Fossil samples archive functional diversity in marine ecosystems: An empirical test from a present-day coastal environment

Significance

Fossils, the primary archive of ancient communities, are used to understand how healthy, unaltered ecosystems functioned in the past. However, the reliability of the fossil record remains uncertain. We compared living marine communities with dead skeletal remains to understand how well fossils capture functional diversity for a wide range of animals. We find that dead remains and fossils faithfully record lifestyles and feeding modes of invertebrate communities, and provide an excellent record of functional structure. This suggests that functional diversity measures based on fossils can be used to study ecosystem evolution, extinction recovery, and factors affecting ecosystem health and functioning. Historical perspectives can inform current conservation efforts, augmenting our understanding of ecological processes operating over a variety of time scales.

Abstract

The fossil record of functional diversity is increasingly used to study ecosystem evolution, extinction recovery, and factors affecting long-term trends in biodiversity. In addition, the youngest fossil record (late Quaternary) can provide insights into the natural range of functional variability of present-day ecosystems, providing a historical framework for conservation and restoration. However, the reliability of common functional diversity measures derived from fossils is uncertain. If fossils yield reasonable estimates of functional diversity, paleontological data could provide information on ecological attributes and trophic structure in past ecosystems allowing for temporally scalable assessments of ecological and evolutionary processes. To assess how well fossils preserve functional diversity across multiple types of marine invertebrates with varying preservation potential, we compared the live benthos (135 species from 6 phyla) with sympatric skeletal accumulations (150 species) and the predicted fossil record (112 species) for 51 coastal sites in North Carolina (USA). High functional fidelity between the live, dead, and fossil assemblages was supported by congruence in quantitative functional diversity indices (e.g., functional richness, redundancy, overredundancy, and vulnerability), overlap in multidimensional functional space, and species distributions among functional groups (ρ > 0.85, P ≪ 0.001). Calculating vulnerability using a threshold of ≤2 species also reasonably approximated the vulnerability in the live assemblages. The results suggest that, despite differential preservation and time-averaging, functional estimates based on fossils may allow for historical assessments of ecological and evolutionary processes, including short-term community responses to human impacts as well as long-term macroevolutionary dynamics of marine ecosystems.

Although ecosystem health has traditionally been assessed using taxon-based approaches, it is increasingly recognized that functional diversity, the number and types of functions in a community (1), may be a better predictor of ecosystem health than species-based estimates (1–6). This is because functional diversity influences ecosystem dynamics, stability, productivity, and functioning (2, 4), and can thus be used to predict ecological change, assess management strategies, evaluate human impacts, and identify reference conditions (7–11). However, to better forecast ecosystem response and effectively manage resources, we must first understand how ecosystems respond to extreme change; yet there are no precedents in the human experience to guide us (12). One obvious way to address this knowledge gap is to use the fossil record, which documents the timing, causes, and consequences of past ecosystem changes (13).

In recent years functional diversity has become increasingly utilized in paleontological research to understand factors affecting ecosystem functioning over various spatiotemporal scales (14–18). On evolutionary time scales, functional diversity can inform macroevolutionary hypotheses (e.g., Red Queen, Escalation Hypothesis, and the Mesozoic Marine Revolution), which require information across multiple groups of organisms and have been formulated and evaluated using taxon-based approaches (19–21). The fossil record can also potentially be used to identify mechanisms driving diversification and community organization and to assess the relationship between number of species and functional diversity, a question at the forefront of ecological research today (22). The youngest fossil record ($\le$2 Mya) is particularly relevant to modern conservation because it archives ecosystems that immediately preceded those that exist today (13, 23–25), and thus can be used to potentially identify human-driven loss of ecosystem functions vital for maintaining ecosystem health (14, 18). At the shortest timescales (decadal-to-millennial), the fossil record has been used to document undisturbed ecosystems prior to the onset of major human impacts (13, 26–29), and both fossil assemblages and surficial skeletal remains (death assemblages) have been increasingly employed to assess changes in trophic structure and functional diversity (e.g., refs. 30–39). High functional fidelity of paleontological data (i.e., a reasonably faithful archiving of functional groups in the fossil record) is a prerequisite for any studies utilizing functional diversity or ecological traits, such as lifestyles or feeding types, derived from the fossil record. However, the fidelity of functional diversity has received limited attention. Studies are sparse and focused on mollusks or theoretical modeling in the marine fossil record (but see refs. 40–42), and we lack quantitative assessments that encompass multiple major taxonomic groups and explicitly focus on functional diversity measures.

