Introduction
Studies devoted to developing richness estimates of marine and continental organisms are crucial for the identification of different macroevolutionary and macroecological patterns and processes operating throughout the Phanerozoic (Sepkoski Reference Sepkoski1984, Reference Sepkoski1993; Benton Reference Benton2001; Bambach et al. Reference Bambach, Bush and Erwin2007; Alroy et al. Reference Alroy, Marshall and Bambach2001, Reference Alroy, Aberhan and Bottjer2008; Bush and Bambach Reference Bush and Bambach2011). The most common approach to examining these patterns and processes is to plot taxonomic richness curves based on the number of taxa present within various stratigraphic intervals (bins). However, methodologically, there are two distinct ways of calculating these numbers: (1) the traditional approach based on counts of the number of taxa present in each bin by compiling the timing of their first and last appearances as published in the literature (Sepkoski Reference Sepkoski1984, Reference Sepkoski1993; Benton Reference Benton1993), and (2) the newly developed occurrence-based sampling which standardizes the count of the number of taxa present in each bin with the aim of reducing the potential spatio-temporal biases that develop due to the nature of sampling (Alroy et al. Reference Alroy, Marshall and Bambach2001, Reference Alroy, Aberhan and Bottjer2008). A number of studies have shown that the large-scale diversity trends derived using the more traditional approach are reliable (e.g., Sepkoski Reference Sepkoski1993, Reference Sepkoski1997), largely reflecting the relatively effective sampling efforts that paleontologists have employed to collect a large amount of adequate data (Foote and Sepkoski Reference Foote and Sepkoski1999; Crampton et al. Reference Crampton, Foote and Beau2006; Benton Reference Benton1998; Paul Reference Paul1998; Benton et al. Reference Benton, Wills and Hitchin2000). However, others have argued that this traditional approach can be significantly biased largely due to variation in the quality, quantity, and taxonomy of the fossils that comprise the various time bins (Alroy et al. Reference Alroy, Aberhan and Bottjer2008) and that these biases can only be compensated for by employing a range of resampling techniques (Alroy et al. Reference Alroy, Marshall and Bambach2001, Reference Alroy, Aberhan and Bottjer2008; Bush et al. Reference Bush, Markey and Marshall2004). Ultimately, both of these approaches rely on the published literature as their potential data sources; sources that are potentially fraught with a spectrum of taxonomic and stratigraphic errors.
Empirical studies (Smith and Patterson Reference Smith and Patterson1988; Sepkoski Reference Sepkoski1993; Wagner Reference Wagner1995; Adrain and Westrop Reference Adrain and Westrop2000; Ausich and Peters Reference Ausich and Peters2005; Wagner et al. Reference Wagner, Aberhan and Hendy2007) as well as simulations (Sepkoski and Kendrick Reference Sepkoski and Kendrick1993) have shown how taxonomic errors can significantly bias the overall diversity estimate by inflating or deflating taxonomic counts (Wagner et al. Reference Wagner, Aberhan and Hendy2007). For example, a published list of taxa may be outdated and potentially invalid because of taxonomic revisions and reassignments of species to different or new genera (Alroy Reference Alroy2002). Additionally, various workers from different parts of the world have likely used various taxonomic approaches and been influenced by different experiences and expertise (Sepkoski Reference Sepkoski1998). Therefore, it is important that when initiating such a study that taxonomic standardization of the raw dataset (quality correction) be a component of the compilation. Given that these data will be subjected to different types of sampling standardization processes, these taxonomic corrections should be performed to minimize the impact of uneven sampling from different time intervals (quantity correction).
This study is primarily focused on investigating how quality corrections can influence reconstructed richness patterns, and it will document how taxonomic richness can vary based on the quality of data as elucidated by performing different types of taxonomy-related corrections on the genera from the molluscan Class Bivalvia. Among the various marine invertebrate groups, the Class Bivalvia is one of the most diverse and abundant classes with a rich fossil record that spans the entire Phanerozoic (Hallam and Miller Reference Hallam and Miller1988; Miller Reference Miller1990a,Reference Millerb). Furthermore, bivalve specimens are found in almost all stratigraphic intervals indicating comparatively high completeness in temporal sampling (Harper Reference Harper1998; Jablonski et al. Reference Jablonski, Roy, Valentine and Price2003; Kidwell Reference Kidwell2005) and have been collected over a broad biogeographic range (Paleobiology Database). Because of this rich record, bivalves have been used for the investigation of macroevolutionary and macroecological patterns and processes in numerous studies (e.g., Stanley Reference Stanley1968; Hallam and Miller Reference Hallam and Miller1988; Miller Reference Miller1990a,Reference Millerb; Fraiser and Bottjer Reference Fraiser and Bottjer2007; Roy et al. Reference Roy, Hunt and Jablonski2009; Harnik and Lockwood Reference Harnik and Lockwood2011), and have been chosen for the present study to document how genus-level taxonomic corrections can alter the overall diversity dynamics of a clade.
Need for an Updated Global Bivalve Dataset
The two largest and most comprehensive datasets available with which to reconstruct the rich fossil record of bivalves are the Sepkoski Compendium (SC) (Sepkoski Reference Sepkoski2002) and the Paleobiology Database (PBDB). Virtually all previous studies on long-term trends and patterns in taxonomic diversity of bivalves have been based on one of these two sources (e.g., Hallam and Miller Reference Hallam and Miller1988; Jablonski et al. Reference Jablonski, Roy, Valentine and Price2003; Harnik and Lockwood Reference Harnik and Lockwood2011; versus Alroy et al. Reference Alroy, Marshall and Bambach2001, Reference Alroy, Aberhan and Bottjer2008; Wagner et al. Reference Wagner, Aberhan and Hendy2007). However, qualitatively, these two datasets differ due to the lack of taxonomic uniformity between them. For example, in many cases, what appear to be identical genera have been placed into different families in these two different datasets; therefore, reconstructions of family-level richness based on these two data sources can vary. More problematically, the genus-level stratigraphic first and last occurrences and, therefore, the summation of their ranges are statistically different between these two sources (t-test; p<<0.01; Fig. 1). The PBDB has more short-ranged (0–10 Myr) taxa than are contained within the SC (Fig. 1). In addition, the PBDB has a prolonged tail of a few, very long-ranging taxa (>200 Myr), which are rare in the SC (Fig. 1). Moreover, each dataset has a relatively high percentage of genera that are absent in the other (see Results). Therefore, reconstructions using either of these two datasets in isolation employ only a subset of the entire data that are available. To develop a comprehensive database from which to examine bivalve richness throughout the Phanerozoic, input from both of these datasets was used to develop a more comprehensive compilation.
