Pathogen growth and virulence dynamics drive the host evolution against coinfections

Significance

Coinfection is widespread in nature, with a significant impact on global health. However, the lack of a strong empirical framework has hindered our understanding of how coinfecting pathogens might influence the host adaptation and evolution of immunity. Our study, which utilized mathematical modeling and experimental evolution with an insect model, shows that the individual growth and virulence patterns of coinfecting pathogens critically regulate the host adaptation to coinfections. Pathogens that grow rapidly and cause quick mortality impede the host’s ability to successfully adapt to coinfection. Furthermore, detailed comparative immune gene expression analyses support our model predictions and observed phenotypic patterns, leading to significant progress in probing the evolutionary patterns and processes in complex real-world coinfection scenarios.

Abstract

The occurrence of coinfections, where hosts are simultaneously infected by multiple pathogens, is widespread in nature and has significant negative impacts on global health. In humans, over one-sixth of the world’s population is affected by coinfections, contributing to several diseases. However, despite the broad ecological relevance and impact on global health, most biomedical research has focused on understanding interactions between a single host and a single pathogen. The extent to which coinfections could impact host adaptation and immune system evolution, particularly in comparison to infections by single pathogens, thus remains largely unknown. Also, what roles do individual pathogen species play in this evolutionary process? To address these questions, in this study, we combined theoretical modeling and experimental validation in a model insect Tribolium castaneum evolving against two coinfecting bacterial pathogens with contrasting growth (e.g., fast- vs slow-growing) and virulence (fast- vs slow-killing) dynamics. Our findings show that fast-growing pathogens causing rapid mortality surges (i.e., fast-acting) can effectively limit the host’s adaptive success against coinfections. While hosts rapidly evolved better survival against slow-growing bacteria causing long-lasting infections, adaptation against coinfections was significantly delayed and resembled the slow rate of adaptation against fast-acting pathogens. Finally, RNAseq analyses revealed that the observed delay in adaptation was associated with the limited scopes for suitable immune modulations against fast-acting pathogens. They might also be costly and pleiotropic (e.g., phenoloxidase activity), posing challenges for further immunomodulation and slowing adaptation. Our study thus highlights how individual pathogens’ growth and virulence dynamics critically regulate adaptive responses against coinfections.

Coinfection of a host by multiple pathogen species is highly ubiquitous (1–3). Although biomedical research has primarily focused on isolated interactions of a single host vs. a single pathogen, growing evidence from natural systems and epidemiological studies indicates the greater ecological importance of coinfecting pathogens in influencing the global health and disease burden (4, 5). For example, in humans, over one-sixth of the world’s population is affected by coinfections, composed of diverse pathogens underlying many globally important diseases such as HIV, tuberculosis, malaria, hepatitis, and leishmaniasis (6–8). In addition to engaging in complex within-host interspecific interactions (1), the evolutionary history of frequent exposure to these coinfecting pathogens can profoundly influence the maintenance and deployment of the host immune system (9, 10). However, despite such natural relevance, the extent to which coinfecting pathogens could influence the evolution of the host immune system differently from infections caused by single pathogens remains unexplored.

A generalized understanding of coinfection outcome is also challenging because of multiple confounding parameters such as the pathogen identity, multiplication rate of individual pathogens during coinfection (11), temporal changes in their relative frequency and damage to the host (12) that influence the dynamics and efficacy of immune activation. For instance, pathogens with divergent antigenic properties, which a single immune strategy cannot control, might lead the host to activate multiple immune components simultaneously during coinfection, increasing the energetic burden (13) and immunopathological risk (14, 15). Moreover, many naturally occurring coinfections can also involve pathogens that vary widely in their growth and virulence dynamics (16). In such cases, the host might evolve temporally separated immune strategies depending on how and at what rate different pathogens multiply inside the host and manifest their virulence (17). For instance, the immune system might experience a strong selection to rapidly eliminate the fast-growing pathogens that induce high mortality rates early in infection (i.e., fast-acting) (18, 19). Recent experiments and theoretical models can support this idea, where the ability to effectively clear pathogens by mounting appropriate immune responses early in the infection can serve as a critical determinant of postinfection survival success (20, 21). However, the evolution of host responses facilitating such early-life fitness advantages against coinfections can be constrained if the response time to fast-acting infections is limited, precluding the timely induction of appropriate immune components at adequate levels (22, 23). Host immune responses can face additional challenges by the co-occurrence of other pathogens that grow relatively slowly, induce a slower mortality rate, and persist longer (i.e., slow-acting), thereby warranting sustained immune responses (24, 25). Consequently, the efficacy of immune adaptation against coinfections can be critically contingent upon balancing the expression of specific immunomodulation against individual pathogens (25, 26). However, experiments accounting for the differences in growth and virulence dynamics between coinfecting pathogens while analyzing their impacts on immune system evolution are missing.

To start understanding these processes, we first built a theoretical model that describes diverse adaptive trajectories of host responses (i.e., changes in the net postinfection survival across generations) against coinfections based on their differences in within-host pathogen growth rate, rate of clearance, timing of immune activation, host mortality rate (i.e., virulence manifestation), and the level of interference due to competitive interactions and immune cross-reactivity between pathogens (27, 28) (SI Appendix, Fig. S1). Our model predicts that for pathogens that do not strongly interfere with each other’s growth or induce strong cross-reactive immunity, within-host growth dynamics of rapidly proliferating pathogens and their effects on host mortality rate at the early infection phase can drive the adaptive dynamics against coinfections. Subsequently, we tested and validated this prediction using experimentally evolving Tribolium castaneum beetles adapting against two coinfecting bacterial pathogens with contrasting growth and virulence dynamics for 30 successive generations (SI Appendix, Fig. S2). At every generation, beetles were infected with either A) fast-growing Gram-positive bacteria Bacillus thuringiensis (Bt), causing a rapid and sharp increase in host mortality followed by rapid clearance within a day (i.e., fast-acting) or B) slow-growing Gram-negative bacteria Pseudomonas entomophila (Pe) that killed the beetles at a slower rate, while causing persistent infection for several weeks (i.e., slow-acting); or C) a combination of both the pathogens (Mx) (see SI Appendix, Fig. S2 for study design). Note that the choice of P. entomophila and B. thuringiensis in beetles was made primarily because they satisfied the criteria for the type of pathogens that show contrasting within-host growth and virulence dynamics in beetles, thereby enabling us to test the predicted outcome. Subsequently, we used an RNA-sequencing approach to investigate the underlying changes in gene expression to gain comparative molecular insights into host adaptations against individual vs coinfecting pathogens.

We speculated two alternative possibilities: A) If variations in the beetle’s ability to clear fast-growing Bt primarily determine their survival probability early in the coinfection (21, 29), selection pressure might act more strongly to resist the infection prevalence of Bt than Pe. Consequently, the rate of adaptation against Mx might closely resemble the responses against only Bt infection. Moreover, there could also be a delay in their rate of adaptation if the scope of immune modulations is limited against the early mortality surges caused by rapid-acting Bt (30, 31); B) Alternatively, since Bt-induced early infection phase is closely followed by a persistent Pe infection phase that interferes with the beetle’s oviposition window, selection can instead be more potent against long-lasting Pe infections to ameliorate its fitness costs during reproduction (32). This could bias the overall adaptive dynamics more toward the responses against Pe.

