Decision letter: Host-pathogen genetic interactions underlie tuberculosis susceptibility in genetically diverse mice

Article Figures and data Abstract Editor's evaluation Introduction Results Discussion Materials and methods Appendix 1 Data availability References Decision letter Author response Article and author information Metrics Abstract The outcome of an encounter with Mycobacterium tuberculosis (Mtb) depends on the pathogen's ability to adapt to the variable immune pressures exerted by the host. Understanding this interplay has proven difficult, largely because experimentally tractable animal models do not recapitulate the heterogeneity of tuberculosis disease. We leveraged the genetically diverse Collaborative Cross (CC) mouse panel in conjunction with a library of Mtb mutants to create a resource for associating bacterial genetic requirements with host genetics and immunity. We report that CC strains vary dramatically in their susceptibility to infection and produce qualitatively distinct immune states. Global analysis of Mtb transposon mutant fitness (TnSeq) across the CC panel revealed that many virulence pathways are only required in specific host microenvironments, identifying a large fraction of the pathogen's genome that has been maintained to ensure fitness in a diverse population. Both immunological and bacterial traits can be associated with genetic variants distributed across the mouse genome, making the CC a unique population for identifying specific host-pathogen genetic interactions that influence pathogenesis. Editor's evaluation This work takes advantage of the genetic diversity of a panel of mice, termed the collaborative cross, to identify those host factors that contribute to heterogeneous outcomes after tuberculosis infection. The authors infect this panel of mouse strains with pools of Mycobacterium tuberculosis transposon mutants, allowing identification of specific host genotypes that confer fitness effects on certain bacterial mutants. The resulting analyses identify loci that affect quantitative immunological phenotypes or fitness of select bacterial mutants. The study is likely to be an important resource to microbiologists in general and those individuals studying the host immune response to tuberculosis infection https://doi.org/10.7554/eLife.74419.sa0 Decision letter eLife's review process Introduction Infection with Mycobacterium tuberculosis (Mtb) produces heterogeneous outcomes that are influenced by genetic and phenotypic variation in both the host and the pathogen. Classic human genetic studies show that host variation influences immunity to tuberculosis (TB) (Abel et al., 2018; Comstock, 1978). Likewise, the co-evolution of Mtb with different populations across the globe has produced genetically distinct lineages that demonstrate variable virulence traits (Gagneux et al., 2006; Hershberg et al., 2008; Wirth et al., 2008). The role of genetic variation on each side of this interaction is established, yet the intimate evolutionary history of both genomes suggests that interactions between host and pathogen variants may represent an additional determinant of outcome (McHenry et al., 2020). Evidence for genetic interactions between host and pathogen genomes have been identified in several infections (Ansari et al., 2017; Berthenet et al., 2018), including TB (Caws et al., 2008; Holt et al., 2018; Thuong et al., 2016). However, the combinatorial complexity involved in identifying these relationships in natural populations have left the mechanisms largely unclear. Mouse models have proven to be a powerful tool to understand mechanisms of susceptibility to TB. Host requirements for protective immunity were discovered by engineering mutations in the genome of standard laboratory strains of mice, such as C57BL/6 (B6), revealing a critical role of Th1 immunity. Mice lacking factors necessary for the production of Th1 cells or the protective cytokine interferon gamma (IFNγ) are profoundly susceptible to Mtb infection (Caruso et al., 1999; Cooper et al., 1993; Cooper et al., 1997; Flynn et al., 1993; Saunders et al., 2002). Defects in this same immune axis cause the human syndrome Mendelian Susceptibility to Mycobacterial Disease (MSMD) (Altare et al., 1998; Bogunovic et al., 2012; Bustamante et al., 2014; Filipe-Santos et al., 2006), demonstrating the value of knockout (KO) mice to characterize genetic variants of large effect. Similarly, the standard mouse model has been used to define Mtb genes that are specifically required for optimal bacterial fitness during infection (Bellerose et al., 2020; Sassetti and Rubin, 2003; Zhang et al., 2013). Despite the utility of standard mouse models, it has become increasingly clear that the immune response to Mtb in genetically diverse populations is more heterogeneous than any single small animal model (Smith and Sassetti, 2018). For example, while IFNγ-producing T cells are critical for protective immunity in standard inbred lines of mice, a significant fraction of humans exposed to Mtb control the infection without producing a durable IFNγ response (Lu et al., 2019). Similarly, IL-17 producing T cells have been implicated in both