The role of self-organization in developmental evolution
- First Online:
- Cite this article as:
- Bozorgmehr, J.E.H. Theory Biosci. (2014) 133: 145. doi:10.1007/s12064-014-0200-4
In developmental and evolutionary biology, particular emphasis has been given to the relationship between transcription factors and the cognate cis-regulatory elements of their target genes. These constitute the gene regulatory networks that control expression and are assumed to causally determine the formation of structures and body plans. Comparative analysis has, however, established a broad sequence homology among species that nonetheless display quite different anatomies. Transgenic experiments have also confirmed that many developmentally important elements are, in fact, functionally interchangeable. Although dependent upon the appropriate degree of gene expression, the actual construction of specific structures appears not directly linked to the functions of gene products alone. Instead, the self-formation of complex patterns, due in large part to epigenetic and non-genetic determinants, remains a persisting theme in the study of ontogeny and regenerative medicine. Recent evidence indeed points to the existence of a self-organizing process, operating through a set of intrinsic rules and forces, which imposes coordination and a holistic order upon cells and tissue. This has been repeatedly demonstrated in experiments on regeneration as well as in the autonomous formation of structures in vitro. The process cannot be wholly attributed to the functional outcome of protein–protein interactions or to concentration gradients of diffusible chemicals. This phenomenon is examined here along with some of the methodological and theoretical approaches that are now used in understanding the causal basis for self-organization in development and its evolution.
KeywordsSelf-organization Gene regulatory networks Morphogenesis Evo-devo Regeneration
The origin of biological form continues to present fundamental challenges to the study of both developmental and evolutionary biology. The basic assumption up until now has been that the elusiveness of this problem is because of the complexity of the relationship between the genotype and phenotype. By understanding exactly how gene products interact with each other, throughout the course of all stages of ontogeny, an adequate solution can be found (Wolpert 1991). Differences in the outcome of embryonic development should, ultimately, be attributable to differences in linear sequences of DNA (Alonso and Wilkins 2005) although few would claim any direct relationship. But this assumption has become precarious in recent years as sequenced genomes belonging to species of diverse physiology have been found to be remarkably similar. Within many taxonomic groups, there exists a substantive degree of phenotypic diversity that belies the observed sequence homology. Genes are often functionally identical across divergent lineages and, allowing for some differences in sequence, most have been evolutionarily conserved (Cooper and Brown 2008). In addition, the genetic basis for complex morphological traits remains largely obscure (Monteiro and Podlaha 2009).
This is surprising given that it is presumed that there must somehow be a correlation, however elaborate, between DNA and morphology. But the once prevailing notion of a genetic “blueprint”, containing all the instructions necessary to build a multicellular organism from a single egg cell, has since receded. Instead, it has been recognized that genes themselves represent only the “building blocks” of life and do not readily explain the higher-order complexity evident at the organismic level (Keim 2010; Britten 2003). The sequence information contained in the genome relates principally to the synthesis and regulation of chemical materials (Hood and Galas 2003); genes themselves are just templates for the amino acid sequences of proteins or the nucleotide sequences of functional RNA transcripts. What they do not encode, as is sometimes misperceived, are actual biological structures and systems, such as hearts and spleens, or the modalities of development (Chandebois and Faber 1987). To illustrate this point, the human brain may contain as many as 1015 synaptic connections between its neurons—roughly 300,000 times more than the number of base pairs in the entire human genome! Thus, while gene products are heavily involved in the make-up and operation of the body, their role in the determination of often distinctly different morphological outcomes is unclear: this is because many proteins are not trait-specific, but are found in various tissue types (e.g. the collagens).
Conversely, in the orthodox dogma of genetic determinism, cells/organisms are “programmed” to behave/function in a certain way because the genes encased within them instruct them accordingly. This metaphor of a genetic “program”, responsible for an organism’s entire development, is indeed a prevalent one among both scientists and philosophers alike. Instead of regarding the nuclear genome as a molecular repository/library of information, which the cell accesses to produce the proteins it needs, the dominant paradigm of the genetic program regards the chromosome-containing nucleus as the “command centre” for the cell and all of its operations. A seminal article by Nijhout (1990) describes how concepts concerning genetic “controls” and “programs” were originally conceived just as metaphors to help define and direct avenues of research, even though it has long been known that enucleated cells may survive for months without genes and yet are capable of effecting complex responses to environmental and cytoplasmic stimuli (Lipton et al. 1991). Another major problem with the idea of a genetic program, and the primacy of DNA, is that genes are not “self-actualizing” entities: just as a program itself needs a compiler or interpreter before it can be executed and run, so genes are in need of something else to control their activation. Yet another problem is that all of an organism’s cells have identical genomes; hence, if cells are programmed in exactly the same way, it is not readily apparent why they should end up having such different fates in development as happens.
One clear example of how researchers have come to terms with the limited role of the gene is in the aftermath of the sequencing of the microscopic nematode worm, C. elegans. It possesses less than 1,000 somatic cells, but contains over 20,000 protein-coding and 16,000 RNA genes in its chromosomes, prompting some to wonder why the lowly worm needs so many (Hodgkin 2001). Remarkably, this is roughly as many as that for humans who have tens of trillions of cells. Moreover, amongst all eukaryotes, the great majority of the protein-coding genes are duplicate variants of each other, many of which may be functionally redundant to a greater or lesser extent (Bozorgmehr 2012).
The cetartiodactyl clade, to use another helpful illustration, includes species as physically (and psychologically) diverse as giraffes, pigs, whales, camels, sheep and hippopotami. Cetacean mammals possess an anatomy entirely suited to a marine life—namely, a blowhole and breathing apparatus that allows them to inhale oxygen and store it deep under water, powerful flukes for aquatic locomotion, a dorsal fin for stabilization, flippers for steering, and an ovoid “melon” for echolocation (Thewissen 1999; Henry et al. 1983). Their close relatives in giraffes possess circulatory, muscular and respiratory systems that afford them the use of long necks to graze on the high leaves of the acacia tree (Simmons and Altwegg 2010). They also possess ruminant digestive systems, quite unlike those of cetaceans or suids. Although many important genetic adaptations that affect metabolism and internal biochemistry have been identified in the minke whale (Yim et al. 2014), the available data also indicate that there is relatively little to distinguish the artiodactyls—in terms of their DNA (Matthee et al. 2001)—compared with members of orders displaying far more homogeneity of form.
The outcome of the human and chimpanzee genome projects shows that, despite tens of millions of base pair substitutions and insertions/deletions since the two species last shared a common ancestor, there are actually few significant functional differences in their respective DNA. Even some of those regions identified as having undergone accelerated evolution in the human lineage have been found to be false positives, due to a compensatory effect or to biased recombination rather than to adaptive innovation (Kostka et al. 2012). All of these evidences suggest that the molecular causal basis for the diversity of form, itself resulting from the process of morphogenesis, is less pronounced than first thought. These observations have theoretical implications for the study of evolutionary developmental biology (evo-devo), which has relied on DNA as the source of morphological information and the means to inherit it: Held (2009) lists over 100 unsolved puzzles in human evo-devo, all arising from the prevailing assumption that genes determine developmental outcomes. But there are also practical considerations, such as for regenerative medicine and bioengineering, since the extent to which genes influence cell behavior and tissue patterning needs to be known. The field of synthetic morphology depends upon being able to induce cells to organize into specific arrangements, structures and tissues (Davies 2008). It so matters whether genes, and their regulatory elements, are the best level at which to try to manipulate shape or if other factors could be more instructive (Levin and Stevenson 2012).