Our results supported the model outputs such that fast-acting pathogens such as Bt imposing early infection costs indeed constrained the adaptation against Mx. Mechanistically, the observed patterns of adaptation against Mx could result from fewer immunomodulatory mechanisms available against its fast-growing Bt counterparts. Together, these are insights into how selection against individual pathogens can determine the trajectory of phenotypic variations vs mechanistic changes while evolving against coinfections.

Results

The Theoretical Model Predicts the Importance of Within-Host Pathogen Growth and Virulence Dynamics in Understanding the Host Adaptation Against Coinfections.

To understand the host adaptation against coinfecting pathogens with contrasting growth dynamics, we began by first simulating the density-dependent growth rate of individual pathogens (e.g., rapid vs slow) leading to the acute infection phase using a Baranyi model (33), as described by Duneau et al. (20). Here, we also considered varying initial inoculation sizes, the lag phase of pathogen growth, and carrying capacity across pathogens and infection types. In this model, a subset of individuals succumbed to infection due to their inability to control the pathogen growth below a threshold density, causing terminal infection. Following this, we also simulated the divergent clearance patterns of the pathogens from the surviving individuals, leading to either rapid clearance or long-lasting persistent infection (i.e., incomplete clearance) using an exponential decline model (20) (See SI Appendix, Table S1 for parameters). Overall, this enabled us to typify within-host growth dynamics patterns of pathogens that could constitute diverse facets of a two-pathogen coinfection system: e.g., rapidly growing pathogens causing acute infections, followed by rapid clearance (Rc) or persistent infection (Rp); Slow-growing pathogens causing acute infection, followed by rapid clearance (Sc) or persistent infection (Sp) (Fig. 1A). We expect that such classifications may encompass a relatively larger number of coinfection scenarios because regardless of the specific pathogen involved and the variability in host responses, many pathogens can still differ significantly in their within-host growth rates, clearance patterns, and the speed at which they induce virulence and lead to host mortality. Subsequently, we paired these pathogens based on their contrasting growth rates (i.e., rapid vs slow) to create the following coinfection combinations: e.g., Rc-Sc, Rc-Sp, Rp-Sc, and Rp-Sp. Note that we also considered a coefficient of interference α, describing the effects of direct competitive interactions (β) between coinfecting pathogens (27) and cross-reactive immune modulations (γ) affecting their overall virulence manifestation driving the postinfection mortality/survival of the host (28). The interference increases when either of the coexisting pathogens attains their peak growth (Fig. 1B). Moreover, in the case of pathogens that can be rapidly cleared by the host, the level of their interference can also decline rapidly, whereas pathogens causing persistent infections can retain their interference longer as they enter chronic infection phase (Fig. 1B). Nevertheless, individual growth dynamics, as well as the nature of interference between pathogens, jointly influence the host survival against coinfection. In the absence of strong interference (i.e., low β- and γ-values resulting in ε-value approaching 0), the early host mortality pattern due to coinfection closely resembles the early host mortality trend against the rapid-acting pathogen that grows and imposes mortality rapidly (Fig. 1C and SI Appendix, Table S2). By contrast, pathogens with very high mutual interference (i.e., high β- and γ-values resulting in ε-value approaching 1) can lead to less severe effects of their coinfection relative to their single infection counterparts.

Fig. 1.

A mathematical model of coinfection landscape and host evolution. (A) Combination of pathogens with divergent growth dynamics during coinfection: fast-growing pathogens causing acute infections, followed by i) rapid clearance (Rc) or ii) persistent infection (Rp); Slow-growing pathogens causing acute infection, followed by iii) rapid clearance (Sc) or iv) persistent infection (Sp), using the combination of Baranyi model and an exponential decline model, developed by Duneau et al. 2017 (n = 250; 10 data points for each of the 25 time points); (B) The shape of the coefficient of interference (α) was simulated for both rapidly cleared vs persistent pathogens, as well as for the various extent of β and γ; The coefficient of interference α between coinfecting pathogens is dependent on pathogen growth dynamics, and the parameter ε [combining the interference due to host immune modulations by the coinfecting counterpart (β) and direct resource-driven competition between the coinfecting pathogens (γ)]; (C) Probable effects of coinfection by different combinations of pathogens (as described in A) on host survival, based on survival patterns against individual pathogens by incorporating conditional probabilities of survival and variable degrees of interference (high interference: ε = 0.95 and low interference: ε = 0.05) between pathogen types (as described in B); (D) The host adaptive trajectories across various combinations of rapid- vs slow-growing pathogens only at low ε values. The trajectories were determined by the growth dynamics of rapidly proliferating pathogens and their total infection window.

Since low interference between pathogens increases the severity of coinfections estimated as postinfection mortality described above (Fig. 1C), we assumed this to be the most relevant condition that maximizes the selection on the host to reduce the infection costs. We thus only modeled the host adaptive trajectories across various combinations of rapid- vs slow-growing pathogens only at low ε-value (i.e., low β- and γ-values) (SI Appendix, Table S3). Overall, our model simulates how the rate of host adaptation (i.e., increase in the net postinfection host survival across generations) against coinfection is determined by the growth and virulence dynamics of rapidly proliferating pathogens and their total infection window (i.e., from the first pathogen exposure to the end of mortality due to infection), where host mortality happens, and fitness costs can be paid under pathogenic infections (Fig. 1D and SI Appendix, Table S3), overriding the effects of growth and virulence dynamics of slow-proliferating pathogens. For example, rapidly growing pathogens such as Rc, which can lead to early mortality surges within a short time, can significantly delay the host adaptation against Rc-Sp or Rc-Sc combinations. In contrast, host evolution (i.e., gain in the survival advantage against pathogens) was faster when the rapid growth phase was followed by persistent infection, causing mortality over a prolonged period (i.e., Rp-Sp and Rp-Sc).

Moreover, we have also examined the model sensitivity by varying all the relevant parameters (described in SI Appendix, Eq. 6) individually while predicting the host adaptive trajectory (SI Appendix, Fig. S3). We found that within biologically plausible ranges of parameter values, our model prediction simulates similar patterns of adaptive trajectories. Subsequently, we have validated the model based on our empirical results (described below) and estimated parameters for within-host growth dynamics, interference effect determinant, and host adaptive trajectory parameters (SI Appendix, Result and Fig. S4 and Table S4 and S5).

Experimental Data Confirm that Rapidly Growing Pathogen Determines the Coinfection Outcome During the Early Infection Phase.

