Blattodea. We follow Djernæs et al. (2020) and Evangelista et al. (2021) regarding the outline of Blaberoidea (i.e. excluding Anaplectidae), and regarding the division of Blaberoidea into the five families Pseudophyllodromiidae, Blattellidae, Ectobiidae, Nyctiboridae, and Blaberidae (i.e. we treat the four former subfamilies that together formed the family Ectobiidae in its older, wider sense as families); and we follow Djernæs et al. (2020) regarding the formal assignment of Attaphila to Blattellidae. This classification agrees with the results of recent molecular-based phylogenetic studies; but we note that for only few of the numerous genera currently assigned to one of the formerly “ectobiid” families this assignment is supported in a phylogenetic sense (taxa in Djernæs et al. 2020: table 4 plus taxa additionally sampled in Evangelista et al. 2021). Regarding the assignment of genera not included in recent phylogenetic studies to the five main classificatory units of Blaberoidea, we follow Beccaloni (2014; exceptions to this are specified below). To address the former “Ectobiidae”, which is occasionally needed for the sake of their shared plesiomorphies and for comparison with older literature, we use the term “non-blaberid Blaberoidea”.

Hymenoptera-Formicidae. Regarding genus- and species-level taxonomy of the reported host ants of Attaphila we follow the catalogue of Bolton (2021). Phylogenetic and evolutionary hypotheses are taken from Schultz and Brady (2008), Branstetter et al. (2017), and Cristiano et al. (2020), who used successively increasing taxon samples. Cristiano et al. (2020) find the leaf-cutting ants and its three genera monophyletic, with the relationships Amoimyrmex + (Acromyrmex + Atta), the genus Amoimyrmex having been newly defined therein (its species were formerly assigned to Acromyrmex).

Soft tissues were removed by treatment with 10% KOH at 40°C for 12 hours. For examination the cleared cuticular parts were either put in a petri dish (direct examination for drawings) or slide-mounted in Euparal using tiny glass rods as spacers between slide and cover slip (for photography). Slide(s) and the remnants of the corresponding specimens got an identification code (Xy or XY numbered) specified in the ‘Material studied’ paragraphs of the species descriptions and in the figure captions; the letter combination Bo is used for material not belonging to the collection of H.B., all other combinations indicate the country of origin: Al Algeria, Cb Colombia, CR Costa Rica, Ma Morocco, Sp Spain.

Regarding photography, the phase contrast images were made with a Sony Nex-5N camera on a Zeiss Photomikroskop II, all other photos were made with a Jenoptic camera (ProgRes SpeedXTcore5) on a Leica microscope (DM 5000B) using software ProgRes CapturePro v.2.8.0 and Helicon Focus 5.3. For drawings, the preparations were examined under a Leica M125 stereo microscope and gradually dissected; initial handmade drawings were scanned and then completed using the computer programs CorelPhotoPaint and CorelDraw. In the figures the orientation of the structures is – unless otherwise stated – with the anterior end on top, or with the base on top (antennae, legs, tegmina); tergites shown in dorsal, sternites in ventral view. For legs and tegmina morphological orientations are given as if they were stretched at right angle from the longitudinal axis of the body towards the side.

Armament of tibiae. The distribution of spines on fore-, mid-, and hindtibia is – as hitherto (see e.g. Bohn et al. 2010) – specified by the following formula: [d·a·v][d·a·v][d·a·v]. Compared to the numbering system for tibial spines introduced by Klass et al. (2009), the explanation of the letters is now read as follows: d number of spines on the dorsal surface outside the apical armament (spines Td excluding Td1m), a number of spines of the apical armament (terminal tibial spines Tt1‒5 plus distidorsal spine Td1m), v number of spines on the ventral surface outside the apical armament (spines Tv in Klass et al. 2009).

Borders of tergites. In the preparations of successive abdominal tergites (e.g. Fig. 6) there are many transversal lines of different kind and distinctness; as some are important in the descriptions, the pattern is briefly explained and illustrated in Supplement 2 (Fig. S1 and associated text). The taxonomic descriptions consider mainly the following lines (n representing the sequential number): The posterior borders of tergites (posterotergal bending lines Tn-p, fixed and discrete); the anterior borders of tergites (anterior margins of tergites Tn-a, fixed but rarely discrete); the lateral borders of tergites (lateral bending lines of tergites, fixed and discrete); and the tergal transversal ridges (trn, fixed and usually discrete). Note that the anteroposterior succession of the transversal lines is not always regular due to a longitudinal shift of part of the series of tergites (as evident from Fig. S1D).

Bristles on tergites. Bristles can be present along the transversal ridge (trn), along the lateral and posterior borders, and on the surface area in between. Those in between are called ‘surface bristles’. The center of a surface area is its middle part both in the longitudinal and the transversal direction.

Male and female genitalia. Selecting a terminology and associated abbreviations is problematic for both sexes. There are, on the one hand, simple terminologies that have been used in recent taxonomic contributions on Blaberoidea, e.g. that of H. Bohn (various papers mainly on Ectobiidae; both sexes: e.g. Bohn 2004; Bohn et al. 2010; Bohn and Chládek 2011; Bohn et al. 2013; Bohn 2019). Their abbreviations are quite arbitrarily designed, as their goal is just cross reference between text and illustrations. However, only the few structures evaluated for taxonomic purposes are named. On the other hand, there is the more elaborate terminology of K.-D. Klass (males: mainly Klass 1997; females: various papers on non-dictyopterans, e.g. Klass and Matushkina 2018; both sexes: Brannoch et al. 2017 for Mantodea). It has the additional goal to express homology hypotheses throughout Dictyoptera or Insecta, and homonomies among segments. Its abbreviations are designed according to a coherent system (e.g. by using different kinds of terms for sclerotisations and elements of shape, such as processes), and the abbreviations actually constitute the terminology. This complex terminology provides names for most elements of the genitalia. However, it has not yet been applied to a broader sample of Blaberoidea, where some homology problems need to be resolved prior to its broad application to this taxon. To cope with this conflict, we apply herein a mixture of the terminologies used by Bohn and Klass. The synonymy between the two is given in the text at first mention and is surveyed in Supplement 3 Fig. S2 (female; synonymy with abbreviations in McKittrick 1964 additionally indicated) and Supplement 4 Fig. S3 (male). The terminologies of Klass are explained, with a focus on Mantodea, in Brannoch et al. (2017: pp. 28‒30, figs 14, 15, supplement 9).

For comparing any body parts between Attaphila and other cockroaches (especially Blaberoidea), we used a variety of taxonomic papers, focally those of H. Bohn, to the extent these include relevant information; and we used several morphological treatments (such as Wipfler et al. 2016 on the head of Periplaneta americana). For the antennae we provide illustrations based on own studies on some Blattellidae species (Fig. 2).

For the genitalia, which are only superficially described in most of the taxonomic literature, we additionally used morphological contributions. The main data source for female genitalia is McKittrick (1964, abbreviated MK64 in the following), where a fairly rich selection of Blaberoidea is covered; in her drawings, however, many spatial relationships between structural elements are unclear, which makes comparison difficult. In addition, the very limited information in Klass (1998: Supella being the only sampled blaberoid) was used; and Brannoch et al. (2017) was taken for interpretation at the Dictyoptera level. The main data sources for male genitalia are MK64 (with the same problems as for female genitalia) and Klass (1997), where cockroach phallomeres are described in great detail, but only for very few blaberoid species. In addition, for some crucial points we provide illustrations from own preliminary studies on genitalia of selected Blattellidae and Ectobiidae (Figs 30–33).

Our own examinations in taxa apart from Attaphila refer to: the BlattellidaeBlattella germanica (Linnaeus, 1767) (ex cult.), Blattella lobiventris (Saussure, 1895) (Gabon), Loboptera decipiens (Germar, 1817) (Spain), Symploce pallens (Stephens, 1835) (ex cult.), Xestoblatta cantralli Fisk and Gurney, 1968 (Costa Rica), Xestoblatta hamata (Giglio-Tos, 1898) (Costa Rica), Pseudomops Serville, 1831 sp. indet. (Mexico), Ischnoptera Burmeister, 1838 sp. indet. (Costa Rica), Lobopterella dimidiatipes (Bolívar, 1890) (ex cult.), and Parcoblatta lata (Brunner v. W., 1865) (USA); and the EctobiidaeEctobius lapponicus (Linnaeus, 1758) (Germany), Dziriblatta haffidi (Bolívar, 1908) [taxonomic status according to Bohn 2019: p. 18] (Morocco), and Dziriblatta kroumiriensis (Adelung, 1914) [taxonomic status according to Bohn 2019: p. 11] (Algeria).

Morphological terms. All abbreviations are listed in Supplement 1; those used in the figures are additionally listed in the associated legends. The abbreviations T + number (abdominal tergite) and S + number (abdominal sternite; S7 = subgenital plate of female; S9 = subgenital plate of male) are frequently used in the text; terms like T6,7 and S1–5 refer to two or several, respectively, tergites or sternites, as indicated by the numbers. Lower-case n in italics is inserted in morphological terms to address all numbered elements in question.

Type specimens. HT – Holotype; LT – Lectotype; PT(s) – Paratype(s); ST(s) – Syntype(s).

Larval stages. L – Larva, larval (L- early larval stage, L+ late larval stage).

