a primary focus of this area
has been on mathematical modeling. A classic, influential paper on trade-offs was written by the
ecologist Richard Levins (1966) and addressed the problem of trade-off in mathematical modeling
of ecological processes. Since then, the trade-off literature has debated the nature of mathematical
representation—more specifically, howdifferent qualities, such as generality, precision, and realism,
are related with each other in mathematical models (Evans et al, 2013; Gelfert, 2013; Levins,
1993; Matthewson and Weisberg, 2009; Odenbaugh, 2003, 2006; Orzack and Sober, 1993).11
However, when we analyze diagrams and tables as visual representations, we should consider not
only how representations are constituted (i.e., what abstraction and idealization are involved), but
also how those representations are configured in a physical, two-dimensional space in a journal
11 Exceptions include Matthewson (2020) and Yoshida (2021), both of whom discuss configurations of
mechanistic models, and Inkpen (2016), who discusses trade-offs in experimental design.
127
article, textbook, conference slide, etc. This is because visual representation consists in presenting
information two-dimensionally. This consideration opens up new philosophical questions. For
example, in the context of visual representation, do we have to consider any modeling desiderata
in addition to the standard set of desiderata that have been discussed in the literature (such as
generality, precision, and realism)? Do we need a new conceptual framework for analyzing the
nature of and relations between modeling desiderata in visual representations? We may ask similar
questions about other forms of representations, such as three-dimensional physical models.
Another related implication concerns how we study scientific diagrams. My account accords
with an integrated approach to diagrammatic representations defended by Ambrosio (2020). Am-
brosio observes that there has been a divide in studies of diagrams. Whereas most of philosophical
discussion about diagrams concentrates on analyzing their representational nature, historians and
scholars of visual culture have criticized such a representational view. They have argued that
diagrams must be understood not as representations of something else, but as objects that exist in
the world and are a target of inquiry in their own right. (Ambrosio calls this the “object-based
view” of diagrams, which is contrasted with the “representational view” common in philosophy.)
But as Ambrosio rightly points out, studying diagrams as representations and treating them as
objects of inquiry in their own right are not incompatible. Although my analysis in this chapter was
based mainly on the representational view, it focused on diagrams and tables that exist in a specific
context (review articles) and examined what influences they exert on interactions in the research
community (cross-fertilization). Furthermore, my discussion of how scientists generalize was de-
pendent crucially on an analysis of not only how individual diagrams represent target systems, but
also how they are configured in a physical space. This example shows that by combining the two
views of diagrams, or incorporating insights from one view to the other, we can acquire deeper
128
understanding of how diagrams function.
Finally, how and to what extent is my account generalizable to other cases in science? This
chapter is based on a single, very specific example, hence certain aspects of my account might be
specific to research on collective cell migration. Nevertheless, I believe that attention to configura-
tion will provide useful insights for analyzing generalization practice in other fields of science. It
seems likely that displaying multiple representations in certain configurations is a generic strategy
to highlight features shared across the systems being represented, and hence employed in many
different fields of science. Although I discussed only four variations of this strategy (juxtaposed
mechanism diagrams, juxtaposed principle diagrams, spectrum presentation, and tables), there will
be more variations. I also expect that in many other areas of research, generalizations facilitate
cross-fertilization among studies of different target systems. This perspective might be particu-
larly useful for studying a field whose subcommunities specialize in specific systems or objects of
research. (Developmental biology is an example, where each researcher or laboratory often special-
izes in one or a few model organisms.) Like the case of collective cell migration, generalizations
might be promoting cross-system comparisons in such fields.
5.5 Conclusion
This chapter has focused on the question of how and why scientists generalize. It addressed
this question by challenging the three assumptions that have been influential in philosophical
discussion about generalization. The analysis of generalizations in review articles about collective
cell migration showed that researchers in this area generalize by employing different types of
visualizations. I illustrated that formulating a single representation with wide applicability is not
129
the only way to generalize; scientists sometimes generalize by configuring multiple representations,
which is a strategy to highlight shared features without eliminating details that differ across systems.
The different types of visualizations are used in a complementary way so that the reader can
understand generality, specificity, and diversity of the mechanisms in an integrated manner. I also
argued that generalizations facilitate cross-fertilization among studies of different target systems.
Multiple generalizations together provide a platform where studies of different biological systems
can informand contribute to each other. This chapter exemplifies that philosophical discussion about
scientific generalizations can be greatly informed by examining representational and investigative
practices of generalization.
130
Chapter 6: Concluding Remarks
Previous philosophical discussion has focused narrowly on analyzing the nature of natural laws and
roles of generalizations in scientific explanations. In contrast, this dissertation inquired into broader
investigative and representational practices of generalizations. In Chapter 1, I highlighted that this
dissertation would examine: (1) how different forms of generalizations are associated with different
investigative practices; (2) how multiple, non-universal generalizations are pursued and utilized
within an area of research; (3) what representational formats are used to express generalizations
and how that matters; and (4) what roles generalizations play in science beyond explanation. I
also emphasized that pursuits of regularities and system-specific details are closely interconnected,
which is another underlying theme of this dissertation.
Chapter 2 offered a basic characterization of generalizations in developmental biology. In
particular, I distinguished two forms of generalization: mechanisms and principles. They are dis-
tinguished in terms of the relevance of abstraction, have different justifications for their generaliz-
ability, and are associated with different investigative practices. This analysis has key consequences
for more general discussions of scientific generalizations. For example, the characterization of the
scope of a generalization must take into account factors specific to disciplinary and investigative
contexts and wide scope for a generalization does not necessarily depend on abstraction.
Chapter 3 discussed presentational and representational aspects of generalization from the
131
perspective of trade-offs among modeling desiderata. It characterized what I call multiple-models
juxtaposition (MMJ), a presentational strategy to coordinate the desiderata of generality and detail.
