Specimens were first collected and observed in Guyana in 2014 (Evangelista et al. 2016). Afterwards, novel field collection was done at five sites across Madre de Dios, Peru. Collections in Peru were made across two expeditions, each occurring during the dry season. Some “friaje”, cold front, events occurred during each expedition, but this did not greatly impede insect activity. In 2021 (June and July), specimens were collected from three sites: Los Amigos Research Station (Los Amigos; 12°33'S 70°06'W), Kapievi Village (Kapievi; 12°36'41.4"S 69°11'47.2"W), and “Camungo” Trail (Camungo; 12°37'48"S 69°11'32"W). Los Amigos is the most remote site, located 99 km from Puerto Maldonado, the nearest city. Los Amigos is a preserved forest comprising multiple habitats (lowland floodplain, terra firma, palm swamps, and numerous oxbow lakes). However, some of the surrounding areas have been destroyed by illegal logging and gold mining. Kapievi and Camungo are within the boundaries of Puerto Maldonado, but outside the most densely populated neighborhoods. Kapievi is an active tourist lodge, with a combination of planted and wild-grown flora on its grounds. Camungo is an unmarked trail ~2 km south of Kapievi, leading to a small palm swamp pond. Camungo has entirely wild-grown flora but is neighboring several tourist lodges, hotels, and plantations.

In 2024 (July), specimens were collected from two rural sites: Finca Las Piedras Research Station (Las Piedras; 12°13'43"S 69°06'52"W) and Kawsay Biological Station (Kawsay; 12°31'37"S 69°00'54"W). Las Piedras is 40 km from Puerto Maldonado but is surrounded by several farms and a Brazilian nut plantation. Las Piedras, a terra firme forest, was selectively logged before being converted to a biological research station, but much of its primary forest remains intact. Kawsay is 21 km East of Puerto Maldonado but is in a relatively undisturbed region of the Madre De Dios, as it borders Tambopata National Park. Forests at Kawsay are mostly lowland floodplain.

The goal of each expedition was to collect as diverse and numerous Blattodea (excluding termites) as possible. This was achieved using three collection methods: manual collection, baited pitfall trapping, and light trapping. Manual collection was the most utilized because previous studies have found it to be the most effective method to collect cockroaches, with an emphasis on diversity (Evangelista et al. 2014). While some manual collection was done during the day, the vast majority was done at night with the aid of a headlamp. Baited pitfall trapping was done with three baits: beer, red wine, and fish. Previous studies have shown beer to be an effective bait for collecting a large abundance of neotropical Blattellidae (Evangelista et al. 2014; Evangelista et al. 2017), and other taxa with less abundance. To our knowledge, wine has not been used to collect cockroaches before, and our efforts confirm that it should not be, because we did not collect any with this bait. Similar to beer, rotten fish is also known to be a strong attractant for cockroaches. All traps utilize plastic cups that are buried in the ground or placed on a tree-trunk at breast height, using twine. In traps with liquid baits, the bait was used both as an attractant and a killing solution. In traps baited with rotten fish, the fish was wrapped in cheesecloth and suspended above the cup, with soapy water in the cup as a killing solution. In 2021, all three baits were used, and all traps were placed on the ground. In 2024, only rotten fish were used, and traps were placed in the ground.

The specimens collected in 2021 were processed in the lab at Adelphi University, and those collected in 2024 were processed in the lab at the University of Illinois Urbana-Champaign. In all cases, specimens were assigned a unique identifying number, and either pinned dry or placed in 70% ethanol and stored in the freezer at –20°C.

Freezer-stored samples were used for gut analyses. The sample was removed from the freezer, restrained with pins on a foam board covered in a sterile cloth (Kim Wipe), and dissected using forceps, micro-scissors, and pins. With the specimen placed on its dorsal side, the abdomen was opened by incising the lateral portion of the sterna. The sterna were reflexed and the intestine revealed. Gut contents were identified as material with dark coloration within the intestinal wall. Effort was made to remove as much gut content while minimizing cockroach tissue in the sample. The removed gut content was then placed in a sterile 1.5 ml tube. The tube was labelled and left open overnight in a fume hood and/or sealed desiccation chamber with silica gel. After drying for at least one full day, the tube was closed and placed in a freezer at –20°C.

Gut content DNA was extracted using standard QIAGEN DNEasy extraction protocols. DNA extracts were used in the following two protocols.