Museums and collections. Below we use abbreviations including the full name of the city (usually following M. = Museum), but here we additionally list the acronyms suggested by Evenhuis (2016). AMNH, M. New York – American Museum of Natural History, New York (USA); MACN, M. Buenos Aires – Museo Argentino de Ciencias Naturales “Bernardino Rivadavia”, Buenos Aires (Argentina); MTD, M. Dresden – Museum of Zoology, Senckenberg Natural History Collections Dresden, Dresden (Germany); MZSP, M. São Paulo – Museo de Zoologia, Universidade de São Paulo, São Paulo (Brazil); NHME, M. Maastricht – Natuurhistorisch Museum, Maastricht (Netherlands); RMNH, M. Leiden – Naturalis Biodiversity Centre, Leiden (Netherlands); TAMU – Department of Entomology, Texas A&M University, College Station, Texas (USA); USNM, M. Washington – National Museum of Natural History (Smithsonian Institution), Washington (USA); ZSM, ZS Munich – Zoologische Staatssammlung München, München (Germany).

Figs 21D, 28

Size very small, 2.5–3.5 mm long. Body rather stout, in dorsal view wide-oval, with strongly vaulted thoracic dorsum. Surface of pronotum, tegmina, and abdominal tergites up to T5 loosely covered with rather long and thin bristles. Colouration almost uniform, in larvae yellowish, in imagines slightly darker, orange-brown; legs always darker than the remaining parts of the body.

Figs 1A–D, 2

Head capsule in frontal view rounded-triangular (Fig. 1A‒C), relatively short, in lateral view not compressed, with well-rounded occiput and frons. The lateral part of the head capsule between the anteroventral margin of the compound eye and the lateral margin of the clypeus forms a rather wide and deep funnel-shaped pit (Fig. 1A, D: cap; Bolívar 1901: p. 334, pl. 6); dorsal margin of the pit forming a fairly sharp edge (capd), walls of the trough otherwise gradually passing over to the surface of the head capsule. Ventromesal half of the pit bearing the antenna insertion, with the base of the scape being surrounded by a fairly wide articulatory membrane and approached by a tongue-shaped antennifer from ventrally. Epistomal ridge absent except for lateralmost parts. Coronal (cc) and frontal (fc) cleavage lines distinct.

Compound eyes (Fig. 1A, D: cpe) placed laterally, very small, not prominent from overall outline of head, with not more than about 70 ommatidia (see also Wheeler 1900). Ocelli missing.

Antennae (Figs 1A, C, D, 2): Scapus (sc) relatively long, with a distinct bend of ca. 70° at its very base (in anterior view: Fig. 1D); pedicellus (pc) and few basal flagellomeres (fl1 and following) rather short; the distally following flagellomeres rapidly increasing in length, reaching their maximal length about at the level of the 7^th flagellomere (ca. 3.4 × as long as wide; Figs 1A, 2F, G); diameter of flagellomeres slightly increasing up to the 6^th or 7^th, then slightly decreasing again; shape of flagellomeres conical, widening towards their apical end. Number of flagellomeres unknown: in imagines in most cases 7–10 (only one specimen found with 11) were present, but all antennae appear as being incomplete (Wheeler 1900; Bolívar 1901, 1905; Brossut 1976), since an intact terminal flagellomere with apically closed cuticle has never been observed (but see larval development in 3.13.). Retained distal flagellomeres according to Brossut (1976: figs 4, 5) provided with a relatively low number of sensory bristles. Hebard (1916) reports that “the joints beyond the first [scapus] are carried normally at a decided angle to it”. Among the specimens available to the authors several showed an almost rectangular upward deflection of the pedicellus versus the scapus (Fig. 1C); the strong deflection is enabled by the scapus having distally a distinct rounded dorsal excavation (Fig. 1D: scex, which is flanked by the two usual scapo-pedicellar articulations, one visible in Fig. 1D: sc-pc). Two fields of sensilla basally on the scapus and one basally on the pedicellus are likely comprised of sensilla chaetica A (Fig. 1D: s-chA), and a few sensilla distally on the pedicellus are likely representatives of a circumferential row of sensilla campaniformia (Fig. 1D: s-cpf; compare Drilling and Klass 2010: S-VL, S-DL, P-D, and oval symbols in fig. 5).

Mouthparts (not studied in detail) with mandibles (md) and laciniae (li; Fig. 1B) shaped as typical for cockroaches (see Wipfler et al. 2016); mandibular dentition asymmetrical as usual in cockroaches: left mandible with 4 teeth, right one with 3 teeth (Fig. 1B inset: I‒IV, including tip and incisivi; compare Wipfler et al. 2016: fig. 9E, F). Maxillary palps with 5, labial palps with 3 palpomeres (I‒V of mp, lp in Fig. 1B), the apical one in both cases the longest and widest, and, according to Brossut (1976: figs 6–9), on the ventral surface densely covered with sensory bristles; mpIV with a distinct basal bend.

Tentorium (Fig. 1A, C, D) of typical blattodean structure (compare Klass and Eulitz 2007: figs 2–8), with anterior transversal bridge (atb) and perforation (tp) behind it; origin of anterior arms (ata) from head capsule in typical position but very narrow (Fig. 1C, D).

Fig. 1E, F

Pronotum (N1) almost completely concealing the head, in dorsal view rounded-trapezoidal, narrowing towards the anterior, with almost straight anterior border. Meso- (N2) and metanotum (N3) in females (in dorsal view) with fairly straight anterior and posterior borders and widely rounded anterolateral corners; in males more or less trapezoidal, narrowing towards the posterior (Fig. 1G, H).

Figs 1G, H, 3

Males with tegmina (Figs 3, 28A‒E) short, apically either transversally or obliquely cut (i.e. either at right angle or from anterodistally to posteroproximally, relative to longitudinal axis of wing), posteriorly scarcely surpassing the metanotum, mesally reaching the thoracic midline, without any venation, loosely covered with long and thin bristles; hindwings (Fig. 1G, H) only consisting of tiny lobes of about half the length of the metanotum. Females without wings (Figs 1E, 21D).

Fig. 4

Rather short and stout. Each coxa with a distinct coxal lobelet (colb) on its distal border (as in most or perhaps all Dictyoptera). Femora (fe) and tibiae (ti) anteroposteriorly compressed. Femora at the base with steeply increasing height (dorsoventral extension). Anterior and posterior walls considerably protruded beyond the narrow ventral surface (edges fane and fpoe), thus forming a proximally flattening groove (femoral groove fegr) which can take up part of the tibia during a strong flexion (Fig. 4A). Apical dorsal fold of tibia (dft; at bases of spines Tt2, Tt3 in Fig. 4E, F) virtually absent (compare dft in Klass et al. 2009: figs 1, 2). Tarsi (ta) rather stout, with five tarsomeres (Fig. 4H of right tarsus; four tarsomeres with vestigial dorsal separation of tarsomeres 1 and 2 in left foreleg, Fig. 4D, G, likely result from regeneration after loss), tarsomeres 1–4 fairly cylindrical, without basal constriction, tightly closed together, borders between them oblique, without euplantulae, ventroapically supplied with small spines in a transversal row. Pretarsus consisting of two symmetrical unspecialised claws (ptcl) and a large arolium (ptar) in between.

Spine armament of femora. Forefemur (Fig. 4D) with only one apical spine, positioned at the anteroventral edge (fane) and proximally followed by a row of more or less strong setae (Type D1, Roth 2003). Midfemur (Fig. 4C) also with only one spine, but apically at the dorsal surface (gs = genicular spine). Hindfemur (Fig. 4A, B) also with one genicular spine and 2–4 spines at the anteroventral edge (fane); one of the latter always near midlength of femur, the others at some distance near apex.

Spine armament of tibiae. Spine formula (for explanation see section 2 and Table 1) [0·5·0][4‒5·6·1][6‒9·5‒6·0‒1]. Apical armament (as) of mid- and hindtibiae (Fig. 4B, C, F) as typical for cockroaches with 5 terminal spines (Tt1–5), all in typical positions (compare in Klass et al. 2009: figs 1, 2 for Blaberus), and a far distally placed middorsal spine (Td1m); in foretibia (Fig. 4D, E) with 4 terminal spines (Tt2–5; spine Tt1 missing) and a distal anterodorsal spine (Td1m^a). Foretibia with no dorsal (besides Td1m^a) and no ventral spine; midtibia with 4–5 dorsal (in addition to Td1m) and 1 ventral spine; hindtibia with 6–9 dorsal (in addition to Td1m) and 1 ventral spine. The numbers of spines on femora and tibiae show fairly wide ranges of variation within the species combined with much overlap among the species; spine armament is therefore unsuitable for species identification.

Figs 5–13

Shape. T1,2 with weakly convex, T3–6 with fairly straight posterior border (Tn-p). T7 of females (e.g. Fig. 6E) in the median half with a short, wide lobe-like posterior expansion, whose posterior border is medially slightly convex or concave, or straight; posterior border T7-p laterally of the lobe concave; transversal ridge tr7 always distinct and in parallel with the posterior tergal border T7-p. T7 of males (Fig. 6A) with median lobe less prominent, transversal ridge tr7 usually distinct, but in some species weakly developed (Fig. 13D) or completely missing (Fig. 13E, F). T8,9 (Figs 6B, 7C) in both sexes rather short, weakly sclerotized, concealed below the preceding tergite T7. T10 (Figs 6B, 19A) rather short, with widely rounded posterior border T10-p; the lateral parts T10p bending to the ventral side (to meet the paraprocts, PP) are very narrow (Fig. 5J, K, arrows point to contact between T10p and PP).

