Mycobacterium tuberculosis-induced monocyte transcriptional responses associated with resistance to tuberculin skin test/interferon-gamma release assay conversion in people with HIV

Abstract

Objective:

To determine whether Mycobacterium tuberculosis (Mtb)-induced monocyte transcriptional responses differ in people with HIV (PWH) who do (RSTR) or do not (LTBI) resist TST/IGRA conversion after exposure.

Design:

We compared ex vivo Mtb-induced monocyte transcriptional responses in a Ugandan tuberculosis (TB) household contact study of RSTR and LTBI individuals among PWH.

Methods:

Monocytes were isolated from peripheral blood mononuclear cells from 19 household contacts of pulmonary TB patients, and their transcriptional profiles were measured with RNA-Seq after a 6-hour infection with Mtb (H37Rv) or media. Differentially expressed genes (DEGs) were identified by a linear mixed effects model and pathways by gene set enrichment analysis that compared RSTR and LTBI phenotypes with and without Mtb stimulation.

Results:

Among PWH, we identified 8,341 DEGs that were dependent on Mtb stimulation (false discovery rate [FDR] < 0.01). Of these, 350 were not significant (FDR > 0.2) in individuals without HIV. Additionally, we found 26 genes that were differentially expressed between RSTR and LTBI monocytes in PWH, including 20 which were Mtb-dependent (FDR < 0.2). In unstimulated monocytes, several gene sets (TGF-β signaling, TNF-α signaling via NF-κB, NOTCH signaling, coagulation, and epithelial mesenchymal transition) were enriched in RSTR relative to LTBI monocytes (FDR < 0.1). These patterns were not observed in individuals without HIV.

Conclusions:

RSTR monocytes in PWH show different gene expression in response to Mtb infection when compared to those with LTBI and RSTR without HIV. These differential expression patterns are enriched in inflammatory pathways.

Introduction

Tuberculosis (TB) is the leading cause of infectious disease-related deaths worldwide and is particularly common and lethal in human immunodeficiency virus (HIV) endemic areas such as sub-Saharan Africa. HIV infection increases the risk of active TB, such that nearly 8% of all TB cases and 187,000 of global TB deaths were HIV-associated in 2021 [1]. Since HIV infection dysregulates T-cell responses to Mycobacterium tuberculosis (Mtb), the causative agent of TB, ineffective adaptive immune responses have been thought to be a main driver of the TB epidemic in this vulnerable population [2, 3]. Interestingly, clinical and epidemiological studies have shown that some individuals are resistant to traditionally defined latent Mtb infection (LTBI) despite intense Mtb exposure [4–7]. In our large-scale TB household contact study in Uganda, approximately 7~9% of close household contacts maintained negative tuberculin skin test (TST) and interferon-γ (IFN-γ) release assay (IGRA) immunoreactivity during extended follow-up, a phenotype we have labeled “resister” (RSTR) [8]. We previously found that HIV-uninfected RSTR in this cohort have differential monocyte transcriptional profiles when compared to LTBI individuals that may identify protective host pathways such as effector responses responsible for early clearance of Mtb [9, 10]. Whether mechanisms of resistance to Mtb infection differs in people with HIV (PWH) is not known. In a recent study conducted in TB hyper-endemic environments in South Africa, a small number of PWH, despite undergoing longitudinal follow-up for repeat IGRA and TST testing, showed no clinical evidence of TB [11]. Regardless of whether these individuals were persistently TB, tuberculin, and IGRA negative or positive, this study found no significant differences in antibody profiles across subclasses/isotypes and Mtb-specific antigens.