MMJ consists in displaying multiple models together in one physical space. This highlights general
features shared across different target systems, while appreciating system-specific details. I also
argued that MMJ is distinct from multiple-models idealization (MMI), which is a widely discussed
idealization strategy to manage trade-offs among modeling desiderata.
Chapter 4 focused on the use of biological model systems in studies of a phenomenon that is
produced by diverse mechanisms. The motivating questions were how biological systems can serve
as models of such a phenomenon and why scientists often continue treating such a phenomenon
as one thing. My analysis of a concrete example showed that studies of such a phenomenon can
formulate multiple, locally applicable generalizations with different but overlapping scope, which
allows multiple model systems to jointly represent the phenomenon of interest. This is a modeling
strategy for investigating generality in mechanistic diversity; it can identify regularities without
ignoring differences across biological systems.
Finally, Chapter 5 integrated insights from the previous chapters to challenge three influential
assumptions concerning scientific generalizations: generalizationsmust be expressed linguistically;
scientists generalize by formulating a single representation with wide applicability; and generaliza-
tions are valuable in science insofar as they enable scientific explanations. My analysis demonstrates
that non-propositional representations, such as diagrams and tables, play unique roles in reasoning
about and communicating generalizations. Importantly, generalizations are often expressed by dis-
playing multiple representations in specific configurations. Furthermore, generalizations facilitate
cross-fertilization among studies of different target systems, which is an important role beyond
enabling scientific explanations.
132
These chapters together provide a picture of how generalizations are practiced in developmental
biology. However, what is even more important is that this dissertation exemplifies an inquiry
into practices of generalization. It suggests a number of ways in which philosophers can study
practices of generalizations, such as: distinguish different forms of generalizations and characterize
and situate them within broader investigative practices; instead of focusing on a few, overarching
generalizations, examine how multiple generalizations are pursued and utilized together within
a particular research context; pay close attention to presentational and representational practices
that are employed to express generalizations; and, inquire into roles of generalizations beyond
explanation, especially how they contribute to further investigations in a research community.
There are aspects of generalization practices that this dissertation was not able to discuss in
detail. For example, most of my discussion (besides the brief discussion in Chapter 2) focused on
mechanisms and principles, which are both explanatory generalizations. (Note that this is different
from the explanation-centered view that I challenged in this dissertation. Explanatory generaliza-
tions indicate that an explanatory account, e.g., a mechanistic model, is applicable across a certain
range of systems or phenomena. In contrast, the explanation-centered view consists in a narrow
focus on how generalizations enable scientific explanations.) However, descriptive generalizations
(which involve practices of mapping observed features, categorizing entities, and naming the cate-
gories) are also a crucial part of scientific research. How are descriptive generalizations practiced?
How are they different from practices of explanatory generalization? For instance, what unique
presentational and representational practices are employed to express descriptive generalizations?
What distinct roles do descriptive generalizations play in science?
Another important question is the extent to which the analyses of this dissertation are applicable
to other fields. Although this dissertation focused on a single field, I have derived generalizable
133
insights and methodological suggestions from the analyses of this particular example. (I also have
considered the applicability of my accounts in different chapters.) At the same time, different fields
have different aims of research, differentmethodological constraints, and different balances between
interests in generality and system-specific details. These factors no doubt influence generalization
practices. Detailed investigations into generalization practices in various local contexts are needed
to develop a general picture of how scientists manage generality, diversity, and specificity in their
study of natural world.
134
References
Abrahamsen A, Bechtel W (2015) Diagrams as tools for scientific reasoning. Review of Philosophy and
Psychology 6(1):117–131, DOI 10.1007/s13164-014-0215-2
Abrahamsen A, Sheredos B, Bechtel W (2017) Explaining visually using mechanism diagrams. In: Glennan
S, Illari P (eds) The Routledge Handbook of Mechanisms and Mechanical Philosophy, Routledge, New
York, pp 238–254
AbzhanovA, Kaufman TC (1999) Homeotic genes and the arthropod head: Expression patterns of the labial,
proboscipedia, and Deformed genes in crustaceans and insects. Proceedings of the National Academy of
Sciences 96(18):10,224–10,229, DOI 10.1073/pnas.96.18.10224
Affolter M, Bellusci S, Itoh N, Shilo B, Thiery JP, Werb Z (2003) Tube or not tube: Remodeling epithelial
tissues by branching morphogenesis. Developmental Cell 4(1):11–18, DOI 10.1016/S1534-5807(02)
00410-0
Affolter M, Zeller R, Caussinus E (2009) Tissue remodelling through branching morphogenesis. Nature
Reviews Molecular Cell Biology 10(12):831–842, DOI 10.1038/nrm2797
Ambrosio C (2020) Toward an integrated history and philosophy of diagrammatic practices. East
Asian Science, Technology and Society: An International Journal 14(2):347–376, DOI 10.1215/
18752160-8538952
Ankeny RA (2007) Wormy logic: Model organisms as case-based reasoning. In: Creager ANH, Lunbeck E,
Wise MN (eds) Science without Laws: Model Systems, Cases, Exemplary Narratives, Duke University
Press, Durham, NC, pp 46–58
Ankeny RA, Leonelli S (2011)What’s so special about model organisms? Studies in History and Philosophy
of Science Part A 42(2):313–323, DOI 10.1016/j.shpsa.2010.11.039