Here, we test the fidelity of functional diversity estimates by comparing living assemblages with sympatric skeletal remains accumulating on the sea-floor (death assemblages), and, secondarily, with a predicted fossil assemblage consisting of species with a documented fossil record. We employed multitaxic benthic macroinvertebrate assemblages (6 phyla and 51 sites) from Onslow Bay, North Carolina (USA). Onslow Bay offers a natural laboratory for comparative fidelity analyses because of its diverse and abundant macrobenthos. Both neontological (43) and geohistorical (44–46) studies suggest that the ecosystem has not been notably altered by human activities, which if present could make it difficult to differentiate human-induced and taphonomic discordances (44–46). Onslow Bay supports a spectrum of nearshore to soft-bottom shallow shelf habitats which are common across the globe today, frequently preserved in the fossil record (40), and increasingly recognized for their significant role in ecosystem functioning (47). This macrobenthic community consists of metazoan groups with varying biomineralization (gastropods, bivalves, scaphopods, annelids, decapods, and regular and irregular echinoids) common in present-day soft-bottom ecosystems (46), and dominant in the Cenozoic marine fossil record (40, 48).

The macrobenthos is the most studied system in marine functional ecology (49). Benthic communities, such as those of Onslow Bay, include abundant bioturbators, which enhance benthic-pelagic coupling and nutrient exchange between sediments and the overlying water column (50), as well as filter-feeders which play a major role in nutrient remineralization (47, 51), nutrient regeneration (52), the cycling of organic matter in sediments (53), water quality enhancement, and phytoplankton biomass control (47, 51, 52, 54, 55). Macrobenthic organisms have also been extensively researched as ecological indicators (56, 57) due to their varied stress-tolerance, and the sensitivity of benthic community structure to environmental change (58–60). In particular, mollusks, annelids, and arthropods (the three dominant taxa in Onslow Bay), are commonly used to monitor aquatic environments due to their wide distribution and relatively small foraging areas (61, 62).

Functional diversity in Onslow Bay was compared across three benthic invertebrate macrofaunal assemblages: 1) the Live Assemblage (LA); 2) the Death Assemblage (DA) consisting of sympatric skeletal material accumulating on the sea-floor; and 3) the Fossil Assemblage (FA) consisting of species in both assemblages with a known fossil record. In addition, we assessed the influence of geographic scale by comparing functional fidelity across the three assemblages (LA, DA, and FA) for the entire study area, with samples constrained to sites in three habitats: offshore, nearshore, and the backbarrier. We examined how much of the functional space was occupied and how species were distributed within functional space in the DA and FA relative to the LA. Functional fidelity was quantified using the presence or absence of species employing two widely applied trait-based approaches: functional entities, which are unique combinations of functional traits (e.g., life habit, motility, or feeding type), and Biological Traits Analysis (BTA) (63). Traits are directly associated with processes such as nutrient cycling, sediment transport, productivity, and trophic support (64), and can commonly be determined for fossil and modern species (14, 65). Species with similar traits are thought to perform similar functions, and differences in trait composition reflect differences in functional composition that relate to trophic structure, tiering, and animal–substrate interactions (SI Appendix, Tables S1 and S2). Traits can be shared by organisms from disparate groups and integrated over large geographic scales, making traits useful for paleoecology (3, 63, 66, 67). Species distributions among functional entities have important consequences for ecosystem functioning, and loss of species does not always equate to loss of functions (5, 68). High functional redundancy, similar functions performed by multiple species (69), can provide insurance and resilience, maintaining functional diversity despite species loss (70). Thus, BTA has become widely utilized in conservation and management to assess whole-ecosystem functioning (7), monitor functional redundancy, and identify vulnerable functional groups maintained by only a few species. These aspects of functional diversity are also quantifiable in the fossil record, and understanding how functional diversity responds to changes in biodiversity over centennial-to-millennial timescales can provide unique insights into ecosystem dynamics and resilience (e.g., refs. 71–73) in the face of unprecedented global change (74). Using the macrobenthos in Onslow Bay, we test whether taphonomic biases substantially alter functional space, quantitative measures of functional diversity, and the resulting functional interpretations. High functional fidelity would support the validity of studies that employ functional diversity approaches using the fossil record.

1. Results

Species in the DA and FA occupied nearly all of the functional entities present in the LA. There was a strong positive correlation between the number of species per functional entity across the LA and DA (Spearman’s $\rho =0.86$, P ≪ 0.001), LA and FA ($\rho =0.88$, P ≪ 0.001), and the DA and FA ($\rho =0.95$, P$\ll$ 0.001) (Fig. 1). Only one out of 17 functional entities in the LA was not occupied by species in the DA (pelagic, fully motile slow carnivore; SI Appendix, Fig. S2 and Table S3). For all three categories of trophic groups (feeding strategy, locomotion, and tiering), when compared to the LA, the number of species per trophic group within the DA and FA fell overwhelmingly within the expected 95% quantiles estimated using a resampling null model which assumes that all three compared assemblages came from the same (LA) assemblage (Fig. 2). The two most diverse functional entities in the LA were disproportionately occupied by mollusks, largely sessile suspension feeders and carnivorous gastropods, consistent with the distribution of species among functional entities in the DA and FA (Fig. 1).

