One third-instar larva of G. adocetus was studied. Guyana, Mazaruni-Potaro district, Takutu Mountains, 6°15’ N, 59°5’ W, 3–10.xii.1983, P. J. Spangler, R.A. Faitoute and P.D. Perkins leg.

For comparative purposes, larvae of Chaetarthria bruchi and unidentified larvae of Chaetarthria from Montana (USA) were examined. Information on C. seminulum comes from the literature (Fikáček 2006). Anacaenini larvae examined: Pseudorygmodus flintispangleri, one pharate third instar larva (the chaetotaxy therefore corresponds to that of the second instar larva), Chubut province (Argentina); Crenitulus suturalis, one second instar larva, La Rioja province (Argentina); Crenitis morata, two third instar larvae, Massachusetts (USA). The material studied is kept in the larval collection of the author and will be deposited in the larval collection of the Laboratory of Entomology, Buenos Aires University, Argentina.

Larval specimens were cleared in warm lactic acid, dissected, and mounted on glass slides with Hoyer’s medium. Observations (up to 1.000 ×), photographs and drawings were made with a Leica S6D dissecting microscope and a Leica DMLB compound microscope, both with a camera lucida and a photographic camera attached.

Different measurements of the head capsule and head appendages were taken with a micrometer. Measurements were used to calculate ratios, which are useful for characterizing shapes. Measured structures were adjusted as parallel as possible to the plane of the objective. The following measurements were taken. TL: total body length; MW: maximum body width, measured at level of prothorax; HL: head length, measured medially along epicranial stem from anterior margin of frontoclypeus to occipital foramen; HW: maximum head width; AL: length of antenna, derived by adding the lengths of the first (A1L), second (A2L) and third (A3L) antennomeres; SeL: length of antennal sensorium; SL: length of stipes; MPL: length of maxillary palpus, obtained by adding the lengths of the first (MP1L), second (MP2L), third (MP3L) and fourth (MP4L) palpomeres; ML: length of maxilla, derived by adding SL and MPL, cardo omitted; LPL: length of labial palpus, obtained by adding the lengths of the first (LP1L) and second (LP2L) palpomeres; LigL: length of ligula; MtW: maximum width of mentum; PrmtL: length of prementum, measured from its base to the base of LP1; PrmtW: maximum width of prementum.

Primary (present in first-instar larva) and secondary (arising in later instars) setae and pores were identified in the cephalic capsule and head appendages. Since only a third instar larva was studied, the primary chaetotaxy was interpreted and coded by comparison with Chaetarthria larvae and also with other hydrophilid genera for which the chaetotaxy is well known (e.g., Enochrus, Hydramara, Tropisternus, etc.) (Fikáček 2006; Fikáček et al. 2008, 2018; Byttebier and Torres 2009; Minoshima and Hayashi 2011, 2012, 2015; Torres et al. 2014; Minoshima et al. 2015; Rodriguez et al. 2015, 2018, 2020; Archangelsky 2016; Archangelsky et al. 2016, 2018; Minoshima 2019). Homologies were established using the criterion of similarity of position (Wiley 1981). Sensilla were coded with a number and two capital letters, usually corresponding to the first two letters of the name of the structure on which they are located. The following abbreviations were used. AN: antenna; FR: frontale; LA: labium; MN: mandible; MX: maxilla; PA: parietale; gAN: group of antennal sensilla; gAPP: group of sensilla on the inner appendage of the maxilla; gFR1, gFR2: group of sensilla on the frontale; gLA: group of sensilla on the labial palp; gMX: group of sensilla on the maxillary palp. To standardize some homology interpretations with regard to setae FR9–10 and FR5–6, we considered FR9 to be the most distal seta closest to FR14 and FR13; FR10 to be the most basal, closest to FR4; FR6 to be more anterior and mesal, closer to FR4 and FR7; and FR5 to be more posterior and closer to FR1–2.

