What Do We Know about Patterns of Within-Lineage Change?

Stasis is common, but not the dominant mode of evolution within lineages. Voje (Reference Voje2016) found stasis to be best supported in 34% of cases. Hunt et al. (Reference Hunt, Hopkins and Lidgard2015) similarly found 38% for uniform stasis models, with many of the other best-supported models also having stasis as a component (e.g., two intervals of stasis separated by punctuation). In total, 63% of time series were best fit by a model with stasis occurring over some or all of its span. As expected, sequences for which stasis is favored show markedly less net change than other sequences (Voje Reference Voje2016); in fact, nearly half of them are better supported by a stricter version of the stasis model consistent with no real evolutionary variation (this means that sampling error on trait means is large enough to account for all observed trait variation; Hunt et al. Reference Hunt, Hopkins and Lidgard2015). However, in other cases, the stasis pattern encompasses substantial evolutionary change, and the average amount of trait change between samples can even be larger for stasis than for other models (Voje Reference Voje2016).

Sustained directional trends, on the other hand, are rather uncommon, comprising 12% (Voje Reference Voje2016) or 9% (Hunt et al. Reference Hunt, Hopkins and Lidgard2015), with the latter figure increasing to 13% if all models that include periods of directional change are included. Cases not best fit by uniform stasis or trends instead favor random walks or models with more complex dynamics. In both studies, these account for a majority of time series (54% for Voje [Reference Voje2016] and 53% for Hunt et al. [Reference Hunt, Hopkins and Lidgard2015]). Thus, most time series conform neatly to neither stasis nor gradual trends.

Which models are best supported can be influenced by the nature of the observational window. Stasis becomes less likely to be favored as the temporal duration of sequences increases (Hunt et al. Reference Hunt, Hopkins and Lidgard2015); if one waits long enough, it becomes increasingly probable that a trait will show some kind of accumulating evolutionary change.

What Do These Patterns Mean for Punctuated Equilibrium?

On balance, these aggregate results offer at least partial support for punctuated equilibrium’s claim about low net evolution during within-lineage change. Stasis is common over the 10⁵–10⁶ year timescales that paleontologists usually examine, and it occurs at much higher frequencies than the directional patterns associated with phyletic gradualism. Moreover, this differential must be even greater than indicated by these compilations, because paleontologists have surely documented trends disproportionately, hoping to capture “interesting” evolutionary change (Gould Reference Gould2002). On the other hand, most cases are not best explained as stasis, and its frequency falls short of what most observers would consider dominant.

Even if stasis is not quite dominant, punctuated equilibrium was important in recognizing stasis as a phenomenon of interest. Indeed, its most lasting contribution to evolutionary biology may well be the realization that net evolutionary rates are so often exceedingly slow compared with instances of rapid evolution in living populations. Although patterns of stasis were noted by some earlier paleontologists (see examples in Gould Reference Gould2002: pp. 745–755), it was not widely anticipated by evolutionary biologists and would not have been discovered without the fossil record. The frequent occurrence of stasis in the fossil record despite sometimes rapid evolution in extant populations (Hendry and Kinnison Reference Hendry and Kinnison1999) and strong directional selection and abundant genetic variance in extant populations has been termed “the paradox of stasis” (Hansen and Houle Reference Hansen, Houle, Pigliucci and Preston2004), and it has been suggested to be one of the significant unsolved problems in evolutionary biology (Williams Reference Williams1992). However, while stasis is commonly observed, it is not universal, and it has become clear that the binary categories of stasis and gradualism are not rich enough to capture the variety of patterns of trait evolution we see in fossil lineages. This observation has several implications to which we will return later.

What Do These Patterns Mean for Understanding Evolution?

Eldredge and Gould (Reference Eldredge, Gould and Schopf1972) cautioned that evolutionary process is difficult to infer from fossil data. Nonetheless, one front in the controversy that followed concerned the microevolutionary implications of the punctuated equilibrium model (Gould Reference Gould1980; Charlesworth et al. Reference Charlesworth, Lande and Slatkin1982; Hancock et al. Reference Hancock, Lehmberg and Bradburd2021). We accept that paleontologists usually lack the information needed to assess microevolution in any particular lineage. Stepping back, however, do the aggregate fossil patterns reveal anything about these underlying processes?

