ABSTRACT Understanding how anatomical structures evolve requires disentangling the roles of integration and modularity in shaping morphological variation. The vertebral column, a serially repeated and regionally differentiated structure, provides a powerful system for investigating these processes. Here, we examine how vertebral morphology evolves in relation to whole-body elongation across the adaptive radiation of Lake Malawi cichlid fishes. We tested for evolutionary integration between the precaudal and caudal domains, as well as assessed the contributions of vertebral count, centrum shape, and intervertebral spacing on body elongation. We find strong evolutionary integration between the shapes of precaudal and caudal vertebrae, with both vertebral shapes varying along similar axes. Despite this, precaudal and caudal vertebral counts evolve independently, indicating a decoupling between the specification of identity and the development of their respective shapes. Whole-body elongation is significantly associated with coordinated changes in vertebral and rib morphology, including proportional increases in centrum size, posterior displacement of neural and haemal spines, and increased rib curvature. In contrast, centrum elongation and intervertebral spacing do not contribute to body elongation across the radiation. These results demonstrate that body elongation in cichlids necessitates integrated, multivariate changes in axial morphology. Our findings highlight the importance of morphological integration in facilitating coordinated evolutionary responses in anatomical systems.

ABSTRACT Phenotypic diversity is often thought to arise from the evolutionary modification of developmental processes. However, developmental processes are tightly coupled in space and time, with each process beginning from conditions set by the one before it. While we know from dynamical systems theory that initial conditions can significantly affect a system’s out-come, their importance as a source of phenotypic evolvability has been largely overlooked. Here we show for the first time, that phenotypic evolution can proceed through changes in developmental initial conditions while the underlying developmental process remains conserved. Somitogenesis is the process by which vertebral precursors, known as somites, are periodically patterned in the pre-somitic mesoderm (PSM). Somitic count (total number of somites) is thought to diversify through the evolution of components of somitogenesis such as the tempo of the segmentation clock or the mechanisms driving axial morphogenesis. Using two closely related species of Lake Malawi cichlid fishes that differ in vertebral counts, we show that somite count evolution has happened without changes to somitogenesis itself, but instead, by altering the size of the PSM at the onset of this process. This work will expand what we consider developmental drivers of phenotypic evolution and highlight the importance of comparative studies to understand the diversification of phenotypes.

Vertebrae arise from somites, transient embryonic segments that rhythmically bud from the presomitic mesoderm during axial elongation. The number and identity of vertebrae are ultimately determined by somitogenesis and subsequent anterior-posterior regionalisation, largely governed by hox gene expression. Interspecific variation in vertebral count and regionalisation therefore reflects evolutionary changes in somite number and homeotic identity following species divergence. While many macroevolutionary studies have examined homeotic and non-homeotic changes in the vertebral column, few have explored these dynamics in teleosts, despite their exceptional species richness. Using African cichlids as a model, we show that shifts in vertebral regionalisation can arise through modifications to anterior-posterior patterning, but that much of the observed variation is driven by changes in somite number, with homeotic effects emerging largely as a by-product of somitic changes. Moreover, low intraspecific variation in vertebral count, lacking phylogenetic structure, suggests that somitic count variation within species is strongly canalised and has remained consistent throughout the diversification of African cichlids. In addition, we find no correlation between intraspecific variation in vertebral counts and mean vertebral counts, and this variation does not consistently scale with body aspect ratio among individuals. Therefore, intraspecific variation is decoupled from both macroevolutionary patterns of vertebral count evolution and body shape diversification. Together, our findings highlight the dynamic interplay between somitogenesis and homeotic transformations in shaping vertebral diversity and underscore the value of cichlids as a model for understanding the developmental basis of axial evolution in teleosts. ### Competing Interest Statement The authors have declared no competing interest. BBSRC, 2445747

Developmental biology seeks to unravel the intricate regulatory mechanisms orchestrating the transformation of a single cell into a complex, multicellular organism. Dynamical systems theory provides a powerful quantitative, visual and intuitive framework for understanding this complexity. This Primer examines five core dynamical systems theory concepts and their applications to pattern formation during development: (1) analysis of phase portraits, (2) bistable switches, (3) stochasticity, (4) response to time-dependent signals, and (5) oscillations. We explore how these concepts shed light onto cell fate decision making and provide insights into the dynamic nature of developmental processes driven by signals and gradients, as well as the role of noise in shaping developmental outcomes. Selected examples highlight how integrating dynamical systems with experimental approaches has significantly advanced our understanding of the regulatory logic underlying development across scales, from molecular networks to tissue-level dynamics.

