We first grouped all higher-ranked taxa of non-benthic metazoans according to their assumed dominant mode of life into demersal, plankton and nekton (Table 1; a discussion of the assignment to ecological megaguilds can be found below). We did not include cnidarian plankton such as scyphozoans and ctenophores because of their low diversity in the Devonian (probably because of a preservational bias). We then analysed the stratigraphical ranges of all genera comprising these ecological megaguilds based on Sepkoski’s compendium (Sepkoski 2002) to assess their diversity trajectories at the stage level. Data from the Paleobiology Database (PaleoDB, http://paleodb.org/) were used to test if the patterns were matched by abundance data estimated from the number of occurrences of each megaguild. Although both Sepkoski’s compendium and the PaleoDB contain taxonomic errors, these are unlikely to affect large-scale diversity patterns (Wagner et al. 2007). Sampling problems are largely taken into account by focussing on proportional rather than raw data (Madin et al. 2006) or by applying rarefaction analyses when detailed occurrence counts were available.

In addition to Sepkoski’s compendium (Sepkoski 2002), unpublished or new databases were available on acritarchs (not included in Sepkoski’s data) and several invertebrates and vertebrates (see below). These data are partially derived from our own investigations and were used to evaluate detailed ecological changes among Palaeozoic marine metazoans. Based on these data, we performed simple diversity analyses including counts of boundary-crossing genera of all groups. As we included rapidly evolving groups such as acritarchs, dacryoconarids and cephalopods in our study, we also tabulated diversity with taxa known from only one stratigraphical interval (singletons).

We assume that the following major taxa lived in demersal habitats: Among the nautiloids, all nautiloids originating during the Cambrian, the Actinocerida, Ascocerida, Discosorida, Ellesmerocerida, Endocerida, Lituitida and Oncocerida are here considered demersal (Table 2); this inference is based on the facies they occur in and morphological features such as coiling and position of hyponomic sinuses (Chen & Teichert 1983; Stridsberg 1985; Westermann 1999; Kröger & Mutvei 2005). Most of the Radiodonta and Eurypterida are also thought to have lived in the demersal zone because of the presence of what probably were swimming and walking appendages. Agnathans (Galeaspida, Osteostraci and Pteraspidomorphi; Tables 3–5) most likely shared this habitat because of their usually dorsoventrally flattened body (Janvier 1996). Cephalochordata were probably demersal like their modern relative, the lancelet (Branchiostoma). Some Devonian fish traces provide evidence for a demersal habitat for the jawless fishes (Morrissey et al. 2006). As far as thelodont genera in the Devonian are concerned, they essentially show the pattern discussed but only one genus extends through the Frasnian-Famennian boundary into the lower-middle Famennian (Märss et al. 2007). At present we include it in Turinia but this has become a waste basket taxon and will probably change. Thelodontidid and turiniid thelodont global distribution is different from other agnathans which might reflect larvae in the plankton, but as adults, many had a demersal and some possibly a nektobenthic mode of life (Table 6).

We included graptoloids, dacryoconarids, homoctenids, orthocerids and bactritids as plankton (Tables 7–10). Many members of these groups occur frequently in black shales and thus certainly lived in the water column. Most orthocerids were probably capable of minor horizontal movements but they were ineffective swimmers and migrated predominantly vertically and/or drifted passively (Hewitt & Watkins 1980; Westermann 1999; Mutvei 2002; Kröger 2003, 2005; Kröger & Mutvei 2005; Mutvei et al. 2007). This is suggested by their poorly differentiated muscle-attachment structures, the absence of significant endosiphonal or endocameral deposits and, in some cases, also shell morphology. Dacryoconarids and homoctenids (small conical shells of unclear systematic affinity) were too small to have been part of the nekton (Li 2000; Berkyováet al. 2007). Graptoloids simply had a colony-morphology unsuitable for active swimming but have been shown convincingly to have lived as vertical migrants (Finney 1979; Rigby & Rickards 1989).

The rich phytoplanktonic nutrient reservoir that existed from the Cambrian until at least the Late Devonian is reflected in the Early Devonian rise and extraordinary abundance of dacryoconarids and homoctenids which persisted until the Givetian (Middle Devonian; Li 2000; Berkyováet al. 2007) as well as in the high radiolarian diversity (new data) which increased from the Givetian until the Early Carboniferous.