Compilation of age ranges for all taxa within the three taxonomic datasets. The distribution in SC differs significantly from the new database and the PBDB (in both cases p<<0.01). In addition, the new database and the PBDB are also statistically significantly different, but the p value is comparably larger (p=0.025).
After combining these two datasets, a thorough check of the data quality is required. It has been shown by many studies that the SC has variable types of taxonomic errors which can be as high as ~75% for some taxonomic groups (Patterson and Smith Reference Patterson and Smith1987; Sepkoski Reference Sepkoski1993; Adrain and Westrop Reference Adrain and Westrop2000; Jablonski et al. Reference Jablonski, Roy, Valentine and Price2003; Wagner et al. Reference Wagner, Aberhan and Hendy2007). The main sources of these errors are that many of the taxonomic names included have been revised and are now considered to be invalid (Sepkoski Reference Sepkoski1993; Alroy Reference Alroy2002). In addition, some of the included taxa are non-marine and/or not a member of the Class Bivalvia (Sepkoski Reference Sepkoski1993). Similarly, the PBDB also has taxonomic errors related to taxonomic inconsistency and synonyms (Wagner et al. Reference Wagner, Aberhan and Hendy2007) as well as a range of additional errors (see below). Because of this, taxonomic re-evaluation is a necessary component of compiling the taxa to be included in the new dataset, and that process has been undertaken here.
Taxonomic Correction and Reassignments
The following protocol was followed when taxonomic corrections were performed on all the genera and subgenera listed in the SC and PBDB as they were combined together to construct an initial raw list. First, all of the subgenera were raised to genus status to reduce the taxonomic bias related to faulty rank assignments as well as to ensure that the results are comparable with previous studies on bivalves (Hallam and Miller Reference Hallam and Miller1988; Harper Reference Harper1998; Jablonski et al. Reference Jablonski, Roy, Valentine and Price2003; Harnik and Lockwood Reference Harnik and Lockwood2011). Second, the taxonomic validity of this initial taxa list was determined applying the following methodology: if a genus was common to both the PBDB and the SC, it was included after correcting for synonymy as well as for alternative spellings and misspellings. Based on this assessment, a number of taxa were excluded related to other issues, such as removing non-marine as well as non-bivalve taxa. These corrections were performed by using a variety of sources, including the bivalve volumes of the Treatise of Invertebrate Paleontology (Cox et al. Reference Cox, Newell and Boyd1969; Carter et al. Reference Carter, Cristian and Anderson2012), Vokes (Reference Vokes1980), the taxonomic information and references included in the PBDB, as well as Google Scholar (a comprehensive and reliable database used to find relevant literature; DeGraff et al. Reference DeGraff, Nicholas and Romesburg2013). Where corrections were made, the sources were noted. For the genera unique to either of the two datasets, a two-step screening process was performed. The first step entailed verification related to a given genus’ taxonomic validity. If the original references of these unique genera were cited in the relevant database, an element not always true for the SC, the taxa were included in this new compilation provided that the reference contained the denoted genus/genera. If no reference was provided, a Google Scholar search was performed to determine whether or not a given genus is commonly reported (i.e., cited in at least two articles). If it was only found in a single article or did not appear in a Google Scholar search, the genus was excluded. In the second step, the taxonomic affinities were scrutinized to correct for synonymy, alternative spellings, and incorrect spellings by using various sources as outlined above.
After the comprehensive generic dataset was updated, the higher-level taxonomic affinities of the various genera (e.g., their relevant families and orders) were determined based on the bivalve treatise (Cox et al. Reference Cox, Newell and Boyd1969; Carter et al. Reference Carter, Cristian and Anderson2012), Vokes (Reference Vokes1980), and the PBDB. If different taxonomic opinions for a single genus were found, the original publications on the taxon were researched to determine its taxonomic history prior to reaching a decision as to where the given genus should be positioned.
Calculation of Taxonomic Diversity
The present study uses genera and subgenera rather than species as there are several problems associated with species-level data (Sepkoski Reference Sepkoski1998; Sepkoski and Kendrick Reference Sepkoski and Kendrick1993). For example, the criteria for species identification vary among generations of workers, cultural/historical vagaries, as well as the group(s) studied (Sepkoski Reference Sepkoski1998). Moreover, sampling of species is uneven in geologic time and space as well as between taxonomic groups (Sepkoski Reference Sepkoski1998). In contrast, using genera as the unit of study has a number of important advantages. Genera can be used as a surrogate for the constituent species as genus-level alpha and beta diversity measures are highly correlated with that of species-level data (Sepkoski et al. Reference Sepkoski, Bambach, Raup and Valentine1981; Gaston and Williams Reference Gaston and Williams1993; Flessa and Jablonski Reference Flessa and Jablonski1995). Furthermore, it has been suggested that in cases where sampling is poor, genera can effectively reflect more detailed overall diversity patterns and lineage histories than their constituent species by better representing the phylogenetic history of the clade (Sepkoski and Kendrick Reference Sepkoski and Kendrick1993). Use of genera and subgenera as the unit of study further enables a comparison of the results obtained here with numerous previous studies, which have been focused on that taxonomic level (e.g., Hallam and Miller Reference Hallam and Miller1988; Harper Reference Harper1998; Jablonski et al. Reference Jablonski, Roy, Valentine and Price2003; Harnik and Lockwood Reference Harnik and Lockwood2011).