Distribution of bristles. T1–5 usually loosely covered with long and thin bristles; the males of A. aptera and A. bergi on T1 without such long bristles. They are in all species arranged in one line along the lateral and posterior borders. The distribution of the remaining bristles on the surface, between the transversal ridge trn and the posterior border Tn-p, is species-specifically different: either in only one distinct transversal line (Fig. 9B, E), in two very irregular transversal lines (Fig. 7A), or more or less irregularly dispersed (Fig. 8C). The bristles along the lateral borders are usually slightly shorter, but stronger than those along the posterior border and on the surface; the bristles along the posterior border of T5 are often more densely arranged than on the preceding tergites (Fig. 7A, D). T6,7 in both sexes along the lateral borders with similar bristles as on the preceding tergites. Size and arrangement of bristles at other places of T6,7 different in the two sexes: Females (Figs 12A–F, 13A, B) always without bristles along the posterior border of both T6 and T7. Transversal ridge tr6 usually with several bristles of small or medium size, tr7 with only two rather small bristles at a distance of about ¼ of tergite width. Surface behind the ridge provided with bristles species-specifically varying in size and number; the bristles are usually arranged in a wide median transversal stripe of varying lateral extension (ranging from slightly less than ½ to about ⅔ of tergite width), on T7 usually in lower density and extension. Males (Fig. 13C–F) with much smaller bristles along the transversal ridge, bristles on the surface or along the posterior border present or absent.

Male tergite glands. Glandular pores occur on T1–5 in the area anteriorly of the transversal ridge, mostly rather dispersed, but in A. aptera in extremely high density (Fig. 5A, B); they are usually tiny, larger ones are found in and near the specialisations on T2 (msl2).

Male tergite specialisations. The males of A. flava, A. fungicola, and A. paucisetosa have a pair of specialisations laterally at the anterior border of tergite T2, each consisting of a shallow transversal trough with a mesolateral extension of about ¼ of tergite width (Figs 9B, 5E–I: msl2). The bottom of each trough shows a more or less complicated relief generated by rather low, rounded ridges crossing the trough. The males of A. aptera and A. bergi have a specialisation medially on tergite T1. In A. aptera (Figs 6A, 5A, B: msp1) this is a small, fairly rounded, weakly sclerotized area with two groups of relatively long bristles pointing anteriorly, located immediately posterior to ridge tr1. In A. bergi (Fig. 5C, D: msa1) the specialisation consists of a pair of small areas in the anterior part of T1 showing a net-like pattern produced by delicate furrows (for the identification as furrows rather than ridges see explanation in Supplement 6 Fig. S4), along which tiny glandular pores are sporadically arranged; specialised areas occasionally with few small bristles (Fig. 5D). The net-like pattern of the specialisation is strongly emphasised microreticulation, which is continuous with much less emphasised microreticulation further posteriorly on T1. T1 in the former three species and T2 in the latter two species without specialisations. T1 in the latter two species additionally characterised by the abovementioned absence of long bristles on surface and lateral and posterior borders.

Tergite T9. In both sexes T9 and T8 are very short and entirely hidden as they are overfolded by the hind part of T7. Dorsolaterally the anterior border of T9 of both sexes forms on each side a distinct semicircular apodeme (ltga9, for males in Figs 6B, 7C; schematic view in Fig. 28G). The ventrally bent lateralmost part of T9 (paratergal part T9p) is narrowed towards the anterior, its terminal part forming an anteromesally directed, slightly mesally curved sclerite arm running along the anterior border of segment 9 (paratergal extension pt9 of male, pt8,9 of female). The tip of this arm closely approaches the lateral margin of S9 in the male, forming a loose articulation (A1) with it (Figs 24A, 25A‒D, 29C, H); in the female it approaches the lateral gonangulum sclerite (gg-l) if this is present, forming a close articulation (A1) with it (Fig. 19A; for further contacts of the female pt8,9 see 3.9.). In both sexes the paratergal extension is strengthened by an internal ridge, which is part of the antecosta of segment 9 (ac9, for male see inserted section in Fig. 29G). We call the arm of the female pt8,9, as generally in Dictyoptera the posterior part of T8 contributes to this structure (although with varied clarity in different taxa, and not resolved for Attaphila; Klass 1998: figs 11–18; Brannoch et al. 2017: TG8+9ε in fig. 14C). In the male we call the arm pt9, as there is no indication of a contribution from T8.

Very short, without any annular divisions; dorsal surface almost plane, smooth, lateral and mesal flanks of cerci not visibly depressed to form a keel (compare Lobopterella in Fig. 29D); bristles and sensilla mostly restricted to the vaulted ventral surface. Outline in ventral or dorsal view egg-shaped (males of all species, Fig. 6B, and female of A. schuppi, Fig. 20C, D); or asymmetrically widened (more strongly laterally) and wider than long (remaining females, Fig. 19A).

Female subgenital plate S7 (Figs 16–18). Anterior part without apodemes. Posterior part located in ventral wall of subgenital lobe (expanded ventral fold vf7; with no delimitation of the lobe in the ventral segmental wall on S7; see vf7 in Fig. 16A and compare MK64: fig. 40A). Subgenital lobe in all species with three short rounded apical lobes, a very wide median one and two much narrower and slightly shorter lateral ones. S7 either semicircular (A. sexdentis, Fig. 17F) or rounded-rectangular (remaining species, Fig. 17B). In semicircular type anterior border strongly curved and lateral borders anteriorly converging, thus all together forming an arch. In rounded-rectangular type anterior border less strongly curved and lateral borders parallel, thus all together being quite rectangular. Subgenital plate towards its anterior border with rather weak, gradually fading sclerotisation; anterior outline in the figures, therefore, not always well visible. The transversal sternal ridge (sr7 in Figs 16, 17), starting latero-posteriorly at the lateral base of each lateral lobe, forms a wide anterior curvature; lateral parts (sr7-l) very steep and reaching far to the anterior (yet converging), either fairly straight (Fig. 16A) or more or less strongly curved mesad around their midlength (Fig. 17A–D), near the anterior border of S7 continuing into the transversal median part of the ridge (sr7-m in Fig. 16A). Median part either continuous across middle (only A. bergi, Figs 16A, B, 18C, D), or with some traces of discontinuities (arrows in Fig. 18E, G), or with a distinct gap of varied width (between bars in e.g. Fig. 18A, I, J). The median part sr7-m of the ridge is best examined at high contrast, because with low contrast parts of it can be difficult to recognise (compare Figs 17E and 18G, which were made from the same object). The different course of the anterior border of S7 in the two types correlates with a different extension of sclerite S7 beyond the lateral parts of the transversal ridge in anterolateral direction: it is very wide in the rounded-rectangular type, but rather limited in the semicircular type with anterior border and transversal ridge running almost in parallel at short distance. Surface of S7 in the posterior 2/3 covered with dispersed rather long and strong bristles, especially densely arranged along the posterior border.

Male subgenital plate S9. Anterior part with a pair of rather long, slender, and strong apodemes (sta9) of about equal length (Figs 24A–D, 25B, C). Posterior part located in ventral wall of subgenital lobe (expanded ventral fold vf9). Subgenital lobe in all five species with males known (Figs 24–26) with a deep excavation along the left side, the conical left stylus (sl^l) inserted at the base of the excavation, not reaching tip of lobe; the more strongly projecting right part of the subgenital lobe tongue-shaped. In A. aptera and A. bergi right part of lobe widely tongue-shaped, without excavation on right side; a small knob-like right stylus (sl^r) present, situated subterminally on right flank of tip (Fig. 24A–D). In A. flava, A. fungicola, and A. paucisetosa right part of lobe narrowly tongue-shaped (and slightly curved towards the left) due to an excavation on right side, which is of similar depth as the excavation on the left side; right stylus absent (Fig. 25A–D). Due to the presence of an excavation on only one side the subgenital lobe appears very asymmetrical in the two former species, whereas due to the presence of an excavation on each side the lobe appears quite symmetrical in the three latter species. On each side the lateral margin of the subgenital plate articulates with the ventral extension of tergite 9 (pt9, e.g. Fig. 24A; see 3.6.).

Male paraprocts. Right paraproct (PP^r) of A. aptera (Figs 5J, 6B) mesally with a sclerotised hook-like projection (hmp), other species with known male without such a differentiation (Figs 5K, 7C, 8B).

Overall structuring largely as typical for Blattodea: There are two cavities in anteroposterior succession, i.e. a large posterior vestibulum (space above subgenital lobe vf7), which continues anteriorly into a narrower genital chamber. The elements of the female genitalia are distributed over the upper and lower walls of these cavities. Problematic interpretations are discussed in Supplement 5.

