Chaetognaths (‘arrow worms’) are tiny (adults approx. 2–120mm long), extremely prolific animals that are often second only to copepod crustaceans in terms of abundance and biomass in the world's oceans (5–15% biomass; Longhurst 1985). They play an important role in the marine trophic web both as consumers and as a food source for larger animals. Their bodies are streamlined, almost translucent and divided into head, trunk and tail regions by internal transverse septa. Their only hard parts with high potential for being fossilized are chitinous circumoral grasping spines and teeth (gs and te, respectively, in figure 1f) attached to lateral plates composed of collagen-like material. These three elements form a powerful feeding apparatus. Chaetognaths are predators and feed on a variety of mesoplanktonic organisms, their most frequent prey being copepod crustaceans (Terazaki 2000). Prey is detected by ciliary receptors, captured by the crown of spines, and forced into the mouth by the teeth. Although the majority of present-day chaetognaths are permanently pelagic (e.g. Sagitta), a few species are epibenthic and can attach to the substrate. Swimming involving rapid dorsoventral undulations of the body alternates with motionless and passive sinking in the water column.

Figure 1 Lower Cambrian and Recent chaetognaths. (a),(b),(d),(e) Protosagitta spinosa from the Chengjiang biota (Yu'anshan Formation, Ercaicun near Kunming, Yunnan Province, China, coll. Yunnan Institute of Geological Science, Kunming, (more ...)

The evolutionary origin of chaetognaths has long been a subject of debate, but recent molecular studies (mitochondrial genes; Helfenbein et al. 2004; Papillon et al. 2004) suggest strong affinities with protostomes. The fossil record of Chaetognatha is extremely patchy and the evolutionary history of the group is still obscure. The oldest putative ancestors of the group are two specimens, both from the Early Cambrian Chengjiang biota: (i) Eognathacantha ercainella Chen and Huang (Chen & Huang 2002) is poorly preserved, making the recognition of most diagnostic features used for chaetognaths questionable (e.g. head), and (ii) Protosagitta spinosa Hu, first briefly reported in Chen et al. (2002) is fully described in the present paper (figure 1). The presence of chaetognaths in the Middle Cambrian Burgess Shale has long been discussed (e.g. Oesia, Amiskwia; Walcott 1911; Conway Morris 1977; Butterfield 2003; Szaniawski 2005) but needs confirmation from the study of unpublished material (D. E. G. Briggs 2005, personal communication). There is a lack of fossil data concerning the group until the Early Carboniferous (phosphatic concretion with numerous chaetognath grasping spines; Doguzhaeva et al. 2002). Chaetognaths possibly occur in the Upper Carboniferous Mazon Creek Lagerstätte (Schram 1973), but the very poor preservation of their body features allows no accurate interpretation. Another important set of fossil evidence that is discussed in the present paper comes from recurved spine-like small shelly fossils (SSF) known as protoconodonts (Bengtson 1976, 1983). Protoconodonts are abundant in the Cambrian of many regions and have been convincingly interpreted as the possible elements of chaetognath grasping apparatuses (e.g. Szaniawski 1980, 1982, 1983, 2002; Andres 1988; Azmi 1996).

The head shows two symmetrical sets of lateral grasping spines, and elliptical features interpreted here as the trace of lateral plates and associated muscles. One set of spines has at least eight elements, each with a recurved shaft and a wider base (best preserved spine, 1.2mm long). The opposite set of grasping spines is more closely packed and single elements more difficult to discern. Straight bar-like features below the left set of spines are interpreted as possible teeth (figure 1d–f). Most of the trunk and tail regions are filled with irregular pinkish to reddish patches that may represent original muscle masses. In Recent chaetognaths, longitudinal muscles comprise approximately 80% of the tissue volume (Shinn 1997; Casanova & Duvert 2002). The posterior part of the trunk of P. spinosa has a brownish elliptical area that contrasts with the overall pale colour of the fossil specimen. This area lies in the same position as the ovaries of Recent chaetognaths and may represent the outline of the fluid-containing space originally occupied by oocytes (figure 1b,c). The trunk of Recent chaetognaths contains no organs apart from the intestine and the ovaries.

