Resources Modulate Developmental Shifts but Not Infection Tolerance Upon Co‐Infection in an Insect System

ABSTRACT

Energetic resources within organisms fuel both parasite growth and immune responses against them, but it is unclear whether energy allocation is sufficient to explain changes in infection outcomes under the threat of multiple parasites. We manipulated diet in flour beetles ( Tribolium confusum ) infected with two natural parasites and used a combination of transcriptomic and phenotypic assays to investigate the role of resources in shifting metabolic and immune responses after single and co‐infection. Our results suggest that relatively benign, single‐celled, eukaryotic gregarine parasites alter the within‐host energetic environment and, by extension, juvenile development time, in a diet‐dependent manner. While they do not affect host resistance to acute bacterial infection, the mRNA‐seq results reveal that they stimulate the expression of an alternative set of immune genes and promote damage to the gut, ultimately contributing to reduced survival regardless of diet. Thus, energy allocation is not sufficient to explain the immunological contribution to co‐infection outcomes, emphasising the importance of mechanistic insight for predicting the impact of co‐infection across levels of biological organisation.

1. Introduction

If we knew how individual stressors affect traits and dynamics at a given biological scale, could we predict outcomes arising from their combination? To unite and explain broad trends of species abundance, persistence and ecosystem functioning in the face of stress and change, ecological theory leans heavily on metabolic and stoichiometric models that rely on assumptions about the flow and use of resources and energy (Ott et al. 2014; Bernot and Poulin 2018). Resource‐focused theory has promoted recent advances in our understanding of ecology within organisms and the maintenance of microbiomes, symbionts and parasites (Rynkiewicz et al. 2015). Whether drawing on simplified resource allocation assumptions or more complex dynamic energy budgets, within‐host models have generated testable predictions for infection outcomes and the oscillation and persistence of parasites (Cressler et al. 2014; Ramesh and Hall 2023) as they directly or indirectly compete with immune systems for resources. As these frameworks grow in popularity, however, it is worth asking about the extent to which resource allocation can fully explain infection outcomes, particularly in the face of multiple stressors such as co‐infection.

Hosts generally endure exposure to parasites at multiple periods in their lives; co‐infection occurs when these exposure events initiate infection with two (or more) parasite species simultaneously (Tate 2019). Since both species exploit host resources and generally induce or modulate immune responses, they can facilitate or antagonise each other and lead to infection and transmission outcomes that differ from single infection scenarios (reviewed in (Rovenolt and Tate 2022)). To what extent can resource theory explain these outcomes? This question largely depends on the relative sensitivity of parasites, immune dynamics and damage repair to resource conditions within the host (Graham 2008; Clay et al. 2023). After all, resource limitation can change allocation among life history traits and reconfigure immune system investment (Adamo et al. 2016), altering the reception of incoming parasites and their transmission potential (Vale et al. 2013). Sometimes, resource availability dominates parasite competition from the bottom up, regardless of immunological regulation. In mice, for example, gut nematodes suppress immune responses, which should facilitate parasite replication and transmission (Griffiths et al. 2015). However, the nematodes primarily limit malaria parasite propagation through resource limitation by destroying red blood cells (Griffiths et al. 2015). On the other hand, helminth co‐infection in African buffalo stimulates a T cell polarisation state away from the optimal regime needed to fight coccidia (Seguel et al. 2023). While it is not clear—and indeed not probable—that this immune shift is primarily resource‐driven, it does lead to a more hospitable gut habitat for the coccidia that ultimately increases parasite shedding at the (presumably energetic) expense of host reproduction (Seguel et al. 2023).

The sensitivity of these molecular mechanisms to resource levels can be more directly tested in experiments that manipulate their availability. Recent examples of co‐infection outcomes in insect systems demonstrate limited (Deschodt and Cory 2022) or mixed effects (Zilio and Koella 2020) of resource limitation on host and parasite fitness‐associated traits, but these studies focus on infection outcomes rather than the mechanisms that drive them. To what extent does a primary infection alter the metabolic and immunological landscape encountered by a second parasite species? Does resource allocation ultimately drive these differences, or should we be selective about generalising co‐infection dynamics from energy budget models?

To address these questions, we turned to a model system for host–parasite population biology—the confused flour beetle Tribolium confusum (Park 1948) and two of its natural parasites. The first parasite, the eugregarine Gregarina confusa, is a single‐cell eukaryote from the Apicomplexa phylum. Its trophozoite stage punctures cells in the host midgut to extract nutrients. It induces a chronic infection that drains host resources and impedes development but has low virulence otherwise, providing tractability without the confounding effects of morbidity and mortality (Detwiler and Janovy Jr. 2008; Thomas and Rudolf 2010). Previous studies have demonstrated that gregarine infection in the flour beetle gut induces the differential expression, including downregulation, of several immune genes involved in antibacterial responses (Critchlow et al. 2019). The second parasite, the entomopathogenic bacterium Bacillus thuringiensis , causes acute mortality (Behrens et al. 2014) and is sensitive to the immune dynamics of the host (Jent et al. 2019).

