Our P. americana colony has been maintained in captivity at the University of Georgia for over a decade. Mixed age and sex stock insects are maintained at room temperature in glass aquarium tanks with wood chip bedding and cardboard tubes for shelter in a 12:12 light:dark cycle. Water via a cellulose sponge fit to a Tupperware reservoir and dog food [Purina ONE chicken & rice formula, guaranteed analysis: 26% crude protein (min), 16% crude fat (min), 3% crude fiber (max)] are provided to the stock colonies ad libitum.

The synthetic diets created for dietary testing were designed to provide balanced nutrition while remaining malleable to component manipulation. The diets contained Vanderzant vitamin mix (catalog #: 903244, MP Biomedicals, Irvine, CA, USA), Wesson salt mix (catalog #: 902851, MP Biomedicals), peptone (catalog #: J636, Amresco, VWR International, Radnor, PA, USA), casein (catalog #: C3400, Sigma-Aldrich, St. Louis, MO, USA), and cholesterol (catalog #: 0433, VWR); the amounts are listed in Table 1 . The dry ingredients were suspended in sufficient volumes of diH₂O to create a batter or dough, formed into pellets, and then dehydrated at 65°C until they were sufficiently dry to maintain shape. Food pellets were stored at -20°C until use.

In most experiments, the only component changed was the carbohydrate source. The polysaccharides used include microcrystalline cellulose (MCC with 51um particle size; catalog #: 435236, Sigma-Aldrich), methylcellulose (catalog #: M0512, Sigma-Aldrich), xylan from corn core (catalog # TCX0078, TCI Chemicals, Portland, OR, USA), pectin from apple (catalog # 93854, Sigma-Aldrich), starch from potato (catalog #: S516, Fisher Chemical), and chitin (catalog #: J61206, Alfa Aesar, Ward Hill, MA, USA). For simple sugar diet variations, cellobiose (catalog #: 22150, Sigma-Aldrich), glucose (catalog #: G7021, Sigma-Aldrich), and xylose (catalog #: 200001-008, Acros Organics, VWR International) were used as the carbohydrate component.

Experimental conditions were prepared as described in Tinker and Ottesen (2016). Briefly, mixed-sex healthy adult insects (n=12/diet) were transferred from the stock colony to plastic tanks containing pebbles and bleached polyvinyl chloride tubes for footing and shelter, respectively. Food and water were provided ad libitum in rigid plastic or glass dishes following two days of food restriction and habituation. Dietary treatments for the four cohorts ( Supplementary Table 1 ) lasted 2 weeks, during which debris, oothecae, and lethargic insects were removed daily.

Upon completion of dietary treatments, all insects were sacrificed for sample collection. Insects were isolated in a sterile culture plate and placed on ice until torpid, upon which they were decapitated and dissected. Sternites were removed with sterile forceps to expose the intact gut and fat body tissue was cleaned away. The cleaned gut was frozen on a sterile aluminum dish on dry ice and divided into foregut, midgut, and hindgut sections for collection in 500-800µL phosphate-buffered saline (1X PBS). Gut contents and tissue-attached bacteria were disrupted with a sterile pestle, and the samples stored at –20°C until DNA extraction. For this study, only the hindgut community was analyzed due to its higher microbial density and activity than in other gut regions.

DNA was extracted from 200µL aliquots of all individual samples using the EZNA Bacterial DNA Kit (Omega Biotek, Norcross, GA, USA) with some modifications. Sample aliquots were centrifuged at 5000g for 10min, with the resulting pellet resuspended in 100µL TE buffer plus 10µL lysozyme (50mg/mL) and incubated for 30 min at 37°C. Following incubation, samples were vortexed with glass beads (25mg, Omega Biotek) for 5 min at 3000rpm, then incubated at 55°C for 1 hour with 100µL TL buffer, 20µL proteinase K, and continuous 600rpm shaking. The kit protocol was followed for additional incubations with BL buffer and DNA isolation using the provided column. DNA was eluted into 50µL of the provided Elution Buffer and quantified using either a Nanodrop Lite spectrophotometer (Thermo Scientific) or the Take3 plate for BioTek plate readers (Agilent).

