We used D. punctata based on availability and as a follow-up to our previous study (13). We maintained individuals (females) on three diets for 28 days. Females were chosen for this study, as starting point for further studies investigating resource allocation in this viviparous species under similar dietary regimes. We assessed potential diet-induced changes in the gut microbiome and microbial EAA provisioning functions of the microbiome at the end of the feeding period and investigated gut microbiome-mediated changes in mass-specific standard metabolic rate (SMR) on days 1 and 28. Diploptera punctata cockroaches used in the colony were obtained from the Evolution, Ecology, and Organismal Biology department and the Insectary and BioSci Greenhouse at The Ohio State University in early 2022. Cockroaches were maintained in ventilated plastic containers, provided with water ad libitum weekly, and fed pulverized dog food (Nestle-Purina, St. Louis, MO). Cockroaches were maintained at 28 °C and a relative humidity of 40%. The individuals used in the study were all females from the main colony.

Thirty females maintained on a dog food diet in the main colony were selected and assigned to three different diets: an optimal composite dog food diet (high-quality diet, DF), a sub-optimal cellulose-amended dog food diet (70% cellulose: 30% dog food) (CADF) (13, 20), and an entirely plant-based diet consisting of granulated root tubers of cassava (Manihot esculenta), called Gari purchased from a local African grocery store in Omaha, NE (n = 10 per each treatment). We determined each female’s standard metabolic rate (SMR) by measuring the CO₂ production twice: on day 1 of the experiment and day 28 after the animals were fed the treatment diet. Before each CO₂ measurement, all females were starved for 24 hours to ensure they were in a post-adsorptive stage. CO₂ production by each female was measured using open-flow respirometry (26). Briefly, insects were weighed and placed into a 10cc syringe as a respiration chamber. The syringe was placed into a dark incubator set to 22 °C and connected to the respirometry system for 20 minutes. Cockroaches were acclimated to the 22 °C temperature for ~30 minutes. The 5–7 °C temperature difference between the insect colony and the incubation temperature for the SMR measurement is within the ∼10 °C range within which mass-specific SMR is not significantly affected by temperature changes in cockroaches (27, 28). Air was pushed through the system at a constant rate of 150 mL/min with a mass-flow meter (SS4; Sable Systems, North Las Vegas, NV, USA). CO₂ and water vapor removed from the air flowing into the insect-containing syringe by passing it through soda lime (Fisher Scientific™, Pittsburgh, PA, USA) and drierite (Drierite Co. Ltd, Xenia, OH, USA), respectively. The air entered the syringe through a rubber stopper in the plunger opening of the syringe barrel and left the syringe through the tip of the syringe. The air containing insect-produced CO₂ was routed through a condensation column and subsequently through drierite (Drierite Co. Ltd, Xenia, OH, USA) to remove water vapor. This vapor-free air was then moved into the CO₂ analyzer (Qubit S151, Kingston, Ontario, Canada). CO₂ production was recorded every second using an electronic interphase (UI-2; Sable Systems, North Las Vegas, NV, USA) and the Expedata software 1.6.0 (Sable systems, North Las Vegas, NV, USA). The baseline was determined before and after each trial by measuring CO₂ content of the air flowing through the empty syringe for an average of 5 minutes.

We calculated the mass-independent CO₂ production (VCO₂) by each insect using the equation: VCO₂ (mL CO₂ h⁻¹) = flow rate (150 mL/min) × (FeCO₂ − FiCO₂)/1 − FeCO₂ × [1 − (1/RQ)] (26). FiCO₂ is the CO₂ content of the air flowing into the syringe, which is 0, and FeCO₂ is the mean amount of CO₂ produced during the 20-minute trial by each insect minus the baseline. The baseline is the mean CO₂ content of the air measured before and after the trial. RQ is the respiratory quotient. We estimated VO₂ (mL O₂ h⁻¹) as: VO₂ (mL O₂ h⁻¹) = RQ × VCO₂ (mL CO₂ h⁻¹). We chose an RQ of 0.85 because the insects had fasted for 24 hours, were postabsorptive, and presumably used mostly fatty acids (29). Furthermore, we included the respiratory (RQ) in the equation to control for errors in the measurement of CO₂ production (VCO₂) when oxygen consumption is not simultaneously measured (26). As a conversion factor to express metabolic rate as energy consumption instead of oxygen use, we used 20.8 J mL⁻¹ O₂ (30). We calculated the mass-independent metabolic rate (MR) using the equation: MR (J h⁻¹) = VO₂ (mL O₂ h⁻¹) × 20.8 (J mL⁻¹ O₂). To determine mass-specific MR, we weighed each cockroach immediately before and after measuring CO₂ production to control for mass change during the CO₂ measurements. We calculated mass-specific MR by dividing the mass-independent MR by the mean mass.

