Criar uma Loja Virtual Grátis

Taxas morfológicas

Taxas morfológicas

Skip to main contentSkip to article


Morphological homeostasis in the fossil record

Under a Creative Commons license
open access


Morphological homeostasis limits the extent to which genetic and/or environmental variation is translated into phenotypic variation, providing generation-to-generation fitness advantage under a stabilizing selection regime. Depending on its lability, morphological homeostasis might also have a longer-term impact on evolution by restricting the variation—and thus the response to directional selection—of a trait. The fossil record offers an inviting opportunity to investigate whether and how morphological homeostasis constrained trait evolution in lineages or clades on long timescales (thousands to millions of years) that are not accessible to neontological studies. Fossils can also reveal insight into the nature of primitive developmental systems that might not be predictable from the study of modern organisms. The ability to study morphological homeostasis in fossils is strongly limited by taphonomic processes that can destroy, blur, or distort the original biological signal: genetic data are unavailable; phenotypic data can be modified by tectonic or compaction-related deformation; time-averaging limits temporal resolution; and environmental variation is hard to study and impossible to control. As a result of these processes, neither allelic sensitivity (and thus genetic canalization) nor macroenvironmental sensitivity (and thus environmental canalization) can be unambiguously assessed in the fossil record. However, homeorhesis—robustness against microenvironmental variation (developmental noise)—can be assessed in ancient developmental systems by measuring the level of fluctuating asymmetry (FA) in a nominally symmetric trait. This requires the analysis of multiple, minimally time-averaged samples of exquisite preservational quality. Studies of FA in fossils stand to make valuable contributions to our understanding of the deep-time significance of homeorhesis. Few empirical studies have been conducted to date, and future paleontological research focusing on how homeorhesis relates to evolutionary rate (including stasis), species survivorship, and purported macroevolutionary trends in evolvability would reap high reward.


Evolutionary constraint

1. Introduction

Developmental systems mediate the construction of phenotype from genotype, often with environmental input. The range of phenotypes expressed by a species is therefore determined by both genetic and environmental variation, and by any interaction between the two [1]. Developmental systems can limit the extent to which stochastic accidents during development, allelic variation, and/or environmental variation translate into phenotypic variation—a capability known as morphological homeostasis [[2][3][4][5]]. The degree of expressed phenotypic variation in organisms, populations, and species that exhibit morphological homeostasis is therefore reduced relative to what would have been produced in the absence of such a property. Mechanisms of morphological homeostasis are thought to evolve under stabilizing selection, when deviant phenotypes are at a selective disadvantage (Section 2.4). Under such a regime the short-term, generation-to-generation fitness advantage of morphological homeostasis is clear, but it is also possible that morphological homeostasis can have a longer-term impact on evolution. Given that phenotypic variation is the raw material upon which natural selection operates, morphological homeostasis has potential to serve as an evolutionary constraint: stronger morphological homeostasis yields less phenotypic variation and thus a weaker response to selection [4,[6][7][8][9]]. Robustness against allelic variation (genetic canalization; Section 2.1) can also permit genetic variation to accrue by shielding it from scrutiny by natural selection. Such cryptic genetic variation could subsequently become exposed—potentially causing a spontaneous and dramatic increase in the range of phenotypes subject to selection—if the homeostatic capability of the developmental system is ever reduced (genetic decanalization or genetic potentiation; [4,10,[11][12][13][14][15],8,[16][17][18][19]]).

Empirical studies of morphological homeostasis have been conducted almost exclusively by neontologists (i.e., researchers studying the modern-day biota). This is not surprising, because unequivocal identification of genetic or environmental canalization requires experimental manipulation of organisms and knowledge of genotype (Section 2). However, such studies are necessarily conducted over relatively short timescales (typically tens of generations) and are thus limited in their ability to scale up from the microevolutionary to macroevolutionary scale. Whether and how morphological homeostasis might have influenced the longer-term evolutionary trajectory of a lineage or the pattern of morphological diversification (disparification) of a clade can then be only indirectly inferred using phylogenetic comparative methods.

The fossil record provides documentation of phenotypic evolution through deep time. The types and resolution of data attainable from fossils differ from or only partially overlap with those attainable from the study of modern organisms (Section 3.1), but if neontological approaches to studying morphological homeostasis can be extended to the fossil record then many intriguing questions could potentially be empirically addressed (Section 4; see also [10,20,21]): How labile is morphological homeostasis? (Or alternatively, for how long can a given pattern of morphological homeostasis constrain phenotypic evolution in a lineage or clade?) Does morphological homeostasis relate to survivorship through ancient examples of environmental change? And if so, can we use lessons from the fossil record to predict the response of extant species to present-day climate change? Are there macroevolutionary trends in morphological homeostasis that are only detectable by examining the fossil record (because extant species might be highly derived relative to extinct basal members of the clades to which they belong)?

This paper explores the potential for a meaningful integration of neontology and paleontology into a deep-time perspective on the evolutionary significance of morphological homeostasis. “Meaningful integration” here refers to the necessity of applying the same terms in the same ways: if the definition of and criteria for identifying morphological homeostasis differ between neontology and paleontology, then any attempt to assess morphological homeostasis across these scales is compromised. First (Section 2), current neontological concepts of morphological homeostasis are summarized. Section 3 highlights differences between paleontological and neontological data, and appraises whether and how morphological homeostasis could be assessed in the fossil record. Finally, Section 4 reviews particular questions pertaining to the evolutionary significance of morphological homeostasis (above) and highlights the contributions that paleontological data can make to answering them. Due to space limitation this paper focuses only on morphological homeostasis in single traits and does not consider integration among traits, despite links between the two phenomena (e.g., [[22][23][24]]).

2. Morphological homeostasis

Morphological homeostasis is a general term referring to the ability of a developmental system to adjust itself to variable conditions [5]. A more nuanced and informative terminology has been developed based on the particular source of variation being buffered against (Figs. 1 and 2 ; [4,5,8,[25][26][27][28][29]]). Such refinement provides a useful conceptual framework for designing research and interpreting data. A summary of the refined terminology is presented below (based on [5]; it is acknowledged that some workers adopt other terminological schemes, sometimes employing the same terms but with different meaning). The types of data required to unambiguously identify buffering against each source of variation are emphasized; this permits critical evaluation of the limits of paleontological data to the study of morphological homeostasis (Sections 3 and 4). Potential relationships among the various buffering processes are discussed in Section 2.4.

