A single-cell atlas of peripheral immunity suggests immune subset attrition during tuberculosis dissemination

Abstract

Objective

This study investigated immune lineage remodeling and intercellular communication to elucidate the mechanisms driving the progression from active tuberculosis to disseminated tuberculosis.

Methods

We integrated single-cell RNA-sequencing datasets of peripheral blood mononuclear cells from healthy controls, patients with active tuberculosis, and patients with disseminated tuberculosis (n = 3 for each group) to perform cellular clustering, functional assessment, and communication network analyses.

Results

Analysis identified 16 clusters across 5 major lineages. Although global cellular proportions remained statistically stable across patient cohorts, disease progression was characterized by distinct cell-intrinsic transcriptomic shifts and subset-specific remodeling. This landscape was highlighted by the systematic peripheral depletion of homeostatic “guardian” monocytes, alongside functional disruptions in natural killer and B-cell states, and a severe-stage-specific accumulation of exhausted T cells. Consequently, circulating intercellular communication networks were broadly attenuated in infected groups, featuring diminished receptor–ligand interactions between peripheral dendritic cells and naïve CD8⁺ T cells, driven by cell-intrinsic transcriptional rewiring rather than shifts in underlying cell frequencies.

Conclusions

Tuberculosis progression involves profound peripheral immune remodeling marked by the systemic attrition of circulating homeostatic populations and altered blood interactome dynamics. This preliminary atlas provides an exploratory framework for potential therapeutic interventions in severe tuberculosis.

Introduction

Disseminated tuberculosis (DTB) is a severe manifestation of Mycobacterium tuberculosis infection, resulting from hematogenous or lymphatic dissemination to multiple organs, including the lungs, liver, spleen, bone marrow, and central nervous system. ¹ DTB predominantly affects immunocompromised individuals, including those with human immunodeficiency virus (HIV) infection or those undergoing immunosuppressive therapy. The pathogenesis of DTB reflects a combination of host immune dysregulation and enhanced bacterial virulence. ²

Despite advances in tuberculosis (TB) research, the immunological mechanisms driving DTB remain poorly characterized. Over the past two decades, pioneering bulk transcriptomic analyses of peripheral blood have successfully mapped the global immune landscape of ATB, consistently uncovering robust types I and II interferon (IFN)-inducible gene signatures, myeloid-driven inflammatory profiles, and transcriptomic correlates of risk or treatment response.^3–5 However, because bulk transcriptomics captures an averaged signature across millions of pooled cells, it inherently lacks the resolution to distinguish whether these signatures stem from shifts in cell-type abundance (compositional changes) or true, cell-intrinsic functional reprogramming. Furthermore, bulk averaging obscures which specific immune subsets are actively driving these protective or pathogenic signatures, and it completely masks the coordinated intercellular signaling networks required to control infection.

Single-cell approaches offer an unprecedented opportunity to complement and advance these bulk findings by dissecting this averaged signal into high-resolution, cell-type-specific transcriptomic layers. However, single-cell RNA sequencing (scRNA-seq) has been scarcely applied to severe manifestations such as DTB, leaving the peripheral immune landscape incompletely mapped. ⁶

To address these gaps, we applied scRNA-seq to peripheral blood mononuclear cells (PBMCs) from healthy controls, patients with active tuberculosis (ATB), and patients with DTB. We centered this study on the hypothesis that the clinical transition from localized pulmonary infection (ATB) to severe, systemic, multiorgan dissemination (DTB) is driven by progressive, stepwise cellular remodeling and a systematic collapse of innate–adaptive immune communication networks. By integrating differential gene expression, pathway enrichment, and intercellular communication analyses, we systematically tested this hypothesis to characterize immune dysregulation, functional exhaustion, and abnormalities in immune networks during TB progression.

Materials and methods

Clinical baseline characteristics

For this study, three groups (n = 3) were selected: healthy controls, patients with ATB, and patients with DTB. The clinical baseline information was compiled into Supplementary Table 1.