The genital chamber (gc) is divided in a dorsal and a ventral subchamber by a flat transversal fold arising from the anterior and lateral walls of the genital chamber (genital chamber fold gcf in Fig. 23A, B, its posterior edge labelled gcf in Figs 19A, B, 21A; fold in same position as the one bearing ‘sp.pl.’ in MK64: fig. 40A of Supella, but much deeper); fold gcf is asymmetrical, projecting further posteriorly on the left side. The gonopore (opening of common oviduct oc) lies in the anterior wall of the ventral subchamber (Fig. 23A, B); there is no genital papilla, but the oviduct widens quite gradually and continues into the lumen of the chamber. The spermathecal plate (sp; SP in Supplement 3 Fig. S2B) lies in the dorsal wall of the fold gcf (Fig. 23A); like the fold it usually exhibits a distinct asymmetry (with a left-side focus in Figs 23A, S2B); a division was not observed. We did not find any paired or unpaired cuticular structures that could reasonably be considered spermathecae, neither on plate sp, nor in any other position. The anterior wall of the dorsal subchamber forms a folded, anteriorly directed pouch (genital chamber pouch gcp in Figs 19–21, 23A) on the side opposite to where the spermathecal plate has its focus.

The left and right valvifers (vlf = part of 8^th-segmental coxal sclerites CX8; Figs 19–21, 23A) in the roof of the genital chamber (gc^d) strongly converge anteriorly, where they are connected across the midline, forming together a single arch-shaped sclerite. The posterior ends show a discrete contact (articulation A5) with the paratergal extension (pt8,9), and the adjacent part of vlf is curved laterally (often showing some asymmetry). The anterior part traversing the midline appears as a discrete ribbon-like continuation of the posterolateral parts in some species (Figs 19A, B, 21B), but is indistinctly delimited, weaker and wider, and perhaps incomplete in others (mesad of arrow in Fig. 20B, D; a distinction between valvifer arch and spermathecal plate, which are placed one above the other in a preparation, is then partly difficult). Note that the area where the valvifer arch crosses the midline is placed morphologically posteriad of the spermathecal plate. Individualised basivalvulae (part of 8^th-segmental coxal sclerites CX8) were not found; these sclerotisations could be included laterally in the sclerite here called vlf, or in the sclerite ls (see below), or be absent (discussion in Supplement 5). In some species the central dorsal wall of the genital chamber (gc^d) bears a microsculpture of small knobs (Figs 19A, 21A, B, shown enlarged in inserts), possibly associated with very weak sclerotisation that appears medially divided (putative mesal border shown by arrow in Figs 19A, 21A, B and their inserts).

The 1^st valves (v1 = 8^th-segmental gonapophyses gp8) show the usual configuration, with their bases (including the basal sclerotisation GP8) reaching far laterally to join articulation A5 (e.g. Fig. 19A).

Of the gonangulum (gg = 9^th-segmental laterocoxal sclerites LC9) the mesal part (gg-m in Figs 19A, B, D, 20B, D, 21A, B) is distinct; it forms the typical articulations A2 (with the posterior lobe pl, see below; Fig. 19A, B) and A3 (with the gonapophyseal sclerotisation GP8 at the dorsal base of the 1^st valve; Fig. 19A). The lateral part (gg-l) forming a hinge-like contact A1 with the paratergal extension pt8,9 (see 3.6.) is present in A. aptera (Fig. 19A; lateral part of LC9 in Fig. S2A, B), where it is completely separated from the mesal part, but appears to be absent in the other species (Figs 19B, D, 20B, D, 21A, B).

The anterior arch (aa = anterior part CX9µ of medially fused 9^th-segmental coxae CX9, compare Fig. 19A and Supplement 3 Fig. S2C for its outline) usually has a darker anterior margin, possibly due to a transversal internal ridge. The shape of the anterior border of aa appears to vary among species, being straight, biconcave, or convex to a varied extent (compare Figs 19A, B, D, 20B, D); however, its shape could be influenced by the angle of view upon the preparation. The posterior lobes (pl = posterolateral parts CX9β of 9^th-segmental coxae CX9, see Supplement 3 Fig. S2A, B) are well developed.

The 2^nd valves (v2 = 9^th-segmental gonapophyses gp9) and the 3^rd valves (v3 = 9^th-segmental gonoplacs gl9) overall show the usual configuration, but their structural details, especially those near the base, are not seen in the preparations due to the overlapping of several elements in the area.

Intercalary sclerites (IC in Fig. 19A, B) are very weak, often indistinctly delimited, limited to the median area, likely medially fused, and close to the paraproct anterior border.

The floor of the vestibulum (= dorsal wall of subgenital lobe vf7; vfl in Fig. 22A, B, 23A) appears to be entirely membranous. It bears membranous folds (which are part of vfl): a pair of longitudinal intersternal folds (isf in Figs 22D, H, 23A) and a transversal ventral vestibular fold (vtf in Fig. 23A) between them. When the membranous floor of the vestibulum is cut off from the sclerotised, stabilising ventral wall of the subgenital lobe, sternite S7, the folds tend to get distorted or to collapse (as in most pictures of Fig. 22).

The laterosternal-shelf area represents the posterior floor of the genital chamber adjoining the floor of the vestibulum. A large W-shaped laterosternal-shelf sclerite (ls in Fig. 23A, halves of W open posteriorly; LG7 + LC8? in Fig. S2E) extends over this area, anteriorly and laterally of the isf folds. The middle part of sclerite ls is U-shaped (U open anteriorly, i.e. the middle peak of the W is rounded or truncate), consisting of a central arch (ls-c) and lateral arms (ls-a in Figs 22A, 23C). The elongated, oblique lateral parts, the wings (ls-w in Fig. 22A), have a plate-like anterior portion, but extend far posterolaterally, where they become much narrower; the apical parts (ls-p, possibly the “posterior extensions” sensu MK64, then part of laterocoxa LC8) are twisted relative to the wing part ls-w (black arrows in Fig. 22A, D, I). Where the middle and lateral parts of sclerite ls approach each other, a deep, anteriorly directed tube-shaped pouch is present on each side, the laterosternal-shelf tube (lst in Figs 22A, 23). The pouch is rolled up and thus has a C-shaped cross section in its anterior part (lowest cross section left of Fig. 23D: the black margins of the C represent the cuticle, the body of the C is external world, the areas surrounding the C – including the area embraced by it – represent the interior of the animal). Both the arm and wing parts of sclerite ls extend into the tube walls and bend along them (shown in Fig. 23C–F and G–J), whereby much of the lst walls are sclerotised; ls-a and ls-w are likely synsclerotic inside the tube (at edge indicated by grey arrows in Fig. 23E–H; but a clear observation was not possible). A laterosternal shelf, i.e. a physical step upward between the floors of vestibulum and genital chamber (see MK 1964: figs 2, 10, 40b, representing the 7^th-segmental genital lobe) is absent.

The laterosternal shelf area shows considerable interspecific variation and can, therefore, serve as an important means for species distinction in the female sex. This concerns the shape of the central sclerotisation ls-c and of the tubes lst (Fig. 22), the anterior extension of the wing part ls-w on tubes lst (blue arrowheads in Fig. 22), and the anteroposterior position of the area where the anterior margin of the arm part ls-a bends from the ventral inner wall of the tube into the dorsal one (red arrowheads in Fig. 22; often associated with a laterally directed angular bend or kink). The shape characteristics of the tubes lst (as seen in preparations: Fig. 22) appear variously reliable due to the composition of the tube walls of sclerotised and membranous parts. For instance, the (inner) lateral border of lst is sclerotised and thus stable anterior to the red arrowheads (reliable), but membranous and thus flexible posterior to them (not reliable; area indicated by black arrowheads in inserts of Fig. 22B, D, E, K); the distinctness of the angular bend depends partly on the mesal bending of the posterior part (compare left and right sides in Fig. 22I, L) and is thus not a very reliable character.

In situ, the lateral wing parts (ls-w) of sclerite ls are positioned beneath the area embraced by the lateral parts of the valvifer arch (vlf), but extend further posterolaterally beneath the paratergal extensions (pt8,9; Fig. 20A, compare labelling on left and right sides). The central part (ls-c) is then placed beneath the anterior arch (aa; compare positions of ls-c in Fig. 20A and aa in Fig. 20B), and the arms ls-a and tubes lst reach anteriorly well beyond the anterior bottom of the dorsal genital subchamber. The intersternal folds (isf in Fig. 22D), which follow behind the ls-c part (upward in Fig. 20A), are in the right place to embrace the group of valves located above them in the roof of the vestibulum, and to form a mould for a new ootheca built in the vestibulum. The case where the central part (ls-c) is placed further posteriorly beneath the central apodeme, and where the arms (ls-a) and tubes (lst) do not exceed the dorsal genital subchamber (Fig. 21C) could be due to artificial shifts during dissection.

The ovarioles of Attaphila fungicola are described by Roth (1968: fig. 17) as being similar to other non-blaberid Blaberoidea (“Blattellidae” therein), with only one oocyte showing incorporation of yolk material.

A female carrying an ootheca was only once observed, among the specimens of A. paucisetosa collected by one of the authors (R.R.G.) in a nest of Atta cephalotes in Colombia (Fig. 21D). The ootheca appeared scarcely sclerotised, with a very low brownish keel (otkl), and contained five eggs; their upright orientation and the dorsal position of the keel signalise that the ootheca was not rotated. Since the female was fixed shortly after its capture, the question of a possible rotation of the ootheca before its deposition could not be resolved. With the very soft sheath and the low keel the ootheca resembles that of ovoviviparous species. These features can be seen as an adaptation to the certainly moist atmosphere in the mushroom chambers of the ants, which makes a strong hardening of the sheath unnecessary. A weakly developed keel was also described by Roth (1971: fig. 81) for the ootheca of A. fungicola.Waller and Moser (1990) placed alates of Atta texana with attached A. fungicola females in jars. Within few days the females produced oothecae, which they deposited at the bottom the jars. Unfortunately, it is not noted whether the oothecae were rotated prior to their deposition.