Metabarcode library construction and sequencing were performed at the Roy J. Carver Biotechnology Center, University of Illinois at Urbana-Champaign. Approximately 1 ng of DNA was used for PCR amplification with the primers shown in Table 1, following the Fluidigm protocol (Fluidigm, CA). Briefly, the PCR conditions were as follows. Every 1 µl of DNA sample was combined with a mastermix of: 0.5 µl of 10X FastStart Reaction Buffer without MgCl2, 0.9 µl of 25 mM MgCl2, 0.25 µl DMSO, 0.1 µl of 10 mM PCR grade Nucleotide Mix, 0.05 µl of 5 U/µl FastStart High Fidelity Enzyme Blend, 0.25 µl of 20X Access Array Loading Reagent, 1.95 µl of water. In the Fluidigm Juno, the following PCR conditions were used: 1 × 50°C 2 min, 1 × 70°C 20 min, 1 × 95°C 10 min; 10 × [95°C 15 s, 55°C 30 s, 72°C 1 min]; 2 × [95°C 15 s, 80°C 30 s, 60°C 30 s, 72°C 1 min]; 8 × [95°C 15 s, 55°C 30 s, 72°C 1 min]; 2 × [95°C 15 s, 80°C 30 s, 60°C 30 s, 72°C 1 min]; 8 × [95°C 15 s, 55°C 30 s, 72°C 1 min]; 5 × [95°C 15 s, 80°C 30 s, 60°C 30 s, 72°C 1 min]. PCR product was harvested, diluted by a factor of 100, and 1 µl of this diluted product was reamplified with Illumina annealing oligos under the conditions: 1 × 95°C 10 min; 14 × [95°C 15 s, 60°C 30 s, 72°C 1 min]; final extension 72°C 3 min. The individually barcoded amplicons generated from each sample were quantified on a Qubit fluorometer (ThermoFisher, CA), and the average size of the amplicons was determined on a Fragment Analyzer (Agilent, CA). Amplicons were pooled in equimolar concentration, size selected on a 2% agarose Ex-gel (Thermofisher) to remove primer-dimers, and extracted from the isolated gel slice with Qiagen gel extraction kit (Qiagen). Cleaned size-selected product was quantitated and run on a Fragment Analyzer again to confirm the appropriate profile and for the determination of average size. The pool was diluted to 5nM and further quantitated by qPCR on a CFX Connect Real-Time qPCR system (Biorad, Hercules, CA) for maximization of the number of clusters in the flow cell. The pool was denatured and spiked with 20% non-indexed PhiX V3 control library provided by Illumina and loaded onto the MiSeq Nano flowcell at a concentration of 8 pM for cluster formation and sequencing. The libraries were sequenced from both ends of the molecules to a total read length of 250 nt from each end.

To investigate the possible presence of protozoa related to the obligate Xylophagodea symbionts, 18S amplicon sequencing was carried out using a two-step protocol with the first-step primers specific for Parabasalia and Oxymonadida. After the first PCR, a second PCR was carried out with barcoded adapter primers as previously described (Jasso-Selles et al. 2020). Barcoded PCR products were quantified on a Qubit fluorometer (ThermoFisher, CA) and pooled at equal concentrations before sequencing on an Illumina MiSeq using v2 2x250 paired-end chemistry at the Arizona State University Genomics Core Facility.

We sequenced the guts of nine Lamproblattasp. nov. individuals from two sites (Los Amigos N=3; Puerto Maldonado N=6) along with a variety of other cockroach species. Other sampled taxa include: 13 Neoblattellini (Trioblattella eudromielloides (Hebard, 1921), cf. Trioblattella castanea (Rocha e Silva, 1958) and Cariblattoides cf. guyanensis Bonfils, 1975), seven Plectopterini (Dendroblatta spp., Euphyllodromia peruana (Saussure, 1864), and others), nine Blattellidae (Ischnoptera sp. and others), five Nyctiboridae (Nyctibora humeralis Dohrn, 1888, Rochaina sp., and undetermined juveniles), one Blaberidae (Epilampra cf. opaca Walker, 1868), and three Blattidae (Periplaneta australasiae Fabricius, 1775, Eurycotis lixa Rehn, 1930, Pelmatosilpha sp.; https://doi.org/10.5061/dryad.0000000g0 for all sample information). All specimens were collected from the same localities as L. mimadelfi in Peru, except for four additional samples (Periplaneta australasiae – Dominican Republic, Eurycotis lixa – lab reared, Epilampra cf. opaca – Guyana, Nyctibora sp. – Guyana). Plant-specific primers yielded very few positive hits for any samples. In Lamproblatta mimadelfisp. nov., metazoan-specific primers yielded mostly self-hit COI sequences, and further examination of other reads showed most were also likely self-hits on COI pseudogenes. Thus, the further analysis detailed below focuses on data obtained from Eukaryota and Fungi primers.

Basic biodiversity metrics were calculated in R using the vegan package (Dixon 2003). We compared the diversity of fungi found in Lamproblatta mimadelfisp. nov. to other Blattodea taxa using the Inverse Simpson (estimates diversity based on abundance among species; Simpson 1949) and Chao-1 (estimates diversity based on under-sampling expectation; Chao 1984).

For each unique contig found in Lamproblatta mimadelfisp. nov., we used NCBI BLAST to identify its most likely taxonomic affiliation and recorded all results in a table. We filtered out all BLAST results with a total score <300 and % identity <90. With this final list of taxa present in the gut contents of Lamproblatta mimadelfisp. nov., we used a literature search to identify the ecology and habit of each consumed taxon. We would then use these to make inferences about the diet of the cockroach species. For instance, if the major component of the gut was a wood-decaying fungus, we might conclude the insect feeds directly on that fungus or the wood itself. Such interpretations would also consider alternative explanations. As another example, if the major component of the gut was entomopathogenic fungi, it could indicate: (i) passive consumption from decaying organic matter, such as leaf litter or decomposing wood, (ii) active foraging on decaying insect matter, (iii) direct predation on another insect, (iv) infection with a pathogenic fungus, or (v) incidental ingestion of environmental sources of DNA through grooming. Further evidence may be needed to support (ii) or (iii) since these are stronger conclusions to be made (e.g., comparisons of gut material between sympatric taxa could be used to identify a preference for animal detritus; e.g., see Evangelista 2016). Point (v) could hopefully be ruled out because we might expect read abundances (i.e., the relative amount of DNA in the sample) to be low when an item is consumed incidentally.

