A remarkable legion of guests: Diversity and host specificity of army ant symbionts

Tropical rainforests are among the most diverse biomes on Earth. While species inventories are far from complete for any tropical rainforest, even less is known about the intricate species interactions that form the basis of these ecological communities. One fascinating but poorly studied example are the symbiotic associations between army ants and their rich assemblages of parasitic arthropod guests. Hundreds of these guests, or myrmecophiles, have been taxonomically described. However, because previous work has mainly been based on haphazard collections from disjunct populations, it remains challenging to define species boundaries. We therefore know little about the species richness, abundance and host specificity of most guests in any given population, which is crucial to understand co‐evolutionary and ecological dynamics. Here, we report a quantitative community survey of myrmecophiles parasitizing the six sympatric Eciton army ant species in a Costa Rican rainforest. Combining DNA barcoding with morphological identification of over 2,000 specimens, we discovered 62 species, including 49 beetles, 11 flies, one millipede and one silverfish. At least 14 of these species were new to science. Ecological network analysis revealed a clear signal of host partitioning, and each Eciton species was host to both specialists and generalists. These varying degrees in host specificities translated into a moderate level of network specificity, highlighting the system's level of biotic pluralism in terms of biodiversity and interaction diversity. By providing vouchered DNA barcodes for army ant guest species, this study provides a baseline for future work on co‐evolutionary and ecological dynamics in these species‐rich host–symbiont networks across the Neotropical realm.


| INTRODUC TI ON
Habitat destruction is a major force of the dramatic biodiversity loss our planet is currently facing (United Nations Environment Programme, 2021). The most biodiverse terrestrial ecosystems-tropical forestscontinue to be degraded at alarming rates, primarily due to the human demand for wood and agricultural land (Crowther et al., 2015;Hoang & Kanemoto, 2021). This loss of tropical forests is widely expected to cause a massive extinction of species at the local scale, and probably also the global scale (Alroy, 2017;. The species-rich communities of tropical forests form complex interaction networks, for example between predators and prey (Gripenberg et al., 2019;Hoenle et al., 2019), hosts and parasites (Esser et al., 2016;Lopes et al., 2020), or plants and frugivores (Menke et al., 2011;Vidal et al., 2013). Identifying highly connected keystone species in these networks is crucial for efficient conservation measurements because their loss can have dramatic effects on community stability by causing the coextinction of many affiliated species (Dunne et al., 2002).
Among social insects, the massive colonies of army ants host an exceptionally species-rich and taxonomically diverse guest fauna, including beetles, mites, spiders, silverfish, millipedes, flies and wasps (Gotwald Jr, 1995;Kronauer, 2020;Rettenmeyer, 1961). Carl Rettenmeyer and colleagues listed about 50 insect guests of the army ant Eciton burchellii (Westwood, 1842) that have been collected either in and around army ants' temporary nesting sites (bivouacs) or in colony emigrations (Rettenmeyer et al., 2011). This outstanding diversity might be explained by the area-diversity relationship of the theory of island biogeography (Darlington Jr, 1957;MacArthur & Wilson, 1967; Mittelbach, 2012)-a pervasive ecological pattern that is applicable to partially isolated ecosystems such as social insect colonies (Gotwald Jr, 1995;Hölldobler & Wilson, 1990;Kronauer & Pierce, 2011;Parmentier et al., 2020;Wilson, 1971). According to the theory (MacArthur & Wilson, 1967), large social insect colonies such as those of army ants are expected to harbour a high diversity of associated species, partly because of high microhabitat heterogeneity and low symbiont extinction rates (Gotwald Jr, 1995;Hölldobler & Wilson, 1990).
Furthermore, such lists often suffer from sampling bias because symbionts of more abundant host species are generally overrepresented (Poulin & Morand, 2014). We have previously reported the first community-based host specificities of army ant guests, but these studies focused on specific taxonomic groups (von Beeren, Maruyama, & Kronauer, 2016a, 2016bTishechkin et al., 2017). As a result, community structure and interaction specificities at the level of the entire army antmyrmecophile community remain unknown.
