Introduction
The characteristics of a community, such as its structure and composition, are at any given time the products of historical processes that are at play locally as the community is assembled and maintained (Ricklefs Reference Ricklefs1987; McPeek Reference McPeek2008). Two mechanisms in ecology that are often theorized to shape how communities assemble are environmental filtering and competitive exclusion, and determining the prevalence of each in natural systems has been the focus of many studies (Cavender-Bares et al. Reference Cavender-Bares, Ackerly, Baum and Bazzaz2004). Environmental filtering occurs when some taxa are excluded from a community because they lack traits that would allow them to sustain themselves under existing abiotic conditions. Competitive exclusion occurs when competition between ecologically similar taxa may be intense so as to prevent their coexistence in a community (Kraft et al. Reference Kraft, Adler, Godoy, James, Fuller and Levine2015). Though each process stresses the importance of either abiotic or biotic factors at the exclusion of the other, both outline avenues by which some taxa may be prevented from becoming established in a community while others are not. By examining how communities are structured and reflect the relative importance of these two processes, we can better understand what factors promote or prevent species coexistence and what drives spatial heterogeneity in biodiversity.
The increasing availability of genetic data has led to the development of new phylogenetic approaches to community ecology (Cavender-Bares et al. Reference Cavender-Bares, Kozak, Fine and Kembel2009). Applications emerging from the merging of these fields include the establishment of phylogenetic diversity as a guiding metric for conservation priorities (Faith Reference Faith1992) as well as indices and null models for characterizing phylogenetic community structure and invasion dynamics (Procheş et al. Reference Procheş, Wilson, Richardson and Rejmánek2007). Early work in defining how communities might assemble within a phylogenetic framework laid out the expectation that environmental filtering and competitive exclusion would produce opposite phylogenetic structures (Webb Reference Webb2000; Webb et al. Reference Webb, Ackerly, McPeek and Donoghue2002). Under environmental filtering, ecologically similar taxa would have similar chances of becoming established in a community or being excluded, depending on their shared abiotic tolerances. As such, taxa occupying communities strongly impacted by environmental filtering would appear phylogenetically clustered or more closely related than expected by chance (underdispersed). Conversely, under competitive exclusion, ecologically similar taxa would be less likely to coexist because of the competitive pressure they would exert on one another. Taxa occupying communities strongly impacted by competitive exclusion would, therefore, appear more phylogenetically evenly spread or less closely related than expected by chance (overdispersed). We now recognize that the proposed dichotomy is an oversimplification in light of coexistence theory (Chesson Reference Chesson2000), as competition may produce both phylogenetic over- and underdispersion depending on average fitness differences between taxa (Mayfield and Levine Reference Mayfield and Levine2010). A pattern of overdispersion is still thought to signal a strong effect of competitive exclusion on how a particular community was assembled, though interpretations continue to be refined. Additionally, underlying most of these new methods is the assumption that phylogenetic distance can act as a proxy for ecological similarity, based on the observation that functional traits tend to be phylogenetically conserved (Wiens and Graham Reference Wiens and Graham2005), although this may not always be the case (Losos Reference Losos2008).
To date, studies of phylogenetic community assembly have focused on modern communities, with some exceptions (e.g., Fraser et al. Reference Fraser, Gorelick and Rybczynski2015). This can be problematic when investigating assembly processes, as signals obtained from modern communities are frequently influenced by human activity (Zalasiewicz et al. Reference Zalasiewicz, Williams, Haywood and Ellis2011; Lim et al. Reference Lim, Crawley, De Vere, Rich and Savolainen2014). The fossil record documents the complete aftermath of colonization events and can help establish regional source pools of successful and unsuccessful colonizers (Zenni and Nuñez Reference Zenni and Nuñez2013; Jackson and Blois Reference Jackson and Blois2015). Furthermore, the abundance and high preservation potential of some morphologically conservative groups, such as bivalves, allow for many recent fossils to be closely associated with extant lower taxonomic units. This opens up unique opportunities to integrate observations from prehistoric immigrations, extirpations, and consequent evolutionary and ecosystem responses from the fossil record with ecological and genetic trait data obtained from extant organisms (Ricklefs Reference Ricklefs1987; Fritz et al. Reference Fritz, Schnitzler, Eronen, Hof, Böhning-Gaese and Graham2013; Jackson and Blois Reference Jackson and Blois2015).