We also analyzed the remainder of the gut meta-barcode data (10.5061/dryad.0000000g0) as a means of comparison against L. mimadelfi. These accounted for hundreds of sequenced OTUs (operational taxonomic units) identified via BLAST search against NCBI GenBank and BOLD databases. To make this a feasible task, we worked with aggregated raw data and did not further scrutinize the OTU identities as we did with the fungal data found only in L. mimadelfi (see above). We merged all identical OTUs, even if the contigs were not identical (e.g., we summed read abundances for contigs identified only as “Cystobasidiomycetes” and treated these as a single OTU). We also summed the read abundances by cockroach taxon. We then categorized all remaining OTUs into ecological categories, focusing on ecological interactions that would be indicative of the consumer’s dietary behaviors. We associated each OTU with a single category through a simple Wikipedia search. If a single taxon had multiple ecological interactions, we categorized it as unknown or other.

To obtain historical taxonomic records, we extracted character information from taxonomic literature and novel observations. Primarily, these were species descriptions (Hebard 1919, 1929; Rehn 1930; Roth 2003a; Fritzsche et al. 2008; Evangelista et al. 2016; Lucañas 2016; Evangelista et al. 2019a; Zhu et al. 2022; Evangelista et al. 2023), but we also compiled information from other systematic publications (Klass and Meier 2006), illustrations (Klass 1997), and photos of types (Hopkins and Beccaloni 2024). Species examined first-hand (Lamproblattasp. nov., Anaplecta intermedia Rocha e Silva, 1966, Eurycotis blattoides Hebard, 1926, E. decipiens (Kirby, 1903), E. bahamensis Rehn, 1906, Pelmatosilpha sp. “Peru”, Nyctibora sp. “Dominican Republic”) were imaged using a Leica Flexacam C1 microscope camera. Images were compiled using ZereneStacker and edited in GIMP v.2.10. Dissected specimens had their genitalia cleared using a 70% NaOH solution for 5–8 hours. Terminology for genitalia follows Klass (1997).

Morphological data were compiled in TaxonWorks V0.40.0. Phylogenetic relationships were inferred under a maximum parsimony framework using multiple methods in PAUP* v4.0a169 (Swofford 2002) and TNT v1.6 (Goloboff & Catalano 2016). In PAUP*, we conducted a heuristic search using Tree Bisection and Reconnection (TBR) branch swapping with 1000 random addition sequence replicates. We did multiple searches, varying the following: coding ambiguous characters as either missing or uncertain; enforcing the topological constraint of (Cryptocercus (Anaplecta + Lamproblattidae s. str.)) or not; considering both ordered and unordered assumptions about character evolution or assuming all states were unordered. From the resulting most parsimonious trees, we calculated both a strict consensus and a 50% majority-rule consensus tree. To assess support for the resulting clades, we performed a bootstrap analysis with 1000 replicates, from which a bootstrap consensus tree was generated. Tree length and standard parsimony metrics were recorded to evaluate tree fit and complexity.

The character states below were scored and compiled for all taxa. The characters, states, evolutionary assumptions (ordered vs. unordered), and inclusion in phylogenetic analyses are defined. Abbreviations are given in Table 2. The resulting character matrix is shown in Table 3.

1. Habitat. 0: Tropical, or subtropical forests; free living on forest floor. 1: Temperate forests; living exclusively in rotten woody galleries (very rarely found roaming free). 2: Tropical or subtropical habitats; found free living in forest floor, sometimes in rotten wood, under bark, tree holes, or other locations. 3: Tropical or subtropical habitats; burrowing in soil; 4: Tropical/subtropical forests; free roaming or aggregating on (or in) wood; may also be found near light sources at night. This character is not included in phylogenetic analysis.

2. Adult body size (measured along the medial line). 0: < 10mm. 1: 10–20 mm long. 2: 20–30 mm. 3: > 30 mm. Use the mean body size among known adults (male and female). Ordered.

3. Body coloration (general color of the body — dorsal and most of ventral; excluding coxae, palps and other small sections). 0: Black, pitch, deep black with red undertones, or another shade of solid black. 1: Shiny brown-black with yellow stripes laterally on thorax; yellow spots on lateral sides of tergites. 2: Buffy to brown. 3: Red to orange brownish. This character is not included in phylogenetic analysis as it is not as specific as character 4, but also somewhat redundant with character 4.

4. Main color of densely pigmented regions of the body. 0: Dull black, with little reddish undertones (e.g., pitch, jet, cynical black, glossy, slate, true black, coal, sometimes coffee black). 1: With variable density of coloration, but primary body color rust, brown, or black with strong red/brown undertones (e.g., coffee black, black bean, HEX #3D0C02, bistre; except e.g. Eurycotis lixa, which is colored more like Lamproblatta). 2: Buff, yellow, or light brown. 3: Primary color red, with orange or buffy undertones. Unordered.

5. Adult forewings (tegmina). 0: Absent. 1: Highly reduced (usually lateral pads not reaching past the posterior edge of the mesonotum, but sometimes spanning width of the thorax, and rarely reaching slightly past posterior edge of metanotum). 2: Moderately reduced but reaching noticably past posterior edge of metanotum, and rarely reaching to SA plate. 3: Large; reaching up to the end of the SA plate or much past it. Ordered.

6. Form of frons between antennae. 0: Slightly raised (as in Periplaneta americana) but not swollen or bulging. 1: Swollen or bulging.

7. Maxillary palp, penultimate palpomere. 0: Slender (tapering homogeneously from distal portion to base; angle of the basal part of palpomere <15°). 1: Robust (more rectangular, tapering less apparent, or only strongly at the base; angle from base of palpomere >15°).

8. Maxillary palp coloration/pigmentation. 0: Weakly pigmented, almost entirely white (adults), or slightly more darkened/pigmented on distal palpomere (juveniles). 1: With dark, more densely pigmented regions.