Here, we present a comprehensive study encompassing an entire army ant-myrmecophile community. Over the course of four years, we systematically collected myrmecophiles from 70 colonies across all six Eciton Latreille, 1804 army ant species occurring at La Selva Biological Station, Costa Rica. Because myrmecophile identification was challenging due to the presence of morphologically extremely similar species (von Beeren et al., 2016a(von Beeren et al., , 2016bTishechkin et al., 2017), we used a molecular approach by supplementing morphological identifications with extensive DNA barcoding to define species boundaries. We then studied community structure as well as host specificity at the species-and at the network-level using weighted ecological network analyses (Blüthgen et al., 2008;Ings et al., 2009;Ivens et al., 2016;Poulin, 2010).

| Collection protocol and research permits
The present study is a synopsis of our work on Eciton myrmeco-  For reasons of feasibility, we restricted our systematic community analysis to myrmecophile species participating in army ant colony emigrations ( Figure 1). Some of the species participating in emigrations can also be found in raids and/or refuse deposits. Species that also participate in raids include myrmecoid Ecitophya and Ecitomorpha rove beetles, as well as the ptiliid Cephaloplectus mus. In addition to standardized collections from emigrations, we haphazardly collected myrmecophile specimens from Eciton raids and refuse deposits.
While these specimens were not examined systematically, we integrated some of them in the current analysis for the sole purpose of increasing sample sizes in genetic and morphological assessments of species boundaries (see Table S1 for a full list). We excluded mites from the current study due to the difficulties of collecting them, their TA B L E 1 Collection information and network specificity parameters Note: Given is the number of colonies from which myrmecophiles were collected for network analyses. Because some colonies were sampled more than once, the number of emigrations is sometimes higher than the number of colonies. Number of specimens describes the total number of myrmecophiles used for morphological and genetic species identifications, which contains specimens from additional colonies including collections from army ant raids and refuse deposits. Numbers in parentheses give the total numbers of myrmecophile specimens used for network analysis. Myrmecophile diversity is given as total number of species, the total number of species separated per arthropod order, as effective Shannon diversity (e H ; Jost, 2006), and as the mean of 100 rarefied e H (e H rare ).
Exclusiveness of myrmecophiles is given as Kullback-Leibler distance (d′; Blüthgen et al., 2006). Network incidences (Inc.) give the total number of occurrences of different myrmecophile species across all colonies of an army ant species.
high diversity and their complicated taxonomy (e.g., Berghoff et al., 2009). Many mites are diminutive and attach to various body parts of army ant workers (Gotwald Jr, 1995;Rettenmeyer, 1963). Conducting a reliable community analysis of mite diversity and host specificity would have required a careful inspection of army ant bodies of hundreds of workers per colony (e.g., Berghoff et al., 2009), followed by a challenging and time-consuming identification process of mite taxa.
This would have been beyond the available resources for this study.
However, with the vouchered material we hope that a communitywide study on mite diversity and host specificity can eventually be included in future work. Furthermore, we did not purposely collect aerial parasitoids hovering over the army ants (Brown & Feener, 1998) because our search was focused on the marching ants and their associates on the ground.
Initially, we followed and intentionally resampled a few colonies, mostly of E. burchellii. This is the species with the largest colonies and the highest ant traffic, making it particularly challenging to spot and collect myrmecophiles (Table 1; Table S1). This resampling indicated that our approach was efficient in collecting most myrmecophile specimens occurring in a colony from a single colony emigration (see Table S1). For example, we collected 24 adult and three larval Vatesus cf. clypeatus sp. 2 specimens from the first observed emigration of the colony EB03 (Table S1). On the next day, we resampled guests from the emigration of the same colony and collected only four additional Vatesus adults and no Vatesus larvae. However, it is impossible to collect all myrmecophiles from a colony with certainty, and we almost certainly missed myrmecophiles in any given colony emigration, especially in those that we did not observe from the very beginning. Nonetheless, our resampling data indicate that,
To streamline our identification process, we used the following three-step protocol. We first applied a morphological prescreening by sorting the collected myrmecophiles per colony to morphospecies, which we defined as morphologically similar specimens that were indistinguishable to us, even with the help of the latest species keys for the group. As a second step, we used, if available, at least five specimens per morphospecies and colony for DNA barcoding.
