Four colonies of Silvestritermes euamignathus and Cornitermes cumulans (Termitidae: Syntermitinae) were collected from pasture areas in the Campinas-SP region (22° 46’29.0”S, 47° 04’06.5”W). To maximize genetic heterogeneity and ecological independence, the colonies were collected within a ~1 km radius from a central coordinate point. This spatial scale is appropriate for these termite species, whose natural dispersal by flight is limited, ensuring that sampled colonies are genetically distinct. The nests were kept in 50 L plastic containers in a climate-controlled room at 26°C, under a 12-hour photoperiod cycle, with water and grass provided as food. Sampling was authorized by the Brazilian Institute of Environment and Renewable Natural Resources (IBAMA), the enforcement agency of the Brazilian Ministry of the Environment (SISBIO permit no. 72676-4).

Feeding experiments were conducted using artificial diets varying in carbohydrate complexity. From each of four colonies, 50 workers and five soldiers were transferred to 10 cm Petri dishes containing sterilized, moistened sand and a 3 cm disc of analytical-grade pure agar (Sigma-Aldrich), supplemented with one of the following treatments to standardize the matrix across all diets:

For each dietary treatment within each colony, we prepared 20 replicate dishes. These were distributed as follows: five dishes were used for gut microbiota sequencing, five for consumption bioassays, and ten for enzymatic assays.

After five days, surviving termites were enumerated, and their digestive tracts were dissected and stored at −22°C for enzymatic and molecular analyses (Menezes et al., 2018). An in-situ sample (50 workers per colony) was retained without experimental manipulation. Food consumption was quantified as the difference between initial and final fresh weight of diet discs, adjusted for humidity loss using control dishes (food without termites). Differences in consumption and survival among treatments were assessed via Generalized Linear Mixed Models (GLMMs) with Gaussian (consumption) and Poisson (survival) error distributions, implemented in the glmmTMB package (Brooks et al., 2017). Pairwise comparisons were conducted using estimated marginal means with the emmeans package (Lenth et al., 2023), and residual distributions were evaluated with the DHARMa package (Hartig, 2022).

The entire guts of 50 workers from each colony subjected to each food treatment and from the in-situ sample were placed in 2 mL tubes containing 1 mL of lysis buffer (500 mM NaCl, 50 mM Tris-HCl, pH 8.0, 50 mM EDTA, and 4% sodium dodecyl sulfate). DNA extraction was performed using a bead-beating protocol as previously described [27]. DNA integrity was assessed via agarose gel electrophoresis (1.0% w/v), and quantification was conducted using a NanoDrop spectrophotometer by measuring absorbance at 260 nm. To optimize the amount of DNA recovered, replicate samples from each treatment were pooled prior to extraction.

The V4 region of the 16S rRNA gene was amplified using primers 515F (5′ TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGGTGCCAGCMGCCGCGGTAA 3′) and 806R (5′ GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGGGACTACHVGGGTWTCTAAT 3′), with Illumina overhangs (underlined) (Menezes et al., 2018). Library preparation involved two PCR steps. In the first step, specific primers were used for each library with Phusion Polymerase (Thermo Scientific) and 40 ng of template DNA per sample. The amplification conditions were: initial denaturation at 98°C for 2 min, followed by 30 cycles of 98°C (30 s), 60.1°C (30 s), and 72°C (40 s), concluding with a final extension at 72°C for 5 min. The second PCR step incorporated Illumina sequencing adapters and dual index barcodes (Nextera Index Kit, Illumina) into the amplified libraries. This reaction used Phusion polymerase (Thermo Scientific) with 100 ng of purified PCR products from the first step as the template, along with indexing primers from Illumina. The amplification conditions included an initial denaturation at 98°C for 3 min, followed by 5 cycles of 98°C (30 s), 55°C (30 s), and 72°C (30 s), ending with a final extension at 72°C for 5 min. Each sample was amplified in triplicate. Pooled samples were purified using Agencourt AMPure Magnetic Beads (Beckman Coulter) and quantified with a Qubit Fluorometer 2.0 (Invitrogen). Sequencing was performed on the Illumina MiSeq platform at the Brazilian Bioethanol Science and Technology Laboratory (CTBE/CNPEM) using a MiSeq Reagent Kit V3 (600 cycles). Afterward, clean reads were processed using the DADA2 v1.16 pipeline (Callahan et al., 2016) with default parameters to generate amplicon sequence variants (ASVs). Taxonomic assignment was performed using the SILVA reference database (Quast et al., 2013). Tables generated by DADA2 were imported into phyloseq package (McMurdie and Holmes, 2013).