9. Spination type on AV margin of profemur (AV proleg spination; sensu Roth 2003b). A: A row of large spines. B: A row of large spines basally, followed anteriorly by a dense row of small spinules, or hairs. C: A row of small spinules or hairs. D: Lacking spines and spinules on the margin (although there may or may not be apical spines). Unordered.

10. Number of apical spines on the apex of the profemur AV margin: 1, 2, or 3. Ordered.

11. Genicular spine of profemur. 0: Absent. 1: Present.

12. Arolium on prolegs (we are specifying the proleg state only for precision; often, the proleg arolia state is the same as in all other legs.). 0: Absent. 1: Present.

13. Large spine on AV margin of metafemur. 0: Absent. 1: Present.

14. 1^st tarsomere of hind leg of adult male: 0: Slender, mostly straight and uniform width throughout. 1: Obviously inflated, but not bulbously. 2: Inflated bulbously. Ordered.

15. Surface of dorsum. 0: Smooth and shining. 1: Partly punctate, shining, and/or partly dull (rarely, entirely smooth). This character is not included in phylogenetic analysis because is it not well-defined (i.e., it doesn’t refer to a specific segment but the dorsum in general, and state 2 is an amalgamation of multiple states that may be variable).

16. 7^th tergite expansion (if present, the SA plate, and part or all of each cercus is hidden underneath the expanded 7^th tergite of adults of both sexes). 0: Absent. 1: Present.

17. Visible parts of tergal gland complex. 0: Absent. 1: Present as a dense patch of hairs and/or fossae in the middle or in the anterior third of the SA plate. 2: Present as a patch of hairs and/or fossae on any of the abdominal tergites 1 to 9 but not 10 (= SA). Unordered.

18. SA plate: long hairs on posterior third of SA plate (dorsal; males). 0: Absent. 1: Present.

19. Color of SA plate. 0: Pigmented (roughly) as other parts of the body. 1: Posterior edge (at least) is depigmented and whitish.

20. SA plate posterior edge shape (male). 0: Deeply notched at midline (i.e., the medial point of the edge is obviously less posterior than the points immediately lateral to it). 1: Posterior edge slightly notched (i.e., the medial point may be slightly less posterior than the points immediately lateral to it). 2: Flat (i.e., medial point is roughly equally posterior as the points immediately lateral to it). Ordered.

21. SA plate shape: shape of lateral edges. 0: Converging posteriorly and rounded/convex. 1: Converging posteriorly and straight or concave. 2: Roughly parallel. Unordered.

22. SA plate width: proportional (the ratio of the width of the posterior edge to the width of the widest, most anterior point visible from an intact specimen). 0: ≤ 0.1. 1: 0.1–0.37. 2: 0.38–0.45. 3: ≥ 0.46. Ordered.

23. SG plate: ratio of stylus length to interstylar space width (if styli are asymmetrical, this uses the average stylus length). 0: < 0.45 (i.e., the styli are no longer than 45% the width of the space between styli). 1: > 0.45 (i.e., the styli are longer than 45% the width of the space between styli).

24. SG plate: how far the medial point of the SG plate projects posteriorly beyond level of insertion of styli, relative to the length of the styli (ratio; if styli are asymmetrical, this uses the average stylus length). 0: > 90% the length of the styli. 1: 40–90% the length of the styli. 2: 0–40% the length of the styli. 3: <0% (i.e., concave; the medial portion of the SG plate is anterior to the level where the styli are inserted). Ordered.

25. Process sra (sclerotized by region R1d) and R1J (fused), R1G and R1H (split) sclerites of right phallomere of male genitalia, shape. 0: sra a spine-like projection facing medially, and not accompanying any other long projections R1J (or R1G if present). 1: sra a medial facing spine-like projection, and accompanied by one or more other long projections R1J (if fused) or R1G (if split). 2: sra lacking, or not a spine-like process, and other long projections on R1 may or may not be present. Unordered.

26. Process pda (sclerotized by region L4l) on left phallomere of male genitalia, shape. 0: A huge, slender, and curved spine. 1: A moderately sized, strongly curved spine. 2: A stouter process/spine (not to be confused with L2, which may have one or more slender spines (paa) projecting posteriorly and not strongly curved). 3: Not as above, not strongly curved or elongated. Unordered.

27. Process paa (sclerotized by region L2d) of left phallomere of male genitalia, shape. 0: A blunt nodule or lobe. 1: A long, spine-like process.

28. Process paa (sclerotized by region L2d) of left phallomere of male genitalia, shape, location. 0: region of L2 forming paa proximal to pda (L4). 1: Region of L2 forming paa not proximal to pda (L4); instead, possibly close to L1 or other sclerite regions.

29. Shape of sclerite region L1a of left phallomere of male genitalia, shape. 0: Level. 1: Hood-shaped, at least weakly so, but lacking anterior plateau. 2: Hood-shaped and with anterior plateau. Ordered.

30. Presence of sclerite region L4n (= sclerotization of and around process nla) of left phallomere of male genitalia. 0: Present; 1: Absent.

31. Extension of membranous basal part ‘30’ of hook hla and resulting retractability of hla of left phallomere of male genitalia. 0: Membrane very narrow, ‘hla’ hardly retractable. 1: Membrane moderately extensive, hla moderately retractable. 2: Membrane very extensive, ‘hla’ (almost) entirely retractable. Ordered.