We challenged larvae in a controlled experiment with a factorial design that included the presence or absence of gregarine exposure and a standard or nutrient‐limited diet. To quantify the extent to which the primary parasite influences the metabolic and immune landscape experienced by the second parasite, we used mRNA‐seq to investigate molecular signatures of metabolic and immunological shifts to gregarine infection and measured diet‐dependent development time and metabolite levels in gregarine‐infected or uninfected larvae. We then infected larvae with B. thuringiensis and investigated transcriptomic signatures of gregarine infection on immune and metabolic trajectories. We used these results to interpret the relative impact of resources and immunity on gut pathology, infection resistance and disease‐induced mortality. Our results suggest that, while some co‐infection‐induced shifts in life history parameters may be approximated with energy budget assumptions alone, others are largely insensitive to resources and rely instead on shifts in pathology associated with immune responses and damage repair. Since these parameters are particularly important for predicting population dynamics and parasite‐mediated apparent competition outcomes at the community level (Johnson et al. 2015; Cortez and Duffy 2020; Rovenolt and Tate 2022), ecologists should account for basic immunology before relying too heavily on resource theory for co‐infection models.

2. Materials and Methods

2.1. Beetle Rearing, Handling and Diet

Tribolium confusum beetles for this experiment were derived from a stock colony collected in 2013 from Pennsylvania, USA (Tate and Graham 2015) and subsequently kept under laboratory conditions (standard diet, 30C, in the dark). To create parental breeding groups, 60–80 adults per group from a colony were allowed to lay eggs in 16 g of flour for 24 h (in petri dishes with 90 mm diameter). We then combined all laid eggs from the parental breeding groups to avoid any block effects. Finally, we distributed the collected eggs randomly to the four experimental diets containing neither, either or both of the treatments (yeast and gregarine spores). Besides using age‐matched larvae from the same egg‐laying period, we also created staggered breeding groups 2 days apart to derive larvae of different ages but equivalent sizes.

The standard diet consists of autoclaved whole‐wheat flour (Fisher) and 5% w/w brewer's yeast (Fleischmann). To reduce the protein and nutrient quality of the diet, we excluded the yeast while providing the same diet quantity by mass (no‐yeast diet). Larvae raised on this restricted diet will still develop into adulthood but generally more slowly (Sokoloff et al. 1966). The extent to which the dietary yeast impacts the flour beetle gut microbiome or gut gene expression is poorly described, but studies that heat‐inactivate the yeast have not noted obvious differences in beetle growth rate or other life history parameters relative to studies in which the yeast is not inactivated (e.g., Korša et al. 2022). In our experiments, all diets were further modified to introduce gregarine infections by reducing new flour and adding flour derived from gregarine‐infected or clean mini‐colonies (for details see below).

2.2. Infection Protocols

The gregarines (Gregarina confusa) used in this experiment originated from an infected stock colony obtained from a teaching lab at Vanderbilt University. We subsequently maintained active parasite cultures by passaging 100 infected larvae and adults from stock colonies to fresh standard flour once per month. To create infectious flour, we added beetles from gregarine‐infected or uninfected stocks to clean flour for 4 days, providing time to deposit gregarine oocysts. We added the conditioned flour to 10 g fresh flour to create the diets (gregarine exposure: 4 g infected flour +2 g uninfected flour; no exposure: 6 g uninfected flour). This method resulted in a 60% + prevalence of gregarine infection in all experiments, as determined by gut microscopy (Thomas and Rudolf 2010), (q)PCR (f: CCTCGAGGAAGTTCGAGTCTAT, r: TTGACAGCTTGGGCACTTTAT, 400 nM efficiency = 99.2%, Tm = 55C) and/or deposited gametocyst counts (Janovy Jr. et al. 2007).

The full B. thuringiensis infection cycle involves host ingestion of cry protein‐containing bacterial spores from the environment; these toxins degrade the integrity of the insect gut to allow vegetative cells to invade the hemocoel and instigate septic infection, whereupon the bacteria replicate rapidly, secrete additional virulence factors, and ultimately kill the host before sporulating again (Garbutt et al. 2011). We chose to circumvent the initial oral stage of the infection cycle, however, to avoid two confounding issues: (1) the unusually high qualitative resistance of T. confusum to typical coleopteran cry proteins, which adds unfeasible stochasticity to the initial colonisation process regardless of dose, and (2) potential direct interactions between the gregarines and the bacteria in the midgut, which may be important ecologically but would completely confound the use of the system to produce general insights into the roles of metabolism and immunity in modulating co‐infection. Therefore, we chose to focus on the septic stage of B. thuringiensis infection, which also allows direct comparison to most other insect models of immunological dynamics and genetics that overwhelmingly use septic procedures. To challenge larvae with bacteria, we produced bacterial cultures as previously described (Jent et al. 2019). Plating of the infection culture confirmed a concentration of 1.8 × 10⁹ CFU/mL, which results in an LD50 dose (Jent et al. 2019). To septically inoculate larvae, we dipped a micro‐dissection needle in the bacterial (or saline control) aliquot and stabbed it into the space between the head and second segments.

2.3. Development and Metabolite Assays and Standardising Age Versus Stage

To monitor development into the next stage, we counted freshly emerged pupae and dead larvae daily from day 22 to 30 post oviposition and removed them to avoid resource‐deprived larvae receiving additional protein from cannibalism (Park et al. 1970). To collect size‐equivalent larvae for the metabolite measurements, we adjusted the collection dates according to the developmental delays in the different treatments. We chose to use size/mass‐matched larvae rather than age‐matched because we noticed systematic differences in the development rate among treatments during pilot experiments, and early‐instar larvae tend to have fundamentally different metabolic profiles from later instar larvae. While we would therefore expect metabolic differences among treatments in age‐matched larvae, it would primarily reflect differences in stage rather than the main effect of infection.