Another important aspect of diversity studies is whether sample-standardized techniques should be used instead of relying on published taxonomic ranges (Sepkoski Reference Sepkoski1993, Reference Sepkoski1997; Alroy et al. Reference Alroy, Aberhan and Bottjer2008). The present study will use the conventional approach of using taxa ranges in the analyses (Sepkoski Reference Sepkoski1984) reflecting the duality of data compiled from the two sources. Whereas the PBDB provides the ranges of the taxa in combination with data delineating the stratigraphic bins from which the taxa have actually been sampled as well as their abundances for a subset of the data entries, the SC provides only the first and the last appearance ages of taxa (i.e., the ranges). Given that a portion of taxa only has range data (i.e., for those taxa unique to the SC), by necessity this study will use that approach and assume that all the bins between the first and last appearance datums (FAD and LAD, respectively) of a given taxon contained a member of that specific taxon. Furthermore, given the limitations on range data imposed by the SC, it is impossible to use sampling standardization for this compilation as this relies on abundance data (Sepkoski Reference Sepkoski1996).
Because the study also contains data on the stratigraphic ranges of taxa, a revision of these ranges is required before the new database can be used. The determination of ranges involves the removal of errors associated with different age spans published for the SC and the PBDB (Fig. 1). In cases where the ranges of a given genus deviated between the two sources, the oldest age available for the FAD and the youngest age available for the LAD were considered as the updated ages. After the ranges of genera within the newly compiled dataset were corrected, the taxonomic richness (genera, families, and orders) was counted for each geological stage. This was done following the ages and names of the geological stages as presented in the Geological Society of America Geological Time scale (v. 4.0) as compiled from Gradstein et al. (Reference Gradstein, Ogg and Schmitz2012) and Cohen et al. (Reference Cohen, Finney and Gibbard2012).
Finally, to produce the taxonomic richness curves, Foote’s (Reference Foote2000) stage-level instantaneous standing diversity metric was used to count the number of taxa (genera, families, and orders) present for each geological stage. Two additional metrics—estimated mean standing diversity (Foote Reference Foote2000) and mean standing diversity (Van Valen Reference Van Valen1984)—give almost identical diversity trajectories when constructed using this new dataset. Given this similarity, we have only used the instantaneous standing diversity metric here to test whether there are any differences in the pattern of the bivalve-genera-richness curves during the Phanerozoic when this new compilation is compared to the two datasets. In addition, this new compilation has been used to depict the proportional richness curve (Bambach et al. Reference Bambach, Knoll and Wang2004) to identify potential episodes of bivalve extinctions and originations. Here, proportional richness has been calculated as the ratio between the difference in the standing richness between two consecutive time bins and the richness in the older bin, therefore indicating proportional changes in richness across each time boundaries with negative changes indicating extinctions and positive changes indicating originations (Bambach et al. Reference Bambach, Knoll and Wang2004).
The presence of singletons (i.e., taxa restricted to only a single time bin) in a dataset can have significant effects on richness and diversity estimates (Foote Reference Foote2000; Fitzgerald and Carlson Reference Fitzgerald and Carlson2006). However, it should be noted that at least some, if not most, of these singleton taxa might actually be taxonomically valid with accurate stratigraphic ranges and do not result from incomplete sampling (Raup Reference Raup1972, Reference Raup1977; Sepkoski Reference Sepkoski1996) or other types of biases (for details, see Fitzgerald and Carlson Reference Fitzgerald and Carlson2006). Therefore, to investigate this issue, the Phanerozoic generic diversity curves were generated with and without singletons (Sepkoski Reference Sepkoski1997; Fitzgerald and Carlson Reference Fitzgerald and Carlson2006; Wagner et al. Reference Wagner, Aberhan and Hendy2007).
In addition to the genus diversity curve, higher taxa (orders and families) were used in the various analyses undertaken to delve into the overall history of the clade (see below). For the order- and family-level curves, only the new dataset has been used. The FADs and LADs of families, orders, and genera were also plotted for each bin to better document the major diversification and extinction episodes of the clade during the Phanerozoic and to compare these results to those obtained from previous studies (Stanley Reference Stanley1968; Bretsky Reference Bretsky1973; Miller and Sepkoski Reference Miller and Sepkoski1988; Miller Reference Miller1990a,Reference Millerb; Harper et al. Reference Harper, Todd and Alphey1997; Nevesskaya Reference Nevesskaya2008, Harnik and Lockwood Reference Harnik and Lockwood2011).
Finally, a number of other corrections were performed to minimize the impact of uneven sampling during different time intervals (Alroy et al. Reference Alroy, Aberhan and Bottjer2008). One of the most significant biases, with the potential to distort the pattern of Cenozoic bivalve diversity, is the so-called ‘Pull of the Recent’ (Cutbill and Funnel Reference Cutbill and Funnell1967; Raup Reference Raup1972, Reference Raup1979; Alroy et al. Reference Alroy, Aberhan and Bottjer2008; but see Jablonski et al. Reference Jablonski, Roy, Valentine and Price2003). Due to better sampling of the Holocene as well as the increased amount of sedimentary exposure of younger units relative to those from earlier in geologic time (Raup Reference Raup1972; Alroy et al. Reference Alroy, Aberhan and Bottjer2008) and increased prevalence of unlithified units as the Recent is approached (Hendy Reference Hendy2011), not only will extant genera have longer ranges than extinct taxa, but there is the likelihood for more complete sampling, which will potentially increase taxonomic richness in the Cenozoic (Raup Reference Raup1972, Reference Raup1979; Alroy et al. Reference Alroy, Aberhan and Bottjer2008; but see Jablonski et al. Reference Jablonski, Roy, Valentine and Price2003). However, the two datasets combined here do not have data from the Holocene. For these reasons and to minimize the effect of this bias in addition to minimizing biases related to temporal differences in the nature of lithification (Alroy et al. Reference Alroy, Aberhan and Bottjer2008; Hendy Reference Hendy2009), the taxonomic richness has been plotted from the Fortunian (oldest age in the Cambrian) through the Piacenzian (youngest age in the Neogene). Furthermore, the inclusion of the Quaternary can have additional problems associated with taxonomy-related issues. For example, extant higher taxa may be subdivided more finely from a taxonomic perspective than their older counterparts because of the ability to identify a potentially broader spectrum of character states. For this reason, it is more likely to find fossil representatives of these extant, finely divided, higher taxa than their extinct, less subdivided counterparts, in the fossil record (see Raup Reference Raup1979). However, the impact of the ‘Pull of the Recent’, at least on bivalves, is debatable (Jablonski et al. Reference Jablonski, Roy, Valentine and Price2003), as the fossil record of bivalves has been documented as showing a high level of completeness (Harper Reference Harper1998; Kidwell Reference Kidwell2005).