Although CD4+ T cells represent the primary reservoir of HIV, innate immune cells derived from the myeloid lineage, such as monocytes and macrophages, are also important targets of HIV [12]. Upon Mtb infection, HIV is thought to adopt distinctive transcriptional machinery in monocytes or macrophages for productive viral replication [13–16]. In the current study, we test our hypothesis that monocytes from PWH have differentially expressed genes (DEGs) and gene sets that distinguish the RSTR from LTBI monocyte response to Mtb infection. First, we first examined whether Mtb-induced monocyte gene expression differs in those living with or without HIV. Second, we compared the transcriptional profiles (DEGs and gene sets) between Mtb-stimulated and unstimulated monocytes in RSTR and LTBI groups among PWH. Finally, we compared these DEGs and gene sets from PWH to our previous study of people living without HIV (hereafter, “non-PWH”) in Uganda [10]. Through such multidimensional comparisons of DEGs that distinguish RSTR status and Mtb stimulation conditions between PWH and non-PWH, we identified HIV-specific RSTR transcriptional responses following Mtb infection.

Methods

Participant recruitment and characterization of clinical phenotypes

This prospective cohort is nested within the Kawempe Community Health Study in Kampala, Uganda. Full details regarding the original study setting, recruitment, informed consent, and ethical approval have been previously described [8, 17]. Briefly, 872 culture-confirmed pulmonary TB (PTB) index cases and their 2,585 household contacts were enrolled between 2002 and 2012 (phase 1) and evaluated at baseline and every 3–6 months thereafter for up to 24 months for evidence of Mtb infection and TB disease progress [8]. Of the total household contacts, 7.6% were PWH (n= 190). During the retracing study between 2014 and 2017 (phase 2), IGRA was added for screening, and 83% were concordantly TST−/IGRA− (RSTR), and 16% converted to TST+/IGRA+ (LTBI) [17]. For the current study, peripheral blood mononuclear cells (PBMCs) from 20 PWH household contacts who had completed serial TST and IGRA were available. Among individuals with a CD4+ T cell count >200/mm³, we defined RSTR phenotype as being persistently TST-negative and IGRA-negative and the LTBI phenotype as being TST and IGRA positive using universal 5 mm induration cut-off criteria for PWH with no signs or symptoms of active TB during this follow-up period. As previously reported, Mtb exposure risk scores were calculated based on the intensity of exposure to the index case, ranging from 0 to 10 for adults and 0 to 9 for children. Both RSTR and LTBI groups were similarly and intensively exposed to PTB in adults (p= 0.37), but we could not compare PTB exposure in children between the two groups due to the absence of children in the LTBI group (Table 1) [18, 19].

Table 1.

Demographic characteristics of study participants

	RSTR (n=9)	LTBI (n=10)	p-value
Number of adults (age ≥15 at phase 2)	7	10
Number of children (age <15 at phase 2)	2	0
Age at phase 2
Median age in adults (min, max)	28 (15, 52)	35 (16, 59)	0.59*
Median age in children (min, max)	7 (6, 8)	-
Sex (N female, %)	7 (77.8%)	6 (60.0%)	0.74∞
Exposure risk score at phase 1
Adults, median (IQR)	6 (6, 8.5)	6 (6, 6.8)	0.37*
Children, median (IQR)	8 (8, 8)	-
Median CD4 counts at phase 2 (IQR)	823 (567, 920)	558 (459, 666)	0.29*

Note. P-values were calculated using the Mann-Whitney test (*) for continuous variables and the chi-square test (^∞) for categorical variables. The exposure risk score was primarily measured during the initial study enrollment (phase 1, between 2002 and 2012), while other variables were measured during the first visit of the retracing study (phase 2, between 2014 and 2017) for each participant.