protective responses and inflammatory tissue damage in TB, but IL-17 has little effect on disease progression in B6 mice, except in the context of vaccination or infection with particularly virulent Mtb (Gopal et al., 2012; Khader et al., 2007). The immunological homogeneity of standard mouse models may also explain why only a small minority of the >4000 genes that have been retained in the genome of Mtb during its natural history promote fitness in the mouse (Bellerose et al., 2020). Thus, homogenous mouse models of TB fail to capture the distinct disease states, mechanisms of protective immunity, and selective pressures on the bacterium that are observed in natural populations. The Collaborative Cross (CC) and Diversity Outbred (DO) mouse populations are new mammalian resources that more accurately represent the genetic and phenotypic heterogeneity observed in outbred populations (Churchill et al., 2004; Churchill et al., 2012). These mouse panels are both derived from the same eight diverse founder strains but have distinct population structures (Saul et al., 2019). DO mice are maintained as an outbred population and each animal represents a unique and largely heterozygous genome (Keller et al., 2018; Svenson et al., 2012). In contrast, each inbred CC strain's genome is almost entirely homozygous, producing a genetically stable and reproducible population in which the phenotypic effect of recessive mutations is maximized (Shorter et al., 2019; Srivastava et al., 2017). Together, these resources have been leveraged to identify host loci underlying the immune response to infectious diseases (Noll et al., 2019). In the context of TB, DO mice have been used as individual, unique hosts to identify correlates of disease, which resemble those observed in non-human primates and humans (Ahmed et al., 2020; Gopal et al., 2013; Koyuncu et al., 2021; Niazi et al., 2015). Small panels of the reproducible CC strains have been leveraged to identify host background as a determinant of the protective efficacy of BCG vaccination (Smith et al., 2016) and a specific variant underlying protective immunity to tuberculosis (Smith et al., 2019). While these studies demonstrate the tractability of the DO and CC populations to model the influence of host diversity on infection, dissecting host-pathogen interactions requires the integration of pathogen genetic diversity. We combined the natural but reproducible host variation of the CC panel with a comprehensive library of Mtb transposon mutants to determine whether the CC population could be used to characterize the interactions between host and pathogen. Using over 60 diverse mouse strains, we report that the CC panel encompasses a broad spectrum of TB susceptibility and immune phenotypes. By leveraging high-resolution bacterial phenotyping known as 'Transposon Sequencing' (TnSeq), we quantified the relative fitness of a saturated library of Mtb mutants across the CC panel and specific immunological mouse knockout strains. We report that approximately three times more bacterial genes contribute to fitness in the diverse panel than in any single mouse strain, defining a large fraction of the bacterial genome that is dedicated to adapting to distinct immune states. Association of both host immunological phenotypes and bacterial fitness traits with Quantitative Trait Loci (QTL) demonstrated the presence of discrete Host-Interacting-with Pathogen QTL (HipQTL) that represent inter-species genetic interactions that influence the pathogenesis of this infection. Together, these observations support the CC population as a tractable model of host diversity that greatly expands the spectrum of immunological and pathological states that can be modeled in the mouse. Results The spectrum of TB disease traits in the CC exceeds that observed in standard inbred mice To characterize the diversity of disease states that are possible in a genetically diverse mouse population, we infected a panel of 52 CC lines and the eight founder strains with Mtb. To enable bacterial transposon sequencing (TnSeq) studies downstream, the animals were infected via the intravenous (IV) route with a saturated library of Mtb transposon mutants (infectious dose of 105 CFU), which in sum produce an infection that is similar to the wild-type parental strain (Bellerose et al., 2020; Sassetti and Rubin, 2003). Groups of three to six male mice per genotype were infected and TB disease-related traits were quantified at one-month post-infection. Data from all surviving animals that were phenotyped are provided in Figure 1—source data 1. The bacterial burden after 4 weeks of infection was assessed by plating (colony-forming units, CFU) and quantifying the number of bacterial chromosomes in the tissue (chromosome equivalents, CEQ). These two metrics were highly correlated (r = 0.88) and revealed a wide variation in bacterial burden across the panel (Figure 1A and Figure 1—figure supplement 1). The phenotypes of the inbred founder strains were largely consistent with previous studies employing an aerosol infection (Smith et al., 2016), where the WSB strain was more susceptible than the more standard B6, 129S1/SvlmJ (129), and NOD/ShiLtJ (NOD) strains. Across the entire CC panel, lung bacterial burden varied by more than 1000-fold, ranging from animals that are significantly more resistant than B6, to mice that harbored more than 109 bacteria in their lungs (Figure 1A). Bacterial burden in the spleen also varied several thousand-fold across the panel and was moderately correlated with lung burden (r = 0.43) (Figure 1—figure supplement 1). Thus, the CC panel encompasses a large quantitative range of susceptibility. Figure 1 with 2 supplements see all Download asset Open asset he spectrum of M.tuberculosis disease-related traits across the collaborative cross. (A) Average lung CFU (log10) across the CC panel at 4 weeks post-infection. Bars show mean ± SD for CFU per CC or parental strain; groups of three to six mice per genotype were infected via IV route (infectious dose of 104 in the lungs and 105 in the spleen as quantified by plating CFU 24 hr post-infection). To compare the field standard B6 mouse strain with the diverse CC mouse strains, bars noted with * indicate strains that were statistically different from B6 (p < 0.05; 1-factor ANOVA with Dunnett's post-test). (B) Heatmap of the 32 disease-related traits (log10 transformed) measured including: lung and spleen colony forming units (CFU); lung and spleen chromosomal equivalents (CEQ); weight loss (% change); cytokines from lung; 'earliness of death' (EoD), reflecting the number of days prior to the end of experiment that moribund strains were euthanized. Mouse genotypes are ordered by lung CFU. Scaled trait values were clustered (hclust in R package heatmaply) and dendrogram nodes colored by 3 k-means. Blue node reflects correlation coefficient R > 0.7; green R = 0.3–0.6 and red R < 0.2. Source files of all measured phenotypes are available in Figure 1—source data 1. (C) Correlation of lung CFU and weight (% change) shaded by CXCL1 levels. Genotypes identified as statistical outliers for weight are noted by #; CXCL1 by † (CC030 is triangle with #†;CC040 is triangle with #; AJ is circle with #; CC056 is circle with †). (D) Correlation of lung CFU and IFNγ levels shaded by IL-17. Strains identified as outliers for IFNγ noted by # (CC055 is left circle with #, AJ is right circle with #, CC051 is triangle with #). Each point in (C) and (D) is the average value per genotype. Outlier genotypes were identified after linear regression using studentized residuals. (E–H) Disease traits measured in a validation cohort (B6 vs CC042, CC032, CC037, and CC027) at 4 weeks after post low-dose aerosol infection (E) lung CFU (log10); (F) Weight (percent change relative to uninfected); (G) CXCL1 abundance in lung (log10 pg/mL homogenate); (H) IFNγ (log10 pg/mL homogenate). Bar plots show the mean ± SD. p-Values indicate strains that were statistically different from B6 (1-factor ANOVA with Dunnett's post-test). Source files of all measured phenotypes in the aerosol validation cohort are available in Figure 1—source data 2. Groups consist of three to six mice per genotype. All mice in the initial CC screen and validation cohort were male. Figure 1—source data 1 CC TB disease phenotypes. TB disease-related phenotypes measured in the CC and parental strains at one-month post-infection. Recorded values are the average and standard deviation of indicated number of mice per genotype ('N of mice infected' at the start of the large screen and 'N of surviving phenotyped animals'). Mice were infected over three batches (denoted by 'block'). 'Freezer days' denotes the number of days prior to the one-month end of infection timepoint that some moribund genotypes were euthanized in accordance with IACUC approved endpoints. 'Blaze' denotes genotypes with white head-spotting coat color trait (WSB haplotype for Kitl; used as a positive control/proof-of-concept for QTL mapping as per Aylor et al., 2011; Smith et al., 2019). https://cdn.elifesciences.org/articles/74419/elife-74419-fig1-data1-v2.zip Download elife-74419-fig1-data1-v2.zip Figure 1—source data 2 Aerosol validation phenotypes. TB disease-related phenotypes measured in B6 and the susceptible CC genotypes (CC027, CC032, CC037, CC042) after infection with Mtb by low-dose aerosol infection. Recorded values are the individual measurements per mouse, designated by genotype. https://cdn.elifesciences.org/articles/74419/elife-74419-fig1-data2-v2.zip Download elife-74419-fig1-data2-v2.zip Comparing various measures of infection progression showed many expected correlations but also an unexpected decoupling of some phenotypes. As an initial assessment of the disease processes in these animals, we correlated bacterial burden and lung cytokine abundance with measures of systemic disease such as weight loss and sufficient morbidity to require euthanasia ('earliness of death'). In general, correlations between these metrics indicated that systemic disease was associated with bacterial replication and inflammation (Figure 1B and Figure 1—figure supplement 1). Lung CFU was strongly correlated with weight loss, mediators that enhance neutrophil differentiation or migration (CXCL2 (MIP-2; r = 0.79), CCL3 (MIP-1a; r = 0.77), G-CSF (r = 0.78), and CXCL1 (KC; r = 0.76)), and more