All interpretations of structural components are unproblematic (i.e. there are no major homology problems relative to other Blaberoidea).

Left phallomere. Hook (h in Fig. 24A, E; process hla bearing L3 sclerite in Supplement 4 Fig. S3A, B) fully retractable due to a long membranous proximal part, which is inverted in the retracted condition (part p, inverted in Fig. 24E, everted but only a short part included in Fig. 25E); sclerotised distal half with a wide basal part (b), a much more slender, variously widely curved intermediate part (n neck), and a claw-shaped apical part (cl) bearing an anterior groove (hge) with a cleft (hcl). Endophallic apodeme (ea in Fig. 24A, F; apodeme lve bearing L2D sclerite in Supplement 4 Fig. S3A, C) long rod-shaped, anteriorly widened. Base of apodeme associated with two posteriorly directed sclerotised processes (Fig. 32F–I), sclerotisation (L2) forked to cover both of them. The left branch of L2 is essentially limited to the virga process (vi in Fig. 32F–I; process via bearing the compound sclerite L2E+L4N), which arises at the L2 fork, is narrowed to a more or less acute apex, and is variously curved; in most preparations one or two longitudinal grooves are apparent (vge in Figs 24A, B, 25A, D; vge1, vge2 in Fig. 32F–I; compare Klass 1997: vge in fig. 273), but their extension, structure, and occurrence in the various species remained quite unclear. The tongue-shaped right branch of L2 extends posteriorly, its right-posterior parts being located in the dorsal wall of the angular or rounded, rightward-directed process psa. The sclerotisation of the virga (L2E) is probably not separated from that of the apodeme (L2D) by an articulation (A10; the apparent separation only in A. aptera seems to be due to a brighter area placed beneath, marked as A10? in Fig. 32F and Supplement 4 Fig. S3A). Opening of ejaculatory duct not unambiguously detected.

Right phallomere. R3 sclerite slender, elongate (Fig. 24A, G), the anterior (a), ventroposterior (v), and dorsoposterior (d) portions are narrowed to arm-like extensions. The short ventroposterior arm is associated with the cleft sclerite (cs; compound sclerite R2+R1S in Supplement 4 Fig. S3A, D), but the articulation was not clearly observed (compare Klass 1997: A7 in figs 282‒284). The longer dorsoposterior arm is distinctly articulated (A3 in Fig. 24G) with the curved dorsal sclerite (R1P) extending along the posterodorsal lobe of the phallomere. The dorsal part R1S of the cleft sclerite has a free end, i.e. is separated from sclerite R1P.

Antennae. These could be studied in 25 larval specimens of various stages, which were roughly determined by measuring the width of the head; the incompleteness and heterogeneity of the material (larvae of several species had to be used) did not allow a clear distribution to specific larval stages. The antennae of the youngest available larva (A. paucisetosa, head width 0.52 mm, Fig. 2A, H, I; head width in adults 0.76–0.89 mm) has a flagellum with 8 flagellomeres, well separated by interflagellomeral constrictions increasing in strength towards the apex. The constrictions cause an unusual shape of the flagellomeres, being rounded at both ends. The last flagellomere in this specimen appears to have a closed cuticle at its terminal end, but histological sections are necessary for a final decision. Diameter of flagellomeres slightly increasing up to the third, reaching there about that of the scapus and remaining constant up to the antennal apex; length of flagellomeres slightly increasing up to the fourth. First flagellomere (called meriston by Campbell and Priestley 1970) incompletely partitioned into three annuli, interflagellomeral membranes already visible, but without corresponding constrictions; the completion of the flagellomere division would be expected to take place at the following moult. The specimen certainly represents a very early, presumably the second larval stage (the presence of a dividing meriston is not expected to be present in a freshly hatched larva). The remaining larvae belong to intermediate and late larval stages (head width 0.61–0.75 mm, Fig. 2C–E) and show with increasing size an increasing approximation to the imaginal structure of the antenna: interflagellomeral constrictions diminished, but flagellomeres still well set off by their conical shape, their length strongly increasing towards the antennal apex (Figs 1A, 2F, G). In two thirds of the larvae signs of a division of the meriston could be found, sometimes restricted to only one of the two antennae. The meriston can be divided into two flagellomeres of different size, the proximal one being much smaller than the distal one (Fig. 2B, observed in seven specimens), or into three flagellomeres of fairly equal size (Fig. 2C, nine specimens). The divisions appear to be incomplete as in the young larva described above and obviously need at least one additional moult for completion. Even then signs of a previous division of the meriston may still be visible as is assumed for the antennae depicted in Fig. 2C, D: size and shape of the proximal flagellomeres are interpreted as showing a weak reminiscence of an earlier division of the meriston into two (Fig. 2C) or three (Fig. 2D) flagellomeres (compare with Fig. 2B, E). There was no evidence for a subdivision of flagellomeres distad of the meriston (no meristonal annuli).

Sex-specific characters. Species determination in larvae is difficult since the larvae are missing most species-specific characters. Larval stages of four species with available larval material (A. aptera, A. bergi, A. fungicola, A. paucisetosa) were studied for species-specific characters. As a result, three character sets were found which, under favourable conditions, may allow an identification at least in late larval stages: the bristle patterns on abdominal T2–5 and on T6,7, and, in the female sex only, the structure of the subgenital plate. The bristle patterns of T2–5 – surface bristles either in one transversal line or dispersed – are the same in imagines (of both sexes) and late larval stages; in earlier larval stages of all species the bristles are arranged approximately in one transversal line. The bristle patterns of T6,7, in the imagines showing strong differences between males and females, are the same in the larvae of the two sexes and correspond to the pattern of the imaginal female; the typical male pattern only appears after the imaginal moult (Fig. 15F–L). On the female subgenital plate S7, the transversal ridge (sr7) has in several species a very specific shape, which is already visible in late larval instars (Fig. 17G, H). For instance, an A. aptera larva (Fig. 17G) already showed the wide median gap of ridge sr7 as present in the adult A. aptera female (Fig. 17C, D), whereas an A. bergi larva (Fig. 17H) had a medially continuous and bisinuate sr7 as the adult A. bergi female (Fig. 16A, B).

Attaphila fungicola shows an unusual high variability in the structure of the median part of the sternal transversal ridge (sr7 in Fig. 18E, F) and is, therefore, not included in the key. Applying the key to its specimens would lead to several places following slot 2’. The females of A. fungicola are otherwise well characterised by the mostly tiny bristles distributed all over the surface of T6. — Note that features of the anteromedian part of the transversal ridge (sr7) should be examined at high contrast (as, e.g., in Fig. 18G compared to Fig. 17E).

The key has to be used with care since adult males are only known from 5 of the 9 species described: Attaphila paucisetosa, A. fungicola, A. flava, A. aptera, and A. bergi.

Data on the biology of Attaphila cockroaches and on the symbiosis with their host ants are quite fragmentary, although with very few aspects studied quite intensely in selected species (see below). In the attempt to combine the available data into a more coherent picture, there are three major problems:

(1) Due to the hidden life within the ant nests, in situ studies on Attaphila biology inside the nests are difficult and therefore quite rare. Observations on Attaphila cockroaches outside the ant nests may partly concern typical behaviours (related to, for instance, dispersal), but may also concern untypical cases of emergency (for instance, after a destruction of the home colony). And results from studies in the laboratory may include to an unknown extent artifacts in some aspects of biology.

(2) In view of the species diversity of both Attaphila cockroaches and their host ants, of the biological diversity of the host ants (e.g. regarding nest size and plants used for fungus cultures, see below), and of the wide distribution spanning different climate zones, some life history traits could well be quite different among the species of Attaphila. The observations on individual species reported below can thus not be generalized to all Attaphila species.

(3) The 9 species of Attaphila recognised herein have been found in colonies of only 10 species of host ants (Table 2; mainly the abovementioned well-studied species) out of ca. 78 extant species of leaf-cutting ants (according to https://antwiki.org). The absence of Attaphila records from the vast majority of leaf-cutting ant species may suggest highly incomplete sampling.

These issues should be kept in mind in the following.

Attaphila are only known from nests of leaf-cutting ants (Atta, Acromyrmex, Amoimyrmex), with one questionable exception, an undetermined Attaphila individual briefly spotted in the nest of an undetermined Trachymyrmex species² (VN personal observation). It is noteworthy that Attaphila individuals were also observed to follow trails of Trachymyrmex (see 6.5.). While leaf-cutting ants farm an obligately symbiotic fungus that is not able to live without the ants, and provide the fungus almost exclusively with fresh plant material, Trachymyrmex and all other non-leaf-cutting Attini (“lower attines”) primarily use detritus (de Fine Licht and Boomsma 2010) to farm an array of different fungi that can also live without their host ants (Schultz and Brady 2008; Branstetter et al. 2017). Leaf-cutting ants typically also have larger bodies and live in larger colonies than the lower attines. Either of these three factors might be responsible for Attaphila being rare or absent in the lower attines.

