Three colonies of Odontotermes sp. were sampled from Linggang Village, Daguokou Town, Chuxiong City, Yunnan Province, China (24°54'46"N, 101°09'30"E, elevation: 1871 m), in January 2025. Three colonies of Macrotermes gilvus were sampled from Chiang Mai University Campus, Chiang Mai, Thailand (18°48'20"N, 98°57'09"E; elevation: 300 m), in April 2025 (Fig. 1A). During sampling, gentle lateral pressure was applied to remove the outer mound casing, minimising disturbance of the fungus combs and termites (Lee and Lee 2015). Intact, clean fungus combs, together with termites, were transferred into temporary plastic boxes (15 cm × 10 cm × 6 cm) that had been sterilised with 75% ethanol and exposed to UV light for 30 min before use. Comb material from each colony was divided amongst three boxes lined with sterile, moist tissue paper. To maintain microbial integrity and minimise loss of microbial biodiversity, isolation was performed upon return to the laboratory, within 3 hours. Termites were identified, based on the morphological characteristics of major soldiers, according to published taxonomic keys (Tsai and Chen 1963; Ahmad 1965).

From each colony, 30 samples of inner portions of mature combs (5 g each) were collected, yielding a total of 180 samples across the six colonies of Macrotermes gilvus and Odontotermes sp. (Fig. 1B). Microbial isolation followed a dilution-and-plating approach (Guswenrivo et al. 2018; Yang et al. 2025), with minor modifications. Each 5 g sample was homogenised in 45 ml of sterile phosphate-buffered saline (PBS: 1.44 g Na₂HPO₄, 0.24 g NaH₂PO₄, 0.2 g KCl, 8 g NaCl, dissolved in 1 litre of distilled H₂O), shaken for 1 min and then centrifuged at 4 °C and 1,000 rpm for 5 min. The resulting suspension was serially diluted to 10^–3 by transferring 1 ml of the suspension into 9 ml of PBS at each step. Subsequently, a 10-µl aliquot of the 10^–3 dilution was spread on to potato dextrose agar (PDA; without antibiotics) and nutrient agar (NA) plates using a sterile L-shaped spreader. Plates were then incubated at 27 °C for one week. Hyphal tips of fungi and small individual bacterial colonies were subcultured on to new PDA and NA plates, respectively (Yang et al. 2025). Bacterial strains isolated from combs of Odontotermes sp. were labelled “CX-Comb” (Suppl. material 1: table SS1), while those isolated from combs of Macrotermes gilvus were labelled “CMU-C” (Suppl. material 1: table S2); fungal strains isolated from combs of Odontotermes sp. were labelled “CX” (Colony 1: CX-CB; Colony 2: CX-CU; Colony 3: CX-CY) (Suppl. material 1: table S3) and those isolated from combs of Macrotermes gilvus were labelled “Cmu” (Colony 1: Cmu-a; Colony 2: Cmu-b; Colony 3: Cmu-c) (Suppl. material 1: table S4). All microbial isolation procedures were conducted in a laminar flow cabinet to prevent external contamination.

Isolation of bacteria and fungi from termite guts followed the procedures described in previous publications with slight modifications (Um et al. 2013; Yang et al. 2025). Active termites (15 major workers and 15 minor workers) were randomly collected from each colony, yielding a total of 180 gut samples across the six colonies of Macrotermes gilvus and Odontotermes sp. Each termite was surface-sterilised by immersion in 75% ethanol for 1 min, followed by three rinses with sterile distilled water and air-drying for 1 min in a laminar flow hood. The whole gut (Fig. 1B) was aseptically dissected using flame-sterilised fine-tip forceps, transferred into 500 µl of phosphate-buffered saline (PBS) and gently rinsed. The resulting suspensions were evenly spread on to PDA and NA plates in triplicate using a sterile L-shaped spreader. Plates were incubated at 27 °C for one week. Emerging fungal hyphae and bacterial colonies were subcultured on to fresh PDA and NA plates, respectively, to obtain pure isolates (Yang et al. 2025). Bacterial strains isolated from Odontotermes sp. were labelled “CX-Gut” (Suppl. material 1: table SS1), while those isolated from Macrotermes gilvus were labelled “CMU-g” (Suppl. material 1: table S2); fungal strains isolated from Odontotermes sp. were labelled “CX” (Colony 1: CX-QJ, Colony 2: CX-JJ, Colony 3: CX-QQ) (Suppl. material 1: table S5), and those isolated from Macrotermes gilvus were labelled “Cmu” (Colony 1: Cmu-ag, Colony 2: Cmu-bg, Colony 3: Cmu-cg) (Suppl. material 1: table S6).