general proinflammatory cytokines (IL-6 (r = 0.80) and IL-1α (r = 0.76)) (Figure 1—figure supplement 1). These findings are consistent with previous work in the DO panel, that found both proinflammatory chemokines and neutrophil accumulation to be predictors of disease (Ahmed et al., 2020; Gopal et al., 2013; Koyuncu et al., 2021; Niazi et al., 2015). The reproducibility of CC genotypes allowed us to quantitatively assess the heritability (h2) of these immunological and disease traits. The percent of the variation attributed to genotype ranged from 56%–87% (mean = 73.4%; (Appendix 1—table 1)). The dominant role of genetic background in determining the observed phenotypic range allowed a more rigorous assessment of strains possessing outlier phenotypes than is possible in the DO population, based on linear regression using studentized residuals that accounts for the intragenotype variation. For example, despite the correlation between lung CFU and weight loss (r = 0.57), several strains failed to conform to this relationship (Figure 1C). In particular, CC030/GeniUnc (p = 0.003), CC040/TauUnc (p = 0.027) and A/J (p = 0.03) lost more weight than their bacterial burdens would predict (Figure 1C; noted by #). Similarly, CXCL1 abundance was higher in CC030/GeniUnc (p = 0.001) and lower in CC056/GeniUnc (p = 0.040), than the level predicted by their respective bacterial burden (Figure 1C; outlier genotypes noted by †). Thus, these related disease traits can be dissociated based on the genetic composition of the host. The cluster of cytokines that was most notably unrelated to bacterial burden included IFNγ and the interferon-inducible chemokines CXCL10 (IP10), CXCL9 (MIG), and CCL5 (RANTES) (Red cluster in Figure 1B; Figure 1—figure supplement 1) (R < 0.3). Despite the clear protective role for IFNγ (Cooper et al., 1993; Flynn et al., 1993), high levels have been observed in susceptible mice, likely as a result of high antigen load (Barber et al., 2011; Lazar-Molnar et al., 2010). While high IFNγ levels in susceptible animals was therefore expected, it was more surprising to find a number of genotypes that were able to control bacterial replication yet had very low levels of this critically important cytokine (Figure 1D). This observation is likely due the inclusion of two founder lines, CAST/EiJ (CAST) and PWK/PhJ (PWK) that we previously found to display this unusual phenotype (Smith et al., 2016). To assess the reproducibility of these findings in an aerosol infection model, we tested four CC genotypes that were susceptible by IV infection, including CC027, CC032, CC037, and CC042. We infected groups of 4–6 mice per genotype with H37Rv strain via low-dose aerosol infection (~100 CFU), including B6 mice as resistant controls. At 4-weeks post infection, we quantified lung CFU, lung cytokine abundance and weight loss as measurements of TB disease. Compared to the resistant B6 mice, the selected CC strains demonstrated higher bacterial burden in the lung (Figure 1E) and significant weight loss (Figure 1F), thus validating disease traits as consistent across both route and dose. Likewise, cytokines that were highly correlated with lung burden in the CC screen (Figure 1B, Figure 1—figure supplement 1) were consistent in the aerosol validation study (Figure 1—figure supplement 2). Notably, CXCL1 was consistently high in the susceptible genotypes, as compared to B6 (Figure 1G), and was highly correlated with lung burden by both IV (R = 0.76) and aerosol (R = 0.92) routes. IFNγ levels were variable across the strains (Figure 1H) and did not correlate with lung CFU (R = −0.22), concordant with findings from the CC screen (R = −0.21). Altogether, this survey of TB-related traits demonstrated a broad range of susceptibility and the presence of qualitatively distinct and genetically determined disease states. TipQTL define genetic variants that control TB immunophenotypes Tuberculosis ImmunoPhenotype Quantitative Trait Loci (TipQTL), which were associated with TB disease or cytokine traits, were identified and numbered in accordance with previously reported TipQTL (Smith et al., 2019). Of the 32 TB-disease traits, we identified nine individual metrics that were associated with a chromosomal locus. Of these, three were associated with high confidence (p ≤ 0.053), and six other QTL met a suggestive threshold as determined by permutation analysis (p < 0.2; Table 1). Several individual trait QTL occupied the same chromosomal locations. For example, spleen CFU and spleen CEQ, which are both measures of bacterial burden and highly correlated, were associated with the same interval on distal chromosome 2 (Table 1, Tip5; Figure 2A and C). IL-10 abundance was associated with two distinct QTL (Table 1). While IL-10 was only moderately correlated with spleen CFU (R = 0.48), one of its QTL fell within the Tip5 bacterial burden interval on chromosome 2 (Figure 2A and C). At this QTL, the NOD haplotype was associated with high values for all three traits (Figure 2E). Similarly, the strongly correlated traits, CXCL1 abundance and lung CFU, were individually associated to the same region on chromosome 7 (Table 1, Figure and In this the haplotype