Introduction

Like most animals, invertebrates form intricate partnerships with diverse microbial communities1, contributing significantly to their evolutionary and ecological success2. This close interconnectedness has led to the concept of animals as “holobionts,” where the host and its microbiota are viewed as a single ecological entity3,4. Recent studies on microbiomes provide further evidence of the widespread prevalence of microbial partnerships across the animal kingdom5,6.

While most invertebrates have microbial associations, their reliance on them varies widely. Termites, for instance, depend entirely on their gut microbiota for nutrition7. Conversely, many other arthropods, such as caterpillars, may lack a resident gut microbiota and develop fully even germ-free8. Most arthropods generally fall between these extremes, relying on their microbiota for some form of support (e.g. cockroaches9,10 or isopods11,12). Detritivorous and xylophagous animals typically rely on gut microorganisms for cellulose digestion. Although animal cellulases are found in some gut systems13, (ligno)cellulolytic bacteria, fungi and protists are generally deemed necessary for hydrolysis and fermentation, releasing short-chain fatty acids, which get absorbed by the host14.

Millipedes (Diplopoda) are crucial detritivores widely distributed and abundant in many temperate and tropical ecosystems15. Despite their status as keystone species in tropical and temperate forests16, millipedes have been understudied compared to other detritivores, particularly concerning their microbiome. Due to the nutrient-poor nature of plant litter, millipedes compensate for low assimilation efficiencies through high ingestion rates17. Similar to other arthropods, millipedes host diverse gut microorganisms18. Notably, the central hindgut was shown to host the highest microorganism density, attaching to its cuticle, while the foregut and midgut contain mostly transient inhabitants19. Various studies suggest that certain millipede gut bacteria possess enzymes for breaking down plant polysaccharides20,21,22,23,24. If millipedes rely on cellulose for their nutrition, extensive fermentation followed by methanogenesis, similar to ruminants or wood-feeding termites, should occur in their guts7,25. However, methanogenesis has only been observed in some millipede species, but not others, with its occurrence correlated to the millipede size26. Despite these findings, direct proof of gut microorganisms supporting the millipede’s nutritional needs has not been experimentally demonstrated. An alternative hypothesis suggests millipedes foster microbial growth in litter, potentially digesting the resulting fungal and bacterial biomass27.

To investigate the role of the millipede gut microbiota, we conducted experiments using two model species: the CH4-emitting Epibolus pulchripes (Spirobolida) and Glomeris connexa (Glomerida), which do not emit CH4. E. pulchripes is a large millipede (130–160 mm) common along the East African coast28, while G. connexa is smaller (10–17 mm) and native to Central Europe29. We assessed the impact of inhibitors on body weight, survival, faecal bacterial load, gut bacterial composition, and CH4 production. Additionally, we identified metabolically active hindgut prokaryotes using 13C-RNA-SIP.

Results

Effect of antibiotic treatment

Feeding millipedes with either sterile or antibiotics-treated feed led to only negligible and insignificant weight change in both species (Fig. 1a, b and Supplementary Data 1). In E. pulchripes, some specific days did show a significant decline in the treated groups compared to the control but the overall change was small (<5%). The treatment also did not significantly impact the millipedes’ survival based on Kaplan-Meier estimates (Supplementary Fig. 1). Despite maintaining a stable weight, faecal production decreased over time in response to antibiotics or sterile feed (P < 2.2e-16 for both species; Fig. 1c, d and Supplementary Data 2). No significant difference was found between the treated groups in E. pulchripes, but in G. connexa the sterile-litter group was different from the antibiotic-treated groups (P < 0.0001). Total faecal colony counts in both millipede species were also consistently higher in the control group compared to the antibiotic-treated or sterile feeding groups at all time points (P < 0.0001; Fig. 1e, f and Supplementary Data 3). The number of CFUs dropped by >90% already at day 9 (when all individuals in both species were still producing faecal pellets) and a much greater percent afterwards (though this was already affected by the decline in the faecal counts). After 35 days for E. pulchripes and 16 days for G. connexa, most animals in the treatment groups ceased faecal production, leading to the cessation of plate counts. Once again, only the sterile-litter group in G. connexa differed from the other treatment groups. Total faecal 16S rRNA gene copies in E. pulchripes were reduced by 61–77% in the treated groups compared to the control group (P = 0.01), while In G. connexa, 34–74% reductions were observed in the treated groups (P < 0.001; Fig. 1g and Supplementary Data 4). In both species, no difference between the treated groups was observed. After noting a substantial decrease in bacterial load, we measured CH4 emission on day 35 (Fig. 1h and Supplementary Data 5). As anticipated, CH4 was measured in E. pulchripes but absent in G. connexa. The control groups displayed a significantly higher CH4 production rate (284.1 ± 58 nmol mg−1 d−1) than the other treatments (P = 0.0008). However, the treated groups saw a 57‒74% reduction in CH4 production without significant differences between them.