Data acquisition and filtering

scRNA-seq datasets were retrieved from the National Center for Biotechnology Information (NCBI) database. scRNA-seq data for healthy controls were obtained from the Gene Expression Omnibus (GEO) under the accession numbers GSM8895292, GSM8895287, and GSM8895289. DTB patient datasets were downloaded from GEO with accession numbers GSM8743124, GSM8743126, and GSM8743123. For the ATB group, raw sequencing data were downloaded from the Sequence Read Archive (SRA) using accession numbers SRR11038989, SRR11038990, and SRR11038991. All datasets used in this study are publicly available and accessible without restriction.

scRNA-seq data processing and integration analysis

All single-cell RNA sequencing data processing and downstream analyses were performed in R (v4.5.1). Raw 10× Genomics output files (features.tsv, barcodes.tsv, and matrix.mtx) were read as sparse count matrices using Matrix::readMM. Gene IDs were stripped of Ensembl version numbers and mapped to gene symbols using AnnotationDbi::mapIds (org.Hs.eg.db), with unmapped IDs retained, and duplicate gene names made unique via make.unique. DropletUtils::barcodeRanks was applied to generate unique molecular identifier (UMI)–rank curves, and barcodes below the knee threshold were removed to exclude empty droplets and low-UMI cells. Filtered matrices were converted to Seurat objects using CreateSeuratObject, and the percentage of mitochondrial genes per cell was calculated. Quality control criteria included removal of cells with <500 or >5000 detected genes, mitochondrial gene content >7%, extremely high nCount_RNA above the 93rd percentile of each sample, and low-abundance genes expressed in fewer than 5 cells. Doublets were identified by converting Seurat objects to SingleCellExperiment (v1.30.1) and running scDblFinder (v1.22.0), retaining only cells annotated as singlets.

Normalization and integration were performed using the Seurat SCTransform (SCT) workflow. SCT was applied to each sample individually to generate SCT assays. Highly variable features were selected via SelectIntegrationFeatures(nfeatures = 2000), preprocessed with PrepSCTIntegration, and sample-specific principal components computed using RunPCA. Integration anchors were identified using reciprocal principal component analysis (RPCA)–based FindIntegrationAnchors(…, reduction = “rpca”), and the integrated expression matrix was obtained with IntegrateData(). Dimensionality reduction was performed on the integrated data using RunPCA followed by RunUMAP(dims = 1:12). Clustering analysis was performed using Seurat with a resolution of 0.4 for the identification of major cell populations. Major cell lineages were defined and tracked using well-established canonical immunology markers: CD3D and TRAC for T cells; CD79A, MS4A1, and CD19 for B cells; NCAM1, NKG7, and GNLY for natural killer (NK) cells; CD14, FCGR3A, and LYZ for monocytes; and ITGAX, HLA-DRA, and CD74 for dendritic cells. Multiple clustering resolutions were systematically evaluated, and the final thresholds were selected based on the detection of distinct, statistically significant marker signatures for each subcluster, ensuring clear lineage segregation while avoiding over-clustering. The resolutions utilized were 0.35 for T cells, 0.4 for B cells, 0.5 for NK cells, 0.4 for monocytes, and 0.2 for dendritic cells. Samples were grouped by source (ATB, DTB, and healthy controls), and stacked barplots were used to visualize the cellular composition of each cluster across groups. Cluster marker genes were identified using FindAllMarkers(only.pos = TRUE, min.pct = 0.25, logfc.threshold = 0.25), ranked by average log₂ fold change, and the top 10 genes per cluster were selected for functional annotation and visualized using expression heatmaps, with hierarchical clustering applied to rows to group similar transcriptomic profiles across the uniform manifold approximation and projection (UMAP) clusters. All intermediate and final Seurat objects are available from the corresponding author upon request to ensure reproducibility.

Gene set variation analysis (GSVA)

To investigate pathway activity differences among cell clusters, GSVA was performed using R (version 4.5.1). First, the average gene expression per cluster was calculated from the “SCT” assay of the integrated Seurat object using AverageExpression() and converted to a double-precision matrix for computational stability. The hallmark gene set (MSigDB v2025.1; file: h.all.v2025.1.Hs.symbols.gmt) was imported using GSEABase::getGmt() and gene lists were extracted with geneIds(). GSVA parameters were set using GSVA::gsvaParam() with a minimum gene set size of 10 and a maximum of 500. GSVA scores were computed using gsva(), followed by row-wise z-score normalization. The prefix “HALLMARK_” was removed from pathway names, and hierarchical clustering heatmaps were generated using pheatmap (row distance: correlation; column distance: Euclidean; clustering method: complete linkage). For each pathway, the cluster with the highest z-score was identified using which.max(), and cluster-specific enriched pathways were summarized and exported as an Excel file (Cluster_HighPathways.xlsx) using openxlsx.

Key R packages and versions used were as follows: ‘GSVA 1.44.3,’ ‘GSEABase 1.70.1,’ ‘pheatmap 1.0.13,’ ‘Seurat 5.3.0,’ ‘openxlsx 4.2.8,’ ‘BiocGenerics 0.54.0,’ and ‘Biobase 2.68.0.’