For the analysis 27 taxa, belonging to 14 tribes of Hydrophilidae were included; Helophorus liguricus Angus 1970 (Helophoridae) was used as the outgroup to root the trees (Supplementary file 1). The resulting matrix had 128 larval characters (Supplementary file 2). The data matrix was built with Mesquite (Maddison & Maddison, 2019), and analyzed with TNT (Goloboff and Catalano 2016) under maximum parsimony (MP) (Supplementary file 3). Traditional morphological characters follow Archangelsky (2004) and Archangelsky et al. (2021); chaetotaxic characters were newly coded for all taxa included. The analysis treated all characters as unordered and equally weighted. Heuristic searches were implemented using ‘tree bisection reconnection’ as algorithm, with 200 replicates and saving 300 trees per replication (previously setting ‘hold 60.000’). Node support was evaluated with Bremer support (Bremer 1994) and bootstrap resampling with 1.000 replications. Optimizations were performed using fast and slow algorithms with WINCLADA-ASADO 1.62 (Nixon 2002). All character state changes supporting the nodes were mapped on the most parsimonious tree.

Since larvae of Paracymus-group appeared nested within Chaetarthriinae in our analyses and considering that based on molecular evidence they are not closely related, we performed a second analysis forcing the monophyly of Chaetarthriinae to examine the effect of Tormus + Paracymus on the topology. The analysis was implemented with the TNT define constraints option, and the constrained searches were carried out using the unconstrained settings.

In addition, we performed a Bayesian inference (BI) analysis using homoplasy as a partitioning criterion of discrete morphological characters (see Rosa et al. 2019). For the partition scheme, the homoplasy scores were calculated as implemented in TNT, with default concavity parameter (k=3). The characters were distributed into 11 partitions based on their homoplasy values (non-informative characters were assigned to their own partition) (see Supplementary file 4). The analyses were conducted in MrBayes 3.2.7 (Ronquist et al. 2012). The Mk model (Lewis 2001) with coding correction lset coding=variable (which corrects for lack of non-variable characters) was used, with equal rate variation across characters. We ran analyses for two independent runs (four chains each) of 5.000.000 generations each and sample frequency of 1.000. Burn-in was set to the first 25% of all samples and convergence was checked in Tracer 1.7.2 (Rambaut et al. 2013). In some of our preliminary and final analyses, Paracymus branches independently from Chaetarthriinae whereas Tormus appears deeply nested as a sister group of Guyanobius. Therefore, in order to avoid artificial topologies due to the inclusion of a taxon with convergent morphology, Tormus was excluded from these analyses.

Larval knowledge of Chaetarthriini is rather limited, of the six recognized genera (four if Apurebium and Venezuelobium are considered variants of Chaetarthria) only Chaetarthria and Guyanobius have known larvae. Within Guyanobius, only the larva of G. adocetus has been described; therefore, within the tribe it can only be compared with larvae of Chaetarthria. Archangelsky (2002) provided a table comparing Chaetarthria and Guyanobius larvae but only a few morphological traits were mentioned. We provide an updated table (Table 2) that contains several morphological and chaetotaxic features that allow us to distinguish between Chaetarthria and Guyanobius larvae. Additionally, because our findings show that Guyanobius may be more closely related to (or potentially a member of) the Anacaenini, we include a comprehensive comparison with known Anacaenini larvae (Pseudorygmodus, Crenitis, Crenitulus).

Chaetarthriinae was raised to subfamily level quite recently (Short and Fikáček 2013), grouping most genera previously included in the tribes Chaetarthriini and Anacaenini by Hansen (1991, 1999), and a few genera originally described in other subfamilies (i.e., Pseudorygmodus, Horelophus, Phelea) (Fikáček and Vondráček 2014; Fikáček and Watts 2015). Out of the 14 genera (13 if Apurebium and Venezuelobium are considered synonyms of Chaetarthria (Short 2009)) that comprise the subfamily only the larvae of five genera (38.5 % of the genera: Chaetarthria, Guyanobius, Crenitis, Crenitulus and Pseudorygmodus) could be included in this analysis since the remaining are unknown.