The V4 region of the 16S rRNA gene was amplified via a 2-step polymerase chain reaction (PCR) from individual hindgut lumen samples (n=8-12/diet) as previously described in (Tinker and Ottesen, 2016; Tinker and Ottesen, 2020; Tinker and Ottesen, 2021). Both PCR reactions used 0.02U/L Q5 Hot Start high-fidelity DNA polymerase (New England BioLabs, Ipswich, MA, USA) with 200µM dNTPs and 0.5µM forward and reverse primers in 1M Q5 reaction buffer. The first 10µL reaction containing 3ng DNA and primers targeting the V4 region (515F: GTGCCAGCMGCCGCGGTAA; 806R: GGACTACHVGGGTWTCTAAT) was performed under the following conditions: activation at 98°C for 30s; 15 cycles of 98°C for 10s, 52°C for 30s, and 72°C for 30s; final extension at 72°C for 2 min. Immediately following amplification, 9µL of the first reaction was added to 21µL of Q5 reaction mix containing barcoded primers with adaptor sequences for Illumina sequencing (Caporaso et al., 2011). Cycling was performed as follows: activation at 98°C for 30s; 4 cycles of 98°C for 10s, 52°C for 10s, and 72°C for 30s; 6 cycles of 98°C for 10s and 72°C for 1 min; final extension at 72°C for 2 min.

After product size verification via gel electrophoresis, samples were cleaned as instructed in Omega Biotek’s Cycle Pure kit, quantified, and pooled for equimolar representation of each sample. Prepared libraries were sent to the Georgia Genomics and Bioinformatics Core at the University of Georgia for 250 base pair paired-end Illumina MiSeq sequencing.

Each dataset collected in this study was processed separately in R (version 4.2.1) by sequencing run using R package DADA2 (version 1.24.0), with the cumulative Amplicon Sequence Variants (ASVs) generated input as a priors table for each successive run (Callahan et al., 2016; RStudioTeam, 2020). To allow for comparison with this dataset, raw data from previous research in the Ottesen lab were reprocessed to generate ASVs following the same procedures as in this current study (Tinker and Ottesen, 2016). All sequence tables produced by these datasets were combined by ASV sequence prior to taxonomy assignment to ensure continuity in naming. Taxonomy was assigned using DADA2 and the ARB Silva v138 classifier to the species level, uniquely numbered, and filtered to remove sequences matching eukaryotic (chloroplast, mitochondria) or endosymbiotic Blattabacterium DNA (Quast et al., 2013; Callahan et al., 2016).

Alpha and beta diversity analyses were performed via the R package vegan (version 2.6-4) (Oksanen et al., 2019). Samples were rarefied prior to diversity analysis to 7,924 reads for comparisons between synthetic and/or whole-food diets, 9,685 reads for analysis of repeat xylan and MCC diet experiments, and 12,274 reads for follow-up experiments exploring nutrient deficiencies and simple sugar carbohydrates. Alpha diversity was measured via the Shannon index, the count of ASVs observed in rarefied samples, and Pielou’s evenness (calculated as Shannon/log(Observed)). Weighted (binary = FALSE) and unweighted (binary = TRUE) Bray–Curtis dissimilarities were calculated, assessed for dispersion, and plotted using the vegan functions vegdist, betadisper, and metaMDS. Unweighted Bray–Curtis dissimilarity, or incidence-based Bray–Curtis dissimilarity, is equivalent to the Sørensen index in that it is based on the number of species shared or unique between groups without accounting for individual species abundance (Legendre and De Cáceres, 2013). Statistics for alpha diversity indices were calculated with the Wilcoxon rank sum test (pairwise comparisons) and the Kruskal–Wallis test (multi-group comparisons). The significance of community composition differences observed in beta diversity measures was assessed using PERMANOVA (vegan::adonis2()). Beta dispersion was further examined through Tukey’s HSD test for pairwise comparisons and ANOVA for multi-group comparisons.

Differential abundance analysis was conducted using DESeq2 (version 1.36) (Love et al., 2014). For identification of ‘diet-characteristic taxa”, raw count data for the synthetic diet set (n=66) were filtered to exclude ASVs present in fewer than five samples and run through the ‘DESeq’ command with parameters ‘fitType = “local”‘ and ‘design = ~ Diet’. Pairwise result tables were obtained for all diet comparisons and filtered for significant data, defined as having an adjusted p-value smaller than 0.05 and a baseMean larger than 10. ASVs significantly upregulated for one diet vs the other five diets were identified as diet-characteristic (n=76) and used to generate the heatmap in Supplementary Figure S3 . For comparison of “synthetic vs whole”, raw count data for both diet sets (n=125) were combined and filtered to exclude ASVs that appeared in fewer than five samples. DESeq2 was run with parameters ‘fitType = “local”‘ and ‘design = ~ Diet_Type’ to identify differentially abundant ASVs between the diet types. The resulting baseMean and log2 fold change were used to generate the MA plots.