Fig. 1.

Functional fidelity and the number of species in ecological guilds for the (A) Live Assemblage (LA), (B) Death Assemblage (DA), and (C) Fossil Assemblage (FA). Ecology was assigned based on feeding, motility, and tiering. Nearly all functional entities observed in the LA are also present in the DA and FA indicating high trophic fidelity. The distribution of richness among functional entities is also well preserved, and functional entities represented by many species in the LA are similarly represented in the DA and FA.

Fig. 2.

The number of species per trophic group in the LA relative to the DA and FA. The number of species (richness) within each functional entity is plotted for the DA and LA (A) and the FA and LA (B). The shaded regions show the 99%, 95%, and 50% quantiles from the distribution produced by a resampling null model which calculated the functional entity richness by sampling the LA species distribution into the actual sample structure of the LA, DA, and FA datasets (10,000 iterations). The black line indicates the median probability. The median probability line closely follows the y$=$x identity line (not shown on the plot because of the near-perfect overlap of the two lines). Abbreviations are as follows: chemosymbiote $=$ chemo, sessile attached $=$ ses-att, facultative slow $=$ facul-slow, sessile unattached $=$ ses-unatt, suspension feeder $=$ sus-feed, shallow infaunal $=$ shall-inf, deposit feeder $=$ dep-feed, fully motile slow $=$ full-mot-sl, and fully motile fast $=$ full-mot-fst.

As the abundance of species can be altered by taphonomic filters, we also examined the abundance of species among functional entities (SI Appendix, Figs. S2 and S3 and Table S3), which was positively and significantly correlated between the LA and DA ($\rho =0.56$, P = 0.03) with strong rank abundance agreement (SI Appendix, Fig. S4). The distribution of abundance among functional entities was similarly consistent between the LA and DA. Functional entities with relatively more individuals in the LA also had more individuals in the DA, and abundance in any given functional entity in the LA was largely comparable to that of the same functional entity in the DA with the exception of suspension feeders, which were notably more abundant in the DA.

In each assemblage, we examined three presence–absence based metrics commonly been used to quantify the number and distribution of traits: Functional redundancy, functional vulnerability, and functional overredundancy (69). These metrics utilize the number of species per functional entity and the total number of functions. When considering the entire study area the results were consistent for redundancy and overredundancy (Table 1).

Table 1.

Functional metrics

	All			Offshore			Nearshore			Backbarrier
	LA	DA	FA	LA	DA	FA	LA	DA	FA	LA	DA	FA
FEs	18	20	19	14	15	18	17	18	20	14	17	19
Sp.	122	138	130	45	75	77	87	116	116	70	116	109
No. single	4	9	8	6	7	6	5	7	6	6	7	6
Mn. Sim. Sgl.	-	1.5	1.5	-	1.7	1.6	-	1.5	1.3	-	1.3	1.3
Vulner.	22%	45%	42%	43%	47%	67%	30%	39%	60%	43%	41%	63%
Mn. Sim. Vln.	-	8%	8%	-	11%	11%	-	9%	7%	-	8%	8%
z	-	6.7	5.8	-	4.8	7.4	-	5.2	8.9	-	5.8	5.8
P	-	$<$0.001	$<$0.001	-	$<$0.001	$<$0.001	-	$<$0.001	$<$0.001	-	$<$0.001	$<$0.001
Red.	6.8	6.9	6.8	3.2	5.0	4.3	5.1	6.4	5.8	5.0	6.8	5.7
Over-Red.	52%	54%	57%	36%	55%	57%	51%	58%	60%	51%	56%	56%
Mn. Sim. Over-Red.	-	31%	31%	-	29%	29%	-	31%	31%	-	31%	31%
z	-	4.9	5.2	-	5.1	5.5	-	5.7	5.5	-	5.0	5.0
P	-	$<$0.001	$<$0.001	-	$<$0.001	$<$0.001	-	$<$0.001	$<$ 0.001	-	$<$0.001	$<$0.001

For each assemblage and habitat, the number of functional entities (FEs), the number of species (Sp.), number of functional entities containing only a single species (No. Single), functional redundancy (the number of species divided by the number of functional entities; Red.), the proportion of functional entities with only a single species (functional vulnerability; Vulner.), and the percentage of species that fill functional entities above the mean functional redundancy (functional overredundancy; Over-Red.) are shown. For the DA and FA, mean values of the number of single species per functional entity (Mn. Sim. Sgl.), vulnerability (Mn. Sim. Vln.), and overredundancy (Mn. Sim. Over-Red.) are shown based on the simulations (1,000 permutations), as well as the associated test statistics (z) and P-values (P) for vulnerability and overredundancy.