Here, we included the ammonoids, cartilaginous and bony jawed fishes, and most coiled nautiloids in the nekton (Tables 11–15). Ammonoids and coiled nautiloids are considered nektonic organisms because of their differentiated muscle-attachment structures and buoyancy devices, their occurrences in black shale facies, and actualistic comparisons (Doguzhaeva & Mutvei 1991, 1996; Klug & Korn 2004; Kröger et al. 2005). Recent studies have convincingly falsified the classical arguments against the high mobility of the ammonoids (Jacobs & Chamberlain 1996) such as the absence of retractor muscles (see also Klug et al. 2008b) and the closer phylogenetic relationship to the coleoids than to the nautilids utilizing different musculature for propulsion. The presence of muscle attachments in ammonoids comparable with those of the nautilids has now been shown for various ammonoids (Doguzhaeva & Mutvei 1991, 1996; Kröger et al. 2005; Klug et al. 2008b) and the idea that a coleoid mantle was hidden in an ammonoid shell is not supported for phylogenetic reasons since the body chamber was reduced long after the ammonoids had evolved from the bactritoids (Fuchs 2006; Kröger & Mapes 2007).

The coiled Tarphycerida and Nautilida are interpreted as nektobenthic or nektoplanktonic based on actualistic comparision of the shell form, muscle-attachment structures and position of the hyponome (Westermann 1999; Kröger & Mutvei 2005).

Among Devonian cephalopods, several striking evolutionary inventions are apparent. During the Early Devonian, the ammonoids, one of the most important groups of fossil marine invertebrate metazoans, evolved from bactritoids (Kröger & Mapes 2007) which had, in turn, just evolved from the orthocerids (Hewitt & Watkins 1980; Sepkoski 2002; Kröger & Mutvei 2005; Kröger & Mapes 2007; Kröger 2008). While the diversity and abundance of most cephalopods with straight conical or incompletely coiled shells decreased significantly towards the end of the Devonian, groups with adaptations to active, horizontal swimming life modes began to diversify in the Early Devonian. In addition to the bactritids and ammonoids, the coleoids (squids and octopods: Fuchs 2006; Kröger & Mapes 2007), or at least their ancestors, evolved. While the nautiloids produced conch morphologies strikingly similar to the earliest ammonoids, the coleoids embarked on a differing strategy by later forming internal shells, the only successful strategy with respect to modern cephalopods. Interestingly, shell coiling or an increase in shell coiling occurred in such different groups as in dacryoconarids, gastropod larvae, bactritoids and ammonoids, both juvenile and adult. This can be interpreted as a reaction to the increasing predation pressure (Nützel & Frýda 2003).

Early Palaeozoic marine vertebrate remains belong predominantly to the agnathans (jawless fishes). Early jawed fishes (Gnathostomata) are rarely documented from strata older than the Silurian (Gagnier 1989; Janvier 1996; Sansom & Smith 2001). Marine gnathostome diversity increased explosively in the Mid Devonian with the radiations of placoderms, cartilaginous and bony fishes (Denison 1978, 1979; Zangerl 1981; Cloutier & Forey 1991; Ginter et al. 2008). Remarkably, these evolutionary patterns coincided with some crucial morphological and ecological alterations (Long et al. 2008). Laterally compressed and thus nektonic body forms became more prevalent than the dorsoventrally flattened forms adapted to a predominantly demersal to benthic mode of life (Janvier 1996). Diversification trends can be grouped according to life habits: (1) demersal forms that display a clear diversity decrease, and in some cases extinction, towards the end of the Devonian (many jawless fishes); (2) nektonic forms, which exhibit low diversity in the Early Devonian, became highly diverse towards the end of the Devonian, and survived the end-Devonian extinctions with little or no loss (most jawed fishes) and (3) intermediate forms which were neither clearly demersal nor truly independent from the seafloor (acanthodians and placoderms). The data presented here require new analyses in the future which incorporates latest publications and results from research in progress (e.g. placoderm data are currently being revised by M. Rücklin, Bristol; reasonably new data were published by Long 1993; Carr 1995 and Blieck & Turner 2000 and were included in our analyses).

The diversity curves of the acritarchs, ammonoids, placoderms, bony and cartilaginous fish track roughly parallel throughout the Devonian with maxima in the Late Devonian, whereas the diversity of jawless fishes and acanthodians fluctuated in different ways.