Gene regulatory networks
Developmental transcription factors (DTFs) are known to play a vital part in morphogenesis through their binding to DNA and recruitment of RNA polymerase with which to control gene expression. Many DTFs comprise a “toolkit” used to activate or repress cascades of target genes, often involving a whole developmental pathway. The basic aspect of a gene regulatory network (GRN) is the functional linkage between these DTFs and the cognate cis-regulatory elements (CREs) of their target genes which provide the integration sites in DNA (Davidson and Erwin 2006). Each node of a gene regulatory network consists of a gene encoding a transcription factor, or a signaling protein, together with the cis-regulatory modules controlling expression for that gene (Davidson and Levine 2008). In addition, many genes implicated in intercellular communication, such as the integrins (Hynes 2002), or those related to cell adhesion, like the cadherins (Angst et al. 2001), are necessary components in the biomechanics of tissue formation which depends upon inductive interactions between the epithelium and mesenchyme. Through a complex matrix of protein–protein interactions, it is believed that embryogenesis proceeds and the basic body plan laid down (Davies 2005). In this way, the creation of biological form is considered to be either consequential or incidental to the aggregate of a fully mechanistic chain of causality, itself necessarily governed by the laws of physics and chemistry.
One of the key concepts behind the GRN in development, and any evolution related to it, is that it is not the genes themselves that necessarily matter the most, since they are shared across the board, but rather how they are used. Any changes in genes and/or DNA elements involved in the transcriptional control of large sets of genes could be seen as providing the basis for major phenotypical change. Two prominent expositions of this idea have been provided by Carroll (2008), Sansom (2011). The latter regards development as being part of a collaborative project of multiple transcription factors that regulate genes in a qualitatively consistent way; the former dwells instead upon the cis-regulatory sites that affect vast networks of transcription factors and the multitude of target genes that they control. Sansom thus argues that developmental evolution can be explained in terms of a framework of interacting genes, whereas Carroll proposes that morphological change is driven by specific mutations within CREs. Both positions on how gene networks “control” development, however, are limited to a view of how development undergoes evolutionary change. Sansom indeed admits that knowledge about gene regulation will not be enough to explain development, but that it is an essential first step. Therefore, to explore the worth of any general hypothesis of developmental evolution requires evaluating GRNs regarding their constituent main elements, i.e., DTFs and CREs.
It is certainly true that DTFs serve as essential tools in development, although it is also apparent that they have a very limited ability to direct and organize cells into the formation of particular organs of distinct shape and size. This is because, as explained above, all they are actually known to do is to bind with DNA, and other proteins, in order to facilitate transcription (as their name denotes). They are indeed found in unicellular life such as yeast (Struhl 1993) and in bacteria (Balleza et al. 2009), and transcription factors also have important non-developmental roles in multicellular organisms such as in immunology networks (Jonsson and Peng 2005) or the constant production of hemoglobin used to transport oxygen in red blood cells. Moreover, any DNA-binding protein can activate any gene so long as the latter contains the appropriate recognition sites in its promoter (Gehring 2002). In a classic case, the homeotic gene Pax6—a gene regarded as necessary for eye and sensory organ development—was transferred between a mouse and a fruit fly but nonetheless induced ommatidium formation (Halder et al. 1995). This is despite the fact that the trans-activating domain of the fly homologue is quite different in sequence. But if the same gene is interchangeable in the construction of structures as radically different as an insect’s compound eyes and those of a mammal, that gene cannot be determining the morphology. Apart from illustrating how cells may be recruited into a particular developmental pathway, homeotic master control genes such as Pax6 fail to sufficiently demonstrate the manner in which complex structures like the eye are actually formed. What they do show is how important transcriptional regulation is to development: reduced expression of Pax6 in the anterior embryonic midline and optic primordia of cavefish, for example, is the major reason for their blindness (Tian and Price 2005); eye degeneration is caused by apoptosis of the embryonic lens which in turn triggers developmental arrest. Therefore, without the proper spatiotemporal expression of Pax6, eye morphogenesis cannot even begin to proceed as it normally would. But its function only represents a first step, signaling the initiation of a process, rather than anything that is more important.
Furthermore, the “mosaic pleiotropism” exhibited by DTFs, and development genes in general, is suggestive of a more basic role rather than a very sophisticated one. This is because most genes regulating development participate in multiple, and often independent processes, in the patterning of morphologically disparate parts. Thus their function is that of a necessary instruction, activating large numbers of genes, but does not determine how they are then employed. For example, the homeobox gene Pitx2 is involved in the activation of genes used in the development of the eye, tooth, heart and abdominal organs (Faucourt et al. 2001); the sonic hedgehog (Shh) signaling protein exhibits a vast array of activity, affecting digit number and polarity, cerebellum development, feather bud formation, spinal cord and teeth in chickens (McMahon et al. 2003). As a result, mutations in this gene can be very detrimental. Given such an extensive range of operation, it is thus difficult to suppose that DTFs are responsible for the physical organization of specific structures. Rather, the same toolkit genes are observed to be needed for morphologically unrelated features, and in anatomies both simple and complex. The lancelet, a primitive chordate, contains as many Hox cluster genes as humans (though not duplicated on other chromosomes) despite lacking a true skeleton (Minguillon et al. 2005a). The basic functionality of the genes remains the same, although their deployment differs among species.
The causal power of transcription factors is thus liable to be over-interpreted. As an example of this, the T-box family is implicated, among many other things, in limb morphogenesis. Tbx4 and Tbx5 genes have been considered to play a role in regulating limb outgrowth and specifying the identity of forelimbs and hindlimbs, respectively (Rodriguez-Esteban et al. 1999). However, more recent research suggests that the paralogs have actually “subfunctionalized”, resulting in their differential expression in the emerging limb buds. It so happens that they both have a common role in initiating outgrowth, but not in determining limb-specific morphologies and identities as was once thought (Minguillon et al. 2005b). Loss of Tbx4 during later limb outgrowth produces no limb defects, revealing only a brief developmental requirement for its function (Naiche and Papaioannou 2007). Pbx homeodomain proteins are also recognized as regulating limb patterning and outgrowth (Capellini et al. 2011) but, as far as explaining morphological differences, this is highly problematic because their coding sequences are exceptionally well conserved in metazoan taxa and their general functionality has remained largely unchanged (Moens and Selleri 2006). This degree of conservation, rather than any variation, is typical for most DTFs (Gerhart and Kirschner 1997). An analysis of the Hox13 genes of the tammar wallaby, a marsupial known for its bipedal hopping, reveals a stringent preservation by selection in both gene structure and predicted proteins when compared to eutherians, such as the bat or mouse, where Hox13 gene expression features in the autopod (Renfree et al. 2012).