IQR = interquartile range

Mycobacterium tuberculosis infection of monocytes and RNA sequencing

Samples were prepared for RNA sequencing (RNAseq) as previously described [10]. Briefly, cryopreserved PBMCs from 20 PWH household contacts were thawed and resuspended in RPMI 1640 medium (Gibco, Grand Island, New York, USA) supplemented with 10% heat-inactivated fetal bovine serum (FBS) and 50 ng/mL of monocyte colony-stimulated factor (M-CSF). CD14+ monocytes were isolated from PBMCs using magnetic bead column purification with negative selection and then incubated with the same media solution for 24 hours. Monocytes were then stimulated with H37Rv Mtb at a multiplicity of infection (MOI) of 1 (i.e., Mtb condition) or medium alone (i.e., unstimulated Media condition). After a 6-hour incubation, monocytes were lysed in TRIzol (Invitrogen, Waltham, Massachusetts, USA), and RNA was isolated using miRNeasy Mini Kit according to the manufacturers’ instructions (Qiagen, Venlo, Netherlands). cDNA libraries were prepared with random hexanucleotide primers and rRNA depletion using SMARTer RNAseq Kit (Takara Bio Inc., Shiga, Japan) and were sequenced on an Illumina HiSeq 2500 platform.

Data processing

Sequences were quality-assessed with FastQC (v0.11.9) [20] and filtered with AdapterRemoval (v2.3.2) [21] to remove adapters and poor-quality sequences (score < 30, length < 15, ambiguous > 1). Sequences were aligned to the human genome (GRCh38 release 102) with STAR (v2.7.9a) [22] and quality-assessed with samtools flagstat (v1.7) [23] and Picard (v2.42.2) [24]. Alignments were filtered with samtools to remove PCR duplicates, unmapped, non-primary, and poor-quality alignments (MAPQ < 30). High-quality alignments were then quantified in gene exons using Subread featureCounts (v2.0.1) [25]. Further analysis and filtering steps were performed in R v4.2.1 [26]. Libraries were quality filtered for sequence alignment > 90%, median coefficient of variation of coverage < 1, and two standard deviations from the mean on principal components (PC) 1 and 2. This removed two low-quality libraries from the same donor, thus resulting in 19 individuals (38 libraries) for analysis. Counts were normalized for RNA composition using the trimmed mean of M values normalization method and filtered to protein coding genes with at least 5% of libraries containing at least 0.8 count per million (CPM). Finally, counts were converted to log2 CPM using R package voom [27].

Bioinformatic and statistical analyses

To identify genes with expression patterns that distinguished RSTR and LTBI phenotypes according to monocyte responses to Mtb infection, we selected a linear mixed effects model that incorporated an interaction term and a random effect for patient: Expression~ Mtb + RSTR + Mtb:RSTR + (1| patient). Models were run using the R package kimma [28] which leverages the lme4 statistical framework [29]. Inclusion of age, sex, exposure risk score or CD4 counts as covariates in the model did not improve model fit for the majority (> 99%) of genes as assessed by Akaike information criterion (AIC) (mean change in AIC, age and sex= 11, CD4 counts= 16). Accordingly, no covariates were included in the differential gene expression analysis to opt for the more parsimonious model. DEGs were assessed at an false discovery rate (FDR) < 0.2; RSTR and interaction significant genes were further assessed in an Mtb:RSTR pairwise contrasts model including RSTR effects within media or Mtb-stimulated libraries and Mtb effects within RSTR or LTBI.

To understand biologic connectivity between significant genes, network analysis was conducted using the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING v11.5) [30]. DEGs were assessed for enrichment against the Human Molecular Signatures Database (MSigDB v2022.1) [31] including Hallmark and C2 canonical pathway (KEGG, Reactome) as well as UniProt Knowledgebase (UniProtKB) keywords [32] using Fisher’s exact test. Gene set enrichment analysis (GSEA) [31] was performed against the MSigDB Hallmark gene sets. Significant gene sets were defined at FDR < 0.1 and if non-significant, also at FDR > 0.2.