In all three assemblages, species were concentrated in only a few functional entities, and the number of species exceeded the mean number of species per functional entity (i.e., redundancy) in less than five functional entities (SI Appendix, Fig. S5). While the percentage of functional entities with only one species (i.e., vulnerability) was low in the LA, it increased notably in the DA and FA. Although not considered vulnerable, 53% of functional entities in the LA were represented by two or less species, an equivalent proportion to both the DA and FA (both equal to 53%). To determine whether the distribution of species among functional entities was an artifact of the number of species, we tested whether observed DA and FA values of vulnerability and overredundancy differed from a random distribution of species in functional entities, as these metrics are influenced by the number of species (see SI Appendix for additional details). Observed values of vulnerability and overredundancy in the DA and FA differed significantly when compared to a distribution of species randomly assigned to functional entities (in all comparisons P ≪ 0.001; Table 1). In all three assemblages a greater number of species accumulated in the highest ranking functional entities than would be expected if species were randomly distributed among functional entities (SI Appendix, Fig. S5). Similarly, the majority of functional entities had fewer species than expected if species were randomly distributed among functional entities.

Relative to the entire study area, the number of functional entities in the three habitat subsets were comparable, despite having fewer species (Table 1). Over redundancy was consistent in all three assemblages within habitats, although the range of values across assemblages within habitats was somewhat greater than that of the entire study area. Functional redundancy was lower in the LA for all three habitats relative to the DA and FA. In contrast, within habitats, vulnerability was generally more comparable within habitats in the LA and DA, but consistently higher in the FA. Observed values of vulnerability and overredundancy within habitats in the DA and FA differed significantly when compared to a distribution of species randomly assigned to functional entities (in all comparisons P$\ll$ 0.001; Table 1).

Multidimensional functional space was assessed using BTA, a multivariate approach which translates information on species distributions into units described by multiple biological traits. Functional space for all three assemblages was highly congruent (Fig. 3), and convex hulls for the DA and FA were only marginally contracted relative to the LA, regardless of habitat type. Species absent in the LA were dispersed throughout the DA and FA functional space, although species in the DA were concentrated in the Bottom Right corner, expanding up toward the Top-Left corner in the FA (SI Appendix, Fig. S6). However, the absence of these species in any given assemblage resulted in minor (DA) or moderate (FA) contraction of the occupied functional space (Fig. 3 and SI Appendix, Figs. S6–S9). The magnitude of the contraction is relatively trivial (especially for the DA) when contrasted to the variability observed among modern communities in other studies. For example, functional richness varied from approximately 0.6 to 0.9 across fish communities (69), whereas the LA and DA differed by only 0.08 despite the fact that this analysis included multiple phyla rather than a single taxon. Within habitats, similarly, the largest difference in functional richness is 0.3 between the LA and FA in the offshore habitat, which is still less than that across fish communities. Whereas this analysis is based only on the presence or absence of species, weighting species by their abundance produced comparable results (SI Appendix, Figs. S10–S13 and Table S5).

Fig. 3.

Distribution of species in multidimensional functional space for the three assemblages. Differences in functional richness across the three assemblages based on PCoA for the complete dataset (A–C), and the three types of habitats: offshore (D–F), nearshore (G–I), and backbarrier (J–L). The first column shows Live Assemblages (LA), the second column shows Death Assemblages (DA), and the third column shows Fossil Assemblages (FA). The relative positions of species (points) reflects their overall similarity in trait composition. The convex hulls marking the functional space filled by the respective assemblages are marked with a black line, while the functional space of the LAs are shown for reference in gray in the death and fossil assemblages. Species are color-coded by phylum, with phylum-level convex hulls color-coded by lighter shades of the same colors. Note that the LA occupies the entire functional space and that the DA and FA are only modestly reduced, and phyla occupy similar functional space in all three assemblages, regardless of habitat.