We next performed a series of experiments to verify the above model predictions, underscoring the role of rapidly growing pathogens in driving the evolution against coinfections. We chose to verify the host adaptive trajectory against a pair of coinfecting pathogens, which the model already predicts to produce the most contrasting effects on host adaptation attributed to their divergent growth and virulence dynamics—i.e., Rc-like pathogen in combination with another slow-growing and slow-killing Sp-like pathogen (Fig. 1D). To this end, we used the model insect T. castaneum infected with a mix of suitable bacterial pathogens, B. thuringiensis (Bt), and P. entomophila (Pe), which we identified as possessing comparable growth and virulence dynamics as that of Rc- and Sp-like pathogens, respectively. Both pathogens eventually killed ~60 to 65% of the beetles within a week, with significant differences in their mortality rates only within the first 24 h of infection (Fig. 2A and SI Appendix, Table S6). While Bt-induced mortality showed a rapid surge within 8 h postinfection (hpi), with most susceptible individuals dying within the first 12 hpi, Pe-induced mortality showed a relatively late onset of around 20 to 24 hpi and continued for the next 7-d. However, beetles infected with a combination of Bt and Pe (Mx) showed a mixed mortality pattern reflecting the individual effects of both pathogens such that there was a sharp Bt-like decline in their survival within the first 16 to 18 hpi, which is then followed by a gradual Pe-like decline for the next 7 d (Fig. 2A). During this, Bt cells showed rapid growth between the first 6 to 8 hpi and then became undetectable by 20 h, whereas the Pe cells reached peak growth around 24 hpi and persisted in high numbers even after a week (Fig. 2 B and C and SI Appendix, Table S7) (Pe cells persisted even after 25 d postinfection in some beetles; SI Appendix, Fig. S5). Moreover, in beetles infected with Mx, early mortalities (within 20hpi) were primarily driven by Bt-induced pathogenicity as dead beetles carried a large abundance of Bt cells (~10⁶ cells/beetle; estimated immediately after death), whereas later mortalities (>20 hpi) were most possibly caused by an overgrowth of only Pe (~10⁷ cells/beetle) (Fig. 2D). These results thus corroborate the model predictions where the severity of coinfection and host mortality patterns during the early infection phase was indeed correlated with the effects of rapid-growing pathogen counterparts (Fig. 1C).

Fig. 2.

(A) Proportion of beetles from baseline population (henceforth, baseline beetles) surviving after infection with bacterial pathogens B. thuringiensis (Bt), P. entomophila (Pe), or a combination of both (Mx) (n = 30 females/treatment). The P-value (P_I) represents differences across different infection treatments (i.e., Bt, Pe, and Mx); Temporal changes in within-host growth dynamics of (B) Bt and (C) Pe load in baseline beetles, both in the context of infections caused by single- (i.e., Bt or Pe) vs coinfecting pathogens (i.e., Mx) (n = 10 replicates with pooled homogenate of 3 females/time points/infection treatment). P-values represent the effects of infection treatment (I) and the assay time (T). (D) The load of Bt vs Pe cells in baseline beetles that succumbed to infection, assayed till first 46 h after Mx infection. Bt cells were detected only in beetles that died within the first 19 h of infection (n = 9). At later time points (>19 to 45 h), beetles (n = 25) only carried Pe cells. (E) Beetle survival across generations (Generation 3 to 30) at the end of the oviposition window (i.e., 8th-day postinfection) in each replicate population of different pathogen-selection regimes (n = 4 replicate populations/selection regime). P-values represent the pairwise differences between selection regimes. Solid black triangles denote the generations where selection response was assayed by comparing the postinfection survival of control (C) vs pathogen-selected regimes (B, P, and M) against their respective pathogens (n = 24 to 60 females/regime/generation). The numbers in the parentheses represent the number of replicate populations from each selection regime that showed significantly improved survival during experimental evolution (also see SI Appendix, Fig. S6). All the assays involved 4 replicate populations, except generation 8, where only 3 replicate populations could be assayed.

Experimental Data Validate that Rapidly Growing Pathogen Drives the Host Adaptive Dynamics Against Coinfection.

We next allowed beetle populations to evolve under strong pathogen selection imposed by either Bt (B-regime) or Pe (P-regime) or a mix of both (M-regime), each with 4 replicate populations (i.e., B1–4; P1–4; M1–4) and tracked their postinfection survival for 30 generations. Beetle response to selection against Pe was the fastest such that within only eight generations, they could rapidly increase their postinfection survival from ~40 to ~75% and then to ~90% by 18 generations (Fig. 2E and SI Appendix, Table S8; also see SI Appendix, Fig. S6). In contrast, B and M beetles required a substantially extended selection period to improve survival. They initially showed large fluctuations in survival (~35 to 60%) for 16 generations and then could steadily increase only up to ~75% by the 24th generation. Control populations that were either pricked with sterile Ringer solution (C) (or maintained as unhandled populations) had a very high survival rate (>98%) throughout the experiment. In parallel, we also directly estimated the relative improvement in postinfection survival of each replicate population, relative to C beetles, at regular intervals between generations 8 to 28 to disentangle the adaptive dynamics across pathogens and infection types (Fig. 2E and SI Appendix, Fig. S7). While at least half of the replicate P-populations showed significantly improved survival within eight generations, followed by the other two populations by 15 generations, the first replicate population of M- and B-beetles could evolve the response only at generations 13 and 18, respectively. The remaining M- and B-populations evolved the response after the 18th and 22nd generations. Overall, while these results highlight the divergence in the rate of adaptation across pathogens (e.g., Bt vs Pe) and infection types (single vs multiple pathogens), they are also in conformity with the theoretical predictions, emphasizing the role of fast-growing Bt-like pathogens in restricting the adaptation against coinfections.

We also found a significant reduction in the bacterial load across pathogen-selected regimes relative to C-beetles, estimated around the onset of mortality after Pe and Bt infections (i.e., 24 and 8 hpi, respectively) (sampled at generation 26) and at two time points after Mx infection (8 hpi and 20 hpi) (sampled at generation 25) (SI Appendix, Figs. S8 and S9 and Table S9–S11). Our result showed that increased postinfection survival of evolved beetles could thus be associated with their improved ability to prevent bacterial growth relative to the unselected control beetles.

Host Populations Evolving against Coinfection Adopted Distinct Strategies to Counter the Severity of Infections Caused by Individual Pathogens.

Next, we also compared the bacterial load of every M- and C-beetle that succumbed to Mx infection and sampled a subset of survivors every 5–8 h for the next 50 hpi to explain their divergent mortality patterns as a function of temporal changes in the pathogen growth dynamics. Contrary to our expectation, live M-beetles did not carry fewer Bt cells than C-beetles, except during the early phase of infection before mortality was initiated (i.e., 6 to 8 h) (Fig. 3A and SI Appendix, Table S12, also see SI Appendix, Fig. S8F). However, they could significantly limit the number of beetle mortalities due to the growth of Bt cells beyond a threshold density at the early phase of infection (i.e., compare the number of dead beetles in M- and C-regime between 12 to 20 hpi; SI Appendix, Fig. S10 and Tables S13). Interestingly, most of the live M- and C-beetles showed complete removal of Bt cells within ~20 h of infection (Fig. 3A), suggesting no differences in their rate of pathogen clearance (SI Appendix, Fig. S11 and Tables S14). Increased efficacy in arresting the Bt growth below the threshold density causing terminal infection, rather than its clearance, thus explained the improved survival of M-beetles relative to C-beetles.

Fig. 3.