Fig. 1: Effect of antibiotic treatment on E. pulchripes and G. connexa.
figure 1

Time series of mean weight loss (mean ± SE ribbon) in (a) E. pulchripes and (b) G. connexa; faecal counts in (c) E. pulchripes and (d) G. connexa; total colony forming units in (e) E. pulchripes and (f) G. connexa; (g) 16S rRNA gene copy numbers in the faeces; and (h) CH4 production rate after 35 days of antibiotics treatment in E. pulchripes. ‘High Conc.’ and ‘Low Conc.’ refer to the concentration of applied antibiotics (see “Materials and methods” section for more details). Different lower case letters in panels g and h denote statistical significance. See “Results” section for a detailed description of the statistical tests performed on the time series (panels a–f).

Full size image

Prokaryotic community compositions after treatment

We sequenced 48 samples of E. pulchripes and G. connexa, consisting of 12 hindguts and 12 faecal samples for each species. The average sequencing depth stood at ca. 40 K reads per sample, post-processing of reads and decontamination (Supplementary Data 6 and 7). The two millipede species differed remarkably in their microbial composition, with the phylum Bacteroidota dominating the hindgut of E. pulchripes and Pseudomonadota that of G. connexa. In each case, these phyla comprised over 50% of the abundance regardless of treatment (Fig. 2a, b and Supplementary Data 8). Pseudomonadota, Bacteroidota and Actinobacteriota dominated both species’ faecal pellets (Fig. 2c, d). On the genus level, E. pulchripes’ hindgut and faecal samples were primarily dominated by Citrobacter, Bacteroides, and Pseudomonas (Fig. 2e–h and Supplementary Data 8). In contrast, G. connexa showed differences between hindgut and faecal sample compositions, with faecal samples appearing more diverse (Fig. 2h).

Fig. 2: Effect of antibiotic treatment on the taxonomic composition of prokaryotes in E. Pulchripes (left) and G. Connexa (right) following treatment.
figure 2

Phylum distribution in the hindguts (a, b) and the faeces (c, d). Distribution at genus level in the hindguts (e, f) and faeces (g, h). ‘High Conc.’ and ‘Low Conc.’ refer to the conc. of antibiotics applied (see “Materials and methods” section for more details).

Full size image

Impact of treatment on prokaryotic community structures

Overall, no significant differences were found in alpha diversity within or between treatment groups in the hindguts (Fig. 3a, b; Supplementary Fig. 2a; and Supplementary Data 9) or faeces (Fig. 3c, d; Supplementary Fig. 2a and Supplementary Data 9) of E. pulchripes and G. connexa. E. pulchripes’ hindgut groups displayed greater bacterial diversity and richness than G. connexa. Constrained analysis of principal coordinates (CAP) revealed significant differences in microbial community composition among sterile feeding or antibiotics treatments in both hindguts and faeces of both species (Fig. 3e–h; Supplementary Fig. 2b and Supplementary Data 9). ANCOM-BC2 analysis identified only a handful of microbial genera with differential abundance between treatments (Supplementary Data 10 and Supplementary Fig. 3), indicating that the antibiotic treatment was relatively non-selective. The few taxa with an increase in the mean absolute abundance (e.g. members of Tannerellaceae, Brucella and Pseudomonas) are known to often possess antibiotic resistance genes30,31,32.