Vertebrates have evolved great diversity in the number of segments dividing the trunk body, however the developmental origin of the evolvability of this trait is poorly understood. The number of segments is thought to be determined in embryogenesis as a product of morphogenesis of the pre-somitic mesoderm (PSM) and the periodicity of a molecular oscillator active within the PSM known as the segmentation clock. Here we explore whether the clock and PSM morphogenesis exhibit developmental modularity, as independent evolution of these two processes may explain the high evolvability of segment number. Using a computational model of the clock and PSM parameterised for zebrafish, we find that the clock is broadly robust to variation in morphogenetic processes such as cell ingression, motility, compaction, and cell division. We show that this robustness is in part determined by the length of the PSM and the strength of phase coupling in the clock. As previous studies report no changes to morphogenesis upon perturbing the clock, we suggest that the clock and morphogenesis of the PSM exhibit developmental modularity.

The number of vertebrae in the axial skeleton of vertebrates is extremely diverse, and reflects adaptations to a diverse range of habitats and lifestyles. The capacity for heritable evolutionary change in the number of vertebrae — its evolvability — is underpinned by the process of somitogenesis, which determines the number of somites that form in the early embryo. However, despite the evolvability of somitogenesis having been crucial for the success of the vertebrates across evolutionary history, the developmental sources of evolvability in somitogenesis are still unknown. Here, we review the evolution of somitogenesis and vertebral number, and attempt to identify sources of evolvability within this important developmental process.

During organ development cells undergo significant morphological and positional changes. Yet, by the end of organogenesis, internal organ structure is typically robustly defined, with cells tightly packed. It remains an open question as to how the three-dimensional (3D) internal structure of an organ emerges reliably, particularly when there are multiple cell types interacting and dynamic boundary constraints. Here, we utilise quantitative live imaging and 3D morphological measures of the developing zebrafish myotome to unravel how early muscle organisation emerges. Contrary to the textbook view of muscle fibres as cylindrical, myocytes undergo an ordered chiral twist, the direction and magnitude of which depends on their position within the myotome. Further, cells skew and rearrange, seemingly to facilitate close packing of neighbouring muscle fibres. Cell movement undergoes a rapid decline in speed once the cells span the myotome segment. We find that cell packing is altered in mutants that disrupt cell fate or cell fusion, even though the final muscle segments remain largely confluent. Biophysical perturbation reveals that the cells are mechanically plastic, able to adjust to changes in the local cellular environment and boundary constraints. Taking these results together, we propose that the early myotome undergoes a structural transition, from a fluid-like state into a frozen state, resembling glass-like behaviour. Cellular plasticity in response to varying boundary constraints may be a general mechanism for ensuring robust organ morphogenesis in dense 3D tissues. ### Competing Interest Statement The authors have declared no competing interest. Biotechnology and Biological Sciences Research Council, BB/W006944/1 Engineering and Physical Sciences Research Council, https://ror.org/0439y7842, EP/W023865/1 Australian Research Council, FL210100107 National Health and Medical Research Council, Ideas Grant 2027559 NSF, PHY-2309135

African cichlids comprise more than 1800 species of freshwater fishes, with remarkable adaptive radiations in Lakes Tanganyika, Malawi, and Victoria that have given rise to extraordinary morphological diversity. However, the evolution of the cichlid axial skeleton has been largely overlooked, despite its high variation and functional significance for locomotion. Here, we present the first macroevolutionary study of axial morphology in African cichlids, based on phylogenetic comparative analyses of 4861 individuals from 583 species. Adaptation to demersal, pelagic, and piscivorous niches has led to the evolution of elongate bodies with high vertebral counts in lacustrine cichlids, emphasising the role of the fusiform body shape in ecological adaptation. However, riverine species occupy a broader axial morphospace than lacustrine species, which is partly explained by a higher stochastic rate of vertebral count evolution in riverine lineages. In addition, the occupied axial morphospace broadly correlates with the estimated age of the lacustrine radiations, suggesting that exploration of axial morphospace is a function of divergence time. However, rates of vertebral count evolution are not the same across the lake radiations. Therefore, accumulated variation in vertebral counts (and more broadly axial morphospace) is not solely a function of divergence time. Finally, we show that the common ancestor of African cichlids possessed a distinctly riverine axial morphology, indicating that the exploration of axial morphospace radiated outward from this ancestral riverine form. These findings highlight the importance of a comparative approach to studying cichlid evolution and underscore the value of African cichlids as a model for investigating the evolutionary and developmental dynamics of the teleostean vertebral column. ### Competing Interest Statement The authors have declared no competing interest. BBSRC, 2445747 Swiss National Science Foundation, https://ror.org/00yjd3n13, 176039, 208002