Cell communication analysis

Cell communication analysis was performed using the R package ‘CellChat’ (version 1.6.1). ⁷ The CellChatDB human was used for analysis. All three groups of samples were normalized together, and then each group was extracted, analyzed, and compared in parallel.

Statistical analysis

Differentially expressed marker genes for heatmap visualization were identified using a nonparametric Wilcoxon rank-sum test, filtering for genes with a minimum expression fraction (min.pct) >0.1 and an average log₂ fold change (logfc.threshold) >0.25. An independent two-sample, two-tailed Student's t-test with equal variance assumed was employed to rigorously evaluate the proportional variations and statistical differences of distinct cell subclusters across individual samples between groups. Multiple testing was adjusted using the Bonferroni method, with significance defined as an adjusted p value <0.05. For GSVA, pathway enrichment scores were calculated using the official human Hallmark gene sets from the Molecular Signatures Database (MSigDB). Significant pathways among the healthy control, ATB, and DTB groups were identified using linear models via the ‘limma’ package (v3.64.3) (moderated t-test), with the significance threshold set at a false discovery rate (FDR) <0.05. For CellChat intercellular communication analysis, interaction probabilities were computed by integrating ligand–receptor expression profiles. Significant signaling pathways were determined using a permutation test, where interactions were retained only if the permutation p value was <0.05, followed by quantitative group-wise comparison of communication information strength.

Results

Peripheral immune cell profiling by scRNA-seq

We analyzed PBMCs from nine individuals—three healthy controls, three patients with ATB, and three patients with DTB—yielding 98,400 single cells. After filtering out empty droplets, doublets, and low-quality cells, 73,226 high-quality cells were retained for downstream analysis. The postfiltering cell counts for each participant were distributed as follows: healthy controls (H1: 7486; H2: 7571; and H3: 3901), patients with ATB (ATB1: 8436; ATB2: 9607; and ATB3: 8978), and patients with DTB (DTB1: 7437; DTB2: 10466; and DTB3: 9344).

As visualized in the UMAP plots (Figure 1(a)), cells from the three groups were well integrated and substantially overlapped with each other. Unsupervised clustering identified 16 transcriptionally distinct clusters, which were broadly intermixed across samples, indicating successful batch correction (Figure 1(b)). Minor sample-specific enrichments likely reflected biological variation rather than technical effects. Using canonical lineage markers, clusters were annotated into five major immune lineages: T cells (clusters 0, 2, 3, 4, 6, 9, and 12), B cells (clusters 7 and 14), monocytes (clusters 1, 10, and 15), dendritic cells (clusters 8 and 13), and NK cells (cluster 5) (Figure 1(b)). The identity of these six major lineages was further validated by the expression of canonical markers (Figure 1(c)). To ensure high data stringency, cluster 11—a minor population comprising less than 1.5% of total cells that exhibited an ambiguous transcriptomic profile and lacked canonical lineage-defining markers—was classified as a low-quality technical artifact and excluded from all downstream immune analyses.

Figure 1.

Dynamic shifts in immune cell composition during TB progression revealed by scRNA-seq. (a) UMAP embedding of the integrated single-cell transcriptomes, colored by sample group; (b) 16 transcriptionally distinct clusters were delineated and assigned to 6 overarching immune lineages; (C) representative marker genes used to define each immune lineage with pairwise statistical comparisons; (D) box plots showing the distribution of the major immune lineages across H, ATB, and DTB samples.

TB: tuberculosis; H: healthy control; ATB: active tuberculosis; DTB: disseminated tuberculosis; UMAP: uniform manifold approximation and projection.

Overall major lineage proportions were broadly stable and showed comparable distributions across the three clinical groups (Figure 1(d)). T cells dominated the PBMC compartment (approximately 60%), whereas other immune cells constituted a smaller proportion. Rigorous statistical testing of individual patient proportions revealed no statistically significant differences in global cell lineage composition among the healthy, ATB, and DTB groups (Figure 1(d)). Although the pooled mean values showed an apparent stepwise downward trend for NK cells, this shift did not reach statistical significance because of notable inter-individual variation within the cohorts. Similarly, the minor visual increases in mean proportions observed for monocytes, B cells, and T cells in both TB-infected states reflected baseline clinical heterogeneity rather than statistically robust compositional remodeling.