Initially we had difficulties distinguishing phorid fly species, and we thus used, if available, at least 10 phorid fly specimens per colony for DNA barcoding. This sampling protocol implied that we did not comprehensively collect abundance data of myrmecophiles (number of specimens per colony) in the present study, even though such data were previously reported for certain taxa (von Beeren et al., 2016a(von Beeren et al., , 2018Tishechkin et al., 2017). We did not incorporate abundance data in the present work because genetic analyses were necessary to distinguish cryptic taxa (e.g., in certain phorid flies; see Results), which would have required barcoding hundreds of specimens per colony. However, we do provide abundance data for those species where fewer than five specimens were collected from a colony.
As a third step, we compared our morphological identifications with genetic clustering results. In most cases the two approaches agreed on the determination of species boundaries. In a few cases, however, contentious results arose as two or more genetic clusters could not be distinguished morphologically. We tried to resolve such controversies by acquiring both additional genetic information from  Table S1. We used the species key for Costa Rican ants of Longino (2010) to identify army ant species.
We extracted DNA from 2,432 myrmecophile specimens using Qiagen DNeasy Tissue Kits, either for single extractions or for 96well plates. We followed the standard protocol with two exceptions: specimens were not homogenized but kept intact and protein digestion was shortened to 2-3 hrs. Except for a few larval specimens and some phorid flies, this procedure preserved specimens in good  Tables S2 and S3). Purification and sequencing of PCR products were outsourced to Macrogen USA and to Macrogen Europe. PCR amplicons were sequenced in forward and reverse directions using Sanger sequencing, which allowed us to link each DNA barcode to a voucher specimen. In cases of low-quality reads, PCR settings were adjusted, and sequencing was repeated.
For sequence analyses we used the laboratory information management system geneious prime programmed by biomatters (version 2020.1; https://www.genei ous.com including the plugin "biocode"; version 3.0.7; Parker et al., 2012). This included assembly of forward and reverse sequences, sequence trimming, sequence editing, sequence alignment using the muscle algorithm (Edgar, 2004), and clustering analyses. We performed several quality checks with the resulting consensus sequences. Sequences with stop codons in the COI alignment were omitted from further analyses as they probably represented nuclear mitochondrial pseudogenes. We compared barcoding results to morphological identifications to detect and omit apparently erroneous sequences due to contamination or pipetting errors (<1% of DNA barcodes). Final consensus sequences were deposited at the Barcode of Life Database systems (BOLD; GenBank accession numbers are given in Table S1).
In a few cases, COI analyses resulted in distinct genetic clusters that we morphologically identified as the same species. In such cases we additionally analysed portions of the nuclear gene wingless (wg) for a subset of specimens with different COI haplotypes (Table S1). Congruency in clustering analyses of mitochondrial sequences (COI) and nuclear sequences (wg), meaning that specimens from distinct COI clusters had distinct wg alleles, was interpreted as support for the coexistence of distinct species. In beetles, specimens of these candidate species were anatomically inspected in more detail to search for diagnostic morphological characters (e.g., aedeagi dissections: von Beeren et al., 2016aBeeren et al., , 2016b. In phorid flies, we relied on genetic species assessment, and a more in-depth morphological analysis remains for future work. In contrast, discordance of genetic clustering, meaning divergent COI clusters shared the same wg sequences, was interpreted as support for the presence of a single species (see also von Beeren et al., 2015Beeren et al., , 2016b. As a consistent amplification of high-quality sequences of the full wg fragment targeted by the commonly used wg insect primers wg550F and wgAbrZ (Wild & Maddison, 2008) failed in phorid flies, we decided to analyse two shorter, nonoverlapping sequences of wg.
Those two portions were located at the start and at the end of the full fragment and produced high-quality signals. The excluded portion in the middle part of the full fragment included an intron, which showed low sequence quality scores in many specimens. We denote these two portions here as wingless gene fragment I and wingless gene fragment II (for primers see Tables S2 and S3). In previous work we additionally analysed a portion of the carbamoyl-phosphate synthetase II gene of the multidomain enzyme CAD (CAD) to determine species boundaries in the genera Ecitophya, Ecitomorpha, Vatesus and Tetradonia (von Beeren et al., 2016a(von Beeren et al., , 2016b(von Beeren et al., , 2018.