We compared the fidelity of a variety of multivariate functional metrics including functional richness, functional divergence, functional evenness, functional dispersion, functional specialization, and functional originality (Table 2 and SI Appendix, Table S5 and Fig. S6). These metrics, which quantify how functional space is filled, were all highly congruent across the three assemblages indicating that functional fidelity was high (SI Appendix, Fig. S6). Relative to the functional space occupied by the LA, all metrics were either equal to, or marginally lower for the death and fossil assemblages for the entire study area (Table 2). Mean values of the six metrics were highly consistent, varying among assemblages by $<$0.03 for all metrics except richness, which was depressed in the FA (0.57) relative to the LA (0.78). The functional richness of the death and fossil assemblages were unlikely to be a random artifact of the number of species as species were significantly clustered in functional space for both the DA (z $=-2.76$, P$=$ 0.99, i.e., 1% of the simulated values were lower than the observed value) and FA (z $=-2.03$, P$=$ 0.98, i.e., 2% of the simulated values were lower than the observed value). Metrics were also largely comparable within habitats (SI Appendix, Figs. S7–S9) varying among assemblages by $<$0.05 for all metrics except functional richness. In nearshore and backbarrier habitats, functional richness was elevated in the DA and FA relative to the LA. These metrics were calculated using only the presence or absence of species to quantify the functional space, however, weighting species by their abundance did not greatly alter the majority of metrics (SI Appendix, Figs. S11–S13 and Table S5). Functional richness was similarly suppressed overall, but higher within habitats. However, evenness in the DA was notably lower both overall (0.27) and within habitats relative to the LA (0.45 overall) with abundance weighting (SI Appendix, Table S5), while originality was consistently higher in the DA within habitats when calculated using abundance weighting. Nonetheless, with abundance weighting, the discordance between the LA and DA still remained relatively small among the other metrics overall ($<$0.08) and within habitats ($<$0.32).

Table 2.

For each assemblage (Asmb.) and habitat, the values of Functional Richness (FRich), Functional Divergence (FDiv), Functional Evenness (FEve), Functional Dispersion (FDis), Functional Specialization (FSpe), and Functional Originality (FOri) are shown with equal weighting (presence–absence)

Equal weighting
Habitat	Asmb.	FRic	FDiv	FEve	FDis	FSpe	FOri
Entire	LA	0.78	0.88	0.70	0.85	0.66	0.19
Area	DA	0.70	0.87	0.67	0.84	0.65	0.18
	FA	0.57	0.86	0.70	0.84	0.66	0.19
	DA P	$\mathbf{0.01}$	$\mathbf{0.001}$	$\mathbf{0.01}$	$\mathbf{0.002}$	0.25	0.09
	FA P	$\mathbf{0.02}$	$\mathbf{0.001}$	$\mathbf{0.12}$	$\mathbf{0.01}$	0.80	0.40
Offshore	LA	0.78	0.88	0.70	0.85	0.66	0.19
	DA	0.70	0.87	0.67	0.84	0.65	0.19
	FA	0.67	0.87	0.68	0.84	0.65	0.18
	DA P	$\mathbf{0.002}$	$\mathbf{0.001}$	0.41	$\mathbf{0.001}$	$\mathbf{0.003}$	0.51
	FA P	$\mathbf{0.01}$	$\mathbf{0.001}$	0.28	$\mathbf{0.002}$	0.09	0.20
Nearshore	LA	0.63	0.89	0.73	0.85	0.65	0.29
	DA	0.74	0.83	0.69	0.82	0.63	0.26
	FA	0.75	0.84	0.69	0.83	0.64	0.27
	DA P	$\mathbf{0.01}$	$\mathbf{0.001}$	$\mathbf{0.04}$	$\mathbf{0.001}$	$\mathbf{0.007}$	0.10
	FA P	$\mathbf{0.04}$	$\mathbf{0.001}$	0.09	$\mathbf{0.001}$	0.06	0.28
Backbarrier	LA	0.55	0.87	0.68	0.89	0.67	0.28
	DA	0.84	0.86	0.68	0.90	0.69	0.29
	FA	0.79	0.86	0.68	0.90	0.69	0.28
	DA P	0.14	$\mathbf{0.001}$	0.25	$\mathbf{0.001}$	0.14	0.81
	FA P	0.17	$\mathbf{0.003}$	0.20	$\mathbf{0.003}$	0.30	0.48

Significance (P) was determined using bootstrapping under the null hypothesis that the probability of the observed DA and FA values were random artifacts of their species richness ($\alpha =0.95$, significant values shown in bold). Note that for nonsignificant values, we cannot refute the null hypothesis that those values are a function of random sorting of species given the number of species in that assemblage.