Here we describe a dataset of freely available, readily processed, whole-body μCT-scans of 56 species (116 specimens) of Lake Malawi cichlid fishes that captures a considerable majority of the morphological variation present in this remarkable adaptive radiation. We contextualise the scanned specimens within a discussion of their respective ecomorphological groupings and suggest possible macroevolutionary studies that could be conducted with these data. In addition, we describe a methodology to efficiently μCT-scan (on average) 23 specimens per hour, limiting scanning time and alleviating the financial cost whilst maintaining high resolution. We demonstrate the utility of this method by reconstructing 3D models of multiple bones from multiple specimens within the dataset. We hope this dataset will enable further morphological study of this fascinating system and permit wider-scale comparisons with other cichlid adaptive radiations.

The study of pattern formation has benefited from our ability to reverse-engineer gene regulatory network (GRN) structure from spatio-temporal quantitative gene expression data. Traditional approaches have focused on systems where the timescales of pattern formation and morphogenesis can be separated. Unfortunately, this is not the case in most animal patterning systems, where pattern formation and morphogenesis are co-occurring and tightly linked. To elucidate patterning mechanisms in such systems we need to adapt our GRN inference methodologies to include cell movements. In this work we fill this gap by integrating quantitative data from live and fixed embryos to approximate gene expression trajectories (AGETs) in single cells and use these to reverse-engineer GRNs. This framework generates candidate GRNs that recapitulate pattern at the tissue level, gene expression dynamics at the single cell level, recover known genetic interactions and recapitulate experimental perturbations while incorporating cell movements explicitly for the first time.

Summary The transition state model of cell differentiation proposes that a transient window of gene expression stochasticity precedes entry into a differentiated state. As this has been assessed primarily in vitro , we sought to explore whether it can also be observed in vivo . Zebrafish neuromesodermal progenitors (NMps) differentiate into spinal cord and paraxial mesoderm at the late somitogenesis stages. We observed an increase in gene expression variability at the 24 somite stage (24ss) prior to their differentiation. From our analysis of a published 18ss scRNA-seq dataset, we showed that the NMp population possesses a signature consistent with a population undergoing a critical transition. By building in silico composite gene expression maps from our image data, we were able to assign an ‘NM index’ to each in silico NMp based on the cumulative expression of its neural and mesodermal markers. With the NM index distributions, we demonstrated that cell population heterogeneity of the NMps peaked at 24ss. We then incorporated stochasticity and non-autonomy into a genetic toggle switch model and uncovered the existence of rebellious cells, which we then confirmed by reexamining the composite maps. Taken together, our work supports the transition state model within an endogenous cell fate decision making event.

The mechanisms underpinning the formation of patterned cellular landscapes has been the subject of extensive study as a fundamental problem of developmental biology. In most cases, attention has been given to situations in which cell movements are negligible, allowing researchers to focus on the cell-extrinsic signalling mechanisms, and intrinsic gene regulatory interactions that lead to pattern emergence at the tissue level. However, in many scenarios during development, cells rapidly change their neighbour relationships in order to drive tissue morphogenesis, while also undergoing patterning. To draw attention to the ubiquity of this problem and propose methodologies that will accommodate morphogenesis into the study of pattern formation, we review the current approaches to studying pattern formation in both static and motile cellular environments. We then consider how the cell movements themselves may contribute to the generation of pattern, rather than hinder it, with both a species specific and evolutionary viewpoint.

As tissues elongate, cell rearrangement alters positional information in manner that must be accounted for to generate gene expression pattern. How this is achieved during paraxial mesoderm elongation is unknown. By reverse-engineering gene regulatory networks that predict single cell expression trajectories across the tissue, we find a network capable of recapitulating the full range of dynamic differentiation profiles observed both in vivo and in vitro. Simulating gene expression profiles on in toto cell tracking data sets reveal that temporal exposure to Wnt and FGF is generated by cell movement. The absence of reversal in gene expression towards a more premature gene expression state predicts the generation of aberrant tbx6 expression in the posterior progenitor zone that we then confirm by quantitative single cell imaging. Taken together, these results demonstrate cell rearrangement tunes the dynamics of mesoderm progenitor differentiation to generate pattern emergence as a function of temporal Wnt and FGF exposure.