For Attaphila we found records of co-occurrence with only 10 of the 78 valid species of leaf-cutting ants (Atta 20, Acromyrmex 55, and Amoimyrmex 3 according to https://antwiki.org). Most leaf-cutting ant species for which no association with Attaphila has been reported had already been formally described (nearly all before 1910) at the time when the labels indicating host ants were produced for collected Attaphila specimens examined herein. This means that the set of reported Attaphila host ants is unlikely to be artificially small because relevant species had not yet been described when the Attaphila were labelled. In addition, many relevant determinations of the ants were conducted (in case of Attaphila and their ants collected by VN or JRG) or tested (in case of determined ants pinned together with formerly collected Attaphila) by ourselves based on literature altogether reflecting up-to-date species-level taxonomy (Santschi 1925; Gonçalves 1961; Borgmeier 1951; Schultz et al. 1998).

The few host records of Attaphila are distributed over the entire leaf-cutting ant phylogeny (Fig. 27A; Cristiano et al. 2020; Bacci Jr. et al. 2009): Atta sexdens (host ant of Attaphila sexdentis), Atta texana (host ant of Attaphila fungicola), and the species pair Atta cephalotes and Atta colombica (host ants of Attaphila paucisetosa) are representatives of the three principal lineages within Atta, i.e. they are phylogenetically as disjunct as possible within the genus. Within the genus Acromyrmex, Acr. octospinosus and Acr. echinatior (host ants of Attaphila aptera) are in a different main clade than Acr. lundii (host ant of Attaphila bergi) and Acr. niger (host ant of Attaphila schuppi), and Acr. lundii and Acr. niger are also not very closely related. This pattern may suggest that Attaphila cockroaches inhabit the nests of far more leaf-cutting ant species than we know of so far. If a targeted search of Attaphila in the nests of a variety of leaf-cutting ant species is successful, it would either reveal additional Attaphila species or wider host ranges (see below) of the species already known. If, in contrast, no Attaphila specimens are found in association with the many further leaf-cutting ant species, the limitation of Attaphila to few disjunct subclades in the leaf-cutting ant clade will pose an interesting biological question. One case in view of this question may be the strictly grass-cutting ants – some species of Atta (within the Epiatta clade) and Acromyrmex that forage grass instead of dicot leaves and flowers, in particular in the grasslands of southern South America (Bacci Jr. et al. 2009; De Fine Licht and Boomsma 2010; Mueller et al. 2017). For these, Attaphila has not yet been recorded. Acromyrmex lobicornis and likely Amoimyrmex silvestrii, the hosts of Attaphila bergi (?) var. minor, forage both grass and dicots (Mueller et al. 2017; no data for Amoimyrmex silvestrii, but the most closely related Amoimyrmex striatus, Fig. 27A, does forage grass and dicots). Whether the absence of Attaphila records from strictly grass-cutting species has biological reasons or is due to limitations in the sampling of these ant species remains open.

Regarding the degree of host specificity, the data available for associations between species of Attaphila and their host ants (Table 2; Fig. 27A) only allow for very limited conclusions. For three of the nine Attaphila species, no specific host ant species have been recorded so far (A. multisetosa, A. sinuosocarinata, A. flava). Two further species have only been recorded once (A. sexdentis, A. schuppi), so that it is not surprising that also only one host species is known. One species, A. fungicola, has been recorded many times and consistently in association with a single ant species, Atta texana; however, there is no other species from the three relevant ant genera that occurs in the distribution area of A. fungicola in Texas and Louisiana (USA). The six foregoing species can therefore not contribute to assessing host specificity of Attaphila species.

Three further species of Attaphila have been recorded on several occasions and in association with more than one species of Attini: (1) A. paucisetosa with Atta cephalotes and Atta colombica; (2) A. aptera with Acromyrmex octospinosus and Acromyrmex echinatior; (3) A. bergi with Acromyrmex lundii and, if the hosts reported for its “var. minor” are considered, also with Acromyrmex lobicornis and Amoimyrmex silvestrii, i.e. across a wide phylogenetic range (Fig. 27A). Case (3), however, is obscure because specimens classified as “var. minor” (all probably larvae) could not be examined in our study, whereby both the conspecifity of the various var. minor specimens (those found with Acromyrmex lobicornis and those found with Amoimyrmex silvestrii) and their assignment to A. bergi remain questionable (see 4.2.). Cases (1) of A. paucisetosa and (2) of A. aptera are thus the only ones demonstrating that host specificity of Attaphila is not necessarily limited to a single ant species. Notably, in both cases the host ants are closely related: Atta cephalotes with Atta colombica, and Acromyrmex echinatior with Acromyrmex octospinosus (Fig. 27A; Sumner et al. 2004; Bacci et al. 2009; Cristiano et al. 2020).

This leads to the current picture that Attaphila species are likely limited to single ant species or to groups of closely related ant species (far below the level of the respective ant genera). With the sparse sampling that is currently available, however, other possibilities cannot be excluded: The closely related species might also share relevant ecological traits, and other, phylogenetically disjunct leaf-cutting ant species with a similar ecological profile might be as useful as hosts for the same Attaphila species. Or, the closely related species might just be the only ones of a larger ant clade (e.g. of a clade classified as a genus) that are available in the distribution area of the Attaphila species concerned, while in other regions the same Attaphila species might (or would) have a wider host range (e.g. at genus-level). Furthermore, the degree of host specificity may vary strongly among the various Attaphila species. We also note that there appears to be some cryptic genetic variation in leaf-cutting ants (Kooji et al. 2018) that in the future might lead to the splitting of species and to an increased species number in leaf-cutting ants. This could also influence our view on the degree of species-specificity of the Attaphila-ant associations.

Host specificity at least at the level of ant genera is, with regard to the mentioned species pairs of Atta and Acromyrmex, especially convincing in the case of the locality Gamboa (Panama). There, all four ant species live in sympatry, and the nests of Atta colombica and two species of Acromyrmex (Acr. octospinosus, Acr. echinatior) occur only few meters from each other (VN personal observations). Despite the close proximity of the nests of all three ant species, A. aptera was never found in the nests of the Atta species, and A. paucisetosa never in those of the Acromyrmex species. On a larger scale, fungicola-group cockroaches have only been found in colonies of Atta ants, and bergi-group cockroaches seem to be restricted to Acromyrmex and Amoimyrmex (Fig. 27A). This may reflect the striking ecological differences between the two ant groups: Atta colonies can reach a size of a house and dominate their ecosystems with their long foraging trails and by defoliating the immediate surroundings of their nests, while Acromyrmex and Amoimyrmex nests are rarely larger than a basketball and rather inconspicuous. On the other hand, experiments under laboratory conditions showed that the cockroaches can survive at least for a short time in colonies of non-host leaf-cutting ants (Moser 1964; Nehring et al. 2016). So far, no specific life-history differences (in e.g. diet, life cycle, or dispersal) among Attaphila species or potential specific adaptations to host ants of a specific clade (or species), or a specific ecological profile have been described.

Wheeler (1900) initially believed Attaphila cockroaches feed on the fungus garden just like their host ants. He concluded this from gut dissections, which yielded a whitish substance that he interpreted as masticated remains of the mycelium (remains of the chitinous hyphal walls are not reported). Attaphila individuals were also observed to manipulate fungus fragments with their mouthparts (Nehring et al. 2016), suggesting the fungus to be at least part of their diet. However, later Wheeler (1910) proposed that cockroaches may lick lipids off the ant cuticle when riding on them (see 6.4.). We submit that this source alone could hardly explain the abovementioned gut contents, and it might appear as too meagre to sufficiently nourish the cockroaches – both regarding the amount and the biochemical diversity of what could be licked from an ant’s surface.

Attaphila cockroaches have frequently been reported to be found deeply inside the nests of their host ants, mainly in the fungal chambers (e.g. Wheeler 1900; Brossut 1976; Waller and Moser 1990; Nehring et al. 2016). Their small compound eyes (relative to those of other cockroaches) may suggest that the cockroaches spend most of their life in the darkness inside the nest (Wheeler 1900) and that they rarely leave this well-protected habitat – although it is hard to estimate the amount of life-time they spend outside the nest (see 6.5.), and the number of leaving events.

Attaphila cockroaches are known to ride on workers within the ant nest (A. fungicola: Wheeler 1900, Phillips et al 2017; A. paucisetosa and A. aptera: Nehring et al. 2016; Fig. 28E; observations mostly made in cultures). This behaviour, enabled by presumably strong attachment abilities via well-developed pretarsal arolia (Brossut 1976), triggered the idea that cockroaches feed on the ants’ cuticular lipids (see 6.3.), and it also serves the intrinsic dispersal, i.e. between fungus chambers within the same nest (Phillips 2021). An ant worker’s back or head may also be a rather safe spot for a cockroach that is always under threat to be killed by its hosts. The cockroaches generally smell like their host colony, probably because they acquire host-specific substances from the ants and/or fungus garden (Nehring et al. 2016), but a genetic disposition to the odour of a specific host species may also be one aspect of the host specialization of the cockroaches. In any case, the cockroaches are still sometimes attacked by their own host colony’s workers, at least when isolated in laboratory set-ups, which may cause stress to the ant workers (Nehring et al. 2016). Tightly clamping onto an ant might allow the cockroach to better blend in with its surroundings and be less of a suspicious particle.