We will not examine this issue in detail (see Uyeda et al. Reference Uyeda, Hansen, Arnold and Pienaar2011; Hansen Reference Hansen, Svensson and Calsbeek2012; Hunt and Rabosky Reference Hunt and Rabosky2014). Instead, we will use Simpson’s (Reference Simpson1944) concept of phenotypic adaptive landscape to sketch out some useful ways to think about species-level morphological time series in the fossil record. Adaptive landscapes have axes that specify phenotype, with mean population fitness represented as elevation above these axes, similar to a contour map of topography. Natural selection pushes populations uphill, toward higher fitness, with drift adding a non-directional component of change. Simpson used this concept qualitatively to explore evolutionary patterns, such as the evolution of hypsodonty in the horse clade, but theory development starting with Lande (Reference Lande1976, Reference Lande1979) formalized this conceptual tool. We now have equations that describe how populations move on specified adaptive landscapes (Arnold Reference Arnold2023), which have become increasingly central to how evolutionary biologists understand microevolutionary processes over long timescales (Arnold et al. Reference Arnold, Pfrender and Jones2001; Hansen Reference Hansen, Svensson and Calsbeek2012; Uyeda and Harmon Reference Uyeda and Harmon2014; Rolland et al. Reference Rolland, Henao-Diaz, Doebeli, Germain, Harmon, Knowles and Liow2023).

In the simplest case of a fixed adaptive peak, populations located far from the adaptive optimum are pushed toward it, rapidly at first, but in a decelerating manner until the population converges to the peak. Even with weak selection, this convergence will usually be too rapid to resolve in the fossil record. Once the peak is attained, the population fluctuates around it in a balance between outward perturbations from drift and the inward pull of selection toward the optimum. These fluctuations are a plausible cause of stasis for traits that display a narrow range of variation (as for cases of strict stasis, noted earlier), but empirical investigations have not found support for this as a general explanation for stasis in the fossil record (Voje et al. Reference Voje, Starrfelt and Liow2018). Meta-analysis of studies of natural selection has found that selection coefficients can vary substantially over the course of several generations (Siepielski et al. Reference Siepielski, DiBattista and Carlson2009), although much of this apparent variation may be attributable to noise in estimating selection (Morrissey and Hadfield Reference Morrissey and Hadfield2012). Regardless, over timescales longer than a few generations, it seems unlikely that environmental and biotic conditions would remain constant. As factors that affect a species change, so does that species’ adaptive landscape. Peaks can change in height and position, and populations will follow them with lags that are generally much shorter than paleontological resolutions. This tracking of peaks may be cut short by population extinction, especially for large changes in the adaptive landscape that indicate substantial mismatch between a population’s traits and its conditions. The resulting population decline and extirpation may thus filter out many of the largest evolutionary responses that might have otherwise occurred (De Lisle et al. Reference De Lisle, Punzalan, Rollinson and Rowe2021). Nevertheless, with dynamic adaptive landscapes, long-term morphological changes are primarily governed by changes in the landscape itself, rather than the relatively quick population dynamics on stable landscapes.