To quantify primary metabolites, we froze individual size‐matched larvae after obtaining their mass, washed them twice in cold insect saline solution (7.5 g/L NaCl +2.38 g/L Na₂HPO₄ + 2.72 g/L KH₂PO₄), and homogenised them (n = 17–21 larvae/diet‐pathogen treatment). A third of each sample went into each of the three performed measurements. We measured total protein content in a Bradford assay (Schulz et al. 2023). For the glucose assay, we used the GO Assay Kit (Sigma) and for lipids, a Vanillin assay (Abcam) (Barr et al. 2023).

We analysed differences in larval development time to pupation among the treatment groups (n = 77–150 larvae/treatment) using log‐rank survival analyses from the ‘survival’ package in R (R Core Team 2012; Therneau 2014). We used linear models to confirm treatment‐wise mass equivalence among larvae selected for metabolite assays (Bates 2010); both models used treatment as the main effect. After standardising lipid, glucose and protein measurements by individual larval mass, we used generalised linear models with gamma distributions to evaluate differences among parasite and diet treatments and their interactions for each metabolite (model structure: metabolite ~ parasite + diet + parasite*diet). For these and all other statistical analyses in this study, we first confirmed that the data distributions conformed to the assumptions and were otherwise appropriate for the model structure.

2.4. Survival Assays

To analyse infection‐induced mortality of gregarine‐by‐diet treated larvae, we infected size‐standardised beetles with an LD50 dose of B. thuringiensis (n = 48 infected and 8 saline‐challenged larvae/treatment/block for 4 experimental blocks). We excluded larvae that died early from the trauma of inoculation and then monitored survival from 6 to 14 h (the usual onset of mortality [Jent et al. 2019]), performing a final check at 24 h, when most larvae will have died or recovered (Tate et al. 2017). We analysed larval survival using Cox proportional hazards (coxme package in R [Therneau and Therneau 2015]) after stratifying by infection treatment (to conform to proportional hazards assumptions). Gregarine exposure, diet and their interaction served as main effects and experimental block as a random effect.

2.5. Gut Integrity Assays

To determine whether the damage or immune responses instigated by gregarine parasites accelerate mortality, we evaluated the gut barrier integrity of larvae. We randomly assigned eggs from breeding groups to the four gregarine‐by‐diet treatment groups in 96‐well microplates. After 20 days, we septically exposed 15–20 larvae per group to an LD20 dose (1.6 × 10⁶ CFU/mL) of B. thuringiensis or mock‐infected them with insect saline. Subsequently, we placed the larvae on blue dye food prepared with 2.5% FD&C blue dye no. 1 (Spectrum Chemicals) using a protocol described by (Zanchi et al. 2020). After 20 h of feeding, we examined the distribution of blue food dye under a microscope to detect leakage in the gut‐intestine barrier, scoring the beetles exhibiting a blue ‘smurf’ phenotype. We analysed smurf proportions using binomial generalised linear models with block as a main effect and then gregarine status, diet and bacterial infection status as main and interacting effects.

2.6. Time Series and RT‐qPCR Analysis of Parasite Loads

Our initial results suggested that diet did not impact disease‐induced mortality (see Results). Therefore, we chose to focus our molecular investigation of immune dynamics on the interactive effects of (co)‐infection and time post infection rather than exponentially diluting our feasible sample size and thus statistical power with the addition of a fully factorial diet treatment.

To evaluate the impact of gregarine infection on host‐ B. thuringiensis dynamics, we first collected gut samples (n = 6 pools of 8 guts/gregarine treatment) from gregarine‐exposed or clean larvae. We challenged larvae from these groups with B. thuringiensis or saline as described above. Beetles were individually sacrificed every 2 h for the 12 h of the acute infection phase (n = 8–12 individual B. thuringiensis ‐infected and 6 uninfected larvae/time point) and stored individually at −80°C. We extracted RNA using Qiagen RNeasy mini‐kits, confirmed RNA concentration using the Nanodrop, and then reverse‐transcribed RNA into cDNA (VILO mastermix). We quantified B. thuringiensis load via RT‐qPCR (SybrGreen) as previously described (Jent et al. 2019; Critchlow et al. 2024); while there is minor non‐specific amplification with these primers, they have been tested against beetle microbiota and common bacterial pathogens and only amplify specifically in the presence of B. thuringiensis (Jent et al. 2019). We devised a threshold of detection for our qPCR primers (see above) on known gregarine‐infected and uninfected samples and used them to categorise gregarine‐infected samples. We log‐transformed the linearised dCt values for normality (Jent et al. 2019) and used linear models in R (‘lm’ function) to analyse the impact of B. thuringiensis exposure, time, gregarine exposure or confirmed infection, and the interaction of time and gregarines on relative B. thuringiensis loads. Because bacterial loads bifurcate over time, and variation in high‐load beetles might not be captured across the entire load distribution, we also used a B. thuringiensis load threshold on samples from 6 to 12 h post infection to characterise beetles as high‐load, and performed logistic regression (lme4 package, glm function, family = binomial and link = logit) on high‐load status vs. gregarine exposure/infection. The results were not sensitive to the chosen threshold or whether larvae were merely exposed to gregarines or actively infected.