After measuring CO₂ production on day 28, cockroaches were killed by freezing and storing at −20 °C. Before DNA extraction, we surface-sterilized insects (DF = 5, CADF = 10, Gari = 10) by washing them in a 1% detergent solution for one minute and two one-minute rinses in deionized water. The digestive tract was removed and used for DNA extraction, and the remaining insect carcass (head, thorax, legs, and abdomen) was frozen at −80 °C for stable isotope analyses. According to the manufacturer’s directions, DNA extraction was done using the Qiagen DNeasy Blood & Tissue Kits (Qiagen Inc. Germantown, MD, USA). We verified the presence of the microbial 16S rRNA marker gene in all extracted DNA samples via PCR using the universal 27F and 1492R bacterial primer pair (31). Samples were subsequently submitted for high-throughput paired-end Miseq library preparation and sequencing at the University of Nebraska Medical Center Genomics Core. Briefly, a limited cycle PCR reaction was performed on each sample to create a single amplicon, including the V4 (515-F) and V5 (907-R) variable region (32). The resulting libraries were validated using the Agilent BioAnalyzer 2100 DNA 1000 chip (Agilent, Santa Clara, CA, USA), and DNA was quantified using Qubit 3.0)(Qubit™, Thermofisher, Waltham, MA, USA). A pool of libraries was loaded into the Illumina MiSeq at 10 pM. The pool was spiked with 25% PhiX (a bacteriophage) at 10 pM for MiSeq run quality as an internal control (33) to generate 300 bp paired ends with the 600 cycle kit (version 3). The raw reads were deposited into the Sequence Read Archive database (BioProject Number: PRJNA1017785).

Mass and mass-specific SMR data were normally distributed. We evaluated the impact of diet on cockroach SMR on day one and day 28 using repeated measures ANOVA in JMP PRO using the Standard Least Squares (SLS) regression approach with a Restricted Maximum Likelihood (REML) estimation method. The model included dietary group (DF, CADF, and GARI) and time (day 1 and day 28) as fixed factors, their interaction, and individual females as random factors. We used an unbounded covariance components structure for the dataset. Data used in the analyses is provided in Supplementary Table S1 .

Acquired fastq primer-trimmed Miseq paired-end reads from the sequencing center were processed using DADA2 (34). Reads with more than two expected erroneous base calls were identified as part of the PhiX bacteriophage genome for quality control, and less than 175 base pairs across both forward and reverse reads were filtered out. Forward reads were truncated to 250 base pairs, and reverse reads to 200 base pairs. Truncation was done to maintain median quality scores above 30 across samples. Reads were merged, and chimeras were subsequently filtered out. We determined amplicon sequence variants (ASVs) and representative sequences against the SILVA 138.1 16S rRNA gene reference database (35). We combined the count and taxonomy information for the generated ASVs into a classical OTU/ASV table, and further analyses were carried out in QIIME v.1.8 (36, 37). Before analyses, we curated the table by removing unclassified reads at the bacterial or archaeal domain level and assigned to Eukaryota. Furthermore, for this analysis, we removed all reads assigned to the cockroach obligate fat body endosymbiont, Blattabacterium sp. This was done to avoid skewing our analyses of gut bacteria with the endosymbiont since our focus was explicitly on EAA provisioning by the gut bacteria and not the endosymbiont. Finally, samples with less than 1000 reads per sample were removed from the table before analyses. We then summarized the filtered and curated ASV table to the family level, and all subsequent analyses were carried out using this table.

For diversity analyses, we rarefied the family-level table to a minimum acceptable number of 1990 reads per sample across all samples to ensure we had enough replicates per treatment group. Furthermore, given the relatively low number of reads across samples, we performed rarefactions at each level up to 1990 a maximum of ten times. The rationale and justification for rarefying have been discussed elsewhere (38, 39). For alpha diversity, the diversity matrices chao1 (40), Simpson’s index (41), and Shannon’s evenness (42) were calculated in QIIME. Significant differences among categorical groupings were determined using the non-parametric Wilcoxon tests in JMP Pro 15 (S.A.S., Cary, NC, USA). We generated the Bray-Curtis dissimilarity distance matrix (43) using the 1990-rarefied table. The distance matrix was used to calculate non-metric multidimensional scales (NMDS) in QIIME. The NMDS scales are used to visualize categorical sample groupings that differ in microbiome composition. After that, we tested for differences among these categorical groupings via permutational multivariate analysis of variance (PERMANOVA) (44) in QIIME using the Bray-Curtis distance matrix as input. We assessed significant differences in abundance of ASV between dietary groups using the group_significance command in QIIME.