This framework suggests a first-order interpretation for the distribution of patterns we see in paleontological time series (Uyeda et al. Reference Uyeda, Hansen, Arnold and Pienaar2011; Hansen Reference Hansen, Svensson and Calsbeek2012; Hunt and Rabosky Reference Hunt and Rabosky2014; Voje et al. Reference Voje, Starrfelt and Liow2018; Arnold Reference Arnold2023). Traits that are best fit as random walks are following peaks in the adaptive landscape that have meandered, and stasis results when conditions instead oscillate or fluctuate. These fluctuations need not be small (Voje Reference Voje2016), but as long as they are bounded and non-directional, a stasis model will be a good description of the pattern. When conditions change heterogeneously over time, evolutionary models that are uniform over time can fail, and complex models (punctuations, mode shifts) may better capture trait dynamics. Trends, which are otherwise difficult to account for (Gould Reference Gould2002), are likely rare, because conditions that species experience tend not to change as simple, linear trends over hundreds of thousands to millions of years. Short-term directional changes are probably common, but likely only rarely sustained long enough to result in accumulating change (Futuyma Reference Futuyma1987) or paleontological trends within species (e.g., Hunt and Roy Reference Hunt and Roy2006). Of course, these evolutionary patterns may be altered or distorted by temporal changes in preservation and sedimentation at a locality (Kidwell and Holland Reference Kidwell and Holland2002; Patzkowsky and Holland Reference Patzkowsky and Holland2012). The net effect of this geological overprint will depend greatly on both sedimentary context and on the true evolutionary pattern. Simulations and other approaches can be used to explore ranges of plausible effects (Holland Reference Holland2000; Hannisdal Reference Hannisdal2006, Reference Hannisdal2007; Hunt Reference Hunt2008; Patzkowsky and Holland Reference Patzkowsky and Holland2012; Hohmann et al. Reference Hohmann, Koelewijn, Burgess and Jarochowska2024), but this remains an understudied area of investigation. The immigration of morphologically similar lineages can also confound anagenetic patterns, but careful morphological analysis can at least sometimes reveal this dynamic (Van Bocxlaer et al. Reference Van Bocxlaer, Van Damme and Feibel2008; Hull and Norris Reference Hull and Norris2009).

This interpretative framework focuses on natural selection as the primary driver of morphological change, which is undoubtedly a simplification of a more complex evolutionary reality. Genetic drift will always operate and may be potent in small populations, but morphological change is far too slow in the fossil record for neutral evolution to be dominant (Hansen Reference Hansen, Svensson and Calsbeek2012; Hunt Reference Hunt2012). Trait variation may also influence the direction of evolution, with evolutionary changes potentially concentrated in more variable traits and combinations of traits (evolvability in its modern sense; Schluter Reference Schluter1996; Hansen and Houle Reference Hansen, Houle, Pigliucci and Preston2004; Hunt Reference Hunt2007a; Love et al. Reference Love, Grabowski, Houle, Liow, Porto, Tsuboi, Voje and Hunt2021; Opedal et al. Reference Opedal, Armbruster, Hansen, Holstad, Pelabon, Andersson and Campbell2023; Voje et al. Reference Voje, Grabowski, Holstad, Porto, Tsuboi, Bolstad, Hansen, Houle, Pavlicev and Pélabon2023; Holstad et al. Reference Holstad, Voje, Opedal, Bolstad, Bourg, Hansen and Pélabon2024). Some workers have also argued that population structure and gene flow may be particularly important contributors to stasis, as genetic exchange across variable environments may average out local adaptive changes (Lieberman et al. Reference Lieberman, Brett and Eldredge1994; Lieberman and Dudgeon Reference Lieberman and Dudgeon1996; Eldredge et al. Reference Eldredge, Thompson, Brakefield, Gavrilets, Jablonski, Jackson, Lenski, Lieberman, McPeek and Miller2005). The relationship between gene flow and evolutionary divergence can be complex (Hanski et al. Reference Hanski, Mononen and Ovaskainen2011; Kopp and Matuszewski Reference Kopp and Matuszewski2014), and testing the importance of spatial structure and gene flow for stasis will require adding to the few studies that document phenotypic evolution within a species across multiple environments or regions (e.g., Cisne et al. Reference Cisne, Chandlee, Rabe and Cohen1980; Lieberman et al. Reference Lieberman, Brett and Eldredge1995; Kucera and Malmgren Reference Kucera and Malmgren1998; Webber and Hunda Reference Webber and Hunda2007).

Despite all these potential complications, simulations by Hunt et al. (Reference Hunt, Hopkins and Lidgard2015) suggest that peak-tracking models have potential to account for many aspects of the aggregated data we have. Data simulated with a simple, calibrated model in which morphology tracks a long-term temperature curve resulted in frequencies of trends, random walks, stasis, and other models that were strikingly similar to the distribution we see in empirical fossil time series.