It has been demonstrated (Gabbott et al. 2004) that pyritization controlled by nucleation crystal growth rates was largely responsible for replicating numerous types of non-mineralized tissues of Chengjiang animals, and that the type of pyrite mineralization reflects the more or less decay-susceptible nature of the tissues involved. Although SEM observations of the iron oxide coating of our unique specimen were not allowed, we suppose that colour differences observed in the fossil chaetognath express differences in the original tissue composition of the animal (e.g. chitinous grasping spines, more labile muscles and possible ovarian structures). In numerous key aspects of its morphology (overall streamlined shape, body division, bilateral symmetry, head armed with grasping spines and assumed teeth, muscles and ovaries), this fossil chaetognath is almost identical to Recent species (figure 1). Although unique, this specimen brings firm evidence that the chaetognath body plan was established in the Early Cambrian and confirms previous assumptions based on poorly preserved material (Chen & Huang 2002; E. ercainella). Whether P. spinosa and E. ercainella, both from the Yu'anshan Formation, are conspecific is uncertain. Most differences observed between the two forms (e.g. tail and fin structure) may result from preservation factors.

Chaetognaths have long been assumed to be present in the Middle Cambrian Burgess Shale, with two plausible candidates, namely Amiskwia sagittiformis Walcott 1911 and Oesia disjuncta Walcott 1911. Although these two forms, especially Oesia, indeed superficially resemble Recent chaetognaths (overall shape, tissue distribution; Butterfield 1990, 2003; Szaniawski 2005), we see strong objections against their placement within the Chaetognatha, especially the lack of clear evidence of a grasping apparatus, the posterior location of the anus and the prominent cephalic tentacles of Amiskwia. The idea that the ancestral chaetognaths may have possessed distinctive features (Conway Morris 1977) is not confirmed by the Early Cambrian P. spinosa from Chengjiang. This displays no major difference from Recent chaetognaths (figure 1).

• Both protoconodont elements and grasping spines of Recent chaetognaths have a recurved conical shape, a well-developed base, an internal cavity reaching the apex and longitudinal ornament (figure 2a,b,e,f; Nagasawa & Marumo 1973; Müller & Hinz 1991; Moreno & Kapp 2003; Qian et al. 2004). The size ranges of P. anabarica (length 1.0–2.5mm), Protohertzina unguliformis (length 1.6–2.9mm) and Mongolodus longispinus (length 0.4–1.3mm) studied herein (basal Cambrian, China) are similar to that of Recent chaetognath grasping spines (length 0.3–2mm).

  Figure 2 Protoconodonts from the lowermost Cambrian Kuanchuanpu Formation, Hexi Shaanxi Province, China (a)–(b) and Kuanchuanpu (c)–(d), both, and grasping apparatus of Recent chaetognaths (e)–(h).(a),(b) Protohertzina unguliformis (more ...)
• The wall of chaetognath grasping spines has a thick middle layer formed by closely packed fibrils of chitin (figure 3a,b) and thin inner and outer layers. A similar three-layered structure is present in all protoconodonts and is remarkably preserved in Phakelodus from Upper Cambrian of Poland (Szaniawski 1983, 2002). Similarly, the middle layer (approx. 50μm thick) of P. anabarica from the basal Cambrian of China has a fibrous phosphatic fabric with elongated, imbricated crystallites running parallel to the shaft axis. The assumed outer layer (approx. 10μm) is featureless with dominant apatite microspherules. The inner layer is often thickened by secondary phosphate growth (figure 3c,d).