Diversity data of the radiodonts, the eurypterids and graptoloids were extracted from Sepkoski’s compendium (2002) and the Paleobiology Database. Gastropods with openly coiled protoconchs formed a considerable, sometimes even dominant, part of Ordovician and Silurian gastropod communities. During the Early Devonian, their number rapidly decreased and their embryonic shells became on average smaller (Nützel & Frýda 2003; Nützel et al. 2007; Frýda et al. 2008). These macroevolutionary trends were followed by the Late Palaeozoic radiation of Neritimorpha, Caenogastropoda and Heterobranchia. Thus, the Devonian was the period which determined the composition of all post-Palaeozoic gastropod faunas.

Both the diversity and abundance data suggest an initial scarcity of macroplankton and nekton, an Ordovician plankton radiation and a Devonian nekton revolution (Figs 1, 2); this is in accordance with the results of Bambach (1999). Although there seems to be a greater proportion of plankton in the early Palaeozoic according to abundance data (Fig. 1B), this may largely reflect differences in the documentation of demersal and planktonic groups in the PaleoDB.

The diversity of acritarchs was clearly lower in the Devonian than in the Ordovician but still high compared with the late Palaeozoic (Servais et al. 2008). During the Late Devonian, acritarchs experienced a last radiation prior to the subsequent strong decline (‘phytoplankton blackout’; Mullins & Servais 2008). Most important groups of Devonian invertebrate meso- and macro-zooplankton are not only morphologically similar (bactritoids, dacryoconarids, homoctenids, orthocerids) but also declined synchronously in diversity towards the end of the Devonian, similar to jawless fish diversity (e.g. Janvier 1996; Märss et al. 2007). Some nektonic groups with a close affinity to the benthos such as many nautiloids, acanthodians and placoderms behaved differently (e.g. Long 1993; Bambach 1999). They underwent minor extinctions and radiations during the Devonian but, except for the placoderms, persisted much longer than the end-Devonian. Taken together, the nektonic groups display a rising diversity through most of the Devonian. Although they suffered during the Late Devonian mass extinctions to a varying degree, they continued to radiate and became highly diverse in the Late Devonian and Early Carboniferous. Remarkably, the Mid- to Late Devonian generic diversification of the nekton was delayed compared with the origins of most of the higher taxa.

After the Cambrian explosion (Seilacher 1999; Butterfield 2001; Hu et al. 2007) and the Great Ordovician diversification (Turner et al. 2004; Harper 2006; Servais et al. 2008, 2009), all nektonic organisms display a steep diversity increase from the Late Silurian to the Early Carboniferous at the expense of demersal and planktonic forms (compare Bambach 1983, 1999). This is accompanied by the radiation of radiolarians and various mollusc clades with planktotrophic juveniles or larvae (cephalopods, gastropods and perhaps bivalves).

For the explanation of the simultaneous explosive diversification of the nekton at the cost of various planktonic and demersal benthic groups, three hypotheses are available:

The first hypothesis is corroborated by the radiation of molluscs with planktotrophic larvae and juveniles; profound changes in larval morphology and thus, reproductive and larval strategies among many molluscs began in the latest Cambrian and intensified during the Devonian (House 1996; Nützel & Frýda 2003). Morphological adaptations of gastropod larvae and juvenile ammonoids to changes in the planktonic habitat are indicated by the closure of a larval or early juvenile umbilical opening, size-decrease of the larval or embryonic shell and occurrences of their shells in anoxic environments (Nützel & Frýda 2003). Regional faunal analyses of the Moroccan Lower Devonian (Klug et al. 2008a) have shown that bivalve associations change from palaeotaxodont-dominated towards pteriomorph-dominated with perhaps planktotrophic larvae (Jablonski & Lutz 1983). Therefore, an increasing proportion of larvae and juveniles from many important mollusc groups fed probably on plankton during the Devonian (Jablonski & Lutz 1983). There is, however, no cross-correlation between originations and extinctions of phytoplankton and the zooplankton (R = 0.2 and 0.38, respectively, not significant) and thus, the first hypothesis is insufficient to explain the observed diversity fluctuations.

Lasting and repeated anoxic episodes are well-documented for the Devonian (Joachimski & Buggisch 1993; Algeo et al. 1998). The main prediction drawn from the second hypothesis would be an extinction of all forms that lived close to the seafloor and were unable of escaping into higher, well-oxygenated water levels. Phytoplankton, zooplankton and nekton should be less affected by anoxia-induced extinctions. Nevertheless, this hypothesis fails to explain why the nekton radiation begins clearly before the most significant anoxia of the end-Givetian and -Frasnian.