What Is Next for Analyzing Fossil Time Series?

We noted earlier that natural selection is usually too fast, and the fossil record is usually too coarse, to resolve evolutionary changes on a stable adaptive landscape. To our knowledge, there is just one such example, in which a varved lake deposit captures the adaptive reduction in skeletal armor in a lineage of stickleback as it invades a new lake (Hunt et al. Reference Hunt, Bell and Travis2008). Targeted work in high sedimentation rate environments, focusing on traits of adaptive significance, may produce more such examples in which paleontologists can hope to infer truly microevolutionary processes. In nearly all cases, however, the models discussed in this paper are better thought of as phenomenological, rather than process models. Their parameters do not reflect population genetic quantities such as selection gradients (although selection can be estimated in fossil populations by other means, at least under special circumstances; e.g., Van Valen Reference Van Valen1963; Hunt and Yasuhara Reference Hunt and Yasuhara2010; Di Martino and Liow Reference Di Martino and Liow2021). Instead, time-series models are tools to extract and quantify general features of evolutionary change such as directionality, pace, and punctuations. Recent papers have developed models to capture new features of interest, for example, initially rapid but then slowing changes in the adaptive landscape (Voje Reference Voje2020), dynamics that reflect underlying but hidden drivers (Reitan et al. Reference Reitan, Schweder and Henderiks2012), and sudden or gradual shifts in the adaptive landscape (Voje Reference Voje2023). There is more work to be done in crafting models to assess other dynamics of interest.

One striking feature about the aggregated literature on fossil time series is how univariate the analyses are. Often, multiple traits are measured, but the analysis almost always proceeds one trait at a time, sometimes after using principal components analysis or other techniques to reduce dimensionality (which has its own complications in this context; Bookstein Reference Bookstein2013; Uyeda et al. Reference Uyeda, Caetano and Pennell2015). The net effect is that we have good compilations of how traits evolve, but less clear of an idea about how morphology, as an integrated whole, changes. For example, it may be unremarkable when a single trait does not change over time. But when all observable aspects of form do not change, that is a more surprising outcome and greater validation of punctuated equilibrium’s contribution to evolutionary thought (Futuyma Reference Futuyma2010). Modeling traits jointly is complicated when different traits within the same lineage show different modes of evolution, which is a common outcome (Hopkins and Lidgard Reference Hopkins and Lidgard2012; Hunt et al. Reference Hunt, Hopkins and Lidgard2015). However, recent developments allow for differing trait dynamics within a unified modeling framework (Voje Reference Voje2023). Although these models can be more challenging to fit compared with univariate models, they open the door to an integrated organismal approach to understanding evolution on multivariate adaptive landscapes.

Although such methodological developments are welcome, the analysis of paleontological time series may be limited more by data availability than by the lack of appropriate models. The hundreds of cases gathered by recent compilations are only a small and somewhat idiosyncratic fraction of what has been collected by paleontologists, and they are insufficient for exploring differences in evolutionary outcomes across taxa, trait types, temporal scales, and other factors of interest. This limitation may be mitigated by the recently established Phenotypic Evolution Time Series database (PETS, https://pets.nhm.uio.no). This online community resource allows researchers to deposit for public use their own (or other published) data on morphological time series. The hope is that, just as for the Paleobiology Database, broad input from the paleontological community will quickly result in a much larger and richer store of data than could be compiled by any one researcher.

Inferring Unobserved Speciational Pulses

Ideally, paleontologists could assess speciational pulses by sampling morphological changes through a cladogenetic event. Eldredge and Gould (Reference Eldredge, Gould and Schopf1972) considered speciation to be rapid—an event in the geological sense. Some mechanisms of speciation do proceed quite quickly compared with paleontological time resolutions (Coyne and Orr Reference Coyne and Orr2004; Hendry et al. Reference Hendry, Nosil and Rieseberg2007), but not all do so, and the reality may be complex (Norris and Hull Reference Norris and Hull2012; Dynesius and Jansson Reference Dynesius and Jansson2014; Etienne et al. Reference Etienne, Morlon and Lambert2014). Nevertheless, only a handful of paleontological studies have claimed to analyze trait evolution in detail as, or shortly after, a lineage splits (e.g., Gingerich Reference Gingerich1976; Kellogg Reference Kellogg1983; Lazarus Reference Lazarus1986; Sorhannus et al. Reference Sorhannus, Fenster, Burckle and Hoffman1988).