UpSetR (version 1.4) was employed to visualize intersecting sets of taxa, providing insights into the distribution of taxonomic features across samples (Conway et al., 2017). For UpSet analysis, samples were rarefied to 7,924 reads and then collapsed together to obtain total counts per diet. Both a presence/absence table and a proportion table were generated from these data, with the presence/absence table used for UpSet graph generation. The relative abundance of each set was calculated using the proportion table with ASVs collapsed per set and visualized as pie charts within the UpSet graph.

Co-correlation analysis was conducted to evaluate the impact of synthetic diets on microbial interaction networks using the SparCC procedure (Friedman and Alm, 2012). Networks were constructed separately for the synthetic diet group and the whole-food diet group using sequence count tables that were filtered to only include ASVs with at least five representatives present in 25% of samples (synthetic: 17 samples; whole food: 15 samples), preventing spurious correlations from rare taxa. SparCC was implemented in R with standard parameters, and the resultant networks were characterized and analyzed with the igraph R package (version 1.5.1) (Csardi and Nepusz, 2006; Csárdi et al., 2024). Networks were pruned to contain only edges with a correlation absolute value of at least 0.4 and exported into Cytoscape for visualization using the edge-weighted spring-embedded layout method (Shannon et al., 2003).

We formulated a series of synthetic diets composed of a fixed base of 25% protein amended with dietary salts, vitamins, and cholesterol while differing only in complex carbohydrate type. Initial experiments utilized five alternative polysaccharide sources: chitin, methylcellulose, MCC, pectin, or xylan. Following the initial analysis of these results, we tested an additional starch-based diet. The prepared diets were readily consumed by the cockroaches in all cases.

To evaluate the impact of these diets on the gut community, each diet was fed to adult cockroaches (n=12/diet) for a period of 14 consecutive days, after which the insects were sacrificed, and their hindgut dissected out for 16S rRNA gene library sequencing. Following library preparation and sequencing, we used DADA2 to obtain 2,321,848 quality-controlled, assembled sequences assigned to 3,308 ASVs after the removal of endosymbiont (Blattabacterium sp.) and mitochondrial sequences (Callahan et al., 2016). At the phylum level, at least 80% of each sample was dominated by Bacteroidota, Firmicutes, and Desulfobacterota, in agreement with previous studies on the cockroach gut microbiome ( Figure 1A ) (Tinker and Ottesen, 2016; Tinker and Ottesen, 2020; Dukes et al., 2023). The relative abundances of these three phyla were similar across all samples excluding xylan-fed cockroaches; these insects hosted notably more Firmicutes and less Desulfobacterota than cockroaches fed other diets ( Figure 1A ).

Alpha diversity, as measured by the Shannon index, evenness, and community richness, significantly differed across diet treatments ( Figures 1D, F ; Kruskal Wallis p<0.001 for each). Pairwise analyses found that chitin-fed insects possessed higher Shannon index values (p<0.05) and community evenness (p<0.01) than those of MCC- and starch-fed insects, while the xylan diet resulted in lower alpha diversity measures than all other diets (p<0.05 for each).

Beta diversity analyses using weighted and unweighted Bray Curtis dissimilarity, which is also known as the Sørensen index, revealed significant impacts of our synthetic diets on gut microbiome composition. On average, between-diet variation was greater than within-diet variation ( Figures 1G, H ), with xylan-fed communities producing distinct communities compared to the other synthetic diets. Ordination analyses using non-metric multidimensional scaling (NMDS) and PERMANOVA analysis showed that samples clustered based on diet composition in both weighted ( Figure 1B ; PERMANOVA: R² = 0.42; p<0.001) and unweighted ( Figure 1C ; PERMANOVA: R² = 0.329; p<0.001) Bray–Curtis metrics, with especially clear separation of the xylan-based diet from other synthetic diets. Removing the xylan-fed samples from the diversity calculations did not eliminate diet-based clustering for weighted ( Supplementary Figure S1A ; PERMANOVA: R² = 0.343; p<0.001) or unweighted ( Supplementary Figure S1B ; PERMANOVA: R² = 0.247; p<0.001) measures, suggesting that each carbohydrate source enriched for a unique community composition.