Candidate gene association analysis

To screen genetic variants in DEGs that are clinically relevant to the RSTR phenotype, we conducted a candidate gene association study using expanded Ugandan household contact cohort data with RSTR non-PWH (n=74) and LTBI non-PWH (n=189) [17]. Odds ratios (ORs) were calculated using a generalized linear mixed model (GENESIS in R [33]), adjusted for age, sex, population structure via principal components and kinship via a genetic relationaship matrix. Statistical comparison of demographic and epidemiologic data between RSTR and LTBI groups were conducted at a two-sided significance level of 0.05 using Stata/IC 15.1 (StataCorp LLC, College Station, Texas, USA) [34].

Results

Transcriptional profiling reveals differentially expressed genes in Mycobacterium tuberculosis-infected monocytes from RSTR and LTBI in people with HIV

To discover DEGs in Mtb-infected monocytes of RSTR and LTBI PWH, we compared the transcriptomes of 19 household contacts of PTB index cases including those who did (LTBI, n=10) or did not (RSTR, n=9) convert their TST/IGRA tests. There were no significant differences in demographic, CD4 counts, or Mtb exposure risk scores between RSTR and LTBI groups (Table 1). Using a linear mixed effects model, we first identified 11,165 DEGs associated with Mtb infection by comparing gene expression in Mtb-infected and uninfected (i.e., stimulated with media only) monocytes (FDR < 0.2, termed “Mtb-dependent DEGs”, Fig. 1A, Supplemental Table 1A). After controlling for Mtb infection status, we found 15 DEGs that distinguished RSTR versus LTBI monocytes regardless of the Mtb stimulation condition (FDR < 0.2, termed “RSTR DEGs”, Fig. 1A,B, and Supplemental Table 1B). For example, ITGB8 was more highly expressed in RSTR than LTBI in both the media and Mtb-stimulated conditions (Fig. 1 C). Finally, we utilized the interaction term (Mtb:RSTR) in our linear mixed model to identify DEGs impacted by both RSTR status and Mtb stimulation. This resulted in 11 additional DEGs, which were further assessed by pairwise contrasts to identify genes where Mtb stimulation effects differed in RSTR versus LTBI and/or where RSTR/LTBI effects differed in media versus Mtb stimulated monocytes (FDR < 0.2, termed “Interaction DEGs”, Fig. 1A,B, Supplemental Table 1C). For example, CDC42BPG was less expressed in RSTR than LTBI in the media condition. Upon Mtb stimulation, however, the expression pattern flipped with RSTR showing higher expression than LTBI (Fig. 1C).

Fig. 1. Mtb and RSTR DEGs identified in monocytes from PWH.

(A) Venn diagram of DEGs at FDR < 0.2. Genes were assessed in the model ~ Mtb + RSTR + Mtb:RSTR + (1 | patient). Total significant genes are quantified for the two main effects (top) and the interaction term (bottom). (B) Volcano plots of RSTR and interaction DEGs. In total, 15 RSTR DEGs (left) and 11 interaction DEGs (right) were identified. Significant (FDR < 0.2) genes are indicated by positive log2 fold change (red squares) and negative log2 fold change (blue triangles). (C) Normalized expression of representative DEGs. Log2 normalized expression of a RSTR DEG (ITGB8) and an interaction DEG (CDC42BPG) are shown. Horizontal bars indicate the group mean with standard deviation error bars. Pairwise contrast significance is indicated by * FDR < 0.2, ** FDR < 0.05

DEG: differentially expressed gene, RSTR: resister, LTBI: latent TB infection, PWH: people with HIV, FDR: false discovery rate, Mtb: Mycobacterium tuberculosis.

We next used a linear regression model toinvestigate potential correlations between Mtb exposure scores and the expression of DEGs that distinguish RSTR and LTBI individuals.. Among the 26 RSTR and interaction DEGs examined, we observed a negative association with the risk score for one Mtb:RSTR interaction DEG (CDH13) and a positive association for one RSTR DEG (QRICH1), specifically under the media condition (FDR < 0.2) (Supplemental Table 1F). These associations lost significance following Mtb infection.