Progressive NK cell exhaustion and the emergence of functionally divergent subsets during TB progression

Secondary clustering identified eight transcriptionally and functionally distinct NK cell subsets (Figure 2(a) and (b)). Among these, the N0 subset showed the most dramatic compositional shift across disease states (Figure 2(c)), accounting for 42.5% of NK cells in healthy controls, decreasing to 20.9% in ATB, and declining further to 5.8% in DTB. This pattern reflects severe depletion of the N0 population in advanced TB. In contrast, most other NK cell subsets increased in frequency in ATB and DTB, suggesting that the overall NK cell contraction in PBMCs is largely driven by the peripheral decline of the N0 subset.

Figure 2.

Functional reprogramming of NK cells across TB disease states. (a) UMAP projection highlighting NK-related populations stratified by disease group; (b) expression patterns of canonical NK markers (NCAM1, NKG7, and GNLY) across subsets; (c) box plots showing the frequencies of individual NK cell subsets across H, ATB, and DTB with pairwise statistical comparisons; (D) heatmap displaying major differentially expressed genes among NK cell subpopulations; (e) GSVA-based enrichment of pathways distinguishing NK cell subsets.

GSVA: gene set variation analysis; NK: natural killer; TB: tuberculosis; H: healthy control; ATB: active tuberculosis; DTB: disseminated tuberculosis; GSVA: gene set variation analysis; UMAP: uniform manifold approximation and projection.

N0 cells were characterized by elevated expression of KIF13A, TLE1, ARPC2, and LINGO2 (Figure 2(d)) and strong enrichment for TGF_BETA_SIGNALING, MITOTIC_SPINDLE, UV_RESPONSE_DN, PROTEIN_SECRETION, and HEME_METABOLISM pathways (Figure 2(e)). These transcriptional signatures align with a transforming growth factor-beta (TGF-β)–driven regulatory NK phenotype, with enhanced metabolic and secretory programs.^8–10 The progressive peripheral attrition of this subset suggest functional attrition of circulating NK cells under chronic inflammatory and immunosuppressive conditions.^11–13

Systemic depletion of a homeostatic monocyte subset during TB progression

Monocytes were partitioned into eight distinct subsets through secondary unsupervised clustering (Figure 3(a)). One subset was transcriptionally separated from the others due to expression of CD16 (FCGR3A), whereas the remaining subsets exhibited a CD14⁺ phenotype (Figure 3(b)). All subsets expressed the universal monocyte marker LYZ. Their distributions varied substantially across disease states, reflecting clear trends alongside individual donor variability.

Figure 3.

Diminished steady-state monocytes in patients with DTB. (a) UMAP-derived visualization of monocyte subclusters; (b) expression profiles of key monocyte-defining markers within each subcluster; (c) box plots showing the frequencies of individual monocyte subsets across H, ATB, and DTB with pairwise statistical comparisons; (d) leading differentially expressed genes characterizing each monocyte subset; (e) GSVA pathway signatures highlighting functional alterations across monocyte subsets.

GSVA: gene set variation analysis; TB: tuberculosis; DTB: disseminated tuberculosis; H: healthy control; ATB: active tuberculosis; UMAP: uniform manifold approximation and projection.

The most striking population was M0, which predominated in healthy controls but decreased sharply in ATB and was nearly undetectable in peripheral circulation in DTB (Figure 3(c)). To rigorously evaluate this attrition phenomenon, we analyzed M0 frequencies across individual study participants and performed statistical comparisons across group cohorts (Figure 3(c)). This analysis confirmed a progressive, stepwise pattern of depletion; despite the expected inter-individual variations within human clinical samples, the statistical metrics directly substantiated the systematic loss of this cell state from the peripheral blood during advanced disease progression (Figure 3(c)). This subset expressed high levels of H1–10, a linker histone variant, and MIDEAS, a chromatin regulatory factor, consistent with an epigenetically stabilized, transcriptionally differentiated state (Figure 3(d)). ¹⁴

Functional enrichment analysis suggested that M0 functions as a systemic homeostatic “guardian” monocyte subset by deploying distinct tissue-protective and patrolling modules (Figure 3(e)). It was strongly enriched for heme-metabolism pathways, which mediate the critical detoxification of pro-oxidant heme species released during infection-induced tissue damage, thereby safeguarding the vascular endothelium from oxidative stress and supporting systemic anti-inflammatory and antioxidant responses. ¹⁵ M0 also exhibited enrichment of mitotic spindle–associated gene sets and robust expression of dynein/microtubule-related genes, including DNAAF9. Rather than indicating proliferative activity, these signatures signify a specialized cytoskeletal organization essential for rapid structural remodeling, continuous endothelial patrolling, polarity changes, migration, and highly efficient phagocytic competence to clear systemic debris. ¹⁶ Additionally, enrichment of lipid-metabolism and adipogenesis programs suggests involvement in systemic metabolic homeostasis and anti-inflammatory myeloid functions that likely act as a metabolic brake against systemic hyperinflammation. ¹⁷

Together, these findings demonstrate that TB pathogenesis involves not only the induction of inflammatory myeloid programs but also the marked peripheral depletion of a homeostatic “guardian” monocyte population, potentially representing a mechanistic tipping point enabling progression from controlled or latent infection to disseminated disease.

