Palaeozoic giant dragonflies were hawker predators

The largest insects to have ever lived were the giant meganeurids of the Late Palaeozoic, ancient stem relatives of our modern dragonflies. With wingspans up to 71 cm, these iconic insects have been the subject of varied documentaries on Palaeozoic life, depicting them as patrolling for prey through coal swamp forests amid giant lycopsids, and cordaites. Such reconstructions are speculative as few definitive details of giant dragonfly biology are known. Most specimens of giant dragonflies are known from wings or isolated elements, but Meganeurites gracilipes preserves critical body structures, most notably those of the head. Here we show that it is unlikely it thrived in densely forested environments where its elongate wings would have become easily damaged. Instead, the species lived in more open habitats and possessed greatly enlarged compound eyes. These were dorsally hypertrophied, a specialization for long-distance vision above the animal in flight, a trait convergent with modern hawker dragonflies. Sturdy mandibles with acute teeth, strong spines on tibiae and tarsi, and a pronounced thoracic skewness are identical to those specializations used by dragonflies in capturing prey while in flight. The Palaeozoic Odonatoptera thus exhibited considerable morphological specializations associated with behaviours attributable to ‘hawkers’ or ‘perchers’ among extant Odonata.

length and width. There are some traces of pin marks around the left ocelli and antenna, but none are visible around the mandibles and compound eyes.
Redescription. Head 14.4 mm long, 13.5 mm wide; compound eye rounded from above, 6.7 mm long, 6.7 mm wide; compound eyes meeting for 3.1 mm; vertex small, rectangular, 0.9 mm long, 1.5 mm wide; lateral ocellus 0.8 mm wide; median ocellus 0.8 mm wide; frons (+ clypeus?) 3.4 mm wide, ca. 2.0 mm long; general shape of vertex, lateral and median ocelli, and bases of antennae similar to those of extant Aeshnidae, due to broadly confluent eyes; mandibles strong, 6.7 mm long, 3.1 mm wide, with first incisor 1.7 mm long, second incisor 0.9 mm long, molar ca. 1.0 mm wide, with two visible strong teeth; space between second incisor and molar space narrow, similar to situation in extant Aeshnidae, while other modern Anisoptera have a broader space 10  Abdomen incomplete, showing four segments at most, preserved part ca. 33.4 mm long, 14.7 mm wide; segment I 5.3 mm long, segment II 10.0 mm long, segment III 11.7 mm long, segment IV 6.4 mm long (as preserved) but incomplete.

Taxonomy
Meganeurula selysii has an elongate ScA and a broad 'subcostal' area between the costal margin and ScA (Fig. 1), at least in the forewing, which is a synapomorphy of the Meganeurinae 1 , while MNHN.F.R53005 has no long ScA and the 'subcostal' area between the costal margin and ScA is short and narrow (Fig. 3). Also, this last fossil clearly preserves the braces CuP and CuA oblique and separated, which are tupine characters. Thus, M.
gracilipes is clearly different from Meganeurula selysii. Also, M. gracilipes does not fit in Piesbergtupinae because of its distinctly broader area between AA and AP 11 . We restore the genus Meganeurites and transfer M. gracilipes in the Tupinae (nov. sit.). Meunier, 1909 and Meganeurina confusa (Handlirsch, 1919) are the two other described Tupinae from Commentry 1 . After Meunier's figure 3 , the preserved part of the forewing of MNHN.F.R53005 is 131.2 mm. As it shows parts of RP's branches, the complete wing can be estimated around 150 mm. The holotype of G. titana has also an incomplete forewing, but ending nearly at the same point as for M. gracilipes; and its preserved part is only 109 mm long. Thus, G. titana has wings shorter than M. gracilipes, and corresponds to a different species. The holotype of M. confusa also has incomplete wings, but the preserved part ends well basal of that of M. gracilipes, and is already 131 mm long, thus the complete wing was certainly longer than that of M. gracilipes. They probably also correspond to different taxa. The incompleteness of all these fossils, especially in their wing venation, prevents us from being more accurate of their generic distinctions. While they have been put in different genera, precise distinctions between these genera remain questionable.

Gilsonia titana
Nevertheless, in the absence of evidence to the contrary, we have maintained these as distinct genera.

Morphology
-Head: Carpentier and Lejeune-Carpentier 6 indicated that the head bears two large contiguous structures ('deux grosses masses contiguës') that they refused to consider as the compound eyes. As these authors had in their hands the ventral part of body showing the legs' insertions, these 'masses' are not so well-preserved as on the counter-part that preserves a dorsal view of the body. These structures are in the exact position of the compound eyes. These are dorsally meeting for quite a long distance (Fig. 1). The reconstructions [12][13]  are not submarginal. In fact, these structures are clearly visible (Fig. 4), but they are not organic but simple breaks in the matrix. This fossil has no lateral filaments (gills?) at all. The Meganeuridae had no lateral abdominal appendages, as previously indicated 1 on the basis of the study of Permian Tupinae. Also, the same authors interpreted some breaks in the matrix at the end of the preserved part of the abdomen as strange abdominal appendages. The abdomen of MNHN.F.R53005 is incomplete (Fig. 3), with only four segments preserved. It is highly probable that the apical part of the abdomen was hidden under the matrix of the part, as it is visible in the photograph of Meunier 3 . Even if we have little information on the structure of the abdomen of these insects, they had 10 abdominal segments as in extant Odonata.
MNHN.F.R53005 has no visible secondary male genital apparatus on the second abdominal segment, but this segment is visible in dorsal view so such structures could be hidden. It could also be a female or it had a manner of mating different from that of the Odonata, as already supposed 13 . The oldest Odonatoptera with a visible male secondary genital apparatus are Permian and belong to the grade 'Protozygoptera' 19 .