Fig. 3: Effect of antibiotic treatment on the alpha and beta diversity indices of the microbial communities in the hindgut and faeces in E. pulchripes (left) and G. connexa (right).
figure 3

Alpha diversity values for each species, stratified by treatment groups for hindgut (a, b) and faeces samples (c, d) from E. pulchripes and G. connexa. The statistical test was based on Kruskal–Wallis (identical letters denote p > 0.05). Dissimilarity between hindgut (e, f) and faeces (g, h) microbial communities in the different treatments using constrained principal coordinates analysis (PcoA) with the model Dist.Mat ~ Treatment for each species and sample type separately.

Full size image

Influence of BES inhibition on methanogenesis in E. pulchripes

Na-BES-treated litter was provided to investigate the importance of methanogenesis in the CH4-emitting E. pulchripes. Methane emissions showed no significant differences on days 0 (P = 0.19) and 7 (P = 0.08; Fig. 4a and Supplementary Data 11). However, by day 14, CH4 production was nearly fully inhibited (P = 2.7 × 10−4) and remained so for an additional 21 days. Upon switching to untreated litter on day 35, CH4 emissions began recovering on day 49 and resumed pre-treatment values by day 63. Despite some average weight increase in the treated groups, no significant difference was detected at any time (Fig. 4b).

Fig. 4: Effect of BES treatment on CH4 emissions from E. pulchripes, animal weight, ciliates and ciliate-associated methanogens.
figure 4

a Emission rates of CH4 over time followed by recommence of methane production after the switch to untreated litters (indicated by the arrow). b Change in the weight of E. pulchripes over time. c Enumeration of symbiotic ciliates found in the faeces following BES treatment. d mcrA gene copy numbers in the faecal samples following BES treatment. e Fluorescence microscopy images of ciliates and the two most-abundant endosymbiotic methanogens in faecal samples of E. pulchripes using DAPI and CARD-FISH probes. ARC915: general archaea, RC281r_mod: Methanomassciillicoccales, and MB311: Methanobacteriales in the 10mM-Na-BES-treated group. f Enumeration of the methanogens associated with ciliates using FISH signals. The scale bar represents 20 µm.

Full size image

After inhibiting methane production for 21 days, a suspension made from fresh faeces was examined under a bright-field microscope, revealing various protists, nematodes, and rotifers ranging from 12 to 100 μm in size (Supplementary Fig. 4). The ciliate abundance averaged 3 × 105 ml−1, regardless of treatment (Fig. 4c; Supplementary Data 12). Quantification of the mcrA gene, pivotal in methanogenesis33, showed a significant reduction in the two Na-BES-treated groups compared to the control (P = 0.02; Fig. 4d; Supplementary Data 12). CARD-FISH was used to detect the presence of free-living (Supplementary Fig. 5) and symbiotic archaea (Supplementary Fig. 6), primarily methanogens, in protists from faecal samples. The amplicon sequencing data indicated that members of the Methanomassciillicoccales and Methanobacteriales were the dominant methanogens in E. pulchripes, and these orders were accordingly targeted. Although mcrA copy numbers declined, positive hybridisation signals for these methanogen orders were observed in both Na-BES treatments. Methanogens were detected on the 0.2 µm filter (Supplementary Fig. 5) and associated with protists as endosymbionts (Fig. 4e; Supplementary Fig. 6), with no significant changes in its count per ciliate (Fig. 4f).

Detection of active microbiota with 13C-RNA-SIP

RNA-SIP was used to identify the active microorganisms in the millipedes’ gut on a temporal scale (Supplementary Data 13). The shift in peak of 16S rRNA towards the denser gradient fractions, indicating label incorporation, was evident by day 3 and more prominently by day 7 for E. pulchripes and day 14 for G. connexa (Fig. 5). Nevertheless, despite feeding on fully-labelled litter for 21 days, a significant portion of RNA remained unlabelled. Surprisingly, the labelling of the fungal biomass, represented by the 18S rRNA peak, shifted earlier towards denser gradient fractions compared to 16S rRNA in both millipede species (Supplementary Fig. 7). However, the lack of pronounced peak deviation compared to the control in some replicates and days does not necessarily imply unsuccessful labelling since the labelled fraction of the community might still be too small. Indeed, there was a significant change in community composition in the heavy fractions of labelled gradients compared to unlabelled ones already by day 3 (Supplementary Fig. 8 and Supplementary Data 14).