Ankeny RA, Leonelli S (2020) Model Organisms, 1st edn. Cambridge University Press, DOI 10.1017/
9781108593014
Baetu TM (2014) Models and the mosaic of scientific knowledge. The case of immunology. Studies in
History and Philosophy of Science Part C: Studies inHistory and Philosophy of Biological andBiomedical
Sciences 45:49–56, DOI 10.1016/j.shpsc.2013.11.003
Baetu TM (2016) The ‘big picture’: The problem of extrapolation in basic research. The British Journal for
the Philosophy of Science 67(4):941–964, DOI 10.1093/bjps/axv018
135
Barker G, Kitcher P (2014) Philosophy of Science: A New Introduction. Oxford University Press, New York
Beatty J (1995) The evolutionary contingency thesis. In: Wolters G, Lennox JG (eds) Concepts, Theories,
and Rationality in the Biological Sciences, The University of Pittsburgh Press, Pittsburth, pp 45–81
BechtelW (2009) Generalization and discovery by assuming conservedmechanisms: Cross-species research
on circadian oscillators. Philosophy of Science 76(5):762–773, DOI 10.1086/605790
Bechtel W (2021) Explaining features of fine-grained phenomena using abstract analyses of phenom-
ena and mechanisms: Two examples from chronobiology. Synthese 198(S24):1–23, DOI 10.1007/
s11229-017-1469-x
Bechtel W, Abrahamsen A (2005) Explanation: A mechanist alternative. Studies in History and Philosophy
of Science Part C: Studies in History and Philosophy of Biological and Biomedical Sciences 36(2):421–
441, DOI 10.1016/j.shpsc.2005.03.010
Bechtel W, Abrahamsen A (2012) Diagramming phenomena for mechanistic explanation. Proceedings of
the Annual Meeting of the Cognitive Science Society 34(34):102–107
Bechtel W, Richardson RC (1993) Discovering Complexity: Decomposition and Localization as Strategies
in Scientific Research. Princeton University Press
Bellusci S, Grindley J, Emoto H, Itoh N, Hogan BLM (1997) Fibroblast Growth Factor 10 (FGF10)
and branching morphogenesis in the embryonic mouse lung. Development 124(23):4867–4878, DOI
10.1242/dev.124.23.4867
Bogen J (2005) Regularities and causality; Generalizations and causal explanations. Studies in History and
Philosophy of Science Part C: Studies in History and Philosophy of Biological and Biomedical Sciences
36(2):397–420, DOI 10.1016/j.shpsc.2005.03.009
Bokulich A (2014) How the tiger bush got its stripes: ’How possibly’ vs. ’how actually’ model explanations.
The Monist 97(3):321–338, DOI 10.5840/monist201497321
Bolker JA (2009) Exemplary and surrogate models: Two modes of representation in biology. Perspectives
in Biology and Medicine 52(4):485–499, DOI 10.1353/pbm.0.0125
Braillard PA, Malaterre C (2015) Explanation in biology: An introduction. In: Braillard PA, Malaterre C
(eds) Explanation in Biology: An Enquiry into the Diversity of Explanatory Patterns in the Life Sciences,
Springer Netherlands, Dordrecht
Brandon RN (1997) Does biology have laws? The experimental evidence. Philosophy of Science 64:S444–
S457, DOI 10.1086/392621
BurianRM,ThieffryD (2000) Introduction to the special issue ’From embryology to developmental biology’.
History and Philosophy of the Life Sciences 22(3):313–323
Buttenschön A, Edelstein-Keshet L (2020) Bridging from single to collective cell migration: A review of
models and links to experiments. PLOS Computational Biology 16(12):e1008,411, DOI 10.1371/journal.
pcbi.1008411
136
Campbell K, Casanova J (2016) A common framework for EMT and collective cell migration. Development
143(23):4291–4300, DOI 10.1242/dev.139071
Carmeliet P, Tessier-Lavigne M (2005) Common mechanisms of nerve and blood vessel wiring. Nature
436(7048):193–200, DOI 10.1038/nature03875
Carroll JW (2016) Laws of nature. In: The Stanford Encyclopedia of Philosophy, fall 2016 edn, Metaphysics
Research Lab, Stanford University
Cheng CW, Niu B, Warren M, Pevny LH, Lovell-Badge R, Hwa T, Cheah KSE (2014) Predicting the
spatiotemporal dynamics of hair follicle patterns in the developing mouse. Proceedings of the National
Academy of Sciences 111(7):2596–2601, DOI 10.1073/pnas.1313083111
Colaço D (2020) Recharacterizing scientific phenomena. European Journal for Philosophy of Science
10(2):14, DOI 10.1007/s13194-020-0279-z
Cooper G (1998) Generalizations in ecology: A philosophical taxonomy. Biology & Philosophy 13(4):555–
586, DOI 10.1023/A:1006508101996
Craver CF (2004) Dissociable realization and kind splitting. Philosophy of Science 71(5):960–971, DOI
10.1086/425945
Craver CF, Darden L (2013) In Search of Mechanisms: Discoveries across the Life Sciences. The University
of Chicago Press, Chicago
Davies JA (2002) Do different branching epithelia use a conserved developmental mechanism? BioEssays
24(10):937–948, DOI 10.1002/bies.10161
Davies JA (2013) Mechanisms of Morphogenesis. Academic Press, Cambridge, MA, DOI 10.1016/
B978-0-12-391062-2.01001-9
Davies JA (2017) Using synthetic biology to explore principles of development. Development 144(7):1146–
1158, DOI 10.1242/dev.144196
Dietrich MR, Ankeny RA, Crowe N, Green S, Leonelli S (2020) How to choose your research organism.
Studies in History and Philosophy of Science Part C: Studies in History and Philosophy of Biological and
Biomedical Sciences 80:101,227, DOI 10.1016/j.shpsc.2019.101227
Elliott-Graves A (2018) Generality and causal interdependence in ecology. Philosophy of Science
85(5):1102–1114, DOI 10.1086/699698
Elliott-Graves A (2022) What are general models about? European Journal for Philosophy of Science
12(4):74, DOI 10.1007/s13194-022-00502-9
Evans MR, Grimm V, Johst K, Knuuttila T, de Langhe R, Lessells CM, Merz M, O’Malley MA, Orzack SH,
Weisberg M, Wilkinson DJ, Wolkenhauer O, Benton TG (2013) Do simple models lead to generality in
ecology? Trends in Ecology & Evolution 28(10):578–583, DOI 10.1016/j.tree.2013.05.022
Ewald AJ, Brenot A, Duong M, Chan BS, Werb Z (2008) Collective epithelial migration and cell re-
arrangements drive mammary branching morphogenesis. Developmental Cell 14(4):570–581, DOI
10.1016/j.devcel.2008.03.003
137
FaganMB (2016) Generativemodels: Human embryonic stem cells andmultiple modeling relations. Studies
in History and Philosophy of Science Part A 56:122–134, DOI 10.1016/j.shpsa.2015.10.003
Fancher TT, Muto A, Fitzgerald TN, Magri D, Gortler D, Nishibe T, Dardik A (2008) Control of Blood
Vessel Identity: From Embryo to Adult. Annals of Vascular Diseases 1(1):28–34, DOI 10.3400/avd.