To minimise contamination from external microbes, the outer portions of fresh termite fungus combs were discarded. To isolate comb-specialised Xylaria, inner portions of mature fungal combs were collected and incubated separately in sterile boxes containing moist sterile tissue. Fifty fragments of mature fungal combs (2 g each) were collected from each of five colonies — two of Odontotermes sp. (China) and three of Macrotermes gilvus (Thailand), resulting in a total of 250 comb samples. Each of these samples was incubated individually in a small, sterilised plastic box (4 × 3 × 5 cm; disinfected with 75% ethanol and exposed to UV light for 30 min prior to use), lined with a layer of moist sterile tissue paper to maintain appropriate humidity. After two weeks of incubation in the dark at room temperature, visible Xylaria strains were isolated by transferring both actively growing aerial mycelia and sclerotial tissues on to PDA plates (Fig. 1B). Xylaria strains from Odontotermes sp. were labelled “CX” and those from Macrotermes gilvus were labelled “CMU” (Suppl. material 1: table S7).

To assess whether incipient colonies established under laboratory conditions showed indications of weedy Xylaria invasion, alates were collected from Macrotermes gilvus colonies from Thailand in May 2025 and paired. The pairs were placed in plastic boxes containing sterile agar (wiped with 75% ethanol and exposed to UV for 30 min) and maintained for five months with sterile food and sterile water. The food was obtained from the same natural colonies and autoclaved three times at 121 °C for 20 min each. Afterwards, all kings, queens and newly-hatched offspring were removed from the boxes (five replicates). The termite-excluded boxes were then incubated at 27 °C for one month to observe whether visible Xylaria mycelia emerged.

Fungal mycelia harvested from two-week-old cultures were transferred into 1.5 ml microcentrifuge tubes for genomic DNA extraction. Fungal DNA was isolated using the Biospin Fungus Genomic DNA Extraction Kit (BioFlux®, Hangzhou, China) following the manufacturer’s instructions. Bacterial DNA was extracted from biomass collected from NA plates using the Biospin Bacteria Genomic DNA Extraction Kit (BioFlux®, Hangzhou, China) according to the manufacturer’s protocol. PCR amplification was performed in a 25 μl reaction mixture containing 12.5 μl of 2× Power Taq PCR MasterMix (Sangon, China), 8.5 μl of sterile double-distilled water, 2 μl of DNA template and 1 μl each of forward and reverse primers (10 pmol). The targeted loci included the 16S ribosomal RNA (16S rRNA) gene for bacterial isolates, the internal transcribed spacer (ITS) region for fungal isolates and the alpha-actin gene (ACT), beta-tubulin 2 (TUB2) and DNA-dependent RNA polymerase second largest subunit (RPB2) for isolates of Xylaria (see Table 1 for primer details). The PCR conditions consisted of an initial denaturation at 94 °C for 4 min, followed by 30 cycles at 94 °C for 30 s, annealing at gene-specific temperatures (see Table 1) for the indicated durations and 72 °C for 1 min, with a final extension at 72 °C for 10 min. PCR products were sequenced by Sangon Biotech Co., Ltd. (Shanghai, China) and the resulting sequences were deposited in GenBank (Suppl. material 1: tables S1–S7).

Raw sequences obtained from the sequencing company were assembled and edited using Geneious (Restricted) (https://www.geneious.com; accessed 27 April 2026). All ITS and 16S gene sequences obtained were subjected to BLASTn searches against the GenBank database (http://blast.ncbi.nlm.nih.gov/; accessed 27 April 2026) to determine the genus-level identification and record the most similar strains (Suppl. material 1: tables S1–S6). A four-locus phylogenetic analysis (ITS, ACT, RPB2 and TUB2) was conducted to determine: (1) whether multiple species of Xylaria can co-exist within a single colony and (2) whether clade differentiation is associated with the termite genus or geographic location (Suppl. material 1: table S7). For bacterial isolates, phylogenetic analyses were conducted using 16S rRNA sequences, including both our sequences and relevant outgroup reference sequences (Suppl. material 1: tables S1, S2). Multiple sequence alignments were generated using the web-based MAFFT server (https://mafft.cbrc.jp/alignment/server/; accessed 27 April 2026) (Katoh and Standley 2013) and manually checked in BioEdit 7.2.3 (Hall 1999). Ambiguous or uninformative regions were trimmed using trimAL v.1.2 (http://trimal.cgenomics.org; accessed 27 April 2026), followed by sequence assembly by BioEdit. The Alignment Transformation Environment (ALTER) digital platform was used to convert FASTA files to PHYLIP format (Glez-Peña et al. 2010). Maximum Likelihood (ML) phylogenetic analyses were conducted using RAxML-HPC2 on XSEDE (v.8.2.12) via the CIPRES Science Gateway v.3.3 (Miller et al. 2010), applying the GTRGAMMA substitution model with 1,000 bootstrap replicates (Stamatakis 2014). Phylogenetic trees were visualised in FigTree v.1.4.0 and further edited using iTOL v.7 (https://itol.embl.de/; accessed 27 April 2026) and node support values were shown based on ML bootstrap analyses.