To identify distinct genetic units in myrmecophile DNA barcodes, we applied the standardized sequence clustering algorithm RESL, which is implemented in the BOLD systems (Ratnasingham & Hebert, 2013

| Deposition of specimens, images and DNA extracts
We deposited 379 specimens of 47 species at 15 museum collections. All other specimens and all DNA extracts are currently stored at −30℃ in C.v.B.'s personal collection at the TU Darmstadt Insect Collection (contact C.v.B. for access). Details on specimen depositories are given in Table S1 and on the BOLD website (http://www. bolds ystems.org/).
In addition to specimen deposition, we uploaded 2,206 focusstacked voucher images of 497 myrmecophile specimens to the BOLD systems (Table S1). Images were taken with three different setups.

| Network statistics
We analysed the diversity and interaction specificity of the army ant-myrmecophile community using standardized analytical metrics provided by ecological network theory (Blüthgen, 2010;Ivens et al., 2016). Overall, 2,113 myrmecophile specimens were included in the evaluation of species-and network-level specificity (Tables 1 and 2; Table S1). The network analysis of the present study was based on a quantitative bipartite interaction matrix (Table S1). An "incidence" was defined as a myrmecophile species occurring in a colony of an army ant species. The "link strength" between an army ant species and a myrmecophile species represented the number of times a given myrmecophile species was detected in different colonies of that host. In other words, the "link strength" summarizes the incidences between an army ant species and a myrmecophile species, serving as a measure of colony infestation frequency in the population. This means that specimens of the same species collected from the same colony were only represented by a single incidence count.
The network therefore contained incidence data of spatiotemporally independent collection events, which yields a conservative estimate of host specificity (Blüthgen et al., 2006;Hoenle et al., 2019). This quantification is important because unweighted networks often underestimate interaction specificity (Blüthgen, 2010;Blüthgen et al., 2006;Hoenle et al., 2019). For instance, when a myrmecophile associates primarily with one host species but is occasionally found with another host, relying solely on presence/absence data would underestimate its true level of host specificity (Blüthgen, 2010;Blüthgen et al., 2006;Hoenle et al., 2019;Poulin et al., 2011). All network statistics were analysed in R (R Core Team, 2020) using the package "bipartite" (Dormann et al., 2008; version 2.08).
Based on the bipartite network, we analysed the network-level specificity and the species-level specificity by calculating the twodimensional Shannon entropy H 2 ′ and the Kullback-Leibler distance d′, respectively. These weighted metrics have several benefits over unweighted network analysis (Blüthgen, 2010;Blüthgen et al., 2006). Values for H 2 ′ and d′ are normalized relative to minimum and maximum possible values, thus ranging from 0 (lowest interaction specificity) to 1 (highest interaction specificity) (Blüthgen et al., 2006 Blüthgen et al., 2006). One example of a rare species with a relatively high d′ value is the histerid Aphanister sp. 1 (d′ = 0.27; N = 1 specimen), which was collected from the rare host E. lucanoides (Tables 1 and 2). Low d′ values are found in host generalists as well as in rare species that are associated with common hosts. For instance, only one specimen of the histerid beetle Aemulister hirsuta was collected in one colony of the more abundant host E. mexicanum. Accordingly, its host specificity was comparably low (d′ = 0.07; Table S1). We tested the metric H 2 ′ against null models based on 10,000 randomized networks using the Patefield algorithm (Blüthgen et al., 2006;Patefield, 1981).
Additionally, we calculated the effective Shannon diversity of interaction partners per species (e H ; Jost, 2006). This metric takes the richness and evenness of interactions into account and thus suitably characterizes link strengths for each species (Blüthgen, 2010).
Furthermore, we calculated the rarefied Shannon diversity (e H rare ) based on 100 permutations of 33 incidences, representing the lowest incidence number of a host species in the present network (E. lucanoides; Table 1). The rarefied metric e H rare improves comparability and accounts for variation in sample sizes between army ant species ( Table 1). Note that myrmecophile rarity is not considered in this metric, and assessment of host specificity needs to be related to sample size when referring to e H . For instance, the histerid beetle Aemulister hirsuta had an e H value of 1.00 host species, which could indicate high host specificity. However, because only one individual was collected, the actual host range and host specificity remains elusive for this species. Table 2 provides a summary of myrmecophile sample sizes for each species, allowing the reader to evaluate host specificity in the context of myrmecophile rarity.