It is often stated that subtle differences due to mutation in elements of non-coding DNA are likely to be responsible for the diversity of form in organisms (Razeto-Barry and Maldonado 2011), and so provide a causal mechanism to explain evolutionary developmental biology. Due to their inherent modularity, the expression pattern of a gene may be controlled by several regulatory elements almost independently of each other. Since changes in cis elements affect only the expression of a gene, rather than its core functionality, they may avoid any negative pleiotropic effects (De Robertis 2008). The majority of the coding genome may remain strongly conserved, but changes in these “switches”, mostly enhancer elements, could effectively reconfigure/rewire gene regulatory circuits leading to novel and adaptive phenotypes (Gellon and McGinnis 1998; Stern and Orgogozo 2008; Levine 2010). But as with DTFs, all that these binding sites (each only 5–15 bp long) actually do is to act as landing platforms for proteins through which the amount of the target gene’s mRNA transcript is modulated.
Also, an increasing number of publications reveal the conservation and functional constraint of these very cis elements (Pennacchio et al. 2006; Rebeiz et al. 2012), even for those without sequence similarity (Hughes and Weirauch 2010; Ayyar et al. 2010; Ruvinsky and Ruvkun 2003; Wittkopp 2006). A notable experiment in transgenesis investigated the function of a CRE controlling the HoxD cluster in gnathostome vertebrates (Schneider et al. 2011). It is speculated that there are molecular mechanisms responsible for the formation of the fins of fish, compared to the limbs of tetrapods, and these could be explained by identifying key differences in regions influencing gene expression. But the cis element affecting the HoxD cluster’s expression was found to be interchangeable between a mouse and a zebrafish: transgenic CsB enhancers were observed to control HoxD activity in a similar fashion, thereby revealing a great conservation of both cis- and trans-acting factors in appendage development. This should not be unexpected when it is appreciated that the enhancer was providing the same degree of transcriptional regulation to the HoxD genes, found operating in the same distal segments, despite the difference in form. The tetrapod-specific CsC enhancer has also been found to promote similar expression in both zebrafish fins and mouse limbs (Freitas et al. 2012). Another transgenic experiment on a sequence in the same cluster, this time between a mouse and coelacanth, produced a similar outcome by driving normal expression in the former (Amemiya et al. 2013). Fish regulatory landscapes have also been shown to trigger HoxA gene expression in the proximal limb segments of tetrapods (Woltering et al. 2014), while research by Sagai et al. (2005) has identified an Shh enhancer with a similar role in the control of fin/limb development. It appears that the same transcriptional machinery is often used irrespective of the physical end result. This detracts from the notion that limb innovation arose primarily through the modification or gain of specific cis elements.
It is still possible that some differences in CREs linked to important genes, especially those related to axial patterning, could result in more significant changes in terms of spatial positioning (Brakefield 2011). Belting et al. (1998) conducted an experimental transgenesis of Hoxc8 in the mouse and chicken and observed significant, albeit limited, alterations in the expression boundaries of genes. These were inferred to be “heterochronic” shifts, i.e., changes in the timing of expression, resulting in the slight posteriorization of spatial domains. There is evidence that modifications to tissue-specific enhancers can produce some morphological change: it has been demonstrated that D. sechellia larvae gradually lose their trichome hairs by the accumulation of changes in transcriptional enhancers of the svb gene (Frankel et al. 2011). A similar development is observed in Nasonia wasp species where changes in ncDNA around unpaired-like (Upd) growth-factor genes lead to correspondingly significant changes in wing width (Loehlin and Werren 2012). Deletions of CREs can also lead to the complete loss of morphological structures, such as the pelvic spines of sticklebacks (Shapiro et al. 2004) or penile spines in the human lineage (McLean et al. 2011). But these facts do not explain the development of novel traits even if they do the adjustment or extension of existing ones. It is for this reason that Hoekstra and Coyne (2007) stress changes in genes coding for structural proteins as being more important. The loss or reduction of features is, however, a major component of evolutionary theory (Shubin and Dahn 2004), and mutations affecting cis elements may have been responsible for the complete loss of limbs in snakes (Mansfield 2013) and hind legs in manatees (Shapiro et al. 2006).
In recent years, in addition to DTFs and CREs, there has also been interest in non-coding RNAs (including micro-RNAs) as post-transcriptional regulators that bind to mRNA and which also interact with proteins. It has been shown that knockdown of lincRNAs has major consequences for gene expression patterns in embryos (Pauli et al. 2011), and that they affect the development of pluripotent stem cells (Guttman et al. 2011). The revelations of the ENCODE project (Birney et al. 2012), which has mapped the functional landscape of the entire human genome, provide ample evidence for a vast array of regulatory elements within ncDNA. The role of DNA methylation and histone modification in the epigenetic control of gene expression and cell differentiation is another growing area of research (Jaenisch and Bird 2003; Khavari et al. 2010). Epigenetic alterations, in contrast to mutations in DNA sequences, may be responsible for some evolutionary change in the great apes (Hernando-Herraez et al. 2013). But while these are relatively new avenues that elucidate upon the complex mechanisms of transcription, they still do not address the principal cause of formation itself.
Shape and the predictive power of genetic data
The strength of any scientific hypothesis lies in its predictive power. If no predictions can be made, then it can run the risk of being termed unfalsifiable. Hence, if genetic information is the primary determining factor in the creation of form, then one would at least, in principle, expect to be able to predict shape emergence based upon a set of molecular data. As previously mentioned, in so much as regenerative medicine and biosynthetic engineering is concerned, it is very useful to be able to control shapes and predict how they emerge. Significant success has indeed been achieved in predicting the tertiary structure of a protein from its amino acid sequence, out of a vast number of possible conformations, by analyzing the evolutionary constraints on residues in protein folds (Marks et al. 2011). The mechanisms by which genes are presumed to control organ shapes, however, remain poorly understood; genetic modulation of tissue polarity organizers, as well as growth rates, has been suggested as one possible mechanism (Green et al. 2010). Merks and Glazier (2005), by contrast, argue that a cell-centered approach is more appropriate since genes can only indirectly influence morphology whereas cells’ biophysical properties, and not their internal chemistry, are more relevant. In any case, not knowing the actual dynamics of shape formation does not mean that predictions cannot be made or that mathematical/computational models (Morelli et al. 2012) cannot be used at all.
There are several morphometric techniques that can be used to try and quantify shape, and thus make any investigation of it more of an exact science. It should therefore be feasible to directly map any differences in DNA to their effects, in terms of shape, and so represent this correlation numerically.