Identification and characterization of a dynamically altered B-cell subset during TB progression

B cells are crucial for limiting M. tuberculosis replication and shaping adaptive immunity. ¹⁸ Secondary clustering identified seven B-cell subsets (Figure 4(a)), all of which uniformly expressed classical B-cell markers CD19, MS4A1, and CD79A (Figure 4(b)). Among these, the B0 subset exhibited the most pronounced compositional change, increasing from healthy controls to ATB but nearly disappearing from peripheral blood in DTB (Figure 4(c)).

Figure 4.

Metabolically active B-cell subsets undergo biphasic modulation with TB progression. (a) UMAP map illustrating transcriptionally distinct B-cell subsets; (b) distribution of hallmark B-cell markers across the identified subpopulations; (c) box plots showing the frequencies of individual B-cell subsets across H, ATB, and DTB with pairwise statistical comparisons; (d) prominent differentially expressed genes associated with individual B-cell subsets; (e) GSVA pathway enrichment patterns revealing functional divergences among B-cell subsets.

TB: tuberculosis; H: healthy control; ATB: active tuberculosis; DTB: disseminated tuberculosis; GSVA: gene set variation analysis; UMAP: uniform manifold approximation and projection.

Differential expression analysis revealed that B0 expressed a suite of genes associated with immune regulation and metabolic adaptation (Figure 4(d)). These included ZBTB16, associated with atypical memory B-cell phenotypes and immune regulation; ¹⁹ FKBP5 and NR3C2, mediators of glucocorticoid and stress hormone signaling;^20,21 IFNG-AS1, linked to IFN-γ–regulated responses; ²² CARMIL1, PALM2-AKAP2, and CEMIP2, involved in cytoskeletal remodeling and cell migration;^23–25 and ST6GALNAC3 and LARGE1, associated with glycosylation and secretory processes.^26,27

GSVA analysis showed significant enrichment for UV_RESPONSE_DN, ADIPOGENESIS, MITOTIC_SPINDLE, and PROTEIN_SECRETION hallmark pathways (Figure 4(e)), which are associated with stress responses, lipid metabolic remodeling, and enhanced secretory activity. ²⁸ Notably, subset 3, which was also markedly reduced in DTB, shared similar pathway features with B0.

Collectively, these findings suggest that B0 represents a metabolically active and stress-adapted B-cell population that undergoes significant functional remodeling in disseminated TB.

Accumulation of naïve T cells and exhaustion of effector T cells in DTB

Secondary clustering of T cells identified 10 transcriptionally distinct subsets (Figure 5(a) and (b)). Although the proportions of most subsets remained relatively stable or exhibited high inter-individual variation without statistical significance, the T0 subset demonstrated a clear progressive expansion from the healthy baseline to the advanced disease states. Importantly, statistical analysis revealed that this enrichment was significantly distinct exclusively when comparing healthy controls directly with the DTB group. In contrast, the baseline changes between healthy controls and ATB and the further accretion between ATB and DTB did not reach statistical significance, establishing that a verified, statistically robust expansion of the T0 subset is a hallmark restricted to severe systemic dissemination (Figure 5(c)).

Figure 5.

T-cell exhaustion and naïve T-cell expansion characterize late-stage TB (DTB). (a) UMAP clustering of T-cell subsets across all samples; (b) expression levels of representative T-cell lineage and functional markers across subsets; (c) box plots showing the frequencies of individual T-cell subsets across H, ATB, and DTB with pairwise statistical comparisons; (d) key differentially expressed genes distinguishing each T-cell subset; (e) GSVA analysis outlining pathway shifts associated with T-cell subset dysfunction and activation state.

TB: tuberculosis; DTB: disseminated tuberculosis; H: healthy control; ATB: active tuberculosis; GSVA: gene set variation analysis; UMAP: uniform manifold approximation and projection.