2. Discussion

Taphonomic biases and time averaging did not appear to alter the representation of functional diversity that could potentially be preserved in the fossil record. Differences in the abundance and richness between assemblages did not lead to differences in the distribution of species within functional entities or functional space in the DA and FA. The transition from live to dead did not lead to a loss of functional groups or parts of the functional space, and indices quantifying the distribution of species within functional entities indicated that fidelity was high, despite the preferential preservation of more skeletally robust suspension feeders. The distribution of species among functional entities in the DA and FA was unlikely to be an artifact of the number of species, as measures of vulnerability and overredundancy of functional entities (which consider species richness) varied substantially from simulated values expected based on the number of species. Furthermore, the overall consistency of multivariate metrics using abundance weighting is remarkable, given that the DA contains $\sim$40,000 more individuals than the LA, which likely contributed to the suppression of evenness in the DA when calculated using abundance.

Fidelity was generally maintained, although functional diversity was somewhat lower at multiple spatial scales (Table 2), and elevated in the DA and FA within habitats. This is likely due to a greater difference between the number of species in the LA relative to the DA and FA within habitats. While in the entire study region, the DA only had 16 more species than the LA, that disparity increased to as much as 46 in backbarrier, 30 in offshore, and 29 nearshore settings. Abundance weighting had a greater effect on fidelity within habitats. Fidelity was similarly lower within habitats for functional richness, which was again elevated, as well as evenness, which was suppressed. In addition, originality was notably elevated, while dispersion was more suppressed in the DA in offshore habitats, which is driven by a disproportionately larger abundance of a single taxon (SI Appendix, Fig. S11). We therefore recommend caution when employing abundance weighting in paleontological assessments of functional diversity within habitats, which can be more sensitive to differences in abundance between species. Functional diversity of nearshore and offshore datasets was very similar, suggesting that compared to the faunal composition and taxonomic diversity of assemblages, functional diversity is less sensitive to water depth and distance from the shoreline.

Across the entire study area, the number of functional entities was positively correlated with the number of species, and was highest in the DA, which also had more species. In contrast, FRic, a multivariate measure of functional diversity, was highest in the LA. Thus, while functional richness is typically positively correlated with the number of species (5, 68), this relationship was not observed here. Measures sensitive to the number of species were unlikely to have been biased by the number of species present in the DA and FA, as observed values of vulnerability in the DA and FA were also significantly higher than expected if species were randomly assigned to functional entities. Similarly, observed values of overredundancy varied significantly from simulated values, indicating that in the DA and FA species were more densely packed into a few functional entities than expected if species were randomly assigned. This suggests that the functional distribution of species in the DA and FA was not merely a product of the number of species, which tend to differ between assemblages (28). Although vulnerability was much lower in the LA relative to the DA and FA, when vulnerable functional entities are defined as having $\le$2 species, as opposed to singletons, the DA and FA provided reasonable approximations of the percentage of vulnerable functional entities in the LA. Thus, studies investigating functional vulnerability in DAs or FAs should instead consider calculating the proportion of functional entities with two or less species. Although a large proportion of functional entities in the LA were represented by only a few species, these functional entities were still observed in the DA and FA. Therefore, taphonomic biases did not appear to have altered the representation of functional diversity, and investigations of functional entities in DAs and FAs can be reliable, particularly in temperate shallow benthic invertebrate communities such as those studied here.

The high functional congruence between LA, DA, and FA reported here is somewhat intuitive, because dead specimens are derived from the live community, representing a preservable subset. In other words, the fossil record benefits from spatial autocorrelation that is inherent to fossilization processes, which typically sequester locally derived skeletal remains. Only if most of the organisms in the live community are nonpreservable (e.g., dominated by soft-bodied organisms), or functional entities are drastically nonrandom in terms of being represented by preservable versus nonpreservable taxa (e.g., numerous functional entities only represented by soft-bodied organisms), would we expect low fidelity. This is not the case here, and the distribution of taxa among traits is not substantially different (Fig. 2, and SI Appendix, Fig. S4). The only functional entity present in the LA but absent in the DA (pelagic, fully motile slow carnivore) was indeed a soft bodied jellyfish. That the FA was not as good an approximation of functional diversity as the DA with respect to some metrics is also not surprising, given that some of the species in the DA (which are, therefore, potentially preservable) are not in the Paleobiology Database (PBDB). This is likely because the PBDB is less complete at the species level and is also missing many genera that are actually found in the fossil record. Comparing the FA based on the presence of LA and DA species in the PBDB is a highly conservative approach, and overestimates the potential effects of taphonomic filters on functional diversity by excluding taxa that are still likely to be preserved as fossils (and were indeed present in the DA). Although only two species of macroinvertebrate parasites were present in the LA, none were preserved in the FA, which could be the result of either taphonomic biases (e.g., ref. 75) or understudy (e.g., ref. 76). Parasites are also typically very small and would largely have been excluded by our sampling methodology which was designed to survey macrobenthic organisms. Only one brachiopod species was found in the LA, the infaunal lingulide Glottidia pyramidata. This brachiopod was very rare in live samples (only seven individuals were found across all sites) and its organo-phosphatic shell does not preserve readily (e.g., ref. 77). This brachiopod was not present in the DA but was present in the FA.