Neither the monophyly of Chaetarthriinae nor that of its tribes is supported by larval characters. Our results are in contrast to those analyses based on molecular data (Short and Fikáček 2013, Clarkson et al. 2019) in which Chaetarthriinae is recovered as monophyletic. The monophyly of Chaetarthriini and Anacaenini needs to be clarified. In the phylogeny of Short and Fikáček (2013)Chaetarthriini was resolved as a monophyletic grouping in the maximum parsimony analysis, but as a paraphyletic grade of two successively branching clades (Chaetarthria-group, Guyanobius-group) subordinate to the Anacaenini in the Bayesian analysis. On the other hand, the preliminary results of a more recent molecular study of Chaetarthriinae, do not support the monophyly of both tribes, being Guyanobius more closely related to a clade formed by Anacaena, Notohydrus and Pseudorygmodus (Clarkson et al. 2019). All analyses performed here support the clustering of Guyanobius with the tribe Anacaenini rather than with the tribe Chaetarthriini. Unconstrained MP analysis placed Guyanobius as sister of the Anacaenini with rather low statistical support. However, the topology is probably affected by the presence of Tormus and Paracymus. When these taxa are forced out of Chaetarthriinae in the constrained MP analysis (Fig. 13), the internal topology changes, and Guyanobius is revealed as sister of Crenitis + Crenitulus, deeply nested within Anacaenini. The same internal pattern appears with strong support in the unconstrained Bayesian analysis when Tormus is excluded (Fig. 14), indicating that (((Pseudorygmodus (Guyanobius (CrenitisCrenitulus))) is the most likely internal topology. Despite the fact that the results of these analyses do not agree with those obtained with the molecular data (Short and Fikáček 2013, Clarkson et al. 2019) some interesting characters can be discussed over the most parsimonious tree of the unconstrained MP analysis (Fig. 15).

The unusual morphology of Chaetarthria places it in a basal position, diverging early from the other taxa. This was expected since these larvae are unique within Hydrophilidae and display a high degree of modifications to the labroclypeus, stemmata, labium and legs, all of which are most likely related to a riparian lifestyle. For instance, these larvae present three unique apomorphies (Fig. 15): character 33(1), prementum incompletely sclerotized, this character is unique for Chaetarthria within Hydrophilidae, with the exception of the sphaeridiine genus Protosternum Sharp, in which second instar larvae have an incompletely sclerotized mentum, first instar larvae are unknown, and third instar larvae have a completely sclerotized mentum (Fikáček et al., 2018); character 39(1), a globular ligula, not found in other hydrophiloid genera; character 43(1), legs reduced, three-segmented, the only other genus with three-segmented larvae is Georissus, belonging to the hydrophiloid family Georissidae, a convergence. Other homoplastic characters worth mentioning are: character 3(1) stemmata closely aggregated forming one or two groups, shared with some Megasternini and Omicrini, although in these taxa the stemmata are clustered in two groups whereas in Chaetarthria the stemmata are aggregated forming only one; character 13(1) teeth of nasale dissimilar, this feature was coded for other hydrophilids but the configuration of the teeth of Chaetarthria is unique with the middle tooth being much larger than the others; character 58(1) FR7 rather long, present also in Crenitis + Crenitulus, Cylorygmus and Acidocerinae; character 68(1) seta FR12 posterior to pore FR13, shared with Guyanobius and only a few unrelated genera belonging to other subfamilies (Enochrus, Agraphydrus and Cylorygmus), a convergence; character 101(2) seta MN1 posterior to retinacular area, this character is unique within Chaetarthriinae, and is shared with only a few Sphaeridiinae genera (i.e., Phaenonotum, Cercyon). These features are strong enough to break the connection between Chaetarthria and the remaining genera of Chaetarthriinae; larvae with highly modified morphologies resulting from adaptations to live in new habitats generate topological problems in phylogenetic reconstructions (Archangelsky et al. 2021; Rodriguez et al. 2021).

The convergent larval morphology of Paracymus and Tormus with Chaetarthriinae (excluding Chaetarthria) also causes many problems in reconstructing the phylogeny of the group. Until quite recently, Paracymus was considered part of Anacaenini based on adult morphology (Hansen 1991, 1999). Subsequent molecular studies placed it close to the Laccobius-group in the subfamily Hydrophilinae (Short and Fikáček 2013). However, the position of the Paracymus-group (which also includes Tormus) of Laccobiini has always been problematic (see in figure 2 of Short and Fikáček 2013, the Paracymus-group is sister to the clade formed by Hydrophilini and Hydrobiusini). Furthermore, in an unpublished recent study based on larval characters (Rodriguez 2021), Paracymus and Tormus nest in a clade together with Chaetarthria and Guyanobius.