The Metarhizium sp. Cmu-b60 isolate, obtained from a Macrotermes gilvus comb, produced a high yield of conidia on PDA and showed high sequence similarity to M. anisopliae strain 090904-2-3, based on ITS BLASTn analysis (Suppl. material 1: table S4). In contrast, Metarhizium sp. CX-CU12 isolated from Odontotermes sp. comb showed a high level of similarity to Metarhizium lepidiotae BUM66.1 (Suppl. material 1: table S3). Metarhizium anisopliae is the most commonly used and studied entomopathogenic species (Zimmermann 2007). Consequently, Metarhizium sp. Cmu-b60 was selected for the entomopathogenicity test. Conidial suspensions were prepared for bioassays against the host termite Macrotermes gilvus. Conidia were harvested from one-month-old, actively sporulating cultures, suspended in sterile distilled water containing 0.02% Tween 80 and then passed through a sterile filter to remove hyphal debris (Wang and Powell 2004; Singha et al. 2011). Conidial density was quantified with a haemocytometer under a compound microscope and the suspension was subsequently diluted with sterile 0.02% Tween 80 carrier solution to obtain a working concentration of 1 × 10⁸ conidia ml^–1. Conidial viability was verified on PDA before testing; only suspensions exhibiting a germination rate of > 95% were used for virulence assays. Control treatments consisted of a sterile carrier solution containing 0.02% Tween 80 without conidia.

To evaluate the virulence of Metarhizium sp. strain Cmu-b60 on fungus-growing termites, active individuals were collected from Macrotermes gilvus colony 2, the colony from which strain Cmu-b60 was originally isolated. Each experimental group consisted of 10 major workers, 10 minor workers, two major soldiers and two minor soldiers. Each group was placed on a piece of moistened tissue paper in 90-mm glass Petri plates (oven-sterilised at 180 °C for 3 h). Ten replicate groups were prepared, and each termite was gently guided using sterile wooden sticks and briefly dipped in a 1 × 10⁸ conidia ml^–1 conidial suspension for 5 s (Syazwan et al. 2021). Control groups were handled the same way and dipped in sterile distilled water containing 0.02% Tween 80 (Singha et al. 2011). The plates were partially sealed with parafilm and maintained in the dark at 27 °C. The plates were monitored at 24-h intervals. Individuals that failed to respond to gentle stimulation, such as prodding the antennae and palps with a fine pin, were recorded as dead, removed from the plates and transferred to fresh Petri dishes containing moistened tissue to observe subsequent fungal colonisation and mycosis development. Mortality was quantified for each 24-h interval and survival rates were calculated using the formula:

Survival rate = N alive N total × 100 %

where N_alive represents the number of surviving individuals at each observation and N_total is the initial group size.

To assess the differences in termite burying responses to the entomopathogenic fungus Metarhizium sp. and the weedy fungus Xylaria sp. in the presence or absence of the termite-fungus garden, a behavioural experiment was conducted in sterilised 90 mm glass Petri dishes (oven-sterilised at 180 °C for 3 h before use). Termites were simultaneously exposed to a PDA control plug, a Metarhizium sp. plug and a Xylaria sp. plug (Fig. 1C). Soil used in the assays was collected from mounds of Macrotermes gilvus colonies and adjacent to nests of Odontotermes sp. The soil was air-dried at room temperature, autoclaved five times at 121 °C for 20 min and subsequently oven-dried at 60 °C prior to use. A piece of sterile, moistened tissue paper was placed in each of the 90-mm glass Petri dishes and 10 g of treated soil was evenly spread over the surface. Metarhizium sp. Cmu-b60 was selected because it shared high ITS BLASTn similarity to M. anisopliae and exhibited strong virulence against termites; Xylaria sp. clades 11 and 20 were selected because they were frequently induced from comb-incubation assays of their respective termite hosts (see Results). Therefore, on one side of the Petri plate (0.5 cm from the edge), either a 0.2 g fresh comb (collected within 30 min) with four attached Termitomyces nodules (treatment) or no comb (control) was placed on a sterile circular filter paper (1.5 cm in diameter) (Fig. 1C). On the opposite side (5.5 cm from the centre), three 0.5-cm-diameter, one-week-old agar plugs of Metarhizium sp. (Cmu-b60), Xylaria spp. (clade 11 or clade 20) or PDA were placed on sterile circular filter paper (1.5 cm in diameter), as shown in Fig. 1C-a (Yang et al. 2025). Prior to transferring termites to the comb-soil plates, termites were acclimatised for 15 min in Petri dishes at 27 °C in a dark, quiet environment to allow allogrooming. Each Petri plate then received one minor soldier, five major workers and five minor workers. The plates were partially sealed with parafilm to allow ventilation and incubated at 27 °C in a dark, quiet environment for 6 hours. Nine replicates were prepared for each termite genus. To quantify the extent to which termites exhibited hygienic behaviour, the weight of fresh soil boluses on and within 1.5-cm-diameter filter paper was measured in each treatment. To ensure analytical precision, the weight of the soil boluses was determined by weighing by difference using the equation:

W soil boluses = W PDA + soil boluses − W PDA

(Katariya et al. 2017; Yang et al. 2025).

To evaluate whether burying responses affect fungal growth, four 0.5-cm-diameter, one-week-old fungal plugs grown on PDA were placed on water agar (WA) plates as controls. In contrast, Metarhizium sp. and Xylaria spp. plugs collected 24 hours after the burying assay, along with the attached soil boluses, were transferred on to separate WA plates. All inoculation points were positioned at the four corners, each 2 cm from the edge of the Petri dish. After one week, Growth Inhibition (GI) was calculated using the formula:

GI ( % ) = R c − R t R c × 100 %

where R_C is the maximum radial growth diameter of Metarhizium or Xylaria on WA control plates and R_t is the maximum radial growth diameter of fungi with encased soil boluses on the test WA plates, including submerged mycelia.

Moreover, to assess whether termites exhibit avoidance behaviour towards Metarhizium-contaminated food substrates, six laboratory-established colonies, each derived from paired alates and containing 20 new individuals (3 months old), were used. Termite food material was collected from natural Macrotermes gilvus colonies in September 2025 and sterilised by five rounds of autoclaving at 121 °C for 20 min each. The food material was then oven-dried prior to use. For the treatment, 1 g of food substrate was mixed with 1 ml of a Metarhizium sp. Cmu-b60 conidial suspension prepared as described above (1 × 10⁸ conidia ml^–1) and placed as a triangular patch in one corner of a 90-mm glass Petri dish lined with moistened sterile filter paper. For the control, 1 ml of sterile distilled water containing 0.02% Tween 80 was used instead of the conidial suspension. Termites from each of the six incipient colonies were separately transferred to glass Petri dishes, lined with moistened sterile filter paper and provisioned with a cotton ball. Termites from each of the three incipient colonies were separately transferred to glass Petri dishes for the treatment group, and termites from another three colonies were used for the control group. The dishes were partially sealed with parafilm to maintain humidity while allowing ventilation. Mortality and the extent of food substrate displacement were recorded after 120 h (5 days).

This study also aimed to screen bacteria associated with fungus-growing termites for inhibitory activity against the entomopathogen Metarhizium sp. Cmu-b60 and the weedy fungus Xylaria sp. clade 20 (a fast-growing Xylaria clade isolated from incubated combs of both Macrotermes and Odontotermes). To determine whether bacterial strains inhibit the growth of Metarhizium sp. and Xylaria sp. in vitro, bacterial genera shared by Macrotermes (Thailand) and Odontotermes (China) were selected. For each genus, two bacterial strains were selected, representing different hosts and distinct clades in the 16S rRNA Maximum Likelihood (ML) phylogenetic tree (Bacillus sp. CX-gut6 and CMU-g10; Brevibacillus sp. CX-gut6-1 and CMU-C11; Burkholderia sp. CX-Comb6 and CMU-C7, Enterobacter sp. CX-gut10 and CMU-g3; Pseudomonas sp. CX-gut7 and CMU-g13; Serratia sp. CX-Comb4 and CMU-g5) (see Results; Suppl. material 1: tables S1, S2, Suppl. material 2: fig. S1; ). Five-millimetre-diameter mycelial plugs from one-month-old Metarhizium sp. Cmu-b60 and one-week-old Xylaria sp. (Clade 20) cultures were placed at the edge of 60- or 90-mm NA Petri dishes. A bacterial colony (0.5 cm wide and 2 cm long) was inoculated at the opposite side of the plate (Fig. 1C). Control plates were inoculated with the fungal isolates only and all the assays were performed in triplicate (Fig. 1C) (Mis et al. 2024). The plates were incubated at 27 °C for 5 d for the bacterium–Xylaria assay and for 15 d for the bacterium–Metarhizium assay. The inhibitory effects of the bacteria against the two fungi were expressed as percentage inhibition (PI) in radial growth, calculated using the formula:

PI ( % ) = R 1 − R 2 R 1 × 100 %

where R₁ is the radial growth distance of the fungus from the inoculation point to the opposite edge in the control plates and R₂ is the minimum growth in the direction facing the bacterium in treatment plates (Mis et al. 2024).