Although the generalizability of a single case study should be viewed cautiously, a literature review revealed that the faunal composition of the assemblages sampled in this study is comparable to those documented in other surveys of various regions and habitat types, which similarly found that mollusks, arthropods, and annelids were the most diverse and abundant macrobenthic groups (SI Appendix, Table S6). Thus, the results reported here are likely applicable to many other types of benthic assemblages. However, communities overwhelmingly dominated by soft-tissue organisms, while less frequent (five out of 27 surveyed studies), have also been documented, and functional fidelity may be lower in these types of assemblages.

High functional fidelity of the fossil and death assemblages that include multiple higher taxa may seem surprising. However, it may simply reflect the robustness of the methods for measuring functional diversity, particularly functional entity and multivariate methods that employ presence–absence data. Functional diversity should be resilient to taphonomic biases because a loss of specific species does not necessarily lead to a loss of functional space occupied by those species. Similarly, taphonomic distortions of abundance largely do not affect key measures of functional diversity. This is contrary to expectations described in previous studies which anticipated that preservation biases, taphonomic filters, and time averaging are likely to alter the preservation of functional groups. For example, previous studies of marine benthic invertebrates highlighted an overrepresentation of suspension feeders and herbivores—particularly mollusks, which are commonly sessile, heavily biomineralized, and have shorter lifespans—relative to detritivores and carnivores, which are predominantly soft-bodied and thus less likely to be preserved (32, 40, 41, 78, 79). Differences in abundance and life-span across species and higher taxa were also thought to lead to differences in functional fidelity.

The observed high functional fidelity supports the use of functional or trait-based approaches in the fossil record to study ecosystem structure and the evolutionary history of functional diversity. While previous studies of changes in Phanerozoic ecospace utilization have largely relied on identifying the equivalent of unique functional entities (e.g., refs. 65, 80, and 81), our findings suggest that more complex analytical approaches, such as BTA, are not only possible with fossil data, but can provide robust assessments. Similarly, functional diversity may prove crucial in conservation paleobiology particularly for assessing anthropogenic impacts, establishing reference conditions, and identifying changes in trophic structure. These results also reinforce the emerging paradigm of conservation paleobiology, a growing consensus that the youngest fossil provides reliable archives of a high informative value to ecosystem management and conservation (e.g., refs. 42, 46, and 82–91).

3. Materials and Methods

The SI Appendix contains additional details of the sample collection and data analysis. To evaluate functional fidelity across multiple higher taxa simultaneously, concurrent live and dead samples were collected in a series of dredges resulting in a total of 220 samples collected from a variety of habitats, depths, and distances from shore (see refs. 13 and 45 for additional sampling details, including locality coordinates). All live macroinvertebrates were counted and identified to the lowest taxonomic level (typically species) capturing live size classes above $\sim$5 mm. Two bulk samples of dead material were collected at each locality (7.5 L) concurrent with live surveys, and later wet-sieved (mesh size 4.76 mm was used to ensure comparability with live materials). Postprocessing reassessment indicated that very few specimens $<$5 mm were included in the dataset and 5 to 6 and 6 to 7 mm size classes were most abundant in the dataset. As this study focused on the macrobenthos, which by definition excludes microscopic organisms, encrusting species such as bryozoans, barnacles, and sponges which were excluded, and were thus unaffected by sieve size. As many organisms have multiple skeletal components likely to disarticulate after death, each component was multiplied by the proportional abundance estimated for an adult live individual of that species (44–46), for example, a crab chelae would be multiplied by 0.5. Skeletal fragments were not counted to prevent inflating the abundance of species easily identifiable from small fragments (92). The fossil assemblage was constructed using the combined live-dead species pool by identifying all species with fossil occurrences documented in the Paleobiology Database (www.paleobiodb.org). While this approach constrains estimates to species known in the study system, and potentially elevates fidelity estimates, it is also conservative because the PBDB is unlikely to have a record for all local species present in the fossil record. More importantly, the fossil assemblage is used here secondarily to evaluate whether the observed patterns persist when limiting data to species that could be preserved (40). To assess the variability of fidelity at multiple spatial scales, analyses were performed using the entire dataset, and repeated using samples constrained to three habitats: offshore (13 open ocean sites, $>$17.5 m water depth), nearshore (28 open ocean sites, $<$17.5 m water depth), and the backbarrier (5 sites in the area behind the barrier islands). This habitat subdivision accommodates a combination of key major drivers of ecosystem structuring along onshore-offshore gradients previously identified in the study system (44, 45), which include water depth and coastal contexts (i.e., while water depth was used to differentiate between habitats, the offshore sites are also open marine systems $\sim$$>$ 3 miles from shore, nearshore sires are also open marine systems $\sim$0 to 3 miles from shore, and backbarrier sites represent sheltered sites between barrier islands and the mainland with quieter waters and fluctuating salinity). Data associated with this manuscript is available at Zenodo (https://doi.org/10.5281/zenodo.15353456) and the previously published (46) abundance data are available at: https://github.com/tylercl/Multi-Taxic-Fidelity (110) (DOI: 10.5281/zenodo.7871639) (111).