AVDrev07011
Ferrari A, Giampietro C (2019) Force and collective epithelial activities. In: La Porta CAM, Zapperi S (eds)
Cell Migrations: Causes and Functions, vol 1146, Springer International Publishing, Cham, pp 31–44
Fontana L, Partridge L, Longo VD (2010) Extending healthy life span—From yeast to humans. Science
328(5976):321–326, DOI 10.1126/science.1172539
Friedl P (2004) Prespecification and plasticity: Shifting mechanisms of cell migration. Current Opinion in
Cell Biology 16(1):14–23, DOI 10.1016/j.ceb.2003.11.001
Friedl P, Gilmour D (2009) Collective cell migration in morphogenesis, regeneration and cancer. Nature
Reviews Molecular Cell Biology 10(7):445–457, DOI 10.1038/nrm2720
Friedl P, Mayor R (2017) Tuning collective cell migration by cell–cell junction regulation. Cold Spring
Harbor Perspectives in Biology 9(4):a029,199, DOI 10.1101/cshperspect.a029199
Friedl P, Locker J, Sahai E, Segall JE (2012) Classifying collective cancer cell invasion. Nature Cell Biology
14(8):777–783, DOI 10.1038/ncb2548
Friedman M (1974) Explanation and scientific understanding. The Journal of Philosophy 71(1):5–19, DOI
10.2307/2024924
Gelfert A (2013) Strategies of model-building in condensed matter physics: Trade-offs as a demarcation
criterion between physics and biology? Synthese 190(2):253–272, DOI 10.1007/s11229-012-0145-4
Georgiadis V, Stewart HJ, Pollard HJ, Tavsanoglu Y, Prasad R, Horwood J, Deltour L, Goldring K, Poirier F,
Lawrence-Watt DJ (2007) Lack of galectin-1 results in defects in myoblast fusion andmuscle regeneration.
Developmental Dynamics 236(4):1014–1024, DOI 10.1002/dvdy.21123
Gerhardt H, Golding M, Fruttiger M, Ruhrberg C, Lundkvist A, Abramsson A, Jeltsch M, Mitchell C,
Alitalo K, Shima D, Betsholtz C (2003) VEGF guides angiogenic sprouting utilizing endothelial tip cell
filopodia. Journal of Cell Biology 161(6):1163–1177, DOI 10.1083/jcb.200302047
Ghabrial AS, Krasnow MA (2006) Social interactions among epithelial cells during tracheal branching
morphogenesis. Nature 441(7094):746–749, DOI 10.1038/nature04829
Gilbert SF (2014) Developmental Biology, 10th edn. Sinauer Associates, Sunderland MA
Glazier DS (2006) The 3/4-power law is not universal: Evolution of isometric, ontogenetic metabolic scaling
in pelagic animals. BioScience 56(4):325, DOI 10.1641/0006-3568(2006)56[325:TPLINU]2.0.CO;2
Glennan SS (1996) Mechanisms and the nature of causation. Erkenntnis 44(1), DOI 10.1007/BF00172853
Glennan SS, Illari P (2017) Introduction: Mechanisms and mechanical philosophies. In: Glennan SS, Illari
P (eds) The Routledge Handbook of Mechanisms and Mechanical Philosophy, Routledge, New York, pp
1–9
138
Gould SJ (1989) Wonderful Life: The Burgess Shale and the Nature of History. W. W. Norton & Company,
Inc, New York
Gray RS, Cheung KJ, Ewald AJ (2010) Cellular mechanisms regulating epithelial morphogenesis and cancer
invasion. Current Opinion in Cell Biology 22(5):640–650, DOI 10.1016/j.ceb.2010.08.019
Green JBA, Sharpe J (2015) Positional information and reaction-diffusion: Two big ideas in developmental
biology combine. Development 142(7):1203–1211, DOI 10.1242/dev.114991
Green S (2013) When one model is not enough: Combining epistemic tools in systems biology. Studies in
History and Philosophy of Science Part C: Studies inHistory and Philosophy of Biological andBiomedical
Sciences 44(2):170–180, DOI 10.1016/j.shpsc.2013.03.012
Green S (2015) Revisiting generality in biology: Systems biology and the quest for design principles.