Paracymus larvae share several character states in common with Chaetarthriinae larvae. However, all of these characters are highly homoplastic and no unique synapomorphy can be mentioned. All of these features are shared with several genera of the family Hydrophilidae such as character 9(1) asymmetric nasale; characters 23(3) and 24(3) right and left mandibles with 3 retinacular teeth; character 61(1) pore FR2 inserted equidistant from FR1 and FR3; character 80(1) short sensilla PA20, character 92(2) pore MX3 in line with MX4 on the stipes; among others. In the case of Tormus, in addition to several homoplastic synapomorphies, there is a unique feature that supports the grouping with Chaetarthriinae (except Chaetarthria): character 19(1) surface of A1 with groups of fine cuticular spicules. Although not present in this work, this character state was also reported for Enigmahydrus and Saphydrus (Hydrophilidae: Cylominae) (see Seidel et al. 2020) and is likely to be a convergent character associated with semi-aquatic environments. Tormus larvae were described by Fikáček et al. (2013) and in many aspects are close to those of Paracymus. In the present study, Tormus is the most problematic taxon as it groups with Guyanobius in all the analyses, modifying the internal relationships of the clade, and does not nest with Paracymus. This could be related to the fact that while Paracymus larvae are aquatic Tormus larvae are not (Fikáček et al. 2013). In fact, Tormus and Guyanobius inhabit much more similar environments such as leaf packs lodged against rocks, logs in small streams near dense rainforest (Guyanobius) and moss within the forest (Tormus) (Archangelsky 1997, Fikáček et al. 2013) although Tormus does not inhabit extremely water-soaked environments as Guyanobius. Therefore, Tormus shows several morphological characters associated with adaptations to a terrestrial life style (i.e. reduction of ligula and dense pubescence on the antennae, maxillae and labium). However, the relationship between Tormus and Paracymus is strongly supported by molecular studies and also by adult characters (Short and Fikáček, 2013; Toussaint and Short, 2018).

In spite of having two alternative positions within the Chaetarthriinae, all analyses agree that Guyanobius appears to be more closely related to Anacaenini than to Chaetarthriini. This clade is supported by nine homoplastic synapomorphies in the unconstrained MP analysis (Fig. 15). Some features worth mentioning are: character 12(0), nasale with five teeth, this is shared with only a few other genera (Amphiops, Hydramara, Limnohydrobius and Cylorygmus), which belong to other subfamilies, a probable convergence; character 38(1) ligula longer than labial palpi, very long ligulas are unique for Chaetarthriinae, (only shared with Paracymus), but in Chaetarthria the ligula is slightly shorter than the labial palpi, related to a change in the shape of the ligula, which is round in Chaetarthria instead of elongated as in the other genera; character 46(1) abdominal segments with lateral conspicuous lobes, this feature is also found in Derallus but in this case the lobes are longer with several cuticular projections and is a convergence; character 106(1) setae LA1 short, only shared with few other unrelated genera (i.e., Sphaeridium, Oosternum and Protosternum); 118(2), ratio A1/A3 in third instar larvae, A1 much longer than A3, with a reversal in Crenitulus that has a slightly shorter A1.

The clade Pseudorygmodus + Crenitis + Crenitulus (Fig. 12) is defined by a unique synapomorphy and more than 10 homoplastic characters. The presence of pronotal lateral projections is a distinctive feature of this group, which is not shared with any other hydrophilid. Crenitis and Crenitulus are strongly supported by a unique synapomorphy, the posterior margin of abdominal tergite VIII trifurcate (character 48(1)), and 10 other homoplastic characters.

This paper provides a first insight into the relationships among the Chaetarthriinae based on larval morphological characters. The results of our work are affected by the low taxon sampling and the effect of evolutionary convergence. However, some interesting conclusions can be mentioned:

(1) Our study does not support the monophyly of Chaetarthriinae and neither of its two tribes. Larval characters place Guyanobius closer to Anacaenini than to Chaetarthriini, making further analysis necessary to test their tribal position.

(2) Larval characters are informative. Nonetheless, they are affected by derived or convergent morphologies related to different life strategies that generate topological problems.

(3) Different sources of characters are necessary (larval, adult, molecular) to generate robust phylogenetic hypotheses.

(4) Homoplasy partitioning seems to be an efficient strategy to analyze morphological data sets. The analysis implementing the homoplasy partitioning scheme proved to be more useful for solving some artifacts generated by convergent morphologies than the other alternative.