Machine learning approaches are becoming increasingly widespread and are now present in most areas of research. Their recent surge can be explained in part due to our ability to generate and store enormous amounts of data with which to train these models. The requirement for large training sets is also responsible for limiting further potential applications of machine learning, particularly in fields where data tend to be scarce such as developmental biology. However, recent research seems to indicate that machine learning and Big Data can sometimes be decoupled to train models with modest amounts of data. In this work we set out to train a CNN-based classifier to stage zebrafish tail buds at four different stages of development using small information-rich data sets. Our results show that two and three dimensional convolutional neural networks can be trained to stage developing zebrafish tail buds based on both morphological and gene expression confocal microscopy images, achieving in each case up to 100% test accuracy scores. Importantly, we show that high accuracy can be achieved with data set sizes of under 100 images, much smaller than the typical training set size for a convolutional neural net. Furthermore, our classifier shows that it is possible to stage isolated embryonic structures without the need to refer to classic developmental landmarks in the whole embryo, which will be particularly useful to stage 3D culture in vitro systems such as organoids. We hope that this work will provide a proof of principle that will help dispel the myth that large data set sizes are always required to train CNNs, and encourage researchers in fields where data are scarce to also apply ML approaches.

Biology is dynamic in nature. From ecological systems to embryonic pattern formation: change is at the centre of any biological phenomenon. The last three decades of molecular genetics have been incredibly successful at identifying the components involved in many biological processes, and now we find ourselves at the advent of very exciting times where new methodologies and technologies are, for the first time, allowing us to address the dynamics of these processes directly. Biologists can now quantify the dynamics of biological processes [1–4], analyse [2,5,6] and image them [7–9] in unprecedented resolution. These and other related advances have been shifting the way we represent biological phenomena, away from static representations and towards increasingly more dynamic and therefore realistic accounts. Biological dynamics are steadily moving to the forefront of many fields in biology. Increasingly more dynamic perspectives and explanations are challenging the validity of static analyses, which although generally more tractable both from a theoretical and an experimental perspective, will have to be justified rather than assumed. The mechanisms underlying biological phenomena will need to address and explain the timing of the processes being investigated as well as their components and spatial distribution. Close interdisciplinary collaborations will be required in order to develop new techniques, methodologies, models, computational tools and conceptual frameworks to address and explain the dynamics that have always characterized biological systems and processes at every level of their organization.

Pattern formation in development has been principally studied in tissues that are not undergoing extensive cellular rearrangement. However, in most developmental contexts, gene expression domains emerge as cells re-arrange their spatial positions within the tissue, providing an additional, and seldom explored, level of complexity to the process of pattern formation in vivo. To investigate this issue, we addressed the regulation of TBox expression in the presomitic mesoderm (PSM) as this tissue develops in zebrafish embryos. Here, cells must differentiate in a manner that leads to well-defined spatial gene expression domains along the tissue while undergoing rapid movements to generate axial length. We find that in vivo, mesoderm progenitors undergo TBox differentiation over a broad range of time scales while in vitro their differentiation is simultaneous. By reverse-engineering a gene regulatory network (GRN) to recapitulate TBox gene expression, we were able to predict the populationlevel differentiation dynamics observed in culture, but not in vivo. In order to address this discrepancy in differentiation dynamics we developed a ‘Live Modelling’ framework that allowed us to simulate the GRN on 3D tracking

The authors, an artist, a mathematician and a biologist, describe their collaboration examining the potential of drawing to further the understanding of biological processes. As a case study, this article considers C.H. Waddington's powerful visual representation of the “epigenetic landscape,” whose purpose is to unify research in genetics, embryology and evolutionary biology. The authors explore the strengths and limitations of Waddington's landscape and attempt to transcend the latter through a collaborative series of exploratory images. Through careful description of this drawing process, the authors touch on its epistemological consequences for all participants.

There is much talk about information in biology. In developmental biology, this takes the form of "positional information," especially in the context of morphogen-based pattern formation. Unfortunately, the concept of "information" is rarely defined in any precise manner. Here, we provide two alternative interpretations of "positional information," and examine the complementary meanings and uses of each concept. Positional information defined as Shannon information helps us understand decoding and error propagation in patterning systems. General relativistic positional information, in contrast, provides a metric to assess the output of pattern-forming mechanisms. Both interpretations provide powerful conceptual tools that do not compete, but are best used in combination to gain a proper mechanistic understanding of robust patterning.