  Figure 3 Wall microstructure of Recent chaetognath grasping spines (a,b) and lowermost Cambrian protoconodonts (c,d), from the Kuanchuanpu Formation, Hexi, Shaanxi Province, China. (a) Eukrohnia hamata, three-layered chitinous structure. (b) (more ...)
• Chaetognath grasping spines grow through the combined secretion of pulp cells and anchor cells, and by basal accretion (Shinn 1997). This growth mode produces discrete oblique increments along the external surface of the shaft that are also visible in protoconodonts (e.g. Szaniawski 1983, 2002; Andres 1988; Müller & Hinz-Schallreuter 1998).
• The most compelling evidence that protoconodonts were indeed the elements of a grasping apparatus comes from natural clusters in acid residues (three-dimensional preservation; figure 2b–d) or flattened in shales (electronic supplementary material 3). These natural clusters typically consist of crescent-like half-apparatuses with 2–13 elements. More rarely, these assemblages are found in two symmetrical sets that strongly recall the bilateral series of grasping spines (4–14 each; figure 1f) that surround the mouth of modern chaetognaths. Protoconodont elements within individual clusters display moderate size and shape variations (Müller & Andres 1976; Müller & Hinz 1991) comparable with those observed in chaetognath apparatuses. A well-preserved Mongolodus cluster from the basal Cambrian of China is shown here (figure 2c,d). It has five hooks oriented in the same direction and all inserted within a single basal structure. This cluster is unlikely to result from the cementation of isolated elements deposited in sediment and, instead, is represented as a natural feature closely resembling the half-grasping apparatus of a chaetognath (figure 2g). The lacunar phosphatized structure that surrounds the base of the hooks is reminiscent of the basal electron-lucent cuticle of chaetognath grasping spines (Shinn 1997), which clearly supports the view that protoconodont hooks, similar to chaetognath spines, had a soft tissue basal support structure. Similar Mongolodus clusters with only two to four elements occur in the lowermost Cambrian of northern India (Azmi 1996). The presence of possible teeth associated with protoconodont natural clusters (Szaniawski 2002) also supports the chaetognath affinities of protoconodonts.
• The grasping apparatus of the chaetognath from the Chengjiang biota (P. spinosa; figure 1) closely resembles protoconodont clusters by its size (electronic supplementary material 3) and its closely packed crescent-like arrangement.

It becomes clear from these five lines of morphological evidence that the animals bearing protoconodont sclerites were chaetognaths or belonged to a group closely related to them. The obvious raptorial function of their apparatus and their lowermost Cambrian occurrence place the chaetognaths s.l. among the earliest marine predators (Szaniawski 2002). The complexity of their feeding apparatus, involving sclerites and associated muscle and neuronal systems, points to the Precambrian origin of the group. Chaetognaths were obviously not restricted to the single species from the Chengjiang biota (figure 1) and were already abundant, widespread and diverse in the earliest Cambrian (electronic supplementary material 3; Qian et al. 2004). Protohertzina unguliformis (figure 2a,b) represents up to 3.6% of the SSF assemblages in southern Shaanxi (up to 1400 specimens per kilogram of rock).

Morphology, biometrics and distribution patterns support a pelagic rather than a benthic lifestyle for the Early Cambrian chaetognaths s.l. Some Recent benthic chaetognaths use adhesive organs to attach to solid substrates (Kapp 1991; Shinn 1997). No such features are present in the Chengjiang specimen (figure 1). The musculature of Recent chaetognaths is essentially longitudinal and concentrated in the trunk and in the tail (Bone & Duvert 1991). Additional transverse muscles, present in all Recent benthic chaetognath species, could not be discerned in the Chengjiang specimen. The relative lengths of the trunk and the tail vary between Recent benthic and pelagic chaetognath species (Shinn 1997; Casanova 1999). We calculated the body ratios of 52 pelagic and 12 benthic Recent species and those of the two specimens from the Chengjiang biota. Multivariate analysis displays a benthic and a pelagic group for the Recent species and places the Cambrian specimens within the pelagic group (electronic supplementary material 1 and 2). The chaetognaths from the Chengjiang biota are preserved in mudstones deposited in a shallow-water environment (approx. 50–200m depth; Hou et al. 2004; Hu 2005). Protoconodonts are recorded worldwide (electronic supplementary material 3). Protohertzina occurs around the margins of several major Early Cambrian palaeocontinents, such as Laurentia, Siberia, Kazakhstan, Mongolia and South China, and its two species, P. anabarica and P. unguliformis (figure 2a,b), have a widespread occurrence over the Yang Tze Platform, being associated with both shallow and deeper water lithofacies (figure 4). These distribution patterns suggest wide oceanic dispersal capabilities and infer a pelagic lifestyle.

Figure 4 Facies reconstruction and distribution of protoconodonts over the Yang Tze Platform, South China, for the earliest Cambrian (Meishucunian Stage). Protohertzina anabarica, P. unguliformis and Mongolodus longispinus.