Many of these methods involve measuring the outline of shape by eigenvector and elliptic Fourier analysis, as well as the positioning of definable “landmarks” (Rohlf 1986). Klingenberg and Leamy (2001) have established a relationship between genetic and phenotypic covariance for the murine mandible. One study has identified five genes as playing a critical role in the development of the facial morphology in humans and the differences among individuals (Liu et al. 2012). It offers up the possibility that phenotypic variations can be directly explained in terms of mutations in DNA. One of the genes, Pax3, is a DTF expressed in the neural crest cells and is believed to determine the position of the nasion. This is corroborated by Paternoster et al. (2012). SNPs, specifically in non-coding intronic regions, have been identified with phenotypic variations in the nasion. This is most likely because they affect the timing of protein production, and so induce subtle positional deviations. It should be noted, however, that nucleotide differences affecting the identified genes served only to disturb an already specified spatial deployment of their products that marked out a geometric plan. The Pax3 product is itself yet another ultra-conserved DTF, and there is a 97 % amino acid identity between human and chicken orthologs. Unless its own cis-regulatory elements, or those of the target genes that it binds to, are sufficiently different in terms of function, and presumably sequence as well, it is hard to see where its role in the development of the human nasion is actually derived from. In the case of the Hox cluster genes, it is often assumed that they somehow encode “positional information” along anatomical axes that are common among organisms as diverse as humans and flies (Held 2010). This is based upon the observation of their precise spatiotemporal distribution. But there is nothing intrinsic about either the gene sequences themselves, or their regulation, that can explain this pattern.
Also in mice, but with relevance to humans, deletions of certain cis-regulatory enhancer elements have been observed that fine-tune craniofacial morphologies (Attanasio et al. 2013). Without these differences, facial shapes could well be identical with one another since the enhancer variations produce subtle, but significant, perturbations in the process of formation by altering the level of gene expression or by affecting its timing. In this way, they may cause deviations from an archetypal form. Thus, while the prospect of measuring shape, and correlating it with DNA sequences, may be useful as far as studying the causes of intraspecies variations is concerned, it is likely to be much more problematic when explaining major anatomical differences that exist among genera and higher taxa.
Nonetheless, there are those like Carroll (2005) who thinks that variations in DNA can potentially generate “endless” forms—even for those not present in Nature or belonging to extinct species. But now, with an increasing amount of genetic data now available, it should soon be possible to fully test this idea. The ability to transplant chromosomes, either through nuclear transfer or even as individual strands (Fisher et al. 2005), has the potential to settle any uncertainty if a trans-genomic experiment between distinct multicellular species could be successfully conducted. If the organism’s embryonic development were to proceed according to that of the maternal host, rather than the form of the species donating the DNA, this could falsify a key tenet of genetic determinism. One such case has already been reported, involving the nuclear genome of a red common carp being inserted into the enucleated egg of a goldfish, both members of the Cyprinidae family but of different genera (Sun et al. 2005). Although many characteristics of the resulting fish resembled the carp donor, the vertebral development, and the number of somites, resembled the recipient goldfish. This outcome shows that, while genetic differences are important, the genome does not itself dictate the course of development.
The definition and place of self-organization in biology
If it seems evident that the regulation of gene expression is by itself insufficient to explain the developmental evolution of intricate and interdependent structures, then what does? The answer to this may perhaps be found by looking beyond a reductionist understanding of biology, where genetic functionality is given paramount importance, and considering a much more holistic approach. This was first articulated by D’Arcy Thompson (1917) who offered a descriptive account that represented organisms as geometrical structures, and suggested that morphogenesis could be explained in terms of physical forces. “Self-organization” is a now a term used in science that refers to spontaneous order arising out of inherent properties rather than from extrinsic factors. There is no universal, or indeed operational, definition of what “self-organizing” means since it is a phenomenon that can be observed in different contexts in both the physical and life sciences. Even so, a slightly modified version of one proferred by Camazine et al. (2001) can be used here: self-organization is identified as a process by which a system, involving several independent components, along with a set of local interaction rules, becomes synergistically combined without reference to any global and coherent pattern. This degree of order arises out of an initially disordered or chaotic state, and the process is not controlled by anything else both from within and outside of the system. The initial conditions and the local rules of interaction serve to constrain relationships among the various components. They are consistent with, but not equivalent to, physical laws as they are not general but depend on the environmental context.
This can lead to non-linear and emergent properties, and the outcome can be complex and difficult to predict since the local rules do not directly specify any of the global properties. It means that the output of the system is not proportional to the input, and that it includes characteristics that differ, both qualitatively and quantitatively, from those of each of the constituent parts when separate to one another. Simple rules governing individual animal behavior, for example, can lead to patterns of collective behavior, such as flocking in birds or swarming in insects, that are not pre-determined but which are emergent. An “attraction–repulsion” principle can explain this type of aggregation (Herbert-Read et al. 2011) where each individual uses simple rules of interaction to respond to its neighbors. It can also be applied to the chemotaxis of cells that move in response to signals (Liu and Wang 2012).
Self-organization is indeed regarded as having a central role in cell biology and development (Karsenti 2008; Kauffman 1993; Chaplain et al. 1999), and not just in more familiar areas like protein folding (Gerstman and Chapagain 2005). The latter is perhaps more akin to “self-assembly” since a set of components converge upon a stable, static structure that has reached equilibrium (Halley and Winkler 2008). Dobrescu and Purcarea (2011) envisage self-organization in morphogenesis as being a dynamic, non-equilibrium, multilevel process and suggest that the order observed, even if produced by agglomerative means, is an inherent one where micro-level rules give rise to macro-level behavior. Complex patterns may, therefore, emerge from the collective and continual interaction of a small number of basic principles even though a system may itself appear to be far more complex (Boettiger and Oster 2009). Physico-chemical systems include many kinetic transitions and non-equilibrium thermodynamical processes (Nicolis 1977). The fall towards equilibrium and maximum entropy is that which drives self-organization and accounts for spontaneity. To adapt to a changing environment the system needs to have many stable states, and a distributed order, to withstand any perturbations.
The spatiotemporal outcomes of these processes tend to be probabilistic, predictable only statistically, rather than fully deterministic. This is also one of the key tenets of quantum theory and its “uncertainty principle” (Heisenberg 1930), in contrast to that of the classical physics that still persists in mechanistic biology. The number of energetically possible patterns is vast, and so offers immense scope for the imposition of order on otherwise stochastic or indeterminate types of activity. Some of these may be quite improbable, occurring as they always do in open and dissipative systems, and it has been shown that apparent transgressions of thermodynamic principles may occur in life to create this order (Schiffmann 1997). Edelmann and Denton (2007) regard self-organization as an innovative agency in evolution, while Batten et al. (2008) suggest that the self-organizing forces of Nature propose multiple possibilities, the least suitable of which natural selection will necessarily dispose of.
Self-organization is also a feature of cellular architecture that ensures structural stability without loss of plasticity (Misteli 2001). Perhaps the most obvious example of this at the organelle level is the cytoskeleton during cell division where mitotic spindle forms dynamically through the use of microtubules (Prost et al. 2010), which function as scaffolding in the cytoskeleton, and whose spatial distribution is important to many cellular activities. Gravity has been shown to trigger this self-organizing process by breaking the symmetry of an initially homogenous state that leads to the emergence of form and pattern (Papaseit et al. 2000). Other intracellular structures, such as the nuclear pore complex, or the proteasome that recycles proteins, represent astonishing examples of this at the molecular level. Although regulatory proteins are implicated in mediating their biogenesis (Murata et al. 2009), their assembly is spontaneous. Scale-invariant models have also demonstrated that self-organizing behavior exists during the earliest stages of embryogenesis (Tiraihi et al. 2011).