Fig. 1

Fig. 1. Among-individual phenotypic variation resulting from variation in genotype and (macro)environment. Top row: Changes in among-individual phenotypic variation resulting from change in allelic sensitivity (genetic canalization; center and right plots) and in allelic diversity (center and left plots). Genotype axis refers to allelic variation at a locus. All individuals raised in identical environment. Slope and shape of relationship (here depicted as linear for simplicity) in each plot represent sensitivity of phenotype to allelic variation. Decreased among-individual phenotypic variation, potentially constraining evolutionary response of trait to directional selection, can result either from loss of allelic diversity (center to left plot) or from decreased sensitivity to allelic variation (center to right plot); only the latter represents morphological homeostasisBottom row: Changes in among-individual phenotypic variation resulting from change in environmental sensitivity (environmental canalization; center and right plots) and in environmental diversity (center and left plots). All individuals share identical genotype. Slope and shape of relationship (here depicted as linear for simplicity) in each plot represent nature of phenotype-environment association (reaction norm, phenotypic plasticity). Decreased among-individual phenotypic variation, potentially constraining evolutionary response of trait to directional selection, can result either from exposure of genotype to fewer macroenvironments (center to left plot) or from decreased sensitivity to macroenvironmental variation (center to right plot); only the latter represents morphological homeostasis.

Fig. 2

Fig. 2. Change in phenotypic variation resulting from change in developmental stability (homeorhesis). Top row: Increased developmental stability is evidenced by closer match between observed phenotype and target phenotype. Target phenotype is predicted phenotype for a given genotype in a given environment. Degree to which an individual achieves its target phenotype reflects ability of its developmental system to buffer against stochastic perturbations during development; imperfect buffering results in developmental instability (developmental noise; i.e., deviation off line of slope = 1, shown). Middle row: Left and right sides of an individual have same genotype and experience same (macro) environmental conditions during development. For a nominally bilaterally symmetrical structure, random difference between phenotypic value on left side and on right side therefore represents developmental instability (assuming no wear-and-tear or environmentally induced asymmetry; [5,63]). Target phenotype is symmetrical, so increased developmental stability is evidenced by closer match between trait value on left side and right side (less deviation off line of slope = 1, shown). Bottom row: Deviations from within-individual symmetry for a trait, represented as a frequency distribution. Within-individual asymmetry (abscissa) is difference between trait value on right and left sides; values may be standardized so as to be independent of trait size and thus be more broadly comparable across individuals or samples [63]. If (size-standardized) values of within-individual asymmetry for a sample are normally distributed with mean of zero, then sample exhibits ideal fluctuating asymmetry (FA). Degree of FA (measure of dispersion within sample; see [63] for various metrics) is proxy for developmental instability.

2.1. Genetic canalization

Genetic canalization is a process by which the sensitivity of a trait to allelic variation is reduced (Fig. 1, top row, middle to right plot; [4,5,25]). This can occur through, for example, the evolution of epistatic interactions among genes [4]. Genetic canalization is identified when two populations, each sharing the same allelic diversity for a given trait and reared in an identical environment, differ in phenotypic variation for that trait. The population with the lower phenotypic variation is less sensitive to the alleles that control the genesis of that trait, and is thus more strongly genetically canalized. The requirement that the number of alleles involved is conserved across the populations is critical: a between-population difference in phenotypic variation that results from a difference in the number of alleles controlling a trait, rather than a difference in buffering against the effect of particular alleles, does not represent genetic canalization (Fig. 1; top row, middle and left plots). Demonstration of genetic canalization therefore requires knowledge of the genotypes of the individuals involved [30].

2.2. Environmental canalization

When reared in different environments, the same genotype can produce different phenotypes. Such phenotypic plasticity is countered by environmental canalization—a process by which the sensitivity of a trait to macroenvironmental variation is reduced (Fig. 1, bottom row, middle to right plot; [5]). Phenotypic plasticity can be trait- and genotype-specific (e.g., [4,7,[31][32][33][34][35][36][37],19]). Although some general mechanistic principles are becoming clearer [19,33,34,38,39], the detailed genetics underlying plasticity are in many cases poorly understood and probably varied, and it is generally unknown whether mechanisms controlling plasticity also control environmental canalization.

Identification of environmental canalization requires knowledge of genotype, because the form of the reaction norm can only be estimated from multiple measurements of the same genotype (or related individuals with known pedigree) taken in different environments [32,33,[40][41][42][43][44]]. Translocation or common garden experiments can assist in determining whether an observed phenotype-environment association results from phenotypic plasticity or from local genetic adaptation (e.g., [45,37,46]). If two genotypes exhibit different degrees of plasticity when reared across the same range of environments, then the genotype with the least environmentally sensitive phenotype is the more strongly environmentally canalized.

2.3. Homeorhesis (developmental homeostasis)

Ideally, a given genotype reared a given environment would consistently produce a single phenotype (the target phenotype). Such perfect within-environment genotype-to-phenotype mapping does not occur in the real world, however, because developmental processes are subject to stochastic perturbations that introduce some instability to a developmental trajectory (e.g., [27,[47][48][49],50]). This individual-scale microenvironmental variation [51,52] results in developmental noise, which can be envisaged as deviation from the target phenotype (Fig. 2). Developmental systems can be buffered against such stochastic perturbations—a phenomenon known as homeorhesis or developmental homeostasis [5]. By reducing the impact of stochastic perturbations, mechanisms of homeorhesis increase developmental stability; that is, they result in a phenotype that more closely matches the target phenotype for that particular genotype within that particular (macro)environment (Fig. 2; [26,27,47,53,53,30,53]).

Developmental instability (noise) therefore represents the phenotypic expression of stochastic perturbations to development that were not buffered by mechanisms of homeorhesis. Developmental instability can be measured as either (1) the variation among genetic clones or isogenic lines when those clones or lines are reared under identical environmental conditions (e.g., [51,52,54,55]), or (2) fluctuating asymmetry (FA; random deviations from symmetry of a symmetric structure; Fig. 2; [24,56,57]). Only the latter option is available to paleontologists (see Section 3), so this review will focus on that.

FA can be used to assess deviation from the target phenotype (and thus as a proxy for developmental instability) because the target phenotype for any individual—symmetry of the trait—is known (Fig. 2). For a bilaterally symmetric structure, the left and right sides of an individual can usually be assumed to be genetically identical and to have experienced a similar (macro)environment, so the only source of variation that differs between them is microenvironmental in nature. The left-right asymmetry of a nominally bilaterally symmetric trait for an individual is therefore the net result of stochastic developmental perturbations (increasing the level of asymmetry) countered by the buffering capability of its developmental system against such perturbations (decreasing the level of asymmetry) [5,8]. This link between FA and homeorhesis has long been known (e.g., [[57][58][59]]) and has been exploited in countless studies; the relationship between the two has been the subject of numerous books and extensive reviews to which the reader is referred (e.g., [10,15,24,60,61]). Methods for identifying and quantifying FA—including standardized measures of relative FA that are independent of trait size and thus more defensibly comparable across samples—are well developed [24,[62][63][64][65][66]] and need not be discussed herein.