Recent arrow worms achieve predation by their grasping apparatus (spines and teeth) in addition to neurotoxins, thus possibly explaining their ability to capture relatively large prey (e.g. fish larvae; Thuesen & Bieri 1987; Thuesen et al. 1988). The presence of a similar feeding apparatus (grasping spines and teeth; figures 1d,e, 2i,j) in the Early Cambrian ancestors of the group is consistent with a function for the capture of small motile or drifting prey of possibly several millimetres long. The exact ecological niche of these early predators is difficult to ascertain but was obviously determined by the availability of prey in the water column. Evidence for potential prey, such as mesozooplankton (0.2–20mm) in the Early Cambrian oceans is sparse and limited to possible filter-feeding crustacean-like organisms (Butterfield 1994, 2001a,b) and small bivalved arthropods (Steiner et al. 1993) found in deeper water black shales. Other swimmers and drifters, such as Isoxys (Vannier & Chen 2001), anomalocaridids (Vannier & Chen 2005), ctenophores (Chen & Zhou 1997) and medusoids (Zhu et al. 2002) were mostly larger (greater than 20mm) and nektonic, and fall beyond the size range of potential prey for chaetognaths. We think that hyperbenthic niches (approx. 1–10m above bottom) may have provided Early Cambrian predator chaetognaths and protoconodont animals with an abundant food source, such as tiny arthropods (e.g. bradoriids and copepod-like Ercaia; Shu et al. 1999; Chen et al. 2001) and the possible meroplanktonic larval stages of various animal groups (e.g. naraoiid arthropods; Zhang et al. 2003). Arthropods account for more than 60% of the species found in the Chengjiang biota (Hou et al. 2004). Our hypothesis is supported by the fact that Recent hyperbenthic communities are intensively exploited for food by mature chaetognaths. For example, submersible dives using video recordings revealed extremely high concentrations of Parasagitta elegans in the hyperbenthic zone within a few metres of the bottom in Conception Bay, Newfoundland (Choe & Deibel 2000). That Early Cambrian chaetognaths lived in the epipelagic and mesopelagic zones remains a plausible alternative. However, no fossil evidence supports the idea that this part of the water column may have provided a suitable and sufficient food source for chaetognaths.

Based on fossil evidence from phytoplankton and assumed filter-feeding pelagic arthropods, Butterfield (1994, 2001a,b) suggested that the introduction of herbivorous zooplankton in the water column during the P–C transition considerably modified the structure of the early marine ecosystem. The introduction of this new tier is likely to have triggered a cascade of effects on the primary production, the phytoplankton evolution (e.g. Knoll 1994) and, eventually, set up the first coupling between pelagos and benthos (e.g. via faecal pellets; Butterfield 2001a,b). Our results suggest that the complexity of the trophic structure may have reached an even more advanced level during the Precambrian–Cambrian transition with the construction of a three-level trophic structure (primary producers, herbivorous and predator mesozooplankton). The widespread introduction of prey–predator relationships into the pelagic realm during the P–C transition via the chaetognaths seems to represent another important innovation in the tiering and functioning of early marine ecosystem that may have laid the foundations of modern-style marine food chains. Other potential pelagic consumers, such as macrozooplankton (e.g. ctenophores and eldoniids) and nekton (e.g. early chordates, vetulicolids and arthropods), were certainly inhabiting the water column, although very little is known of their exact lifestyles, diet and feeding strategies. There remains the crucial and unresolved question of when and how metazoans entered the plankton (Rigby & Milsom 1996). This key event may have resulted from different factors, such as possible changes in larval development (e.g. meroplanktonic stages) and pressure from benthic predators (Butterfield 1997). Hyperbenthic niches close to the sea bottom may have represented transitional habitats for the pioneer invaders of the water column (e.g. chaetognaths) and a step towards a subsequent and extensive occupation of the pelagic realm.

The authors thank H. Szaniawski for fossil material, S. Bengtson, M. Duvert, C. Eckert, D. Janussen, H. Kapp, G.-X. Li, B. Neuhaus, F. Paris, Y. Qian, D. J. Siveter and M. Williams for their information and constructive criticism. Funding was provided by the French Ministries of Foreign Affairs, Education and Research (DRIC) and the Chinese Ministry of Science and Technology (MOST) via the PRA T03–04 Program and by the Deutsche Forschungsgemeinschaft (DFG).