Analyses of fungal and bacterial communities were performed using the R (v.4.5.1) (R Core Team 2025). Fungal isolates were identified by BLAST analysis of the ITS sequence against the NCBI database. Relative abundance of each genus was calculated per nest as the percentage of isolates assigned to that genus amongst all isolates obtained from that nest. A bar chart of mean relative abundances and a heatmap of the top 50 genera were generated using the R packages “microeco” and “pheatmap”, respectively (Liu et al. 2021). For the heatmap, abundance data were log₁₀(x+0.001)-transformed and row-wise Z-score scaled, without column clustering. Alpha diversity indices (Shannon and Chao1) were compared between nest and gut samples within each termite species using Wilcoxon rank-sum tests. Beta diversity was assessed using Bray-Curtis distance matrices and visualised via Principal Coordinate Analysis (PCoA). Permutational multivariate analysis of variance (PERMANOVA) was performed to assess community dissimilarity, using 9999 permutations and the Bray–Curtis distance matrix as input (Bray and Curtis 1957). To identify genera with high discriminatory potential between nest and gut, a random forest (RF) classification model (ntree = 1000; the model was run 25 times) was performed using the randomForest package in R and the top 10 genera ranked by Gini importance were selected as discriminative features (Liaw and Wiener 2002). Importantly, to avoid confounding between termite species and geographic origin, all statistical comparisons (alpha diversity, beta diversity and differential abundance analyses) were performed separately for the two termite species.

In tests of termite burying responses to weedy Xylaria and entomopathogen in the presence or absence of termite comb, the normality assumption was met for M. gilvus and violated for Odontotermes sp. (Shapiro-Wilk: P = 0.258, P < 0.001) and the assumption of homogeneity of variance was violated for both species (Levene’s Test: P = 0.022, P = 0.004). Therefore, a Generalised Linear Model (GZLM) was fitted using a gamma distribution with a log link function and robust standard errors in IBM SPSS Statistics for Windows v. 29.0 (IBM Corp., Chicago, IL, USA). All main effects and interactions were evaluated using Type III Wald χ² statistics within the GZLM framework. When pairwise comparisons were required, simple-effects analyses were performed using estimated marginal means (EMMs) with a Bonferroni correction to control the family-wise Type I error rate.

For the analysis of whether deposited soil boluses inhibited the growth of Xylaria sp. and Metarhizium sp., the generated data met assumptions of normality and homogeneity of variance in SPSS (Shapiro-Wilk: P > 0.05, Levene’s Test: P > 0.05). Therefore, one-way ANOVA was applied; when the ANOVA indicated a statistically significant difference, post hoc pairwise comparisons were conducted using Tukey’s HSD test.

Likewise, for analysis of whether the termite entomopathogen (Metarhizium sp. Cmu-b60) and Xylaria sp. clade 20 were inhibited by 12 selected bacterial strains in SPSS, the generated data met the assumptions of normality and homogeneity of variance. Therefore, a one-way ANOVA was applied; if the ANOVA revealed statistically significant differences, post hoc pairwise comparisons were performed using Tukey’s HSD test.

Kaplan-Meier curves and survival plots were generated using Prism (GraphPad Prism v. 9.0; GraphPad Holdings, San Diego, CA, USA) (Motulsky 2003).

In this study, 59 bacterial strains were isolated from six colonies of two species of termites. From Odontotermes sp. (China), 28 strains were isolated; 12 strains from inner combs and 16 strains from worker guts. From Macrotermes gilvus (Thailand), 31 bacterial strains were isolated; 17 strains from the inner combs and 14 strains from worker guts (Suppl. material 1: tables S1, S2; Suppl. material 2: fig. S1). The ML phylogenetic tree, based on 16S rRNA, revealed that the 59 strains belong to 14 distinct genera (Fig. 2A). Six genera — Bacillus (15 isolates: 25.4%), Brevibacillus (3.4%), Burkholderia (5.1%), Enterobacter (7 isolates: 11.9%), Pseudomonas (14 isolates: 23.7%) and Serratia (5 isolates: 8.5%) were common to both Odontotermes sp. and Macrotermes gilvus. In contrast, four genera: Pandoraea (1.7%), Paraburkholderia (3.4%), Priestia (6.8%) and Rhodococcus (1.7%), were exclusively isolated from Odontotermes sp., while Citrobacter (1.7%), Pantoea (1.7%), Providencia (1.7%) and Kosakonia (3.4%) were unique to M. gilvus. Furthermore, substantial overlap was observed between comb- and gut-associated isolates across the two termite species (Fig. 2).

At the family level, the 59 bacteria were distributed amongst nine distinct families. Six families (Bacillaceae, Burkholderiaceae, Enterobacteriaceae, Paenibacillaceae, Pseudomonadaceae and Yersiniaceae) were shared between Odontotermes sp. and Macrotermes gilvus. Nocardiaceae was exclusively associated with Odontotermes sp., whereas Erwiniaceae and Morganellaceae were unique to M. gilvus. Moreover, considerable overlap in family-level composition was observed between comb- and gut-associated bacterial isolates from the two termite species (Fig. 2B).

At the higher taxonomic level, the 59 bacterial isolates were classified into five distinct orders, four classes and three phyla. Four orders (Bacillales, Burkholderiales, Enterobacterales and Pseudomonadales) were shared between Odontotermes sp. and Macrotermes gilvus, but members of the Mycobacteriales were exclusively isolated from Odontotermes sp. (Fig. 2).