32. Presence of process nla (sclerotized by region L4n) of left phallomere of male genitalia. 0: Absent. 1: Present.

33. Location of right insertion of muscle l2 of male genitalia. 0: On top of pouch ‘pne’. 1: In left, left-ventral, or dorsal wall of pouch ‘pne’, or in corresponding area. 2: On utmost base of hook ‘hla’. Unordered.

34. Location of left insertion of muscle l2 of male genitalia. 0: In the posterior two thirds of left edge of left complex on sclerite region L4l (in some taxa on individualized sclerite L4K or L4U, or on corresponding membranous area). 1: In anterior left edge of left complex on sclerite region L4l (sclerite L4K). 2: In anterior left edge of left complex, in membrane anterior to sclerite region L4l (and sclerite L4K). 3: in left anterior ventral wall of left complex on sclerotization (sclerite region L4x). Ordered. Note that only states 0, 1, and 2 are utilized here but the character state descriptions (and numbering) are preserved from Klass and Meier (2006) for consistency.

35. Presence of muscle s7 of male genitalia. 0: Absent. 1: Present.

We reviewed iNaturalist for observations of Lamproblattidae. The two native Blattidae genera of the Neotropics, Eurycotis and Pelmatosilpha, are more widespread, better-known, and superficially similar to Lamproblattidae (particularly immatures). To find Lamproblattidae misidentified as Blattidae, we reviewed all Neotropical observations of Blattoidea identified between the superfamily and genus levels. This assumes that all observations identified to the species level would be correctly identified to genus (although we also reviewed pre-existing Lamproblattidae observations of any taxonomic level). To narrow our search more effectively, we appended the following to the search URL “&without_taxon_id=118903,82233,154214”. This omitted termites, Periplaneta, Blatta, and Anaplectidae from the search results. We assumed that these taxa would rarely, if ever, be misidentified from a Lamproblattidae (although Blatta orientalis does have superficial similarities with Lamproblattidae). We did not look for any Lamproblattidae misidentified as Corydioidea or Blaberoidea.

We manually reviewed the 1000+ results for any superficially like Lamproblattidae (black body, blattid-like shape, and lacking wings). For photo observations that were visibly male, we used the following features to identify Lamproblattidae. We compiled these features based on our review of the literature using the taxa mentioned in the previous section, and those in Figure 1: the size of the styli (small and posterio-medially projecting in Lamproblatta, long and posteriorly projecting in Blattidae), the morphology of the SA plate (with lightly pigmented posterior edge, and variably shaped in Lamproblatta, often deeply medially notched in Blattidae). For non-adult males, or photographs where the terminalia were not visible, we used the following characters: coloration of the maxillary palps (largely white and lacking pigmentation in Lamproblatta, and densely or variably pigmented in Eurycotis), and body coloration (predominantly black, smooth and shining with little red undertones in Lamproblatta. In Eurycotis, with variable coloration, or, if black, with strong reddish undertones and punctations). In particular, the SA plate morphology was most heavily relied on compared to the other characters. Also, we have noticed that Lamproblatta does not often have the whitish secretions that Eurycotis and some other cockroaches often have on their dorsal-posterior abdomen. This factored into some identifications but was a minor consideration. For observations of smaller juveniles, or photos with all important features obscured, we proposed an identification of Blattoidea to account for uncertainty. For observations where the important features were visible, we identified the observation to the species level, relying mostly on the SA plate shape, while disregarding biogeography. All observations were added to the iNaturalist project Lamproblattidae (https://www.inaturalist.org/projects/lamproblattidae).

We then synthesized iNaturalist occurrence data with additional data from taxonomic literature (Hebard 1919, 1929; Rehn 1930; Gautier and Deleporte 1986; Roth 2003a; Evangelista et al. 2016; Evangelista et al. 2019a), the Global Biodiversity Information Facility (GBIF), and new field collection.

In total, we had 526 records of Lamproblattidae (including Lamproglandifera and Eurycanthablatta) obtained from the previous section (details in section 3.3): 54 of these were from iNaturalist, all of which appear to be Lamproblatta spp.; 48 occurrences were novel observations of L. mimadelfi; and 205 occurrences were from published literature, and the remainder were downloaded from GBIF. We removed the duplicate observations for every location and records without coordinates or with a low level of precision (≤ 2 decimals), resulting in 128 records. We also removed Eurycanthablatta because of systematic concerns (see section 3.1). Since habitat suitability models can be influenced by clustering of records (Sillero and Barbosa 2021), we performed a spatial thinning using the spThin package (Aiello-Lammens et al. 2015). To avoid oversampling localities within a 10 km radius, we considered a minimum distance of 20 km among localities. The home ranges of Lamproblatta spp. individuals may be very narrow (< 20 m; Gautier and Deleporte 1986), but our observations of Lamproblattasp. nov. in Peru demonstrate that metapopulations can span large areas (> 90 km). Also, we expect that Lamproblatta spp. have wider ranges than Cryptocercus (Nalepa 1999; Kambhampati and Peterson 2007; Che et al. 2016), the only well-studied and comparable close relative. Cryptocercus spp. are much more limited by the distribution of suitably rotten logs, whereas Lamproblatta spp. are less specialized. However, without precise knowledge of gene flow and dispersal dynamics in Lamproblattidae, we are forced to choose an intermediate buffer range of 10 km arbitrarily – larger than the very short home ranges, but smaller than the total geographic range of populations. Afterwards, we drew a 500 km buffer area around all the sampling points so that the extent of the calibration area (used to extract background data randomly) covers the whole Neotropical region. For this, we used sf package (Pebesma 2018). A total of 8441 random background points were obtained with the package dismo (Hijmans and Elith 2021). To assess the habitat suitability across the Neotropical region, we used the 19 bioclimatic variables and the elevation data obtained from WorldClim (2024) with a resolution of 30 seconds (~1 km²). We checked if the climatic variables were strongly correlated (correlation coefficient >0.8) with the package usdm (Naimi et al. 2013). To do this, we previously drew a 10 km radius buffer area around each record, and these areas were used in the correlation analysis. Buffers were created with the package terra (Hijmans 2024). After excluding correlated ones, we utilized the following variables: BIO 2 (mean of diurnal temperature variation), BIO 4 (temperature seasonality, a measure of how much temperature change there is throughout the year), BIO 13 (precipitation of wettest month), BIO 14 (precipitation of driest month), BIO 18 (precipitation of warmest quarter), BIO 19 (precipitation of coldest quarter), and altitude.