Results and Discussion
The Newly Compiled Dataset
Taxonomic corrections, which may increase or decrease the number of genera within a dataset (Wagner et al. Reference Wagner, Aberhan and Hendy2007), were made during the compilation that forms the foundation of the present study. The SC and PBDB represent compilations of 2823 and 2627 bivalve genera and subgenera, respectively. Among this list of 5450 total genera, only 1949 were contained within both two datasets, and the remainder was unique to one of them. Reducing the number of total genera by the amount in common between the two datasets resulted in a count of 3501 genera. Among these, 874 and 678 genera were unique to the SC and PBDB, respectively. This total (i.e., 3501) is higher than the total richness in either the SC or the PBDB separately, and can be considered as a case of “gaining” taxa because of the pooling of data from two major compilations (Fig. 2) (Wagner et al. Reference Wagner, Aberhan and Hendy2007). Therefore, previous studies that have used either of these two datasets for their analyses (e.g., Hallam and Miller Reference Hallam and Miller1988; Jablonski et al. Reference Jablonski, Roy, Valentine and Price2003; Harnik and Lockwood Reference Harnik and Lockwood2011; but see Alroy et al. Reference Alroy, Marshall and Bambach2001, Reference Alroy, Aberhan and Bottjer2008; Wagner et al. Reference Wagner, Aberhan and Hendy2007) have only used a subset of the entire data available for bivalves. More importantly, it is impossible to simply combine the diversity curves derived from these two different groups of studies using two different datasets with only 40% genera in common. Therefore, it should be emphasized that when either the SC or the PBDB are employed, they do not represent a complete compilation of the available data; for implications of this lack of completeness, see Patterson and Smith (Reference Patterson and Smith1987), Sepkoski (Reference Sepkoski1993), Adrain and Westrop (Reference Adrain and Westrop2000), and Ausich and Peters (Reference Ausich and Peters2005).
Genus richness curves based on three datasets are presented. Note the major variations among these datasets are at the Mesozoic and Cenozoic time intervals.
Effect of Taxonomic and Other Corrections
In the next phase of developing this compendium, 205 PBDB-unique genera were excluded from the list, because although these taxa were present in the PBDB, they were not associated with any accompanying stratigraphic information even after a thorough search through various sources. Similarly, 198 genera in the SC-unique list were also excluded. These SC genera failed to qualify based on reference criteria established, as their source was not explicitly stated (in many cases the reference is cited as ‘ZR’ [i.e., the Zoological Record], without referring to a published source).
The final stage involved taxonomic correction for synonymy, removal of alternate spellings, misspellings, taxonomically invalid subtaxa, and the exclusion of non-marine bivalves as well as those that are not actually in the Class Bivalvia from the composite list. These processes excluded an additional 204 genera: 71 invalid subtaxa, 83 synonyms, and 47 typographic errors/alternative spelling, and three non-bivalve taxa. This resulted in a final tally of 2894 genera that were subsequently used in this study’s analyses (Supplementary Table 1). Therefore, the taxonomy-related correction removed ~18% of the genera from the raw combined list. This number is well within the range of percent error reported in previous studies that have done similar taxonomic corrections on the PBDB and the SC for various groups. For example, studies on echinoderm and fish richness covering the interval from the mid-Permian to the Recent, and trilobite richness in the Ordovician and Silurian have revealed that the SC can have a level of taxonomic error as high as ~70% (Patterson and Smith Reference Patterson and Smith1987; Adrain and Westrop Reference Adrain and Westrop2000). Jablonski et al. (Reference Jablonski, Roy, Valentine and Price2003) also found taxonomic errors in the SC dataset when they studied Plio-Pleistocene bivalves, although the error was substantially lower. In comparison, in the case of North American fossil mammal species, the PBDB contained ~25–30% taxonomic error (Alroy Reference Alroy2002). A species-level study on Jurassic and Cenozoic bivalves using the PBDB shows that the error can vary from ~20% to ~40% (Wagner et al. Reference Wagner, Aberhan and Hendy2007). Because taxonomic errors must be expected, especially within an old dataset, due to continuing taxonomic revisions and newly discovered fossil material, it is a good practice to start with a taxonomic revision, unless one can assume that the error is evenly distributed throughout the time bins (Adrain and Westrop Reference Adrain and Westrop2000; but compare with Ausich and Peters Reference Ausich and Peters2005).
Effect of Stratigraphic Range Corrections and Singletons
If the mean stratigraphic ranges of all taxa are compared between the SC and the PBDB, they are statistically different from each other (t-test; p<<0.01). This is also true of the newly compiled dataset relative to the SC (p<<0.01). However, although the differences in the taxa ranges of the PBDB and the new dataset are statistically significant, the p value is comparatively larger (p=0.03). Whereas the mean ranges of all the SC and PBDB genera are ~31.8 Myr and ~36.9 Myr, respectively, in the newly compiled dataset this increases to ~50.3 Myr (Fig. 1) portraying the range extension of genera (Wagner et al. Reference Wagner, Aberhan and Hendy2007) that results from the approach used in taking the oldest FADs and youngest LADs (see above). This pattern of range extension after taxonomic corrections associated with updating both an old dataset with additional data and taxonomic opinions has been documented in general (Sepkoski Reference Sepkoski1993; Adrain and Westrop Reference Adrain and Westrop2000) as well as specifically for bivalves (Wagner et al. Reference Wagner, Aberhan and Hendy2007). It should be noted that the ‘tail’ present in Figure 1, due to genera with >200 Myr temporal ranges, represents, to some extent, overestimates of the true ranges likely the result of misidentification or errors in data inputs. However, such taxa represent <5% of the total number of genera in the newly compiled database. Moreover, because these ‘tail’ genera are long-ranged, they should not have a selective bias towards any particular time interval on the richness curve. For the present study, we have thus decided to include them in all the analyses.