Using the STRING database, we examined the network of protein-level interactions of the 26 RSTR and interaction DEGs but found no strong connections (STRING score < 400, Supplemental Fig. 1A). These DEGs showed no enrichment in Hallmark or UniProtKB keyword gene sets but were weakly enriched in several C2 canonical pathways including cell adhesion, IFN signaling and extracellular matrix organization (FDR < 0.3, Supplemental Fig. 1B, and Supplemental Table 2A–C). Together, these data suggested that RSTR and LTBI monocytes from PWH have different transcriptional profiles in response to Mtb infection, but the functional pathways corresponding to these transcriptional differences remain undefined.

Mycobacterium tuberculosis-induced gene set enrichment reveals distinct pathways between RSTR and LTBI in people with HIV and nonpeople with HIV cohorts

We next utilized GSEA to identify pathways associated with Mtb resistance and susceptibility in PWH (Supplemental Table 2D) and compared their transcriptional profiles to our previous study of Ugandan cohorts without HIV to determine HIV-specific pathways in response to Mtb (RSTR, n=49 and LTBI, n=52) [10]. In unstimulated PWH monocytes, TGF-β signaling and TNF-α signaling via NF-κB (Fig 2A,B, section i) as well as NOTCH signaling, coagulation, and epithelial mesenchymal transition (Fig 2A,B, section iv) gene sets were more highly expressed in RSTR relative to LTBI (FDR < 0.1, Fig. 2A). With Mtb infection, the degree of enrichment of these gene sets increased independently of RSTR phenotype (Supplemental Fig. 2). Intriguingly, the differences in TGF-β signaling and TNF-α signaling via NF-κB, wherein the expression was higher in RSTR individuals among PWH whereas it was lower among non-PWH (Fig 2A,C, section i). Adipogenesis and oxidative phosphorylation (section ii) and allograft rejection (section iii) gene set enrichments showed consistently higher or lower expression, respectively, in RSTR monocytes regardless of Mtb stimulation nor HIV status (Fig. 2A,B).

Fig. 2. GSEA of transcriptional profiles of RSTR versus LTBI monocytes with and without Mtb infection.

GSEA of MSigDB Hallmark gene sets from monocyte transcriptional profiles within media or Mtb-infected conditions. PWH (A,B) and non-PWH (C,D) in the Ugandan cohort as reported in Simmons et al. (10). All gene sets significant in at least one condition in PWH (FDR < 0.1) are shown in addition to those with concordant direction in individuals without HIV. From the top panel: section i includes gene sets with differences in expression between PWH and non-PWH; sections ii and iii include gene sets that are consistent across HIV status and cohort, split by those down-regulated (ii) or up-regulated (iii) in response to Mtb (Supplemental Fig. S2); section iv includes gene sets with inconsistent expression trends across HIV status and/or cohort.

GSEA: gene set enrichment analysis, RSTR: resister, LTBI: latent TB infection, PWH: people with HIV.

The IFN-γ response gene set was highly enriched in PWH LTBI monocytes after Mtb stimulation, as was also observed in a similar Uganda cohort without HIV. Importantly, this enrichment in LTBI was also seen in the absence of Mtb antigens under unstimulated condition (Fig. 2A,B, FDR < 0.1, section iii) yet Mtb-stimulation resulted in increased expression (i.e., a positive enrichment score in Supplemental Fig. 2) of the IFN-γ response genes across both phenotypes when compared to the unstimulated condition. Such directionality is expected since our phenotypes are defined by an Mtb-induced IFN-γ response based on the enrollment criteria of the cohort via IGRA tests. Similarly, IFN-α response showed consistently higher expression in LTBI than RSTR with and without Mtb stimulation (Fig. 2A,B). As we previously described in non-PWH, core enrichments of IFN-α gene sets among HIV transcriptomes overlapped with IFN-γ response genes (Supplemental Fig. 2 in Simmons et al. [10]). Among monocytes from PWH in this study, the IFN-α response mostly overlapped with the IFN-γ response with 78% of IFN-α leading edge genes also in the IFN-γ leading edge. In contrast, IFN-γ contained many genes distinct from IFN-α with only 40% of its leading edge found in both gene sets, suggesting IFN-γ signaling is the driver of the IFN-α enrichment among LTBI monocytes from cohorts with and without HIV. Together, these data suggest that gene sets in several pathways are expressed differentially between RSTR and LTBI in PWH and non-PWH monocytes in response to Mtb stimulation.