To find the best parameter configurations, we combined different feature classes and regularization multipliers (from 0.5 to 5, increment of 0.5; Morales et al. 2017). The feature classes we used were (linear – L, quadratic – Q, hinge – H, product – P, and threshold –T): L, Q, H, P, T, LQ, LQH, LQT, LQHP, LQPT, and LQPTH. We partitioned occurrences into two groups for data training with a background checkboard pattern and 2 folds (all data used for training). We used two criteria to select the best model: (1) the Akaike information criterion corrected for small sample sizes (AICc), and (2) the cross-validation method using the lowest omission rate and the highest average validation area under the curve (AUC) (Radosavljevic and Anderson 2013; Kass et al. 2019). To calculate the significance of these models, we performed a series of null models (1000 iterations) to calculate the significance (α = 0.05) of the performance metrics (omission rates and AUC) as suggested by Raes and Steege (2007), and the modifications of Bohl et al. (2019) and Kass et al. (2019). The distribution model, as well as the model selection and the null models, was performed using the package ENMeval 2.0.5 (Kass et al. 2021) in R 4.4.1 software (R core Team 2024).

The character matrix (Table 3) analyzed contained 28 taxa and 32 characters (11 ordered, 21 unordered), of which 6 were not parsimony informative.

Relationships recovered varied across the analyses and are summarized in Table 4. The major differences between analyses were in the placement of Anaplecta and the internal relationships within Lamproblattidae s. str. Eurycanthablatta pugionata was placed outside of Kittrickea in all analyses. Anaplectidae’s placement depended on how ambiguous characters were coded. When treating ambiguities as polymorphisms or coding them as missing data, Anaplectidae was within Kittrickea. Otherwise, it was placed outside of Kittrickea. Anaplectidae was supported as sister to Lamproblattidae by the shape of the male genital feature pda (on sclerite L4). Ignoring Anaplectidae, Cryptocercus was sister to, or in a polytomy with, Lamproblattidae in all analyses.

Within Lamproblattidae s. str., relationships were highly volatile, likely owing to the large amount of missing data, and the low number of characters in the matrix. L. meridionalis, which had 23/32 characters missing, was placed outside of Lamproblattidae s. str. in a few analyses. Otherwise, Lamproglandifera and Lamproblatta formed a single monophyletic group in all analyses, with Lamproblatta being paraphyletic with respect to Lamproglandifera. Lamproglandifera flavoglandis was sister to L. neuque in some analyses. Lamproblatta mimadelfi was sister to L. mimetes in most analyses. Otherwise, there was little consistency between the trees, and majority rule consensus calculations demonstrated that there is little-to-no resolution within Lamproblattidae s. str. We present the results of one of the constrained analyses as a representative phylogeny (Fig. 7).

Under ideal circumstances, taxonomic experts would have no issues differentiating among the Neotropical taxa treated here. Yet, parataxonomy is often done without physical specimens, access to a microscope, or even high-quality photos. Under these circumstances, Lamproblattidae may be hard to identify, and can be easily confused with Eurycotis. We have observed this issue first-hand on iNaturalist, where Lamproblatta observations are not particularly rare, but correct identifications were lacking. To facilitate better identification, we present a key here.

128 unique occurrences (Fig. 10) showed that Lamproblattidae are most common in Central America and the northern parts of South America. However, we expect that sparse records of Lamproblattidae throughout the Amazon and the remainder of South America are sampling bias. We incorporated all unique occurrences as presence data to infer niche model range predictions for all Lamproblattidae (including Lamproglandifera). We did not predict the ranges of individual species due to weak species definitions for each and limited occurrence data for nearly all of them. Also, we exclude Eurycanthablatta due to systematic uncertainty.

The optimal model setting was LQPT 1 and T 4.5 according to the AICc and cross-validation criteria (Table SS1). However, only the model with the LQPT 1 configuration was statistically significant (highest average validation AUC: P < 0.01, percentile 10 of the omission rate: P = 0.045). Thus, this model is the one presented here (Figs 11, 12). We found that habitat suitability for Lamproblattidae is positively correlated with higher precipitation of wettest and warmest months (Fig. 11A, C). The suitability also increased in regions with precipitation ranging from 500 to 1500 mm in the coldest quarter of the year (Fig. 11D). Conversely, suitability decreased in regions with the driest month having rainfall more than 100 mm (Fig. 11B). Lamproblattidae exhibited low tolerance to temperature variation. The model showed a drastic suitability decrease in regions with daily temperature variation greater than 8°C (Fig. 11E) and annual standard deviation greater than 2°C (Fig. 11). Additionally, the suitable altitude from Lamproblattidae ranges from 0 to 2500 m above sea level (Fig. 11G).

