Abstract: Fishes include more than half of all living animals with backbones, but large-scale palaeobiological patterns in this assemblage have not received the same attention as those for terrestrial vertebrates. Previous surveys of the fish record have generally been anecdotal, or limited either in their stratigraphic or in their taxonomic scope. Here, we provide a broad overview of the Phanerozoic history of fish diversity, placing a special emphasis on intervals of turnover, evolutionary radiation, and extinction. In particular, we provide in-depth reviews of changes during, and ecological and evolutionary recovery after, the end-Devonian (Hangenberg) and Cretaceous–Palaeogene (K–Pg) extinctions.
‘Fishes’, the paraphyletic assemblage of primitively aquatic backboned animals, comprise over half of all living vertebrate species (Nelson 2006), with a fossil record extending to the early Cambrian (Shu et al. 1999, 2003). Compared with marine invertebrates (Sepkoski 2002) and many other vertebrate groups (e.g. Alroy 1996; Lloyd et al. 2008), our understanding of large-scale diversity patterns in fossil fishes is limited. This is somewhat surprising, because fishes were the first group for which patterns of palaeobiodiversity were systematically documented, more than a century and a half ago by Louis Agassiz in his monumental Recherches sur les Poissons fossiles (1833–1844). Fishes should represent a obvious study system for exploring macroevolutionary contrasts between freshwater and marine environments within a single group (e.g. McKinney 1998; Vega and Wiens 2012), and the considerable body of work dissecting form–function relationships in living forms generally, and teleosts specifically (e.g. Wainwright and Bellwood 2002), make fishes a particularly attractive system for exploring questions related to ecology over geological timescales (e.g. Bellwood 2003; Friedman 2010; Anderson et al. 2011; Sallan et al. 2011; Sallan and Friedman 2012). Additionally, the impact of fishes as predators, competitors and prey on the long-term macroevolution and macroecology of other clades in the same ecosystems (e.g. invertebrates and tetrapods) has likely been substantial (Vermeij 1977; Brett and Walker 2002; Stanley 2008; Bush and Bambach 2011; Sallan et al. 2011). Ignorance of historical patterns of fish diversity, therefore, impedes a general understanding of aquatic biodiversity and evolutionary processes.
Several factors have prevented a synoptic picture of macroevolutionary patterns in fishes from materializing. Key among these is the sheer scope of the problem: no other vertebrate assemblage encompasses as much taxonomic richness and morphological disparity, distributed over such a long geological interval and represented by such a diverse range of preservational styles as fishes. But this extensive record remains relatively understudied, as a review of the table of contents in any issue of a palaeontological journal will illustrate. Since the time of Agassiz, the small community of fish palaeontologists has focused primarily on taxonomy, description and interrelationships. As a result, fish workers were among the early adopters and innovators of cladistic methods late in the 20th century (e.g. Nelson 1969) and among the first to apply then novel ‘tree-thinking’ approaches to the fossil record (Patterson and Rosen 1977). By contrast, the coincident ‘Palaeobiological Revolution’ (Sepkoski and Ruse 2009), with its emphasis on common macroevolutionary trends among disparate clades, was barely noted.
Macroevolutionary studies have been perceived as of secondary importance by most fish workers for many reasons: the very real need to catalogue available and newly discovered material, assumptions of scarcity relative to the invertebrate record, and unfamiliarity with palaeobiological methods. In addition, systematic and palaeobiological analyses impose distinct demands on the worker. Cladistic analysis requires detailed investigations of a small set of exemplars, selected on the basis of group membership and phylogenetically informative characters. Macroevolutionary analysis requires large-scale surveys of diverse taxa selected on the basis of factors generally not specific to a particular group (e.g. faunal membership, ecology and age or environment).
As a result of this dichotomy, the literature concerning patterns of extinction and recovery in fishes is small and diffuse, making it difficult for workers both within and outside the field to make meaningful statements about patterns of diversity. Under the theme of extinction and recovery, our goal is to provide a broad overview of major patterns of turnover in the fossil record of fishes, so far as they can be discerned at present, and highlight areas of persistent uncertainty that should be the target of future research. While we have attempted to present a balanced picture of Phanerozoic fish diversity patterns, this topic is vast and our review is inevitably biased towards our own areas of expertise, centred on the end-Devonian (Sallan) and end-Cretaceous (Friedman) extinctions. Our taxonomic scope encompasses the vertebrate total group exclusive of tetrapods (Fig. 1). There is mounting evidence that cyclostomes are monophyletic (Heimberg et al. 2010; Janvier 2011; Ota et al. 2011), meaning that the vertebrate crown group includes hagfishes in addition to gnathostomes and lampreys. In contrast to most previous surveys of fish diversity in the fossil record, we include conodonts in our review. There is some conflict concerning the precise systematic position of the group (e.g. Turner et al. 2010), but all explicit hypotheses of conodont relationships place them as members of the vertebrate total group. A summary of hypothesized relationships between major groups discussed elsewhere in this paper is given in Figure 1.
An overview of the history of fish diversity
A detailed picture of diversity patterns in the fossil record of fishes is lacking, despite synoptic attempts dating back to the 19th century (Agassiz 1833–1844; Marsh 1877). Broad overviews are presented as range charts in relevant chapters of The Fossil Record (Andrews et al. 1967) and The Fossil Record 2 (Aldridge and Smith 1993; Cappetta et al. 1993; Gardiner 1993a, b; Halstead 1993; Patterson 1993a; Schultze 1993; Zidek 1993). Richness-through-time plots have been produced for different fish groups (Fig. 1) during all (osteichthyans, or bony fishes: Patterson 1994) or part (Jurassic-Recent neoselachians (anatomically modern sharks and rays): Cappetta 1987a; Late Cretaceous-early Palaeogene neoselachians: Kriwet and Benton 2004; Late Jurassic – Late Cretaceous actinopterygians (ray-finned fishes): Cavin et al. 2007 Jurassic actinopterygians: Cavin 2010; Middle Triassic – Late Cretaceous neoselachians: Kriwet et al. 2009; Ordovician–Triassic vertebrates excluding conodonts: Blieck 2011; Devonian vertebrates excluding conodonts: Long 1993; Devonian–Carboniferous jawed vertebrates: Sallan and Coates 2010) of their evolutionary history. More synoptic spindle diagrams or ‘Romerograms’ (so-called because of their ubiquity in Romer’s classic vertebrate palaeontology text) are prolific, appearing in various formats since Agassiz (1833–1844), but these are generally schematic to the point that no precise statements might be drawn from them, and their connections to empirical counts of taxonomic diversity often seem murky at best (e.g. Carroll 1988, figs 3.1, 4.4, 5.2, 6.7). Clear diagrams of fish diversity over the Phanerozoic, where the axis representing taxonomic richness is explicitly scaled, are comparatively rare, and we are aware of only two examples. The first was constructed by Thomson (1977, fig 1) at the genus level, based largely on taxonomic appendices to Romer’s Vertebrate Palaeontology (1966). The second derives from family ranges taken from The Fossil Record 2 and appears as a small figure in an inset text box in various editions of Benton’s Vertebrate Palaeontology (2000, p. 34, 2005, p. 36).
Until the advent of the Paleobiology Database (PBDB), Sepkoski’s (2002) genus-level compendium represented the standard global resource for investigating extinction and recovery in the marine record, but it appears that the specific pattern of diversification for fishes recorded by this dataset has never been published. Rather, fish groups have been included as a notable but minor component of marine biodiversity curves, such as Sepkoski’s marine evolutionary faunas (1982; Bush and Bambach 2011). Here, we include a plot of Sepkoski’s data (Fig. 2), which reveals a trajectory similar to that of Thomson (1977) and Benton (2000, 2005). There are two major peaks of richness: one in the Late Devonian and another approaching the Recent. Actinopterygians, specifically teleosts, are responsible for the latter. A shallow trough, extending from the Carboniferous to mid-Cretaceous, lies between these two intervals of apparently high richness. This long span is punctuated by a series of minor peaks and valleys that, for the most part, are not consistent between the three datasets.
Using Foote’s (2000) expressions, we made estimates of per capita rates of extinction and origination from all of Sepkoski’s (2002) fish data, as well as calculating these rates separately for the two major groups of living fishes, chondrichthyans (cartilaginous fishes, or sharks, chimaeras and their relatives) and osteichthyans (actinopterygians and lobe-finned sarcopterygians, the latter exclusive of tetrapods here) (Figs 2–4). These rates cannot be calculated for all intervals, but where they can, we have estimated 95 per cent confidence intervals about them based on binomial probability. Intervals of high extinction and origination rates are discussed in the relevant sections. These estimates appear robust to the inclusion or exclusion of conodonts. Rates of origination (raw estimates: Spearman’s rho = 0.77, p = 7.9.10−7; first differences: Spearman’s rho = 0.77, p = 1.0.10−6) and extinction (raw estimates: Spearman’s rho = 0.67, p = 2.9.10−5; first differences: Spearman’s rho = 0.68, p = 3.2.10−5) based on Sepkoski’s data including and excluding conodonts are strongly correlated over the early Ordovician to end-Jurassic interval, representing the span for which meaningful estimates can be made for both datasets and which precedes the total extinction of conodonts at the Triassic–Jurassic boundary.
Reliability of apparent diversity patterns
Historically, some of the harshest criticism levelled against taxonomic databases of fossils and the research programmes built around them have come from prominent palaeontologists (Patterson and Smith 1987, 1989; Smith and Patterson 1988), including fish workers. Such critiques of taxic palaeobiology have probably done much to discourage palaeoichthyologists from exploring large-scale diversity patterns in the fossil record of fishes, but the question remains: how unreliable are these datasets? Patterson and Smith (1989) estimated that roughly 65 per cent of the fish data used in Sepkoski’s then unpublished generic compendium represented ‘noise’: para- or polyphyletic groups, incorrectly dated monophyletic groups, and various classes of other errors. This figure agrees closely with Adrain and Westrop’s (2000) estimate of the error rate for trilobites in this same database (Sepkoski 2002). Importantly, this error appears to have been randomly distributed throughout the trilobite data, resulting in little net change between patterns reconstructed on the basis of Sepkoski’s compendium and expert-vetted versions. This confirmed the assumptions of diversity-focused researchers that random taxonomic biases should be cancelled out in sufficiently large datasets and that real signal should appear above a threshold level of sampling. Patterson and Smith (1987, 1989) and Smith and Patterson (1988) were largely concerned with the reliability of last appearances, and it is not clear from their analyses the degree to which their screened fish data produce richness trajectories comparable with those reconstructed directly from Sepkoski’s data.
We have not comprehensively vetted Sepkoski’s genus compendium, but note the following broad issues that should be addressed in future attempts to build upon his efforts and more completely catalogue Phanerozoic fish diversity. The first is a lack of nonmarine and many euryhaline taxa. This oversight might seem obvious when discussing a marine diversity curve. However, euryhaline vertebrates are common, and many lineages have alternated between marine and freshwater habitats during their evolutionary histories, some more than once. Thus, consideration of only marine occurrences may truncate taxon ranges, creating artificial extinction and origination events. This bias is likely compounded during periods when vertebrate-bearing marine outcrop area is low, such as during the later Palaeozoic (McGowan and Smith 2008).
The second issue concerns the exclusion of major marine lineages and localities. Many taxa and faunas are missing, such as the diverse actinopterygians found in Mississippian deposits in Glencartholm, Scotland (Dineley and Metcalf 1999; Sallan and Coates 2010). This oversight is puzzling because this same Lagerstätte contains a range of invertebrate taxa (Schram 1983). However, the absence of other sites might be chalked up to taphonomic bias: in many cases, shelly faunas tend to occur in layers where articulated vertebrates do not (Schram 1979; Briggs and Clarkson 1983; Baird et al. 1985). By contrast, Sepkoski’s coverage of marine chondrichthyan ichthyoliths (isolated elements, such as scales and teeth) appears less spotty, perhaps because enamel-bearing elements are likely to co-occur with the hard parts of invertebrates (Sallan et al. 2011). An alternative, nontaphonomic, explanation for this pattern might be that comprehensive summaries of fossil chondrichthyan diversity have been catalogued in two volumes of the Handbook of Palaeoichthyology available to Sepkoski (Zangerl 1981; Cappetta 1987b), while four planned osteichthyan volumes (Dipnoi (lungfishes), Crossopterygii (nonlungfish sarcopterygian fishes), Actinopterygii I, Actinopterygii II) have never been completed or published.