Even so, providing a comprehensive and quantitative explanation for the causes of self-organization related to biological form remains a daunting task. Fractal geometry, as formulated by Mandelbrot (1982), allows natural structures to be quantitatively characterized in geometric terms even if their form is not regular since it deals with the geometry of hierarchies and random processes. It has been suggested as a “design principle” for living organisms that can describe the surfaces of membranes and blood vessels (Weibel 1991). Tissue growth based on fractal concepts has been proposed as a model for clustered patterning evident in lung morphogenesis (Nelson and Manchester 1988). Approaches that involve self-organized percolation, whereby systems make phase transitions in a lattice, and also analyses of lacunarity, showing how fractal patterns fill space, have produced useful results (Alencar et al. 1997). Empirical and theoretical approaches can at least provide a foundation for understanding the relationship between the genetic and non-genetic factors that control development. Some of the most impressive evidence for self-organization can be found in the regeneration of complex biological structures, both in vivo and in vitro. What is significant about it is that it shows how dynamic organisms, and their constituent cells, can be within an abnormal situation.
Regeneration and autonomous formation
All life has at least some capacity to regenerate, although it is more pronounced in some organisms than it is in others. Thousands of cuttings can be taken from a plant, each of which can then grow into a new plant. This salient fact serves to distinguish living organisms from man-made machines. Understanding the dynamic laws of biological regeneration, and the self-organization that takes place therein, may prove indispensable to providing an insight into the causal basis of ontogeny. It also has deep implications as far as regenerative medicine is concerned (Levin 2011), especially in the use of scaffolds and molecules that guide a self-organization of host or exogenous cells into tissues in vivo. The process of regeneration reveals that organisms possess a wholeness that is more than just the some of their parts; parts can be removed, and yet wholeness is restored. One of the most outstanding examples to illustrate regeneration, at least among vertebrates, is when a portion of a newt’s eye is surgically removed—something unlikely to happen in the wild and whose restoration natural selection would have favored. Normally developed out of the epidermis at the embryonic stage, the lens re-emerges in a stepwise fashion from the edge of the iris instead (Henry and Tsonis 2010). Neither age nor repeated amputation seems to diminish this regenerative capability (Eguchi et al. 2011). But it turns out that an injection of the growth factor, FGF2, is all it takes to trigger the onset of development (Hayashi et al. 2004), along with the subsequent expression of other genes. Usually, this fibroblast growth protein would be produced via a transcription factor like EGR-1 (Jimenez et al. 2004). Regeneration is thus observed to be a quite spontaneous process, in response to the appropriate biochemical and environmental stimuli, by making adjustments to the normal course of development.
This does not mean that patterning in development and regeneration always use dissimilar means. There is evidence that regeneration can also recapitulate and redeploy the normal developmental process when the limbs of the axolotl, a salamander, are amputated (Roensch et al. 2013). But it is still true that, during regeneration, cells can find themselves in an abnormal position and have to respond by creating an alternative to the route used in embryogenesis that entails different starting conditions. Though served by a set of identical genes, the ways proximal–distal patterns are achieved for leg development and regeneration in Drosophila are distinct from each other (Bosch et al. 2010). An even more profound example is evident in planarian flatworms that have an almost unlimited ability to regenerate: adult tissue can redevelop into a whole new organism, as recent experiments have dramatically demonstrated (Wang et al. 2011). Scimone et al. (2011) have identified a set of genes that encode transcription factors required for planarian protonephridia regeneration. During this stage, stem cells are induced to form a group of cells within blastemas, expressing those genes used in excretory system formation. There is, therefore, a transcriptional regulatory mechanism that is necessary to allow regeneration to occur in as much as it activates the right genes and that provides the material means. While essential, it cannot be regarded as causally directing this tightly organized process. It is a matter of interpretation, of course, and some would prefer a gene-based explanation as it satisfies their own preconception of development. However, as has been explained in some detail in the previous section, the transcriptional control of genes explains next to nothing of the self-formation that proceeds subsequent to the production of proteins. Moreover, the manner in which regeneration often occurs along developmental routes not evident in embryogenesis is reflective of the variability of the self-organizing process in reaction to a new environmental situation. It suggests a deep principle that is not inextricably tied to any specific signaling pathway (Yoshida and Kaneko 2009).
Aside from regeneration in vivo, which has been characterized now for over a century, the self-organization of structures in vitro has become potentially more revealing and interesting. It certainly has profound implications in the study and application of biomedical engineering and technology. A notable observation of the autonomous formation of an optic cup structure, arising out of a culture of mouse embryonic stem cell aggregates (Eiraku et al. 2011), offers one very important insight. It builds upon previous research by the same team into the self-formation of cerebral cortical tissues, also in culture (Eiraku et al. 2008). Organogenesis requires the intricate orchestration of multiple cellular interactions to create collective cell relationships in the developing tissue. A self-formed retinal epithelium was observed to acquire distinct biochemical, mechanical and expression pattern properties in a domain-specific manner during the early stages of eye-cup morphogenesis. It entailed a complex process, consisting of a latent intrinsic order, with dynamic self-patterning driven by a sequential combination of local rules and internal forces emanating from within the epithelium itself.
Interestingly, retina tissue architecture in the culture emerged in a spatiotemporally regulated manner mimicking that of in vivo development, as if the behavior were remembered. This would detract from any notion that forces present in vivo alone are a major determining factor. The inference made is that a “self-organizing program” provides the necessary direction and synchronization, and which is itself not merely the causal outcome of gene expression and protein–protein interactions: localized parts harmonized in a very ordered and also highly coordinated course to develop the whole organ’s shape.
Signaling interactions between tissues are not as important as the developmental routines inherent within a particular tissue. Indeed, the entire process can be seen as progressing towards a seemingly pre-determined morphological goal. The expression of genes such as Pax6, Chx10 and Dkk1, involved in the control of key signaling pathways, was, as expected, an essential aspect. As a consequence, photoreceptor genes such as rhodopsin were properly expressed in the latter stages of the process. Other experiments have indeed shown that self-organization in epithelial tissues is driven by (E)-cadherin as well as actin expression (Chanson et al. 2011; Wei et al. 2007). But gene expression levels alone do not explain the specific sequential steps observed for which a “relaxation-expansion” model has been used to interpret the tissue dynamics (Eiraku et al. 2012). This model is based on three local changes in morphology, rigidity and growth at the cell and tissue levels. Flexible neural retinal epithelium, is driven by its own expansion leading it to buckle inside the shell of rigid retinal pigment epithelium, thereby causing a spontaneous invagination, although the exact control of this process is not known.