Thus, barring unusually protracted speciation captured by a fortuitous fossil window, the evidence used to assess speciational pulses will be indirect. The most important paleontological studies on this question were made by Alan Cheetham and colleagues on the bryozoan genus Metrarabdotos (Cheetham Reference Cheetham1986, Reference Cheetham1987; Jackson and Cheetham Reference Jackson and Cheetham1990; Cheetham et al. Reference Cheetham, Sanner and Jackson2007). These studies are widely seen as the best test of punctuated equilibrium because of their detailed stratigraphic and geographic sampling, careful delimitation of species, quantitative genetic analysis of traits, and demonstrated correspondence between genetic and morphological species. We would argue, however, that the most critical advance was that these studies integrated ancestor–descendant sequences into a phylogenetic framework, rather than analyzing them in isolation. This approach allowed Cheetham and colleagues to estimate changes within species and to compare them explicitly to the changes between species. Doing so indicated that Metrarabdotos showed only minor fluctuations within species and much larger differences between species, supporting the claim that speciation was associated with an extra burst of morphological change.

The status of Cheetham et al.’s conclusions is more clouded now, as a subsequent study (Voje et al. Reference Voje, Martino and Porto2020) revealed that the finding of speciational pulses depended on methodological decisions related to missing data and the combination of continuous and discretely measured traits in the same analysis (see that paper for details). Regardless of the outcome in this one case study, the approach of stitching ancestor–descendant sequences into a phylogenetic framework is clearly a promising one. It is therefore surprising to us that only a handful of subsequent studies have used this approach in the nearly 40 years since Cheetham first employed it (Polly Reference Polly, MacLeod and Forey2002; Pachut and Anstey Reference Pachut and Anstey2009; Hunt Reference Hunt2013).

We present a brief summary of one of these studies (Hunt Reference Hunt2013) to illustrate the potential—and the challenges—of this kind of study. Like Cheetham’s work, this one focused on a single genus, that of the deep-sea ostracode Poseidonamicus. This study characterized morphology through 10 geometric landmarks (Fig. 2A) that were largely independent of the features used to define species, avoiding the potential circularity of looking for speciational pulses in the same characters that were used to separate species in the first place (Levinton and Simon Reference Levinton and Simon1980). Fifty-one populations from 14 different species-level lineages were sampled (Fig. 2B). Visually, some of these species displayed little morphological change (e.g., P. rudis, P. miocenicus), whereas others experienced anagenetic changes comparable to the differences between closely related species (Fig. 2B). These data were fit to an evolutionary model, first proposed by Bokma (Reference Bokma2002), with anagenetic evolution within lineages that followed a random walk, plus an additional, instantaneous pulse of change that occurs at each speciation. A version of this model with stasis within lineages, rather than random walks, turned out to be better supported, and we focus on those results here.

Figure 2. A, Representative of the ostracod genus Poseidonamicus with 10 landmarks used in morphological analysis; scale bar is 250 microns. B, Inferred phylogeny of the genus, with sampled anagenetic portions of the clade in solid lines. Population means are plotted vertically according to their mean scores on PC 1 and horizontally by their geological ages. C, Plot showing maximum-likelihood estimates (circles) and confidence intervals (vertical bars) for the proportion of evolutionary change in species attributable to speciational pulses. PC 1 and PC 2 accounted for 25% and 17% of the variation in the morphometric dataset, respectively. Body size was measured as log-transformed valve length.