The presence of singletons can significantly impact diversity estimates (Foote Reference Foote2000; Fitzgerald and Carlson Reference Fitzgerald and Carlson2006). However, for the present study, only ~3% are singleton genera present in the updated dataset. This low percentage of singletons is comparable to values previously documented (~5%) (Harper Reference Harper1998). This percentage of singletons is quite low for a sampling interval of approximately 540 Ma with 95 geological subdivisions resulting in an average of only ~0.9 singleton genera per temporal bin and is also clearly reflected when richness with and without singletons is plotted (Fig. 3A). However, although in general the differences between the two curves are very slight, they show slightly more pronounced differences during three intervals: the Middle Triassic, the mid-Cretaceous, and just prior to the K/Pg extinction (Fig. 3A). At this point, there is no specific explanation for why these three intervals are relatively rich in singletons.
Phanerozoic richness curves for bivalves. A, Compares the curves with and without singletons demonstrating that the two are virtually identical with the most pronounced, although only minimal deviations denoted by the arrows. B, Depiction of the sample-standardized richness curves of bivalves and gastropods, based on the PBDB dataset, also displays a steep Cenozoic rise in richness (modified from Alroy [Reference Alroy2010]) with the Miocene decline likely an artifact of insufficient sampling for that interval.
The Phanerozoic History of the Class Bivalvia
Given that this new compilation of bivalve data has been subjected to a range of quality corrections, a critical question that must be addressed first is the validity of the reconstructed richness pattern and trends. There are several types of inherent geologic and collection-based biases, which can potentially impact the reconstructed Phanerozoic richness of a clade (Smith et al. Reference Smith, Gale and Monks2001; Smith Reference Smith2001; Alroy et al. Reference Alroy, Marshall and Bambach2001, Reference Alroy, Aberhan and Bottjer2008; Dunhill et al. Reference Dunhill, Benton and Twitchett2012). One correction that has been promoted is the application of sample-standardized richness instead of more direct raw richness counts. In many cases, it has been argued that paleontologists have done an excellent job in collecting a large amount of adequate data to effectively document Phanerozoic changes in biodiversity (Foote and Sepkoski Reference Foote and Sepkoski1999; Crampton et al. Reference Crampton, Foote and Beau2006; Benton Reference Benton1998; Paul Reference Paul1998; Benton et al. Reference Benton, Wills and Hitchin2000), whereas others have argued that biases overprint the data to such an extent that simply considering the raw data is at best problematic (Raup Reference Raup1972, Reference Raup1979; Alroy et al. Reference Alroy, Aberhan and Bottjer2008). A recent impetus to more fully address this debate comes from the proposed ‘common-cause hypothesis’ (Peters and Foote Reference Peters and Foote, M.2001, Reference Peters and Foote, M.2002; Peters Reference Peters2005, Reference Peters2006), which argues in favor of common global forcing which strongly controls both the rock record and biodiversity. According to this view, the fossil record is biased because of spatio-temporal variation in various geological processes, such as sea level, tectonics, as well as lithologic differences, and, therefore, should be treated with caution.
However, for marine bivalves, it is thought that the biases that can impact richness estimates of various other groups are minimal. Qualitatively, the fossil record of bivalves includes both calcitic and aragonotic taxa, even under calcite-sea conditions (Ros and de-Renzi Reference Ros and Renzi2005). Furthermore, their record consists not only of actual skeletal material, but it is further augmented by numerous taxonomically identifiable external and internal molds. This suggests that the commonly cited aragonite bias cited for other groups may not strongly overprint the fossil record of this clade (Alroy et al. Reference Alroy, Aberhan and Bottjer2008) resulting in the generally effective preservation of many bivalve genera (Foote and Sepkoski Reference Foote and Sepkoski1999; but see Harnik and Lockwood Reference Harnik and Lockwood2011). A further element of this is that taxa with aragonitic shells and organic microstructure have been documented even from the Paleozoic (Kidwell Reference Kidwell2005). Quantitatively, the fossil record of bivalves is highly complete, as supported by the following lines of evidences: (1) most of the family- and genus-level data are complete with low proportion of gaps, i.e., they have been found in almost all stratigraphic bins in between their FADs and LADs (Harper Reference Harper1998); (2) as suggested by Paul (Reference Paul1982), if the group was characterized by relatively high levels of stratigraphic incompleteness, we should expect more short-lived taxa in our dataset. In this bivalve compilation, singletons are rare (i.e., ~3%; see also Harper Reference Harper1998), further supporting a robust record; (3) if the mineralogical composition of fossil taxa inhibits preservation (i.e., aragonitic versus calcitic), we should also expect more singleton taxa with aragonitic shell material with low-preservation potential. However, for bivalves, singletons are not exclusively aragonitic (Kidwell Reference Kidwell2005); (4) if the fossil record of a group is fragmented, it should result in fossils being so distinct from each other that they will be placed in two separate higher groups, increasing monotypy (Paul Reference Paul1982). For bivalves, less than 15% of families are monogeneric (Harper Reference Harper1998); (5) most of the families and genera are extant (Harper Reference Harper1998; Valentine et al. Reference Valentine, Jablonski, Kidwell and Roy2006). In addition, 95% of the living families and 76% of the living genera are known from the fossil record, again indicating high completeness for bivalves; these values are higher as compared to other marine molluscan groups with ~60% and ~50%, for families and genera, respectively (Foote and Miller Reference Foote and Miller2007). From a quantitative perspective, the bivalve record is conducive to paleobiological analyses for several reasons: overall a relatively complete collection of material sampled over a broad range of latitudes, geographic areas, and time intervals (Alroy et al. Reference Alroy, Aberhan and Bottjer2008; Paleobiology Database); readily available data sources, and very low influence of the ‘Pull of the Recent’ (Jablonski et al. Reference Jablonski, Roy, Valentine and Price2003; Alroy et al. Reference Alroy, Aberhan and Bottjer2008). This overall robustness of the bivalve record is likely the reason that the overall richness patterns documented in previous studies are remarkably similar to each other irrespective of the method and data used (Stanley Reference Stanley1968, Hallam and Miller Reference Hallam and Miller1988; Miller Reference Miller1990a,Reference Millerb; Nevesskaya Reference Nevesskaya2008; Harnik and Lockwood Reference Harnik and Lockwood2011).