3.1. Functional Entities.

Biological traits are measurable properties of organisms describing their morphology, physiology, and behavior (63, 93–95), and community trait composition is an effective measure of various aspects of benthic ecosystem functioning, such as productivity (96). For each species feeding type, motility, and tiering were classified to create unique functional entities (SI Appendix, Table S1). These traits capture important aspects of ecosystem functioning and can commonly be determined for fossil and modern species (65, 80, 81, 97). This approach has been widely employed in paleoecology, making this study readily comparable with other Phanerozoic assessments of the functional diversity of marine macroinvertebrates. In each assemblage, we examined three commonly used metrics to quantify the abundance and distribution of traits (see SI Appendix for details regarding trait assignments and the calculation of metrics): Functional redundancy (the number of species divided by the number of functional entities), functional vulnerability (the percentage of functional entities with only one species), and functional overredundancy (the percentage of species that fill functional entities above the mean functional redundancy) (69, 98). All of these metrics are based on the presence of a species regardless of its abundance.

3.2. BTA.

BTA is a multivariate approach increasingly used in conservation, which translates information on species distributions into units described by multiple biological traits (7). Functional attributes of the macrobenthic community were described using 8 biological traits, subdivided into 38 modalities (SI Appendix, Table S2), known to influence ecosystem processes (e.g., nutrient cycling, sediment transport, productivity, and trophic support) (99), and describing an organism’s morphology (size, shape, protection) and behavior (life habit, motility, degree of attachment, feeding habit, tiering). As a given trait may vary within a single species, for example due to ontogenetic variation or plasticity, a single value cannot always be assigned (100). In such cases, it is necessary to discretize the trait into multiple categories; This is referred to as “fuzzy coding” and allows for nonlinear relationships between variables (101). Fuzzy coding was employed for size, life habit, motility, feeding habit, and tiering. Traits were determined from various online databases (Polytraits Team 2018, Encyclopedia of Life, WoRMS, SealifeBase, Biotic, Paleobiology Database) and the literature.

Standardized trait scores were used to calculate pairwise functional distances between species traits computed using the Gower’s distance (102), which allows mixing of different types of variables while giving them equal weight (103). A Principal Coordinates Analysis was then performed using the functional distance matrix (104, 105). Using the multidimensional functional space, the following functional diversity indices were calculated for each assemblage using six dimensions: Functional Richness (FRich), Functional Divergence (FDiv), Functional Evenness (FEve), Functional Dispersion (FDis), Functional Specialization (FSpe), and Functional Originality (FOri). FRich estimates the proportion of functional space used, and corresponds to the volume of the convex hull, with high values indicating greater functional heterogeneity (104, 106). FDiv quantifies how species are distributed in trait space, or the distribution of extreme trait values, based on species deviation from the centroid (104). FEve measures the regularity of the distribution of species in trait space using the sum of branch lengths of the minimum spanning tree required to connect all species in an assemblage (104). FDis is the mean distance to the average species position in a given assemblage divided by half the maximum distance among all species in all of the assemblages (104). FSpe is the average position of all species among assemblages divided by the maximum distance to the centroid (8). FOri is the weighted mean distance to the nearest species from the species pool, divided by the maximum distance to the nearest neighbor (8). In all three assemblages, species were equally weighted using presence/absence, and an additional analysis was performed for the LA and DA with species weighted based on their abundance to establish the reliability of the results based on presence absence data alone, which is an approach commonly employed in paleoecological analyses. High agreement between functional diversity indices across the three assemblages would suggest high functional fidelity. As the above metrics can vary depending on the number of species, randomly permuted values were calculated for the death and fossil assemblages by randomly selecting n species from the total species pool without replacement, where n represents the number of species in a given assemblage, and calculating each metric for 1,000 permutations (14, 107, 108). Significance was determined by comparing the observed value with the distribution of simulated values (additional methodological details regarding BTA can be found in SI Appendix; 109).

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Datasets have been deposited in Zenodo (DOI: 10.5281/zenodo.15353456) (111).