If two populations differ in (relative) FA for a given trait, then it is commonly inferred that the population with higher (relative) FA consists of individuals whose developmental systems are, on average, less effective at reducing microenvironmental variation for that trait. However, some studies have found that (macro)environmental stress negatively impacts the capacity of a given developmental system to buffer against developmental noise for a particular trait, in which case homeorhesis for that trait could be modeled as a reaction norm [41]. Among-population differences in (relative) FA would then be context-specific, because they would depend not only on the relative buffering capacities of the development systems represented but also on the particular (macro)environments to which those systems were exposed.

It is difficult to draw general principles regarding the expected behavior of FA in the face of genetic or (macro) environmental variation, especially because for the vast majority of traits studied to date the mechanistic controls that determine the level of FA are unclear. Reviews indicate that the level of FA is not always a consistent indicator of (macro)environmental stress; nor does FA consistently relate to genetic parameters such as degree of heterozygosity or genetic quality of individuals [24]. The level of FA (and, by inference, developmental instability) is therefore most likely a trait-, population-, and taxon-specific property [15,24,31,60,[67][68][69][70]]. Comparison of (relative) FA across samples, and interpretation of any differences in terms of homeorhesis, must be done with caution.

2.4. Morphological homeostasis and phenotypic evolution

Genetic canalization, environmental canalization, and mechanisms of homeorhesis reduce the sensitivity of the phenotype to allelic variation, environmental variation, and stochastic developmental perturbations, respectively. In situations where phenotypic deviants are at a selective disadvantage within and/or across particular environments—i.e., when a population is subject to stabilizing selection—one or more of these processes might therefore be expected to evolve ([3,4,10,25,51,59,[71][72][73][74][75],7,11,17,52,[76][77][78][79][80][81][82]]; although canalization or homeorhesis of a given trait under stabilizing selection can be limited by factors such as mutation-selection balance and fitness trade-offs between integrated traits, e.g., [4,22,83,84]). Conversely, in the face of changing or fluctuating environmental conditions, the capacity of a developmental system to produce phenotypic deviants might be a selective advantage. Processes resulting in morphological homeostasis might reduce fitness and thus might not be favored by selection. In such situations, properties such as phenotypic plasticity can be adaptive (e.g., [19,[38][39][40],85]). A relaxation of stabilizing selection within and/or across particular environments, a switch to directional selection, or a breakdown of the buffering capacity of a developmental system under stress can lead to increased phenotypic variation resulting from genetic potentiation (genetic decanalization), an increase in phenotypic plasticity (environmental decanalization), and/or greater developmental instability (Figs. 1 and 2; [4,7,8,12,[14][15][16][17],37,59,79,86]).

Genetic and environmental canalization can limit the phenotypic variation of a trait among individuals within and across particular (macro)environments, respectively (Fig. 1), with obvious potential relevance to evolution: a less variable trait has less capacity to respond to directional selection. But whether and how homeorhesis relates to phenotypic evolution is less clear. Selection can alter developmental (in)stability (e.g., [82,87], and references therein; Fig. 2), but does the magnitude of developmental (in)stability of a trait affect the capacity of that trait to evolve (i.e., its evolvability; [88])? Several observations offer provisional support for a link between developmental (in) stability and trait evolution. First, developmental instability can account for a non-trivial proportion of phenotypic variation [79,89,90] and so might be “visible” to natural selection. Second, some among-population or between-environment comparisons have found that the level of within-individual variation (FA) for a trait positively correlates with the among-individual variation of that trait (e.g., [91,92] [but see critique by [93,[93][94][95][96][97]]). Third, the ability of an individual to achieve the target phenotype is a component of fitness [79], and modeling has shown that developmental noise can increase the fitness of a genotype in the face of environmental variation by broadening its environmental tolerance [98] thus suggesting that reduced homeorhesis might be favored in changing environments. However, some studies have found that directional selection on a trait does not produce a consistent change in developmental instability (e.g., [86,99]), and others have found either no relationship or even a negative relationship between the level of within-individual variation (FA) for a trait and the among-individual variation of that trait [7,53,80,93,100]. It therefore seems that a relationship between homeorhesis and among-individual variation is trait- and taxon-specific [24]. The relevance of homeorhesis to phenotypic evolution should be assessed on a case-by-case basis.

Is there an association between processes buffering against the different sources of variation? In cases where within-individual variation (FA) positively correlates with among-individual phenotypic variation, is it conceivable that the same (or related) mechanisms control buffering against microenvironmental variation, macroenvironmental variation, and genetic variation [92]? Unraveling the potential mechanisms by which genetic canalization, environmental canalization, and homeorhesis are achieved is an area of active research. Oft-cited mechanisms include genetic modifiers that mask the expression of alleles [25]; genetic redundancy (e.g., by gene duplication; [27,28]); molecular chaperones such as heat-shock proteins that regulate protein folding and correct for genetic or stress-induced errors ([7,21,[101][102][103][104][105][106][107]]); and microRNAs that can regulate gene expression [[108][109][110]]. A full discussion of such mechanisms is beyond the scope of the present paper; the reader is instead referred to several summaries (e.g., [15,17,19,24,[32][33][34],60,77,111]).

It has been argued that genetic canalization might be an indirect result of selection for environmental canalization [76,80], although the two need not always be tightly coupled because experimental data indicate that environmental canalization can be modified without affecting genetic canalization [97]. There is also mixed empirical support for an association between homeorhesis and either genetic or environmental canalization, and any such relationship is probably trait-specific [8,26,27,29,52,54,55,73,79,80,[93][94][95],97,[112][113][114]]. This means that a complete exploration of how the evolution of a trait might have been influenced by mechanisms of morphological homeostasis (genetic canalization, environmental canalization, or homeorhesis) should involve testing for robustness to all sources of variation: rejection of a hypothesis that the trait evolution was influenced by homeorhesis, for example, is interesting in its own right but does not necessarily mean that the trait was not influenced by genetic or environmental canalization.