Fig. 5: Bacterial 16S rRNA copies recovered from each fraction in the SIP gradients.
figure 5

rRNA copies relative to the total number of rRNA copies obtained from the entire gradient against the buoyant density of each fraction. Labelled RNA is expected to be found in fractions with density >1.795 g ml−1.

Full size image

For comparing heavy fractions in labelled versus unlabelled gradients of 16S rRNA, an average of 1305 ± 59 and 579 ± 41 ASVs were used for E. pulchripes and G. connexa per time point after filtering (Supplementary Data 15). Surprisingly, the model identified, on average, only 22% of the ASVs in E. pulchripes and 24% in G. connexa as labelled. These values were consistent over time. Therefore, the shift in copy-number peaks towards denser fractions, as observed in Fig. 5, was due to increased labelling in already labelled ASVs rather than a change in the proportion of labelled ASVs.

Diversity of active microbiota in a heavy fraction of 13C-RNA-SIP

In agreement with the general bacterial diversity in the gut, the major phyla whose members were flagged as labelled were Actinobacteriota, Bacillota, Bacteroidota, and Pseudomonadota (Fig. 6; Supplementary Data 15). In E. pulchripes, Bacillota comprised 35 to 55.3%, Bacteroidota 13.1 to 15.1% and Pseudomonadota from 13.8 to 23% of the total labelled ASVs. In G. connexa, Bacillota comprised 20.4 to 45.9% of total significant ASVs, Pseudomonadota ranged from 20 to 51.6%, Actinobacteriota from 15.1% to 22.6%, and Bacteroidota from 3.2 to 10.8%. Supplementary Figs. 9–15 shows the phylogenetic distribution of the labelled ASVs across the samples in each of the major bacterial classes. Despite our expectation for gradual labelling of the microorganisms with time, similar numbers and, in many cases, the same ASVs were consistently labelled throughout the incubation. In E. pulchripes, members of the classes Clostridia and the orders Bacteroidales, Rhizobiales, Enterobacterales, Desulfovibrionales, Pirellulales, Verrucomicrobiales and Victivallales were most prominent in the labelled fraction. In G. connexa, members of the class Clostridia and the orders Bacteroidales, Rhodobacterales, Enterobacterales, Pseudomondales and Micrococcales were most prominent prominent in the labelled fraction.

Fig. 6: Differentially abundant ASVs between the labelled and unlabelled gradients of the SIP experiments.
figure 6

Comparison of the relative abundance of each ASV from E. pulchripes and G. connexa. Each subfigure represents a triplicate. The plot shows the most abundant phyla in the dataset in decreasing abundance. The differential abundance of any particular ASV is given in Log2 fold change. “Rare” indicates phyla with mean relative abundance below 0.45%.

Full size image

Discussion

The gut microbiota, crucial for the ecophysiology of arthropods34, is especially vital for detritivores relying on recalcitrant plant polymers with low nitrogen content, such as senescent leaves. Building on culture-based35 and recent molecular studies21,23,36,37, the findings underscore a generally stable gut microbiota, specific to the millipede species. Variations in closely related arthropods may arise from gut conditions like pH, oxygen availability38, and gut topography19. Specifically for millipedes, hindgut volume differences, influencing redox potential, likely contribute to microbiota variations, promoting fermentation and methanogenesis in larger species (e.g., E. pulchripes and T. aoutii) but not in smaller ones (e.g., G. connexa)21,26.