The parameters of this model can be converted into a metric that summarizes the percent of total evolutionary change in a species that can be attributed to the speciational pulse at its origin. The best estimates for this metric differed by trait: PC 1’s estimate implies almost all change is speciational, whereas PC 2 shows the opposite pattern (Fig. 2C). Body size is intermediate, with nearly equal speciational and anagenetic contributions to evolutionary change. Except for PC 1, however, the confidence limits on these estimates are very broad, spanning solutions from near-zero to almost 100% speciational evolution (Fig. 2C). Thus, this dataset, even with its rather good phylogenetic and stratigraphic coverage, cannot discriminate speciational and anagenetic evolution for most traits.

The lack of similar paleontological studies makes it difficult to know how general these findings are. For extant species, there is a parallel literature in which the Bokma model or other tree-based approaches are used to infer speciational change (Mooers and Schluter Reference Mooers and Schluter1998; Pagel Reference Pagel1998; Ricklefs Reference Ricklefs2006; Bokma Reference Bokma2008; Mattila and Bokma Reference Mattila and Bokma2008; McPeek et al. Reference McPeek, Shen, Torrey and Farid2008; Ingram Reference Ingram2011; Ingram et al. Reference Ingram, Harrison, Mahler, Castaneda, Glor, Herrel, Stuart and Losos2016). Another strand of research tests for punctuated equilibrium more indirectly by looking for a correlation between lineage diversification rates and morphological divergence (e.g., Ricklefs Reference Ricklefs2004; Adams et al. Reference Adams, Berns, Kozak and Wiens2009; Rabosky et al. Reference Rabosky, Santini, Eastman, Smith, Sidlauskas, Chang and Alfaro2013), although this correlation can be positive even in the absence of speciational change (Rabosky Reference Rabosky2012). Some, but not all, of these studies find evidence for speciational evolution. Moreover, those studies that include confidence intervals often report that the data cannot well constrain the importance of speciational pulses (e.g., Mattila and Bokma Reference Mattila and Bokma2008; Ingram Reference Ingram2011; Ingram et al. Reference Ingram, Harrison, Mahler, Castaneda, Glor, Herrel, Stuart and Losos2016). This uncertainty may be unsurprising for modern studies, because they need to overcome the lack of data on within-species change with strong evolutionary assumptions (see Hunt Reference Hunt2013). But the Poseidonamicus case study shows that this uncertainty can persist even with good paleontological data constraining anagenetic evolution.

We conclude that punctuated equilibrium’s second core claim cannot yet be assessed with any degree of confidence (see also Rolland et al. Reference Rolland, Henao-Diaz, Doebeli, Germain, Harmon, Knowles and Liow2023). Limited fossil evidence, along with broader but less direct evidence from extant clades, indicates that pulses of morphological evolution associated with speciation occur. But the importance of these pulses compared with anagenetic change within lineages is still unresolved. Moreover, we foresee that this question will likely remain a challenging problem, because it faces inherent difficulties. Speciation is inferred indirectly, with uncertain timing, from an incomplete record of fossil occurrences. Models of morphological change are needed to convert morphological observations into parameters that are relevant for testing, but we do not know yet whether results will generally be robust across plausible models.

Furthermore, all these analytical sources of uncertainty are layered on top of the complexities of just recognizing the input units—species—in the fossil record (Allmon and Yacobucci Reference Allmon and Yacobucci2016). A long-standing criticism of punctuated equilibrium concerned its potential circularity: if species are defined on the basis of morphology, then rapid trait evolution will lead to new named species, and therefore a spurious association with “speciation” (Gingerich Reference Gingerich1976; Levinton and Simon Reference Levinton and Simon1980; see also Hopkins and Lidgard Reference Hopkins, Lidgard, Allmon and Yacobucci2016). Such a scenario is falsified when the putative ancestor persists beyond the origin of the descendant; thus, the stratigraphic configuration of species became an important line of evidence in these studies (Gould Reference Gould2002: p. 795). Recently developed phylogenetic methods incorporate stratigraphic data and are able to compare the support for hypotheses of anagenetic versus cladogenetic relationships among species (Wright et al. Reference Wright, Wagner and Wright2021). Although practical challenges remain, we are optimistic about future progress in this area.