Biology & Philosophy 30(5):629–652, DOI 10.1007/s10539-015-9496-9
Green S, Wolkenhauer O (2013) Tracing organizing principles: Learning from the history of systems
biology. History and Philosophy of the Life Sciences 35(4):553–576
Green S, Levy A, Bechtel W (2015) Design sans adaptation. European Journal for Philosophy of Science
5(1):15–29, DOI 10.1007/s13194-014-0096-3
Green S, Dietrich MR, Leonelli S, Ankeny RA (2018) ‘Extreme’ organisms and the problem of gen-
eralization: Interpreting the Krogh principle. History and Philosophy of the Life Sciences 40(4):65,
DOI 10.1007/s40656-018-0231-0
Green S, Dam MS, Svendsen MN (2021) Mouse avatars of human cancers: The temporality of
translation in precision oncology. History and Philosophy of the Life Sciences 43(1):27, DOI
10.1007/s40656-021-00383-w
Halina M (2018) Mechanistic explanation and its limits. In: Glennan SS, Illari P (eds) The Routledge
Handbook of Mechanisms and Mechanical Philosophy, Routledge, New York, pp 213–224
Harmer SL, Panda S, Kay SA (2001) Molecular bases of circadian rhythms. Annual Review of Cell and
Developmental Biology 17(1):215–253, DOI 10.1146/annurev.cellbio.17.1.215
Hellström M, Phng LK, Hofmann JJ, Wallgard E, Coultas L, Lindblom P, Alva J, Nilsson AK, Karlsson
L, Gaiano N, Yoon K, Rossant J, Iruela-Arispe ML, Kalén M, Gerhardt H, Betsholtz C (2007) Dll4
signalling through Notch1 regulates formation of tip cells during angiogenesis. Nature 445(7129):776–
780, DOI 10.1038/nature05571
Hempel CG (1965) Aspects of Scientific Explanation and Other Essays, 1st edn. Free Press, New York
Hempel CG (1966) Philosophy of Natural Science. Princeton Hall, Englewood Cliffs
Hodgkin AL, Huxley AF (1952) A quantitative description of membrane current and its application to
conduction and excitation in nerve. Journal of Physiology 117:500–544
Hopwood N (2009) Embryology. In: Bowler PJ, Pickstone JV (eds) The Cambridge History of Science: The
Modern Biological and Earth Sciences, Cambridge University Press, Cambridge, pp 285–315
139
Huebner RJ, Ewald AJ (2014) Cellular foundations of mammary tubulogenesis. Seminars in Cell & Devel-
opmental Biology 31:124–131, DOI 10.1016/j.semcdb.2014.04.019
Inkpen SA (2016) Like Hercules and the Hydra: Trade-offs and strategies in ecological model-building
and experimental design. Studies in History and Philosophy of Science Part C: Studies in History and
Philosophy of Biological and Biomedical Sciences 57:34–43, DOI 10.1016/j.shpsc.2016.02.019
Iruela-Arispe ML, Beitel GJ (2013) Tubulogenesis. Development 140(14):2851–2855, DOI 10.1242/dev.
070680
Khalil AA, Friedl P (2010) Determinants of leader cells in collective cell migration. Integrative Biology
2(11-12):568, DOI 10.1039/c0ib00052c
Kimelman D, Martin BL (2012) Anterior-posterior patterning in early development: Three strategies:
Anterior-posterior patterning in early development. Wiley Interdisciplinary Reviews: Developmental
Biology 1(2):253–266, DOI 10.1002/wdev.25
Kitcher P (1981) Explanatory unification. Philosophy of Science 48(4):507–531, DOI 10.1086/289019
Kitcher P (1989) Explanatory unification and the causal structure of the world. In: Kitcher, Philip, Salmon,
Wesley C (eds) Scientific Explanation, University of Minnesota Press, Minneapolis, pp 410–505
Kondo S, Miura T (2010) Reaction-diffusion model as a framework for understanding biological pattern
formation. Science 329(5999):1616–1620, DOI 10.1126/science.1179047
Krogh A (1929) The progress of physiology. Science 70(1809):200–204
Kuhn TS (1977) The Essential Tension: Selected Studies in Scientific Tradition and Change, 11th edn. Univ.
of Chicago Press, Chicago
Lange M (2013) What would natural laws in the life sciences be? In: Kampourakis K (ed) The Philosophy
of Biology, vol 1, Springer Netherlands, Dordrecht, pp 67–85
Levins R (1966) The strategy of model building in population biology. Amrican Scientist 54(4):421–431
Levins R (1993) A response to Orzack and Sober: Formal analysis and the fluidity of science. The Quarterly
Review of Biology 68(4):547–555, DOI 10.1086/418302
Levy A, Bechtel W (2013) Abstraction and the organization of mechanisms. Philosophy of Science
80(2):241–261, DOI 10.1086/670300
LevyA,CurrieA (2015)Model organisms are not (theoretical)models. TheBritish Journal for the Philosophy
of Science 66(2):327–348, DOI 10.1093/bjps/axt055
Liao C, Peng R, Luo Y, Zhou X, Wu X, Fang C, Chen J, Li B (2008) Altered ecosystem carbon and nitrogen
cycles by plant invasion: A meta-analysis. New Phytologist 177(3):706–714, DOI 10.1111/j.1469-8137.
2007.02290.x
Linquist S, Gregory TR, Elliott TA, Saylor B, Kremer SC, Cottenie K (2016) Yes! There are resilient
generalizations (or “laws”) in ecology. The Quarterly Review of Biology 91(2):119–131, DOI 10.1086/
686809
140
Llimargas M (1999) The Notch pathway helps to pattern the tips of the Drosophila tracheal branches by
selecting cell fates. Development 126:2355–2364, DOI 10.1242/dev.126.11.2355
Love AC (2010) Idealization in evolutionary developmental investigation: A tension between phenotypic
plasticity and normal stages. Philosophical Transactions of the Royal Society B: Biological Sciences
365(1540):679–690, DOI 10.1098/rstb.2009.0262
Love AC (2014) The erotetic organization of developmental biology. In: Minelli A, Pradeu T (eds) Towards
a Theory of Development, Oxford University Press, pp 33–55, DOI 10.1093/acprof:oso/9780199671427.