We used R version 3.5.0 (R Core Team, 2021) to conduct analyses using different packages. Downstream analysis, including α- and β-diversity analysis was calculated using the microeco and MicrobiotaProcess packages (Liu et al., 2021; Xu et al., 2023). The ASVs table was merged with relevant metadata into a microeco object. Generalized linear mixed models (GLMM) were performed to check for overall significant differences of α-diversity estimates and the total number of ASVs among samples. We fitted multivariate generalized linear models (mvabund package) (Wang et al., 2012) to test the effects of the diet on the microbial relative abundance. Models were fitted with a negative binomial distribution using 999 bootstrap iterations. The manyglm function was used to carry out the analysis. Ordination analyses were used to evaluate microbial composition at the ASV level. We used GLMM analysis to identify ASVs that define the differences between microbiomes.

Functional predictions based on the 16S rRNA amplicon data were conducted using the Tax4Fun2 version 1.1.6 package (Aßhauer et al., 2015) and the reference profiles (Ref99NR) from the Kyoto Encyclopedia of Genes and Genomes (KEGG) database. Similarities greater than 97% were considered orthologous KEGG groups (KOs). The output included CAZy encoded genes (https://www.cazy.org/).

The enzymatic activity against cellulose and hemicellulose substrates was evaluated using crude enzyme extracts from the gut of workers from four colonies, each subjected to different dietary treatments, following the methodology described by Franco Cairo et al. (2011). The assays aimed to assess the activity of the soluble fraction of protein extracts against both natural polysaccharides and synthetic oligosaccharides with varying monomer compositions and branching patterns. Crude protein extraction was performed using the guts of 50 workers from each colony. The samples were homogenized in 2 mL of 100 mM sodium acetate buffer (pH 5.5) and then centrifuged at 20,100 × g for 30 minutes at 4°C. The supernatant was collected, and 1 µL of Protease Inhibitor Cocktail (Anresco) was added per mL of crude extract to prevent protein degradation. Protein concentration in each extract was determined using the Bradford method (Bradford, 1976).

Each enzymatic reaction consisted of 10 µL of crude protein extract (3.2μg/μL), incubated at 37°C for 40 minutes with 40 µL of 50 mM sodium acetate buffer (pH 5.5) and 50 µL of a 0.5% specific natural polysaccharide solution (in water), performed in triplicate (Franco Cairo et al., 2011). The carbohydrate substrates used in the enzymatic assays included CMC (carboxymethyl cellulose, low viscosity) (β-1,4-carboxymethylglucan), β-glucan from barley (β-1,4-glucan), xylan from oat spelt (β-1,4-xylan), rye arabinoxylan (α-2,3-arabinose-β-1,4-xylan) from citrus, and pNP-G (4-nitrophenyl β-1,4-D-glucopyranoside). To terminate the enzymatic reactions, 100 µL of dinitrosalicylic acid (DNSA) was added, followed by heating at 99°C for 5 minutes. The resulting color change was measured at 540 nm using a TECAN M2000 plate reader. Blank reactions were prepared using the same procedure, except that DNSA was added before incubation with the protein extract. Enzymatic activity was expressed in mmol of glucose equivalents produced per mg of protein. For the assays using synthetic oligosaccharide the same enzymatic reaction was performed as described above, however using 2 mM pNP-G (4-nitrophenyl β-1,4-D-glucopyranoside) as substrate. The reaction was stopped with 100 µL of 1M sodium bicarbonate, followed by heating at 37°C for 30 minutes. The color change was analyzed at 412 nm using a TECAN M2000 plate reader. p-nitrophenyl were used for standard curve construction. Enzymatic activity was expressed in mmol of p-nitrophenyl released per mg of protein. To evaluate the effect of diet on the enzymatic activity of termite workers, a Wilks’ Lambda-type nonparametric multivariate inference was conducted using the npmv package (Ellis et al., 2017).