A total of 600 culturable fungal strains (excluding comb-specialised Xylaria isolates) were isolated from six colonies of two termite species. Of these, 300 fungal strains were from Macrotermes gilvus, with 200 strains associated with combs and 100 isolated from worker guts. Similarly, 300 strains were isolated from Odontotermes sp., 200 from the combs and 100 from worker guts (Suppl. material 2: figs S2–S5). Based on ITS sequences and BLASTn searches, the isolated fungal strains belong to three phyla — Ascomycota (413 isolates, 68.8%), Mucoromycota (171 isolates, 28.5%) and Basidiomycota (16 isolates, 2.7%). Ascomycota isolates were far more abundant in fungus combs than in guts (Macrotermes, 170 vs. 61; Odontotermes, 148 vs. 34). In contrast, Mucoromycota isolates were more frequent in guts than in combs (Macrotermes, 49 vs. 66; Odontotermes, 22 vs. 34). Furthermore, the isolated fungal strains were distributed into 11 classes, 26 orders, 52 families and 79 genera (Suppl. material 1: tables S3–S6). At the genus level, Trichoderma was the most dominant genus, accounting for 164 isolates (27.3% of all fungal isolates). This was followed by Mucor (9.8%), Aspergillus (5.5%), Fusarium (5.1%), Linnemannia (4.7%), Curvularia (3.5%), Cunninghamella (3.3%), Penicillium (3.2%), Neopestalotiopsis (3%) and Umbelopsis (2.8%) (Fig. 3; Suppl. material 1: tables S3–S6).

The relative abundance (%) of genera and phyla associated with four isolate sources (combs and worker guts of Macrotermes gilvus; combs and worker guts of Odontotermes sp.) is shown in Suppl. material 2: fig. S6. The top 50 dominant fungal genera are shown in Fig. 3. Alpha diversity indices (Shannon and Chao1) did not differ significantly between nest and gut samples within either termite species (Wilcoxon rank-sum test, P > 0.05; Suppl. material 2: fig. S7). Subsequent PERMANOVA analysis showed that the effect of sample part (comb vs. gut) on the fungal community structure showed a strong, but statistically non-significant trend in both termite species. For Macrotermes gilvus, it explained 30.2% of the variance (PERMANOVA, F = 1.736, P = 0.1, R² = 0.302) (Fig. 4A). For Odontotermes sp., the explanatory power was even higher, accounting for 50.4% of the variance (F = 4.066, P = 0.1, R² = 0.504) (Fig. 4B). A random forest (RF) analysis identified the key fungal genera that differed between termite guts and combs within each termite genus. In Macrotermes termites, Bartalinia, Geotrichum and Phomopsis were isolated exclusively from the combs, whereas Irpex was isolated only from worker guts (Fig. 4C). Amongst the genera shared between combs and guts, most exhibited higher abundance in the combs, with the exception of Cunninghamella. In Odontotermes termites, Chaetomium, Cladosporium, Penicillium, Phialomyces and Scleroconidioma were isolated only from the combs and, amongst the shared genera, most were more abundant in combs, except for Linnemannia (Fig. 4D). Overall, the cultivable fungal community composition did not differ significantly between gut and comb isolates from the same genus or location (Fig. 4).

Using the dilution plating technique, no Termitomyces strains were isolated from either worker guts or comb samples. One Xylaria strain (CX-CB35/GMBCC1340) was isolated and showed 100% ITS sequence similarity to the Xylaria grammica isolate Otu0315, as determined by BLASTn analysis. However, this isolate did not belong to the termite comb-specialised Xylaria (TE) clade. One strain (CX-CU12/ GMBCC1448) isolated from combs of Odontotermes sp. (China), showed 100% ITS sequence similarity to Metarhizium lepidiotae isolate BUM66.1, based on BLASTn analysis, while another strain Cmu-b60 isolated from combs of Macrotermes gilvus (Thailand), showed 99.6% ITS sequence similarity to M. anisopliae 090904-2-3, based on BLASTn analysis.

A total of 112 strains of Xylaria were isolated from incubated combs of two fungus-growing termite genera; these were 50 strains from two colonies of Odontotermes sp. in China and 62 strains from three colonies of Macrotermes gilvus in Thailand (Suppl. material 1: table S7; Suppl. material 2: figs S8, S9). Multi-locus phylogenetic analyses were performed using sequences generated from four loci (ITS, ACT, RPB2 and TUB2), together with sequences of 76 comb-specialised Xylaria taxa and outgroup species retrieved from GenBank (Li et al. 2025). Thus, a total of 188 termite-associated Xylaria strains were included in the phylogenetic reconstruction (Suppl. material 1: table S7). The dataset comprised ITS sequences for all strains, while ACT, TUB2 and RPB2 sequences were available for 175, 138 and 145 specimens, respectively. The aligned lengths of ITS, ACT, RPB2 and TUB2 were 670, 297, 1,183 and 1,784 characters, respectively, yielding 2,441 distinct alignment patterns. The proportion of gaps and completely undetermined characters accounted for 33.81% of the total alignment.