003.0003
Love AC (2018) Developmental mechanisms. In: Glennan SS, Illari P (eds) The Routledge Handbook of
Mechanisms and Mechanistic Philosophy, Routledge, New York, pp 332–347
Love AC (2022) Developmental biology. In: Zalta EN (ed) The Stanford Encyclopedia of Philosophy,
summer 2022 edn, Metaphysics Research Lab, Stanford University
Lu P, Lu Y (2021) Born to run? Diverse modes of epithelial migration. Frontiers in Cell and Developmental
Biology 9:704,939, DOI 10.3389/fcell.2021.704939
Lu P, Werb Z (2008) Patterning mechanisms of branched organs. Science 322(5907):1506–1509, DOI
10.1126/science.1162783
Machamer P, Darden L, Craver CF (2000) Thinking about mechanisms. Philosophy of Science 67(1):1–25,
DOI 10.1086/392759
Matthewson J (2011) Trade-offs in model-building: A more target-oriented approach. Studies in History
and Philosophy of Science Part A 42(2):324–333, DOI 10.1016/j.shpsa.2010.11.040
Matthewson J (2020) Detail and generality in mechanistic explanation. Studies in History and Philosophy
of Science Part A 80:28–36, DOI 10.1016/j.shpsa.2018.06.001
Matthewson J, Weisberg M (2009) The structure of tradeoffs in model building. Synthese An International
Journal for Epistemology, Methodology and Philosophy of Science 170(1):169–190, DOI 10.1007/
s11229-008-9366-y
Mayor R, Carmona-Fontaine C (2010) Keeping in touch with contact inhibition of locomotion. Trends in
Cell Biology 20(6):319–328, DOI 10.1016/j.tcb.2010.03.005
Mayor R, Etienne-Manneville S (2016) The front and rear of collective cell migration. Nature Reviews
Molecular Cell Biology 17(2):97–109, DOI 10.1038/nrm.2015.14
Méhes E, Vicsek T (2014) Collective motion of cells: From experiments to models. Integrative Biology
6(9):831–854, DOI 10.1039/C4IB00115J
Metzger RJ, Krasnow MA (1999) Genetic control of branching morphogenesis. Science (New York, NY)
284(5420):1635–1639, DOI 10.1126/science.284.5420.1635
Min H, Danilenko DM, Scully SA, Bolon B, Ring BD, Tarpley JE, DeRose M, Simonet WS (1998) Fgf-10
is required for both limb and lung development and exhibits striking functional similarity to Drosophila
branchless. Genes & Development 12(20):3156–3161, DOI 10.1101/gad.12.20.3156
141
Mishra AK, Campanale JP, Mondo JA, Montell DJ (2019) Cell interactions in collective cell migration.
Development 146(23):dev172,056, DOI 10.1242/dev.172056
Mitchell SD (1997) Pragmatic laws. Philosophy of Science 64(S4):S468–S479, DOI 10.1086/392623
Mitchell SD (2000) Dimensions of scientific law. Philosophy of Science 67(2):242–265, DOI 10.1086/
392774
Montell DJ (2008) Morphogenetic cell movements: Diversity from modular mechanical properties. Science
322(5907):1502–1505, DOI 10.1126/science.1164073
Morgan MS (2017) Narrative ordering and explanation. Studies in History and Philosophy of Science Part
A 62:86–97, DOI 10.1016/j.shpsa.2017.03.006
Morgan MS (2019) Exemplification and the use-values of cases and case studies. Studies in History and
Philosophy of Science Part A 78:5–13, DOI 10.1016/j.shpsa.2018.12.008
MorrisonM (2011) One phenomenon, manymodels: Inconsistency and complementarity. Studies in History
and Philosophy of Science Part A 42(2):342–351, DOI 10.1016/j.shpsa.2010.11.042
Muñoz-Chápuli R (2011) Evolution of angiogenesis. The International Journal of Developmental Biology
55(4-5):345–351, DOI 10.1387/ijdb.103212rm
Murray MJ (2015) Drosophila models of cancer. AIMS Genetics 2(1):97–103
Nogawa H, Morita K, Cardoso WV (1998) Bud formation precedes the appearance of differential cell pro-
liferation during branching morphogenesis of mouse lung epithelium in vitro. Developmental Dynamics
213(2):228–235, DOI 10.1002/(SICI)1097-0177(199810)213:2<228::AID-AJA8>3.0.CO;2-I
Norden C, Lecaudey V (2019) Collective cell migration: General themes and new paradigms. Current
Opinion in Genetics & Development 57:54–60, DOI 10.1016/j.gde.2019.06.013
Ochoa-Espinosa A, Affolter M (2012) Branching morphogenesis: From cells to organs and back. Cold
Spring Harbor Perspectives in Biology 4(10):a008,243–a008,243, DOI 10.1101/cshperspect.a008243
Odenbaugh J (2003) Complex systems, trade-offs, and theoretical population biology: Richard Levin’s
“Strategy of model building in population biology” revisited. Philosophy of Science 70(5):1496–1507,
DOI 10.1086/377425
Odenbaugh J (2006) The strategy of “The strategy of model building in population biology”. Biology &
Philosophy 21(5):607–621, DOI 10.1007/s10539-006-9049-3
Olson HM, Nechiporuk AV (2018) Using zebrafish to study collective cell migration in development and
disease. Frontiers in Cell and Developmental Biology 6:83, DOI 10.3389/fcell.2018.00083
Orzack SH (2005) Discussion: What, if anything, is “The strategy of model building in population biology?”
A comment on Levins (1966) and Odenbaugh (2003). Philosophy of Science 72(3):479–485, DOI
10.1086/498475
Orzack SH, Sober E (1993) A critical assessment of Levins’s The strategy of model building in population
biology (1966). The Quarterly Review of Biology 68(4):533–546, DOI 10.1086/418301
142
Painter KJ, Hunt GS, Wells KL, Johansson JA, Headon DJ (2012) Towards an integrated experimen-
tal–theoretical approach for assessing the mechanistic basis of hair and feather morphogenesis. Interface
Focus 2(4):433–450, DOI 10.1098/rsfs.2011.0122
Park WY, Miranda B, Lebeche D, Hashimoto G, Cardoso WV (1998) FGF-10 Is a chemotactic factor for
distal epithelial buds during lung development. Developmental Biology 201(2):125–134, DOI 10.1006/
dbio.1998.8994
Perini L (2005) Visual representations and confirmation. Philosophy of Science 72(5):913–926, DOI
10.1086/508949
Prasad M, Wang X, He L, Montell DJ (2011) Border cell migration: A model system for live imaging and
genetic analysis of collective cell movement. In: Wells CM, Parsons M (eds) Cell Migration, vol 769,
Humana Press, Totowa, NJ, pp 277–286
Pratt CW, Cornely K (2014) Essential Biochemistry, 3rd edn. Wiley, Hoboken, NJ
Press J (2015) Biological explanations as cursory covering law explanations. In: Braillard PA, Malaterre C
(eds) Explanation in Biology, vol 11, Springer Netherlands, Dordrecht, pp 367–385
Qin L, Yang D, Yi W, Cao H, Xiao G (2021) Roles of leader and follower cells in collective cell migration.