Curing or sterilising arthropods to assess their dependence on gut microbiota has been conducted in various species, yielding diverse outcomes. Not surprisingly, for wood-feeding termites, exposure to high oxygen levels results in the disappearance of flagellates, leading to starvation7,39. This is because wood-feeding termites rely on short-chain fatty acids, which are the products of cellulose fermentation, for their nutrition. Cured arthropods in other studies exhibited moderate responses, including decreased feeding and altered microbiota, observed in desert millipedes24, Carabidae members40, and egg-hatching cockroaches with ootheca41. In contrast, larval Lepidoptera, exclusively feeding on fresh leaves and likely relying on simple sugars, showed no physiological response to antibiotic curing42. Both millipede species in this study maintained a stable weight, despite dramatic drops in both viable and total gut bacteria, suggesting they might not require fermentation products for nutrition. However, the notable decrease in faecal production and the relatively unchanged taxonomic composition indicated a potentially significant role in the microbiota. The decrease in faecal production could result from either a (dis)function of the gut microbiota or, more likely, considering the equal effect of sterilisation only, a result of the leaves becoming less palatable in the absence of microbial and fungal colonisation. Regarding antibiotics treatment, despite the relative stability of the community at high taxonomic levels, there was a shift in abundance towards antibiotic-resistant bacterial strains, such as Citrobacter and Bacteroides in E. pulchripes43,44 and Pseudomonas and Achromobacter in G. connexa].

This study validated CH4 release in E. pulchripes, aligning with previous findings26,45. Antibiotics decreased CH4 emission, likely disrupting bacterial fermentation, a phenomenon observed in cockroaches when bacteria and flagellates were targeted46. As expected, the application of BES, a specific methanogenesis inhibitor47, reduced CH4 production to undetectable levels without apparent effects on E. pulchripes fitness. As CH4 production serves as a hydrogen sink in anaerobic systems driving syntrophic fermentation processes48, it supports the notion that removal of gut fermentation bioproduct via metahnogenesis is non-essential for millipede nutrition. A recent study showed that while SCFA (especially acetate) are present in the mM they are negligible in the hemolympgh44. The dominant methanogens, Methanobacteriales and Methanomassiliicoccales, in our millipedes are known gut inhabitants in arthropods45. Surprisingly, despite BES suppressed CH4 production and a 10-fold drop in mcrA gene copy numbers, methanogen density in the faeces remained unaffected. In dynamic gut systems, members must continue to proliferate to avoid being flushed out, methanogens likely live as symbionts of protists, directly benefiting from fermentation products, similar to the case in termites49,50.

In the SIP experiment, RNA labelling was slow and gradual, leaving a substantial portion unlabelled even after 21 days, indicating the inefficiency of the millipede gut system in degrading leaf litter and assimilating carbon. Similar rates of labelling using isotopically-labelled cellulose were observed for forest and agricultural soils, indicating that millipede guts are not hotspots for cellulose degradation51,52. In contrast, fungal biomass exhibited faster and higher labelling, especially in G. connexa. Soil litter decomposition studies suggest fungi thrive first on recalcitrant and nutrient-poor litter, with bacteria flourishing later on nutrient-rich and more labile litter53,54. In the hindgut of both millipede species21 and Telodeinopus aoutii23, Ascomycota and Basidiomycota dominate, mirroring soil decomposition patterns53,55,56.

Despite millipedes’ ability to hydrolyse polysaccharides, lipids, and proteins through salivary gland enzymes alongside their resident microbes (as in many other detritivores13,57,58) and conditions, and methanogenesis in the digestive tract26,45,59, cellulose digestion significance in millipede metabolism remains inconclusive. Quantitative data, including low metabolic rates in millipedes fed pure cellulose, suggest challenges in maintaining a positive energy balance60.

The labelled microbiota in E. pulchripes and G. connexa, primarily Bacillota, Bacteroidota, and Pseudomonadota, show distinctive patterns associated with polysaccharide degradation, consistent with recent millipede studies21,23. Similar labelling of these phyla was observed in a scarab beetle study using 13C-cellulose61. Although certain labelled taxa (e.g., Bacteroidales, Burkholderiales, and Enterobacterales) are recognised for their role in (ligno)cellulose fermentation in millipedes21,23,37, others (e.g., members of Desulfovibrionales) are hindgut microorganisms involved in processes like sulfate reduction and are likely unrelated to fermentation. Although senescent leaves are not exclusively comprised of (ligno)cellulose, these polymers constitute approximately 50–75% of litter material62. In the near absence of other terminal electron acceptors in the gut, most other simpler carbon sources will also need fermentation for metabolism. Consequently, we conclude that while cellulolytic fermentation occurs in the millipede gut, it likely contributes minimally to the host’s diet.