This study examines the phylogenetic community structure in the San Joaquin Basin (SJB), a restricted embayment that persisted for approximately 24.5 Myr in the late Cenozoic in what is now part of the southern half of the Central Valley in California. At times covering a large geographic extent (175 km long by 100 km wide), the SJB had limited faunal and hydrological exchange with the adjacent coastal marine environment through a relatively narrow strait, which was largely controlled by eustatic sea-level cycles, subsidence and infill of the basin, and seismic activity stemming from movement along the San Andreas Fault (Bartow Reference Bartow1991; Bowersox Reference Bowersox2005). Previous work has reconstructed a number of freshwater sources entering the basin from the surrounding uplifted ranges (Stanton and Dodd Reference Stanton and Dodd1997) and indicates that the SJB had generally brackish waters, with more marine conditions during sea-level highstands.
Following peak late Cenozoic global temperatures during the mid-Miocene, roughly 17–14 Ma (Zachos et al. Reference Zachos, Pagani, Sloan, Thomas and Billups2001), global temperatures steadily declined, leading to increased provinciality in mollusks on the North American Pacific coast and overall southward shifts in the geographic ranges of most of those taxa (Hall Reference Hall2002). These shifts would have brought new faunas in contact with the semirestricted embayment of the SJB. Multiple pulses of colonization and extirpation documented in the SJB throughout this period of time have been associated with sea-level highstands and lowstands, respectively, and suggest ongoing opportunities for environmental or biotic filtering to occur (Bowersox Reference Bowersox2005). Indeed, the distinctiveness of the SJB fauna compared with the coastal marine fauna has been noted, as well as the persistence of a relictual warm-water fauna within the SJB (Hall Reference Hall2002; Bowersox Reference Bowersox2005). These observations indicate incomplete biotic mixing between the two environments and suggest the presence of barriers to establishment. Differential responses to the brackish or near-marine environmental conditions of the SJB or competition with the persisting fauna within the basin could lead to such barriers, which may be reflected in the phylogenetic structure of the SJB community. While the SJB and nearby coastal region have been the subject of biogeographic analyses, no treatment of this system has been done within an explicit phylogenetic framework.
Here, we apply methods in phylogenetic community ecology to the late Cenozoic fossil record of bivalves along the California Pacific coast. We focus on bivalves because of their rich fossil record in this system and the high proportion of fossil bivalve genera that have living representatives. We leverage the availability of modern genetic sampling with the fossil record of a large-scale natural experiment to examine whether the SJB community is phylogenetically over- or underdispersed and whether new colonizers are more or less related to native fauna than expected by chance. Studies such as this one have the potential to provide insight into our ability to detect phylogenetic community structure in the fossil record and the mechanisms by which taxa may be filtered.
Methods
Fossil Community Dataset
Presence data for the California coastal region during the late Cenozoic were obtained from Hall (Reference Hall2002), a compendium of marine invertebrates along the Californian coast from the late Oligocene (27 Ma) to the late Pliocene (2.5 Ma). The compendium notes latitudinal ranges for species and whether the species can be found east and west of the San Andreas Fault for each of six temporal bins (27–23, 23–17, 17–13, 13–8, 8–5, and 5–2.5 Ma). Analyses were conducted at the genus level to reduce sensitivity to taxonomic uncertainty, uneven taxonomic efforts, and uneven sampling across time. Working at the genus level also, crucially, allows for many extinct species to be associated with molecular data from closely related extant species, broadly assuming that genera are monophyletic. Because using higher taxonomic levels ignores ecological variation in congeneric species, there is a risk that environmental filtering and competitive exclusion operating at the species level will be masked. However, several ecological traits have been found to be moderately to highly conserved within genera (e.g., Hopkins et al. Reference Hopkins, Simpson and Kiessling2014; see “Discussion”), and thus barriers to establishment in a community contingent on species ecology should be similarly reflected at the genus level.
We considered a genus whose range includes the region east of the San Andreas Fault as present in the SJB community. For each time bin, genera were assigned to one of three categories: (1) genera known only outside of the SJB, or west of the San Andreas Fault, (2) genera that remained present within the SJB since the previous time bin, or (3) genera that are present in the SJB but not present in the previous time bin, henceforth referred to as “colonizers.” Because it is unclear from this dataset whether taxa in the earliest time bin (27–23 Ma) persisted in the SJB from a previous time or were colonizers, that time bin was omitted from any analyses that required identification of colonizing taxa.