Molecular Biology of the Cell 32(14):1267–1272, DOI 10.1091/mbc.E20-10-0681
Raerinne J (2013) Stability and lawlikeness. Biology & Philosophy 28(5):833–851, DOI 10.1007/
s10539-013-9386-y
Raerinne J (2015) Evolutionary contingency, stability, and biological laws. Journal for General Philosophy
of Science 46(1):45–62, DOI 10.1007/s10838-014-9271-7
Reichert H (2009) Evolutionary conservation of mechanisms for neural regionalization, proliferation and
interconnection in brain development. Biology Letters 5(1):112–116, DOI 10.1098/rsbl.2008.0337
Reutlinger A (2014a) The generalizations of biology: Historical and contingent? In: Kaiser MI, Scholz
OR, Plenge D, Hüttemann A (eds) Explanation in the Special Sciences, vol 367, Springer Netherlands,
Dordrecht, pp 131–153
Reutlinger A (2014b) Natural law and universality in the philosophy of biology. European Review
22(S1):S145–S162, DOI 10.1017/S1062798713000811
Robert JS (2004)Model systems in stem cell biology. BioEssays 26(9):1005–1012, DOI 10.1002/bies.20100
Rørth P (2009) Collective cell migration. Annual Review of Cell andDevelopmental Biology 25(1):407–429,
DOI 10.1146/annurev.cellbio.042308.113231
RyanTJ, Grant SGN (2009) The origin and evolution of synapses. Nature ReviewsNeuroscience 10(10):701–
712, DOI 10.1038/nrn2717
Saraiva JE, Barriga EH (2021) The basics of collective cell migration: Unity makes strength. In: Pajic-
Lijakovic I, Barriga EH (eds) Viscoelasticity and Collective Cell Migration, Academic Press, Cambridge
MA, pp 1–19, DOI 10.1016/B978-0-12-820310-1.00001-X
143
Scarpa E,MayorR (2016)Collective cellmigration in development. Journal of Cell Biology 212(2):143–155,
DOI 10.1083/jcb.201508047
Schumacher L (2019) Collective cell migration in development. In: La Porta CAM, Zapperi S (eds) Cell
Migrations: Causes and Functions, vol 1146, Springer International Publishing, Cham, pp 105–116
Schumacher LJ, Kulesa PM, McLennan R, Baker RE, Maini PK (2016) Multidisciplinary approaches to
understanding collective cell migration in developmental biology. Open Biology 6(6):160,056, DOI
10.1098/rsob.160056
Sekine K, Ohuchi H, Fujiwara M, Yamasaki M, Yoshizawa T, Sato T, Yagishita N, Matsui D, Koga Y,
Itoh N, Kato S (1999) Fgf10 is essential for limb and lung formation. Nature Genetics 21(1):138–141,
DOI 10.1038/5096
Shellard A, Mayor R (2019) Supracellular migration – beyond collective cell migration. Journal of Cell
Science 132(8):jcs226,142, DOI 10.1242/jcs.226142
Sheredos B, Bechtel W (2017) Sketching biological phenomena and mechanisms. Topics in Cognitive
Science 9(4):970–985, DOI 10.1111/tops.12290
Sheredos B, Burnston D, Abrahamsen A, Bechtel W (2013) Why do biologists use so many diagrams?
Philosophy of Science 80(5):931–944, DOI 10.1086/674047
Siekmann AF, Lawson ND (2007) Notch signalling limits angiogenic cell behaviour in developing zebrafish
arteries. Nature 445(7129):781–784, DOI 10.1038/nature05577
Siekmann AF, Covassin L, Lawson ND (2008)Modulation of VEGF signalling output by the Notch pathway.
BioEssays 30(4):303–313, DOI 10.1002/bies.20736
Slack JMW (2006) Essential Developmental Biology, 2nd edn. Wiley, Hoboken, NJ
Smart JJC (1963) Philosophy of Scientific Realism. Routledge & Kegan Paul, London
Soler L, Zwart S, Lynch M, Israel-Jost V (eds) (2014) Science after the Practice Turn in the Philosophy,
History, and Social Studies of Science. Routledge Studies in the Philosophy of Science, Routledge, New
York
Spurlin JW, Nelson CM (2017) Building branched tissue structures: From single cell guidance to
coordinated construction. Philosophical Transactions of the Royal Society B: Biological Sciences
372(1720):20150,527, DOI 10.1098/rstb.2015.0527
Steinert B, MacCord K (2018) Visualizing the cell: Pictorial styles and their epistemic goals in general
cytology. In: Matlin KS, Maienschein J, Laubichler MD (eds) Visions of Cell Biology: Reflections
Inspired by Cowdry’s "General Cytology", University of Chicago Press, pp 134–155, DOI 10.7208/
chicago/9780226520650.003.0007
Stewart TA, Bhat R, Newman SA (2017) The evolutionary origin of digit patterning. EvoDevo 8(1):21,
DOI 10.1186/s13227-017-0084-8
Strevens M (2011) Depth: An Account of Scientific Explanation. Harvard University Press, Cambridge,
MA
144
Stuelten CH, Parent CA, Montell DJ (2018) Cell motility in cancer invasion and metastasis: Insights from
simple model organisms. Nature Reviews Cancer 18(5):296–312, DOI 10.1038/nrc.2018.15
Sutherland D, Samakovlis C, Krasnow MA (1996) Branchless encodes a Drosophila FGF homolog that
controls tracheal cell migration and the pattern of branching. Cell 87(6):1091–1101, DOI 10.1016/
S0092-8674(00)81803-6
Tee SH (2018) Mechanism diagrams and abstraction-by-aggregation. Studies in History and Philosophy
of Science Part C: Studies in History and Philosophy of Biological and Biomedical Sciences 71:17–25,
DOI 10.1016/j.shpsc.2018.10.003
Tekin E, Hunt D, Newberry MG, Savage VM (2016) Do vascular networks branch optimally or ran-
domly across spatial scales? PLOS Computational Biology 12(11):e1005,223, DOI 10.1371/journal.