T0 cells were marked by increased expression of genes associated with functional attenuation, including UBASH3B, and RNF175, which mediate ubiquitin-dependent negative regulation of T-cell receptor signaling (TCR) signaling; ²⁹ MAL and RAB30, which regulate membrane trafficking and receptor localization; ³⁰ and ATG10, a regulator of autophagy linked to T-cell stress and impaired effector function. ³¹ Upregulation of metabolic and glycosylation-related genes such as GDP-mannose 4,6-dehydratase (GMDS) further supports adaptive metabolic remodeling in response to chronic antigen exposure (Figure 5(d)). ³²

GSVA analysis revealed enrichment for ALLOGRAFT_REJECTION, COAGULATION, ESTROGEN_RESPONSE_EARLY, IL6_JAK_STAT3, and KRAS pathways (Figure 5(e)), reflecting sustained immune activation and systemic inflammatory stress. ³³ Although classical exhaustion markers such as PDCD1, LAG3, and HAVCR2 were not upregulated, the transcriptional profile indicates a noncanonical exhausted or functionally impaired CD8⁺ T-cell state. ³⁴ These cells likely experience exhaustion via mechanisms involving post-transcriptional regulation, autophagy, and altered signaling and trafficking rather than through the classical inhibitory-receptor pathway. ³⁴

Inflammatory remodeling of circulating dendritic cells during TB

Single-cell analysis identified five dendritic cell (DC) subpopulations in peripheral blood (Figure 6(a) and (b)). Two subsets, D0 and D1, demonstrated opposing compositional and functional dynamics across disease stages (Figure 6(c)). D0 comprised a small fraction of DCs in healthy controls, expanded to approximately half of DCs in ATB, and became dominant in DTB. Conversely, D1 was abundant in healthy controls but declined sharply in ATB and DTB.

Figure 6.

Disrupted DC homeostasis following TB progression. (a) UMAP visualization showing transcriptionally defined DC subsets; (b) expression of canonical DC markers across the identified subsets; (c) box plots showing the frequencies of individual DC subsets across H, ATB, and DTB with pairwise statistical comparisons; (d) major differentially expressed genes characterizing DC subtypes; (e) GSVA-derived pathway enrichment signatures indicating altered DC functionality.

DC: dendritic cell; GSVA: gene set variation analysis; TB: tuberculosis; H: healthy control; ATB: active tuberculosis; DTB: disseminated tuberculosis; UMAP: uniform manifold approximation and projection.

D0 cells expressed numerous genes related to phagocytosis, cytoskeletal remodeling, and inflammatory activation (Figure 6(d)) and were enriched for IL6/JAK/STAT3, TNFA/NF-κB, and interferon-response pathways (Figure 6(e)), consistent with a highly inflammatory DC phenotype.^35–37 In contrast, D1 cells showed weaker inflammatory signatures but upregulated genes involved in autophagy and antigen processing, such as GABARAPL1 and ISG15.^38,39 Their enrichment for metabolic and TGF-β signaling pathways suggests a homeostatic DC state important for maintaining antigen processing and peripheral immune equilibrium.

Together, these observations demonstrate profound DC remodeling during TB progression, marked by expansion of inflammatory DCs (D0) and peripheral depletion of homeostatic DCs (D1), indicating that circulating DCs shift toward inflammation rather than effective antigen presentation during systemic mycobacterial infection.

Severe defects in canonical signaling pathways required for naïve CD8⁺ T-cell activation

Given the extensive immune dysregulation observed in TB, we next examined intercellular communication among immune subsets. We reannotated the first-round clusters into 10 functional immune cell subtypes (Figure 7(a)). To evaluate whether downstream signaling dynamics were driven by gross alterations in cell abundance, we quantified the baseline proportions of these 10 refined subsets across the disease states. Rigorous statistical testing revealed that the relative composition of all 10 functional lineages remained highly stable across cohorts, with no statistically significant differences detected in any pairwise comparison. This notable lack of compositional divergence indicates that the dysregulated intercellular communication networks detailed below are primarily driven by cell-intrinsic transcriptional rewiring rather than simple shifts in underlying cell-type frequencies (Figure 7(b)).

Figure 7.

Impaired T-cell activation linked to weakened DC–T-cell communication. (a and b) Reclustering of the integrated dataset to resolve major immune lineages and comparison of their proportional distribution in H, ATB, and DTB groups; (c) chord diagrams illustrating the overall inter-lineage communication networks across the three disease states; (d) scatterplots depicting incoming and outgoing interaction strengths for each major immune subset in different groups; (e) bubble plot summarizing ligand–receptor–mediated communication signals directed toward naïve CD8⁺ T cells.

DC: dendritic cell; H: healthy control; ATB: active tuberculosis; DTB: disseminated tuberculosis.