Regardless of which dataset is used (i.e, ones derived from the SC, the PBDB, or the new compilation herein), the overall genera-level richness curves for bivalves show a progressive rise throughout the Phanerozoic (Fig. 2), a trend which reflects the fact that the majority of bivalve families ever present are still extant (Stanley Reference Stanley1970; Harper Reference Harper1998) as well as a secular decline in origination and extinction rates throughout the Phanerozoic (Hallam and Miller Reference Hallam and Miller1988; Harnik and Lockwood Reference Harnik and Lockwood2011). Interestingly, the curves derived from the three databases are identical for the Ordovician, after which they diverge displaying three trajectories (Fig. 2). Throughout the Paleozoic and Mesozoic, the curve based on the SC data is always the lowest, largely reflecting the presence of a higher number of short-ranging taxa than the PBDB (Fig. 1). This result differs from what has been documented for all marine invertebrate genera by Alroy et al. (Reference Alroy, Aberhan and Bottjer2008) in that they showed that the curve based on the entire Sepkoski compendium, without any sampling correction, is higher than their raw, unstandardized curve based on their dataset, throughout most of the Paleozoic. The SC-derived bivalve curve, however, rises to higher levels of richness than the sample unstandardized PBDB curve for all marine groups during the Cenozoic, which, according to Alroy et al. (Reference Alroy, Aberhan and Bottjer2008), generally results from Cenozoic oversampling. This Cenozoic variation signifies that other marine fossil groups are more poorly represented in the PBDB than in the SC. However, addition of data into the PBDB after 2008 is still inadequate for the Cenozoic. As to be expected, the curve based on the newly compiled dataset is always higher than the other two curves throughout, although the gap between this compilation and the SC lessens during the Cenozoic. The overall height of the new curve results from range extensions reflecting the revision of stratigraphic ranges of the taxa as well as from combining data from two sources with many unique taxa (see above).
When the trends of the three bivalve curves are compared (Fig. 2), that derived from the SC is mostly flat throughout the Paleozoic, then displays a steep rise in richness in the Mesozoic, followed by an even steeper increase in the Cenozoic. In case of the PBDB curve, it documents a slow, progressive rise in richness throughout the Phanerozoic. The most unique feature of this curve is a lack of a substantial Cenozoic increase, and the Cenozoic richness remains similar to that recorded for the end-Cretaceous. This pattern compares favorably to that reported for all marine invertebrates when a sample-standardized dataset was used (Alroy et al. Reference Alroy, Aberhan and Bottjer2008). However, the sample-standardized curve of only bivalve richness based on the PBDB dataset displays a steep Cenozoic rise in diversity (Fig. 3B) (Alroy Reference Alroy2010). It, therefore, indicates that the Cenozoic radiation absent in the raw PBDB curve appears when various biases are accounted for and strongly suggests that the PBDB dataset may be undersampled, as argued by Jackson and Johnson (Reference Jackson and Johnson2001) and Vermeij and Leighton (Reference Vermeij and Leighton2003).
When the other major details of the curves are compared, all of them depict a drop in richness associated with the end-Ordovician, the P/T, and the K/Pg mass extinctions (Fig. 2). Interestingly, although the new dataset also captures a fall in richness during the Late Devonian coinciding with the Frasnian-Fammenian mass extinction, the other two curves do not record this event. Another mismatch between these curves occurs at the T/J mass extinction—whereas the PBDB curve shows a large reduction in richness, the SC curve shows a small decline and the new curve shows no effect of this event. Moreover, the overall patterns of these extinctions are different among these datasets (Fig. 2; also see below).
Finally, in addition to the genus-richness curve, it is also important to study the diversification history of higher taxa, which here is focused on orders and families along with determining their FADs and LADs to document major diversification and extinction episodes of the clade. The order and family level richness curves based on the new dataset show different trajectories (Fig. 4). The overall trajectory in richness for bivalve orders shows an initial and dramatic diversification of numerous orders (>10) during the Ordovician (Fig. 5A), followed by a gradual rise in richness which was maintained until the Early Jurassic, after which richness remained almost constant with a value of ~20 orders (Fig. 4). In comparison, the number of families shows an initial rise in the Ordovician, although not nearly as pronounced, in a relative sense, as for the orders, followed by a gradual rise throughout the Paleozoic that continued through to the Early Jurassic, after which it remained virtually static (~120 families) until the mid-Cretaceous, and then started to decrease in number (Fig. 4; but see Hallam and Miller Reference Hallam and Miller1988). The span from the Jurassic until the mid-Cretaceous is the interval over which these curves reached a richness plateau, whereas the overall trend of genus richness reflects a progressive, yet mostly gradual increase through this interval (Fig. 2).
Richness of bivalve orders and families through the Phanerozoic based on the new compilation.
Numbers of first (FAD) and last (FAD) appearances of orders and families (A and B), and of genera (C) derived from the new compilation from this study.