The third hypothesis implies an initially high abundance and diversity of organisms in the demersal zone, followed by a selection for a mobility increase, ultimately leading to the rise of descendants of demersal organisms. This is in accordance with the observed diversity and occurrence fluctuations. Around the Silurian-Devonian boundary, the predominantly demersal jawless fish were rather diverse (Janvier 1996). In general, nektonic organisms with some demersal affinity such as several groups of nautiloids, acanthodians, placoderms and thelodonts display repeated minor diversifications and extinctions. The acanthodians have their maximum diversity in the Early Devonian, followed by a long-term decline. Many placoderms had thick bony armour and are morphologically intermediate between dorsoventrally flattened agnathans and laterally compressed bony fish, indicating a demersal to nektonic habitat (Janvier 1996). These mobile predators possibly suppressed many demersal animals (such as some cephalopods). The placoderm diversity peak (Fig. 2) predates the diversity and abundance maxima of the perhaps more meso- to epipelagic nekton such as some derived nautiloids (Kröger 2005), ammonoids (Klug & Korn 2004), chondrichthyans (sharks and relatives; Miller et al. 2003; Ginter et al. 2008) and bony fish. Potentially, the nektonic predators subsequently suppressed representatives of the zooplankton, causing a partial zoo- and phytoplankton extinction (graptoloids, dacryoconarids, homoctenids, ‘phytoplankton blackout’) or at least profound changes. Simultaneously, the proportional diversity of benthos decreased relative to that of inhabitants of the water column (Fig. 3).

We tested our conclusions by plotting the proportional diversity of benthic animals versus those inhabiting the water column (Fig. 3). The resulting graph clearly shows that diversity changes among taxa inhabiting the water column cannot be explained simply by a general increase in the diversity of marine taxa.

Additionally, we produced regressions for boundary crossers, genera with and without singletons for all marine genera, benthos only and nekton only (Fig. 4). The regressions support our conclusions that the diversity of nekton increases at the cost of other marine genera including benthos and plankton.

Major macroecological changes usually require some kind of trigger: The appearance of early mobile predators induced the evolution of skeletons and thus the Cambrian radiation, the Ordovician abundance of microplankton formed the basis for the radiation of larger plankton, the Ordovician to Devonian radiation of land plants enabled animals to follow on land. The evolutionary acquisition of the capability to swim actively in the pelagic realm thus followed the evolution of demersal and planktonic innovations, which occurred in several groups of macroscopic metazoans more or less simultaneously in the Cambrian and Ordovician (e.g. various arthropods, early cephalopods and graptoloids). The timing of diversification of various demersal and nektonic groups suggests that the Devonian Nekton Revolution was initiated by an escalation in the benthic and demersal zones (Fig. 5). The synchronous decline of benthic animals (Fig. 3) might be sufficient to explain why the overall diversity is not strongly affected. Increasing nutrient supply from the increasing terrestrial biomass probably supported the diversification of microplankton, organisms with planktotrophic larvae (Cambrian pelagic predators are here considered as parts of the plankton because of their small size; compare Butterfield 2001 and Hu et al. 2007) and nekton, which suppressed many demersal organisms and the macroplankton from their habitats. Consequently, among the marine metazoans, almost only nektonic predators of most scales and the diverse benthos remained following the explosive nekton radiation in the Early Devonian while the proportion of demersal and planktonic taxa decreased dramatically.

We thank M. Rücklin (Bristol) for information on placoderms. A. Nützel (Munich) and two anonymous reviewers provided valuable comments. C.K. was supported by the Swiss National Science Foundation (project No. 200021-113956/1). Synthesys funded a trip to Berlin by C.K., representing the starting point for this project. G.L.M. thanks the Leverhulme Trust (grant F/00212/F, awarded to R. J. Aldridge) and W.K. the Volkswagen-Stiftung. S.T. thanks for the opportunity to work as a guest researcher at the Institut für Geowissenschaften (Tübingen). This study is also a contribution to IGCP projects 491 and 503. J. Křiž shared his knowledge of Devonian bivalve ecology and M. Ginter made his shark occurrence data available. P. Janvier (Paris) generously let us use his diversity database of jawless fish. K. De Baets (Zürich) also discussed the manuscript with us.