Before quantifying δ¹³C_EAA, Diploptera carcass samples were lyophilized under vacuum at −80 °C for 48 hrs. Samples were then pulverized and submitted for analysis at the Stable Isotope Facility at the University of California, Davis (Davis, CA, USA). Briefly, samples were first acid-hydrolyzed for 70 min under a N₂-gas headspace in 6M HCl at 150 °C. Samples were then derivatized as N-acetyl methyl esters via methoxy carbonylation-esterification (NACME) (45, 46). Essentially, derivatized samples were injected into a splitless liner at 260 °C and separated on an Agilent DB-35 column (60 m × 0.32 mm ID × 1.5 µm film thickness) at a flow rate of 2 mL/min under the following temperature program: 70 °C (hold 2 min); 140 °C (15 °C/min, hold 4 min); 240 °C (12 °C/min, hold 5 min); and 255 °C (8 °C/min, hold 35 min). Compound-specific isotope ¹³C-amino acid analysis (δ¹³CA) was carried out using a Thermo Trace GC 1310 (GC; Thermo Fisher Scientific, Waltham, MA, USA) coupled to a Delta V Advantage isotope ratio mass spectrometer via the GC IsoLink II (Thermo Electron, Bremen, Germany) (see 46 for analytical details). All samples were analyzed in duplicate. Replicates of the quality control and assurance reference materials are measured every five samples. Standard exogenous carbon addition, kinetic isotope effects from derivatization reagents procedures, and normalization to the international reference for δ¹³C VPDB, as well as calibrated amino acid mixture, UCD AA3, and multiple natural materials were carried out as per facility practices. δ¹³C-EAA data from all samples were obtained for isoleucine, leucine, lysine, phenylalanine, threonine, and valine. The mean standard deviation for all samples was ± 0.21 ‰, well below the established quality control value of ± 0.44%. Final accuracy, as determined by the mean absolute difference in the measured and known δ¹³C values of EAAs from a quality assessment mixture of amino acids, was within ± 0.46 ‰. Analyses of δ¹³C_EAA values and enrichment among insect samples and dietary groups (Gari, DF, and CADF) were carried out using ANOVA with insect treatment groups and amino acids (all six EAAs) as factors, following normalization to respective dietary δ¹³C_EAAs in JMP (SAS).

To investigate EAA provisioning by gut microbiota, we first examined the δ¹³C offset between consumers and their diets (47). A δ¹³C offset exceeding 2‰ typically signals that the gut microbiota supply EAAs to the host. Without such microbial input, the host would solely depend on its diet for these vital nutrients (3, 4). To pinpoint the origins of de novo synthesized EAAs in host tissues, we employed a method known as ¹³C-EAA fingerprinting (5). This technique utilizes linear discriminant function analysis (LDA) and relies on δ¹³C-EAA training data with bacterial, fungal, and plant origins. Our training data was adapted from Larsen et al. (6) following interlab calibrations. Subsequently, we carried out a supervised discriminant analysis to categorize insect and dietary samples into their respective classifier groups (6), followed by a Wilks Lambda test of the accuracy of separation of classifiers. This method also tests for significant differences among centroids of the classifiers. The LDA was executed using the R package MASS (R version 4.2.2.; http://www.R-project.org).

The Dietary group × time interaction on body mass change was not significant (P = 0.29), and neither was the dietary group (P = 0.89), but time (F _(1,14) = 7.66, P = 0.015) was significant. Overall, average body mass significantly increased over time, ranging from a 3.48% increase in the DF group to an 8.85% increase in the CADF group and a 17.04% increase in the Gari group from day 1 to day 28. This change in body mass between day 1 and 28 on all three diets is represented in Figure 1A .

We found a significant Diet × Time interaction on SMR of cockroaches (F-value _(2,15) = 3.87; P = 0.045), as well as a significant time effect on SMR (F-value _(1,15) = 4.76; P = 0.045). Mass-specific SMR was constant for DF-fed cockroaches, whereas SMR decreased in CADF-fed and Gari-fed cockroaches by 16.3% and 33%, respectively ( Figure 1B ).