Considering these large and small-scale trends of diversification, as identified in the present study as well as in other previous studies, the overall history of the Class Bivalvia can be divided into several major episodes after their first appearance in the Cambrian (see Fang Reference Fang2006 for details). The first of these occurred during the Ordovician when the clade experienced a major diversification (Bretsky Reference Bretsky1973; Stanley Reference Stanley1968; Nevesskaya Reference Nevesskaya2008; Alroy Reference Alroy2010) characterized by the first appearance of many orders and families (Fig. 5A). This diversification was a component of the well-documented Ordovician radiation (Connolly and Miller Reference Connolly and Miller2001; Webby et al. Reference Webby, Paris, Droser and Percival2004; Bambach et al. Reference Bambach, Knoll and Wang2004), which has been tied to multiple biotic and abiotic causes (Webby et al. Reference Webby, Paris, Droser and Percival2004; Servais et al. Reference Servais, Harper and Munnecke2009). A major ecological cause that likely triggered this radiation was the invasion of more offshore settings combined with an overall increase in onshore diversity (Miller Reference Miller1988, Reference Miller1990a,Reference Millerb; see also Crame Reference Crame2000). However, the number of bivalve genera was relatively low during this initial phase of diversification (Fig. 2) (Fang Reference Fang2006; Fang and Sanchez Reference Fang and Sánchez2012). This event is unique in that no other major phase of order-level diversification has been recognized at any point in the subsequent evolutionary history of the class.
At the family level, a second phase of bivalve diversification occurred during the Triassic and Jurassic (Fig. 5A) (Bretsky Reference Bretsky1973; Stanley Reference Stanley1968; Nevesskaya Reference Nevesskaya2008). According to Stanley (Reference Stanley1968), this radiation was initiated by the evolution of siphons as a consequence of mantle fusion that triggered a tiering expansion in bivalves through their invasion of the relatively deeper part within the sediment (i.e., infaunalization). This major bivalve radiation is a component of the well-known Mesozoic Marine Revolution (MMR; Vermeij Reference Vermeij1977, Reference Vermeij1987; Thayer Reference Thayer1979) when several groups of organisms diversified, especially those considered elements of the ‘Modern’ fauna (Walker and Brett Reference Walker and Brett2002). Additionally for bivalves, relatively minor phases of family-level radiation occurred during the Devonian, Carboniferous, and Cenozoic (Fig. 5A).
The genus-level curve shows another important feature common to many other marine groups: a mid-Paleozoic interval of comparatively stable generic richness (Stanley Reference Stanley1968; Hallam and Miller Reference Hallam and Miller1988; Nevesskaya Reference Nevesskaya2008; Alroy Reference Alroy2010). This has been termed the ‘Paleozoic Plateau’ and has been identified in the global genus-level analysis of all marine organisms (Sepkoski Reference Sepkoski1984; Bambach Reference Bambach1999; but see Alroy et al. Reference Alroy, Aberhan and Bottjer2008 for an opposing perspective). Given that a spectrum of groups were diversifying during this interval largely in response to extinction-related ecological reorganization (Bambach et al. Reference Bambach, Knoll and Sepkoski2002) and the invasion of previously unexploited ecological niches (Ausich and Bottjer Reference Ausich and Bottjer1982; Bambach Reference Bambach1983; Bambach et al. Reference Bambach, Bush and Erwin2007) during the so-called ‘mid-Paleozoic Marine Revolution’ (Signor and Brett Reference Signor and Brett1984), it is not clear why global diversity remained almost unchanged instead of displaying an overall increase. However, for bivalves, there are at least three ecological aspects that potentially explain this pattern: (1) this clade suffered minimal richness reductions associated with the major Paleozoic extinctions (Hallam and Miller Reference Hallam and Miller1988; Nevesskaya Reference Nevesskaya2008); (2) a limited number of new ecological niches were exploited by bivalves during this phase of the Paleozoic; and (3) the ecologic dominance and incumbancy of the already abundant and diverse brachiopods (Steele-Petrovic Reference Steele-Petrovic1979) in the marine realm, which was severely disrupted during the P/T crisis (Sepkoski Reference Sepkoski1996; but see Gould and Calloway Reference Gould and Calloway1980; Clapham and Bottjer Reference Clapham and Bottjer2007). After that, bivalves filled much of the ecospace previously dominated by brachiopods and became a major component of marine ecosystems.
During the Cenozoic, a huge rise in the number of genera is seen (Fig. 5B); a rise that appears to be ubiquitous to virtually all approaches to examining the group’s Phanerozoic richness and can be seen in studies by Nevesskaya (Reference Nevesskaya2008) and Alroy (Reference Alroy2010). However, despite the pronounced increase in genera there is no concomitant increase in the FADs of orders (Fig. 4), and, somewhat surprisingly, it is associated with a reduction in family-level richness (Fig. 4). Origination of as many as 150 and 125 new genera occurred during the Paleogene and the Neogene, respectively (Fig. 5C) and reflects this Cenozoic radiation. However, for bivalves, the genus-level taxonomic radiation initiated in the Permian, and the greatest pre-K/Pg increase occurred during mid-Cretaceous time (Figs. 2, 5).
The overall genus-level patterns of Phanerozoic bivalve and gastropod richness based on Alroy’s (Reference Alroy2010) calculations are comparable with the Phanerozoic trajectory of Sepkoski’s (Reference Sepkoski1984) Modern Fauna. This fauna, consisting mostly of bivalves and gastropods along with other groups, such as Osteichthyes, Malacostraca, Echinoidea, Gymnolaemata, Demospongiae, and Chondrichthyes, mirrors the Phanerozoic pattern of bivalve richness in displaying a substantial diversification during the Ordovician, a Paleozoic plateau, a small reduction at the K/Pg extinction, and an ever-increasing diversification during the Mesozoic into the Cenozoic (Sepkoski Reference Sepkoski1984). Sepkoski’s (Reference Sepkoski1984) family-level diversity trajectories are highly comparable with the present study: a gradual increase until the Jurassic, after which it became stable (Fig. 4).