Using CellChat, we reconstructed intercellular communication networks among these subsets (Figure 7(c)). Quantitative analysis revealed that almost all subsets exhibited attenuated communication strength in ATB and DTB (Figure 7(d)). Despite their increased numbers, naïve CD8⁺ T cells lost both incoming and outgoing signaling interactions in infected groups, suggesting a functionally quiescent or isolated state within the immune network.

To pinpoint pathways contributing to this communication deficit, we examined the canonical activation signals received by naïve CD8⁺ T cells (Figure 7(e)). In healthy controls, these cells received complete activation inputs from DCs, including Signal 1 (HLA–CD8B) for antigen presentation and Signal 2 (CD86–CD28) for costimulation.^40,41 In ATB, major histocompatibility complex 1 (MHC-I)–mediated antigen presentation (Signal 1) was completely lost, although costimulation remained intact. Strikingly, both activation signals were absent in DTB. In addition to losing activation cues, naïve CD8⁺ T cells in ATB and DTB received alternative inhibitory inputs, including LGALS9–CD45, LGALS9–CD44,^42,43 and CD99–CD99 signaling. ⁴⁴

Together, these findings strongly indicate that circulating naïve CD8⁺ T cells exhibit a profound deficit in their systemic activation potential during TB progression. This is evidenced by the systemic loss of essential molecular hardware required for antigen presentation and costimulation in the peripheral pool, compounded by the gain of alternative immunosuppressive cues, which collectively preprogram or sustain their functionally quiescent state before transitioning into lymphoid tissues.

Discussion

This study presents a comprehensive single-cell transcriptional atlas of peripheral immune remodeling across the tuberculosis disease spectrum, from health to active and disseminated infection. By integrating changes in cellular composition, gene expression, and signaling pathway activity, our data outline a continuum of immune dysregulation characterized by depletion of homeostatic subsets and expansion of stress-adapted, functionally impaired populations. These findings suggest that TB pathogenesis reflects not only immune activation but also a progressive loss of regulatory balance and coordinated breakdown of innate–adaptive communication, which was not addressed by previous reports.^6,45 The resulting landscape is an atypical feature of chronic inflammatory stress, as shown in other chronic infectious diseases.^46,47

Innate immunity has long been implicated TB.^48,49 The frequencies of most major immune cell types were not significantly affected during TB progression. However, the frequency of NK cells progressively diminished during TB progression, consistent with previous publications.^49,50 Our further in-depth analysis identified a dominant NK population (N0) in healthy individuals that progressively contracted in ATB and DTB. This population is enriched for TGF-β signaling and heme metabolism, and these pathways have been linked to regulatory and metabolically poised NK states, suggesting that they represent a homeostatic, tissue-surveillance–type NK pool rather than a terminal effector subset. ⁵¹

Previous work has established that human monocytes are heterogeneous and that the balance of classical, intermediate, and nonclassical subsets shifts in TB. ⁵² We identified a homeostatic monocyte subset (M0) that is present at high frequency in healthy PBMCs but diminishes in patients with ATB. Phenotypically, this subset is CD14⁺/LYZ⁺; lacks CD16; and exhibits a transcriptional program marked by high H1–10 and MIDEAS expression, heme-metabolism signatures, dynamic microtubule/dynein gene expression, and lipid-metabolism/adipogenesis pathways. These features support the concept that M0 represents a highly differentiated, epigenetically stabilized “guardian” monocyte population that contributes to systemic immune and metabolic homeostasis.^53–56 Taken together, these observations suggest that TB pathogenesis involves not only induction of proinflammatory monocyte programs, as widely reported, ⁴⁸ but also catastrophic depletion of a chromatin-stabilized, heme- and lipid-metabolism–competent homeostatic monocyte compartment. From a translational perspective, quantitative loss of M0 could serve as a blood-based biomarker of TB progression risk, potentially complementing existing transcriptomic signatures that largely capture interferon and inflammatory modules, ⁵⁷ and might help identify patients at imminent risk of dissemination or poor outcomes.