Besides examining specificity and diversity, we used two complementary approaches to explore the extent to which myrmecophiles partition the available Eciton host niche space. First, we measured the degree of network modularity (metric Q calculated by the QuanBiMo algorithm; Dormann & Strauss, 2014). This metric quantifies to what extent data support the division of a network into modules (Dormann & Strauss, 2014). Modules are characterized by a high density of within-module links and few to no between-module links. The metric Q is normalized and ranges from 0 (no more links within modules than expected by chance) to 1 (perfectly modular networks). We tested Q against randomized null models as described previously (Hoenle et al., 2019;Schleuning et al., 2014). Second, we tested whether myrmecophile faunas differed between army ant species pairs by comparing H 2 ′ values of each army ant species pair to 1,000 randomized networks. Species pairs consisted of two army ant species and all their associated myrmecophiles (see also Wehner et al., 2018).
The literature on host associations (e.g., Ivens et al., 2016;Kistner, 1982;Kistner & Jacobson, 1990) and additional collections of army ant associates by us indicate that the selected Ecitonmyrmecophile network represents an almost closed network (or a well-defined module) within the entire army ant-myrmecophile network at LSBS. In other words, the myrmecophiles of Eciton army ants generally do not occur in colonies of other army ant genera. This is TA B L E 2 Overview of Eciton-associated myrmecophiles and their host specificity Note: Sample size (N) gives the total number of myrmecophile specimens per species used in network analysis. Note that additional specimens collected from army ant raids and refuse deposits were studied for species identifications (Table 1; Table S1). Network incidence (Inc.) indicates the number of times a myrmecophile was found in different host ant colonies. Diversity calculations are based on the bipartite army ant-myrmecophile interaction network (matrix provided in Table S1). Host diversity is given by the absolute number of host species (Hosts) and as effective Shannon diversity (e H hosts ). Exclusiveness of host associations is given as standardized Kullback-Leibler distance (d′). Two species were denoted here as "false Lomechusini," which is a group of Neotropical rove beetles formally placed in the tribe Lomechusini (Elven et al., 2012). Asterisks (*) mark either species that have already been scientifically described as part of this project, or species where we have evidence via morphological inspection (including aedeagi dissections) that they represent species not yet scientifically described. Taxonomic species descriptions are in progress/planned for the following species: Aphanister sp. 1 (AKT), Clientister sp. 1 (AKT), Euclasea sp. 1 (AKT), Sacosternum aff. lebbinorum (M. Fikáček), the two Ecitomorpha species (MM), and all Vatesus species (CvB). The two Ecitomorpha species have erroneously been synonymized by Kistner under the species name Ecitomorpha arachnoides and need to be formally re-erected to species status (see von Beeren et al., 2018). In the two species complexes Vatesus cf. clypeatus and Ecitophora cf. comes, at least two of the three cryptic species are new to science. A careful taxonomic comparison to type material is necessary to evaluate this, which was beyond the scope of this study. A taxonomic revision of the subfamily Cephaloplectinae and Nossidiinae is currently in progress (WEH), including the taxonomically challenging ptiliid genera Limulodes and Nossidium. Abbreviations: EB = E. burchellii, ED = E. dulcium, EH = E. hamatum, EM = E. mexicanum, EL = E. lucanoides, EV = E. vagans.

TA B L E 2 (Continued)
of all six local Eciton species at LSBS (von Beeren et al., 2016b). This species was also found in one Nomamyrmex esenbeckii raid and one Neivamyrmex gibbatus emigration (Table S1).

| Species identification and diversity of army ant myrmecophiles
We identified 2,355 myrmecophile specimens to the species level (Table S1). Of these, 61 specimens were identified based solely on morphological characters, while DNA barcodes were available for all other specimens as an additional character for identification (Table S1). Seventy-seven specimens could not be identified to the species level because DNA barcoding failed and because they belonged to taxa that we were not able to identify via morphology alone (54 phorid flies, 21 ptiliid beetles, two staphylinid beetles; Table S1).