3. The challenges of demonstrating morphological homeostasis in the fossil record

Unfortunately, paleontological use of morphological homeostasis-related terms has often been incommensurate with how those terms are applied by neontologists (Section 2). For example, paleontologists have used the term “canalization” to describe a macroevolutionary trend towards restricted evolution—usually a trend from early lability to subsequent fixation of morphological traits in clades. This is often interpreted in terms of a lineage settling into an adaptive zone with subsequent progressive fine-tuning of its genotype or phenotype. The taxonomic level at which such a trend has been assessed varies widely: in some cases differences in the level of intraspecific variation were examined (e.g., a progressive decrease in variation of thoracic segment number within and among trilobite species through time; Section 4.3); in other cases the focus was on macroevolutionary trends in interspecific disparity (e.g., head appendage disparity among arthropods through time; [115]). But in all cases the source(s) of the variation/disparity and the mechanism(s) by which that variation/disparity was reduced (e.g., loss of genetic or environmental diversity, or decreased allelic or environmental sensitivity) cannot be unambiguously identified (Fig. 1). The convergent evolution of similar phenotypes has also been referred to as “canalization” in the fossil record [116], apparently because the homoplasies (often involving independent reversals to plesiomorphic character states) indicated a limitation on phenotypic disparification. Such a pattern might indicate character state exhaustion [11], but is not equivalent to canalization as currently understood (Sections 2.12.2).

The term “plasticity” has been used in the paleontological literature to describe intraspecific variation (e.g., [[117][118][119]]). However, intraspecific variation does not necessarily result from phenotypic plasticity (in the sense of a reaction norm)—it could result from genetic variation among individuals—and conflation of the two should be avoided. Similarly, phylogenetic lability in developmental processes, with attendant high disparity of resulting phenotypes, has sometimes been referred to as “developmental plasticity” (e.g., [120]), but this does not necessarily mean that a given developmental system was environmentally sensitive or was weakly canalized. Phenotypic variation should only be described as plastic when it is a reaction norm to environmental variation (Section 2.2).

Stringent requirements must be met in order to demonstrate the presence, type, and level of morphological homeostasis (Section 2). This section discusses whether and how paleontological data can meet those requirements. An understanding of how paleontological data differ from neontological data (Section 3.1) provides insight into how morphological homeostasis can be identified in the fossil record (Section 3.2).

3.1. The nature of paleontological data

Paleontological data differ from neontological data in that fossils have been exposed to various biostratinomic (post-death, pre-burial) and diagenetic (post-burial, pre-discovery) processes prior to their study. Those taphonomic processes limit the amount and type of data available, and can destroy, blur, or distort the original biological signal. It is imperative to understand how taphonomic processes limit our ability to meaningfully extend the study of morphological homeostasis into the fossil record. Several key factors are discussed in the following sections. Some additional factors that must be controlled for in any study of morphological homeostasis, such as the size and/or developmental stage of the specimens for which trait variation is assessed, are common to both neontology and paleontology and are not discussed herein.

3.1.1. Genetic data are unavailable

Rapid decay of non-mineralized organics means that genotypic data are not preserved in the fossil record. Genetic differences between (non-clonal) fossil specimens are therefore unknown. Within-colony variation among clonal organisms such as corals, bryozoans, or graptolithines offers potential insight into environmentally sourced variation on the scale of a single colony (e.g., [121,122]). However, the contribution of genetic versus non-genetic variation to among-colony variation in those groups cannot be determined.

3.1.2. Phenotypic data are available but not always reliable

The fossil record can provide reasonable documentation of the evolution of “hard part” phenotypes, particularly in environments of high sedimentation rate where bioclasts can experience a relatively short pre-burial exposure time. But diagenetic processes can modify the phenotype, complicating assessment of its original form and distorting estimates of phenotypic variation within and between fossil samples. Given the need to assess phenotypic variation and deviations from symmetry in studies of morphological homeostasis (Section 2), such taphonomic overprint on morphology is a serious concern.

Tectonic or compaction-related deformation, for example, will affect fossil shape. The effect of taphonomic compaction on fossil shape can be dramatic: a comparison of non-compacted and compacted samples of the trilobite Olenellus gilberti found (controlling for specimen size and ontogenetic stage) a significant difference in mean cephalic shape, and that the variance of cephalic shape in the compacted sample was more than twice that in the non-compacted sample ([123]; see also [124]). Compacted fossil material is therefore unlikely to yield reliable estimates of biological shape variation or of fluctuating asymmetry in trait shape or size. However, the condition of qualitative or meristic traits (e.g., presence or absence of a given structure, or thoracic segment count) might still be ascertainable on compacted or distorted fossils. The range of traits for which the level of within- or among-individual variation can be reliably ascertained will therefore depend on the style and quality of fossil preservation, which can vary spatiotemporally.

3.1.3. Time-averaging limits temporal resolution

A fossil collection from a single stratigraphic bed or bedding surface is typically not equivalent to a deme. Even if the fossil-bearing interval was deposited as a single, rapid depositional event (such as a submarine mudslide) that simultaneously suffocated and buried contemporaneous individuals, the resulting fossil-bearing stratum is also likely to include cast-off molts and the remains of individuals that died prior to the event—perhaps a long time so—that were lying at or near the sediment surface and so were entombed along with those unfortunate victims of the event. A given fossil collection will therefore consist of a naturally time-averaged (temporally pooled) assemblage of individuals that inhabited (or perhaps were transported into) the same geographic location but were not necessarily all alive simultaneously.

The degree of time-averaging exhibited by a particular fossil collection depends on net sedimentation rate and the extent of physical or biological reworking of sediments prior to lithification. Degree of time-averaging can be inferred from detailed analysis of fossil preservation, sedimentary structures, diagenesis, and sequence stratigraphic position ([125,126,[127][128][129]]). In cases of high sedimentation rate and low reworking, fossil collections can be minimally time-averaged and come closest to approximating an ecological census (e.g., [128,130,131]), but such cases are rare. Most tightly constrained, “single-bed” fossil collections probably represent aggregates of individuals pooled over tens to hundreds of years [132,133]. Sometimes data from several stratigraphically distinct collection intervals must be pooled by a researcher to yield sufficient sample size for rigorous analysis (e.g., [130,131]). The effect of such analytical time-averaging is analogous to having sampled a more coarsely naturally time-averaged fossil-bearing interval.