Phylogenetic Inference
To assess the relationships among taxa in this study, all available sequence data matching genera in the fossil dataset were downloaded from GenBank (accessed 27 January 2020; Benson et al. Reference Benson, Karsch-Mizrachi, Lipman, Ostell and Sayers2011) for five genes: two nuclear ribosomal subunits (18S and 28S), one mitochondrial ribosomal subunit (16S), one mitochondrial cytochrome c oxidase subunit I (COI), and one nuclear protein–encoding gene (histone H3). Additional sequence data were downloaded for one gastropod genus (Haliotis), which served as the outgroup. For each genus, we retained the longest returned sequence per gene. However, if returned results included species that matched species present in the fossil dataset, the longest sequence per gene for those taxa was preferentially retained. We aligned sequences from each gene with MUSCLE (v. 3.7; Edgar Reference Edgar2004) using a ClustalW sequence-weighting scheme and UPGMB clustering method.
Phylogenetic relationships among bivalve taxa were reconstructed using a maximum likelihood (ML) approach in RAxML (v. 8.2.9; Stamatakis Reference Stamatakis2014). The concatenated alignment was partitioned by gene with the GTR + G model assigned to each partition. Aligned sequences belonging to members of the same genus were concatenated to reduce the amount of missing data. Concatenating species in the same genus also increases overlap between genera represented in the fossil dataset and the tree, but assumes that sampled genera are monophyletic. Topological support was determined using 1000 nonparametric rapid bootstrap replicates.
Divergence times were estimated using the Bayesian program MCMCtree in PAML (v. 4.8a; Yang Reference Yang2007), using the ML tree from RAxML as a guide tree. Eight calibration points based on fossil data were used to constrain the minimum age of the following clades: Bivalvia (bound: 530–520.5 Ma), Anomalodesmata (truncated Cauchy: 483.45 Ma), Arcida (truncated Cauchy: 480.2 Ma), Cardiidae (truncated Cauchy: 216 Ma), Mactroidea (truncated Cauchy: 125 Ma), Ostreoidea (truncated Cauchy: 245 Ma), Palaeoheterodonta (truncated Cauchy: 474.5 Ma), and Tellinoidea (truncated Cauchy: 199.3 Ma) (see Bieler et al. [Reference Bieler, Mikkelsen, Collins, Glover, González, Graf, Harper, Healy, Kawauchi, Sharma, Staubach, Strong, Taylor, Tëmkin, Zardus, Clark, Guzmán, McIntyre, Sharp and Giribet2014] for descriptions of each). Dating analyses were run under an HKY + G model of nucleotide substitution with independent rates. Two analyses were run until 60,000 samples were obtained from the posterior. Convergence was assessed using the R package RWTY (v. 1.0.2; Warren et al. Reference Warren, Geneva and Lanfear2017).
Phylogenetic Community Structure
The goal of this study was to determine whether the SJB community exhibited phylogenetic structure that deviated from that expected due to random assembly in any of the six designated late Cenozoic time bins. To do this, we characterized the phylogenetic community structure in the SJB system to address two objectives. The first objective was to determine whether the SJB community represents a nonrandom subset of the available taxon pool. We considered the source pool to contain all taxa known in our dataset for that time bin and calculated metrics of phylogenetic community structure for the SJB community subset. The second objective was to determine whether colonizers were more or less closely related to native taxa than expected by chance. For each time bin, taxa that persisted in the SJB since the previous time bin were considered native and fixed in the community, and the source pool composition consisted of all other taxa in the region. Metrics of phylogenetic community structure were then calculated for the colonizers only.