We identified a total of 2,222 bacterial ASVs (508,218 reads) and 90 archaeal ASVs (18,147 reads) in the gut microbiota of S. euamignathus workers, as well as 982 bacterial ASVs (289,091 reads) and 52 archaeal ASVs (14,812 reads) in C. cumulans workers ( Supplementary Tables S1 , S2 ). The most abundant phyla—Bacillota, Actinomycetota, and Bacteroidota—accounted for 67% and 69% of sequence reads in S. euamignathus and C. cumulans, respectively ( Figures 1C ; Supplementary Figure S1 ). At the genus level, Papillibacter was the most abundant in S. euamignathus, while Acinetobacter dominated in C. cumulans ( Supplementary Figure S1A ). The gut microbiota composition differed significantly between these termite species (manyglm; Dev = 35.21; df = 1; p < 0.001) ( Figure 1D ), with only 6.3% of ASVs shared between them ( Supplementary Figure S1B ). Diversity indexes were not significantly different between the two termite species ( Supplementary Figure S1C ; Supplementary Table S3 ).

Diet variation had no significant effect on food consumption or survival in either the litter-feeding Cornitermes cumulans or the polyphagous Silvestritermes euamignathus ( Supplementary Figure S2 ; Supplementary Table S4 ). However, in C. cumulans, artificial diets led to notable shifts in both diversity ( Supplementary Figures S3 , S4 ) and relative abundance of specific bacterial taxa, particularly within the phyla Bacillota and Pseudomonadota, which varied across feeding regimes. In contrast, S. euamignathus showed no substantial changes in taxonomic abundance across diets ( Figures 2A, B ; Supplementary Tables S1 , S2 ). It is worth noting that no data were available for the agar-only control in C. cumulans, as individuals from this treatment did not feed and suffered complete mortality during the assay. Despite these taxonomic fluctuations in C. cumulans, overall gut microbiota composition remained relatively stable in both species, as indicated by non-significant results from multivariate analyses (S. euamignathus: Deviance = 9.292, p = 0.068; C. cumulans: Deviance = 4.890, p = 0.092) ( Figures 2C, D ). These results suggest that C. cumulans exhibits some taxon-specific sensitivity to dietary changes, while maintaining a broadly stable community structure, whereas S. euamignathus displays resilience at both compositional and abundance levels under dietary stress.

The predicted functional inference of gut microbiota revealed differences between the two termite species in response to dietary treatments. In S. euamignathus, the dextrose diet strongly induced enzymes involved in simple sugar metabolism, including alpha-amylase (EC. 3.2.1.1) and alpha-glucosidase (EC. 3.2.1.20) ( Figure 3A ). However, individuals fed complex fiber-rich diets such as CreaFibe and Arbocel, or the wild treatment, showed moderate to higher profiles of ligninases (EC. 1.11.1.9 and EC. 1.11.1.15) and hemicellulases (EC. 3.2.1.10 and EC. 3.2.1.23) ( Supplementary Table S5 ). Enzymatic activity assays also showed consistent profiles across dietary treatments, although individuals fed the dextrose diet exhibited slightly elevated activity for some substrates, retaining higher activity on β-glucan and pNP-G substrates even when exposed to artificial diets (Wilks’ Lambda = 7.91; df = 15, 36.29; p < 0.001) ( Figure 3C ; Supplementary Table S6 ).

In contrast, Cornitermes cumulans displayed a pronounced reduction of enzymes involved in polysaccharide degradation—particularly those related to starch, cellulose, and hemicellulose—when exposed to artificial diets ( Figure 3B ; Supplementary Table S7 ). This functional suppression was reflected in significantly lower enzymatic activities against substrates such as xylan, CMC, PNPg, and arabinoxylan compared to wild-fed individuals (Wilks’ Lambda = 3.14; df = 20, 130.29; p < 0.001) ( Figure 3C ; Supplementary Table S8 ). Altogether, these results demonstrate that S. euamignathus is more resilient to dietary shifts, maintaining functional digestive enzyme activity across a range of nutritional environments. C. cumulans, in contrast, exhibits more pronounced reductions in enzymatic performance under simplified or artificial diets, suggesting a narrower metabolic scope and lower tolerance to dietary change.