Within-host bacterial growth dynamics in control vs selected beetles. Temporal changes in (A) B. thuringiensis (Bt) (B) P. entomophila (Pe) load of live (n = 6 to 10 replicates with pooled homogenate of 3 females/selection regime/time point/bacteria) and dead beetles sampled at various time points after coinfection in M regime relative to C regime until 50 h postinfection (hpi); Temporal changes in (C) Bt and (D) Pe load sampled from live (n = 6 to 10 replicates with pooled homogenate of 3 females/selection regime/time point/bacteria) and dead beetles in B and P regime, relative to their C counterparts. Bt load from every dead B vs C beetle was recorded till 12 hpi, whereas live individuals were monitored till 18 hpi as Bt cells are usually cleared by beetles by this time. In contrast, Pe load from dead and live P vs C beetles was recorded only till 30 hpi (beetle mortality beyond this point was not tracked for bacterial load assay) and 72 hpi, respectively. In each case, bacterial load of dead beetles was extracted from individual beetles. The total number of beetles that died is indicated in parentheses. In each panel, P-values either represent the effects of the selection regime (SR) and assay time (T) on bacterial load derived from live individuals; or the main effect of the SR on bacterial load upon death.

We noted that the number of M-beetles that died due to Pe overgrowth after 16 h was also drastically reduced (Fig. 3B and SI Appendix, Table S12), but now, in contrast to the Bt-infection phase, surviving M beetles always carried a much lower density of Pe cells (Fig. 3B and SI Appendix, Table S12), indicating that evolved beetles cleared Pe more efficiently. This, in turn, enabled them to prevent the Pe load of surviving beetles from exceeding the threshold density, leading to lethally acute infection (Fig. 3B). Also, M- and C-beetles that succumbed to infection did not differ in their Bt or Pe burden, suggesting that the threshold pathogen density needed to cause mortality was comparable across regimes (Fig. 3 A and B and SI Appendix, Table S12). Overall, these results broadly corroborated the patterns of bacterial growth dynamics in B- and P-regimes as well (Fig. 3 C and D and SI Appendix, Table S12), suggesting that the outcome of Mx infection in the M-regime might be additively determined by both their initial success in controlling the Bt overgrowth as well as maintaining lower Pe burden in the later phase of infection.

Immune Gene Expression Profiles in Host Populations Adapted against the Coinfection Resembled More with those Evolving Against the Slow-Growing Pathogen.

To gain mechanistic insights into divergent responses evolving across pathogens and infection types, we next conducted RNAseq using beetles across selection regimes, collected around the onset of their mortality after respective infection treatments (i.e., 8, 16, and 24 h after infection with Bt, Mx, and Pe, respectively). This allowed us to compare the gene expression changes underlying nearly comparable fitness consequences across diverse beetle lines and infection types. Overall, the number of differentially expressed genes (DEGs) upon infection was considerably higher in M-beetles and P-beetles compared to B-beetles, both before (No. of genes: M = 427, P = 439, B = 165) and after (No. of genes: M = 374, P = 472, B = 171) the experimental evolution (SI Appendix, Fig. S12 A and B). Also, the evolved M-beetles showed a significantly higher number of overlapping DEGs with that of P-beetles (N = 119) than B-beetles (N = 29), which might indicate similar mechanisms using a shared set of candidate genes between M- and P-beetles (SI Appendix, Fig. S12C). We found 77 and 81 DEGs common across infection treatments in control and evolved populations, respectively. Those common set of genes possibly played pervasive roles across pathogens and infection types, including immune-related molecules such as peptidoglycan recognition proteins (PGRP SC2), gram-negative bacteria binding proteins, antimicrobial peptides (AMPs; Attacin 2, Coleoptericin, and Defensin 3) as well as key metabolic genes, namely, glucose dehydrogenase and fatty acyl CoA reductase. We also identified 65 DEGs upon infection with known immune functions across pathogens and selection regimes (SI Appendix, Table S15). However, to disentangle their roles, we divided them into five broad categories based on their immune-related functions (i.e. immune categories; see SI Appendix, Table S15 for gene lists) a) pathogen and immune receptors; b) immune regulators; c) inducible immune effectors, including AMPs and lysozymes; fast-acting constitutively expressed d) melanization response involving phenoloxidase pathway; and e) production of reactive oxygen species (ROS) (Fig. 4A), followed by a MANOVA to test effects of infection status, pathogen identity, and selection regimes on each of these immune categories (SI Appendix, Tables S16–S20). While the effects of selection regimes varied across immune categories (e.g., significant main effects only in receptors and inducible effectors), we found a consistent two-way interaction between pathogen identity and infection status across all immune categories, suggesting that the deployment of immune responses can be pathogen-specific.

Fig. 4.

RNA sequencing and molecular insights into the evolved responses. (A) Heatmaps denoting the differentially regulated genes with known immunological function in insects (described in SI Appendix, Table S14) after respective infection treatments (e.g., sham infection vs either Bt, Pe, or Mx) in control vs pathogen-selected regimes (B-, P-, or M-regimes), divided into five broad functional categories: a) pathogen and immune receptors; b) immune regulators; c) inducible immune effectors, including antimicrobial peptides (AMPs) and lysozymes; d) melanization response involving phenoloxidase pathway; and e) production of reactive oxygen species. (B) Cumulative gene expression profile of differentially expressed immune genes based on linear discriminant analysis (LDA). The first axis of LDA is considered as gene expression profile for various categories of differentially expressed immune-related genes. Here, we compared expression profile changes between beetles with sham infection and infection with respective pathogens (Bt, Pe, and Mx) in pathogen-selected (S) beetles (B-, P-, M-beetles) vs their respective unselected (C) beetle populations. In each panel, significantly different groups are connected with different alphabets. Alphabet assignments are not comparable across pathogens (separated by dotted lines). (C) Correlation of phenotypic profile (based on combined estimates of postinfection survival and bacterial load) with cumulative expression levels of diverse immune function categories described in panel A, using canonical correlation analysis. Correlations are shown separately for each immune gene category. Corresponding statistics, including canonical R-square (Can) and R-squares representing only significant pathogen-specific trends, are also shown for each immune gene category. The abbreviations used in the figure legend refer to the specific regime (Control denoted by hollow and pathogen selection denoted by solid) followed by the infection treatment (e.g., C-Bt = control beetle infected with Bt, B-Bt = B-beetle infected with Bt and so on).

To further explore these associations, we performed a canonical discriminant analysis to obtain a linear combination of the expression profile of immune-related genes, separating the effects of infection across pathogen types and selection regimes in each immune category. We corroborated the statistical differences in expression profile due to infection treatment as found in MANOVA (SI Appendix, Tables S16–S20), where effects were again found to be pathogen-specific. For example, infection resulted in similar patterns of gene expression changes in both P- and M-beetles across all immune categories (note the direction of changes relative to respective sham-infected treatments) (Fig. 4B). However, infected B-beetles displayed different responses in inducible immune effectors and receptors, where gene expression patterns went opposite to those observed in the infected P-beetles and M-beetles (Fig. 4B). Immune gene expression patterns thus seem to reemphasize the relatively greater functional overlaps in immune repertoires between M- and P-beetles relative to B-beetles.