Quality trimming and filtering of the raw data resulted in 3,053,626 combined reads retained from an initial 4,543,250 reads (67%) that were then assigned to 3,580 ASVs. Subsequent curating of the ASV table yielded 3,529 ASVs (mean reads per sample = 35,268; Minimum: 1991, Maximum: 95,875) distributed across 22 insect samples (DF = 5, CADF = 9, Gari = 8), and a total of 775,896 reads. Rarefaction curves indicated that microbial diversity was sufficiently covered across samples.

There were no significant differences in gut bacterial species richness estimates across diets (chao1, χ2 = 0.75, P = 0.68; Shannon’s index, χ2 = 0.60, P = 0.74; Simpson’s index, χ2 = 0.45, P = 0.80). We did not find significant differences in microbial community composition of the cockroach gut microbiomes across the three diets (PERMANOVA, test statistic = 1.04, P = 0.37) ( Figure 2A ). However, we found differences in the abundance of 20 ASVs belonging to five bacterial phyla (Firmicutes, 47.88%; Bacteroidota, 37.69%; Planctomycetota, 13.27%; Proteobacteria, 0.68%; and Desulfobacterota, 0.45%) and fourteen (14) bacterial families in the gut microbiome of females fed the three diets (P < 0.01) ( Figure 2B ) ( Supplementary Table S2 ). In the CADF diet group, microbiomes were dominated by Firmicutes (Streptococcaceae, 68.7%), Proteobacteria (Xanthomonadaceae, 11.7% and Enterobacteriaceae, 9.82%). The gut microbiomes of the DF diet group were dominated by Bacteroidota (Paludibacteraceae, 31.4%, Williamwhitmaniaceae, 10.82%, Tannerellaceae, 1.1%), Planctomycetota (vadinHA49, 24.03%), Firmicutes (Oscillospiraceae, 10.18%, Erysipelotrichaceae, 3.68%, Christensenellaceae, 3.04%), and Desulfobacterota (Desulfovibrionaceae, 1.52%). Finally, the gut microbiomes of cockroaches fed on Gari were dominated by Bacteroidota (Paludibacteraceae, 52.3%, Dysgonomonadaceae, 8.2%), Planctomycetota (vadinHA49, 14.43%), and Firmicutes (Unassigned Clostridia, 8.03%, Christensenellaceae, 3.5%, and Enterococcaceae,1.8%).

We obtained δ¹³C_EAA data for six EAAs. Our data analysis revealed significant δ¹³C enrichment relative to diets across all six EAAs for all three treatments (F_(5,102) = 6.41, P < 0.0001). Insects fed on Gari exhibited the highest δ ¹³C-enrichment (4.2 ± 0.4, mean ± S.E. across all EAAs), followed by those on the DF diets (2.6 ± 0.4 across all EAAs). The CADF-fed insects showed the least enrichment (1.8 ± 1.0 across all EAAs) ( Figure 3A ). In terms of individual EAAs in the Gari treatment, Lys had the highest δ ¹³C enrichment (4.2 ± 0.6), followed by Val (3.9 ± 0.6) and Ile (2.6 ± 0.6). The δ¹³C-enrichments in EAAs of sampled insects relative to the three diets are shown in Figure 3A . The observed significant ¹³C offsets in these EAAs suggest gut microbial provisioning to the host of these essential nutrients. We investigated the origins of these de novo synthesized EAAs in the studied cockroaches using the δ ¹³C-EAA fingerprinting approach. The model based on training data from Larsen et al. (6) successfully separated bacteria (N = 11), fungi (N = 9), and plants (N = 12) into distinct groups (F = 55, P < 0.0001; Wilk’s lambda = 0.007, a test of the appropriateness of separation of classifiers and for group membership prediction of non-classifier samples). The supervised LDA with samples with unknown EAA origins ( Figure 3B ) assigned the DF and CADF diets near the fungal classifier group. This can be attributed to the composite nature of the DF and CADF diets. The Gari diet was assigned between the fungi and plant classifier groups. For insect samples, the Gari-fed samples exhibited the most displacement from the dietary source towards bacterial and fungal classifiers, indicating possible fungal and bacterial sources of EAAs. In contrast, the DF-fed and CADF-fed insect samples were close to the fungal classifier group, possibly reflecting the reliance on fungal crude protein in the diet.