We used three established metrics to measure the degree to which membership in the SJB community or colonizer status is phylogenetically over- or underdispersed in a particular time bin. Terminal branches were truncated so as not to extend past the most recent boundary age of that bin. Genera that were not recorded as present during the time bin of interest were pruned from the tree to restrict the available source pool to taxa known to be proximate to the SJB. The first two metrics, the net relatedness index (NRI) and nearest taxon index (NTI), summarize community structure by comparing observed branch lengths to branch lengths produced by a null model of random assembly (Webb Reference Webb2000; Webb et al. Reference Webb, Ackerly, McPeek and Donoghue2002). The NRI is defined as
$${\rm NRI} = {-}{\rm 1 \,\ast } \,\displaystyle{{{\rm MP}{\rm D}_{{\rm obs}}-{\rm mean\lpar MP}{\rm D}_{{\rm random}}\rpar } \over {{\rm SD\lpar MP}{\rm D}_{{\rm random}}\rpar }}$$where MPD is the mean phylogenetic distance, or the mean total branch length, between all pairs of coexisting taxa in a community. NRI is the standardized effect size of MPD. In the case where our focus was on the SJB community as a whole, MPDobs was the MPD between all taxa recorded in the SJB, and MPDrandom was the MPD between those same taxa following the randomization of their positions on the tips of the tree and, thus, the randomization of the pairwise distances. In the case where our focus is on the colonizers alone, native taxa are fixed in the tree, and all other taxa not known in the SJB in the previous time bin comprise the source pool. MPDobs is the MPD between each observed colonizer and the native taxa, and MPDrandom is the MPD when colonizers were sampled from the source pool at an equivalent number to the number of colonizers observed, with each potential colonizer receiving equal chance of inclusion.
A similar metric to the NRI, the NTI is defined as
$${\rm NTI} = {-} \, {\rm 1 \, \ast } \,\displaystyle{{{\rm MNT}{\rm D}_{{\rm obs}}-{\rm mean\lpar MNT}{\rm D}_{{\rm random}}\rpar } \over {{\rm SD\lpar MNT}{\rm D}_{{\rm random}}\rpar }}$$where MNTD is the mean nearest taxon distance, or the total branch length, between each taxon and its closest relative in the community. NTI is the standardized effect size of MNTD. MNTDobs and MNTDrandom were calculated for the same observed and random communities as described for MPDobs and MPDrandom.
For each time bin, 1000 communities were simulated to determine the means and standard deviations of MPDrandom and MNTDrandom values. Positive values of NRI and NTI indicate co-occurring taxa are overall more closely related than expected by random inclusion in the SJB (clustered or underdispersed), while negative scores indicate co-occurring taxa are less closely related than expected from random assembly (evenly spread or overdispersed). While NRI provides a measure of how over- and underdispersed members of a community are across the whole phylogeny, NTI is more sensitive to patterns closer to the tips. Significance of NRI and NTI was assessed using Wilcoxon rank-sum tests and rankings among simulated values.
A third metric, D, uses independent contrasts and a null model of trait evolution under Brownian motion to yield a measure of phylogenetic signal for binary traits (Fritz and Purvis Reference Fritz and Purvis2010). This method was favored over other measures of phylogenetic signal (e.g., Pagel's λ), as most other measures currently in use are suitable for continuous traits only. As with NRI and NTI, we calculated D for each time bin both for the entire SJB community, using a tree including all taxa known in the region and treating presence in the SJB as a discrete trait, and for colonizers only, using a tree of only SJB occupants and treating status as a new colonizer as a discrete trait. A D value greater than 1 indicates the trait distribution is less clustered than expected from random (overdispersed), a D value equal to 1 indicates the trait distribution is indistinguishable from random, a D value of 0 indicates the trait distribution is as expected under a Brownian motion model of evolution, and a D value less than 0 indicates the trait distribution is more phylogenetically conserved than expected under Brownian motion (underdispersed). Though occupation in a community is unlikely to be thought of as a trait that evolves, as this approach models, phylogenetic distance is frequently treated as a proxy for ecological similarity. Thus, detection of a strong phylogenetic signal in community occupancy may point to structure in underlying, associated ecological traits. The probabilities that observed values of D differ significantly from random or Brownian motion expectations were determined using 1000 random permutations of the trait on the tree tips. Calculations of D and permutations were conducted using the R package caper (v. 1.0.1; Orme et al. Reference Orme, Freckleton, Thomas, Petzoldt, Fritz, Isaac and Pearse2018).
Phylogeny construction using MUSCLE and RAxML was conducted through the CIPRES gateway (Miller et al. Reference Miller, Pfeiffer and Schwartz2010). Unless otherwise stated, all remaining analyses were conducted in the R programming environment (v. 3.4.0; R Core Team 2017).