However, subsequent analyses revealed that infection affected evolved beetles differently from their respective control populations, although the effects varied across pathogen identities and immune categories (Fig. 4B and SI Appendix, Tables S16–S20). For example, the level of divergence in the expression patterns of inducible effectors, receptors, and immune regulators in evolved P beetles either increased or decreased relative to C beetles after Pe infection (Fig. 4B and SI Appendix, Tables S16–S18). It also induced significant changes in phenoloxidase response-related genes that were initially nonresponsive in control beetles (Fig. 4B and SI Appendix, Table S19). In contrast, evolution against Bt produced changes limited to only fast-acting phenoloxidase response and ROS (Fig. 4B, SI Appendix, Tables S19–S20). Note that in the case of gene expression profile of receptors for P-beetles and phenoloxidase response for M-beetles, the extent of divergence across selection regimes after infection appeared unclear from multiple comparisons using Tukey’s HSD (Fig. 4B and SI Appendix, Table S16 and S19). We thus corroborated the divergence by estimating the effect size of infection, which was higher in selected regimes for both the immune categories (Cohen’s D test; Receptor: Control = 3.27, P-beetles = 5.14; Phenoloxidase response: Control = 1.89, M-beetle = 3.91).

Finally, we applied canonical correlation analyses, followed by linear regression analyses, to determine whether the observed changes in the gene expression profile of the aforementioned immune categories predicted the phenotypic variations between the control vs selected regimes across pathogens. In each case, we used a joint estimate of the bacterial load of individual beetle hosts and the infection susceptibility, estimated as the hazard ratios (34) of infected vs sham-infected beetles (where Hazard ratios > 1 denote higher mortality in the infected beetle), to gain an integrated view of postinfection fitness outcomes as a function of pathogen burden and concomitant survival costs. We assumed significant correlations between gene expression values and the phenotypic changes to imply whether the concerned category of immune molecules can explain the observed variation in phenotypic traits during experimental evolution. Overall, phenotypic variations of Pe-infected beetles correlated with a maximum number of immune categories, including receptors, immune regulators, and inducible immune effectors, followed by Bt-infected beetles that correlated with regulators and melanization responses, and Mx-infected beetles that correlated with only receptors (Fig. 4C and SI Appendix, Tables S21–S25). Similar patterns also emerged when we separately analyzed the associations of infection susceptibility with the gene expression profile using linear regression analyses. Pe-infected beetles still had more correlations (i.e., with both receptors and melanization response) than B and M beetles that correlated either with melanization response or receptors, respectively (SI Appendix, Fig. S13A and Table S26). In contrast, no such correlation existed with the bacterial load of Bt-infected beetles, as opposed to Pe- and Mx-infected beetles, where, in addition to exhibiting correlations to receptors and regulators respectively, they also showed association with melanization response (SI Appendix, Fig. S13B and Table S27). While these correlations suggest a larger scope of modulating immune responses at various functional levels in P-beetles to increase postinfection fitness, they also reemphasized the potential divergence of immune strategies adopted in B-beetles from that of P- and M-beetles to control the pathogen growth.

In addition, we also used KEGG enrichment analyses to reveal broad similarities in how several key metabolic pathways responded against Pe and Mx infection (SI Appendix, Fig. S14). For instance, unselected C-beetles infected with Mx and Pe showed downregulation of glutathione and several components of amino acid (e.g., valine, leucine, and isoleucine), and carbohydrate (e.g., amino sugar and nucleotide sugar) metabolism. They also showed downregulation of glycolysis and upregulation of phagosome maturation pathways, contrasting with Bt-infection. In the evolved M- and P-beetles, we noted a downregulation in both the citrate cycle and OXPHOS pathway (SI Appendix, Fig. S14). Also, while evolved B-beetles overexpressed mismatch and nucleotide excision repair pathways, both M- and P-beetles produced no changes in their expression. These results thus suggest the possibility of common metabolic bases underlying overlapping immune responses against Mx and Pe.

Discussion

Despite the ubiquity of coinfections and their direct relevance to many infectious diseases (2, 4), it is still unclear how they influence host adaptive trajectories against pathogens and concomitant immune system evolution. Also, what are the specific drivers of such evolutionary effects of coinfections? Here, we first used mathematical models to propose within-host growth rate, the rate of virulence manifestation, and the infection-driven mortality window of individual pathogens (i.e., between first and last host mortality) as critical determinants of adaptive success against coinfections. In the absence of strong competitive interference or cross-reactive immunity between pathogens (27, 28), rapidly growing pathogen counterparts, imposing acute mortality surges early in the infection, determined the course of evolution against coinfections. Moreover, if such an early surge of survival costs is also expressed rapidly within a short infection window, the rate of adaptation against coinfections can be delayed. Our model shows this as a possible scenario when appropriate immune responses are unavailable or cannot be induced against such fast-acting pathogens within the available infection window to curb their acute early-infection costs (31, 35).

Subsequently, we validated the model outputs using replicated populations of T. castaneum evolving against bacterial pathogens of distinct Gram-types (i.e., Bt and Pe) with contrasting within-host growth dynamics, virulence manifestation rates, and differential host immune modulations (36, 37). Bt grew faster early in the infection, inducing early and rapid mortality surge within 12 h (i.e., fast-acting), followed by rapid clearance by the host. By contrast, Pe grew relatively slowly, causing long-lasting persistent infections with mortality beginning around 24 h postinfection (i.e., slow acting). We found the rate of adaptation to be fastest against Pe, with half of the replicate populations evolving resistance as early as generation 8, whereas resistance evolution against fast-growing Bt was delayed the most. Also, as predicted by the model, the rate of adaptation against coinfection by Mx indeed appeared to be constrained by fast-acting Bt such that M-beetles followed almost a similar evolutionary trajectory as that of beetles infected with Bt alone (where most replicate populations took 15 to 22 generations to evolve resistance). Another striking aspect is that while the survival success of the P-beetles rose to ~90%, the survival of both M- and B-beetles could not increase beyond ~75% despite a continuous strong selection for 30 generations. This indicated the constraints associated with evolving resistant alleles against Bt cells present in both M- and B-beetles during their early infection phase, restricting their net fitness gain to much below that of their P-beetle counterparts (38, 39). However, it should be noted that we used a septic injury method to infect beetles by pricking, which may mimic the effects of opportunistic infections through wounds that can occur in nature. Nevertheless, a more natural method of infection via the oral route might have altered the rate of host adaptations due to changes in the interactions among pathogens, host immune responses, and the pattern of virulence manifestation. Yet, we opted against the oral mode of infection as it could also lead to correlated evolution of divergent feeding rates as part of pathogen avoidance strategies across treatments, thereby confounding the effects of pathogen selection on the immune system across beetle lines.

Based on our observations, an emerging question is—how might Bt cells drive the dynamics of adaptive evolution against Mx? We noted that beetles infected with Mx showed a sharp decline in survival early in infection (between 16 to 20 h), which broadly resembles the mortality pattern of beetles that were only infected with Bt. Also, beetles that succumbed to infection within this early timeframe predominantly carried many Bt cells (~10⁵–10⁶ cells/female), linking the overgrowth of Bt to lethal infections. Interestingly, the estimated levels of the bacterial load causing such terminal infection did not correlate with the time postinfection at which death occurs. Hence, they also denoted the maximal Bt load which beetles could tolerate before they died (20). Several dead beetles also carried Pe cells, but neither their frequency nor their within-host Pe density was sufficient to explain all the beetle mortality observed during the early infection phase, hinting at the limited role of Pe in driving the early survival costs of coinfection. In contrast to dead beetles, surviving beetles early in infection either had a much lower Bt burden than their dead counterparts or cleared the infection below the detection level within 20 h. The ability to restrict the growth of Bt below their threshold density, which otherwise could lead to terminal infections, followed by rapid clearance, was thus critical for these beetles to survive the early phase of coinfection. These results also conform to recent studies with D. melanogaster, where similar binary infection outcomes have been reported across pathogens (20, 21, 26), underscoring the pivotal roles of rapidly induced immunity in effectively curtailing the pathogen overgrowth early in infection.