It should be noted that preservational issues and unfamiliarity are also likely to account for an underestimate of vertebrate occurrences in the PBDB. Vertebrates are less likely to appear in productive invertebrate collections or localities, while those that do occur may be underreported by workers unfamiliar with fish taxonomy. For example, at the time of writing the PBDB contained just three records for the Devonian antiarch placoderm (armoured fish) Bothriolepis, which is represented by more than seventy nominal species globally (L. C. S., pers. obs. 2011) and is often the most abundant taxon at Late Devonian vertebrate localities (E. Luksevics, pers. comm. 2009). In addition, it lacked records for any fish taxon at most of the Palaeozoic Sites of Special Scientific Interest compiled by the British Geological Survey (Dineley and Metcalf 1999).
Another issue concerns the incorrect assignment of taxon ages. As an illustration of this problem, Sepkoski designated many long-ranging actinopterygian lineages of the later Palaeozoic as exclusively Tournaisian. Some of these genera, such as Eurynotus, have been known to occur in multiple stages ever since they were first described by Agassiz (1833–1844) or other 19th century palaeoichthyologists. Alternatively, other genera are given undue longevity, assigned to bins such as Carboniferous–Permian, when their more restricted ranges can be found in the literature. An additional class of error relates to outdated higher-level taxonomic assignments, which are particularly common for Palaeozoic and Mesozoic actinopterygians recorded in the PBDB. Remedying these problems will require extensive recompilation of the primary data.
Finally, there are reasons for concern about geological and taxonomic biases in all databases of raw taxonomic diversity, including Sepkoski’s compendium (Raup 1979; Peters and Foote 2001; McGowan and Smith 2008). Biodiversity trends in single environments and regions can be linked to the relative amount of outcrop area available in each bin, perhaps explaining to such phenomena as multistage troughs (e.g. the Permo-Carboniferous; Fig. 1 and below), and the ‘Pull of the Recent’ (Raup, 1979; Alroy et al. 2008). The latter is a marked, even exponential, increase in biodiversity over the Mesozoic–Cenozoic (see osteichthyans in Fig. 2). This has been linked to the appearance of diagnostic characters shared with better-sampled living taxa and an increase in pristine outcrops towards the present (Raup 1979; Alroy et al. 2008). However, it is possible that the correlations between deposit availability and biodiversity might reflect an interaction or a ‘common cause’ rather than a bias (Hannisdal and Peters 2011), and for some groups, the increase in richness towards the Recent might reflect a real pattern rather than an artefact (Jablonski et al. 2003; Alroy et al. 2008). Lastly, there is the potential that differential worker output (sampling intensity) has affected taxonomic counts in different intervals and environments (Butler et al. 2011; Lloyd et al. 2011, 2012b). However, sampling standardization, use of higher taxonomic levels (above species), total coverage of habitats and provinces can sometimes mitigate this effect (Alroy et al. 2008; Wall et al. 2009; Lloyd et al. 2011, 2012b). The total impact of sampling biases on our knowledge of evolutionary history continues to be an area of heavy debate in palaeobiology, and one that is yet to be explored in any meaningful way for fishes.
These problems aside, we present Sepkoski’s (2002) data unaltered here as a reflection of the state of the art, and as a motivation for fish workers to combine their efforts and expertise in order to produce refined, accessible databases for investigating biodiversity patterns throughout the Phanerozoic.
Early to mid-Palaeozoic events
The Cambrian event: the origin of vertebrates
Vertebrates, like many other metazoan groups, seem to make their first appearance in the fossil record during the Cambrian (542–488 Million years ago (Ma); throughout, interval dates based on Gradstein et al. 2004) explosion. Various soft-bodied taxa, largely from the lower Cambrian Chenjiang and middle Cambrian Burgess Shale Lagerstätten have been implicated in vertebrate origins, with Cathaymyrus, Haikouichthys, Metaspriggina, Nectocaris, Pikaia, mylokunmingiids, and yunnanozoons all interpreted at some point as members of either the vertebrate total or crown group (Shu et al. 1999, 2003; Donoghue and Purnell 2009; Sansom et al. 2010b). Of these, Haikouichthys seems to be the most likely vertebrate, with preserved features identified as definitive synapomorphies such as paired sensory capsules and vertebrae (Shu et al. 2003; Sansom et al. 2011), although the interpretation of structures preserved in these soft-bodied fossils is fraught with difficulty (Donoghue and Purnell 2009; Sansom et al. 2010b).
Late Cambrian to mid-Ordovician (488–444 Ma) deposits from Euramerica and Australia contain what might be the earliest armoured vertebrate: fragments of layered, dentine-bearing tissue mostly attributed to the genus Anatolepis (Repetski 1978; Smith et al. 1996; Young et al. 1996; Clark et al. 1999). This material resembles, both superficially and histologically, remains of Porophoraspis and Arandaspis, armoured agnathans (jawless fishes collectively called ‘ostracoderms’) known from the later Ordovician (Repetski 1978; Smith et al. 1996). Such fragments might mark the earliest records of the ‘Arandaspida’, a potentially paraphyletic assemblage branching near the base of the gnathostome stem and usually allied with Silurian and stratigraphically younger heterostracan agnathans (Janvier 1996; Sansom et al. 2001; Fig. 1). If the identity of Anatolepis and other early fossil material is correct, it would indicate that the split between cyclostomes (lamprey and hagfish) and jawed vertebrates (including all armoured agnathans along their stem) had occurred by this point (Janvier 1996; Smith et al. 1996), a conclusion supported by late Cambrian conodonts (Aldridge and Smith 1993; Donoghue et al. 2000).
The Ordovician and Silurian: the origin of gnathostomes
As in the Cambrian, Ordovician vertebrate macrofossils are rare, with the most complete available body fossils assigned to a small number of arandaspid agnathans and allied genera (Arandaspis, Sacabambaspis, Eriptychius, Astrapis, Porophoraspis). These earliest definitive, articulated armoured agnathans occur in second half of the Ordovician and are found at just a handful of sites in Australia (Young 2009) and the Americas (e.g. the Harding Sandstone of Colorado; Sansom et al. 2001; Bolivia: Gagnier et al. 1986). Fragments of histologically similar material have been recovered from further afield (Australia, Arabia, Siberia, South China, and possibly Mongolia and South Africa: Wang and Zhu 1997; Blieck and Turner 2003; Karatajute-Talimaa and Smith 2004; Sansom et al. 2009).
While the macrofossil record might indicate that fishes were neither abundant nor diverse during the Ordovician, ichthyoliths have been used to suggest otherwise (Sansom et al. 2001, 2009). Conodont elements, the tooth-like feeding apparati of soft-bodied agnathans, are widespread as they are throughout the Palaeozoic and are joined by a host of additional microfossils. Histological analyses have been used to identify isolated bone fragments as heterostracan- and placoderm-grade taxa, while ‘odontodes’ and spines have been given presumptive chondrichthyan or acanthodian (‘spiny shark’) status on the basis of broad morphological similarities (Sansom et al. 1996, 2001, 2012; Young 1997; Smith et al. 2002). Such assignments have been used to argue for the origin of major gnathostome divisions (and indeed the gnathostome crown; Fig. 1) in the Ordovician or even Cambrian (Young 1997; Smith et al. 2002), tens of millions of years before the earliest definitive jawed body fossil (see below; Zhu et al. 2009). Contrary to this hypothesis, no Ordovician jaw fragments or teeth are known, and tissues (e.g. dentine) and morphological traits (e.g. pore systems) once thought to be restricted in their distribution are now known to be phylogenetically widespread (Sansom et al. 1992; Young et al. 1996). Microfossils of this age assigned to various jawed vertebrate groups lack definitive morphological characteristics and occur alongside very similar ichthyoliths belonging to thelodont and arandaspid agnathans (Sansom et al. 1996; Young 1997; Blieck and Turner 2003; Donoghue et al. 2003).
Unfortunately, the relative absence of identifiable articulated specimens, especially of putative jawed vertebrates, means little can be said about either the taxonomic or the ecological impact of the end-Ordovician Hirnantian mass extinction (444 Ma) on fishes aside from conodonts, which suffered extreme loss that is clearly expressed in our estimated rate of extinction from Sepkoski’s (2002) data (Fig. 2; see also Bambach 2006). This glacial event had significant and severe impacts on marine invertebrates perhaps equivalent in scope to the end-Permian and end-Cretaceous (Raup and Sepkoski 1982; Bambach et al. 2004; Bambach 2006), although ecosystem restructuring may be been light relative to these other major events (Droser et al. 2000; McGhee et al. 2012). As most known Ordovician fish material derives from similar marine settings, our present inability to make more definitive statements about how this extinction affected fishes more broadly is frustrating. Some remedy might be forthcoming, as a number of new Ordovician Lagerstätten have recently been reported. These include Burgess Shale/Chengjiang-type style preservation that could deliver the body remains of naked vertebrates such as Haikouichthys, stem-lampreys and hagfishes, and conodonts (Gabbott et al. 1995; Young et al. 2007; Van Roy et al. 2010) along with taxa bearing mineralized external skeletons.
Mass extinctions like the end-Ordovician might have long-term impacts even on clades suffering minimal direct effects, as common ecosystems are altered and new opportunities arise in the aftermath (Friedman 2010; Sallan et al. 2011). The Silurian (444–416 Ma) record presents a number of potential roadblocks to detecting such patterns. It is undergoing intense stratigraphic and age revision because of isotope volatility, unconformities at various boundaries, a lack of major biozonation taxa (graptolites) in those facies which preserve most other fossils, and a large number of apparent marine extinction pulses that strongly affected those same index taxa (graptolites and conodonts; Munnecke et al. 2011). Despite these issues, the Silurian record does contain a greater (and better supported, in terms of precision of possible phylogenetic placement) richness of fishes than the Ordovician (Qu et al. 2010), allowing some diversity patterns to be parsed with appropriate caveats.
Vertebrate fossils are uncommon in the early part of the first Silurian series, the Llandovery (444–428 Ma), yet articulated material appears by the middle of the interval. These primarily consist of scaled agnathans representing the earliest articulated thelodonts (Loganellia: Dineley and Metcalf 1999), with scale morphologies matching Ordovician ichthyoliths (Sansom et al. 1996), and naked jawless forms (Jamoytius) attributed to lineages originating near the base of the vertebrate tree (Janvier 1996; Sansom et al. 2010a). The earliest armoured agnathan specimens in the Silurian appear in China (galeaspids at multiple localities: Zhao and Zhu 2007) and Canada (the possible heterostracan Athenagis: Janvier 1996; Soehn et al. 2001) in the last stage of the Llandovery. Articulated material from the subsequent Wenlock series (428–423 Ma) represents a phylogenetically broad sample of agnathan diversity, adding heterostracans and osteostracans to the body fossil record (Janvier and Blieck 1993; Wilson and Caldwell 1993; Von Bitter et al. 2007). This raw increase in richness, following what seems to be an earliest Silurian gap (Talimaa’s Gap: Blieck 2011) in the vertebrate record lasting some 3–10 Ma, might suggest that vertebrates underwent a diversification event during or after a post-Ordovician recovery interval (Janvier 1996). However, the aforementioned uncertainties about the severity of the Hirnantian extinction, united with ambiguities surrounding the true phylogenetic affinities and abundance of some fossil material (macro and micro) from the Llandovery and Ordovician, and uncontrolled sampling, mean that the significance of this pattern remains unclear.