2.7. Transcriptome Assembly and Annotation

In addition to the gut samples, we chose whole‐body larval RNA samples from the time series (Table S1) that exhibited a near‐median bacterial load for the treatment and time point to avoid introducing load‐induced variance (Tate and Graham 2017). 150 bp paired‐end libraries were produced using the Illumina TruSeq kit and sequenced at the Vanderbilt VANTAGE core on the Illumina NovaSeq 6000 (complete statistics in Table S2). Sequencing data is publicly available on NCBI Sequence Read Archive (accession PRJNA771764). We first assessed RNAseq read quality using fastqc (Andrews 2010). There is currently no published annotated genome for T. confusum ; so using only samples not infected with gregarines, we assembled a de novo transcriptome using Trinity with default settings (contig statistics in Table S3); quality filtering was performed within Trinity and reads were assembled in paired‐end mode (Grabherr et al. 2011). We clustered highly similar transcripts using cd‐hit (Fu et al. 2012). To assess the quality of the assembly, we realigned reads to the assembled transcriptome using bowtie2 and calculated the ExN50 statistic within Trinity (Grabherr et al. 2011). We used BUSCO (Benchmarking Universal Single‐Copy Orthologs, Table S4) to assess the completeness of the assembly against an insect gene set (Manni et al. 2021).

We used kallisto v 0.48.0 to quantify gene expression (Bray et al. 2016) by performing pseudo‐alignment of RNA‐seq reads to the assembled transcriptome of T. confusum and summing count or transcript per million (TPM) values across isoforms. Because we were interested in achieving a high degree of accuracy for analyses specific to antimicrobial peptides (AMPs), we also used Coleoptera AMPs as training sets (Table S5) and constructed Hidden Markov Model (HMM) profiles using HMMER (Finn et al. 2011) to annotate AMPs in the T. confusum proteome. Since some of our analyses relied on annotation data from a well‐developed genome (Herndon et al. 2020), we filtered bit scores to combine results from BlastP and Blastx to identify T. confusum orthologs of T. castaneum genes (Figures S1–S6, S2). Full methodological details for analytical pipelines and sample processing are described in the Supplementary Methods.

2.8. Differential Expression Analyses

We used the DEseq2 (v 1.36) package in R (Love et al. 2014) to run three differential expression analyses. First, we identified differentially expressed genes in the gut upon gregarine infection relative to uninfected guts (sample details in Tables S1, S2). In the second and third analyses, we identified differentially expressed genes upon B. thuringiensis or co‐infection in the whole‐body samples at six (n = 4) or eight (n = 6) hours post infection with B. thuringiensis relative to uninfected beetles (n = 8) or beetles infected only with gregarines (n = 8). Within each time point, we modelled differential gene expression as expression ~ B. thuringiensis status + gregarine status + their interaction (false‐discovery rate [FDR] corrected p‐value < 0.05 [Benjamini and Hochberg 1995]). Because we were concerned about type II error after FDR adjustment due to the large number of annotated T. confusum genes, we also used the seSeq (2.30) package in R (Hardcastle and Kelly 2010) to investigate the AMPs specifically. To this end, we divided samples into four groups: genes not differentially expressed across samples, genes differentially expressed in samples infected by B. thuringiensis , genes differentially expressed in the gregarine‐infected samples, and genes differentially expressed in co‐infected samples relative to other samples. We reported the posterior probability of differential expression. We tried identifying TF binding sites upstream of relevant AMP genes but we have low confidence in the accuracy so the results are not presented here.

We performed weighted gene co‐expression network analysis (WGCNA; [Langfelder and Horvath 2008]) on genes with orthologs in T. castaneum to identify modules of co‐expressed genes. Genes with zero count values across all replicates were removed before analyses. Next, we constructed a signed correlation matrix for each analysis (merging threshold = 0.25, minimum module size = 30) using the count data and identified positive or negative correlations between the expression of genes in the network. We calculated the Pearson correlation of the module eigengenes across samples to identify modules of co‐expressed genes. Using the associated T. castaneum ortholog gene ids, we performed gene ontology (GO) analyses with DAVID (Huang et al. 2007) on co‐expressed modules. In addition, we performed KEGG (Kyoto Encyclopedia of Genes and Genomes) analyses on differentially expressed genes using the clusterProfiler package in R (Yu et al. 2012).

3. Results

3.1. Restricted Diet and Gregarine Infection Prolong Development, but Hosts Compensate Metabolically Unless Doubly Stressed

For optimal development, flour beetles rely on additional protein sources in their carbohydrate‐rich diet. These additional resources can come from the addition of yeast to the flour substrate. Here, we investigated how yeast restriction, gregarine infection or both affect the development and metabolic profiles of beetle larvae. In isolation, both yeast restriction (Figure 1A, Log‐rank test, developmental hazard ratio (HR) = 0.76(0.6–0.96), p = 0.024) and gregarine infection (HR = 0.73(0.55–0.97), p = 0.029) significantly and equivalently prolonged larval development time by approximately 1 day relative to reference larvae (Figure 1B). This effect was exacerbated when gregarine infection and yeast restriction were combined, leading to significantly slower development times than all other treatments (HR = 0.46(0.35–0.62), p < 0.001).

FIGURE 1.

Effects of chronic gregarine infection, resource limitation and their interaction on larval development and metabolic profiles. (A) Log‐rank statistics (hazard ratio of development time to pupation, 95% CI, and p values; note that a smaller HR means they develop more slowly) and (B) development curves for the rate of larval development to the pupal stage for gregarine‐infected larvae under standard or yeast‐restricted diets relative to uninfected larvae on standard diets. In mass‐standardised larvae (C), the effect of gregarine infection and diet on lipid (D), glucose (E) and protein content (F) were analysed using GLMs with gamma distributions (Table 1); post hoc pairwise test (BH‐corrected) bins appear in lowercase letters.