Phylogenetic analysis, based on a concatenated four-locus dataset, resolved the 112 strains into 20 distinct operational taxonomic units (OTUs, hereafter referred to as clades). Ten clades corresponded to previously described species — clade 3 (Xylaria acuminatilongissima), clade 4 (X. neonigripes), clade 5 (X. rogersionigripes), clade 6 (X. escharoidea), clade 7 (X. cf. escharoidea), clade 9 (X. iriomotensis), clade 12 (X. minima), clade 16 (X. reinkingii var. microspora), clade 18 (X. siamensis) and clade 20 (X. ochraceostroma). The remaining ten clades (1, 2, 8, 10, 11, 13, 14, 15, 17 and 19) likely represent novel Xylaria taxa. Only two clades — clade 16 (X. reinkingii var. microspora) and clade 20 (X. ochraceostroma) — were isolated from both termite genera, whereas the remaining 18 clades were each restricted to either the Chinese (Odontotermes sp.) or Thai (Macrotermes gilvus) hosts (Fig. 5). In Odontotermes sp. combs, the four most frequently isolated were clade 11 (21 isolates), clade 9 (10 isolates), clade 16 (six isolates) and clade 7 (five isolates). In Macrotermes gilvus combs, the four most abundant clades were clade 20 (16 isolates), clade 8 (eight isolates), clades 16 and 19 (six isolates in each) and clade 12 (five isolates) (Fig. 5).

Notably, only clade 12 (X. minima) was observed to form miniature sclerotia (3–6 mm diam.; n = 5) on two-week incubated combs of Macrotermes gilvus (Suppl. material 2: fig. S10). Such structures were, however, not observed on Odontotermes combs. The sclerotia were globose, initially white and gradually darkened to black upon maturation (Suppl. material 2: fig. S10). Vertical sections revealed white internal tissue consisting of densely entangled mycelia, with no spores or conidia observed (Suppl. material 2: fig. S10). The sclerotia were directly and abruptly attached to the fungal comb, with few or no visible hyphal development in the early stage. At later stages, hyphae emerged from the sclerotial crust, growing upwards with minimal branching until the sclerotia became dry and hollow. X. minima strains grew readily on PDA and WA plates (Suppl. material 2: fig. S10). As in a recent study, this sclerotium-forming Xylaria minima was also observed in Macrotermes annandalei and M. gilvus (Yang et al. 2026).

Meanwhile, nine paired-alates rearing boxes successfully produced eggs which hatched into the first generation of termites under laboratory conditions. With continuous provision of sterilised food, one colony developed initial comb-like structures (Suppl. material 2: fig. S11). However, after the termites were removed and the comb-like structures were incubated in the laboratory for one week, no Xylaria mycelial growth was observed (Suppl. material 2: fig. S11).

When grown on PDA, the Cmu-b60 isolate formed yellow to dark green conidial bundles upon maturation. The conidia were unicellular and clavate, measuring 6–8 × 2–3 μm (mean = 7 × 2.5 μm, n = 20), with blunt apices (Fig. 6A).

To evaluate the virulence of the Cmu-b60 isolate, a bioassay was conducted against its original host, Macrotermes gilvus, using a conidial suspension of 1 × 10⁸ conidia ml^-1. The survival rate of the infected group (n = 10 replicates) was compared with that of the negative control group (n = 10 replicates) over a 96-hour period. The control group maintained a consistently high survival (> 96.67%) throughout the experiment. In contrast, the infected group exhibited a rapid decline in survival, beginning at 95% survival 24 hours post-inoculation and reaching 100% mortality by 96 hours. Fungal subculture, DNA extraction and ITS sequencing from corpses confirmed that the strain was identical to Cmu-b60. Statistical analysis confirmed the highly significant difference in mortality. Survival analysis using a log-rank test confirmed that the difference between the treatment and control groups was statistically significant (χ² = 475.6, df = 1, P < 0.0001). The Hazard Ratio (HR) test indicated an approximately 49-fold increase in mortality in the infected (T) group compared with the control (C) group (HR = 48.88; 95% CI, 37.82–63.16) (Fig. 6; Suppl. material 1: tables S8, S9).