Considering this ichthyolith evidence, it is interesting that no articulated jawed vertebrate material is known from pre-Ludlow sites with taphonomy similar to Xiaoxing or later Palaeozoic Lagerstätten. For example, conodonts and heterostracans are the only vertebrates known from the soft- and hard- tissue-preserving layers of the Eramosa fauna of the Canadian Wenlock (Von Bitter et al. 2007), while only conodont elements appear at the similarly aged hard-tissue-biased Waukesha locality in Wisconsin. Similarly, the Llandovery deposits of Lesmahagow, Scotland contain only the thelodont Loganellia and the naked Jamoytius (Dineley and Metcalf 1999). Devonian (416–359 Ma) and Carboniferous (359–299 Ma) sites with such taxa and/or preservational modes also bear articulated chondrichthyans, osteichthyans and placoderms (e.g. Hunsrückschiefer, Germany; Miguasha, Canada; Gogo, Australia; Cleveland Shale, USA; Waterloo, South Africa; Bear Gulch, USA; Mazon Creek, USA: Gess and Hiller 1995; Janvier 1996; Baird 1997; Gess et al. 2006). Collectively, these observations imply that if jawed vertebrates had arisen by the earliest Silurian, they existed at sufficiently low frequencies to evade preservation. An alternative explanation is that early jawed gnathostomes inhabited environments that either have been little sampled by palaeontologists or are poorly represented in the stratigraphic record. However, the abundance of gnathostome microfossils, and later macrofossils, in Silurian faunas bearing corals and other well-sampled invertebrates, and aforementioned scarcity in preservational Lagerstätten of similar age suggests that sampling is not the main issue (Janvier 1996; Zhao and Zhu 2007; Žigaite et al. 2011).
Late Silurian to mid-Devonian: the rise of ecologically modern fishes
Even though major jawed gnathostome divisions originated by the mid-Silurian, fish assemblages exclusive of conodonts were still, by and large, dominated by jawless forms into the Early Devonian. As illustrated earlier, definitive jawed vertebrate remains and associated ichthyoliths are rare in Ordovician–early Devonian deposits, with jawed fishes generally representing minor contributors to vertebrate diversity even at the best-sampled localities (Boucot and Janis 1983; Anderson et al. 2011). By contrast, vertebrate assemblages from the close of the Early Devonian to the Recent are composed almost exclusively of jawed gnathostomes (Anderson et al. 2011). This major faunal turnover is therefore not coincident with the origin of anatomically modern gnathostomes, and anecdotal scenarios linking armoured agnathan decline to jawed gnathostome competition have not been corroborated by quantitative investigation (Janvier and Blieck 1993; Anderson et al. 2011; Purnell 2001a).
Significantly, the earliest diverging branches of the jawed vertebrate tree represent dorsoventrally compressed forms (antiarch and petalichthyid placoderms: Friedman 2007; Brazeau 2009) that probably had benthic ecologies not dissimilar from those of the most crownward jawless members of the gnathostome stem. The origin of modern jawed vertebrates and their subsequent evolutionary success is therefore probably as much about the shift from a benthic to nektonic lifestyle in the late Silurian – Early Devonian as it is about the origin of jaws (Anderson et al. 2011; Klug et al. 2011). However, this is not the whole story, as eel-like conodonts and cyclostomes, fusiform and even deep-bodied unarmoured and scale-bearing agnathans (e.g. anaspids and thelodonts), and streamlined heterostracans were likely already nektonic in the Cambrian–Silurian (Briggs et al. 1983; Janvier and Blieck 1993; Wilson and Caldwell 1993; Purnell 2001a, b; Botella and Farina 2008). Many jaw-bearing fishes, both early and modern, are ram, suspension or suction feeders, acquiring mobile and planktonic prey without any major processing by the jaws or dentition (Anderson et al. 2011). Nektonic jawless fishes cannot be excluded a priori from these and other trophic strategies (Purnell 2001b), especially in the light of the functional capabilities of living agnathans (Clark and Summers 2007; Clark et al. 2010).
A nonecological explanation has been also been offered for the abrupt diversification of jawed vertebrates lineages long after their origin: an increase in atmospheric oxygen levels in the late Silurian and/or early Devonian that presumably allowed more energetic life modes and larger taxa (Dahl et al. 2010; Qu et al. 2010). Anderson and colleagues (2011) have noted that fish jaws reached near-peak functional diversity in the Devonian before one of the proposed causal maxima (Emsian (407–397 Ma) peak in Dahl et al. 2010). Regarding other peaks (e.g. Lockhovian (416–411 Ma) and Pragian (411–407 Ma): Berner et al. 2007; Qu et al. 2010) and long-term changes (e.g. Clement and Long 2010), it is not clear that there is, or should be, a general relationship between atmospheric oxygen and fish diversity. Chondrichthyans, and possibly many of the acanthodians that populate their stem (Brazeau 2009), do not possess lungs (Nelson 2006). It is not known if the most dominant and environmentally widespread group of Devonian gnathostomes, the placoderms, were capable of directly accessing atmospheric oxygen either; previous evidence of such organs in Bothriolepis is now regarded as ambiguous (Arsenault et al. 2004; Janvier et al. 2007). The limiting factors for oxygen uptake in living fishes seem to be physiological, morphological and behavioural (e.g. lung and gill surface area, structure and efficiency, ram speed, buccal pumping, energy spent on surfacing, trade-offs with other gill functions) rather than environmental (available oxygen) under normal conditions (Piiper 1982; Kramer 1987; McNab 2002).
If there were a relationship between oxygen levels and gnathostome success, one would expect a mass depletion of fishes coincident with a fall in oxygenation later in the Devonian (Clement and Long 2010; Qu et al. 2010). This is especially true because, aside from increasing energy from aerobic respiration, the link between oxygen levels and diversity is thought to run through positive correlations with plankton abundance (a key factor in most aquatic ecosystems; Berner et al. 2007). However, jawed vertebrate functional, faunal and size diversity remains consistent or even increase throughout multiple hypoxic events and long-term oxygen decline (Dahl et al. 2010; Sallan and Coates 2010; Anderson et al. 2011). Indeed, a previous correlation between a nadir in tetrapod abundance in the early Carboniferous (‘Romer’s gap’: Coates and Clack 1995) and low oxygen levels (Ward et al. 2006) might instead be interpreted as a postextinction trough with associated environmental effects (see below; Sallan and Coates 2010).
Devonian extinctions and carboniferous recovery
Newell (1952, 1962) was the first to identify a Late Devonian (385–359 Ma) mass extinction and associated losses among fishes, claiming 90 per cent family-level extinction while providing few details as to specific groups. McLaren (1970) indicated the marine extinction as occurring at the Frasnian–Famennian boundary (374.5 Ma), and subsequently that horizon, or the surrounding Kellwasser deposits, was placed among the Big Five mass extinction events, albeit as the least statistically significant member (Raup and Sepkoski 1982). This uncertainty is compounded, if not caused, by the nexus of distinct bioevents in the Late Devonian that encompasses the 25-million-year ‘Late Devonian Biodiversity Crisis’ (Stigall 2010). The most heavily investigated of these continues to be the Kellwasser event, linked to the loss of 60 per cent of genera and up to 82 per cent of species (McGhee 1996). However, large events have also been identified at the end-Givetian (the Taghanic event; 385 Ma) and the end-Famennian (the Hangenberg event; 359 Ma) (McGhee 1996; Caplan and Bustin 1999; Racki 2005; Sallan and Coates 2010).
It has long been apparent that the vertebrate fauna changed over the later Devonian into the Carboniferous. Spurred by Agassiz (1833–1844) and confronted by a wealth of appropriately aged British and North American fossil material, 19th century fish palaeontologists noticed a number of faunal differences between the ‘Age of Fishes’ and the ‘Age of Amphibians’. Long before the concept of a Devonian mass extinction was introduced, O. C. Marsh (1877, p. 214) remarked on the complete extinction of placoderms by the end-Devonian, followed by the rise of previously ‘subordinate’ chondrichthyans in ‘both numbers and size’. However, this early recognition of change did not lead to further investigation of the connections with various global events, even after acceptance of vertebrate mass extinction elsewhere in the record. Nearly 120 years later, Janvier (1996) could only note that there were indeed large radiations of holocephalan chondrichthyans (chimaeras and their relatives), actinopterygians and tetrapods in the early Carboniferous. Their relationship with any extinction-associated losses of marine placoderms was merely hypothetical and the cause of such extinctions was in doubt.
In fact, many fish workers have argued for other sources of faunal change. Janvier (1985) and Ahlberg and Milner (1994) attributed the loss of Late Devonian Old Red Sandstone deposits and their associated fish fauna, and equivalent facies elsewhere, to lack of deposition, or even destruction, of appropriate sediments. Several subsequent authors proposed that a Euramerican regional biota had been swept away by invasions during a faunal interchange with Gondwana (Clement et al. 2004; Janvier and Clement 2005; Clack 2006). Rhizodontid and megalichthyid sarcopterygians and gyracanthid acanthodians, known from both Devonian and Carboniferous deposits, supposedly caused turnover in previously isolated ecosystems by outcompeting or preying upon placoderms and early tetrapods (Clement et al. 2004; Clack 2006). Similarly, changeover from a Devonian antiarch placoderm-porolepiform sarcopterygian biota to a Carboniferous chondrichthyan-actinopterygian assemblage in Russia was dismissed as a local phenomenon (Alekseev et al. 1994; Clack 2006).
Even when the possibility of global turnover has been considered, change was usually assumed to be gradual and limited to the marine realm. Freshwater deposits, by contrast, were thought act as a refugium because of the absence of known extinction impacts in those environments (this is, of course, a truism, reflecting the fact that mass extinctions have been inferred on the basis of marine invertebrates). Zhao and Zhu (2007) placed this faunal turnover in the Devonian, providing no biological explanation and attributing the pattern to the inevitable replacement of stem by crown. Blieck (2011) considered marine durophagous placoderms and sarcopterygians to have been replaced by holocephalans and actinopterygians, respectively. Even Romer’s gap (Coates and Clack 1995) has been linked to a lack of nonmarine or continental deposits (Clack 2007; Blieck 2011). This demonstrably false claim (Peters and Foote 2001; McGowan and Smith 2008) was made in spite of concerted exploration of such deposits (Garvey 2005), similar patterns among other terrestrial forms (e.g. arthropods: Ward et al. 2006), and a coincident bottleneck in tetrapod phylogeny (Ruta et al. 2006; Coates et al. 2008).
This is not to say there have not been any attempts to quantify extinction impacts on Devonian vertebrates. Stratigraphic studies provide a wealth of occurrence data, and analyses of marine impacts often include fishes present within large databases or found at invertebrate sites (McGhee 1996; Hallam and Wignall 1999; Stanley 2008). Long (1993) devoted an entire chapter of a volume on the Devonian record to the impact of mass extinctions. Using family-level and range data from published compendia, including The Fossil Record (Andrews et al. 1967) and various volumes of the Handbook of Palaeoichthyology (HoP), origination and extinction rates were reconstructed for all vertebrates as well as major traditional taxonomic subgroups (Long 1993). Blieck (2011) augmented this with family-level diversity curves based on The Fossil Record 2. Most recently, Sallan compiled genus-level occurrence and species-level faunal composition databases for the mid-Devonian through Mississippian (Sallan and Coates 2010; Sallan et al. 2011). These databases cover both marine and nonmarine taxa, dealing with the aforementioned issues concerning euryhaline taxa and exclusion of continental faunas from larger datasets. Here, we discuss the impacts of the Kellwasser and Hangenberg events based on these studies.
The Kellwasser event
Conodonts, armoured agnathans and all manner of marine gnathostomes have been counted among the victims of the Kellwasser extinctions, which Long (1993) estimated as the most major of all Devonian bioevents for vertebrates. Conodont faunas experience nearly complete turnover, as usual for any stage boundary (and necessary for their use in biozonation). Aldridge (1988) estimated losses of 89 per cent of conodont species, including all those bearing coniform elements, while stratigraphically important genera were reduced further. McGhee (1996), citing Sandberg and colleagues (1988), suggested that losses among conodonts took place over a time span of a few days to 20 000 years.