When not actively attached to them, the cockroaches appear to avoid contact with ant workers as much as possible. The cockroaches flee when touched by ants and otherwise hide in the fungus garden with its multiply folded surface providing many crevices (Nehring et al. 2016).

Attaphila cockroaches have to leave the ant nest at least for their extrinsic dispersal to other ant colonies, which could either be already existing ones or newly founded ones. There are basically two ways to reach another colony: the cockroach could either join the ants in their dispersal activities (vertical transmission; e.g. by phoresis during mating flights); or it could conduct its own activities independent of the ants (horizontal transmission; e.g. by leaving its natal nest and searching for another).

Females and – less commonly – larvae (and in a single reported case a male: Phillips et al. 2017) of Attaphila fungicola (Moser 1967) as well as females of Attaphila paucisetosa (see 4.6. “Material”) have been seen attached to swarming virgin ant queens that were about to found new colonies. Attachment to alates has additionally been reported for Attaphila bergi (Bolivar 1901). Female A. fungicola that had experimentally been separated from the virgin ant queens prior to their mating flight produced oothecae within a few days (Waller and Moser 1990), indicating that they were mature and presumably also inseminated, so that they would have been able to populate the newly founded host colony. Attaphila fungicola females have indeed been found in newly established colonies (Moser 1967). This all is in line with a vertical transmission system where inseminated cockroach females disperse with host ant alate females to establish new populations in newly founded ant colonies. If vertical transmission were the major or only dispersal mechanism, all cockroaches in an Atta colony would be expected to be the offspring of one or few cockroach females that were transported by the colony’s queen.

In contrast, in studies on Attaphila paucisetosa and its host Atta cephalotes in Colombia (Rodríguez et al. 2013; JRG unpublished observations), Attaphila specimens were only found in nests older than two years and larger than 30 m². Their absence from newly founded ant nests indicates vertical transmission to be either uncommon or not very effective. Observations by Phillips et al. (2017) that Attaphila fungicola females indeed do not survive well in newly founded Atta nests corroborate this.

The latter cases suggest that horizontal transmission is also (or even more) important. One prerequisite for this seems to be present since Attaphila fungicola can follow ant pheromone traces in the laboratory (Moser 1964) and generally track ants (Sánchez-Peña 2005), and Attaphila schuppi has been found on ant foraging trails in the field (Bolívar 1905). There seems to be little specificity of trail following behaviour since Attaphila fungicola cannot only follow the trails of its host Atta texana, but also those of Trachymyrmex ants (Moser 1964). However, trail following alone would not be sufficient for dispersal because the trails of different Atta nests are unlikely to be connected.

Phillips (2021) suggests a combination of extrinsic dispersal via swarming queens and via ant trails: He observed Attaphila fungicola females to dismount the Atta texana queens after the mating flight and to search for ant trails instead of remaining with the queen. Once on a trail – most likely one of a foreign colony – the cockroaches would not walk to the nest themselves, but mount foraging ant workers, or even leaf fragments carried by the ants, as vehicles. Such a two-step dispersal makes use of long-range dispersal via swarming queens but avoids the problem that a high rate of foundress nests will fail, and is in line with all observations above. Riding into a new colony on the back of a worker may be a way for the cockroaches to avoid their hosts’ nestmate recognition system (see 6.4.). Ants recognize intruders by their smell, and in laboratory experiments leaf-cutting ant workers indeed attacked and killed Attaphila paucisetosa and A. aptera originating from other colonies than their own (Nehring et al. 2016). However, being carried by a nestmate worker that is carrying food to the nest might allow the cockroaches to “fly” under the radar and avoid detection.

Few details of the Attaphila life cycles are known, mostly from Attaphila fungicola in Louisiana, where observations indicate that the Attaphila life cycle is linked to that of the host ant: Waller and Moser (1990) report that the ratio of mature females to larvae in an ant nest is much lower after the mating flight than before, indicating that many mature females have left the nest. It thus appears plausible that there is a gradual maturation of the Attaphila population in a nest throughout the year; during the swarming of the ants, inseminated mature Attaphila females can disperse with virgin ant queens (see 6.5.); in their new host colonies, Attaphila females then produce oothecae, and larvae hatch, which develop into adults before the next ant mating flight in the following year. Then, another cycle begins with the appearance of small larvae. Females can be collected from the colonies throughout the year and can live for longer than one year in laboratory colonies (Waller and Moser 1990), suggesting that they can go through more than one reproductive cycle.

While both males and females have been collected from Atta texana colonies in Texas, only females have been reported from Louisiana (Waller and Moser 1990). It is unclear whether the lack of males is due to incomplete sampling or due to a potential local evolution of parthenogenesis.

The species of Attaphila are in both sexes provided with a series of interspecifically variable characters allowing in most cases a clear identification (summarised in Table 3). In addition to the sex-specific characters listed below there is one which is identically expressed in both sexes: the arrangement of the surface bristles on abdominal tergites T2–5.

Important male characters are shape characteristics of the tegmina including the orientation of the apical border; the shape of the hindwings; the presence of long bristles on T1, the structure and position of tergite specialisations; the size and arrangement of bristles along the posterior border and on the surface of T6,7; the development of the transversal ridge of T7; the presence of a hook on the right paraproct; and the structure of the subgenital lobe (with one or two styli, with a lateral excavation on one or on both sides) and the phallomeres (curvation of virga, shape of hook). The variability in the latter two body parts is, at least within the two species groups (see below), astonishingly low; A. flava, A. fungicola, and A. paucisetosa are according to their male genital characters almost indistinguishable (Figs 25A, C, D, 26A–C).

In the females distinguishing features can be found in the size and arrangement of the surface bristles of T6,7; in the development of the transversal ridge of T7; and in the features of the subgenital plate (S7 shape and (dis)continuity of transversal ridge) and of the genitalia (shape characteristics of laterosternal shelf area, gonangulum sclerites, spermathecal plate). In the latter, the laterosternal shelf area is most important, showing the highest variability; the females of the eight species with known females are all well characterised by specific details of this area (Fig. 22A–L). This is a rather unusual situation for Blattodea, which contain many genera (like the ectobiids Ectobius and Phyllodromica) in which species determination of females via morphology is extremely difficult or even impossible. Nevertheless, a similar case as in Attaphila is also found in the blattellid genus Loboptera (Bohn 1991), where the variable structures of the female genitalia, including also the laterosternal shelf area, allow an unequivocal determination of almost all species.

For most of the characters that distinguish Attaphila species (see 7.1.), outgroup comparison with other blaberoid taxa is problematic for a variety of reasons (often in combination): either the elements concerned are likely unique to Attaphila (e.g. tubes of laterosternal shelf); or they show a unique condition (middle part of laterosternal shelf sclerite); or data on other blaberoids are insufficient (concerning structural detail or the number of taxa studied). In other cases outgroup comparison is conceivable, but conflicting (e.g. gonangulum sclerites); homoplasy can occur to a considerable extent. Relationships in Attaphila can thus to a large extent only be discussed without indications on character polarity, or based on polarity hypotheses derived from the specific kind of structural differences (e.g. conditions with stronger reduction, asymmetry, or segmental differences as putative apomorphies).

According to male characters (see Table 3), A. aptera, A. bergi, A. flava, A. fungicola, and A. paucisetosa (the species with males known) can be sorted into two species groups: the bergi-group, as we may call it, containing A. aptera and A. bergi, and the fungicola-group with A. flava, A. fungicola, and A. paucisetosa (matching colours in Table 3). The bergi-group is characterised by the transversal apical border of the tegmina, the absence of long bristles on T1, the presence of a median specialisation on abdominal T1 (though this is represented by different cuticular structures in the two species), the presence of transversal ridge tr7, a subgenital lobe with an excavation only on the left side and two styli, and the distinct sinusoidal excurvation of the phallomere virga. The fungicola-group is characterised by the oblique apical border of the tegmina, the presence of long bristles on T1 (unknown for A. flava), the presence of lateral specialisations on abdominal T2 (with corresponding cuticular structures in the species), the absence of transversal ridge tr7, a subgenital lobe with an excavation on both sides and only one stylus, and the only weakly curved or bent virga. The differing sets of features are accompanied by a different host specificity: the species of the bergi-group live in nests of Acromyrmex species, those of the fungicola-group in nests of Atta species. The genus specificity in the host selection seems to be very strong, while a comparable species specificity does not seem to exist (see 6.2.).

The idea of a division into two species groups does not get much support by characters of the females. At first glance, the size and arrangement of the surface bristles of T6,7 appear to be in agreement with this grouping. The bristles are numerous and large in the bergi-group, even the largest in the fungicola-group are smaller and they are much less numerous. These differences, however, become obsolete when the remaining species ‒ A. multisetosa, A. schuppi, A. sexdentis, and A. sinuosocarinata, i.e. those with only females known ‒ are included in the considerations. The size of the bristles in these females varies strongly from slightly smaller than in the bergi-group to much smaller than the largest in the fungicola-group; a separation into two well defined groups by this character, therefore, is not possible.