Material and methods
We performed two PCA (Principal Component Analysis), a multivariate method conventionally used in morphometry and biometrics 20 widely used in zoology, botany, and evolutionary biology. We analyzed and visualized a dataset corresponding to measurements taken from four fossil taxa and 21 extant Odonata. These were chosen for their representativeness of the sizes of bodies, wings, and eyes among extant Odonata. The studies are based on 12 measurements of the head (compound eyes), thorax (Th), prothoracic legs (P1), and wings (Extended data fig. 3).
These measurements were taken with vernier calipers (0.1 mm precision) on collection specimens (Arthropods Collections, MNHN Paris) and have been preferred to automated methods that are sources of error. Asymmetry was not considered; all measurements were done on the right side, by default. Sex was not considered in this analysis because it is not frequently known for all compression fossils, but it needs to be considered in studies of extant taxa (sexual dimorphism vs. polymorphism). The measurements made are the smallest denominator between the fossil and the extant specimens. The final dataset has no missing data. The list of taxa measured is given in Supplementary  Table 2). The largest representatives (wingspan) of the extant Odonata have been taken into account (Petaluridae and Pseudotigmatidae). Hunting behaviors are taken into account, H (hawker) for hunters in flight, and P (perching) for species that hunt from a support or catch prey that do not fly. Some species are also known for their twilight behavior, as in the genus Gynacantha (Libellulidae).

Results
The analysis 1 (Extended data fig. 4)  In the log shape ratio analysis (Extended data fig. 6), despite of the transformation, the size, and especially the wing size (Wingspan, wing length and width) strongly influence the positions of large species (Meganeurites and Erapsiteroides).

Discussion
The extant Odonata are good models for studies of morpho-functional evolution of particular organs but it is essentially the wings that are taken into account 30 . Few studies combine other characters than wings in the morphometrics of Odonata.
Even less markedly, fossils are rarely taken into account in this type of study, which is confined to the micro-evolutionary level of population or phylogenetic studies in a restricted set (effects of migration and various life traits, including biomechanics) 31-32 . The insertion of fossils in this type of study is promising and allows morpho-functional inferences as in the case of mimicry in fossil orthopteran wings 33 .
The multivariate analysis of 12 morphological variables of 25 Odonatoptera, including four fossil taxa (Extended data figs. 4-5) shows a very strong influence of size at the expense of the shape of the structures that we seek to implement. However, we consider size as an important variable because it has been defined as an apomorphy to characterize the Meganeuridae 2 . Nel et al. 1 , when describing Meganeuridae less than or equal to 100 mm in size (wingspan), discussed this trait, which no longer can be accepted as an apomorphy of this family. However, we must consider that the size of these organisms also influences their functional morphology and thus the shape of certain organs. The arrangement of the compound eyes, not much contiguous to broadly contiguous (Ey-con), and their distance when they are separated (EE-Dist), discriminate quite well two taxonomic sets (Extended data fig. 4). This arrangement also distinguishes the two sets of extant Odonata represented by hawkers (Libellulidae, Petaluridae) and perchers (mainly Zygoptera) (Extended data fig. 5).
Meganeuridae are well differentiated in this scheme, approaching the morphology of hawkers (shape of the head and legs, including the length of the protibia). MegF (Meganeurula selysii) is divergent from the other Meganeuridae in the context of this study, since this specimen is 'over-prepared' at the level of the head (eyes and especially the mandibles exaggeratedly prominent). Its abnormal position confirms this fact. So it must be removed from analyses and discussion. Erapsiteron is well separated from all the other taxa. It is also not well preserved; leading to a problem of interpretation, and must be reconsidered in this sense. The Jurassic fossil Bellabrunetia (Belcat) is also quite well separated from other taxa, probably by its size larger than in the extant Odonata. The ANOSIM and ADONIS (a MANOVA) tests confirm the partitions of the morphospace and the relevance of the analysis by a high significance that also confirms the relevance of the dataset (characters and specimens) for this issue. Further analysis with expanded data needs to be performed to improve this approach.
Several ways exist to reduce the size effect in this type of analysis 28 . However, in this first analysis, taking into account axes 2 and 3 (more than 30% of inertia) makes it possible to highlight the conformation (shape) by excluding the component which carries the greatest part of inertia due to the correlation with size (Extended data fig. 5). The distinction of taxa into perching hunters or hawkers makes it possible to distinguish two morphologically differentiated groups (P, H). To avoid influencing the analysis, we did not make a priori assignments for any of the four fossil taxa. In this analysis, Meganeurites [Meg] is found in the H group that practice hawker hunting type, especially due to the conformation of the compound eyes: i.e., contiguous compound eyes (little contiguous to broadly contiguous) that marks this group and this character.
Extended data fig. 6 illustrates an analysis of log shape ratio, a method to remove the size effect from a morphometric data set 27 . This allows to highlight several groupings (more or less by organs: eyes, head, thorax, legs) and oppositions of variables, such as the inverse relationship between the length and the thickness of the protibia. The size effect is always present and discriminates all fossil species including [Meg]. The shape of the compound eyes is less discriminating in this analysis.
This first preliminary study will make it possible to set up morpho-functional comparative methods including fossils and extant lineages by increasing the sampling, in the extant one and for the fossils that have their head preserved. A more complete analysis is in progress in this direction and will detail this new comparative approach, including phylogenetic considerations.
The selected morphometric characters are relevant to discrimine taxa and morpho-functional features in our dataset, including fossils and in particular M. gracilipes. This unique specimen is critically important for deducing the paleobiology of this lineage. The size variable played probably a functional role in the life traits of these organisms (to determine their ecological niche). It may be relevant to take their sizes into account in comparative morpho-functional approaches as demonstrated herein.