Results
Of the 199 bivalve genera present in the fossil SJB community dataset, 188 are extant, and 135 of these genera had matching sequences in GenBank that were included in the phylogeny estimation (Supplementary Table S1). Following alignment, concatenated sequences consisted of 6888 nucleotides (Table 1) and 67% missing data. The resulting ML tree is well resolved, with 65% of nodes with bootstrap support greater than 50 (Fig. 1). Clades associated with calibration points were recovered as monophyletic in the ML tree, and most of the reconstructed relationships between major clades are in agreement with previously published phylogenies (Sharma et al. Reference Sharma, González, Kawauchi, Andrade, Guzmán, Collins, Glover, Harper, Healy, Mikkelsen, Taylor, Bieler and Giribet2012; Bieler et al. Reference Bieler, Mikkelsen, Collins, Glover, González, Graf, Harper, Healy, Kawauchi, Sharma, Staubach, Strong, Taylor, Tëmkin, Zardus, Clark, Guzmán, McIntyre, Sharp and Giribet2014; González et al. Reference González, Andrade, Bieler, Collins, Dunn, Mikkelsen, Taylor and Giribet2015). Though many genera included in this study are known to be speciose, the median species-to-genus ratio for taxa on our tree ranges between 1 and 3 species across the time bins, likely due in part to the restricted geographic scope of this study (Supplementary Fig. S1).
Time-calibrated molecular phylogeny (left) and community membership (right) for bivalve genera along the late Cenozoic Californian Pacific coast. Relationships were reconstructed in RAxML using a maximum likelihood analysis of five genes (18S, 28S, 16S, COI, and H3). Divergence times were estimated using Bayesian methods in MCMCtree. Black dots indicate fossil calibration points, and asterisks (*) indicate node support greater than 75 based on 1000 bootstrap replicates. Colors on the right indicate San Joaquin Basin (SJB) community membership for each genus during six Cenozoic time bins (27–23, 23–17, 17–13, 13–8, 8–5, and 5–2.5 Ma) from Hall (Reference Hall2002). In each column, blue boxes indicate genera known only outside of the SJB (i.e., west of the San Andreas Fault) during that time bin, orange boxes indicate genera with continued presences in the SJB (i.e., east of the San Andreas Fault) from the previous time bin, and red boxes indicate genera that are newly colonizing the SJB. Unfilled boxes indicate genera that are not recorded in the SJB or adjacent region during that time bin. (Color online.)
Number of included genera and sequence length for each partition used to generate a time-calibrated maximum likelihood tree.
Null distributions for mean phylogenetic distance remained consistent over time when examining both the SJB community and colonizers only, indicating representation of deep nodes did not undergo dramatic increases or decreases in the Pacific system as a whole (Fig. 2). When considering all taxa in the SJB community, the two phylogenetic distance-based metrics, NRI and NTI, were not statistically different from the null expectation generated by a process of random assembly using two-tailed Wilcoxon rank-sum tests (Fig. 3, Table 2). The observed value of the NRI in the 8–5 Ma time bin ranked in the lowest 2.5% of all values obtained from 1000 simulated communities, even when simulation sets were repeated multiple times. This drop in the NRI is not reflected to a similar degree in the NTI for the same time bin. When looking at whether colonizers were more or less closely related to native SJB taxa than noncolonizing taxa, we found that observed NRI and NTI values were not significantly different from values obtained if colonizers were drawn from the available taxon pool at random.
Null distributions and observed values of mean phylogenetic distances (MPD) and mean nearest taxon distances (MNTD) for the San Joaquin Basin community and for new colonizers across six time bins. Null values were obtained from 1000 simulated communities generated under a model of random community assembly. Diamonds indicate the median, gaps indicate first and third quartiles, and lines extend to 1.5 times the interquartile range. Black dots indicate observed values.
Net relatedness index (NRI) and nearest taxon index (NTI) for the San Joaquin Basin community and for new colonizers across six time bins. Asterisks (*) indicate NRI or NTI values ranked in upper 2.5% or lower 2.5% of values for 1000 simulated communities. No NRI or NTI values were found to differ significantly from zero based on two-tailed Wilcoxon rank-sum tests. The y-axis has been reversed for consistency with Fig. 4.
Net relatedness index (NRI), nearest taxon index (NTI), and associated measures of significance for the San Joaquin Basin community and for new colonizers across six time bins. p-values indicate no significant difference between observed values of mean phylogenetic distance (for NRI) and nearest taxon distance (for NTI) and values obtained from 1000 communities simulated through random assembly using two-tailed Wilcoxon rank-sum tests. Rank indicates the ranked position of observed values among simulated communities.