was associated with both low bacterial burden and CXCL1 (Figure At both Tip5 and we found statistical that the of the associated QTL were different = = et al., 2019). These observations support the role of a single variant at each that is for a mapping can both additional statistical support for QTL (p values by combined = = and suggests mechanisms of disease Figure 2 Download asset Open asset Host loci underlying TB disease-related traits. genome QTL of (A) spleen CEQ, spleen CFU and IL-10 (B) lung CFU and (C) of chromosome two (D) of chromosome were determined by permutation and lines represent = = and = 0.2. Scaled phenotype value per haplotype at the QTL Each represents the mean value for a genotype. Table 1 Tuberculosis ImmunoPhenotype QTL QTL within the same interval and clear effects are designated with the same TipQTL p-Values are determined by (Churchill and QTL, quantitative trait of the CEQ, chromosomal start end number of factors can the statistical of QTL identified in the CC population, including small effect genotype and the genetic complexity of the We an to assess the of the lung CFU QTL on chromosomes 7 and and Table 1). that the at both QTL were by the haplotype (Figure we an population based on two CC strains, and that at and (Figure and The validation cohort = were et al., and infected with the Mtb strain H37Rv route with infectious dose of 105 CFU, as per the CC At 1 post infection, lung CFU was and we QTL mapping in et al., to identify host loci underlying bacterial burden in the We identified a QTL significantly associated with lung CFU = < 0.05; on chromosome that with thus validating this as a of bacterial In this complexity cross, we did not a QTL on chromosome This may be due to the B6 haplotype at this in which did not represent the phenotypic to in the mapping validation we identified a new lung CFU) on chromosome = by the with the This QTL was not in the CC due to the low of the haplotype at that in the CC cohort Altogether, this as a of lung CFU, rigorous validation of may require a more optimal of parental strains. Figure 3 Download asset Open asset to QTL underlying lung CFU. (A) of and CC strains at and (B) at The population = based on these were infected with Mtb infectious dose by IV as per the CC and lung CFU was quantified at 1 post-infection. (C) QTL mapping identified significant (p < loci on = on at with and a new on = were determined by permutation and lines represent = = and = 0.2. Source files of genotypes are available in Figure data phenotypes are available in Figure data 1. Figure data 1 genotype genotype data from mice from The infected cohort included both male and mice, as Download Figure data 2 phenotype Lung CFU data quantified by plating CFU from Mtb infected lungs from mice at 1 post infection with genotype data in Figure data 1). The infected cohort included both male and mice, as Download Mtb to diverse hosts by distinct This survey of traits demonstrated that the CC panel encompasses a number of qualitatively distinct immune phenotypes. To determine different bacterial were necessary to adapt to these we leveraged transposon sequencing (TnSeq) as a high-resolution phenotyping to the relative abundance of individual Mtb mutants after in each CC host genotype. To as of known immunological we also in B6 mice that were lacking the mediators of Th1 immunity, and IFNγ or were lacking the mediators that control disease by et al., or the et al., 2018). The relative of each Mtb mutant in the library the library from each mouse spleen after one-month of infection was quantified by et al., 2015). of saturated Mtb transposon were 60 distinct mouse genotypes (Figure data 1). this we identified Mtb genes that are required for or of Mtb in B6 mice, based on significant of the mutant after four weeks of in percent of these genes with a similar previous analysis in mice (Bellerose et al., the of the All but one of the genes found to be important in B6, were also required in the mouse panel, confidence in this Mtb (Figure and While the number of genes found to be necessary in each genotype across the diversity panel was largely the composition of these Mtb varied As more CC strains, and more distinct immune states, were included in the the number of bacterial genes necessary for in these animals also This at after the inclusion of approximately mouse genotypes (Figure additional of B6 not the number of genes identified as necessary for in that genotype (Figure supplement the presence of selective across the CC The number of genes important for fitness in the CC panel the genes identified in the B6 and strains combined (Figure and Figure data 1). Figure 4 with 1 supplement see all Download asset Open asset Mtb genetic requirements vary across diverse (A) The number of Mtb genes required for or in each diverse mouse strain across the panel the mutants required for each strain; the as each new host strain is (B) the composition of Mtb required in each of host only required in the CC panel required in specific immunological mice and genes required in B6 mice is required in B6 and In to be in each mouse strain, Mtb genes had to be significantly over or in at two (C) Each