We used DESeq2 to identify 76 microbes that exhibited significantly higher abundance in a single synthetic diet across pairwise comparisons against all other treatments, which we termed “diet-characteristic taxa” ( Supplementary Figure S2 ) (Love et al., 2014). Diet-characteristic ASVs were primarily assigned to Firmicutes (n=48) and Bacteroidota (n=20); other phyla with diet-responsive taxa include Fusobacteriota, Deferribacterota, Desulfobacterota, Fibrobacterota, and Spirochaetota. We found that the chitin and methylcellulose diets were not associated with any diet-characteristic taxa by this definition, while the diets made with xylan, MCC, starch, and pectin enriched for 45, 10, 13, and 8 ASVs respectively.

To confirm that the diet-associated gut community differences we observed are genuine rather than artifacts of natural variation in the insect colony (“cohort effects”), we prepared fresh MCC- and xylan-based diets and repeated the 2-week diet experiment with a new cohort of adult cockroaches. These diets were selected for follow-up experiments due to both their contrasting molecular structure and the dissimilarity they generated in the Bray–Curtis analyses ( Figures 1B, C ). Data from the first and second experiments exhibited similar alpha diversity measurements ( Figures 2A–C ), with the repeated cohorts maintaining the significant shifts in alpha diversity (p<0.001) observed in the initial experiment between these two diets while showing no difference between the same-diet cohorts. Beta diversity analysis showed that samples clustered by both cohort and diet ( Figures 2D, E ). Diet had large effects on both weighted (PERMANOVA: R² = 0.34; p<0.001) and unweighted (PERMANOVA: R² = 0.20; p<0.001) Bray–Curtis dissimilarity, while the cohort explained minimal effect sizes of 3.7% and 5.8% for the weighted and unweighted measures respectively, with only unweighted reaching significance (p<0.01).

Because unweighted Bray–Curtis calculations only consider whether an ASV is present or not in an individual community, we hypothesized that the difference between the cohorts was driven by low-abundance microbes that had lesser impact in the calculation of abundance-weighted beta diversity. Using Venn diagrams ( Figure 2F ), we confirmed that most of the ASVs recovered were in fact shared across cohorts. While 616 ASVs (27%) were unique to cohort 1 and 423 ASVs (20.2%) unique to cohort 2 ( Figure 2F , grey pie slices), these ASVs represented only a small fraction of the total sequences ( Figure 2F , black pie slices) obtained from each cohort. Further, 66.7% (cohort 1) and 61% (cohort 2) of these ‘cohort-specific’ taxa appeared in only one sample ( Supplementary Figure S3 ), indicating that most of the differences in composition due to the time between the studies stemmed from transient, rare taxa. Separating the diets for these comparisons confirmed the overall findings, with rare taxa contributing few sequencing reads despite comprising 19.7%-33.7% of the unique ASVs ( Figure 2F ). Altogether, these results show that synthetic diets reproducibly alter the gut microbiome composition in cockroaches.

While fiber is the primary component of undigested material that reaches hindgut microbiota, other macro- and micronutrients are known to influence gut community structure in non-cockroach host systems (Zhu et al., 2015; Zhao et al., 2017; Akimbekov et al., 2020). To confirm the role of polysaccharides as key modulators of the gut community, we leveraged our synthetic diets to test the impacts of xylan and MCC-based diets deficient in protein or micronutrients. For protein-deficient diets ( Table 1 ), casein and peptone were replaced by mass with either xylan or MCC, while in vitamin-deficient diets, both the vitamin mixture and cholesterol were replaced with additional polysaccharides. We also created simple sugar versions of these two diets to test whether replacing long-chain fibers with their component backbone sugars results in different gut communities: xylose for comparison with xylan, and cellobiose and glucose for comparison with MCC.