In most time bins, observed values of Fritz and Purvis's D significantly differed from expected D values under a Brownian motion model of evolution, but not from expected D values expected if community membership was randomly distributed across phylogeny tips (Fig. 4, Table 3). When examining all members of the SJB community, D was found in two time bins (27–23 and 8–5 Ma) to be significantly different from random but not significantly different from Brownian motion. When examining colonizers only, we could not distinguish the observed distribution of inclusion in the SJB across the tree from distributions resulting from random processes or trait evolution under Brownian motion in the 23–17 Ma time bin. Under this approach, no D value was found to be significantly greater than 1 (overdispersed) or significantly less than 0 (underdispersed).
Phylogenetic signal, D, of colonization success for the San Joaquin Basin community and for new colonizers across six time bins. D > 1 when trait is overdispersion, D = 1 when trait is randomly distributed, D = 0 when trait evolves Brownian motion, and D < 0 when trait is underdispersion. Asterisks (*) above and below the dot indicate values that are significantly different from random expectation and Brownian motion expectation, respectively. Significance was assessed through 1000 permutations. See Table 3 for D and p-values.
Phylogenetic signal, D, of colonization success for the San Joaquin Basin community and of new colonizers across six time bins. p-values, p random and p BM, less than 0.05 indicate observed values of D differ significantly from random (D = 1) or from Brownian expectation (D = 0), respectively. Significance was assessed using distributions of values generated from 1000 permutations.
Discussion
Early frameworks for interpreting phylogenetic community structure presented a dichotomy in which patterns of over- and underdispersion were attributed directly to the processes of competitive exclusion and environmental filtering, respectively. Attempts to reconcile phylogenetic community ecology and modern coexistence theory have recognized that environmental filtering and competitive exclusion can in fact produce similar phylogenetic patterns of underdispersion (Chesson Reference Chesson2000; Mayfield and Levine Reference Mayfield and Levine2010). Thus, overdispersion would be observed only if competitive exclusion overwhelms the combined effects of weak competition and environmental filtering. Furthermore, environmental filtering and competitive exclusion do not necessarily act independently. Not only do both processes operate concurrently (McClain et al. Reference McClain, Stegen and Hurlbert2012), but the frequency or intensity of species interactions such as competition may increase across an environmental gradient. Examples of this type of complex relationship have been detected in modern ecosystems (Callaway et al. Reference Callaway, Brooker, Choler, Kikvidze, Lortie, Michalet, Paolini, Pugnaire, Newingham, Aschehoug, Armas, Kikodze and Cook2002).
Our results show the SJB community was neither phylogenetically over- nor underdispersed across the late Cenozoic. The lack of phylogenetic structure indicates competitive exclusion did not play an overwhelmingly strong role in shaping the composition of the SJB community. We find one possible exception to the general lack of structure from 8 to 5 Ma, when observed values of D indicate taxa within the SJB community were more phylogenetically clumped than expected from random assembly, though not to the point of underdispersion. This is reflected, as well, in a low community NRI value for that time bin. However, determining whether the SJB community was the result of random assembly or a mix of processes that acted in opposing directions requires an in-depth understanding of how different ecological traits and strategies function in the specific physical conditions of the semirestricted SJB embayment and the biotic environment with which new colonizers may have had to contend.