As expected, the ability to prevent Bt overgrowth early in coinfection also increased in beetles adapted against Mx infections. When challenged with Mx-infection, fewer individuals from evolved M-populations carried the lethally high Bt density (~10⁵–10⁶ cells/beetle) relative to C-beetles, thereby explaining the reduction in their early-infection mortality. However, increased survival of evolved beetles was not achieved by merely clearing the Bt cells, as their number, by and large, did not vary considerably between the live M- vs C-beetles. Subsequently, C-beetles that survived the infection could clear the Bt cells at a nearly equal rate to M-beetles. This suggests that pathogen selection on the beetles did not improve their pathogen clearance ability. Instead, the selection favored mechanisms in hosts to arrest the Bt growth below the critical density, otherwise leading to lethal infections (20). This is likely also the reason why transcriptome analyses of beetles challenged with only Bt infection had several differentially expressed immune effectors upon infection (e.g., AMPs Attacin 2, Coleoptericin, Defensin 3, and Tenecin 1), corroborating previous studies with the same infection model (40–42) and still, none of them responded differently in evolved B beetles, suggesting no added contribution toward experimental evolution against Bt. Also, the overall changes in the gene expression profile of different immune effector groups, including AMPs, PO, or ROS, did not explain the variation in the overall bacterial load across beetle lines. Perhaps more relevant changes in Bt-resistant beetles were detected in terms of their higher basal expression levels (i.e., without infection) of apolipophorins, facilitating phagocytosis and pathogen pattern recognition (43) or chymotrypsin, which is known to arrest the growth of Gram-positive bacteria (44), including neutralization of Bt-toxins (45). Indeed, similar roles of cellular immunity in curbing Bt infection were also implicated in another beetle study by Behren and coworkers (40). Increased circulation of these molecules, even in the nonimmune challenged state of B beetles, can thus play a more critical role in the early detection and prevention of Bt overgrowth.

By contrast, immune strategies against Bt in M-beetles may be more complex due to confounding effects of immune responses against chronic Pe infections persisting throughout the oviposition window of these experimentally evolving beetles (i.e., 3 to 8 d postinfection). Moreover, unlike Bt infection, surviving M-beetles consistently had reduced Pe load relative to C-beetles, suggesting the potential immune activation against Pe to minimize the infection costs while reproducing (46). Finally, despite receiving a lower infection dose (M- vs P-beetles: ~10³ vs 10⁴ cells/female), Pe cells in M beetles grew at an equivalent level as that of P beetles (~10⁵ cells/female within 12 h), which indicates that both beetle populations eventually experienced similar selection pressure from the severity of Pe infection. This is further corroborated by comparing the reproductive costs of each infection type in the unselected beetles (SI Appendix, Fig. S15 and Table S28). In the case of both Pe and Mx infections, the persistence of Pe cells during the oviposition window was also associated with a reduction in reproductive outputs. However, this contrasts with beetles challenged with Bt infection. Bt-infected beetles that survived until the oviposition window reproduced as much as their uninfected control counterparts, attributed to their ability to clear the infection completely by then. Based on these observations, we thus speculated strong selection pressure on both M- and P-beetles from the beginning of their selection treatment to evolve counterstrategies to reduce the reproductive costs imposed by a common pathogen that persists longer inside the host (47). More specifically, to this end, M-beetles could evolve more similarities with P-beetles vis-à-vis their immune responses rather than temporally compartmentalizing immunity against individual participating pathogens (48). Our transcriptome analyses that revealed larger overlaps in the set of genes and their expression profile against Pe and Mx infection, both before and after experimental evolution, supported this idea.

The possibility of mechanistic congruence between M- and P-beetles is further highlighted by the linear discriminant analyses of immunity-related gene expression data (Fig. 4B) (49). Many of them, classified into various functional categories ranging from sensing the pathogen or pathogen-associated molecular patterns (receptors) and regulating the immune responses to immune effectors such as AMPs, lysozyme, phenoloxidase cascade, and ROS production (50, 51), showed more concurrent gene expression patterns between M- and P-beetles. These patterns, however, did not always match with that of B beetles, as some of these functional categories, such as AMPs and lysozymes, immune regulators, and receptors, showed changes either in the opposite direction to that of M- or P-beetles or produced no changes after infection. Also, unlike in M- and P-beetles, none of the immune groups correlated with the changes in the overall bacterial burden before and after the evolution against Bt. Together, all these patterns thus hint at distinct functional implications of these immune groups in B-beetles relative to both M- or P-beetles. However, it remains possible that the observed patterns in gene expression are partly due to different time points chosen for sample collection for mRNA sequencing. Further investigation on the temporal dynamics of gene expression across selection regimes is thus needed to confirm these underlying mechanistic bases.

The similarity in immune responses against Pe and Mx infection was also reflected by their resemblance in metabolic changes. For example, KEGG enrichment analyses revealed the downregulation of several important components of carbohydrate (e.g., glycolysis, amino sugar and nucleotide sugar metabolism) and amino acid (e.g., valine, leucine, and isoleucine) metabolism in unselected C beetles (42). Besides, evolved P- and M-beetles showed reduced OXPHOS metabolism and increased glycolytic enzyme hexokinase 2 expression, suggesting shifting energy metabolism to support immune activation in these beetles (52). However, such metabolic patterns were reversed in B beetles, which corroborates why we failed to detect increased expression of immune effectors after experimental evolution. Instead, the enrichment of pathways related to increased DNA repair (53) and phagosome maturation (54) indicated strategies to reduce the DNA damage caused by immune activation (55) and the use of alternative immune strategies in B beetles [e.g., cellular immunity (36)] respectively.

Finally, a detailed comparison of phenotype-by-immune gene expression correlations across pathogens and infection types offered critical molecular insights into their divergent adaptive dynamics. For example, strong correlations between the combined phenotypic changes (i.e., postinfection survival and bacterial load) in P-beetles and diverse categories of immune-related molecules such as receptors [e.g., PGRP SC1a/b-like, PGRP2 (56)], regulators [e.g., Relish (57)] and inducible effectors [including Attacin 1, Attacin 2, Tenecin 1, and Ctenidin 1 (58)] suggested a wider scope for selection, acting parallelly and more effectively across various functional levels of their immune signaling cascade (59, 60). This, in turn, can accelerate their rate of adaptation. This notion can also be supported by previous analyses where immune molecules, particularly those involved in pathogen recognition and immune regulation, have been shown to evolve more rapidly under strong positive selection than other nonimmune genes (61). Moreover, the multilevel immune crosstalk between receptors, regulators, and effectors driving phenotypic variations against Pe corroborates the assumptions of our theoretical model. For instance, faster adaptation against slow-acting Pe-like pathogens was possible because slower mortality costs expressed over a prolonged infection window enabled beetles to employ functionally more diverse phenotype-by-immunological modulations under pathogen selection (25).