To analyse metabolic profiles across the treatments, we minimised the confounding effect of development time discrepancies by controlling for larval mass (Figure 1C; n = 21; mass range 1.3–2.6 mg; p > 0.6 among all pairwise comparisons) rather than instar, which is indeterminate in flour beetles. Neither diet nor gregarine treatment significantly affected mass‐corrected lipid levels (Table 1), but the interaction of the two was significant, as the infected and no‐yeast group had reduced lipid stores (Figure 1D; interaction p = 0.027). Diet and its interaction with gregarines significantly predicted glucose levels in opposite directions (Figure 1E), as no‐yeast diet larvae had a significantly higher glucose level than the reference group (posthoc BH‐corrected p = 0.003) but the infected and no‐yeast group had significantly lower glucose levels (interaction p < 0.0001). Protein (Figure 1F) made up a significantly higher proportion of larval body mass in no‐yeast (p = 0.005) and gregarine‐infected groups (p = 0.002); the interaction effect was not significant. These values were not significantly dependent on individual larval mass within treatments except for lipids in gregarine‐exposed individuals (Figure S3), which increased for larger larvae in the gregarine‐infected standard‐diet treatment and decreased in the gregarine‐infected low‐protein diet.

TABLE 1.

The impact of diet, gregarine infection and their interaction on within‐host metabolites in mass‐matched larvae.

	Estimate	SE	t	p
Mass
No yeast/uninfected	0.047	0.097	−0.49	0.63
Standard/infected	−0.03	0.097	−0.31	0.76
No yeast/infected	−0.011	0.097	−0.11	0.91
Lipid
Intercept	0.926	0.100	9.3	< 0.0001
Gregarine exposure	−0.088	0.140	−0.6	0.53
No‐yeast diet	0.067	0.140	0.5	0.63
Gregarine: No‐yeast	−0.412	0.197	−2.1	0.029
Glucose
Intercept	0.965	0.087	11.2	< 0.0001
Gregarine exposure	0.216	0.122	1.8	0.082
No‐yeast diet	0.392	0.122	3.2	0.0020
Gregarine: No‐yeast	−0.960	0.173	−5.5	< 0.0001
Protein
Intercept	1.423	0.051	28.0	< 0.0001
Gregarine exposure	0.134	0.039	3.4	0.00064
No‐yeast diet	0.121	0.039	3.1	0.0021

Note: Individual larval mass was analysed using an lm evaluating all groups relative to the standard diet uninfected reference group. All metabolites were modelled using gamma‐distributed glms and diet, gregarines and their interactions as main effects. The interaction effect was dropped from the protein analysis as it was not significant, and its exclusion yielded a better model fit. Values in bold: p < 0.05.

3.2. Gregarine‐Infected Guts Reveal Altered Metabolic and Immunological Profiles

A principal component analysis (PCA) of transcriptomic profiles revealed a clear separation of gregarine‐infected and uninfected gut samples (Figure S4), reflecting overall differences in both immune and metabolic processes. Differential gene expression analysis using DESeq2, followed by the identification of enriched gene ontology (GO) terms revealed upregulated genes were enriched for ribosomal, cuticular and glycolytic proteins, while downregulated genes included a bacterial recognition protein (GNBP‐1) as well as metabolic and digestive enzymes such as apolipoproteins, lipases, trehalose transporters, cytochrome P450s, cathepsin B and juvenile hormone binding proteins (Table S6). Focusing on immune effector responses, we identified the significant differential regulation of four upregulated and tightly co‐expressed antimicrobial peptides (AMPs) in the gut upon gregarine infection (Figure 2A, Table S7: Defensin‐1, Attacin‐2, Attacin‐3 and Cecropin‐3) as well as two downregulated AMPs (Cecropin‐1 and PR5‐3, which contains a thaumatin domain; Figure 2B). Weighted correlation network analysis (WGCNA), which identifies clusters of co‐regulated genes and relates terms to biological traits, did not identify any specific gene modules that were significantly associated with gregarine infection (Table S8). However, a KEGG (Kyoto Encyclopedia of Genes and Genomes) analysis, which identifies pathways associated with the differentially expressed genes, revealed significant enrichment of the ribosome. There were also non‐significant hits on steroid synthesis and signatures of altered carbohydrate metabolism (Table S9).

FIGURE 2.

Effector protein (AMP) expression patterns upon infection with B. thuringiensis , gregarines and co‐infection. (A) The Pearson correlation matrix of expressed AMPs across all samples includes highlighted clusters of co‐expressed effectors. (B) Differential expression analysis of effectors across treatments showing log2‐fold change of effectors in the gut and at 6 and 8 h post B. thuringiensis infection relative to naïve reference; DE genes delineated with asterisks. Four AMPs identified as DE upon co‐infection using Bayseq are shown with red ‘B’. (C) Expression patterns over time and treatments for three AMPs that each represent a cluster of co‐expressed effectors (underlined in A). The Y‐axis shows normalised counts (transcripts per million; TPM) and the x‐axis shows treatments. Significant differences relative to naïve are shown with asterisks.

3.3. Transcriptomic Profiles From Gregarine‐Infected and Co‐infected Larvae Reveal Altered Physiology and Unique Immune Responses to Single Versus Co‐infection

We evaluated transcriptomic profiles of whole‐body samples from individual gregarine‐infected or uninfected larvae 6 or 8 h post co‐infection with Bacillus thuringiensis (Table S1). Overall, the results suggest that co‐infected beetles express a distinct immunological and metabolic profile that is not simply an additive effect of each individual infection.