Overall, we identified 62 myrmecophile species participating in
Eciton army ant emigrations at LSBS: 25 rove beetles (Staphylinidae), 17 clown beetles (Histeridae), six featherwing beetles (Ptiliidae), one water scavenger beetle (Hydrophilidae), 11 scuttle flies (Phoridae), one millipede (Stylodesmidae) and one silverfish (Nicoletiidae) ( Table 2). Among these, at least 14 species were new to science (Table 2). New species discoveries came in essentially two types: F I G U R E 2 Molecular species identification in beetles. RAxML trees based on COI barcode data of staphylinid (rove) beetles, histerid (clown) beetles and ptiliid (feather-winged) beetles. Grey boxes show cases where morphological identification and DNA barcode clustering agreed on the presence of a single species. Green boxes depict singletons. Red and purple boxes highlight cases in which specimens initially identified as a single species split into two or more COI clusters. Additional morphological and genetic data suggested that those specimens either belonged to a single species (red boxes) or to different species (purple boxes; see also Supplementary Results). Scale bars show expected nucleotide substitutions per site as inferred by the RAxML algorithm. Bootstrap support values are shown at major nodes (1,000 repetitions) cryptic species resolved by DNA barcoding, and new species that were morphologically clearly distinguishable but had not been collected or scientifically described.

Histerid beetles
We obtained 2,294 high-quality COI consensus sequences of army ant myrmecophiles (mean sequence length = 636 bp; range sequence lengths = 157-700 bp; Table S1). Amplification of COI repeatedly failed, or sequences showed low quality or stop codons in 138 specimens resulting in a 6% failure rate in DNA barcoding. BOLD's clustering algorithm RESL grouped 2,287 myrmecophile barcodes into 66 BINs, while no BIN was assigned to seven DNA barcodes due to short sequence lengths (Table S1).
For 52 of these BINs, genetic clustering agreed with our morphological species identifications, meaning that morphologically distinguishable species were clustered in a single BIN. These BINs were also recovered in the RAxML tree analyses (Figures 2 and 3).
Interestingly, species of the two genera Tetradonia and Ecitophora showed a high intrageneric variability in host specificities. Host associations ranged from two to six species among the five Tetradonia rove beetles (Table 2; Table S1), with lowest host specificity in the species T. laticeps (e H : 4.92, d′: 0.08; N = 151 specimens from 28 Eciton colonies; Table 2) and highest host specificity in the species T.

| Network-level specificity and network modularity
The associations between army ants and their guests differed from purely random associations (H 2 ′ tested against 10,000 randomized networks: p < .001; incidences in the network matrix based on 2,113 specimens and 70 Eciton colonies; Table S1) and showed an overall moderate level of network specificity (H 2 ′ = 0.47; Figure 4). This is also reflected in a moderate level of interaction specificity or exclusiveness of interactions of each army ant species (d′: 0.38-0.53; Table 1), resulting from the fact that each Eciton species harboured myrmecophiles covering the entire range from host specialists to host generalists (Figure 4).
The army ant-myrmecophile network was significantly modular (Q tested against 10,000 randomized networks: p < .001) and showed a moderate degree of modularity (Q = 0.43; Figure S3). Each army ant species formed a single module within the network, except for the species pair E. mexicanum and E. dulcium, demonstrating a certain degree of host partitioning among myrmecophile species ( Figure S3). We additionally tested for host partitioning by comparing network subsets consisting of two army ant species and all their associated myrmecophiles against null models of these network subsets. Each of the 15 possible network subsets differed significantly from randomized network models, showing that every Eciton species hosted a composition of myrmecophiles distinct from other Eciton species (all pairwise comparisons: p ≤ .018; 10,000 permutations).