B-cell involvement in TB is not as well understood as that of other immune cells. ⁵⁸ However, a B-cell-deficient murine TB model showed that M. tuberculosis was more virulent and difficult to clear than wild type (WT) mice, indicating that B cells play an essential role in TB. ⁵⁹ Our results identify a B-cell subset (B0) that has heightened metabolic/stress-adaptation capacity and secretory dynamics, dramatically increases from healthy controls to active TB but nearly vanishes in disseminated TB. This dynamic transition echoes conflicting reports about B-cell frequency in TB, with studies reporting increases, decreases, or no change.⁵⁹

Profound transcriptional remodeling was observed in the T-cell compartment. T cells in ATB and DTB were enriched with multiple negative regulators of TCR signaling, such as UBASH3B, MAL, RAB30, and RNF17. Although these factors are not canonical T-cell negative regulators, such as PD-1, TIM-3, and LAG-3, which have been implicated in TB development, ⁶⁰ they might represent noncanonical T-cell exhaustion regulators. Indeed, a previous study has described a heterogeneous exhausted CD8⁺ T-cell population in DTB, some of which strongly express exhaustion signatures but others with more limited inhibitory-receptor expression but impaired function as well as altered effector states associated with protection or dissemination. ⁶ Therefore, our findings are consistent with this emerging picture of exhaustion heterogeneity but extend it by defining a specific CD8⁺ T-cell subset in the peripheral blood of patients with DTB that appears functionally attenuated despite lacking robust expression of canonical exhaustion markers.

A large body of previous work has established that DCs are deeply involved in TB pathogenesis, demonstrating reduced DC numbers in blood and accumulation of DCs in granulomas, ⁶¹ highly proinflammatory DC states in in vitro–infected monocyte-derived DCs ⁶² and phenotypic alterations that differ between latent and active TB. ⁶³ These findings collectively suggest that TB skews DCs toward dysfunctional inflammatory phenotypes, but most prior studies have focused on bulk DC populations or in vitro models rather than in vivo human DC subset remodeling in TB. Our single-cell analysis adds resolution to this picture by revealing a reciprocal shift between two circulating DC compartments, “inflammatory” and “homeostatic/antigen processing.” Inflammation-oriented DCs expand, whereas autophagy/TGF-β-associated, metabolically biased DCs contract with progression to disseminated TB, suggesting that TB disrupts a fundamental balance between inflammatory and homeostatic DC circuits in the blood.

Beyond individual immune cell phenotyping, we also emphasized immune cell communication in this study. Our data clearly demonstrated broadly weakened cell–cell communication across all major immune cell types. We focused on T-cell–DC communication and found that naïve CD8⁺ T cells displayed both attenuated incoming and outgoing signals in ATB and DTB. Our further analysis revealed progressive defects in the canonical “Signal 1” and “Signal 2” between DCs and naïve CD8⁺ T cells in advanced TB. This finding is consistent with previous work showing that M. tuberculosis infection can delay antigen processing, MHC loading, and T-cell priming, leading to slow and quantitatively suboptimal effector generation. ⁶⁴ Furthermore, naïve CD8⁺ T cells also received alternative inhibitory circuits featuring LGALS9–CD45/CD44 and CD99–CD99 pathways. LGALS9 has been implicated as an important glyco-immune checkpoint in TB and HIV/TB coinfection, with elevated circulating levels in extrapulmonary TB and experimental data showing that galectins, including LGALS9, modulate host responses to M. tuberculosis. ⁶⁵ Likewise, CD99 has emerged as a key regulator of T-cell activation, immune synapse formation, and leukocyte trafficking via homophilic CD99–CD99 interactions. ⁶⁶

Overall, this study provides a comprehensive single-cell transcriptional atlas of peripheral immune remodeling across the TB disease spectrum, spanning healthy, active, and disseminated infection. Future longitudinal and spatially resolved investigations incorporating proteomic, epigenomic, and functional approaches will be crucial to validate these cellular dynamics, clarify causal mechanisms, and evaluate the clinical applicability of the identified biomarkers and therapeutic targets.

Conclusion

In summary, the progression from ATB to DTB is driven by profound systemic peripheral immune attrition. This process is characterized by the attrition of homeostatic subsets, including “guardian” monocytes, NK cells, and stress-adapted B cells, alongside the expansion of inflammatory DCs and functionally exhausted T cells. Circulating naïve CD8⁺ T cells show a dramatic deficit in their systemic activation potential during TB dissemination.

Nevertheless, several critical limitations inherent to this study must be acknowledged. First, due to the extreme rarity and logistical complexities of capturing clinical specimens from patients with disseminated manifestations, this study utilized a limited sample size of three donors per group. Consequently, this cohort size is insufficient to fully capture broader human population variance and clinical heterogeneity. Second, our findings remain exploratory and preliminary, as the current study lacks an independent cohort for external validation. Finally, although our high-resolution single-cell atlas provides a robust foundational blueprint, these transcriptomic patterns have not yet been cross-verified using orthogonal, multimethod validation, such as flow cytometry. Therefore, these insights should be interpreted as an introductory framework until validated in larger, multicenter cohorts using multimethod approaches.