Armoured agnathans, including members of several divisions of jawless stem gnathostomes such as heterostracans, osteostracans, galeaspids and thelodonts (Fig 1), are often considered victims of this same event; Long (1993) and McGhee (1996), working from different data, indicated that the youngest armoured agnathans are Frasnian (385–374 Ma) in age. However, both authors admitted that these groups had been in decline for quite some time, reduced to a handful of mostly near-shore dwellers by the Frasnian. It is now apparent that the last appearances of armoured agnathans are staggered throughout Late Devonian. Osteostracans cross the Givetian–Frasnian boundary only to make their final appearance at the Miguasha Lagerstätte, before or around the mid-Frasnian (Janvier 1996). Those agnathan groups which last appear closer to the end of the Frasnian–heterostracans and galeaspids – had been restricted in both diversity, abundance, habitat and range since the mid-Devonian (Halstead 1993; Janvier 1996; Jia et al. 2010). In addition, a thelodont ichthyolith taxon has been recovered from the early-mid Famennian (374–359 Ma) of Iran (Hairapetian and Turner 2003). It is therefore unlikely that a sudden, abiotically driven Kellwasser event was the primary cause of armoured agnathan extinction in the Devonian.
Turning to jawed vertebrates, marine placoderms were long counted among the dead. Using HoP data compiled by Denison (1978, 1979), Long (1993) and McGhee (1996) calculated losses of up to 50 per cent of species and cosmopolitan families, with another third forced into nonmarine refugia. Diversity curves published by Sallan and Coates (2010), based on an aforementioned genus-level compilation, and Blieck (2011), based on family data from the Fossil Record 2 (Gardiner 1993a), show similar reductions in placoderm diversity. However, many of these losses are attributable to the loss of endemic species and monotypic families found only at well-sampled Lagerstätten situated well before the Kellwasser horizon (e.g. Gogo, Australia and Bad Wildungen, Germany: Sallan and Coates 2010) (Fig. 4), as well as a relative lack of similarly productive marine localities within the Famennian. In fact, the proportion of placoderm species at Famennian marine faunas was unchanged relative to the Frasnian (Sallan and Coates 2010). Similarly, recorded losses among other jawed fishes were largely limited to families with low diversity, many of them in decline long before the Kellwasser event (e.g. acanthodians, diplocercid coelacanths and cheirolepidid actinopterygians: Long 1993; Sallan and Coates 2010).
In summary, there does not seem to be evidence indicating a single, sudden vertebrate extinction event at the Frasnian–Famennian boundary. Instead, differences between Frasnian and Famennian diversity were the culmination of millions of years of gradual environmental and faunal change viewed through the lens of imperfect sampling. Sallan and Coates (2010) showed that there were no significant differences in the composition of late Frasnian and early Famennian jawed vertebrate assemblages at higher taxonomic levels. The most substantial change was an increase in the number of chondrichthyans, which might be attributable to the relative abundance of ichthyolith-bearing black shales in the Famennian. Likewise, Anderson and colleagues (2011) failed to detect a statistically significant reduction in ecological diversity based on the disparity of the gnathostome jaws and dentition. The diversity of durophagous lungfishes is substantially reduced in the Famennian compared with the Frasnian (Anderson et al. 2011), but this might simply reflect a failure to sample suitable shallow marine environments (Sallan and Coates 2010).
Therefore, observed diversity changes in the later Devonian are probably due to greater cosmopolitanism and gradual environmental trends, as well as basic sampling issues (Long 1993). As Pangaea formed, the supercontinents of Laurussia and Gondwana were brought together around the Southern tropics, destroying the near-shore and marine habitat and facilitating migration between once isolated faunas (Copper 1986; Young 2011). Marine invertebrate losses over the later Devonian have been attributed to a long phase of low origination rather than significantly heightened extinction rates (Bambach et al. 2004), a phenomenon linked to invasive migrations and subsequent reduction in endemism (Stigall 2010). However, there was at least one significant bioevent around the Kellwasser that cannot be explained by these circumstances: the loss of most metazoan reef builders (e.g. corals and stromatoporoid sponges), which did not fully recover until the Carboniferous (Copper 2002; Kiessling and Simpson 2011). It is surprising that marine vertebrate lineages occupying Frasnian carbonate platform reefs were not much affected by their loss; most simply moved on to Famennian replacement habitats, such as calcified algal structures and crinoid-dominated carbonate ramps, sometimes in the same vicinity (e.g. the Canning Basin complex and other Frasnian-Famennian localities around Australia; Wood 2000; Copper 2002; Waters and Webster 2009; Young et al. 2010; Sallan et al. 2011).
The Hangenberg event
While evidence for a Kellwasser event supports the cautiousness of most Devonian fish workers towards bioevents, the Hangenberg appears to represent a real, abiotically driven mass extinction. Acknowledgement of the large magnitude of the end-Devonian event is recent and not yet universal; the extinction was long considered a second-order pulse, no larger than others throughout the Famennian (Racki 2005). The first review of the biotic impact of the event was published in 1999 (Caplan and Bustin 1999) and as recently as 2010 geologists and palaeontologists could declare the Hangenberg and associated abiotic events ‘underrated’ among Devonian extinctions (Racki 2005; Bambach 2006; Brezinski et al. 2010). Part of the issue was stratigraphic; Hangenberg black shales were assumed to occur within the Devonian praesulcata conodont zone, perhaps representing the boundary between the Famennian and the ‘Strunian’. Exact correlation was impeded by a lack of biozonation taxa, such as ammonites, conodonts and palynomorphs: all victims of the Hangenberg (Caplan and Bustin 1999; Korn 2000; Kaiser et al. 2008). This stratigraphic uncertainty has enabled claims that classic Devonian vertebrates (e.g. placoderms) and faunas continued above the Hangenberg horizon (Alekseev et al. 1994; Young 2003; Blom et al. 2007).
Recent work has shown that Hangenberg sediments comprise both anoxic black shales and the sandstones above them, with composition and thickness dependent on environment (Buggisch and Joachimski 2006; Brezinski et al. 2008, 2010; Kaiser et al. 2008). The Hangenberg horizon is now designated as the Devonian–Carboniferous boundary and thus the focus of concerted stratigraphic work (Kaiser et al. 2006; Brezinski et al. 2010). The unfossiliferous formations of varying height above many important end-Famennian vertebrate Lagerstätten, such as the Cleveland Shale in Ohio, the Chaffee Formation in Colorado, Red Hill in Pennsylvania, are Hangenberg event sediments laid down in <100 000 years (perhaps East Greenland as well: Theron 1993; Clack 2006; Brezinski et al. 2008, 2010; Myrow et al. 2011; J. E. A. Marshall, pers. comm. 2011). Many of these, particularly formations correlated with the ‘Hangenberg Sandstone’, are ice age-associated deposits. Glaciers at sea-level at both the poles and near the tropics left characteristic diamictites and mud flats, while significant sea-level fall drove the formation of deep river valleys directly atop older marine deposits (including the Cleveland Shale; Algeo et al. 2007; Brezinski et al. 2008, 2010; Isaacson et al. 2008). At other vertebrate localities such as Andreyevka-2 in Russia and the Witpoort Formation in South Africa, Devonian–Carboniferous unconformities begin at the level of the Hangenberg horizon (Theron 1993; Alekseev et al. 1994, 1996) and it is likely that deposition was halted by marine regression (Alekseev et al. 1994; Hallam and Wignall 1999; Algeo et al. 2007). Global correlation of the Hangenberg horizon, and therefore the Devonian–Carboniferous boundary, is an ongoing project (Kaiser et al. 2008; Brezinski et al. 2009, 2010) which could reveal further examples of vertebrate faunal hiatus or sudden taxonomic turnover in even conformable deposits.
Abrupt vertebrate faunal changes were noted from around the Devonian–Carboniferous before recent stratigraphic work. Thomson (1976) pointed to a reduction from the numerous finned sarcopterygian genera at late Famennian sites to just a few megalichthyids and rhizodonts in the entire early Carboniferous (Mississippian; 359–318 Ma). Janvier (1977) remarked on the sudden end-Devonian extinction of large arthrodire placoderms, which he suggested were equivalent in scope to the loss of marine reptiles at the end-Cretaceous. This loss was linked with the Hangenberg regression (Algeo et al. 2007), as well as the same vague ‘internal and external’ factors used to explain the extinction of dinosaurs prior to the discovery of the Cretaceous–Palaeogene (K–Pg) iridium anomaly. In contrast, the coincident disappearance of durophagous ptyctodont placoderms was attributed to a shift of habitat into the deep sea (a habitat poorly represented in the fossil record), based on their mistaken placement as the ancestors of living bathypelagic holocephalans (and neglecting the diversity of holocephalans in shallow marine Carboniferous habitats; Janvier 1977; Sallan and Coates 2010; Sallan et al. 2011).
The proposed arthrodire extinction was later expanded to encompass all placoderms. Long’s (1993) family-level diversity data showed that placoderms in all habitats faced extinction over the end-Devonian or shortly thereafter. Hallam and Wignall (1999) were the first to directly attribute placoderm extinction to the Hangenberg event, even positing that the extinction enabled reported increases in actinopterygian diversity in all Carboniferous marine settings. Long (1993) also suggested an end-Devonian extinction of ischnacanthids among acanthodians, and porolepiforms, onychodonts, a few osteolepidid families, and marine lungfishes among sarcopterygians (Long 1993). The Devonian–Carboniferous extirpation of lungfishes from marine environments has been noted by a number of authors, but most discussions have failed to link it to any abiotic event (e.g. Campbell and Barwick 1986; Lloyd et al. 2012a). This is despite evidence for coincident and sudden changes in the habitat ranges and environmental associations of gyracanthids, rhizodonts, elasmobranchs (Figs 5 and 6) and even eurypterids (Lamsdell and Braddy 2010). The large number of such shifts might be a mark of random refugia in different environments.
Looking to direct stratigraphic and faunal evidence, Alekseev and colleagues (1994) dated the Andreyevka-2 fauna at Tula, Russia as latest Famennian, noting the compositional similarity of the tetrapod-bearing assemblage to similarly aged sites elsewhere. This is followed by the previously mentioned, likely Hangenberg-related, hiatus in sediment deposition, which is found elsewhere in Russia (Alekseev et al. 1994, 1996; Algeo et al. 2007). In contrast to the Famennian fauna, the earliest Carboniferous assemblage in the Tula sections is similar to actinopterygian-dominated Russian and Scottish Lagerstätten from later in the Carboniferous. These sediments also contain chondrichthyan, megalichthyid and rhizodont sacropterygians remains, lacking any evidence of the Porolepiform sarcopterygians antiarch placoderms and tetrapods that had previously inhabited a similar environment at the same spot (Alekseev et al. 1994). Alekseev suggested that the early Carboniferous actinopterygian-chondrichthyan fauna had originated at distinct Devonian localities and passed through the boundary unscathed.