In unweighted Bray–Curtis analyses, which consider only the presence/absence of ASVs ( Figure 3B ), communities from all MCC-fed insects overlapped with those of the sugar diets while retaining separation from xylan-fed samples (PERMANOVA: R² = 0.36; p<0.001) suggesting that MCC and the sugar diets supported a shared set of microbiota that were absent in the xylan-fed insects. When abundance of ASVs is accounted for via weighted Bray–Curtis ordination ( Figure 3A ), the sugar-fed communities formed their own distinct cluster, while both xylan-fed and MCC-fed samples clustered by polysaccharide regardless of vitamin or protein content (PERMANOVA: R² = 0.434; p<0.001). The alpha diversity profiles of the deficient diets matched the standard MCC or xylan diets as well, as measured by the Shannon index ( Figure 3C ), evenness ( Supplementary Figure S4A ), and number of observed ASVs ( Supplementary Figure S4B ). The cellobiose communities displayed slightly higher Shannon index values than the standard and protein-deficient MCC communities, while the xylose-fed insects possessed noticeably more even and diverse communities than the xylan diets. At the genus level, ( Figure 3D ) the community composition of the insects fed diets containing the same polysaccharide resembled each other, while sugar-fed microbiota reflected each other more than the polysaccharide they were derived from. Despite the xylose-fed samples clustering with the other sugar diets, the pentose diet did enrich for an abundant Lachnoclostridium ASV ( Supplementary Figure S5 ) that is heavily associated with xylan, while MCC-associated taxa such as Fibrobacter and Ruminococcus remained at low levels in the glucose and cellobiose diets. Overall, these results suggest that the communities observed in the fiber diets are driven by the long-chain structures of the polysaccharides rather than their component sugars.

The different microbial communities triggered by our synthetic diets were unexpected given that previous experiments examining the impact of whole-food diets with strongly differing macronutrient profiles did not produce substantially different gut microbiome compositions (Tinker and Ottesen, 2016). Therefore, we compared the samples from this current study (“synthetic” diet type) to samples from the previous study (“whole food” diet type) of cockroaches fed butter, tuna, honey, white flour, or whole wheat flour (Tinker and Ottesen, 2016). Both studies fed diet treatments ad libitum to groups of adult mixed-sex American cockroaches for 2 weeks, following the experimental setup described in Methods section 3.3.

We found that the gut microbiome samples from cockroaches fed synthetic diets exhibited higher ASV richness (p<0.01) but lower evenness (p<0.001) and Shannon index values (p<0.01) than those from insects fed whole foods ( Supplementary Figures S6A-C ). Synthetic and whole-food diets produced distinct diet type clusters in the NMDS ordinations ( Figures 4A, B ) for weighted (PERMANOVA: R² = 0.105; p<0.001) and unweighted (PERMANOVA: R² = 0.152; p<0.001) Bray–Curtis dissimilarities. When we analyzed the samples by diet, we found that diet explained more variation in the NMDS ordination than diet type as interpreted from PERMANOVA R² values (weighted: R² = 0.393; unweighted: R² = 0.369).

Beta dispersion analysis of variation within diet types showed that, together, the gut microbiota of cockroaches fed synthetic diets was more variable than that observed among whole-food-fed cockroaches ( Supplementary Figures S7A, B ; Tukey’s HSD: p<0.001). However, when diets were analyzed individually, they were equally dispersed in weighted Bray–Curtis dissimilarity ( Supplementary Figure S7A ) but not unweighted measures ( Supplementary Figure S7B ; ANOVA: p<0.001). The xylan-fed cockroaches exhibited significantly greater within-group variability than the ten other diets (Tukey’s HSD range: p = 0.037 – 3.17e^-06), with no significant differences observed in the pairwise comparisons of all other diets. When synthetic diets were compared to whole-food diets without including xylan-fed samples, we observed no significant differences in unweighted beta dispersion ( Supplementary Figure S7D ) or Shannon index values ( Supplementary Figure S6B , red boxes). However, significant differences remained in weighted Bray–Curtis dispersion ( Supplementary Figure S7C ), richness and evenness ( Supplementary Figures S6A, S6C ), and in both weighted and unweighted Bray–Curtis ordination analysis ( Supplementary Figures S8A, B ), highlighting that the altered microbiomes produced by the synthetic diet type were not solely due to biases produced by xylan-fed samples.