In contrast, the scope for such phenotype-by-immunological modulations in M- and B-beetles was limited. For example, unlike P-beetles, their immuno-competence phenotype correlated with either receptors or regulators but not with both, which might reduce the number of potential loci available to evolve rapidly under selection. Scopes for selection can be even more constricted in B beetles, as their phenotypic variations also correlated with PO response (62), which, in addition to serving as a critical insect immune defense component, exerts multiple pleiotropic roles in insect physiology (63, 64). An overall reduction in PO enzymatic activity in evolved B-beetles (SI Appendix, Fig. S16 and Table S29), mediated via lower expression levels of phenoloxidase 2 and tyrosine decarboxylase transcripts (63, 65), might impose significant development and reproductive costs (63, 64). Besides, evolved B beetles also showed more divergent expression profiles of ROS-related genes after Bt infection than the unselected beetles, driven primarily by downregulation of Glutathione S-transferase 1 after Bt infection, which could incur higher cytotoxicity (66), thereby adding significant costs to faster adaptation against Bt.

In summary, this study introduces a unique integrated framework, combining theory and experiments, to identify drivers of host adaptive dynamics against coinfecting pathogens. However, note that, in this work, we could experimentally test the effects of two coinfecting bacterial entomopathogens using an insect model only as a specific case study. Numerous other possibilities could exist across host–pathogen systems and their varied infection outcomes, where hosts may encounter pathogens with similar growth and infection dynamics, synergistic effects, or coinfection involving diverse pathogen types (e.g., virus, fungi), which remain unexplored. Nonetheless, the coherence between theoretical predictions and our empirical datasets, while establishing the importance of pathogen growth dynamics and virulence manifestation patterns in driving phenotypic and mechanistic trajectories during coinfection, substantiated the broader implications of our findings. Also, unlike our experimental observations based on the specific example of beetles coinfected with Bt and Pe where pathogen growth dynamics largely explained the evolved survival advantages, the theoretical framework used in this study relies on the phenomenological representation of the overall effect of infection in terms of host survival only and is agnostic to host strategies to counter the within-host pathogen growth. We did not specify in the model assumption whether the predicted survival advantages after evolution are derived from restricting the pathogen load through better pathogen clearance mechanisms or tolerating the pathogen burden without a pronounced infection cost (67). This potentially enabled our theoretical model to include a relatively wider range of host–pathogen interactions independent of the variations in host strategies against pathogen growth. Another striking outcome of our work is the decoupling of the overall rate of phenotypic evolution vs mechanistic bases against coinfecting pathogens relative to the effects of individual pathogens. This eventually highlighted the asymmetry in why and how individual pathogens might unequally bias the adaptive dynamics against coinfection vs underlying genetic mechanisms rather than their simple additive effects (68), offering exciting avenues for future theoretical models to encompass other infection types and more mechanistic explorations. Finally, our systematic investigation of host adaptations against multiple pathogens and infection contexts in a rare comparative framework may instigate more fundamental work to fill the gaps in our understanding of how innate immune features might evolve across pathogens and infection types.

Materials and Methods

Mathematical Simulation of Host Survival and Adaptation Against Coinfecting Pathogens With Contrasting Growth and Virulence Dynamics.

To formulate a mathematical framework to understand host adaptation against coinfection, we implemented the following steps (Detailed explanations and relevant equations describing each step are provided in SI Appendix, Supplementary Methods; also see SI Appendix, Fig. S1 for visual representation)—Step 1: We began by simulating within-host growth dynamics of individual pathogen species in ancestral host populations, followed by pairing up two pathogens with contrasting growth (e.g., slow- vs fast-growing) and clearance (e.g., complete clearance vs persistent infection) dynamics in various coinfection scenarios; Step 2: We then simulated the degree of interference between coinfecting pathogens by estimating a coefficient of interference based on the individual growth dynamics of each participating pathogen; Step 3: Next, we simulated the host survival patterns against infections caused by each pathogen separately by matching the temporal phases of their respective within-host pathogen growth dynamics. Note that the survival patterns are not simulated as a function of pathogen growth dynamics described in step 1 (See SI for explanations). Subsequently, we simulated survival against coinfection based on the simulated host survival pattern against individual pathogens constituting the coinfection and their interference coefficient; Step 4: Finally, we predicted the host adaptive trajectory based on the parameters derived from host survival patterns against single vs coinfecting pathogens. Note that here we considered the net postinfection survival of the host as the evolvable fitness trait across generations. Simulation results based on this framework are shown in Fig. 1, and all the simulation parameters are enlisted in SI Appendix, Table S1–S3. We simulated all the parameters, visualized the sensitivity of the parameters and validated on our empirical results using R.

Experimental Quantification of Virulence and Growth Dynamics of Coinfecting Pathogens.

We used a large, outbred population of Tribolium castaneum adapted to laboratory conditions for >2 years before commencing the experiments (see SI Appendix for more details on baseline population maintenance and assays described below). Also, based on the observations from other experiments (32, 69), we chose two bacterial entomopathogens that are likely to show contrasting growth dynamics and rates of virulence manifestations within insect hosts: fast-growing B. thuringiensis DSM2046 (Bt) vs slow-growing P. entomophila L48 (Pe), causing a rapid vs slower onset of mortality respectively. To quantify their effects on beetle hosts, we followed a septic injury method (32), wherein we pricked individual 10-d-old virgin females between their head and thorax from baseline populations with a needle dipped in a bacterial slurry composed of either Bt (~ 8 × 10⁷ cells/µl) or Pe (~ 4 × 10⁹ cells/µl), or a mix (1:1) of both bacterial cells (Mx) and monitored their survival for 8 d (See SI Appendix for input infection dose) (Dataset S1). We used sham-infected beetles pricked with sterile insect Ringer solution as procedural control for our infection assays (n = 30 females/infection treatment. Note that we did not use oral infection, where individual differences in feeding rates could cause large variations in the pathogen dose received by the beetle.

We also tracked the changes in growth dynamics of these pathogens inside surviving beetles sampled at regular intervals for the next 24 h for Bt (or ~3 d for Pe), both in the context of infections caused as individual vs co-occurring pathogens (n = 8 to 10 replicates with pooled homogenate of 3 females/infection treatment/time point), using established protocols in the lab (Dataset S2). We differentiated the Bt and Pe cells by their distinct colony sizes and morphologies on the Luria agar plates (SI Appendix). For each pathogen, we analyzed the postinfection survival data using Cox proportional hazard analysis, using infection treatment as a fixed effect in “survival” package in R. In addition to analyzing the survival data until 8 d postinfection, we also analyzed the beetle survival for the first 24 h after infection to highlight the differences in mortality patterns only during the early infection phase. We log-transformed bacterial load dynamics data using a generalized linear model (‘glm’ function in R) fitted to a gamma distribution with infection treatment and time of bacterial load estimation as fixed effects. Separately, we also assayed the bacterial load of a subset of females that succumbed to Mx-infection within the first 48 h to estimate the bacterial load upon death and the relative contribution of Bt vs Pe burden in causing mortality during coinfection (n = 34 females).

Experimental Evolution Paradigm.