There is stratigraphic evidence of extinction and turnover within chondrichthyans and actinopterygians. A vertebrate fauna in latest Famennian red sandstones at Hakkari, Turkey contains actinopterygian microfossils and ‘ctenacanth’/‘cladodont’ chondrichthyan ichthyoliths, alongside groenlandaspid placoderms, onychodontid sarcopterygians and chirodipterid lungfishes (Janvier et al. 1984). However, the only fossil known from the overlying Early Carboniferous formation is a cochliodont chondrichthyan tooth, appearing above an unfossiliferous shale that might be Hangenberg-related (Janvier et al. 1984). In Ohio, the diverse chondrichthyan fauna of the Cleveland Shale is largely absent from both the overlying Hangenberg-associated deposits and the early Tournaisian (359–345 Ma) Sunbury black shale (Hansen 1996; Brezinski et al. 2008, 2010). The latter has similar taphonomy to the Famennian formations, but the only chondrichthyan material is preliminarily assigned to Stethacanthus, belonging to a genus and clade (Symmoriiformes; figs 1, 6) common in Carboniferous Euramerica. Similarly, Randon and colleagues (2007) noted the Hangenberg-related loss of a phoebodont-jalodont chondrichthyan-dominated ichthyolith fauna in the Italian Carnic Alps, and its replacement by a taxonomically depauperate symmoriiform assemblage in conformable Tournaisian sediments. The Famennian fauna was similar to those described elsewhere and also contained actinopterygian and placoderm material. Ginter (1990) hypothesized a global Hangenberg-associated extinction of phoebodont chondrichthyans based on such samples (Ginter and Ivanov 1995). However, because there is little direct association between phoebodont ichthyoliths and macrofossil material, it is not clear what these patterns might mean in an ecological sense.
Until recently, these studies represented the full extent of investigation into vertebrate faunal change over the Hangenberg event, representing a sparse and uneven sample relative to the number of vertebrate-bearing deposits from the Famennian and Tournaisian. Sallan and Coates (2010) documented the global extent and depth of faunal turnover at the Hangenberg by using both global jawed vertebrate diversity data and analyses of faunal composition at well-sampled localities from both sides of the horizon. While diversity curves indicated a 32 per cent reduction in generic diversity over the Famennian–Tournaisian, faunal analyses showed the complete loss of 44 per cent of major gnathostome clades immediately around the Hangenberg horizon (Sallan and Coates 2010) (Fig. 6). There appears to have been little selectivity and few refugia: all Devonian and Carboniferous faunas were significantly different regardless of inferred environment (continental or marine) (Figs 5 and 6). The victims included sarcopterygians, acanthodians and placoderms found in different habitats and exhibiting divergent ecomorphologies (Sallan and Coates 2010; Anderson et al. 2011) (Fig. 6). Tournaisian and later Carboniferous sites were found to contain a highly homogeneous and temporally distinctive assemblages dominated by actinopterygians and chondrichthyans, with fewer lungfishes, rhizodonts, megalichthyids, coelacanths, acanthodid and gyracanthid acanthodians, and scant tetrapod material (Coates and Clack 1995; Coates et al. 2008; Sallan and Coates 2010) (Fig. 6).
The Hangenberg event seems to have restructured vertebrate ecosystems worldwide and marked a bottleneck in the evolution of modern clades (Figs 5 and 6). Diversity curves show large increases in actinopterygian and chondrichthyan richness starting in the Tournaisian, followed by a postbottleneck diversification of tetrapods (Coates et al. 2008; Sallan and Coates 2010; Sallan et al. 2011). These three groups have dominated all subsequent vertebrate faunas (Nelson 2006; Sallan and Coates 2010) (Figs 5 and 6). Romer’s gap might therefore be interpreted as the post-Hangenberg recovery interval, based on the existence of a depauperate homogeneous global gnathostome fauna (Sallan and Coates 2010). This is also supported by the relative rarity of both macrofossil sites and tetrapod material, particularly during the first 10 million years post-extinction, despite concerted searches (e.g. Brazeau 2005; Garvey and Turner 2006; Smithson et al. 2012) and the continued availability of suitable depositional environments (Hansen 1996; McGowan and Smith 2008). At the moment, diversity patterns for all three major post-Devonian divisions, tetrapods, actinopterygians, and chondrichthyans, are suggestive of a general recovery curve (Sallan and Coates 2010; Sallan et al. 2011): little new diversity, significant morphological differentiation and/or abundance at both ichthyolith and macrofossil levels in the early Tournaisian, a small number of superficially similar but ecologically differentiated lineages at faunas in the late Tournaisian (e.g. the Foulden fauna of Scotland and the Horton Bluff fauna of Canada; Brazeau 2005; Sallan and Coates 2010; Smithson et al. 2012), and finally a peak in the appearance of novel taxa, body forms and niches by the middle to late Visean (e.g. the various Visean tetrapod and fish sites of the United Kingdom, such as Glencartholm and East Kirkton; Dineley and Metcalf 1999).
The Tournaisian marks the beginning of a major round of morphological diversification in both actinopterygians and chondrichthyans, featuring the appearance of novel body shapes, jaw mechanics and dentitions (particularly durophagous and benthic types: Sallan and Coates 2010; Sallan et al. 2011; Finarelli and Coates 2012; Sallan and Friedman 2012) (Fig. 4). This suggests the existence of new opportunities facilitated by the loss of incumbent Devonian vertebrates at all ecological levels (Fig. 6). It is not clear why other survivors, such as acanthodians and tetrapodomorph fishes (rhizodonts and megalichthyids), did not diversify into new ecospace (Janvier 1996; Sallan and Coates 2010). Rather, a handful of genera persisted for several million years before disappearing, illustrating the postextinction phenomenon of ‘Dead Clade Walking’ (Jablonski 2002).
Late Palaeozoic to late Mesozoic events
The late Palaeozoic to late Mesozoic interval is punctuated by two mass extinctions (the Permian–Triassic (251 Ma) and Triassic–Jurassic (200 Ma); Raup and Sepkoski 1982) and a purported reorganization of marine ecosystems (Vermeij 1977), coupled with environmental perturbations ranging from ocean anoxic events (Toarcian (183–176 Ma), early Aptian (125–112 Ma), and Cenomanian–Turonian (100–89 Ma): Schlanger and Jenkyns 1976; Jenkyns 1988) to rapid changes in global temperatures (Price 1999; Jenkyns 2003). However, the response of fishes to these major climatic and biotic shifts is unclear. Here, we review outstanding patterns that are presently apparent in fish diversity during this lengthy interval and suggest whether these features are likely real or merely artefactual.
The late Mississippian and Permian richness trough: genuine or artefactual?
Marine invertebrate palaeontologists have proposed a bioevent equivalent to the Hangenberg and ecologically more severe than the end-Ordovician at the Visean–Serpukhovian boundary (328 Ma), coincident with the onset of the Late Palaeozoic Ice Age (Bambach 2006; McGhee et al. 2012). However, Sallan and Coates (2010) found jawed fish faunal composition to be stable in the later Mississippian, perhaps in line with the onset of stasis in vertebrate morphological and taxonomic diversity (Sallan et al. 2011; Sallan and Friedman 2012). There is a significant reduction in marine ichthyolith sampling and the number of macrofossil localities beginning at the Serpukhovian (328–318 Ma) and persisting into the later Palaeozoic (Sepkoski 2002; Sallan et al. 2011). However, the Mississippian–Permian trough found in many marine generic diversity curves might be related to a general lack of appropriate marine deposits in Europe (Sepukhovian; McGowan and Smith 2008; McGhee et al. 2012). The absence of fossiliferous marine facies was likely driven by both the destruction of Euramerican coastlines during the formation of Pangaea and regression-driven hiatuses during long glacial cycles (which last up to 150 000 years in the Pleistocene; Falcon-Lang et al. 2009; McGhee et al. 2012). There is also a lack of nonmarine Permo-Carboniferous (c. 318–251 Ma) deposits relative to the temporal span: most productive fish localities represent coal swamps and similar nonmarine habitats which were only widespread during short interglacials (lasting at most 20 000 years as in the Pleistocene) and cyclically replaced by seasonally or perennially dry environments (Falcon-Lang and DiMichele 2010).
Therefore, the late Palaeozoic depletion apparent in Sepkoski’s genus-level data might represent a sampling artefact. However, it does not occur in all data. Serpukhovian vertebrate taxon counts are boosted by the Lagerstätten at Bear Gulch, USA and Bearsden, Scotland, which exhibit a higher taxonomic and morphological diversity of chondrichthyans than might assumed from ichthyolith sites (Sepkoski 2002; Sallan and Coates 2010; Sallan et al. 2011) (Fig. 4). Blieck (2011) showed that Pennsylvanian vertebrate familial diversity rises above Mississippian levels in data from The Fossil Record 2, perhaps driven by the sporadic appearance of members of certain families better known from other Palaeozoic and Mesozoic intervals.
It is also possible that the Permo-Carboniferous trough is driven by the existence of stable cosmopolitan marine and freshwater faunas around Pangaea, rather than abiotic events or rock volumes. Permo-Carboniferous actinopterygian genera, such as Elonichthys, Platysomus, Aeduella, Amblypterus, Acrolepis and Palaeoniscum, seem especially widespread and long-lived, appearing at multiple disparate localities and persisting for tens of millions of years. Likewise, similar chondrichthyan teeth are known from around the Permo-Carboniferous world, with a few ranging through the entire interval (e.g. some petalodonts; Sepkoski 2002). Many of these apparently stable late Palaeozoic taxa have not undergone critical taxonomic revision in decades, with their longevity perhaps more indicative of neglect rather than true stratigraphic duration. However, their ecomorphologies do appear static (L. C. S., pers. obs.) and the same pattern occurs in marine invertebrate data, with Permo-Carboniferous benthic assemblages being particularly stable and conserved over large temporal and geographical scales (Schram 1979; Bonuso and Bottjer 2006). Bonelli and Patzkowsky (2008) noted highly homogeneous faunal composition at a single Serpukhovian site, with similar taxa appearing at all depths along a coastal gradient. In summary, reduced richness of fishes during the Permo-Carboniferous appears attributable to a lack of ecological opportunities and available habitat leading to low origination, perhaps exacerbated by poor sampling (McGowan and Smith 2008; Falcon-Lang and DiMichele 2010; McGhee et al. 2012).
The end-Permian event and Triassic diversification
Although classically regarded as the most catastrophic of Phanerozoic mass extinctions (Raup and Sepkoski 1982; Erwin 1990), the effects of the Permian-Triassic event on fishes remain deeply ambiguous (Janvier 1996; Blieck 2011). This reflects, we suspect, the relatively poor understanding of late Permian fishes (particularly actinopterygians; see discussion in Hurley et al. 2007), combined with a failure to adequately quantify patterns derived from available fossil data. Recent discussion in the literature touching on the end-Permian event as it relates to fishes is generally couched in terms of extinction recovery (Tong et al. 2006; Brinkmann et al. 2010; Hu et al. 2011; Wen et al. 2012). This is largely because Triassic (251–200 Ma) fishes are abundant and well-studied in comparison with the sparse record available for the Permian. The obvious problem is that such a research programme presupposes that patterns apparent among fishes represent recovery from extinction, without having first clearly demonstrated that fishes themselves were affected in such a way that it might be necessary to invoke recovery. We emphasize this point here because palaeontologists studying both elasmobranchs (Mutter et al. 2007; Mutter and Neuman 2008, Mutter 2009) and osteichthyans (Schaeffer 1973), as well as total fish diversity (Janvier 1996; Blieck 2011), have found little evidence for major taxonomic shifts associated with the end-Permian event, and no striking decline in richness is apparent in broader surveys (Fig. 1). Marine conodonts, which have best global record of any vertebrate group, were apparently little affected by the Permo-Triassic apart from a dip in abundance (Clark 1983; Clark et al. 1986; De Renzi et al. 1996). Indeed, the familial data within the Fossil Record 2 show consistent levels of fish diversity through the Permian and into the early Triassic, leading Blieck (2011) to conclude that there was no real loss. Sepkoski’s (2002) data give no indication of elevated extinction in the late Permian (Fig. 2), and his osteichthyan data are so scant that rates cannot be estimated for this major group (Fig. 3). A literal reading of Sepkoski’s data would suggest that the earliest Triassic (Induan; 251–249.5 Ma) was an interval of intensive turnover among fishes (Figs 1–3), perhaps indicative of turbulent postextinction recovery. Here too, small taxonomic samples urge caution: Sepkoski (2002) records only six chondrichthyan and 13 osteichthyan genera from this interval. In the face of such evidence, we question claims that ‘fishes were severely affected by the end-Permian mass extinction’ (Wen et al. in press), at least in aggregate.