When fungus-growing termites were simultaneously exposed to the two potential threats — the comb weed Xylaria clades 11 and 20 and the entomopathogenic fungus Metarhizium sp. Cmu-b60 — along with a PDA control plug, their burying responses were markedly different (Fig. 7; Suppl. material 1: tables S10, S11). The presence or absence of the fungal comb had no significant effect on the weight of soil boluses deposited (GZLM: Macrotermes gilvus: Wald χ² = 1.280, df = 1, P = 0.258; Odontotermes sp.: Wald χ² = 1.185, df = 1, P = 0.276). In contrast, fungus identity exerted a strong effect on burying behaviour (M. gilvus: Wald χ² = 97.989, df = 2, P < 0.001; Odontotermes sp.: Wald χ² = 169.779, df = 2, P < 0.001), whereas the interaction between comb presence and fungus identity was not significant (Macrotermes gilvus: Wald χ² = 1.292, df = 2, P = 0.524; Odontotermes sp.: Wald χ² = 1.291, df = 2, P = 0.525). Regardless of the presence or absence of termite comb, exposure to Metarhizium elicited significantly more bolus deposition than either PDA or Xylaria in both Macrotermes gilvus and Odontotermes sp. (Bonferroni-corrected GLM, all P ≤ 0.001) (Suppl. material 1: table S12). In contrast, no significant difference was detected between Xylaria and PDA in either species (all P ≥ 0.215) (Suppl. material 1: table S12).

In descriptive comparisons, based on mean ± SD, Macrotermes gilvus deposited more soil boluses than Odontotermes sp. across all three treatments. The largest difference was observed under Metarhizium exposure (168.17 ± 65.57 vs. 130.28 ± 44.81), followed by PDA (66.24 ± 12.47 vs. 35.57 ± 10.51), while the difference was less pronounced under Xylaria (73.78 ± 24.29 vs. 54.24 ± 21.81) (Suppl. material 1: tables S10, S11).

The results obtained in this study showed that soil-bolus deposition effectively inhibited the growth of both Xylaria and Metarhizium in vitro. On WA control plates after one week, Xylaria sp. clade 20 produced upright, long, aerial mycelia or stromata, as well as immersed mycelia within the agar, whereas Xylaria sp. clade 11 and Metarhizium sp. Cmu-b60 produced only immersed mycelia (Suppl. material 2: fig. S12). In contrast, in WA plates treated with soil boluses, fungal growth was strongly suppressed, with only limited mycelial or stromatal growth emerging from the plugs of Xylaria sp. clade 11 (GI ± SD (%): 66.21% ± 9.32) and clade 20 (60.7% ± 7.9). However, although some plugs still produced conidia, the immersed mycelia were rarely observed. For Metarhizium plugs, growth was strongly inhibited regardless of whether the soil boluses were deposited by Macrotermes gilvus or Odontotermes sp. (GI ± SD: 68.32% ± 5.68 and 66.75% ± 6.34, respectively) and unidentified bacterial growth was occasionally observed around the plugs (Suppl. material 2: fig. S12). Despite these differences in fungal identity, the degree of radial growth inhibition by deposited soil boluses did not differ significantly between Xylaria sp. clade 11 and clade 20 and Metarhizium sp. Cmu-b60 (one-way ANOVA: F₃,₁₂ = 0.796, P = 0.519).

The plates were examined after 120 hours (5 days). In the control groups, the termites remained highly active throughout the experimental period, with no mortality or signs of fungal infection and newly-hatched termites were observed. The continuous movement and feeding activity of the termites led to extensive dispersal of the provided food material, creating a visibly disordered plate environment (Suppl. material 2: fig. S13A). In contrast, termites exposed to Metarhizium-contaminated food exhibited 100% mortality and dead individuals were covered by dense white mycelial growth indicative of successful fungal infection; subculture, DNA extraction and ITS sequencing confirmed that it was the same strain as Cmu-b60. Termites showed no intensive movement or feeding activity, as evidenced by the undisturbed, neatly accumulated food at the corners of the plates. These results indicate that first-generation termites from the lab innately avoid Metarhizium conidia-contaminated food, and that this avoidance is accompanied by a rapid, lethal pathogenic effect of Metarhizium under laboratory conditions, occurring even in the absence of direct physical contact between newly-introduced termites and the fungal inoculum (Suppl. material 2: fig. S13B).

All 12 tested bacterial isolates (Suppl. material 1: tables S1, S2) exhibited varying degrees of growth inhibition against the termite entomopathogen Metarhizium sp. Cmu-b60. Four strains of the genera Bacillus, Brevibacillus and Burkholderia suppressed fungal growth by more than 50%. Interactions with Bacillus were characterised by pronounced inhibition zones, indicating strong antagonistic activity (Fig. 8; Suppl. material 1: table S13). In contrast, the same bacterial isolates displayed either weak or no inhibitory effects against the comb weed Xylaria sp. clade 20. Only six isolates from the genera Brevibacillus, Burkholderia, Pseudomonas and Serratia showed growth inhibition exceeding 50%. Xylaria rapidly produced abundant aerial mycelia, sometimes growing above colonies of Bacillus and Enterobacter isolates. In contrast, interactions with Brevibacillus, Burkholderia, Pseudomonas and Serratia resulted in a clear concave mycelial retreat, consistent with active suppression of fungal growth (Fig. 8; Suppl. material 1: table S13).