Next, we used the experimental evolution paradigm for 30 successive generations to examine the adaptive dynamics against coinfecting pathogens (32). We used the baseline beetle population to create five selection regimes: namely I) Unhandled regime (U-regime): populations that did not undergo any treatment; II) Control regime (C-regime): Unselected control populations sham-infected with sterile Ringer; III) Infected with Bt (B-regime); IV) Infected with Pe (P-regime); V) Infected with a mixed culture of both Bt and Pe (M-regime), with each of these regimes having four independently evolving replicate populations (i.e., C1–4, B1–4, P1–4 & M1–4; See SI for infection doses). We infected (or sham-infected) 9 to 10 d old virgin adult male and female beetles between the head and the thorax following the septic injury method. Three days later, we combined the surviving beetles into 75 pairs and allowed them to oviposit for another 5 d (i.e., beetle reproductive window; day 3 to 8 postinfection). Note that although we maintained ~75 breeding pairs for each selection regime, we infected an excess of virgin beetles (~300 to 400 beetles/replicate population) for every generation, as we expected high mortality after the respective infection treatments. During experimental evolution, this ensured we had sufficient individuals to set up the 75 mating pairs to oviposit every generation. Also, we adjusted our infection doses to induce ~60 to 65% mortality within 8 d postinfection across infection treatments, which enabled us to initiate beetle lines with comparable selection pressure across pathogen-selected regimes until their reproductive window (as described above) and eventually disentangle the relative impacts of single vs coinfecting pathogens on host adaptative trajectories. After three weeks of egg incubation, we isolated male and female pupae from each population and allowed the eclosed adults to initiate the next generation after the relevant selection treatment. We handled the four replicate populations from each selection regime on different days (but C_i, B_i, P_i, and M_i, where i = 1 to 4, were handled together on the same day) and maintained continuous divergent pathogen selection. For each of the 4 replicate populations across selection regimes, we also estimated the proportion of the surviving adults (out of the total number of infected beetles) pre- (i.e., day 3 postinfection) and postreproductive window (i.e., day 8 postinfection) at every generation (except the first two generations) to track the overall changes in survival postinfection across selection regimes (Dataset S1). Moreover, to understand the contrast between diverging evolutionary trends across generations from different selection regimes, we used a generalized linear model fitted to Gaussian distribution followed by performing pairwise comparisons using “emmeans”.

Quantifying the Evolved Responses against Coinfection.

To assay the evolved responses, we performed a series of experiments spread across different generations along the experimental evolution timeline. For all our experiments, we collected virgin adults as pupae after one generation of relaxation of pathogen selection from the generation of interest to minimize the nongenetic or epigenetic effects (i.e., standardized beetles) (32). For this purpose, we collected a separate set of male and female pupae from respective selection regimes, allowed them to attain sexual maturity and put them in wheat for mating and egg-laying without infecting them.

To understand the dynamics of adaptation against single pathogens vs coinfections, we repeatedly assayed postinfection survival of each replicate population across selection regimes against their respective infection treatments during experimental evolution and compared with that of control unselected beetles (i.e., C vs P; C vs B or C vs M beetles after Pe, Bt, and Mx infection respectively), at multiple generations (e.g., generations 8, 13, 15, 18, 20, 22, and 28) (Dataset S1). We used 9 to 10 d old standardized females (for logistical reasons, we could not test males, except generation 28 when both sexes were assayed) (n = 24 to 60 beetles/regime/population/generation). For each pathogen-selected regime, we compared them separately with the C-regime after the respective infection treatments, using the mixed-effects Cox model (with the selection regime as a fixed effect and replicate populations as a random effect), followed by analyzing each replicate population separately, using Cox proportional hazard analyses (with the selection regime as a fixed effect). Besides, we also quantified the bacterial load of evolved beetles when all the replicate populations showed improved postinfection survival (P- and M-beetles at generation 18; B-beetles at generation 22) to test whether their improved survival can be explained by lower bacterial burden relative to their control counterparts (N = 10 to 15 replicates with pooled homogenate of 3 females/selection regime) (Dataset S2). Since different pathogens might manifest their virulence at different rates with divergent within-host pathogen growth dynamics, we sampled 9- to 10-d-old females from B and P regimes (and their corresponding Control regimes) after the onset of the first 10–15% mortality (information derived from postinfection survival curves) after respective infection treatments. For the M regime (and its corresponding control), we sampled females at two time points to obtain an adequate number of both Bt and Pe cells (See SI Appendix for detailed methods and analyses).

However, to explain the mortality patterns in more detail, we next characterized the dynamics of within-host bacterial load in one of the replicate populations from both evolved and control beetles by assaying 9- to 10-d-old standardized females (collected at generation 25 for Mx infection and 26 for Bt and Pe infection) every few hours (See SI Appendix for detailed methods) (Dataset S2). We tracked Bt cells in B vs C regimes until 18 h (or Pe cells in P vs C regime until 72 h), whereas, for Mx infection, both the bacterial cells were assayed until 50 h (n = 6 to 10 replicates with pooled homogenate of 3 females/selection regime/time point/infection treatment). Simultaneously, we also noted the beetle death during these experiments and estimated their bacterial load as soon as they succumbed to respective infection treatments across selection regimes to understand the link between growth dynamics, virulence manifestation, and the maximum pathogen load that beetles could tolerate before death (n = 13 to 80 replicates/selection regime/infection treatment). We analyzed the bacterial load data using a generalized linear model fitted to a gamma distribution, with selection regime and time of assay as fixed effects for live beetles and only selection regime as fixed effects for dead beetles.

Transcriptome Analyses.

To identify mechanistic insights for the evolved immune responses, we sampled sham-infected vs infected females as virgins across the control vs respective pathogen-selected regimes (i.e., B-. P- and M-beetles (n = 4 replicates; each comprised 10 females pooled together from each replicate population/infection treatment/selection regime) (See SI Appendix for sampling details). The sequencing was performed on an Illumina Novaseq 6000 platform using a 150 bp paired-end chemistry. RNA-seq data were analyzed using a custom pipeline (70), as described in SI method. We performed differential gene expression analyses and generated heatmaps using the “DESeq2” package in R. The annotation of reads was performed based on reference genome Tcas 5.2 (71). The sequencing results have been deposited in the National Center for Biotechnology Information’s BioProject (BioProject Id PRJNA1110986) (72).

Next, we used these normalized expression values to perform principal component analysis to identify inherent differences in gene expression profiles across different selection regimes and infection treatments. Subsequently, we performed a functional enrichment analysis to investigate the changes in the biological pathways associated with evolved immune response. We further estimated a metagene expression profile using canonical discriminant analyses for each treatment (49). Further, to understand how these gene expression profiles correlated with phenotypic variations (either individually with the hazard ratios as a proxy of survival response, or bacterial load, or as a combined estimate of both hazard ratio and bacterial load for each infection treatment and selection regime), we performed linear regressions and canonical correlation analysis (49). Detailed description of analyses is provided in SI Appendix.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Dataset S02 (XLSX)

Dataset S03 (XLSX)

Dataset S04 (XLSX)

Dataset S05 (XLSX)

Dataset S06 (XLSX)

Dataset S07 (XLSX)

Dataset S08 (XLSX)

Dataset S09 (XLSX)

Data, Materials, and Software Availability

The raw data files obtained from mRNA sequencing are deposited and available in the NCBI database under the BioProject (Project Id PRJNA1110986) (72). All other data are included in the manuscript and/or supporting information.