The prediction of habitat suitability for Lamproblattidae is shown in Figure 12. Suitable areas for Lamproblattidae are bounded by the Sierra Madre del Sur Province in Central America in the North, by the Atlantic Ocean in the East, by the Andes highlands in areas south of El Parramo Province in the West, and by the Chaco and Pampanean Province in the South (classification according to Morrone et al. 2022). The modelled distribution is mostly tropical and subtropical lowland forests with the highest suitability in Central America, Cuba, northern South America, the Atlantic coast of South America (including the Atlantic Forest region), and an area around Tinis National Park and Carrasco National Park in Bolivia. Numerous islands are also suitable areas for Lamproblattidae, both in the Atlantic and Pacific oceans (i.e., San Cristobal Island and Española Island). In Central America, the suitable areas for Lamproblattidae include both the Pacific and Atlantic coasts. However, in South America, they generally occur only on the Atlantic coastline (but see Ecuador coast). Surprisingly, the coastal regions of the Atlantic Forest are predicted to have high habitat suitability, despite few, if any, known records in that biome.

On 1 July 2021, a few hours after sunset, we observed one Lamproblatta mimadelfi (AUDE-PE-7-56, adult female) on the forest floor inside a small burrow along with two superficially similar juvenile cockroaches of an undetermined species (one we think is a Blattellidae, the second was not captured). During the time we observed these cockroaches, the adult female would emerge from the hole, walk freely on the ground, and retreat back into the original hole, or another nearby hole, when disturbed. We captured this on video (https://www.youtube.com/playlist?list=PLI7KLkHwkPbX0MI0_Fc2dWIGjS85lKggv; https://doi.org/10.5061/dryad.0000000g0; Fig. 13). Upon further investigation, we discovered 23 total holes in the immediate vicinity (4 m²; Fig. 14). By blowing air into the holes with an insect-collecting aspirator, we determined that many shared the same connecting tunnel. Most tunnels seemed to have only two openings, but at least one had three openings.

We excavated a few tunnels by hand, paying special attention to the tunnel where we originally observed the adult female Lamproblatta mimadelfi, which we will refer to as the “burrow”. Fig. 14 illustrates the burrow, which was about 15 inches long and reached a depth of about 13 cm below the ground’s surface. The burrow had two entrances that we were able to identify. The tunnel was 1–1.5 cm wide and passed underneath the root of a tree. We exhumed the portion of the root that intersected with the burrow. The root tissue was living and did not appear to have any insect damage or visible fungal rot.

It is possible, but perhaps unlikely, that the burrow was more extensive, and our excavation obscured branching points. We have no evidence that this burrow was created by the Lamproblatta mimadelfi individual, or that the individual occupied it prior to us approaching. The remaining tunnels had similar shapes and proportions, and roughly half of them were also along other large roots of the same tree. Observations by Gautier and Deleporte (1986) corroborate that Lamproblatta spp. have consistent burrows, and those burrows may be in soil.

Gut metabarcode data (10.5061/dryad.0000000g0) from L. mimadelfi suggests they feed on a wide variety of fungi (Table 5; Fig. 15). While primers targeting plant and animal DNA did have some meaningful hits in L. mimadelfi and other taxa (Fig. 15), they were so few that we hesitate to draw conclusions from them. For the animal primers at least, we cannot rule out that this is due to degradation of animal tissue in the gut, as opposed to an entirely fungivorous diet. Imaging of gut contents (Fig. 16) certainly shows insect parts (Fig. 16D) and possibly shows solid plant parts as well (Fig. 16F). Beta-diversity analyses did not reveal any variance in diet by location, although we had only a handful of samples from each locality. Fungal diet diversity (Inv. Simpson and Chao) was similar to what we observed in Dendroblatta spp. and Neoblattellini spp. but lower than some other cockroaches we sequenced (e.g., cf. Xestoblatta spp., Neoblattellini spp., and Periplaneta spp.)

The most comprehensive review of Lamproblattidae systematics was done by Roth (2003). This work, being published near his death, was incomplete, largely relied on superficial characteristics of other taxonomists (Rehn 1930), only briefly acknowledged work by McKittrick (1964, 1965) and McKittrick and Mackerras (1965), and did not consider the detailed morphological work done by Klass (1997). The present work has now achieved this, as well as integrating all this with newer systematics findings (Fritzsche et al. 2008; Evangelista et al. 2019a; Evangelista et al. 2019b). Yet, the ongoing saga of Anaplectidae (Djernaes et al. 2015; Legendre et al. 2015; Evangelista et al. 2018; Djernæs and Murienne 2022; Evangelista et al. 2023; Liu et al. 2023; Evangelista et al. 2024) and their potential close relationship to Lamproblattidae (Fig. 7), or the possibility of new family-level lineages in Kittrickea (Evangelista, pers. obs.; Fritzsche et al. 2008; Herrera et al. 2026) would necessitate further systematic revisions. We leave these large undertakings for future work, but we do make some progress here.

We show weak morphological support for Anaplectidae within Kittrickea (Fig. 7) without even invoking the shared SA plate morphology between Lamproglandifera and Anaplectidae (see results section). We also further establish the monophyly of Lamproblattidae s. str. (ignoring the placement of Lamproblatta meridionalis due to missing data) and demonstrate the usefulness of the SA plate morphology for systematics.

While the iNaturalist observations comprise only 10% of the total Lamproblatta spp. observations, they provide among the largest geographical spread of observations. Observations from the literature are numerous, but highly clustered in a handful of study areas (e.g., 133 observations by Gautier and Deleporte 1986 from a single site). Thus, iNaturalist observations were among the most useful data sources to predict range limits through niche modelling (Fig. 12).