Time-averaging limits the temporal resolution at which phenomena such as morphological homeostasis can be studied in the fossil record. Even single-bed paleontological samples are almost invariably multi-generational. This raises the obvious concerns that any trends or fluctuations in phenotype occurring at a temporal scale shorter than that over which the time-averaging occurred will be unresolvable, and that phenotypic variation within a fossil sample might be inflated relative to that of any included deme. Fortunately, this impact seems to be fairly minimal. Short-term temporal phenotypic variation at a single locality is often less than geographic variation within a species [57,134,135]. Modeling has also shown that phenotypic variation within a sample is not dramatically affected by time-averaging [136]. This is consistent with empirical data [137] showing that time-averaging on the scale of tens of years to tens of thousands of years typically inflated phenotypic variance in fossil samples by only approximately 5% relative to population-level variance. Analytical time-averaging of fossil assemblages into bins spanning up to nearly 4 million years yielded a similar conclusion [137]. The degree of intraspecific phenotypic variation in fossil samples is thus often comparable to that of modern samples (e.g., [135,138,136,131,137]). Indeed, the “averaging out” of shorter-term fluctuations in a time-averaging fossil sample can even be advantageous for some questions [139], particularly if the focus is on estimating longer-term variability rather than the standing level of variation at any particular time-slice [140].

3.1.4. Environmental variation is hard to study and impossible to control

It is unrealistic to attempt to sample the full geographic or environmental range of an extinct species at any particular time-slice of its stratigraphic range. The limited distribution of fossil-bearing localities that remain preserved and accessible today means that undersampling of the original range is virtually guaranteed. At the finest scale of stratigraphic resolution, studies of geographic variation within fossil species are typically restricted to spatial scales of several tens of kilometers. This is because (1) it is difficult or impossible to trace a single bed over greater distances, and (2) an assumption of synchronous deposition of a bed is more prone to violation at larger spatial scales. Studies at larger geographic scales typically involve regional correlation of thicker stratigraphic intervals and are therefore of lower temporal resolution. The relatively small spatial scale of high-resolution studies must result in underestimation of the full range of geographic variation that was exhibited by the species in life. Nevertheless, a diversity of environments can sometimes be sampled over such a relatively small spatial scale, and geographic variation has been documented over similar (and even smaller) scales among extant organisms (e.g., [[141][142][143]]).

Several studies have sampled time-equivalent strata at fine temporal resolution in order to assess phenotype-environment associations in the fossil record (e.g., [[144][145][146][147][148]]). However, interpreting any geographic/environmental variation in a fossil taxon is fraught with difficulty. First, such variation might be produced, reduced, or exaggerated by taphonomic processes. Lateral facies change in a fossil-bearing interval indicates a spatial change in environment, but this can also be associated with a change in preservational mode of those fossils (e.g., non-compacted specimens in a shallow water limestone grading into compacted specimens in deeper water shale) that can induce a non-biogenic component to among-environment phenotypic variation (Section 3.1.2). Specimens can also be transported from one site to another before they get buried. This can blur any differences between populations in a taphonomic analogy to dispersal and gene flow, or can create differences between populations by shape- or size-sorting. Fortunately, careful analysis of the fossils and the entombing sediments can reveal evidence of transportation (e.g., [128,129]). Second, it can be difficult to discern whether phenotypic differences between fossil samples represent intraspecific variation or interspecific disparity. Geographic/environmental variation of a fossil species might be underestimated if polyphenic ecomorphs or members of an incompletely sampled ecomorphocline are (mis)interpreted as distinct species [149], or overestimated if cryptic species are mistakenly treated as conspecific. (This problem is of course not unique to paleontology.) Finally, even if the observed geographic/environmental variation is considered to be a genuine biological phenomenon, it could represent phenotypic plasticity and/or local genetic adaptations (with limited gene flow). Without genetic data these alternatives cannot be distinguished.

A surprising amount of environmental information for a fossil-bearing locality can be obtained from careful study of the nature and geochemistry of the fossils and the entombing sediments. But the types and resolution of such paleoenvironmental data are still crude compared to those available in neontological studies: fossil-bearing localities that seem identical in terms of the available paleoenvironmental data could still have markedly differed in ways that are not resolvable in the fossil record. This, combined with the impossibility of subjecting fossil species to common garden or transplant experiments, severely limits our ability to confidently identify and control for environmental determinants of phenotype-environment associations in the fossil record.

3.2. Implications for identifying morphological homeostasis in the fossil record

3.2.1. Genetic canalization

The inability to control for genotype (Section 3.1.1) means that it is impossible to rigorously test a hypothesis that fossil samples differ in their degree of genetic canalization. Even if it is assumed that all individuals experienced an identical environment, any difference in among-individual phenotypic variation between those samples could result from either a difference in allelic sensitivity (i.e., genetic canalization) or from a difference in allelic diversity (i.e., genetic variance) (Fig. 1, top row). Stabilizing selection against deviant phenotypes could result in either an increase in genetic canalization or a decrease in genetic variance, but without the ability to assess allelic diversity it is not possible to distinguish between these alternatives (Section 2.1). A stratigraphic trend towards decreased among-individual phenotypic variation within a given environment would therefore be consistent with, but not uniquely diagnostic of, increasing genetic canalization. It would be interesting and important to explore whether selection tends to remove deleterious alleles more rapidly than developmental systems evolve genetic canalization [25,76], but paleontology cannot contribute towards finding the answer.

3.2.2. Environmental canalization

It is not possible to rigorously test a hypothesis that a fossil species changes (or that fossil species differ) in degree of environmental canalization. That is because it is not possible to unambiguously determine the form of a reaction norm for a fossil species. A reaction norm for a trait can be genotype-specific (Section 2.2). Genetic variation among individuals within outbreeding wild-type populations—and thus fossil samples—is expected. Even if an observed phenotype-environment association in a fossil species is deemed genuine and reliable (as opposed to having been produced or modified by either taphonomic processes or taxonomic practices), it is impossible to know to what extent that association resulted from phenotypic plasticity (i.e., represented a reaction norm of a single genotype) as opposed to genetic differences between individuals (i.e., local genetic adaptation; Section 3.1.4). The inability to resolve among-genotype from within-genotype variation cripples any attempt to resolve the form of a reaction norm for a fossil species and thus to compare reaction norms. Intraspecific change or interspecific difference in among-individual phenotypic variation of a trait over a given range of environments could result from either a difference in environmental sensitivity (i.e., environmental canalization) and/or a difference in genetic variation. Distinguishing between these alternatives is difficult even in neontological studies, and is impossible with paleontological data.