If fermentation products are not a primary nutritional source for the millipede, their main nutrient origin remains a question. Classical 14C-labelling studies indicated bacterial assimilation into the millipede’s biomass surpassing that of plants but focused on lab-grown strains and omitted fungi27. Woodlice, another detritivore, exhibits a preference for fungi- or bacteria-colonised leaf tissues over natural litter63,64. Genomic and transcriptomic screening of the studied millipede species revealed glycoside hydrolases (GH) capable of degrading chitin and peptidoglycan as abundant as, or even more so than, cellulose-degrading GHs21. The decrease in ergosterol levels post-digestion supports significant fungal digestion in the millipede gut65 and some millipede species exhibit a preference for fungal fruiting bodies, algae, and lichen films66. The midgut fluid of millipedes is also known to effectively kill bacteria67. Coprophagy in millipedes may provide access to fresh microbial and fungal biomass resulting from a partial breakdown of recalcitrant plant material68. Additionally, millipedes produce endogenous GHs in their salivary glands and midgut for digesting non-structural plant material23,37,58. Certain millipede species feed on living plants, especially the agricultural crops69. These findings don’t exclude other roles of the millipede gut microbiota, such as detoxification of plant toxins70, essential compound production23, protection against pathogens36 and even acquiring new genes through horizontal transfer71.

This work demonstrates that cellulose fermentation likely plays a minor role, at best, in the millipede’s nutrition. Further work is needed to decipher their exact trophic function in nature and the potential role their microbiota plays in their survival and modulating greenhouse gas emissions.

Materials and methods

Animal collection and maintenance

We used juvenile E. pulchripes from our lab breeding colony and wild-caught G. connexa from Czechia (forest near Helfenburk u Bavorova; 49o8’10.32“N, 14o0’24.21”E). No specific permit was required for the collection. Species identification relied on morphological features72,73). Before use, the animals were kept in the lab for several weeks. Both species were housed in perforated plastic terraria, filled with commercial sand as a substrate, broken terracotta pots for shelter, and locally collected or purchased Canadian poplar (Populus × canadensis) leaf litter (see below). Canadian poplar leaves were chosen since they are favoured by the millipedes, easy to obtain both in labelled and unlabelled forms and to reduce the complexity of the experiment by using a a single source of plant. Moisture (50–60%) was maintained by spraying with tap water every other day. Both species experienced a 12-hour photoperiod. E. pulchripes were housed individually in a box (19.3 × 13.8 × 5 cm) at 25 °C and in a climate-controlled room. Meanwhile, five G. connexa individuals were kept in each box (15 × 10 × 4 cm) in an incubator (TERMOBOX LBT 165, Vanellus s.r.o.) at a temperature of 15 °C.

Antibiotic curing

Each millipede species comprised 40 individuals split into four groups of ten: Control, Sterile, diluted antibiotics (2×-Diluted in E. pulchripes and 5×-Diluted in G. connexa) and undiluted antibiotics (Undiluted in E. pulchripes and 2×-Diluted in G. connexa). Briefly, dry senescent leaves of Canadian poplar were collected from a nearby park during the autumn of 2020, air dried and kept in a plastic container at room temperature for 2 months. The Control group was fed the with unamended leaves sprayed with 500 µl of tap water 24 h feeding. For the treated groups, the leaves were autoclaved and then either sprayed with autoclaved water (Sterile group) or antibiotics solution containing penicillin G: 10,000 units ml−1, streptomycin sulfate: 10 µg ml−1 and amphotericin B: 25 µg ml−1, undiluted, 2×-Diluted or 5×-Diluted (Thermo Fisher Scientific), following Zimmer and Bartholme74. The terraria, sand, and litter were replaced weekly to maintain hygiene. E. pulchripes groups were fed around 2.4 g of litter, while G. connexa groups received 0.5 g.

The animal fitness was followed for 42 days by aseptically measuring their weights (±0.01 g). During feeding, three fresh faeces pellets (0.15–0.19 g for E. pulchripes and 0.01–0.02 g for G. connexa) were sampled from the terraria, suspended in phosphate buffer (2 ml; pH 7.4), plated in triplicates on LB-agar plates and incubated at 25 °C. After 16 h, the colonies were counted and used to quantify the bacterial load. The remaining faecal material was kept at −20 °C for DNA extraction (see below). Methane emission was also monitored (see below).