To verify that the inclusion of whole-food dietary components alone was not sufficient to eliminate fiber-dependent gut microbiome configurations, we tested the impact of diets mimicking our synthetic diets but with the purified protein components replaced with canned tuna. These diets induced community compositions similar to those observed in polysaccharide-matched diets containing purified proteins rather than supporting protein-associated communities ( Supplementary Figure S9 ). Xylan-containing diets generally produced communities with lower alpha diversity scores ( Supplementary Figures S9D-F ) and clustered away from the MCC-containing diets and dog food-fed insects we included as controls in weighted ( Supplementary Figure S9B ; PERMANOVA: R² = 0.395; p<0.001) and unweighted ( Supplementary Figure S9C ; PERMANOVA: R² = 0.392; p<0.001) analyses. Despite the discordant structural complexities between tuna fish and purified casein/peptone amino acids, the protein portion of the synthetic diets exerted less influence than the polysaccharide source.

Given these strong differences in community structure, we utilized the R package UpSetR to determine how the ASVs in the different diet types overlap (Conway et al., 2017). UpSet plots are akin to Venn diagrams, considering only the presence/absence of an ASV. Rarefied count tables were aggregated by diet and ASVs were marked as either present or absent per diet. ASVs present in the same subset of diets were grouped into “Sets”, with the phylum-level composition per set depicted as stacked bar charts labeled with the number of included ASVs ( Figure 4C ). We supplemented the UpSet plot with pie charts illustrating the relative abundance (calculated as the fraction of total reads recovered from the collapsed treatment group) of reads assigned to ASVs within each set ( Figure 4A ), in addition to pie charts representing the “core” ASVs present in all whole-food or all synthetic diets, regardless of presence in the other dietary group ( Figure 4D ). For simplicity’s sake, Figure 4 and Supplementary File 1 show only the five largest intersecting sets as well as all single-diet sets; additional sets are presented in Supplementary Figure S10 .

A total of 492 ASVs (“Set 1”) were shared across all diet treatments ( Figure 4C ). These ASVs made up over half of the sequences recovered for all diets except the MCC diet, for which they represented 49% of sequences ( Figure 4C , pie charts). Only 43 ASVs (“Set 4”) were exclusive to the whole-food diets, contributing between 0.9% and 1.65% of the reads in these diet sets. The 57 ASVs (“Set 3”) identified as exclusive to the synthetic diets made up 1.6%-3.4% of the reads recovered from the starch-, pectin-, chitin-, and methylcellulose-fed cockroaches, and 7% of the xylan-fed and 13% of MCC-fed cockroaches. Together, these results indicate that the synthetic diets did not eliminate the core taxa present in the guts of cockroaches fed whole foods, nor did they result in hindgut colonization by a large new set of microbial taxa.

Similarly, individual synthetic diets were not associated with hindgut colonization by large groups of unique microbes. In general, taxa that were unique to individual synthetic or whole-food diets represented a very small proportion of sequences recovered ( Figure 4C ). ASVs unique to the MCC diet formed the second largest set overall, with 160 diet-unique taxa, yet they only represented 0.4% of total recovered sequences ( Figure 4C ). A xylan-based diet, which repeatedly produced the largest community dissimilarities ( Figures 4A, B ; Supplementary Figure S7 ), was associated with 90 diet-specific taxa comprising only 0.63% of reads. Interestingly, our analysis revealed that a xylan-based diet did result in the loss of 136 taxa that were present in all other diets in abundances ranging from 4.78%-10.14% (“Set 2” in Figure 4C ). However, other sets that excluded individual diets were substantially smaller ( Supplementary Figure S10A ) suggesting that this was not a common mechanism underlying the diet-driven differences in gut microbiome composition. Instead, synthetic diet-driven differences in gut microbiome composition were primarily associated with high relative abundance of individual taxa that were consistently present in the cockroach gut regardless of dietary treatment.

To follow up on these observations, we used DESeq2 to assess the enrichment of individual ASVs between synthetic- vs whole-food-fed cockroaches. Of the 492 ASVs found, often at high abundances, in all 11 diets, 95 ASVs were significantly enriched in cockroaches fed synthetic diets while 38 ASVs were enriched in cockroaches fed whole-food diets ( Figure 4E ; padj < 0.001). The magnitude of these ASV-level differences across diet type were modest (log fold change <5), consistent with the high proportion of reads recovered in all diets that belonged to Set 1 ( Figure 4C ). Bacteroidota, Desulfobacterota, and other phyla showed similar enrichment distributions to each other in terms of both the number of ASVs and abundance within samples, but the 49 Firmicutes enriched in the synthetic diets included far higher individual abundances than the six enriched in the whole-food diets.