3.2.3. Homeorhesis (developmental homeostasis)

Deviation from the target phenotype can be quantified for some traits on some fossils, and this permits evaluation of homeorhesis in fossil samples. This is restricted to nominally symmetric traits, for which the target phenotype (i.e., symmetry) is known. Fluctuating asymmetry can be (and in rare cases has been) measured on fossil groups such as vertebrates and trilobites that exhibit the required trait symmetry ([20,57,[150][151][152][153][154][155][156]]). However, estimates of FA will be extremely sensitive to taphonomic overprint resulting from fossil compaction or deformation (Section 3.1.2). Only pristine, non-compacted and non-deformed specimens are appropriate for inclusion in an analysis of FA.

To maximize comparability to neontological studies, homeorhesis of a fossil species should be assessed from a minimally time-averaged sample (Section 3.1.3). A time-averaged sample might consist of pooled populations that differed in their levels of (relative) FA, because of either (1) a relatively rapid evolutionary change in homeorhesis, and/or (2) environmental change within the interval of time spanned by the time-averaged sample (because the level of FA can be influenced by environmental factors). In that case, the distribution of asymmetry values within the time-averaged sampled would be more leptokurtic than would that of any of the component populations. The ability to detect any differences in (relative) FA between fossil samples will depend on the magnitude of the effect (i.e., intersample difference in homeorhesis) relative to any “noise” introduced by the pooling of multi-generational data into each time-averaged sample. The distributions of (size-standardized) asymmetry values within the samples under comparison should not differ in kurtosis (in fact, each would be normally distributed if they exhibit “ideal FA” [Fig. 2; [[62][63][64]]; but see [24]]).

The small sample size available in most paleontological studies limits the ability to robustly estimate the level of (relative) FA within a fossil sample and to compare (relative) FA between samples [157]. Pooling specimens from more inclusive stratigraphic intervals in order to increase sample size comes with the caveat that such analytical time-averaging can bias the estimate of FA in a similar way to natural time-averaging.

It cannot be assumed that homeorhesis affects trait evolvability (Section 2.4). When investigating whether the evolution of a particular trait within a particular species or clade was constrained by homeorhesis, the first step should therefore be to empirically explore the relationship between the level of within-individual variation (FA) and the level of among-individual (symmetric) variation for that trait across samples (Fig. 3C; both FA and among-individual variation should be standardized to remove any size-related differences between samples). If a positive relationship is found, then a defensible claim can be made that developmental stability limited among-individual phenotypic variation and thus that homeorhesis might have been relevant to the evolvability of that trait. The next step would then be to test a hypothesis that trait evolution was influenced by homeorhesis. For example, large or rapid shifts in trait mean value—consistent with response to directional selection—should be associated with weak homeorhesis (high relative FA values) relative to a control (Fig. 3). Conversely, temporal stability in trait mean value combined with stable or decreasing among-individual variation—consistent with response to stabilizing selection—should be associated with strong or increasing homeorhesis (low or decreasing relative FA values) relative to a control (Fig. 3). Ideally, the control in either case should be a conspecific (or at least closely related) sample that showed a contrasting pattern of trait mean values (i.e., was not under the same inferred selection regime) as the focal samples.

Fig. 3

Fig. 3. Hypothetical paleontological example consistent with trait evolution being influenced by homeorhesis. A: Data for stratigraphic (time) series of seven samples, showing shift in mean phenotypic trait value within a lineage. B: Samples from fourth to sixth stratigraphic horizons from base (associated with largest changes in mean trait value in A) exhibit higher relative variation in trait value than do other samples (associated with little change in mean trait value in A). Among-individual variation expressed as coefficient of variation to control for differences in mean trait size. C: Within-individual variation (relative FA) positively correlates with among-individual variation (coefficient of variation) for trait across samples. Ability of developmental system to buffer against stochastic perturbations during development (homeorhesis) could have influenced production of phenotypic variation on a scale “visible” to selection. Note that in this example the positive relationship would not have been observed if only the four samples to the left of the plot had been examined—the relationship only becomes clear if samples spanning a wide range of relative FA or among-individual variation values are examined. D: Highest degree of within-individual variation (greatest developmental instability; weakest homeorhesis) is expressed in samples from stratigraphic interval over which largest changes in mean trait value (A) occurred. Positive association among degree of developmental instability (relative FA), magnitude of among-individual variation (coefficient of variation), and magnitude of change in mean trait value (A to D) is consistent with hypothesis that trait evolution was influenced by homeorhesis. One plausible scenario is that samples in oldest three horizons were under stabilizing selection, with lineage subsequently shifting into regime of either directional selection or drift (next three horizons) and back into regime of stabilizing selection at new, lower optimum trait value (youngest samples).

4. The role of morphological homeostasis in phenotypic evolution through deep time

In the fossil record, morphological homeostasis has been invoked as a control over phenotypic evolution on scales ranging from change in eye lens number within a single trilobite species [158], to shifts in dental growth rate during hominin evolution [157], to the establishment of major animal body plans during the Cambrian Radiation [159]. However, the ability to rigorously test such hypotheses in the fossil record—in a way that allows meaningful comparison to neontological studies of morphological homeostasis—is limited. Neither genetic nor environmental canalization can be unambiguously identified from fossil samples (Section 3.2.1, Section 3.2.2), so the contribution of changes in allelic and environmental sensitivity to deep time evolution is impossible to assess. The identification of changes in homeorhesis is possible in groups that exhibit trait symmetry, but requires the analysis of multiple minimally time-averaged samples, each comprising many individuals of exquisite preservational quality (Section 3.2.3)—requirements that will not often be met. Nevertheless, great insight into the relevance of homeorhesis to deep time phenotypic evolution could be gained from exploiting any instances where those requirements are met, as now discussed.

4.1. Within-lineage evolutionary rate

Since its initial popularization [160,161], the phenomenon of stasis—that fossil species commonly exhibit no net morphological change throughout their stratigraphic ranges, sometimes spanning several millions of years—has attracted much attention in the evolutionary sciences [162]. Morphological homeostasis has been invoked as a potential mechanism for stasis ([72,160,163,164,11,102,140]): a species that experienced stabilizing selection might evolve a homeostatic developmental system that buffered the species against variation and thus produced long-term stasis. If the degree of morphological homeostasis of the developmental system subsequently became lessened—as for example might occur in a peripheral population that experienced a novel and/or stressful environment—then the expression of previously masked variation could permit rapid phenotypic evolution and perhaps speciation (e.g., [19,38]). Alternation of increasing and decreasing morphological homeostasis could result in the stepwise evolutionary pattern of punctuated equilibria [16,38,160]. Many other mechanisms have also been proposed to explain stasis (see summaries by [140,162,165,166]), and the relative contributions of stabilizing selection and morphological homeostasis (if any) are uncertain. Gould [165] considered it unlikely that morphological homeostasis could buffer against variation for many thousands or even millions of years, but I am not aware of any published study that has explicitly tested the hypothesis that morphological homeostasis (and homeorhesis in particular) was responsible for stasis in the fossil record.