Self-organizing principles of development
Determining the precise contributing factors to self-organization is far from easy, as it almost certainly involves a diversity that are not all readily discernible. Instead of investigating any specific mechanisms involved, an appreciation for self-organization firstly requires an exploration of broad principles that can provide the foundation for future research. Moreover, as well as needing to be properly understood, self-organization should be distinguishable from models of genetic determinism where gene products are assumed to provide instructions that causally direct development. Over the course of the study of developmental biology four notable approaches, that include both genetic and non-genetic elements, have been established although there is considerable overlap between them. The unifying principle underlying all self-organization is the variation that governs the dynamic system by exploring the different regions in the state space until a particular configuration is reached.
Alexander Gurwitsch was one of the first developmental biologists to propose the hypothesis of “morphogenetic fields” with which to explain the emergence and nature of the wholeness of living organisms (Beloussov 1997). It was an integrative “top-down” model, in contrast to the reductionist, and “bottom-up” approach that now prevails within mainstream biology. It has since been accepted as a helpful construct that describes a collection of interacting cells out of which a structure is formed. The usage of the term “field” is deliberate, since it refers to non-localized spatial influence denoting both informational and regional relationships (Weiss 1939). This information exists not at the point which is under influence, but rather at more distant regions. Moreover, Goodwin (1985) postulates that living organisms more generally can best be modeled as holistic fields of organization rather than as complex assemblages of disparate parts: if a limb field was cut in half, and the two halves transplanted to different locations, each half would form a complete limb just as a magnet does when it is divided. This is also a concept similar to that of regeneration, and morphogenetic fields attract developing or regenerating systems towards a particular form or pattern. Gilbert et al. (1996) contend that these fields exemplify the modular nature of developing embryos, and are proposed to mediate between the genotype and the phenotype. Just as the cell, rather than its nuclear genome, is considered the major unit of organic structure, so the morphogenetic field may be a unit of ontogeny.
Morphogenetic fields represent chemical and physical determinants translated into spatial domains, the boundaries of which are presumed to be delineated by distributed transcription factors that collectively control numerous developmental pathways (Gilbert 2006). The familiar “chemical gradient”, which will be discussed in more detail later on, is actually a field model as it refers to changes in the level of a substance across a spatial domain as opposed to a single concentration level at a particular locality. A cell is defined by its position within its respective morphogenetic field, and responds to signals and environmental factors; these may be chemicals, like gene products or retinoic acid, but are likely to include more diverse factors. The fields, which may form a nested hierarchy, comprise dynamic entities replete with their own specialized developmental processes (Bolker 2000), and mathematical field-based models have been proposed to describe the physics of embryonic processes like cleavage (Goodwin and Trainor 1980). The self-regulating tendency of morphogenetic fields, that confer this degree of versatility and robustness, may possibly be explained by opposed transcriptional regulation located at opposite poles of the embryo (Reversade and De Robertis 2005).
But while an appropriate conceptualization of the process of morphogenesis, it is still only a way of abstracting unspecified self-organizing properties among cells that arise in sync with gene expression.
Formal morphogenetic field models that incorporate specific mechanisms, and which make testable predictions, are not common. This fact, along with the onset of the details provided by modern genetics, may have diminished some of the interest in field theory as a paradigm of ontogeny with many regarding it as too vague for practical and experimental purposes, but it has recently received more attention. Levin (2012a) proposes that morphogenetic fields reflect a higher level of organization, that is distinctly non-local in nature, and which conveys subtle pre-patterning and positional information. The field is characterized as essentially a conflation or synthesis of various biochemical, biomechanical and bioelectrical signals and forces. Particular emphasis is placed on the endogenous bioelectric communication between cells, and the properties they have for implementing morphogenetic fields by imposing patterning information on tissue during growth. This builds on the hypothesis of Shi and Borgens (1995) that electric fields generated by the embryo itself may serve as cues for morphogenesis. It has been demonstrated that membrane voltage gradients may, in fact, also provide a “bioelectric code” for gene expression, just as or even more instructive than chemical gradients (Pai et al. 2011). Bioelectrical signals may be determinants of shape during pre-patterning, where the templates of structures are set up, and observations of frog embryos indicate that fluxes of charged particles may function by precisely outlining the areas that will become particular facial features (Vandenberg et al. 2011). As such, any empirical research into the nature of morphogenetic fields may focus on the physical reality of electric fields, derived from membranal potential levels that communicate spatial coordinates in the developing embryo and so regulate sequences of events.
Zygotic structural patterning
An alternative account of formation is based on the premise that the body is a mosaic enlargement of self-organizing patterns engrained within the physical outline of the egg itself. This touches upon now obsolete notions of preformationism—namely that biological form arises from out of the patterned structure of a primordial germplasm, and that morphogenesis is controlled mainly by these epigenetic cytoplasmic determinants. These material sub-structures, first described by Weissmann (1892), do not contain the actual form of the adult organism, but may provide a template that gives rise to it and shapes tissues during development. Also, changes in cell shape, such as that affecting cytoskeletal tensional homeostasis, can induce gene expression (Rosette and Karin 1995; Lavagnino and Arnoczky 2005). Cell shape and cytoskeletal structure are coupled to tissue growth, and cells likely use mechanical cues, like changes in cytoskeletal tension, and non-linear dynamics that affect cell shape, to control the activation of genes and signal pathways (Ingber 2005; Bizzarri et al. 2013a).
Wells (2011) argues that, although developing embryos require the precise spatial deployment of specific cellular functions, this spatial information is not itself provided through the expression of maternal effect genes: this geometry in fact determines the initial shape of any zygotic gene expression (Schiffmann 2012). Rather, positional coordinates are themselves dependent on the prior establishment of an anteroposterior body axis by antecedent asymmetries that are, in turn, derived from the centrosome and cortex. Epigenetic spatial information in the embryo could be traced back to asymmetrically distributed molecules deposited in the egg during its production in the ovary. Therefore, gene regulatory networks require initial inputs that are themselves not derived and inherited from DNA sequences. Gene expression thus occurs within an already-established floor plan.
A similar observation is made in the initiation of the formation of the two main axes of the embryo due to internal anisotropies built into the egg itself (Davidson 2006). According to Poyton (1983), although the proteins used in a membrane pattern are themselves encoded in DNA, the membrane pattern itself pre-exists their synthesis. Moreover, sperm entry position may provide a cue for spatial patterning of the embryo by specifying the axes of the emerging blastocyst (Piotrowska and Zernicka-Goetz 2001). The main problem, of course, with this approach is that it sets up a causal regress that proves to be unsatisfactory. This is because the forces that built the ovary and testis, which in turn produced the egg and sperm, require a proper explanation themselves. But the ultimate cause remains unknown and describing development based on these determinants must account for their own origin.
Pivar (2011) offers a highly speculative membrane-toroid model involving topological distortions that show how the embryo is affected by complex physical forces to generate various morphological structures. It does not seem to attribute anything of consequence concerning the emergence of form to GRNs. Central to this concept is that egg cells are internally pressurized and inflated spherical membranes, and the embryo is formed in the sudden collapse of the equilibrium between the pressurized contents of the cell and the ability of the elastic membrane to contain it. It does not actually describe what is observed in embryogenesis, but rather provides a conceptual framework that seeks to illustrate the historical evolution of form in contrast to a gene-centric approach (Pivar 2009).