However, at least one important segment of Permo-Carboniferous fish diversity is missing from Triassic localities: nonchimaeroid holocephalans (Cappetta et al. 1993; Stahl 1999; Sepkoski 2002). The durophagous holocephalans became particularly diverse in the aftermath of the Hangenberg extinction (Figs 5 and 6), and many assemblages, including petalodonts, cochliodonts, deltoptychiids, pristiodonts, menaspids, and helodonts, persist through the late Palaeozoic (Stahl 1999; Sepkoski 2002; Sallan et al. 2011). The majority of these Palaeozoic holocephalan groups have last appearances in the late Permian. Indeed, petalodont ichthyoliths and even body fossils (e.g. Janassa) were widespread in end-Permian formations from Japan to Germany to Arizona (Goto 1994; Stahl 1999). Many of these holocephalan taxa are effectively repositories for morphologically similar teeth arising in different lineages (Smith and Patterson 1988; Sallan et al. 2011; Finarelli and Coates 2012). However, most represent clades at various levels and may serve as a proxy for ecological or functional diversity (Stahl 1999; Sallan et al. 2011).
In reconciling these losses with the lack of extinction signal for other fishes, it is notable that the vast majority, if not all, of Palaeozoic holocephalan material is marine (Stahl 1999) (Fig. 4). These groups co-existed with, and fed upon, known invertebrate victims of the extinction (Moy-Thomas and Miles 1971; Stahl 1999; Sallan et al. 2011). In contrast, the majority of unaffected elasmobranch and osteichthyan lineages were either euryhaline (e.g. hybodonts) or completely freshwater (e.g. xenacanths) in ecology (contraSchultze 2009) and are well represented in the coal measures, deltas and shallow basins of Palaeozoic–Mesozoic Euramerica (Hook and Baird 1988; Dick 1998; Zajic 2000; Kriwet et al. 2008) (Fig. 6). As exceptions that might prove the rule, several deep-bodied, durophagous actinopterygian lineages widespread in Permo-Carboniferous sediments (e.g. Eurynotus, Amphicentrum; Moy-Thomas and Miles 1971; Dineley and Metcalf 1999), suddenly disappear at or before the end-Permian (Gardiner 1993a). An additional datum supports marine, but not durophage, selectivity: the brush-backed symmoriiform chondrichthyans (stethacanthids) (Fig. 1), nektonic predators once widely dispersed in Palaeozoic seas (Figs 1 and 6), make their last appearance in the late Permian despite high diversity in earlier assemblages (Goto 1994; Ivanov 2005).
It is therefore possible there was a selective loss of marine and/or durophagous fishes between the Permian and Triassic. Whether these extinctions were centred on the end-Permian event or occurred more gradually is another issue. Mapes and Benstock (1988) noted a decline in holocephalan lineages over the later Permian, and Ivanov (2005) suggested that chondrichthyan tooth assemblages from that interval are less diverse than in the Carboniferous. Indeed, compendia show that the last appearances of some holocephalan lineages, such as cochliodonts, deltoptychiids and chondrenchelyiforms, occurred in the early Permian or even the Late Carboniferous (Cappetta et al. 1993; Stahl 1999; Sepkoski 2002).
These last appearances could represent random extinctions expected over any geological interval. After all, other fishes sometimes counted among the victims of the end-Permian event – acanthodians and megaliththyid sarcopterygians – represented depauperate lineages that disappeared in geographically stepwise fashion long before the boundary (Janvier 1996; Mutter and Richter 2007; Sallan and Coates 2010). Alternatively, holocephalan losses before the end-Permian could be due to the Signor-Lipps effect: backsmearing of extinction because of incomplete sampling of the record (Signor and Lipps 1982; Foote 2007), perhaps magnified by the poor condition of the late Permian fish record. Indeed, articulated marine fishes from that interval derive almost exclusively from the Marl Slate and Kupferschiefer of the northern European Zechstein Sea (Dineley and Metcalf 1999; Diedrich 2009) with a few scattered examples from further afield (China: Liu and Wei 1988; Greenland: Stemmerik et al. 2001), and these have been little studied. However, it is clear even from this scant record that many actinopterygian genera typically associated with Triassic age assemblages (Saurichthys: Liu and Wei 1988; Bobasatrania: Stemmerik et al. 2001) have ranges extending into the late Permian.
By contrast, the Triassic is marked by a series of well-documented and widely distributed marine Lagerstätten in Austria, Canada, China, Greenland, Italy, Madagascar, Spitsbergen, and Switzerland, with more minor assemblages in the United States (Tintori 1998; Brinkmann et al. 2010; Hu et al. 2011; Mutter 2011). Such Triassic deposits have been particularly important for calibrating the radiation of neopterygians, the clade that includes nearly all modern ray-finned fish diversity (Nelson 2006). In addition, the early Triassic also seems to record the onset of the neoselachian elasmobranch radiation (Underwood 2006), which includes all living chondrichthyans apart from holocephalans.
Both the neopterygian total group and crown appear to extend well into the Palaeozoic (Hurley et al. 2007), but crown representatives are minor components of faunas of this age, in terms of both richness and abundance (Diedrich 2009). Holosteans (bowfin, gars and their relatives) remain rare and teleosts are absent until the Late Triassic, at which point some localities record a great diversity of trophically divergent crown neopterygians. Particularly, striking is the proliferation of durophagous taxa both among total group teleosts (Pycnodontiformes) and holosteans (Macrosemiidae, Seminotidae) along with incertae sedis neopterygian groups (Dapediidae), occurring subsequent to the last appearances of so many Palaeozoic durophages (Tintori 1998).
In summary, although the magnitude of the end-Permian event for fishes is unclear, there is a possibility of selective extinction of marine (and particularly durophagous) fishes. In contrast, euryhaline elasmobranch and actinopterygian lineages, and marine conodonts, seem to have been largely unaffected, with multiple lineages present in devastated marine ecosystems in the Early Triassic. Just as many groups underwent habitat shifts post-Hangenberg, surviving fish clades might have moved into new marine environments, even though they would have found few resources in the immediate aftermath. Indeed, elasmobranch ichthyoliths from some Early Triassic localities are said exhibit a ‘Lilliput effect’, a reduction in size often noted after mass extinction and associated with poor productivity (Twitchett 2001; Chen et al. 2007; Mutter 2009). However, the significance of this reduction is unclear, because ichthyoliths from a single taxon can be variable, the size of pre-event ancestors is not always known, and oversampling of stressed environments in recovery intervals can affect recorded averages (McGowan et al. 2009). Holocephalans do not reappear in the record until the later Triassic, when they largely exhibit a form very similar to living chimaeroids (Stahl 1999). Mesozoic holocephalans did achieve some degree of disparity, ranging from dorsoventrally flattened skate-like forms in the Early Jurassic (Stahl 1999) and giant, but anatomically more conventional, species in the Late Cretaceous (Cicimurri et al. 2008). However, they never completely refilled lost morphospace or recovered to their Palaeozoic abundance, at least outside of the deep-sea environments that Recent forms inhabit and for which Mesozoic records are poor (Stahl 1999). The apparent diversity patterns discussed previously are therefore suggestive of selective extinction and replacement of fishes at the Permo-Triassic. These are in need of detailed investigation.
Middle and late Mesozoic diversity patterns
The middle Mesozoic commenced in the aftermath of the Triassic–Jurassic extinction, but the effect of that event upon fishes seems minimal (McCune and Schaeffer 1986; Bambach et al. 2004). Not a single osteichthyan family listed in The Fossil Record 2 by Gardiner (1993b) or Patterson (1993a) fails to cross the boundary (Hallam 2002). However, there is one potentially major exception: conodonts (Fig. 2). While Triassic conodont diversity paled in comparison with their Ordovician and Devonian–Missisissippian peaks (Fig. 2), conodont origination preceding the Triassic-Jurassic boundary was greater than in any interval because the Mississippian and their biostratigraphic utility was undiminished until the end of the period (Clark 1983; De Renzi et al. 1996).
Conodonts seem to undergo a long-term decline in the second half of the Triassic (Fig. 2) (Clark 1983; De Renzi et al. 1996; Sepkoski 2002). During this same interval, conodont elements became more ‘generalized’ or homogeneous in form, waned in abundance, were reduced to a single family and were extirpated from faunal regions in stepwise fashion (Trammer 1974; Clark 1983; Hallam 2002; Tanner et al. 2004). The latter trends could be reflective of a Signor-Lipps effect driven by facies change near the extinction boundary (Signor and Lipps 1982; Clark 1983; De Renzi et al. 1996; Tanner et al. 2004). However, the selective disappearance of conodonts by the end-Triassic is not widely attributed to an abiotic marine event, but rather to long-term loss of appropriate environments and greater provincialism because of the break-up of Pangaea (De Renzi et al. 1996). Losses have also been credited to an unfortunate run of random extinction, after-effects of the Permo–Triassic, competition with the modern evolutionary fauna, and even anecdotal scenarios invoking predation by ichthyosaurs (Clark 1983; De Renzi et al. 1996).
It is difficult to favour any single or combination of options because relevant data are largely unavailable, including that for the diversity of other fishes with similar environmental preferences and ecologies. However, the Triassic–Jurassic interval is generally defined by such gradual taxonomic and ecological turnover (Hallam 2002). In fact, the event has been termed a mass depletion without a single cause, akin to the Frasnian–Famennian (Bambach et al. 2004). On that note, Kriwet and colleagues (2009) have suggested that the Early Jurassic radiation of neoselachians reflects opportunistic radiation into ecological roles cleared by extinction, whether sudden or gradual, a repeated pattern by this point in the record.
The Jurassic–Cretaceous interval (200–65.5 Ma) is punctuated by several ocean anoxic events (OAEs; Schlanger and Jenkyns 1976; Jenkyns 1988), but there has been no suggestion that any of these resulted in either major extinction or faunal turnover among fishes. Indeed, some OAEs are flanked by well-studied faunas (e.g. the Hettangian–Sinemurian of Lyme Regis, England and the Toarcian of Yorkshire, England, Germany, and France in the case of the early Toarcian OAE; Wenz 1967; Hauff and Hauff 1981; Dineley and Metcalf 1999), and some even occur within continuous units that have been historically well-sampled for fishes (e.g. the English Chalk in the case of the Cenomanian–Turonian OAE; Dineley and Metcalf 1999).
There are two clear peaks in raw richness trajectories for Mesozoic fishes: one at the close of the Jurassic and another in the early Late Cretaceous (100–65.5 Ma). Both of these correspond to intervals with well-studied exceptional deposits, yielding numerous articulated actinopterygians (Late Jurassic: Cerin, France, and Solnhofen, Germany; early Late Cretaceous: Djebel Tselfat, Morocco, English Chalk, UK, Hakel, Hadjoula, and Namoura, Lebanon; Lambers 1999; Forey et al. 2003). Phylogeny-based techniques indicate that the Late Jurassic peak is probably a preservational artefact, at least for actinopterygians (Cavin 2010). However, the same phylogenetic test suggests that a genuine radiation among actinopterygians occurred during the Cenomanian (100–94 Ma; Cavin and Forey 2007), and Cavin et al. (2007) hypothesized such a pattern might have been driven by high sea surface temperatures.
The end-Cretaceous extinction and Palaeogene recovery
The end-Cretaceous event (65.5 Ma) is perhaps the most thoroughly studied and best understood of the canonical Big Five Phanerozoic extinctions, due in large part to its relative recency combined with the charismatic nature of its most conspicuous victims. The literature concerning the mechanism of extinction is extensive, and we refer readers to the recent review of the prevailing bolide impact hypothesis presented by Schulte et al. (2010) and responses to it by advocates of more nuanced drivers (Archibald et al. 2010; Courtillot and Fluteau 2010; Keller et al. 2010).