If homeorhesis influenced the evolution of a trait within a lineage (by constraining the among-individual variation for that trait; Section 2.4), then a positive relationship should exist between the level of developmental instability (FA) and the evolutionary rate of that trait within that lineage (Section 3.2.3Fig. 3). In the particular case of lineage in strict stasis (i.e., experiencing stabilizing selection on a stationary adaptive peak), that lineage should, all else being equal, show lower developmental instability than a lineage experiencing directional selection (Fig. 3; [20,72]). The ability to detect such a relationship depends on the nature of the “stasis” (strict versus bounded fluctuations) and on sampling resolution relative to the duration of phenotypic response to selection. Long-term net stasis in a lineage that resulted from shorter-term bounded (perhaps stochastic) fluctuations in location of an adaptive peak—i.e., stasis involving no net trend in a fluctuating evolutionary trajectory rather than absolute stability of mean trait value—might not fit this prediction. Modeling has shown that, although morphological homeostasis can be favored under short-term bounded fluctuating selection [167], the level of morphological homeostasis under such circumstances can be weaker than under stabilizing selection [81]. Indeed, empirical data show that evolutionary rate along a single segment of a fluctuating trajectory can be high [168]. The response of a lineage to a shift in adaptive peak location, or a transition from one peak to another, can be rapid (103 to 104 years, or even down to a single generation in the case of adaptive polyphenic ecomorphs) relative to typical paleontological sampling [38,162]. The probability of sampling a fossil species during this transition is therefore low: the transition might not be sampled at all, or could be represented within a single coarsely time-averaged sample that would contain individuals adapted to each optimum. Nevertheless, there are documented examples of fossil species apparently “caught in the act” of undergoing a phenotypic response to directional selection either into or out of an interval of stasis, two of which are highlighted below.

In one example, Hunt et al. [131] documented a highly resolved time series of sticklebacks from Miocene lake sediments in Nevada that evolved armor reduction towards a new adaptive optimum over approximately 17,000 years. The average time gap between successive analytically time-averaged collections was 250 years, although the actual stratigraphic sampling was made at a resolution of years to decades [130]. As the lineage approach the new adaptive peak it experienced a transition from directional to stabilizing selection, and the rate of change in the evolving traits progressively declined. The lineage exhibited an apparent lag of approximately 3500 years between invading the lake (when selection for pelvic reduction was presumably first experienced) and the actual onset of pelvic reduction; there was little to no variation in the armor traits prior to the onset of that trend. In the absence of any other data it might be tempting to infer that this lag represented the time taken to increase the sensitivity to variation of a developmental system that initially been well buffered against such variation, and that morphological homeostasis had initially served as a constraint. Such a hypothesis could potentially be tested (albeit only for homeorhesis, and only if preservational quality was sufficiently high) by tracking any temporal change in within-individual variation (relative FA) of the evolving traits (Fig. 3). However, in this particular example morphological homeostasis of any kind is unlikely to have been responsible for the lag, because the cause of the phenotypic change has been tentatively identified (based on studies of extant sticklebacks) as a segregation of the Pitx1 gene: the lag probably resulted from the low initial frequency of the recessive Pitx1 allele [130,131]. That the evolution involved a change in the relative frequency of, rather than sensitivity to, the allele argues against a potential hypothesis that genetic canalization served as a constraint.

A second example involves a progressive reduction in mean eye lens number in the Devonian phacopid trilobite Acuticryphops acuticeps over less than one million years, perhaps as an adaptation to decreasing illumination at the seafloor during a sea level rise [158]. As mean lens number decreased from eleven to five, the intra-sample coefficient of variation increased. This is consistent with a relaxation of stabilizing selection and the onset of drift or directional selection. Intriguingly, left-right asymmetry in lens number was only observed in the two youngest samples, and the authors noted that this was consistent with developmental instability controlling trait variation. Unfortunately, sample size was insufficient to rigorously test the hypothesis that the phenotypic trend was associated with a decrease in relative FA. Nevertheless, the study represents a promising potential case of morphological homeostasis (specifically, homeorhesis) influencing evolution in the fossil record. Other cases of left-right asymmetry in lens number within phacopid species are known (e.g., [[169][170][171][172][173]]). Changes in lens number in the species studied by [160,174]) form one of the classic cases for punctuated equilibria [160], and it would be fascinating to more thoroughly explore the distribution of asymmetry in lens number in those samples.

4.2. Survivorship

High phenotypic variation has been assumed to decrease extinction probability (e.g., [71]). This association might arise because more variable traits can respond faster to selection (i.e., enhanced evolvability), or because intraspecific variation presumably positively correlates with niche width and/or geographic range size which in turn promote survivorship. The assumption has gained empirical support: a study of veneroid bivalves found that, in a series of species-pair comparisons in which one species survived the Plio-Pleistocene extinction event and the other did not, it was the more variable species that tended to survive [175]. A counterexample might be represented by study of Cambrian trilobites which found that the longest stratigraphic durations were associated with the least variable species [176], although that relationship could conceivably be explained by a higher rate of evolution (and thus pseudo-extinction) of the more variable species.

If survivorship is influenced by phenotypic variation, this raises the question of whether morphological homeostasis (in addition to genetic variation) plays a role in determining extinction risk [114,143,[177][178][179]]. This could only be investigated in the fossil record for homeorhesis, and the demands on sampling are extremely high: within-individual variation (relative FA) would have to be reliably measured in enough victim and survivor species to yield sufficient statistical power to meaningfully test for its effect on survivorship. Interpretation of the results might be complicated if the level of relative FA exhibited by a species increased in the stressful conditions of an extinction event [20], or if intraspecific geographic variation in developmental (in)stability was present [114]. Nevertheless, study of any system that permits such an investigation would pay dividends: empirical support for or against the potential link between developmental (in)stability and extinction risk in the fossil record is currently lacking, but could have great relevance to predicting biotic response to present-day environmental perturbations.

4.3. Primitive versus derived developmental systems

Perhaps the greatest contribution that paleontology can make to our understanding of the evolution of development comes from the opportunity to study the developmental systems of organisms that were far less derived than are those of the extant biota. Developmental systems of fossil organisms offer a chance to ground-truth ancestral state reconstructions inferred by comparative methods applied to modern taxa. Ancient developmental systems also operated within and were adapted to ecosystems and environments for which there is not always a modern-day equivalent, and might thus exhibit novel states that are beyond our familiarity and would not be predicted from the study of modern taxa. For example, it has been claimed that ancient developmental systems “undoubtedly exhibited less genetic redundancy and metabolic integration and homeostasis” relative to the derived developmental systems of modern organisms ([159], p. 290).