Transcriptional profiling of Mycobacterium tuberculosis-infected monocytes reveal distinct differentially expressed genes and pathways when comparing those with or without HIV

We further investigated HIV- and RSTR-specific transcriptional responses to Mtb by comparing Mtb-dependent DEGs between PWH and non-PWH cohorts in the Ugandan household contact study. To ensure rigor, we employed astringent FDR threshold of <0.01 to determine significance in the PWH cohort and, conversely, a threshold of >0.2 in the non-PWH cohortto denote non-significance. Among the 8,341 Mtb-dependent DEGs from PWH (FDR < 0.01), 350 were not significant (FDR > 0.2) in non-PWH (Fig. 3, top panel). Conversely, of the 9,110 Mtb-dependent DEGs (FDR < 0.01) from non-PWH, 797 were not significant (FDR > 0.2) in PWH. Therefore, we designated these two sets of DEGs as putative HIV-specific, Mtb-dependent DEGs, and hereafter, “Mtb-HIV specific DEGs”. (Supplemental Table 3A). The remaining DEGs were significant in both cohorts (FDR < 0.01, n=7,154), near the significance threshold in one cohort (0.01 < FDR < 0.2, n=526 PWH, n=1,054 non-PWH), or failed QC in one cohort (n=339 PWH, n=105 non-PWH). Conversely, there was no overlap across cohorts among the Mtb:RSTR interaction DEGs (FDR < 0.2, Fig. 3, bottom panel) and each RSTR DEG in PWH was unique since none were identified among non-PWH.

Fig. 3. Monocyte DEGs comparing people living with or without HIV from Ugandan household contact study.

The plot shows the number of genes assigned to each statistically significant category based on FDR cutoffs in PWH and non-PWH in the Ugandan cohort. The color represents the following categories from top to bottom: pink indicates DEGs that are only significant in the PWH group; orange indicates DEGs that are significant in PWH and borderline significant in the non-PWH group; yellow indicates significant DEGs found in both groups; blue indicates DEGs that are significant in non-PWH and borderline significant in the PWH group; and purple indicates DEGs that are only significant in the non-PWH group. The top panel shows a stacked bar graph of overlapping Mtb-specific DEGs in PWH and non-PWH groups. Significance was assessed at FDR < 0.1 and non-significance at FDR > 0.2. A total of 350 DEGs (pink) were found to be specific to Mtb infection in monocytes from PWH group, while 797 (purple) were specific to the non-PWH group. The bottom panel shows a stacked bar graph of overlapping Mtb:RSTR interaction DEGs in PWH and non-PWH groups. Significance was assessed at FDR < 0.2 and non-significance at FDR > 0.4. A total of 11 DEGs (pink) were found to be specific to Mtb infection and RSTR status in monocytes from PWH, while 252 were specific to individuals in non-PWH group (purple).

DEG: differentially expressed gene, RSTR: resister, LTBI: latent TB infection, PWH: people with HIV, FDR: false discovery rate, Mtb: Mycobacterium tuberculosis.