Inhibition of methanogenesis

Thirty E. pulchripes individuals were divided into three groups of ten. The Control group was fed on untreated litter, while the other two groups were fed litter treated with 5 mM or 10 mM of Sodium 2-bromoethanesulfonate (Na-BES; Sigma-Aldrich) to inhibit methanogenesis. Moisture was maintained by spraying with sterile tap water or Na-BES solution every other day. The animals’ weight and CH4 production were regularly monitored for 64 days. Methane emission measurements were conducted by placing the millipedes in sealed glass bottles with wet filter paper pieces to maintain humidity (130 ml bottle for E. pulchripes; 30 ml for G. connexa; Thermo Fisher Scientific) for 4 h at 20 °C. The control was glass vessels without animals. Headspace samples (0.5 ml) were collected at the start and the end of incubation using a gas-tight syringe and analysed on a gas chromatograph (HP 5890 series II; Hewlett Packard) equipped with a 2 m Porapak N column at 75 °C and an FID detector. The difference in CH4 concentration between start and finish was used to calculate the production rate.

Identification and enumeration of protists and symbiotic methanogens

Fourteen days post-CH4-inhibition, fresh E. pulchripes faecal pellets were crushed using a sterilised mortar and pestle, vortexed in 5 ml of 1X phosphate buffer saline (PBS) solution (pH 7.2), and then incubated at room temperature for 2–6 h to dissolve the aggregates. After spin-down, 2 µl of the supernatant was examined under a bright-field microscope(20x) using a Neubauer chamber (Sigma-Aldrich). Protists were identified and enumerated. Part of the supernatant was also used for enumerating the ciliate-associated archaea and methanogens of the Methanobacteriales and Methanomascilliicoccales orders using Catalysed Reporter Deposition Fluorescence in situ Hybridization (CARD-FISH; see Supplementary material for further details).

Stable isotope labelling of RNA

For the SIP experiment, three replicates from separate terraria were used for each species. E. pulchripes had one individual per replicate, while G. connexa had five to adjust for size differences. Millipedes were fed >96% dried, senescent 13C-labelled Canadian-poplar leaves (IsoLife, Netherlands) conditioned by spraying them with tap water and kept in an open plastic box for 5 days. Control groups were fed unlabelled leaves conditioned in the same way. Rearing conditions were maintained as described above. Control groups were fed unlabelled leaves. Rearing conditions were maintained as described above. Faecal samples were collected every 2 days for isotopic labelling analysis. Then, 1.9 g of faeces from each millipede species were vacuum dried in a SpeedVac DNA130 (Thermo Fisher Scientific) at 45 °C for 3 h, and 25 µg was transferred into triplicate tin capsules. Isotopic labelling was assessed at the Stable Isotope Facility, Biology Centre CAS, using a Thermo ScientificTM 253 PlusTM 10 kV IRMS equipped with a SmartEA Isolink and GasBench II (Thermo Fisher Scientific). The 13C at% was calculated following Hayes75). Animals were sacrificed and dissected on days 3, 7, 14, and 21 following Sardar et al.23 and stored at −20 °C for subsequent analysis. RNA was extracted from frozen hindgut samples, purified and quantified according to Angel et al.76. Hindgut samples from the SIP experiment measured 0.677–1.108 g for E. pulchripes and 0.083–0.092 g for G. connexa. See Supplementary material for further details.

Isopycnic ultra-centrifugation of 13C labelled RNA

Following RNA purification, density gradient centrifugation was performed in caesium trifluoroacetate (CsTFA) density gradients following a previously published protocol77. See Supplementary material for further details.