Other female characters usable for the elucidation of interspecific relationships concern the laterosternal shelf area with its high interspecific variability. The great similarity in the shape characteristics of the laterosternal shelf area between A. fungicola and A. paucisetosa (Fig. 22E, F, H; represented as “type 2” in Table 3), for example, may well be interpreted as an indication of a close relationship between the two species, confirming their assignment to the same species group. The species A. bergi and A. schuppi represent another pair with very similar laterosternal shelf areas (Fig. 22C, D, J; “type 1” in Table 3) also suggesting a close relationship. The deviating shape characteristics of the laterosternal shelf area in A. aptera (Fig. 22A, B) do not necessarily contradict the suggestion to place the three species together in a bergi species group. The high variability in the shape characteristics of the laterosternal shelf area can be seen as a sign for rapid evolutionary changes in these structures, in contrast to male genital structures (subgenital lobe, phallomeres) showing only small differences within each of the two species groups. Thus, while the male genital structures thanks to their slower evolution might still indicate a close relationship between A. aptera and A. bergi, these indications may no longer be present in the structure of the laterosternal shelf. The suggested assignment of A. schuppi to the bergi-group is also supported by its ant host belonging to Acromyrmex (Acr. niger; Table 2).

Female characters cannot contribute to the clarification of the position of the other three species with unknown males: A. multisetosa, A. sinuosocarinata, and A. sexdentis; neither the laterosternal shelf area nor the subgenital plate allow conclusions on their relationships with other species. However, two of the species, A. multisetosa (with an Atta host ant) and A. sinuosocarinata (with host ants unknown), show similarities in the arrangement of the surface bristles on abdominal T2–5 with A. paucisetosa. The bristles are arranged in a strict line in A. paucisetosa, less strictly in A. sinuosocarinata, with singular bristles being slightly apart; in A. multisetosa bristles arranged in one line are only found in the median third of T2–4. The similarities might be interpreted as signs of close relationships suggesting the assignment of the two species to the fungicola-group. The suggestion would in case of A. multisetosa also get support from its host, a species of the genus Atta. But the proposed assignment remains doubtful for two reasons: first, the respective character, the arrangement of bristles in one line, is not very complex and thus potentially prone to homoplasy; second, since the larvae of species having dispersed bristles in later stages also have bristles in a single line in early stages, it could be a plesiomorphy (though alternatively a paedomorphic apomorphy).

For some of the male characters separating the two species groups, tentative conclusions on their polarity can be made. (1) The presence of a stylus on each side of the male subgenital lobe in the bergi-group would appear plesiomorphic, and the lack of a stylus on the right side as an apomorphy of the fungicola-group, as a pair of styli is part of the basic body-plan of Blattodea. A lack of the right stylus only sporadically occurs in other Blaberoidea (e.g. Roth 1999: fig. 7G). (2) In both groups the male subgenital plate has an excavation on the left side bearing the left stylus; the formation of an additional excavation on the right side in the fungicola-group may be considered as apomorphic in comparison with the bergi-group, as such an excavation only sporadically occurs in other, likely phylogenetically remote blaberoid taxa (e.g. Bohn 2019: fig. 19K). However, polarity within Attaphila is a bit doubtful, because it is unknown from which position the right stylus was lost in the fungicola-group: either from a proximal position comparable to the left stylus of the bergi-group (the distal shift of the right stylus then being an apomorphy of the bergi-group, and the presence of an excavation on both sides possibly representing a plesiomorphic symmetrical condition); or from a distally shifted position as the right stylus of the bergi-group (the distal shift of the right stylus then being a groundplan feature of Attaphila, and excavations of the two sides not representing a symmetrical condition, the right one being an independent apomorphy). (3) The absence of male tr7 might be an apomorphy of the fungicola-group, as it constitutes a difference between segments (male tr2–6 are in all species well developed). (4) The transversally cut tegmina of the bergi-group may be considered as more plesiomorphic than the obliquely cut tegmina of the fungicola-group – if the latter can reasonably be considered as including a further advanced reduction of the posterior part of the tegmen. (5) The lack of long bristles on male T1 might be an apomorphy of the bergi-group, as it constitutes a difference between segments (all Attaphila males have long bristles on T2–5). Characters (1)–(4) tentatively support the Atta-associated fungicola-group as monophyletic. This may then additionally be supported by the shared specialisations on male tergite T2 (for which outgroup comparison is not conclusive, see 7.3.3.). On the other hand, only one character (5) suggests the monophyly of the Acromyrmex-associated bergi-group.

Paraphyly of the bergi-group is more strongly supported than its monophyly by the retention of three putative plesiomorphies only in A. aptera but not in A. bergi (and other Attaphila species): the presence of the lateral part of the gonangulum (gg-l in Fig. 19A, clearly a plesiomorphy by outgroup comparison, see 7.4.2. – although with some instances of homoplasy); the well-developed male tr7 (see character (3) above; only weakly developed in A. bergi); and the presence of a hook on the right paraproct. Paraproctal hooks are ubiquitous in Blaberoidea (except for Pseudophyllodromiidae) and most likely an autapomorphy (under exclusion of Pseudophyllodromiidae if this is the sister group of the remaining Blaberoidea); its presence in A. aptera may represent a unique plesiomorphy within Attaphila (though only with regard to species with known male sex). In this case the aberrant laterosternal shelf of A. aptera (see above in 7.2.) could rather be seen as a further indication of this species being an early offshoot. On the other hand, the fully continuous condition of the transversal ridge on the female subgenital plate only in A. bergi (sr7 in Fig. 16A, B) is possibly a plesiomorphic condition suggesting this species to be the earlier offshoot; yet, both continuous and interrupted sr7 ridges occur in various other blaberoid genera, indicating a high degree of homoplasy for this character. As a third alternative, A. schuppi is supported as the basalmost offshoot within Attaphila by the probable lack of the genital chamber pouch (gcp). This structure has not been reported from other Blaberoidea (but may have been overlooked) and may thus appear as an autapomorphy of Attaphila excluding A. schuppi.

The last remaining species, A. sexdentis (only known from the female holotype), differs from all other species by the semicircular female subgenital plate S7 (Figs 17F, 18L, Table 3), which is based on the throughout rounded-converging course of the lateral plus anterior borders in short distance to the transversal ridge sr7. This is another character for which outgroup comparison is difficult. The first reason is conflicting outgroup comparison, as in other blaberoids the lateral borders of S7 variously diverge to the anterior, are parallel, or converge to the anterior (red lines in Fig. 30A–F, compare Fig. 30G, H showing the two conditions occurring in Attaphila). The anterolateral extension of S7 sclerotisation beyond tr7, which in A. sexdentis is much shorter than in other Attaphila species (compare green lines in Fig. 30G and H), also varies strongly in other blaberoids (green lines in Fig. 30A–F) and is often unclear in addition due to gradual fading (indicated by dashed parts of green lines in Fig. 30). These and other shape characteristics of the female S7 are generally difficult to compare between Attaphila and the blaberoids shown in Fig. 30 due to the aberrant course of ridge sr7 in Attaphila, with very steep lateral parts. The evidence from S7 of A. sexdentis on the grouping of Attaphila species thus remains unclear. The observed association of A. sexdentis with a species of Atta supports the assignment to the fungicola-group, and its host ant Atta sexdens (Table 2) is, in addition, more closely related to those of A. paucisetosa (Atta colombica and Atta cephalotes) than the latter ants with the host ant of A. fungicola (Atta texana) (Fig. 27; Cristiano et al. 2020: fig. 2).

In conclusion, (i)A. flava, A. fungicola, and A. paucisetosa likely form a clade (supported by four potential apomorphies in males), to which A. sinuosocarinata and A. multisetosa may also belong (based on similarities among the females); this is the Atta-associated (unknown or doubtful, respectively, for the two latter species; Table 2) fungicola-group. (ii)A. aptera, A. bergi, and A. schuppi, the members of the Acromyrmex-associated bergi-group, are poorly supported as a clade, while plesiomorphies support each of these species to be the sister taxon of the remaining Attaphila; this position is most strongly supported for A. aptera. (iii) There is no evidence on the position of A. sexdentis in this grouping of species, except that its association with Atta favours its assignment to the fungicola-group. This picture is very preliminary and partly contradictory. It mainly suffers from the lacking knowledge of the male sex in several species and from the limited availability and/or ambiguity of outgroup comparison with other Blaberoidea, especially Blattellidae.

It is thus too early for conclusions on a possible co-evolution between Attaphila and its host ant genera. Branstetter et al. (2017: p. 4, fig. 1) date all splits from that between Atta + Acromyrmex + Amoimyrmex (Acromyrmex striatus therein) and its Trachymyrmex sister clade down to that between Acromyrmex and Atta to a fairly short time span of ca. 20–17 Ma ago; this places the origin of leaf cutting in ants to about 20–18 Ma ago. This was then possibly also the time when, in close succession, the life habits of Attaphila originated and the early dichotomies within Attaphila have occurred together with the dichotomies in the ants. Alternatively, these events in Attaphila evolution could have occurred later if one or several host shifts from one ant genus to the other are involved despite the apparent present-day stability in host choice. Due to the above finding of Acromyrmex-associated Attaphila possibly being paraphyletic and Atta-associated Attaphila more likely being monophyletic, a host shift from Acromyrmex to Atta is more likely than one in the opposite direction.