Although sampling biases may be a factor behind this Cenozoic rise (as documented, for example, in Raup Reference Raup1972, Reference Raup1979; Alroy et al. Reference Alroy, Aberhan and Bottjer2008), several other lines of evidence suggest that this increase may be real, at least as displayed by the bivalves. First, because the tropics have the highest diversity (Pianka Reference Pianka1966; Valentine and Ayala Reference Valentine and Ayala1978; Bambach Reference Bambach1990; Valentine and Jablonski Reference Valentine and Jablonski1991) and these regions are generally considered to be under-sampled, most specifically during the Cenozoic (Jackson and Johnson Reference Jackson and Johnson2001; Johnson Reference Johnson2003; Vermeij and Leighton Reference Vermeij and Leighton2003) or underrepresented by Cenozoic deposits (Allison and Briggs Reference Allison and Briggs1993), the expectation is that this would bias richness towards lower levels during the Cenozoic. Second, all stratigraphic bins (i.e., stages) during the Cenozoic are shorter than the remainder of the Phanerozoic (~3.4 as compared to ~6.08 and Ma for the Cenozoic and Paleozoic-Mesozoic, respectively; p<0.05), another feature that would be predicted to dampen Cenozoic diversity (Raup Reference Raup1972). However, there is no correlation between taxonomic richness and age durations during each geologic era, and this correlation for the entire Phanerozoic is weak and negative (Pearson ρ=−0.24, p=0.02), indicating that uneven binning is not a potential explanation for this rise. Third, although the number of genus-level occurrences (derived from data contained in the PBDB) during the Paleozoic is significantly lower than the Mesozoic and Cenozoic (t-test, p<0.01), the correlation between the number of genera and the number of occurrences during the Cenozoic, excluding the Quaternary, is weak (Pearson ρ=0.243, p=0.347), lending further support to the Cenozoic rise being real and minimally impacted by sampling biases. Fourth, habitat heterogeneity and rock volume decreased during the Cenozoic (Peters and Foote Reference Peters and Foote, M.2001, Reference Peters and Foote, M.2002; Peters Reference Peters2005, Reference Peters2006), which should negatively affect richness by decreasing sampling intensity from this interval. Moreover, the sample-standardized curve for bivalves based on the PBDB dataset also displays a steep Cenozoic rise for bivalves (Fig. 3B) (Alroy Reference Alroy2010), which corroborates the reality of the Cenozoic rise.
Another important aspect of the overall diversity dynamics is the effect of mass extinctions on bivalve evolutionary history. It has been argued that the class has been relatively immune to the five major mass extinctions, as reflected in the fact that the majority of bivalve families are extant (Stanley Reference Stanley1970; Harper Reference Harper1998). However, the new dataset suggests that the K/Pg extinction had a relatively significant impact when as many as 250 genera became extinct (Fig. 5C) (Nevesskaya Reference Nevesskaya2008; Harnik and Lockwood Reference Harnik and Lockwood2011); it can also be seen in the proportional diversity curve (Fig. 6) as well as in the significant drop in family-level diversity (Fig. 4) and even in the orders as several went extinct during this mass extinction (Fig. 5B). Among the most prominent of the latter was the extinction of the Hippuritoida (i.e., rudistids), which were an important component of Cretaceous tropical marine communities (e.g., Steuber Reference Steuber2013). In contrast, the P/T extinction (compare Fig. 5B with 6) resulted in the extinction of only a few families without the loss of a single order (Bretsky Reference Bretsky1973; Hallam and Miller Reference Hallam and Miller1988; Nevesskaya Reference Nevesskaya2008; Harnik and Lockwood Reference Harnik and Lockwood2011). In terms of extinction intensity, the next largest extinction occurred during the end-Ordovician, but no significant drop in richness is documented (Figs. 2–3). In case of the Frasnian-Fammenian (Late Devonian) mass extinction, extinction intensity was ~11%, and consisted of multiple extinction phases (Harnik and Lockwood Reference Harnik and Lockwood2011). However, its impact on bivalve diversity was quite limited. This may be explained by a rapid diversification when a few new genera and families appeared (Figs. 5A,C and 6) (Bretsky Reference Bretsky1973; Nevesskaya Reference Nevesskaya2008; Harnik and Lockwood Reference Harnik and Lockwood2011). A few other minor extinction phases can also be identified at the middle Carboniferous, middle Cretaceous, middle Paleogene, and in the Neogene (Fig. 6). However, the effect of these extinction events and the various bioevents documented throughout the Phanerozoic had minor effects on the genus-, family-, and order-level diversity as documented in the richness trends (Figs. 2, 4).
Proportional changes in generic diversity are presented, following the new dataset.
Conclusion
The study clearly indicates the need to carefully construct a dataset before undertaking any studies on global, macroevolutionary patterns, as these trends are likely to be influenced by different types of time-dependent geological, sampling, and taxonomic biases. Because most of the available datasets are outdated and fraught with a relatively higher precentage of taxonomic and stratigraphic errors, it is necessary to revise the raw data so that these errors can be minimized. Instead of accepting the incompleteness of data, major online databases should be assembled to improve taxic-stratigraphic-geographic coverage of the raw data, so that a ‘true’ global database can be compilated.
Based on this new compilation, the Phanerozoic history of bivalves shows a progressive rise in richness without much effect of the major mass extinctions, with the K/Pg as notable outlier. The present study has identified several major macroevolutionary episodes in the history of bivalves: the Cambrian-Ordovician radiation, followed by a Paleozoic plateau, a Mesozoic diversification, and a steep Cenozoic rise. During both of the Paleozoic and Mesozoic phases of diversifications, higher taxa (e.g., new orders and/or families) appeared during the early phase, followed by diversification of lower taxa (e.g., genera).
Acknowledgments
The manuscript was greatly improved by comments and suggestions from F. Zong-jie, D. Geary, M. Olney, and J. Wynn. S. Mondal acknowledges a Geological Society of America Student Research Grant (2013) and the Paleontological Society of America’s Stephen Jay Gould Student Grant (2013) for financial support.
Supplementary Material
Supplemental material deposited at Dryad: doi:10.5061/dryad.2ph0t