Gene quantification, amplicon library construction and sequencing

Pooled faecal pellet samples from the antibiotics curing and inhibition of methanogenesis experiments used for DNA extraction were 0.43–0.59 g for E. pulchripes and 0.20–0.40 for G. connexa. DNA extracts from the antibiotics treatment experiment (24 samples per species) were subjected to 16S-rRNA-gene quantification using the QX200 AutoDG Droplet Digital PCR System (ddPCR; Bio-Rad), primers 338 F—805 R and the 516 P FAM/BHQ1 probe78. DNA extracts from the methanogenesis inhibition experiment were used for quantifying the mcrA gene as a marker for methanogens using primers mlas_mod and mcrA-rev, according to Angel et al.79. Before sequencing, the cDNA from the SIP fractions (160 samples for each millipede species) was used for quantifying the 16S rRNA copies of bacteria using the same method as mentioned above and the 18S rRNA copies of fungi using the FungiQuant system80. For amplicon sequencing, the V4 region of the 16S rRNA gene was amplified and sequenced in a two-step protocol on an Illumina MiniSeq platform (Mid Output Kit; Illumina) according to Naqib et al.81. PCR amplification was performed on 10 ng of DNA or 2 µl of cDNA with primers 515F_mod and 806 R82, synthesised with the Fluidigm linkers CS1 and CS2 on their 5′ end. Sequencing was performed at the DNA Services Facility at the University of Illinois, Chicago, USA. See Supplementary material for further details.

Bioinformatic and statistical analyses

Unless mentioned otherwise, all bioinformatic and statistical analyses were done in R V4.1.183. A linear mixed-effects model84 was fitted to determine the effect of treatments and time on the millipede weight and microbial load. Differences between treatments in terms of total faecal pellet production, methane emission, mcrA and 16S rRNA copies were evaluated using an ANOVA model85 followed by Tukey’s HSD test for pairwise comparisons86. Survival analysis of the animals was computed using the Kaplan-Meier estimates87. Sequencing data was analysed as follows: primer and linker regions were removed from the raw amplicon reads using Cutadapt (V3.588). The raw reads were processed, assembled and filtered using the R package DADA2 (V1.28) with the following non-standard filtering parameters: maxEE = c(2,2) in the filterAndTrim function and pseudo pooling in the dada function89. Chimaeras were removed with the removeBimeraDenovo option. The quality-filtered pair-end reads were classified to the genus level using SILVA V13890, and those not classified as bacteria or archaea were filtered out. Heuristic decontamination was done using the decontam R package91, and unique sequences were identified and clustered in an amplicon sequence variant (ASV) table. The resulting tables were imported into the R package Phyloseq92. Read counts were normalised using median sequencing depth before plotting taxa abundance and after excluding ASVs without taxonomic assignments at the phylum level and those below a 5% prevalence threshold. Alpha diversity indices were computed using the vegan package on unfiltered and non-normalised data93 and evaluated using the Kruskal–Wallis test94 and Dunn’s test95. Corrections for multiple testing were made using the Benjamini-Hochberg (BH96) method. Values were compared and converted to a compact letter using the cldList function in the rcompanion package97. Beta diversity was calculated with a constrained analysis of principal coordinates (CAP98) from a Bray-Curtis distance matrix. Lastly, a permutational multivariate ANOVA (99; function vegan::adonis) and pairwise adonis (pairwise.adonis2 function100) were calculated using Bray–Curtis distance matrix to assess combined treatment and pairwise effects on the microbial community.

Differentially abundant genera were identified after sterile feeding or antibiotic treatment using ANCOM-BC2101 after removing all ASVs not present in at least two samples or with an abundance of less than 2. Only genera with adjusted P-values ≤ 0.05 and those passing the pseudo-count-addition sensitivity analysis were plotted.

Identification of isotopically labelled ASVs in the SIP experiment using differential abundance analysis followed Angel102. After initial processing as described above, rare taxa (with <100 total reads, present in <2 SIP fractions in a given gradient and its unlabelled counterpart). The DADA2 output sequences were aligned using sina 1.7.2103 against the SILVA V138 DB, and a maximum-likelihood phylogenetic tree was constructed using IQ-TREE V2.1.1104 with the ‘-fast‘ option. The 16S rRNA copies were plotted against the density and used to calculate absolute ASV abundances. Fractions with densities >1.795 g ml−1 (’heavy’ fractions) from each labelled sample at each time point were compared against their unlabelled counterparts using DESeq2 V1.40.1105, using the parametric fit type and the Wald significance test. Log2 fold change (LFC) shrinkage was applied using the function lfcShrink106, and the results were filtered to include only ASVs with a positive log2 fold change and a p-value < 0.1 (one-sided test).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.