PCA on the whole transcriptomes revealed separation of samples based on B. thuringiensis or gregarine infection at 6 h post‐infection, but this separation became weakened by 8 h post‐infection (Figure S4). Focusing first on immune‐specific analyses, we discovered that two AMPs (Coleoptericin‐1 and Attacin‐1) were highly associated with the main effect of B. thuringiensis infection at both six and 8 h post‐infection (Figure 2B,C). Meanwhile, gregarine‐only infection in the whole‐body samples was associated with the downregulation of the same cecropin (Cecropin‐1) originally downregulated in the gregarine‐infected gut samples, although this effect was muted in the co‐infected samples (Figure 2C). The same AMPs that were upregulated in the gut upon gregarine infection were also uniquely upregulated in the whole‐body co‐infection samples at both 6 and 8 h time points (Figure 2B,C; Defensin‐1, Attacin‐2, Attacin‐3 and Cecropin‐3). While p‐values for these four AMPs were not significant after the correction for multiple testing (adj p > 0.05) (Tables S10, S11, Figure S5), we hypothesised that this is due to a type‐II error on the false discovery rate (FDR) because of the unusually large number (> 200 k) of draft‐annotated T. confusum genes. Therefore, we added a Bayesian approach to calculate the posterior probabilities instead of p‐values (Figure 2B), which dispenses with the need for multiple corrections. Posterior probabilities of upregulation due to co‐infection for these genes were close to one, signalling a high probability of upregulation (Figure S5) but were larger at 6 h post‐infection compared to 8 h post‐infection, suggesting that their expression is reduced after 6 h (e.g., TPM count in Figure 2C).

We ran co‐expression network analyses to identify modules of genes, rather than just individually differentially expressed genes, that are uniquely co‐regulated according to each infection treatment (Tables S12, S13). We limited our analyses to genes with orthologs in T. castaneum because Gene Ontology terms are not annotated for T. confusum . Our co‐expression results at 6 h post infection were generally consistent with our AMP‐specific analyses, such that differentially expressed AMPs within each analysis belonged to the same modules (e.g., Attacin‐1 and Coleoptericin‐1 both belong to the purple 6 h module; Table S14). Two modules showed significant associations with B. thuringiensis (purple) and co‐infection (cyan) (Figure S6), and the expression of genes within these modules was almost exclusively linked to these specific treatments. The top GO and individual differential gene expression terms associated with bacterial infection included cell adhesion proteins and wound healing/immune defence (e.g., hemocyanin activity, mucins, PGRP‐SC2, AMPs, serine proteases and serpins; Table S15). KEGG analyses indicated significant enrichment of the ribosome and oxidative phosphorylation pathways for the main effect of B. thuringiensis , while the main effect of gregarines was once again enriched for ribosomes and co‐infection for sphingolipid metabolism. GO terms most strongly associated with co‐infection (cyan module) were primarily related to ribosomes, translation and protein synthesis (Table S15). Other co‐infection associated modules (darkgreen, green, lightcyan and lightgreen) were enriched for chitin binding and various metabolic processes, although these were not significant after correction for multiple testing (Table S15). The eight‐hour differential expression, KEGG and co‐expression results generally recapitulated the six‐hour and gut results but had fewer modules with weaker associations, possibly due to the resolution of the acute immune response (Tables S11, S13).

3.4. Gregarine Infection Increases Disease‐Induced Mortality Upon Co‐Infection and Reduces Gut Tolerance to Damage

In size‐matched larvae, gregarine infection significantly increased B. thuringiensis‐induced mortality relative to uninfected and well‐fed larvae and exhibited no significant interaction effect with diet (Figure 3A; Table 2). To determine whether the extra mortality was due to differences in resistance among gregarine treatments, we measured bacterial load via RT‐qPCR across the acute phase when most mortality is initiated (Figure 3B). While bacterial load significantly increased and bifurcated over time, as previously described in this and other insect species (Duneau et al. 2017; Tate et al. 2017; Franz et al. 2023), gregarine samples were equally represented in high bacterial load trajectories (Bernoulli glm; z = 0.69, p = 0.49) relative to gregarine‐uninfected samples, and neither gregarine exposure, confirmed gregarine infection, nor the interaction of gregarines and time predicted bacterial load overall (Table 2, Table S16). Thus, gregarine‐infected individuals are not less resistant to B. thuringiensis .

FIGURE 3.

The impact of gregarines on host outcomes after B. thuringiensis infection. (A) Survival curves illustrate that saline‐challenged larvae (dotted lines; same colours as Bt‐infected legend) have high rates of survival regardless of treatment, but gregarine‐infected larvae (yellow) are more likely to die than clean larvae (blue) after B. thuringiensis infection regardless of diet (dashed: No yeast). (B) Bacterial load relative to housekeeping gene (18 s) was quantified via qPCR in B. thuringiensis ‐infected (blue: No gregs, yellow: Gregs) and saline control larvae (purple: No gregs, green: Gregs), revealing no difference in bacterial load based on gregarine status; all stats in Table SJJ. Note that while the B. thuringiensis ‐uninfected groups technically have values, all were hovering around the threshold of detection and show some non‐specific amplification; thus, seeming differences among the green and purple boxes are artefacts of background amplification. (C) The proportion of larvae revealing smurf (leaky gut) phenotypes by gregarine, diet and B. thuringiensis infection treatment.