Using pathway enrichment analysis, we identified C2 canonical pathways and UniProtKB keywords associated with Mtb-HIV specific DEGs (Supplemental Table 3B–D). The 350 Mtb-HIV specific DEGs in PWH were uniquely enriched in several transcription related gene sets such as transcription regulation, DNA-binding, zinc-finger, and TCF transactivating complex as well as DNA repair and replication such as the ORC complex and base excision repair (FDR < 0.05, Supplemental Fig. 3A). Using STRING, we found that 175 of the 350 Mtb-HIV specific DEGs had medium to high confidence potential protein-protein interactions (STRING score > 400) and the majority of these were members of one or more enriched gene set (Supplemental Fig. 3B). The 797 Mtb-HIV specific DEGs from those without HIV were uniquely enriched in fewer gene sets including phosphoprotein and histone acetylation (FDR < 0.05, Supplemental Fig. 3C). Together, these data suggest that Mtb-infected monocyte transcriptional profiles contain both similarities and differences when comparing individuals living with or without HIV.

Differentially expressed gene polymorphisms and association with the RSTR phenotype

We also used a candidate gene association study to determine whether RSTR phenotype in the Ugandan cohort without HIV was associated with any of the 1,480 genetic variants in the 26 DEGs, which included both 15 RSTR DEGs and 11 interaction DEGs. We found that 63 single nucleotide polymorphisms (SNPs) residing in 12 DEGs (including 5kb flanking regions) were significantly associated with the RSTR versus LTBI clinical outcome with a nominal p value (p < 0.05, Supplemental Table 4). Three DEGs had associated SNPs with a more stringent threshold of p < 0.01, including CDC42BPG, BCO2, and CDH13. Among these SNPs, rs916606 is an expression quantitative trait locus (eQTL) in the CDC42BPG genein brain and skin tissues in a public database (GTExPortal, https://www.gtexportal.org). However, none of these findings were significant after correction for multiple comparisons. Together, these data suggest that genetic variants in RSTR and interaction DEGs may not be associated with RSTR versus LTBI clinical status in household contacts living without HIV. Due to the small sample size, larger cohort studies are needed for a more robust assessment of genetic contributions to the RSTR-LTBI distinction.

Discussion

PWH in TB endemic areas are at higher risk of TB incidence and TB-associated mortality and morbidity. To gain insight into the molecular mechanisms of resistance to Mtb infection during HIV infection, we analyzed monocyte transcriptional profiles associated with persistently TST/IGRA-negative resisters (RSTR) and LTBI individuals from rigorously characterized PTB household contacts living with HIV. We identified 26 DEGs that discriminate RSTR and LTBI transcriptional responses, with 20 of them specifically responding to Mtb in PWH. In addition, we identified 350 Mtb-HIV specific DEGs by comparing Mtb-induced genes from individuals living with and without HIV. These genes as well as overall pathway expression suggest that multiple mechanisms of inflammatory processes can influence clinical resistance or susceptibility to Mtb infection in this population.