| DISCUSS ION
The core of the present study was a biodiversity inventory of an army ant-symbiont community in a single population, which allowed us to describe the network of host-symbiont species interactions in unprecedented detail. Our survey unearthed numerous new army ant guest species. We have already described three of these species taxonomically (von Beeren et al., 2016b;, and we have erected one subspecies to the species level . One of the new species is the charismatic histerid beetle Nymphister kronaueri with its exceptional mechanism of phoresy (von Beeren & Tishechkin, 2017) ( Figure 1b). The discovery of many new species-in fact every fifth species in our survey was unknown to science (Table 2)-might at first seem surprising, in particular because LSBS is one of the best-studied tropical field sites and researchers have repeatedly collected army ant myrmecophiles there (e.g., Disney & Rettenmeyer, 2007;Jacobson & Kistner, 1998;Kistner & Jacobson, 1990;Kistner & Mooney, 2011;Tishechkin, 2007). However, we did expect to discover many new species due to the integration of DNA barcoding, a tool that facilitates the discovery of new species in taxonomically challenging groups (Hajibabaei et al., 2006;Hebert et al., 2004;Pečnikar & Buzan, 2014). In army ant guests, this molecular technique had previously only been applied to very few selected taxa (Caterino & Tishechkin, F I G U R E 3 Molecular species identification in phorid flies. RAxML tree based on COI barcode data of phorid (hump-backed) flies. Grey boxes show cases where morphological identification and DNA barcode clustering agreed on the presence of a single species. The green box depicts a singleton. Red and purple boxes highlight cases in which specimens initially identified as a single species split into two or more COI clusters. Additional genetic data suggested that those specimens either belonged to a single species (red boxes) or to different species (purple boxes; see also 2006; Pérez-Espona et al., 2017). One example of cryptic species detected by DNA barcoding in our community survey were the three species included in the Vatesus cf. clypeatus complex (Figure 2). Each species had a high host specificity ( Figure 4; identifications and network analyses can be efficiently combined. The integrative approach revealed considerable variation in host specificities among myrmecophile species. We found four species of host generalists-a silverfish, a ptiliid beetle, a phorid fly and a rove beetle-that associated with all Eciton host species in the community ( Figure 4; Table 2). Note that host specificity is a relative term and the herein detected generalist species might still be relatively host-specific at the genus or tribal level when compared to myrmecophile species that infiltrate colonies of different ant subfamilies (e.g., Molero-Baltanás et al., 2017). The majority of myrmecophiles were, however, only detected from colonies of a single Eciton species ( Figure 4; Table 2), resulting in a left-leaning host distribution curve ( Figure S2)-a typical pattern of host-symbiont systems (Combes, 2005;Poulin & Keeney, 2008;Poulin et al., 2006;Schmid-Hempel, 2011). This pattern partly arose because many species were rare. For those species, including for example most histerid beetles, the actual host specificity at the population level remains elusive. In these cases, historical collections from other populations can sometimes help to assess the host specificity in more detail (e.g., von Beeren Kistner, 1979;Kistner & Jacobson, 1990;Seevers & Dybas, 1943). For instance, we only collected four specimens of the histerid beetle Euxenister caroli in two E. burchellii colony emigrations (Table 2). However, additional collections from multiple Neotropical locations (>50 specimens from >11 E. burchellii colonies) corroborate that E. burchellii is indeed the species' preferred and possibly only host (Akre, 1968;Rettenmeyer, 1961).
Good examples of perfect host specialists-which we define here as myrmecophiles occurring regularly in host colonies of a single species-are the four ant-mimicking beetle species of the genera Ecitomorpha and Ecitophya (Figure 4; Table 2). Myrmecoid beetles show striking signs of specialization compared to their free-living relatives by mimicking host workers anatomically, behaviourally and chemically von Beeren et al., 2018;Maruyama et al., 2009;Maruyama & Parker, 2017;Parker, 2016). The evolution of such elaborate levels of specialization usually increases a symbiont's fitness on that particular host, but it comes at the cost of a reduced host range, which increases the risk of local coextinction events (Schmid-Hempel, 2011). As a consequence, specialists often coevolve and cospeciate with their primary host (Combes, 2005; Schmid-Hempel, 2011), which has recently been suggested for several species in the genera Ecitomorpha and Ecitophya (Pérez-Espona et al., 2017).