TABLE 2.

The impact of co‐infection on survival, Bt resistance and gut integrity.

Factor	Exp (coef)	SE (coef)	z	p
Survival post Bt or saline challenge1
Bt challenge	9.47	0.25	9.05	< 0.0001
Gregarine exposure	1.39	0.14	2.39	0.017
No‐yeast diet	1.15	0.14	0.98	0.32
Gregarines: no‐yeast	0.93	0.19	−0.39	0.7

	Estimate	SE	t	p
Bacterial load over time by gregarine exposure status 2
Intercept	−19.30	1.04	−18.57	< 0.001
Time post Bt challenge	0.33	0.13	2.56	0.012
Gregarine exposure	−0.77	1.42	−0.54	0.59
Time: gregarines	0.08	0.18	0.45	0.65
Bacterial load over time by confirmed greg infection status2
Intercept	−19.43	0.83	−23.32	< 0.001
Time post Bt challenge	0.34	0.11	3.12	0.0023
Gregarine infection	−1.02	1.56	−0.66	0.51
Time: gregarines	0.15	0.20	0.76	0.45

	Estimate	SE	z	p
Proportion smurfs by gregarine exposure, Bt infection and diet3
Intercept	−2.18	0.75	−2.90	0.0037
Experimental block	−0.39	0.31	−1.28	0.20
Gregarine exposure	2.21	0.85	2.60	0.0093
Bt challenge	2.23	0.83	2.68	0.0074
No‐yeast diet	0.80	0.92	0.86	0.39
Gregarines:Bt	−2.14	1.01	−2.13	0.033
Gregarines: diet	−0.46	1.09	−0.42	0.67
Bt:diet	−0.93	1.07	−0.87	0.39
Gregarines:Bt:Diet	1.54	1.35	1.14	0.25

Note: 1. Full Cox proportional hazards model is survival ~ Bt + gregarines × diet + (1|block). Exp(coef) is the hazard ratio. Stratifying by bacterial versus saline challenge did not qualitatively change results. 2. Full linear model is log2(relative Bt load) ~ time × gregarines. Includes only those larvae challenged with Bt; background amplification rates in controls were not significantly impacted by any factors (Table SJJ). Gregarine infection was confirmed by qPCR. 3. Full logistic model is smurf status ~ block +gregarines × diet × Bt. Reduced models did not yield a significantly better fit, so the full factorial model is presented here. Values in bold: p < 0.05.

Abbreviation: Bt, Bacillus thuringiensis.

To understand whether tolerance mechanisms might instead account for the difference in mortality, we employed a smurf assay, which indicates gut leakiness through failure to maintain gut integrity or repair damaged structures (Figure 3C). We found that both gregarine infection (Table 2; z = 2.6, p = 0.0093) and B. thuringiensis infection (z = 2.7, p = 0.0074) individually predicted greater gut leakiness. B. thuringiensis infection modestly increased smurf outcomes in co‐infected beetles but not as drastically as in gregarine‐free beetles (z = −2.1, p = 0.033). Neither diet alone nor its interaction terms contributed significantly to smurf status (Table 2).

4. Discussion

To what extent does a primary infection alter the metabolic and immunological landscape encountered by a second parasite species, and does resource allocation ultimately drive these differences? These questions are critical for building generalisable frameworks to predict the consequences of co‐infection in natural populations and at different levels of biological organisation. By manipulating resource availability and monitoring both metabolic and immunological facets of the host response to co‐infection, we tested the resource sensitivity of key infection outcome parameters in a model system. Our results suggest that host development time, which contributes to age‐structured infection susceptibility (Clay et al. 2023) and population intrinsic growth rate (Pearl et al. 1941; Park 1948), is exacerbated by the dual effects of gregarines and resource limitation. On the other hand, disease‐induced mortality, which influences epidemiological dynamics and competitive outcomes in co‐infected assemblages (Cortez and Duffy 2020; Rovenolt and Tate 2022), is not as sensitive to resources; differences are instead attributable to pathology during co‐infection. Thus, mechanistic models that rely primarily on metabolic theory or energy budgets to predict co‐infection dynamics are likely to underestimate the contribution of immunological shifts.

In our study, both resource quality and gregarine infection affected development time, presumably through slower storage of resources needed to grow. After accounting for development rate (i.e., with mass‐ rather than age‐matched larvae), the metabolic state of diet‐restricted and well‐fed but gregarine‐infected larvae largely catches up to their reference peers. This leaves only the dual‐stressed group with major metabolic consequences, including lower lipid stores and a consequently higher proportion of body mass composed of structural protein. This indicates that gregarines are capable of starving their hosts or forcing them to purge metabolites to avoid oxidative stress (Li et al. 2020), but the effects are dramatic only under resource‐limited conditions; otherwise, the larvae appear to compensate by feeding more over a longer developmental window. This result aligns with our general understanding of gregarine infections as ubiquitous but relatively benign resource‐exploiting parasites that inflict noticeable costs on their insect hosts only under multiple stressors or high parasite burdens (Randall et al. 2013; Wolz et al. 2022), and sets the stage for the phenotypes we observe upon B. thuringiensis infection.