pcbi.1005223
Theveneau E, Linker C (2017) Leaders in collective migration: Are front cells really endowed with a
particular set of skills? F1000Research 6:1–11, DOI 10.12688/f1000research.11889.1
Theveneau E, Mayor R (2011) Can mesenchymal cells undergo collective cell migration? The case of the
neural crest: The case of the neural crest. Cell Adhesion & Migration 5(6):490–498, DOI 10.4161/cam.
5.6.18623
Theveneau E, Steventon B, Scarpa E, Garcia S, Trepat X, Streit A, Mayor R (2013) Chase-and-run between
adjacent cell populations promotes directional collective migration. Nature Cell Biology 15(7):763–772,
DOI 10.1038/ncb2772
Torii KU (2012) Two-dimensional spatial patterning in developmental systems. Trends in Cell Biology
22(8):438–446, DOI 10.1016/j.tcb.2012.06.002
Trinkaus JP (1969) Cells into Organs: The Forces That Shape the Embryo. Prentice Hall, Englewood Cliffs
Tufte ER (1990) Envisioning Information. Graphics Press, Cheshire, CT
Turchin P (2001) Does population ecology have general laws? Oikos 94(1):17–26, DOI 10.1034/j.
1600-0706.2001.11310.x
Turing A (1952) The chemical basis of morphogenesis. Philosophical Transactions of the Royal Society of
London Series B, Biological Sciences 237(641):37–72
Uechi H, Kuranaga E (2017) Mechanisms of collective cell movement lacking a leading or free front edge
in vivo. Cellular and Molecular Life Sciences 74(15):2709–2722, DOI 10.1007/s00018-017-2489-x
VarnerVD,NelsonCM (2014)Cellular and physicalmechanisms of branchingmorphogenesis. Development
141(14):2750–2759, DOI 10.1242/dev.104794
Waddington CH (1956) Principles of Embryology. The Macmillan Co., New York
Wang RN, Green J, Wang Z, Deng Y, Qiao M, Peabody M, Zhang Q, Ye J, Yan Z, Denduluri S, Idowu O,
Li M, Shen C, Hu A, Haydon RC, Kang R, Mok J, Lee MJ, Luu HL, Shi LL (2014) Bone Morphogenetic
Protein (BMP) signaling in development and human diseases. Genes & Diseases 1(1):87–105, DOI
10.1016/j.gendis.2014.07.005
145
Wang S, Sekiguchi R, Daley WP, Yamada KM (2017) Patterned cell and matrix dynamics in branching
morphogenesis. Journal of Cell Biology 216(3):559–570, DOI 10.1083/jcb.201610048
Waters CK (1998) Causal regularities in the biological world of contingent distributions. Biology & Philos-
ophy 13(1):5–36, DOI 10.1023/A:1006572017907
Weisberg M (2004) Qualitative theory and chemical explanation. Philosophy of Science 71(5):1071–1081,
DOI 10.1086/428011
Weisberg M (2006) Robustness analysis. Philosophy of Science 73:730–742, DOI 10.1086/518628
Weisberg M (2007) Three kinds of idealization. The Journal of Philosophy 104(12):639–659, DOI 10.5840/
jphil20071041240
Weisberg M (2013) Simulation and Similarity: Using Models to Understand the World. Oxford Studies in
Philosophy of Science, Oxford University Press, New York
Wimsatt WC (2012) Robustness, reliability, and overdetermination (1981). In: Soler L, Trizio E, Nickles T,
Wimsatt W (eds) Characterizing the Robustness of Science, vol 292, Springer Netherlands, Dordrecht,
pp 61–87
Woodward J (2001) Law and explanation in biology: Invariance is the kind of stability that matters.
Philosophy of Science 68(1):1–20, DOI 10.1086/392863
Woodward J (2003) Making Things Happen: A Theory of Causal Explanation, oxford univ. press paperback
edn. Oxford Studies in Philosophy of Science, Oxford Univ. Press, Oxford
Woody AI (2004) More telltale signs: What attention to representation reveals about scientific explanation.
Philosophy of Science 71(5):780–793, DOI 10.1086/421416
Woody AI (2014) Chemistry’s periodic law: Rethinking representation and explanation after the turn to
practice. In: Soler L, Lynch M, Israel-Jost V (eds) Science after the Practice Turn in the Philosophy,
History, and Social Studies of Science, Routledge, pp 123–150, DOI 10.4324/9781315857985
Woody AI (2015) Re-orienting discussions of scientific explanation: A functional perspective. Studies in
History and Philosophy of Science Part A 52:79–87, DOI 10.1016/j.shpsa.2015.03.005
Wu MY, Hill CS (2009) TGF- superfamily signaling in embryonic development and homeostasis. Devel-
opmental Cell 16(3):329–343, DOI 10.1016/j.devcel.2009.02.012
Yoshida Y (2021) Multiple-models juxtaposition and trade-offs among modeling desiderata. Philosophy of
Science 88(1):103–123, DOI 10.1086/710054
Yoshida Y (2023) Joint representation: Modeling a phenomenon with multiple biological systems. Studies
in History and Philosophy of Science 99:67–76, DOI 10.1016/j.shpsa.2023.03.002
Zegers MM, Friedl P (2014) Rho GTPases in collective cell migration. Small GTPases 5(3):e983,869,
DOI 10.4161/sgtp.28997
146