“Rosa’s rule” is a purported macroevolutionary trend towards progressive reduction of variation through time: traits that are variable within ancestors become fixed in descendants ([6,71,123,[180][181][182]]; the “flexible stem hypothesis” of West-Eberhard [38] is similar but more restrictive in that it stresses that the variation expressed within ancestors resulted in particular from phenotypic plasticity). To the (limited) extent that “Rosa’s rule” is supported by data, those data have come from comparison of stratigraphically old members of a clade to stratigraphically younger members of a clade. This of course requires the study of fossils, because all extant members of a clade are separated from their last common ancestor by an equal amount of time.

Most support for “Rosa’s rule” has come from trilobites. Mature thoracic segment number often varies within species or among closely-related species in Cambrian trilobites, but becomes almost invariable within post-Cambrian clades [123,[183][184][185][186]]. [117,118] further found that the Cambrian trilobite Dikelocephalus minnesotensis exhibited greater phenotypic variation in many traits than the Devonian trilobite Eldredgeops rana, and suggested that the developmental systems of Cambrian trilobites were generally less well regulated or less strongly canalized than were those of post-Cambrian trilobites. In conformity with this pattern, the early Cambrian trilobite Olenellus gilberti also exhibited a relatively high degree of among-individual variation in many traits [123]. However, some aspects of cephalic growth in Olenellus gilberti were consistent with being under tight developmental regulation, emphasizing the need to consider trends in phenotypic variation on a trait-by-trait basis [123]. A phylogenetically and stratigraphically broad survey of trait evolution in trilobites [182] found that early and middle Cambrian trilobites exhibited significantly higher levels of intraspecific variation than post-Cambrian trilobites. There is still much work to be done to rigorously test whether trait evolution in trilobites followed “Rosa’s rule,” and analogous trends in other fossil groups remain unexplored, but the scant data currently available suggest that the macroevolutionary trend might be real (at least for some traits) and further investigation is justified.

Mechanistically, declining phenotypic variation within a clade could be produced by partitioning of geographically-structured variation in an ancestor among two daughter lineages at an allopatric speciation event [139,181,187]. That pattern could also result from exhaustion of genetic variation or from increased morphological homeostasis [25,38,71,117]. This raises the fascinating possibility that a net trend towards increased morphological homeostasis, spanning 107 to 108 years, might characterize major metazoan lineages, potentially causing a progressive net decline in phenotypic variation and evolvability. [188], p. 372) stated that “the notion that biological systems ought to evolve to a state of higher stability against mutational and environmental perturbations seems simple enough, but has been exceedingly difficult to prove.” Studying developmental instability (relative FA) through clade history offers one potential way to assess changes in homeorhesis through deep time [62]. Of course, for any such trend to be detected it would have to be of greater effect size than any inter-sample differences in homeorhesis that result from other, shorter-term influences on developmental instability (e.g., whether the samples were under stabilizing versus directional selection, or experienced stress-inducing versus non-stress-inducing environmental conditions).

A pioneering study testing for a long-term temporal trend in homeorhesis in trilobites failed to detect a significant change in levels of (relative) FA through deep time [151]. However, that study included only nine species spanning the Cambrian through Silurian, representing five orders without controlling for inter-ordinal differences in variation (that were subsequently found by Webster [182]). The nine species also inhabited a wide range of environments, had disparate life modes, and exhibited vastly different (and often difficult to homologize) cephalic morphologies. Taphonomic control was also inadequate: seven species were represented by non-compacted silicified specimens but the other two were represented by compacted specimens in shale, and some of the samples had suffered mild tectonic deformation (personal observation). These issues raise concerns about the robustness of the results. A similar study to that of Smith [151], employing denser taxonomic sampling with stricter phylogenetic and taphonomic control, is greatly needed.

5. Conclusions

Extending the study of morphological homeostasis back through deep time has obvious appeal: it would allow hypotheses regarding its long-term (104 to 108 years) influence over evolutionary rate and survivorship to be empirically tested, and could reveal insight into the nature of primitive developmental systems that would not be predicted from the study of modern organisms alone. Meaningful integration of neontological and paleontological data requires application of common terms and methods, and an awareness of potential limits of comparability between datasets. Taphonomic processes restrict the amount and type of data available in paleontological studies, and can destroy, blur, or distort the original biological signal: genetic data are unavailable; phenotypic data can be modified by tectonic or compaction-related deformation; time-averaging limits temporal resolution and can reduce short-term evolutionary signal into intra-sample “noise”; and environmental variation is hard to study and impossible to control. These facts impose strong limitations on the ability to study morphological homeostasis in the fossil record.

Identification of either genetic or environmental canalization requires knowledge of genotype. Hypotheses that phenotypic evolution was influenced by change in either genetic or environmental canalization therefore cannot be unambiguously tested with fossil data, because neither makes predictions that are not also made by a hypothesis involving a change in genetic diversity.

Homeorhesis can be assessed in ancient developmental systems by measuring the level of fluctuating asymmetry (FA) in a nominally symmetric trait—a proxy for developmental instability. The opportunities to do this will be limited, because it requires the analysis of multiple, minimally time-averaged samples of exquisite preservational quality. The ability to detect any differences in (relative) FA between fossil samples will depend on the magnitude of the effect relative to any “noise” introduced by the pooling of multi-generational data into each time-averaged sample. The detection of any long-term trend in homeorhesis will similarly depend on its effect size relative to that of shorter-term influences. Furthermore, there is no guarantee that within-individual variation (FA) influences the among-individual (symmetric) variation—and thus the evolvability—of a trait: such a relationship must be assessed on a case-by-case basis.

Nevertheless, studies of homeorhesis in the fossil record could reap high reward. Very few such studies have been conducted to date, and any future study that meets the criteria to test hypotheses regarding how developmental (in)stability relates to evolution would stand to make a valuable contribution to our understanding of the deep time, macroevolutionary significance of morphological homeostasis.


This work was supported by the National Science Foundation (research grant EAR Integrated Earth Systems 1410503).


Vincent Debat and Arnaud Le Rouzic generously invited this contribution. Editor-in-chief John Davey efficiently smoothed the navigation through the submission process. Constructive, helpful reviews were provided by Vincent Debat, Arnaud Le Rouzic, David Pfennig, and A. Richard Palmer.


View Abstract