The mRNA‐seq data suggest that gregarine infection alters the immune environment in the gut through the differential regulation of antimicrobial peptides and other effectors. This result largely concurs with a previous study on gut gene expression after gregarine infection in the related flour beetle T. castaneum , although the latter exhibited much broader downregulation of antibacterial genes (Critchlow et al. 2019). This contrast raises an interesting hypothesis that gregarines may differentially affect susceptibility to co‐infection in these two co‐occurring and competing host species (Park 1948; Rovenolt and Tate 2022). The transcriptional data also suggest that the gregarines affect the gut metabolic environment more generally, which may be important for nutrient processing and damage repair. The smurf assay indicates that diet does not significantly affect gut integrity, whereas gregarine‐infected individuals have a greater baseline ‘leakiness’ regardless of diet or co‐infection status. This is clearly not enough to kill them in isolation, since control larvae have low mortality rates regardless of their gregarine status (Figure 3A). B. thuringiensis infection initiates significant damage to the guts (also shown in (Critchlow et al. 2024)), but after B. thuringiensis infection, the impact of gregarine status on additional gut leakiness is greatly diminished (Figure 3C). Thus, there must be another contributor to pathology in gregarine‐infected individuals to explain the difference in B. thuringiensis‐induced mortality.

Is it the altered immune environment? One set of co‐expressed AMPs is specific to gregarine infection (as determined by gut expression) while another set is induced by B. thuringiensis but not gregarines in the whole body. It is interesting that at the whole‐body level, the first set is highly expressed specifically upon co‐infection (rather than gregarines alone), suggesting that co‐infection activates a separate immune program in the fat body and other tissues beyond the gut. Moving beyond specific AMPs, the WGCNA modules significantly associated with B. thuringiensis infection feature many players previously identified in RNA‐seq studies of B. thuringiensis in flour beetles (e.g., bacterial recognition, immune signalling and defence molecules, cytochrome P450s, serine proteases, glycolysis enzymes (Behrens et al. 2014; Tate and Graham 2017)). The modules most significantly associated with co‐infection, however, are full of protein synthesis and metabolic genes, suggesting a struggle to effectively manage the physiological response. Interestingly, there is not an observable difference in bacterial load between gregarine‐infected and uninfected larvae, suggesting that neither mortality differences nor gene expression patterns are attributable to differences in resistance. Instead, these co‐infection‐specific modules point to differences in infection tolerance (Louie et al. 2016), possibly due to increased pathology of the alternate immune responses and/or their co‐expressed genes or an increased struggle to maintain homeostatic metabolic and tissue repair programs in the gut. As we further improve the annotation of the T. confusum genome, we will be able to test these hypotheses with functional genomics approaches.

It is worth noting that gregarines are parasites of the midgut but are not the only residents there. Previous studies have suggested that Tribolium beetles require an intact gut microbiome to form proper immune responses and prime against microbes previously encountered (Futo et al. 2016). The combination of resource limitation and the presence of gregarines might lead to dysbiosis of the gut microbial communities and thereby increase host susceptibility to secondary infections, even to microbes outside the gut, like the bacterial infection in this study. Further research is needed to clarify the interaction of eugregarines, which are ubiquitous parasites of arthropods, with the microbiome of their hosts.

While we did not directly measure the production and spread of transmission stages, our results do hint at the consequences of co‐infection for parasite fitness. B. thuringiensis is an obligate killer and relies on making spores in its dying or newly dead host to achieve transmission (Garbutt et al. 2011). Gregarines, on the other hand, mate in the living host gut to produce oocysts that are shed into the environment, and a dying host also spells a dead end for gregarines (Janovy Jr. et al. 2007). Thus, the exacerbated mortality in co‐infected hosts undoubtedly hurts gregarine fitness, but it is not entirely clear that B. thuringiensis benefits because bacterial loads were not higher in co‐infected individuals at the time of peak mortality. Future studies would benefit from new protocols for accurately quantifying gregarine transmission so that we can understand how this class of parasites, ubiquitous in the arthropod world (Rueckert et al. 2019), influences disease dynamics for biopesticides and vectored infections that preoccupy agricultural and biomedical efforts.

In conclusion, resources clearly matter—both the mRNA‐seq and phenotype data suggest that the gregarines are indeed acting like parasites in depriving their hosts of resources and altering metabolic efficiency. Based on the metabolite data, the host can compensate for the parasitism when resources are not strictly limiting, but gregarine presence does change the immunological landscape in the face of secondary infection and may exert additional pathological effects. When it comes to infection mortality outcomes, the shifting immune landscape and physical damage inflicted by the gut parasite overshadow the importance of variance in resources. To the extent that many parasites across taxa inflict tissue damage and alter the immune landscape, either through manipulation or host‐mediated alternative responses, this is likely to be a generalisable problem. Thus, mechanistic models should allow for resource‐independent contributions of immune responses when predicting or generalising co‐infection dynamics.

Author Contributions

A.T.T. conceived the project and provided funding; N.K.E.S. and A.T.T. designed the experiments. N.K.E.S. and S.L. executed all experiments except the gut assay (conducted by A.P.). S.S.L.B. and A.M.W. assembled the T. confusum transcriptome. D.A. performed all differential expression and most statistical analyses and made the figures, with contributions from A.T.T. and N.K.E.S., A.T.T., D.A. and N.K.E.S. wrote the manuscript, with edits from all authors.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figures S1–S6

Tables S1–S16

Data Availability Statement

RNA‐seq data is publicly available in the NCBI Sequence Read Archive (project accession PRJNA771764). Experimental data, derived data needed to produce the figures, and associated R code are available in Zenodo (Tate 2024; DOI: 10.5281/zenodo.13313067).