The high degree of host specificity in some of the studied myrmecophiles translated to strong signatures of host partitioning at the network level (Figure 4; Figure S3). A good example for host niche differentiation in closely related species are the four species of Vatesus beetles, which almost perfectly partitioned the available host niche space (Figure 4; Table 2). On the other hand, species in the rove beetle genus Tetradonia and the phorid fly genus Ecitophora showed substantial overlap in their host range ( Figure 4; Table 2).
This pluralism of host-myrmecophile interactions resulted in an overall moderate level of interaction specificity at the network level (H 2 ′ =0.47; Figure 4). Because H 2 ′ is a standardized metric, it allows for comparisons across communities and study systems (Blüthgen, 2010 (Brown & Feener, 1998;Glasier et al., 2018;Hölldobler & Wilson, 1990;Kistner, 1982;Molero-Baltanás et al., 2017;Parker, 2016). Among army ant associates, facultative associations exist in the detritivores and scavengers occupying the army ant middens, while myrmecophiles that have specifically adapted to follow the ants' emigrations are thought to form obligate associations (Gotwald Jr, 1995;Rettenmeyer et al., 2011). As we focused on emigrationfollowing myrmecophiles, we expected most myrmecophiles would be obligate associates. However, at least some of the detected species might in fact form facultative associations, which seems particularly likely for several of the low-density myrmecophiles (see also Molero-Baltanás et al., 2017). For instance, we report here the first record of a millipede (Calymmodesmus montanus) in Eciton emigrations. These unusual guests are commonly found in army ant colonies of the genera Nomamyrmex and Labidus and are considered scavengers that do not depend on the ants for survival, but occasionally follow army ant emigrations to take advantage of the abundant food resources (Rettenmeyer, 1962).
In the studied community, species of two genera, Tetradonia rove beetles and Ecitophora phorid flies, showed high intrageneric variability in host specificities ( Figure 4; Table 2). Apparently, phylogeny did not constrain the evolution of host specificities in these two genera and pre-adaptations allowed for both trajectories, the evolution towards host generalists and towards host specialists. As in the studied host-symbiont system, the underlying mechanisms leading to such variation in the degree of ecological specialization within a phylogenetic lineage remain unknown in the majority of cases, and understanding these mechanisms is still one of the most pressing issues in ecological and evolutionary parasitology (Poulin et al., 2011;Schluter, 2000;Thompson, 2005).
Adopting a phrase from E. O. Wilson, here we have unearthed "a remarkable legion of animal species" (Wilson, 1971, p. 389) that exploit the colonies of Eciton army ants in a Costa Rican community.
Like so many of their cohabitants in tropical ecosystems, this legion is threatened by severe human interference with the natural world.
The existence of many army ant associates depends on the presence of host ants (Gotwald Jr, 1995;Kronauer, 2020;Rettenmeyer, 1961). These, however, are sensitive to habitat degradation (Gotwald Jr, 1995;Kronauer, 2020) and local extinction of the ants will thus probably go hand in hand with an extinction cascade of numerous specialized, host-specific species, including many of the swarmfollowing birds and the diverse fauna of obligate myrmecophiles (Boswell et al., 1998;Brown & Feener, 1998;Koh et al., 2004;Kronauer, 2020;Kumar & O`Donnell, 2007;Pérez-Espona, 2021;Willis, 1974). Such coextinction cascades are indeed most severe in species forming tight symbiotic interactions, where at least one species depends entirely on the presence of one or a few others (Dunn et al., 2009). Here we have shown that many of the army antassociated guests are highly host-specific and parasitize a single or a few host ants. These specialists certainly face high coextinction risks when host species disappear locally. Hence, we must enhance our efforts to protect tropical rainforests if we want to preserve army ants and their marvelous symbiont fauna. Open Access funding enabled and organized by Projekt DEAL.

CO N FLI C T S O F I NTE R E S T S
The authors have no competing interests.

DATA AVA I L A B I L I T Y S TAT E M E N T
Sequences are deposited in GenBank and the Barcode of Life Data systems (for accession numbers see Table S1). Voucher specimens are deposited at 15 registered museum collections and the insect collection of the TU Darmstadt currently curated by C.v.B. (Table   S1). Specimen images are deposited at BOLD systems (see Table S1).