We also examined the enrichment of ASVs that fell outside of Set 1 but were both somewhat abundant (baseMean > 1) and present in at least five samples ( Supplementary Figure S11 ). These 1,270 ASVs generally had smaller abundances than the members of Set 1 ( Figure 4E ), but greater log-fold changes between diet types. The ASVs Ruminococcus_NA and Fusobacterium ulcerans were exceptions, being both highly abundant (baseMean = 1020 and 651, respectively) and unique to the synthetic diets. In contrast, the Christensenellaceae ASV (R7_NA.68), unique to whole-food diets, had a high log-fold change, but a baseMean of only 4.29. Apart from a few highly abundant ASVs in the synthetic diets, most of the abundant, diet-enriched taxa were shared by all diets, supporting that change is driven by common gut bacteria restructuring the community rather than interloping bacteria disrupting the community.

To explore the impact of diet on microbial taxa associated with fiber degradation, we evaluated the dietary responses of abundant taxa within the Firmicutes and Bacteroidota. For this analysis, we selected the two most abundant representatives of each of these phyla from each diet and examined their abundance across all diets ( Figure 5 ). We identified 16 Firmicutes and 11 Bacteroidota as most or second-most abundant in at least one diet.

The most abundant taxa from both Firmicutes and Bacteroidota represented a small fraction of reads across the whole-food diets, consistent with the higher Shannon diversity and evenness of the gut microbiome for whole-food-fed cockroaches ( Supplementary Figures S6B, C ). In contrast, the synthetic diets produced strong ‘blooms’ of individual ASVs, particularly among the Firmicutes ( Figure 5A ). Several abundant Firmicutes were both present in Set 1 ( Figure 1C ) and enriched in individual synthetic diets ( Supplementary Figure S2 ), namely Lachnoclostridium_NA (xylan), Lachnoclostridium_NA.1 (xylan), Enterococcus_NA (xylan), Enterococcaceae_NA (xylan), and ChristensenellaceaeR7_NA (pectin). In contrast, the Firmicutes identified as enriched in the MCC and starch diets were not typically found across all diets: Ruminococcus_NA (MCC), Lachnospiraceae_NA.5 (starch), and Ruminococcaceae_NA (starch). Among the Bacteroidota ( Figure 5B ), we observed greater overlap in the most abundant taxa present in each diet group, with all but one (3M1PL1.52termite_NA) of the abundant Bacteroidota ASVs belonging to Set 1, and only two classified as diet-characteristic: 3M1PL1.52termite_NA (MCC) and Bacteroides_NA.12 (pectin).

We constructed co-correlation networks with SparCC to examine the community structure underlying synthetic and whole-food microbiome data sets (Friedman and Alm, 2012). To filter out noise and reduce spurious correlations, the datasets were filtered to include ASVs present in at least 25% of the samples for each diet type, resulting in 976 ASVs for whole-food diets and 700 ASVs for synthetic diets. The networks were further pruned to contain edge weights with absolute values larger than 0.4, removing 75 and 168 ASVs from the whole-food and synthetic networks respectively. Both positive and negative edges were retained for network layout formation ( Supplementary Figure S12 ), but for analysis, only positive edges were considered ( Figure 6 ). After negative edge removal, the whole-food network contained 875 ASVs with 9,515 edges forming two connected components ( Figure 6A ), while the synthetic diet network contained 497 ASVs with 2536 edges that formed six connected components ( Figure 6B ). Networks at SparCC correlation levels of 0.5, 0.6, and 0.7 are included in Supplementary Figure S13 .

Gut microbiota from the whole-food-fed cockroaches formed an extensive and dense interaction network ( Figure 6A ), with higher edge counts ( Figure 6C ), node counts ( Figure 6D ), node degree ( Supplementary Figure S14A ), network strength ( Supplementary Figure S14B ), and betweenness scores ( Supplementary Figures S14C, D ) than those fed synthetic diets at most levels of filtering based off SparCC correlation values. High-degree nodes, or ASVs with large numbers of neighbors, were present throughout the whole-food network structure, while in the synthetic diets, two primary clusters of ASVs appeared with fewer connecting ASVS. When a range of inclusion cutoffs was considered, the synthetic diet network degraded more quickly than the whole -food network into separate connected components ( Figure 6E ; Supplementary Figure S13 ) displaying greater fragility and a tendency to fragment. These overall network structures suggest that the synthetic diets disrupted the stability of the cockroach microbiome.