Unlike our previous studies with Ugandan household contacts without HIV, gene expression related to TGF-β signaling was associated with RSTR status in PWH. Transforming growth factor-β (TGF-β) is an anti-inflammatory and profibrotic cytokine that is persistently elevated in PWH [35, 36]. The active form of TGF-β inhibits T-helper (Th) 1 and Th2 cell differentiation, and in concert with other cytokines, promotes Th17 or regulatory T (Treg) cell differentiation [37–42]. Here, we found that in the absence of Mtb infection, ITGB8 (a subunit of integrin αvβ8) is significantly more expressed in RSTR than LTBI monocytes. These data suggest that integrin αvβ8-mediated-TGF-β activation may be more dominant in RSTR than LTBI monocytes, given that myeloid cell integrin αvβ8 mediates TGF-β activation [42, 43]. Regardless of HIV status, both RSTR and LTBI monocytes have heightened proinflammatory transcriptional signatures following Mtb infection (Supplemental Fig. 2), characterized by genes involved in TNF-α-NFκB signaling and inflammatory response. In addition, the enhanced TGF-β signaling in RSTR, potentially together with IL-6 [35, 44], may promote the differentiation of Th17 cells to fight against Mtb infection. Furthermore, TGF-β and NOTCH signaling, which was also more highly expressed in RSTR at baseline, contribute to group 3 innate lymphoid cell (ILC3) differentiation [45]. Like Th17 cells, ILC3s are involved in lung defense by generating IL-17A [46], resulting in excessive neutrophil recruitment and augmenting airway inflammation [47, 48]. To date, although a host-protective role of neutrophils against Mtb infection is controversial, uncontrolled neutrophil influx and inflammation at the infection site can cause severe tissue damage and airway remodeling [49–51]. Concurrently, we identified significant enrichment of genes in the epithelial mesenchymal transition (EMT) pathway in RSTR PWH. Together with the TGF-β results, this may indicate that TGF-β-induced EMT plays an important role in airway epithelial repair after flares of inflammation during early infection [52–54]. CDC42BPG, a downstream effector of CDC42 GTPase, was differentially expressed in RSTR and is thought to be involved in TGF-β-induced EMT [55]. Collectively, these data suggest that RSTR monocytes in PWH are more likely to be involved in epithelialization and tissue remodeling, presumably through a TGF-β-dependent manner mediated by CDC42BPG overexpression. Thus, TGF-β, TNF-α, NOTCH signaling, and EMT pathways that are enriched at baseline among RSTR monocytes may help clear early Mtb infections but possibly at the expense of inflammation and tissue damage.

This study has several limitations. First, the accurate classification of RSTR remains challenging as neither TST nor IGRA is a direct measure of Mtb infection in asymptomatic individuals [56]. To minimize RSTR phenotype misclassification, we confirmed that both RSTR and LTBI groups had high and similar Mtb exposure through rigorously characterized exposure risk scores as previously reported [8]. We also traced them with serial tests after 8–10 years of follow-up to identify LTBI individuals who take longer to convert. Additionally, all participants included in our study had CD4+ T cell counts above the previously reported cutoffs (minimum 216/mm³, median 583/mm³), which reduces the chance of misclassifying RSTR individuals due to an anergic T cell IFN-γ response as measured by TST/IGRA, although such misclassification remains possible [57–60]. Second, our study had a relatively small sample size of PWH (n=19), resulting in limited statistical power and making direct comparison with non-PWH cohort (n=101) difficult. Also, our model did not include potential covariates of HIV-associated TB susceptibility due to data availability constraints. Future studies with larger sample sizes and more comprehensive data collection on potential covariates, such as viral load, duration of living with HIV infection, and antiretroviral therapy, are needed to enable more robust comparisons with HIV uninfected counterparts. Third, as previously reported [9], we observed evidence of potential contamination with T cells, identified by lymphocyte cytokine gene expression (e.g., IFNG and IL2) in our transcriptional data. CD14 selection resulted in highly enriched monocyte populations, but minor T cell contamination cannot be excluded. Finally, since our analysis compared Mtb-specific transcriptional differences in RSTR and LTBI monocytes, it will be necessary to confirm transcriptional differences in other primary myeloid cell types, including alveolar macrophages and dendritic cells that have distinct ontogeny and differentiation pathways.

Despite these limitations, our study provides unique insights that can inform the potential mediators of Mtb resistance in the context of HIV-associated chronic immune activation. Many genes and pathways that distinguish the RSTR phenotype involve TGF-β mechanisms of Mtb resistance in conjunction with HIV-infection. These findings open many potential avenues for future investigation in TGF-β-mediated IFN-γ-independent protective mechanisms in early Mtb infection.

Data availability

FASTQ and count table data are available through the NCBI database of Genotypes and Phenotypes (dbGaP) data browser (https://www.ncbi.nlm.nih.gov/gap/) under accession phs002445. However, as part of the study participants’ consents, authorization for data access must first be approved by a local data access committee (DAC), the contact for which can be found on the dbGaP page. Once local DAC approval is granted, a letter of collaboration from the authors will be provided and can be presented to dbGAP to gain access to file download.