In contrast to older extinction events, patterns of turnover in fishes centred on the K–Pg boundary have been the focus of numerous studies in both marine and continental environments, at scales ranging from individual sections to global surveys. Most work has focused on either marine or freshwater taxa, but the handful of analyses addressing both (e.g. Cavin and Martin 1995; Cavin 2001a) have inferred contrasting patterns in these two environmental settings. We have divided our discussion of the K–Pg extinction to reflect this dichotomy.
Patterns in continental settings
The picture of turnover in freshwater fishes is based largely on regional patterns, with a particular emphasis on sections of the Maastrichtian (71–65.5 Ma) Hell Creek Formation and overlying Danian (65.5–61 Ma) Tullock Formation in eastern Montana (Bryant 1989; Sheehan and Fastovsky 1992, 1993; Archibald 1993, 1996). Late Cretaceous to Palaeocene (84–56 Ma) continental deposits in western North America generally represent high-energy palaeoenvironments, with the consequence that most fish fossils consist of isolated, often badly abraded, bones, teeth, spines and scales recovered by screen washing (Grande and Grande 1999). This taphonomic filter generally prevents precise identification of many specimens, particularly of teleosts, although more detailed anatomical analysis has demonstrated that finer taxonomic assignment is possible (Wilson et al. 1992; Brinkman and Neuman 2002; Neuman and Brinkman 2005; Newbrey et al. 2009). Previous studies have therefore been conducted at the genus level, with probable overrepresentation by nonteleostean taxa characterized by distinctive and easily identifiable microremains (chondrichthyans, chondrosteans, holosteans). There is compelling evidence that teleosts were more significant members of Late Cretaceous freshwater communities, both in terms of numbers of individuals and taxonomic richness, than current faunal lists tend to indicate (Neuman and Brinkman 2005; Redman et al. 2010); consequently, patterns of fish extinction reconstructed from K–Pg sections in the Western Interior must be considered with this important caveat in mind.
Numerous studies have examined the distribution fossil fishes above and below the K–Pg boundary (Archibald and Bryant 1990; Sheehan and Fastovsky 1992, 1993; Archibald 1993, 1996), but all share a common empirical foundation: a vertebrate occurrence database initiated by Bryant (1989) that draws on more than 150 000 individual specimens. She found that 9 of the 18 genus-level fish lineages represented in the latest Maastrichtian make their last local appearance in this interval, with eight found in deposits immediately postdating the extinction event (Bryant 1989, table 1). One additional taxon, Belonostomus, occurs in still younger deposits, but it has been argued that the single specimen upon which this range extension is based was reworked from Cretaceous sediments (Cavin 2001a). Similar patterns are reported from updated versions of this compendium (Archibald and Bryant 1990; Archibald 1996). Although limited in terms of taxonomic richness, these data have led to contrasting interpretations of extinction dynamics. Bryant (1989) parsed her fish data into divisions based on assumptions about environmental preference and salinity tolerance, noting the highest rate of extinction among the suite of putative marine-to-brackish taxa and higher survivorship in those sets of taxa inferred to have stronger ties to strictly freshwater environments. She argued that this pattern reflected the local unavailability of suitable environments for marine fishes during the Palaeocene, associated with the regression of the Western Interior Seaway, rather than any particular consequence of a global extinction, a conclusion subsequently echoed by Archibald and Bryant (1990) and Archibald (1996). Pruning these marine forms along with rare taxa, Sheehan and Fastovsky (1992, 1993) argued for higher rates of survival among freshwater taxa than terrestrial vertebrates, a pattern they attributed to the buffering of detritus-based fluvial ecosystems against collapses in primary productivity associated with postimpact reductions in insolation. Archibald (1993) questioned this analysis and its conclusions, but despite conflicting opinions, it seems clear that the best-characterized freshwater fishes from the Western Interior appear to have been little affected by global K–Pg events.
To the extent that it can be compared with microremains, the sparse articulated body fossil record of Late Cretaceous freshwater fishes in western North America provides a complementary picture of negligible turnover associated with the K–Pg extinction. With the possible exception of the incertae sedis rhinobatoid elasmobranch (guitarfish) Myledaphus (Langston 1970; Neuman and Brinkman 2005; see qualifications about possible Eocene occurrences in Cappetta 1987b) and indeterminate holosteans (Brinkman 1990), all of the fish families known from articulated remains in freshwater Campanian and Maastrichtian deposits of western North America (all are actinopterygians; Acipenseridae (sturgeons): Hilton and Grande 2006; Polyodontidae (paddlefishes): Grande and Bemis 1991; Lepisosteidae (gars): Grande 2010; Amiidae (bowfins): Grande and Bemis 1998; Osteoglossidae (goldeyes): Li 1996; Megalopidae (tarpons): Bardack 1970; Gonorynchidae (beaked sandfishes): Grande and Grande 1999; Sorbinichthyidae (armoured herrings): Newbrey et al. 2010) survive the end-Cretaceous extinction (e.g. Grande 1980), with some individual genera persisting regionally to the Eocene or later (the lepisosteid Lepisosteus, the amiid Cyclurus, and the gonorynchid Notogoneus; Grande and Bemis 1998; Grande and Grande 1999; Grande 2010). These rare body fossils provide a particularly important window into teleost diversity that has historically gone unrecognized in screen-washed samples and suggest that this important component of the freshwater fish fauna did not undergo radical changes associated with the K–Pg boundary in North America.
The Bolivian faunas, although potentially less rich, offer the advantage of having been subjected to comprehensive palaeoichthyological review yielding reliable taxonomic identifications (Gayet and Meunier 1998; Gayet et al. 2000). Of the 23 mostly species-level lineages reported on the basis of fragmentary remains from the Maastrichtian of this sequence, 13 are also present in the overlying Palaeocene deposits (Gayet and Meunier 1998, table 1; Gayet et al. 2000, table 3). Taxa which fail to cross the boundary in this sequence are predominantly chondrichthyans (6 of 10), with the others belonging to actinopterygian clades typically associated with marine environments (enchodontids, pycnodonts). None of these taxa occur in the uppermost Maastrichtian section of the El Molino Formation, suggesting that, like North American continental boundary sections, apparent patterns of turnover at this scale are likely attributable to changes in depositional environment rather than any global biodiversity crisis.
Our understanding of K–Pg extinction among freshwater fishes at a global level is limited to taxonomically coarse surveys, but reveal patterns congruent with those apparent at finer scales of taxonomic and geographic resolution. Cavin and Martin (1995) and Cavin (2001a) have argued that freshwater taxa were less affected by the K–Pg event than fishes living in other environments, based on high survivorship of families containing taxa believed to inhabit brackish or freshwater settings. Cavin (2001a, fig 3) in particular interpreted this pattern as further support for Sheehan and Fastovsky’s (1992) hypothesis that detritus-fed continental aquatic environments were buffered to a greater extent than marine ecosystems from the drivers of the K–Pg extinction.
Patterns in marine environments
Patterns of turnover among marine fishes at the end of the Cretaceous have been addressed at scales ranging from individual sections and regions (Gallagher et al. 1986; Gallagher 1991, 1993, 2002, 2003; Noubhani and Cappetta 1997; Zinsmeister 1998) to global analyses (Cavin and Martin 1995; MacLeod et al. 1997; Cavin 2001a; Kriwet and Benton 2004; Friedman 2009, 2010). Reflecting classical divisions of research emphasis, along with profoundly different taphonomic profiles, chondrichthyans and osteichthyans have generally been treated separately outside of broad reviews that explore coarse family-level data (e.g. MacLeod et al. 1997). At this scale, patterns between the two major divisions of fish diversity appear similar. Estimates for extinction rates among actinopterygians at the family level vary between 10 (MacLeod et al. 1997) and 19 per cent (Cavin 2001a), with the larger of these two values probably representing a more accurate value. Family-level extinction rates for neoselachians are of a similar magnitude, with Kriwet and Benton (2004) reporting 17 per cent of families known from the Maastrichtian making their final appearance in that stage. For actinopterygians, raw figures for the Maastrichtian compare closely with the percentage of families making their last appearances in the preceding Campanian (12 per cent; Cavin 2001a), which has been used to argue that the K–Pg extinction was a nonevent among fishes (MacLeod et al. 1997). However, the K–Pg is perhaps the only clear example in Sepkoski’s fish data where a recognized mass extinction horizon is preceded by elevated rates of extinction followed by a possible recovery phase characterized by increased rates of origination (Figs 2–4).
Kriwet and Benton (2004) provided the most comprehensive overview of extinction and recovery patterns within neoselachian elasmobranchs on a global scale. A literal reading of their diversity curves (Kriwet and Benton 2004, fig 1) suggests a Maastrichtian peak in richness, followed by a steep decline in the Danian and Selandian, with recovery to near-Maastrichtian levels by the Thanetian (59–56 Ma). Eighty-four per cent of Maastrichtian species make their last appearance in that stage, which translates losses of 56 and 17 per cent and the genus and family level, respectively. Kriwet and Benton (2004, fig 2) presented a qualitative argument for selectivity. A curious assemblage of pelagic apex predators, shallow-water carpet sharks and durophagous rays are said to have suffered disproportionately. These putative patterns of selectivity rest on verbal arguments rather than quantitative analysis, so we have built upon Kriwet and Benton’s (2004) work by testing for phylogenetic selectivity using the protocols outlined by Roy et al. (2009). Such a taxonomic approach, of course, does not put us in a position to comment on specific patterns of trophic selectivity; this would require quantification of ecologically relevant features that could be derived from fossil neoseleachians, which, in practice, means isolated teeth. It does, however, allow us to determine whether extinctions were clustered phylogenetically. If so, this might suggest selection against common aspects of ecology that characterized closely related taxa. This procedure rejects the directional null hypothesis that victims are not clustered phylogenetically (p = 0.032; Fig. 7), lending some quantitative support to Kriwet and Benton’s (2004) assertion of selective extinction among neoselachians.
There are also indications of selective extinction among actinopterygians at the K–Pg. From a phylogenetic perspective, comparisons drawn between actinopterygian faunas from the Late Cretaceous and early Palaeogene reveal some striking contrasts. Cretaceous assemblages, including those from the Maastrichtian (David 1946; Gallagher et al. 1986; Chalifa and Lewy 1992; Gallagher 1993, 2002, 2003; Khaddumi 2009; Friedman 2012), contain crown teleosts as well as an ecologically diverse array of groups that branch from the teleost stem: pycnodontiforms, pachycormiforms, aspidorhynchiforms, ichthyodectiforms and crossognathiforms (the final group has variously been assigned to the teleost stem and crown by different workers; Patterson and Rosen 1977; Cavin 2001b). With the exception of pycnodontiforms, which persist until the middle Eocene (56–34 Ma; Poyato-Ariza and Wenz 2002), none of these stem teleost clades are definitively known from post-Maastrichtian sediments. The pruning of radiations from teleost stem between the Cretaceous and Palaeogene is complemented by the taxonomic and morphological proliferation of some crown-group radiations in the Palaeocene and Eocene. Most prominent among these are the acanthomorph, or spiny-finned, teleosts, a group that today numbers nearly one of every three living vertebrate species (Stiassny et al. 2004; Friedman 2010; Sallan and Friedman 2012).
Many of the stem teleosts making their last appearance in the Maastrichtian represent some of the most conspicuous members of Late Cretaceous marine fish faunas (Fig. 7). These include a series of large-bodied taxa (>1 m) that appear to have had wide geographical distributions during the final stage of the Late Cretaceous: the pachycormids Protosphyraena and Bonnerichthys (Friedman et al. 2010; Friedman 2012), and the ichthyodectiforms Xiphactinus (Schwimmer et al. 1997), Saurodon (Taverne and Bronzi 1999), Saurocephalus (Bardack and Sprinkle 1969; Friedman 2012) and Gillicus (Tanimoto and Kikyo 2001). Large teleosts making their last appearance in the Maastricthian are not limited to members of the stem, with crown representatives including the pachyrhizodontid Pachyrhizodus (David 1946; Friedman 2012), possibly the tselfatiiform Pentanogmius (Chalifa and Lewy 1992; Gallagher 2003), and the aulopiforms Apateodus, Cimolichthys and Enchodus (David, 1946; Chalifa and Lewy 1992; Khaddumi 2009; Friedman 2012). Most of these genera have dense records leading up to the end-Cretaceous, with occurrences known from all stages between their first and last appearances (Friedman 2009). Because of their robust bones and teeth, these large taxa are generally the easiest to recognize on the basis of isolated remains, suggesting that their complete absence from Palaeocene deposits is genuine rather than artefactual. This pattern is not simply attributable to a reduction in Chalk facies after the K–Pg. Fossils of these teleost groups are found in a range of lithologies (including shales and marls: David 1946; Russell 1988) in the Late Cretaceous and are completely unknown from early Palaeocene (Danian) chalks (Davis 1890).