Previous studies, particularly on invasive species, have associated successful establishment with both distant relatedness (Strauss et al. Reference Strauss, Webb and Salamin2006; Schaefer et al. Reference Schaefer, Hardy, Silva, Barraclough and Savolainen2011; Park and Potter Reference Park and Potter2013) and close relatedness (Duncan and Williams Reference Duncan and Williams2002; Li et al. Reference Li, Cadotte, Meiners, Hua, Shu, Li and Shu2015) with native taxa. Several factors make identifying consistent patterns of over- or underdispersal across multiple studies difficult. One factor is the degree to which the underlying assumption that phylogenetic relatedness is a direct proxy of ecological similarity is supported. The relationship between phylogenetic relatedness and ecological similarity is a complex one requiring careful consideration on a case-by-case basis (Losos Reference Losos2008). Observed values of D in the SJB community indicate that, in most time bins, inclusion in the SJB community was not distributed across the tree in a way that significantly differed from random assignments. This observation suggests that, if there were underlying ecological traits associated with successful occupation in the SJB, those traits either are not strongly phylogenetically conserved or only weakly influenced the chances of becoming established. Previous studies have, however, identified some traits linked to invasion potential in bivalves. For example, North American Pacific coast bivalves with larger body sizes appear more likely to become established outside their initial ranges since the Pleistocene (Roy et al. Reference Roy, Jablonski and Valentine2001, Reference Roy, Jablonski and Valentine2002). In recent, human-mediated introductions into the Mediterranean Sea, successful invaders from the Red Sea were larger than native taxa (Nawrot et al. Reference Nawrot, Albano, Chattopadhyay and Zuschin2017). Because we know body size to be phylogenetically conserved in bivalves (Saulsbury et al. Reference Saulsbury, Moss, Ivany, Kowalewski, Lindberg, Gillooly, Heim, McClain, Payne, Roopnarine and Schöne2019), these patterns suggest colonizers may be more closely related to one another than expected from random. However, our study does not find that to be the case. This may be explained by a number of factors that warrant further examination, including the lower temporal resolution we used or, in fact, the specific dynamics of the SJB fauna and environment.
Direct tests for phylogenetic signal in ecological traits only improve interpretations of phylogenetic community structure. Recently, time-calibrated phylogenies of Bivalvia have begun to encompass the taxonomic coverage required to measure phylogenetic signal of ecological traits across the whole clade. One recent study noted that growth rates appear to be phylogenetically conserved (Saulsbury et al. Reference Saulsbury, Moss, Ivany, Kowalewski, Lindberg, Gillooly, Heim, McClain, Payne, Roopnarine and Schöne2019). Additional studies of bivalve subclades have also demonstrated conservatism in some ecological traits, such as commensal and free-living lifestyles and associated morphologies in extant bivalves of the clade Galeommatoidea (Li et al. Reference Li, Foighil and Strong2016). Furthermore, substrate affinity appears conserved within genera across marine invertebrate groups (Hopkins et al. Reference Hopkins, Simpson and Kiessling2014), and body size and geographic range of paraphyletic morphogenera are broadly congruent with those of the crown group (Jablonski and Finarelli Reference Jablonski and Finarelli2009). In bivalves, geographic range size appears to be conserved across genera (Roy et al. Reference Roy, Hunt, Jablonski, Krug and Valentine2009b), though how closely it reflects species range size varies across space and is tied to macroevolutionary rates (Tomašových and Jablonski Reference Tomašových and Jablonski2017). These studies suggest that ecology is conserved within and across higher taxonomic levels and that it is not unreasonable to expect that filtering processes may be detectable above the species level. Relatively low species-to-genus ratios also alleviate concerns that genus ranges do not broadly reflect species ranges in the geographic context of our study (Supplementary Fig. S1).
Several studies have documented strong phylogenetic signal in how extinctions have been distributed across the tree of life in several diverse clades, including bivalves, with many of these studies detecting phylogenetic clustering (Roy et al. Reference Roy, Hunt and Jablonski2009a; Fritz and Purvis Reference Fritz and Purvis2010; Hardy et al. Reference Hardy, Fara, Laffont, Dommergues, Meister and Neige2012; Harnik et al. Reference Harnik, Fitzgerald, Payne and Carlson2014). The disconnect between the presence of structure in extinctions and the lack of structure we observed in community occupation suggests the underlying traits most frequently associated with extinction risk and the underlying traits most frequently associated with successful establishment in novel environments are decoupled.
Furthermore, consideration of traits provides a deeper understanding of how strongly associated an organism is with the resources (such as food and predator-free space) and environmental conditions locally present. In such cases where the association is weak or where the number of ecological roles is low or imbalanced, we would not expect competition and environmental filtering to lead to differential rates of establishment. Thus, the interpretation that random community assembly results from neutral interactions can be corroborated through paleoecological studies. The strength of association must, necessarily, be assessed on a clade by clade basis, but even localized understanding contributes toward the aim of community phylogenetics, which is to identify the presence and strength of filtering processes, using phylogeny as a proxy for ecology. Similarly, the inclusion of trait data provides valuable additional information that can help identify filtering processes not explained through relatedness metrics (Thuiller et al. Reference Thuiller, Gallien, Boulangeat, De Bello, Münkemüller, Roquet and Lavergne2010; Schaefer et al. Reference Schaefer, Hardy, Silva, Barraclough and Savolainen2011; Marx et al. Reference Marx, Giblin, Dunwiddie and Tank2016). Plans for future work include integrating trait data with the community and phylogenetic datasets for the taxa specifically examined in this study to better understand the relationship between ecology and phylogeny in our study system.