The reaction–diffusion theory of self-organizing formation, first proposed by Alan Turing (1952) in his seminal work, remains the standard exposition of the chemical basis of morphogenesis. Turing argued that pattern formation is possible for two interacting substances that diffuse with different rates, and where a small random fluctuation can become amplified due to positive feedback kept localized, and to long-ranging inhibition by way of negative feedback. Mostly a mathematical model, it continues to exert influence in developmental biology (Maini et al. 2002; Kondo and Miura 2010). Turing emphasized the role of chemical “morphogens” as substances, almost all now identified as signaling proteins, secreted by one group of cells that caused a specific change in the growth and differentiation of another group of cells. These are the diffusible factors that govern tissue morphology since cells respond to them in a concentration-dependent manner. Diffusion creates spatial inhomogeneities in reacting chemicals which is then translated into patterns. The gradients of these morphogen concentrations determine gene expression according to certain threshold levels (Wartlick et al. 2009); genes still control the production of morphogens, and so influence anatomical form, but patterning occurs because of differential diffusivity and accumulative chemical reactions (Elsner and Tsonis 1989; Müller et al. 2012). It is also possible that mechanical processes, such as those evident in motor proteins, if coupled with chemical ones, can have a scalar effect on shape (Howard et al. 2011). The “activator-inhibitor” model of Meinhardt and Gierer (2000), first proposed in the 1970s, has elaborated on Turing’s idea though with more emphasis on the chemistry. Local activation of a chemical process is produced by autocatalysis, is self-enhancing leading to further production, and also activates an inhibitor that disrupts this autocatalytic process. As the two substances diffuse through the system at different rates, the inhibitor migrates faster thereby localizing the activation and preventing catalysis spreading until a stable pattern is formed. However, lateral inhibition is insufficient for pattern formation, and may require non-linear self-enhancement so that it can self-regulate. Clearly, the principles of reaction–diffusion have been established even if many of the particularities have not been established. Problems remain in using this approach to account for the patterning of specific structures, although recent evidence has implicated Turing’s model in pentadactyl formation (Sheth et al. 2012; Vogel 2012) where the dose of distal Hox genes modulates the digit period or wavelength. It is speculated that a modification to this may explain the emergence of tetrapod digits.
Attempts to suggest diffusing chemicals as positioning markers within the developing embryo have, however, been less convincing. Unlike in Turing’s model, where reaction and diffusion determine pattern formation, the alternative hypothesis of Wolpert (1969) proposes that the concentration gradient of a particular morphogen “instructs” cells of their exact location in spite of the fact that they all possess identical genomes. Cells are thus differentiated according to their position relative to the source of the morphogen. This model requires an exact and direct correlation between inputs and outputs, namely the gradient and tissue response to it. However, although the nature of any positional signaling and cues may now be understood, the molecular basis for specifying exact positional value and spatial coordinates is deeply problematic (Wolpert 1989) because there are clearly stochastic variables involved in biochemical environments that can serve to make diffusion both unreliable and messy as a process (Kerszberg and Wolpert 2007). There also mechanistic problems regarding how any positional gradient based on signaling is interpreted by the cell and so regulates differential gene expression in a concentration-dependent manner (Wolpert 2010). Morphogens tend to be very dilute and imprecise, and so it is very unlikely that concentration gradients alone determine the fate of cells. Conversely, Niehrs (2010) has suggested that morphogen gradients of perpendicular Wnt and BMP signaling, distributed along the respective anteroposterior and dorsoventral axes in invertebrates, may confer a precise Cartesian coordinate system of positional information and accurately specify cell polarity. The central idea behind this is that the bodily axes correspond with x and y axes, and that the concentration levels of both signaling proteins at any point can determine the position from the origin. It should be recognized, however, that the hypothesis does not itself account for the deployment of the signaling proteins along the axes that would make this mechanism possible if in fact it does work. More generally, because many expression patterns depend largely on previous expression patterns, there results a regressive problem of explaining what caused the prior differences in gene expression. Explaining differential expression in terms of differential diffusion appears to be a circular argument.
Dynamical patterning modules
A similar concept to that of the morphogenetic field is the dynamical patterning module (DPM) (Newman and Bhat 2008; Newman 2012). It is presented as a “physicalist” approach that builds upon the more established “chemicalist” one of reaction–diffusion theory in that it recognizes that cells can affect each other by secreting diffusible signals and by coupling (Salazar-Ciudad et al. 2003). It differs in as much as it postulates that, while morphogens may help explain pattern formation, they are unable to account for the forces determining how groups of cells move, how tissues elongate, bend or undergo all kinds of vibrations (Lecuit 2008). Rather than being encoded in genes, form emerges when a cluster of cells, and certain molecules, mobilize physical forces, effects, and processes within a multicellular context (Newman and Linde-Medina 2013); these in turn generate a particular aspect in the cluster’s form or pattern. DPMs are thus multicellular determinants that mediate such effects as cell adhesion and phase separation of differentially adhesive cell groups, and the generation of structural anisotropy across cells with the system properties of an intercellular microenvironment. This is analogous, at the level of cells and tissues, to the physical behavior of condensed materials. The central idea behind DPMs owes less to transcriptional control and more to the interaction of those gene products involved in the mechanico-chemical specifics of cell aggregation and the physics set in motion by them. These processes may include cohesion, viscosity, elasticity, diffusion and oscillation that are all necessary for the formation of cell aggregates. Genes specify some of the key components (i.e., RNAs, proteins) that participate in the complex physico-chemical reaction networks and transitions that occur during embryogenesis. Acting individually, or in association, DPMs constitute a “pattern language” that have been responsible for the origination of metazoan body plans throughout the course of evolutionary history. As a quintessentially structuralist account of the generation of biological form, it is predicated on the inherent self-organizing plasticity and excitability of material systems: the elongation of tissues by convergent extension is known to affect organogenesis (Keller 2006). DPMs may have particular relevance in relation to the popular “clock-and-wavefront” model of pattern formation (Cooke and Zeeman 1976; Kaern et al. 2004). This involves the progressive arrest of a periodic oscillation at a moving boundary along the body axis. It has been proposed for the sequential formation of somites and segmental patterning (Baker et al. 2006), although new research indicates that the “clock” might not be needed, due to localized self-organizing interactions (Dias et al. 2014), for which the physics of the DPM model could help explain.
It is obvious that the control of protein synthesis does play a substantive role in morphogenesis, and that gene regulation is required for the development of complex multicellular organisms. Transcription factors in a tissue layer activate the expression of growth factors and other genes in response to signaling produced from adjacent tissue, thus forming a signaling network replete with intricate feedback loops that regulate development (Zhang et al. 2005). But while instrumental in gene regulation and expression, DTFs do not explain any patterns of organization. Gene expression by itself does not determine why particular structures form as they normally do even if it explains how the formative process is initiated and governed throughout the course of development. In this respect, gene networks are a necessary but not sufficient cause. Developmental switches and signals without question have an integral function, but if the goal is to capture the downstream causes of systemic properties, the gene is the wrong level of organization. The genome contains a set of instructions for enabling and disabling the synthesis of certain proteins, over certain periods of time, and when particular concentrations of signaling chemicals are reached. What it does not contain is any spatial information related to the organizing of cells into functional structures (Harold 2005; Noble 2013).