The concentration of last appearances of large-bodied marine teleost groups in the Maastrichtian was first highlighted by Cavin and Martin (1995) and Cavin (2001a). Along with the decimation (dercetids) or complete extirpation (aspidorhynchids) of clades containing smaller-bodied but piscivorous taxa, these authors inferred selection against pelagic predators, which they linked to collapse of primary productivity in the open ocean. Friedman (2009) provided a quantitative test of selectivity using body size and measurements of jaw mechanics, and found that marine teleost genera making their last appearance in the Maastrichtian represented a nonrandom sample of taxa known from that stage. This result was largely driven by the extinction of all taxa occupying a region of ecomorphospace characterized by large body size and low-to-moderate jaw-closing mechanical advantage (typical of piscivorous taxa; Wainwright and Richard 1995).
The inferred extinction of many large-bodied marine teleosts during the Maastrichtian is largely built upon the recognition of disarticulated or fragmentary remains of these taxa (Friedman 2012, but see Khaddumi 2009 for articulated examples from coeval deposits), and we presently lack a clear understanding of patterns of turnover at the smaller end of the size spectrum, where articulated remains are particularly important for determining taxonomic placement. A better understanding of evolutionary patterns among small taxa forms important historical context for exploring the origins of modern reef fishes, which are increasingly used as model systems in biomechanical and evolutionary studies (e.g. Wainwright and Bellwood 2002) and appear to trace their origin to the Late Cretaceous–Palaeogene interval (Friedman 2010).
The latest well-documented Cretaceous marine site that yields complete, small fish specimens in any abundance is the latest Campanian–earliest Maastrichtian Nardò locality in southern Italy (Taverne 2010 and references therein). Younger Cretaceous assemblages yielding articulated fishes are known (e.g. Khaddumi 2009), but these specimens have not yet been formally studied. With the exception of Chanos and Chanoides (Taverne 2005, 2009), the small-bodied fish genera found at Nardò are not found in geologically younger deposits, and many of these are assigned to monotypic families. The Nardò assemblage is particularly unusual in comparison with both pre-Maastrichtian and Cenozoic marine assemblages in its diversity of putative ostariophysian teleosts (minnows, carp, characins, catfishes, knifishes and their allies), which number nearly one quarter of the genera known from this site. Whether this represents a peculiarity of the Nardò fauna or instead reflects a more general feature of latest Cretaceous marine ichthyofaunas remains unclear.
Ecological and evolutionary recovery: building modern fish biodiversity
Some indication of the temporal scale for recovery of abundance (rather than richness) in pelagic fish communities in the aftermath of the K–Pg is provided by accumulation rates of teeth in marine cores from the South Atlantic (Shackleton et al. 1984; D’Hondt 2005). These show a steep drop centred about the K–Pg boundary, and only return to Maastrichtian levels some 4 million years into the Palaeocene. Such observations suggest a timescale for recovery to pre-extinction levels of productivity, but estimating the span required for evolutionary recovery is not as easily assessed and can only be discussed in qualitative terms at present.
Patterns of recovery on this scale are best illustrated by examples of putative replacement of extinct groups with new clades that appear to occupy comparable ecological roles. The loss of large-bodied, predatory taxa at or before the K–Pg represents a pattern common to both actinopterygians and chondrichthyans (Fig. 8). Similarly, both groups are marked by the Palaeocene–Eocene proliferation or appearance of large pelagic predators still found in today’s oceans: carcharinids (requiem sharks; Kriwet and Benton 2004), lamnids (makos; Kriwet and Benton 2004), and scombroids (tunas and their kin; Dartevelle and Casier 1949) in the early Palaeocene, carangids (jacks; Bannikov and Parin 1997; Bonde 1997) and xiphioids (billfishes; Bannikov and Parin 1997; Sytchevskaya and Prokofiev 2002; Friedman and Johnson 2005) in the late Palaeocene, and sphyraenids (barracudas; Casier 1966; Blot 1980) in the early Eocene. But not all large predators in early Palaeogene oceans are familiar inhabitants of modern seas. Most striking is a short-lived invasion of marine environments by osteoglossomorph teleosts (bony tongues) in the Palaeocene and early Eocene (Bannikov and Parin 1997; Bonde 2008; Forey and Hilton 2010). These taxa were widely distributed geographically, with remains reported from North America, Asia, Europe and Africa and appear to represent multiple osteoglossomorph clades. A marine proliferation of large, predatory osteoglossomorphs is particularly striking given that modern members of this radiation are tightly linked to freshwater environments (Hilton 2003). Compelling remains of osteoglossomorphs are unknown from marine deposits either before or after this Palaeocene to mid-Eocene window, and it seems likely that their brief incursion into the oceans during this time represented an opportunistic replacement of the predatory fish groups so strongly impacted at or before the K-Pg.
Other patterns of functional replacement associated with the K–Pg emphasize the need to examine extinction and subsequent recovery at a broad phylogenetic scale, extending across traditional taxonomic divisions of palaeontological research. The diversity of suspension-feeding chondrichthyans and whales in the Cenozoic (Sanderson and Wassersug 1993) has raised questions about the apparent absence of ecologically analogous vertebrates during most of the Mesozoic (Colin and Janis 1997). However, it is now clear that the role of large vertebrate suspension feeders was occupied by a subset of pachycormiform fishes until the Maastrichtian, with the familiar groups of living suspension feeders appearing in the Palaeogene (whale sharks and manta rays: Palaeocene; basking sharks: Eocene; baleen whales: Oligocene: Friedman et al. 2010).
Beyond these specific examples, a literal reading of any taxonomic databases documenting the Phanerozoic diversity patterns in fishes suggests a striking rise in richness in the Cenozoic (Figs 2–4), with much of this attributable to a sharp increase in the number of species of acanthomorph teleost (and especially the diverse acanthomorph subgroup Percomorpha: Patterson 1993a, b, 1994; Stiassny et al. 2004), which show a commensurate increase in morphological disparity (Friedman 2010; Sallan and Friedman in press). Ostariophysians, like percomorph acanthomorphs, are another exceptional radiation of living teleosts (Alfaro et al. 2009) whose fossil record points to a Cenozoic diversification (Patterson 1994). Unlike acanthomorphs, ostariophysians are predominantly a freshwater radiation, with the consequence that their fossil record is comparatively sparse. A complete picture of the evolution of modern fish diversity demands a better understanding of the palaeontological histories of these two major clades – Acanthomorpha and Ostariophysi – which collectively make up nearly half of all living vertebrate species (Stiassny et al. 2004; Alfaro et al. 2009).
Whether the striking increase in bony fish richness in the Cenozoic reflects genuine signal, pull of the Recent, an artefact of the sedimentary record, or some combination of the three remains unclear. If this pattern survives subsequent analyses that take into account these and other biasing factors, then the true ‘Age of Fishes’ is not the Devonian, but instead, and perhaps surprisingly, the Palaeogene and Neogene.
Conclusions and a roadmap for future research
1 So far as can be discerned from available data, many major marine invertebrate and terrestrial (e.g. tetrapod) extinction intervals either appear to have left much of fish diversity unscathed, or are so poorly constrained for fishes as to preclude any definitive statements. These include several of the Big Five mass extinctions, such as the Frasnian–Famennian, end-Permian and end-Triassic events. This suggests extinction patterns cannot be translated from one group to another, or different clades used as proxies for one another, even when they co-exist or interact. Morphological (and presumably ecological) diversification of fishes is common in the intervals following mass extinctions, whether or not fishes themselves appear to have been strongly impacted. These radiations usually involve movement into new environments and trophic roles, as well as ecological replacement of putative extinction victims.
2 At present, the extinctions whose impacts upon fishes are best-characterized are the end-Devonian (Hangenberg) and the end-Cretaceous (K–Pg) events. The Hangenberg extinction appears to have resulted in a phylogenetic bottleneck that radically restructured vertebrate faunas in both marine and continental environments. Selectivity appears to have been a minimal factor; taxa of every major vertebrate division, ecology and environment were lost. Chondrichthyans, actinopterygians and tetrapods radiated after this event, and these same clades continue to dominate vertebrate biodiversity today. The K–Pg extinction seems to have had strikingly different effects upon fishes inhabiting continental and marine ecosystems. Although data are scant, a series of regional studies targeting patterns of turnover among freshwater fishes across the K–Pg boundary show few, if any signs, of catastrophic extinction. By contrast, there is clear evidence for selective extinction among both marine chondrichthyans and actinopterygians. Large-bodied and predatory taxa appear to have been particularly hard hit in both clades, with new groups arising to fill vacated functional roles in the early Palaeogene.
3 The effects of geological megabiases on the fish fossil record remain unquantified, although our intuition is that many peaks and troughs in fish diversity curves are more related to sampling issues than genuine biological patterns. Assessing the strength of the relationship between the rock record and fish palaeobiodiversity will require new databases recording individual occurrences, rather than available compendia that archive only first and last appearances. Fish subgroups have different taphonomic profiles, preservational modes and fossilization rates (Foote and Sepkoski 1999), meaning that patterns recovered for one clade cannot be easily translated to fishes as a whole.
4 Fishes, particularly nonmarine forms, are poorly represented in current compendia of fossil diversity (Sepkoski’s genus database and the PBDB being the most important and accessible) and the global diversity curves derived from them. Consequently, fishes are also neglected in large-scale studies of mass extinction and reviews of palaeobiodiversity. There are many databases that have been assembled by fish workers to address specific palaeobiological problems, but these are variable in both taxonomic and stratigraphic scope and generally include only taxon ranges (rather than occurrences). A better understanding of patterns of fish diversity will require experts on various clades and geological intervals to produce a synoptic, standardized, occurrence-based dataset for fishes that will bring the empirical dataset for this group up to modern standards. These data should be of interest to palaeobiologists more generally, because fishes are an important component of biodiversity and have played a nontrivial role in influencing evolutionary processes and ecological histories of many other clades (Stanley 2008; Sallan et al. 2011; Bush and Bambach 2011).
Acknowledgements. We thank Richard Twitchett for the invitation to submit this review of fish diversity patterns stemming from a symposium on biotic crises in the fossil record held at the Palaeontological Association Annual Meeting in Plymouth in December 2011. Our understanding of extinction and recovery in fossil fishes has been enriched by conversations with many colleagues, but we would like to highlight the contributions of Bill Ausich, Dave Bapst, Kevin Boyce, Martin Brazeau, Lionel Cavin, Michael Coates, Michael Foote, David Jablonski, Tom Kammer and Graeme Lloyd. We accessed Sepkoski’s genus compendium at http://strata.geology.wisc.edu/jack/, and thank Shanan Peters for maintaining these data online. Lastly, we thank Philip Donoghue, Alistair McGowan, Robert Sansom, and one anonymous reviewer for their constructive comments on an earlier version of this manuscript. Support for this research comes from the Natural Environment Research Council (NERC NE/I005536/1, to MF), the John Fell Fund of the University of Oxford (MF), The National Science Foundation (DEB-0917922 and DDIG DEB-1011002, to LCS), The Evolving Earth Foundation (LCS), The Palaeontological Association (LCS), The Palaeontological Society (LCS), The Evolving Earth Foundation (LCS) and the University of Chicago (LCS).