Another factor that complicates the search for consistent support for phylogenetic over- or underdispersal is the use of different spatial and temporal scales across studies. Meta-analyses of invasion studies have found that the frequency at which a phylogenetic signal can be detected, and the direction of that relationship, is sensitive to the spatial scale used (e.g., Procheş et al. Reference Procheş, Wilson, Richardson and Rejmánek2007; Thuiller et al. Reference Thuiller, Gallien, Boulangeat, De Bello, Münkemüller, Roquet and Lavergne2010; Ma et al. Reference Ma, Li, Pu, Tan, Liu, Zhou, Li and Jiang2016). Other meta-analyses have found that relatedness may better predict invasion success at larger spatial scales than smaller ones (Gallien and Carboni Reference Gallien and Carboni2017). Thus, caution must be taken when comparing measures of phylogenetic structure across studies at different spatial scales. Our use of a consistent spatial scale allowed us to compare metrics of phylogenetic structure across time bins, though we recognize that these measures are non-independent across time and are reluctant to draw strong conclusions from first-order differences. Furthermore, the coarse spatial scale used in this study treated the SJB as one community, ignoring environmental heterogeneity within the embayment and potentially masking strong environmental filtering taking place on very local scales. More data is necessary to determine the degree to which this treatment impacts our findings.
The temporal scale of this study introduces unique benefits and complications not present in similar studies of extant communities. In studies of modern systems, the available species pool is often reconstructed by spatial averaging, assuming the presence of unobserved taxa, or aggregating repeated observations across historic timescales (Cornell and Harrison Reference Cornell and Harrison2014). Likewise, communities are often defined by observations of the biota at a given locality, possibly over the course of years or decades. These approaches could be problematic in situations, such as when observed taxa eventually fail to become established, when important environmental variability is not captured within the time frames studied, when regional pools are dynamic, and when community members evolved in response to their changing biotic or abiotic environment. Time-averaging in paleontological datasets can provide a more complete view of the available species pool and communities on longer timescales (Jackson and Blois Reference Jackson and Blois2015). However, time-averaging introduces challenges as well. For example, colonizers may arrive staggered over the course of one time bin lasting several million years, each impacting the biotic environment for subsequent colonizers (Fukami Reference Fukami2015). Such priority effects are unrecognizable without historical information but are similarly difficult to detect with highly time-averaged data. Still, paleontological evidence provides a key perspective for chronicling colonization events and environmental changes at evolutionary timescales (Mergeay et al. Reference Mergeay, De Meester, Eggermont and Verschuren2011), a necessary step toward determining how temporal resolution impacts our ability to detect community assembly processes.
Conclusions
This study presents one example of how modern phylogenetic and ecological theory may be applied to fossil systems with the goal of ascertaining how historic processes led to the coexistence of a specific subset of taxa. We combined the rich molluscan fossil record with molecular data to ask whether the SJB community was phylogenetically structured at multiple points across the late Cenozoic and whether colonizers of the SJB were more or less related to the native fauna than expected by chance. In both cases, we found little evidence of strong phylogenetic signal, indicating competitive exclusion was not the overwhelming force shaping community composition in the SJB and suggesting that the SJB community was the result of either random assembly processes or a mix of competitive exclusion and environmental filtering. As the field of community phylogenetics continues to develop, there is great potential to integrate the historical perspective offered by the fossil record into a theoretical framework for how filtering mechanisms impact phylogenetic structure across multiple spatial and temporal scales.
Acknowledgments
We would like to thank J. Lamsdell and C. Congreve for the invitation to contribute to the Paleobiology special issue on phylogenetic paleoecology. We are thankful to D. Fraser and one anonymous reviewer for comments that greatly improved the quality of this article; J. Lim and B. Muddiman for valuable discussions; and C. Marshall, D. Lindberg, and R. Gillespie for comments on an early version of this article. This work was supported by the University of California Museum of Paleontology, National Science Foundation Graduate Research Fellowship (grant no. DGE-1106400), and the Smithsonian Institution National Museum of Natural History Peter Buck Deep Time Fellowship.