The high habitat suitability for Lamproblattidae in Central America, and the northern region of South America is unsurprising considering the high number of known occurrences in these areas. Similarly, L. albipalpus is known to be invasive in Cuba (Gutierrez 1995). More surprising are the southern hotspots, which suggest that Bolivia and Brazil’s Atlantic Forest region are suitable habitats for Lamproblattidae, perhaps harboring undiscovered diversity. Yet, the latter species may already be extinct, as the Atlantic Forest is an endangered biome facing fragmentation in Brazil (Ribeiro et al. 2009). Revisions in natural history collections are necessary to verify these hypothetical species’ existence (Anderson 2012). These collections provide historical data on species distribution, aiding in confirmed true absence in localities to improve model performance (Barbet-Massin et al. 2012). Our model considered only meteorological parameters, but not habitat loss and fragmentation (important variables to determine current distribution patterns). Future ecological models should include habitat loss, climate change projections, and genetic information to study current and future distribution variation (Segan et al. 2016; Ikeda et al. 2017; Van Moorter et al. 2023). We did not include soil variables in our analysis, although they may prove to be informative for their distributions. Our own sampling in the coastal rainforests of Guyana did not yield any Lamproblatta spp. (Evangelista et al. 2014), but they are known from the interior of the country (Evangelista et al. 2016). This is relevant because of the high sand content of soils in coastal Guyana in comparison to the interior. Considering the potential importance of soil burrowing to their lifecycle (Gautier and Deleporte 1986) a loam with more structural integrity may be necessary for their habitat suitability. Similarly, this may be one of the driving factors behind our finding that they require low to moderate precipitation during the driest and coldest parts of the year. Perhaps increased soil saturation with water would make it difficult to maintain effective burrows.

Lamproblattidae are among the closest relatives of Xylophagodea (Legendre and Grandcolas 2018; Evangelista et al. 2019b; potentially, along with Anaplectidae, Eurycanthablatta, or other undescribed taxa; Fig. 7). Currently, our best way to understand the transitions from an omnivorous, solitary cockroach to a eusocial xylophagous termite is to study the close relatives of termites in close succession. In this light, much work has been done on Cryptocercus (e.g., Nalepa et al. 1997; Nalepa 1999; Grandcolas et al. 2001; Nalepa 2003; Kambhampati and Peterson 2007; Klass et al. 2008; Carpenter et al. 2010; Che et al. 2016; Yaguchi et al. 2021), and now we comprehensively look at Lamproblattidae. Nalepa (2003) discussed the importance of fungi in predigesting wood and preparing the microclimate of logs for suitability by Cryptocercus. Beza-Beza et al. (2024) reviewed evolutionary pathways towards wood-feeding but surmised that the transition from xylomycophagy to xylophagy would be unlikely due to the very different physiological requirements of each feeding mode. However, our findings here suggest this is an equally probable scenario as the alternative (xylomycophagy being derived in Lamproblattidae).

With current knowledge, there is no way to rule out either of these two hypotheses. If a lineage could be reliably placed as sister to Lamproblattidae + Xylophagodea (e.g., Eurycanthablatta, Anaplectidae, or a new family) and a consistent diet could be reliably assessed (e.g., entirely xylomycophagous or not) then it would shift confidence towards or away from the evolutionary story presented below. The following scenario assumes that xylomycophagy is not derived in Lamproblattidae, and that the natural history of living Lamproblattidae is indicative of the natural history of their common ancestor with Xylophagodea (i.e., the ancestor was xylomycophagous). With these big assumptions in mind, we might consider the following. 150 mya (Evangelista et al. 2019b) the ancestor of Xylophagodea was likely a moderately large, black, free-living cockroach. North America was in the process of splitting from Pangea and South America was still connected to Africa (Scotese 2001). We currently cannot know where this ancestral population would have been distributed, but based on the suitable habitats for Lamproblattidae, it may have been a relatively warm, low elevation (< 2500 m), wet habitat subject to seasonal precipitation and a relatively narrow variation of temperature. It was likely a tropical or semi-tropical wet forest. It would have been solitary or “weakly gregarious” sensu Gautier and Deleporte (1986), having consistent burrows in soil or wood. Its parental care behaviors are largely a mystery, but we do have evidence that it first would (A) lay its ootheca, (B) dig a cradle, and then (C) deposit its ootheca in the cradle (as opposed to B, A, C, which is done outside of Kittrickea; McKittrick 1964; Evangelista et al. 2019b). It would have fed opportunistically on dead organic matter. A significant portion of this diet may have been eating wood-associated fungus at an early to intermediate stage of wood-decay. As such, pre-existing Parabasalia and/or Preaxostyla in the hind-gut would have co-adapted with the cockroach to gain more nutritional benefits from eating the wood-fungus “sandwich”.

The largest caveat to these findings would be changes to the phylogeny of Kittrickea via inclusion of Eurycanthablatta, a different placement of Lamproglandifera (Herrera et al. 2026), characterization of Anaplectidae diets, or the existence of yet more undescribed family-level lineages. These could either reinforce the evolutionary scenarios described above or undermine them. Gaining a robust insight to the evolutionary precedents to xylophagy and sociality will require decades more study to resolve some of the big phylogenetic questions, behavioral and dietary studies of more taxa, and more surveys of cockroaches and parabasalians in nature. Also, our scenario suggests that fungal interactions in rotting wood may be more important to the ecology of living Cryptocercidae, and this should be further investigated.