This is often overlooked by developmental biologists, possibly due to popular misconceptions about the causative power of DNA, but also due to the practical appeal of gene centrism. Researchers have tended to focus on the GRNs only because they are eminently easier to study, and since the right approach and metaphor to understand the generation of order is not readily available. While current knowledge of the architecture of gene regulatory networks is still limited, and the efforts of the ENCODE consortium have a long way to go in uncovering all of the details, recent evidence does not indicate that these hold the key to understanding embryonic development. GRNs do not specify the 3-dimensional geometry of anatomy, even if changes in them can disrupt it, and therefore it is necessary to understand the rules at work in large-scale development. Gene expression in a cell population is an intrinsically stochastic process (Elowitz et al. 2002); yet biophysical cues drive the system toward a coherent and stable behavior, eventually leading to a specific form. The new field of “evo-devo” has relied too much upon the assumption that certain changes in regulatory DNA, encompassing the idea of “molecular tinkering” (Jacob 1977), can readily explain the diversity of morphological themes and body plans. But differential transcription alone seems unlikely to explain the profound anatomical diversity among phyla. Gene expression profiles are observed to be very similar between humans and mouse tissues despite the obvious difference in the size and anatomy of both species (Dowell 2011).
Development is itself fundamentally a cellular process whereby cells divide, elongate, migrate, deform, aggregate etc. Cells bring about changes in embryonic form by generating patterned forces, and by differentiating the tissue’s mechanical properties that harness these forces in specific ways (Keller et al. 2003). Unlike chemical signaling, cell mechanics can control tissue-scale structures by changing the topology and structure of the environment (Guo 2013). This can also facilitate coordination of cell morphology through interactions with the structure of the underlying substrate. Shape-dependent control of cell growth and function appears to be mediated by tension-dependent changes in the cytoskeleton (Huang and Ingber 2000). While gene products do, of course, influence all of these different mechanisms as prime facilitators they are not themselves the principal causes.
The dynamic control and organization of physico-chemical forces, however, can provide a proper context with which to investigate the cause of ontogeny. Variability within the process of self-organization, itself influenced by changes to non-genetic factors, can begin to address this. Emergent phenomena (like shape) may be extremely sensitive to small differences and micro-environmental factors at the lowest level, quite possibly at the quantum level itself. However, this also raises the pertinent question as to where this patterning information actually exists. All manner of modes have been suggested, from physical forces (magnetic fields, pressure, tension/stress, ion fluxes and even gravity) to general features of complex system behavior and artificial intelligence. What is clear from this evidence is that there exists an all-important middle layer that cannot be viewed just as the consequential by-product of the regulation of gene expression. Some have proposed that this “black box”, a combination of epigenetic processes such as embryonic inductions, tissue interactions, and various other bio-signals, may collectively represent the bridge between genes and form (Hall 2003).
Hans Driesch was among the first biologists to realize that there was a propensity for each of the cells in the developing zygote to become a holistic entity in itself. Driesch (1908) carried out experiments on sea urchin embryos and demonstrated that even if several cells are destroyed at a very early stage, just one could still proceed towards a seemingly predefined morphological goal, namely the development of the whole organism. Therefore, if parts of the system were removed, and the normal course of ontogeny disturbed, the main developmental process was sustained. Driesch called the force behind the phenomenon he observed “entelechy”, a word termed by Aristotle that connotes the existence of a goal. He described it as a non-spatial, and non-energetic, causal agency that organized and controlled physico-chemical processes during morphogenesis, largely by affecting their timing. Driesch regarded entelechy as something with an inherent wholeness that acted on the living system but was not a material part of it. It appears to be similar to what Darwin (1875) referred to as “the coordinating power of the organization” which brought all parts into harmony with each one another. But the notion that there may be non-material forces of organization at work has been dismissed and condemned as invoking vitalist beliefs that posit the presence of a mysterious life force. The field of developmental biology is now focused on strictly molecular mechanisms as a point of methodological enquiry. But it is unclear if this approach can provide valid explanations and not just add more details.
Like the “attractors” advanced by Thom (1972), Waddington believed that chreodes could be explained in terms of the concentrations of chemicals and their stabilization. There is evidence that suggests that negative feedback, an important but not essential characteristic of self-organizing systems, involving the syn-expression of inhibitors in gene signaling cascades, may account for this apparent channeling (Paulsen et al. 2011). But embryonic development has also been observed to adjust its morphogenetic processes to counteract perturbations that are not caused by chemicals. Induced disruptions to development in tadpoles can be corrected by the embryo in a manner quite unlike any normal movements (Vandenberg et al. 2012). This demonstrates that there are no fixed set of movements, but rather there is a flexible process that can recognize any deviations and perform appropriate actions to correct this. These self-monitoring and adaptive decision-making capabilities of tissues detract from the notion that development is really the product of an encoded genetic program.
It is now broadly accepted that embryos and organisms should be treated as fundamentally self-organizing systems (Wennekamp et al. 2013) replete with hierarchical levels of complexity (Bizzarri et al. 2013b). Whereas chemistry is concerned with substances and how they react, biology appeals to much deeper concepts such as information and organization (Davies 2013). The nature of genetic information in DNA is only one aspect of this. A common objection made against “self-organization” is that it is mostly phenomenological, rather than mechanistic, and so lacks sufficient explanatory power. Some may even be inclined to regard talk of self-organization as recourse to magic. But this is only because the precise mechanisms and causal factors remain unclear, or unknown, even if many general principles of self-organization are well established. One of the main problems associated with purely molecular explanations of developmental evolution lies in understanding how patterning and development are coordinated and synchronized when nothing relating to this really exists in DNA sequences. The same can be said, more generally, for the intense choreography evident in cell division, or the remarkable orchestration now observed in human brain development (Pletikos et al. 2013), and in the self-organization of axial polarity during neocorticogenesis (Kadoshima et al. 2013).
Too much credit has so far been given to diffusing chemicals and their purported role in determining cell identity. But even though models and principles of self-organization can offer more insight into the developmental process and its evolution than can gene expression levels alone, they do not necessarily address the elusive origin of organismic complexity and the architecture of the body. Some would propose that the fundamental outline of the body plan is embedded within the patterns of the egg itself, but this tends to be limited to explaining the polarity and topology of the embryo. Practically, it is perhaps more sensible to isolate those forces at work in specific instances of morphogenesis which can then be comprehensively examined. As well as experimentation, this may involve drawing up models that relate to localized patterns of development (Meinhardt 2012). Some of the examples of regeneration and autonomous formation, as discussed above, can also become avenues for future investigation along with novel research into bioelectric-induced patterning that has hitherto not been given due attention (Levin 2012b). These focus on inherently self-organizing factors that work in conjunction with protein functions to produce the features evident in biological form.