Abstract

Objective

   Despite advances in understanding systemic immune responses to
   Mycobacterium tuberculosis (Mtb), the localized immune dynamics within
   infected lymph nodes, particularly cell-type-specific transcriptional
   reprogramming and intercellular crosstalk, remain poorly defined,
   impeding targeted therapy development. By means of single-cell
   transcriptomics, the objective was to dissect the immune
   microenvironment and map intercellular crosstalk in Mtb-infected
   cervical lymph nodes, with the purpose of uncovering the mechanisms of
   localized immunity and immunopathology.

Methods

   Paired Mtb-positive cervical swollen lymph nodes (SLN) and adjacent
   Mtb-negative normal-appearing lymph nodes (NLN) from five cervical LNTB
   patients were analyzed using single-cell RNA sequencing (scRNA-seq).
   Computational network modeling (CellChat) and flow cytometry validation
   were employed to map immune cell heterogeneity and cell–cell
   communication.

Results

   ScRNA-seq identified ten T cell subsets, ten B cell subsets, and six
   myeloid subsets, revealing conserved frequencies but substantial
   transcriptional reprogramming in SLN. The SLN exhibited extensive
   upregulation of pro-inflammatory pathways across T, B, and myeloid
   cells, accompanied by minimal alterations in subset frequencies.
   IL1B + macrophages in SLN showed an enrichment of genes associated with
   oxidative phosphorylation, antigen presentation, and
   inflammasome-related genes. The SLN demonstrated an increased cell–cell
   communication driven by the crosstalk between macrophage and
   CD8 + T/NKT cells. Validation through flow cytometry confirmed
   comparable proportions of immune subsets between the SLN and the NLN,
   which was consistent with the findings of scRNA-seq.

Conclusions

   This study delineates a spatially coordinated immune strategy in
   cervical LNTB. In this context, Mtb infection induces transcriptional
   and metabolic reprogramming rather than subset redistribution. The
   strengthened interaction between macrophages and T cells emphasizes
   that cellular immunity serves as the main impetus for bacterial
   containment. Nevertheless, there are trade-offs between inflammation
   and tissue integrity. These insights provide a framework for developing
   therapies that target intercellular networks to achieve a balance
   between immunity and pathology in LNTB.

Supplementary Information

   The online version contains supplementary material available at
   10.1186/s12865-025-00763-y.

   Keywords: Cervical Lymph Node Tuberculosis, Single-cell RNA sequencing,
   Transcriptional reprogramming, Cell–Cell communication

Introduction

   Globally, it is estimated that one-third of people are infected with
   Mycobacterium tuberculosis (Mtb), the agent responsible for
   tuberculosis (TB). Generally, the lungs are the most common site of
   infection and disease, but extrapulmonary tuberculosis (EPTB) accounted
   for 16% of notified TB cases in 2020, as reported by the World Health
   Organization (WHO) Global Tuberculosis Report [[42]1]. In clinical
   practice, EPTB at atypical sites is frequently diagnosed
   incidentally[[43]2]. The challenges in diagnosis arise from nonspecific
   clinical manifestations and the frequent absence of systemic symptoms,
   which often lead to delayed recognition and the initiation of treatment
   [[44]3, [45]4]. Notably, EPTB demonstrates particular predilection for
   immunocompromised populations including patients with diabetes or HIV
   infection as well as vulnerable age groups [[46]5]. Of the EPTB,
   cervical lymph node TB (LNTB) is one of frequent forms [[47]6].
   However, they paradoxically also harbor Mtb replication and elimination
   [[48]7–[49]11]. Despite advances in understanding systemic immune
   responses to Mtb, the localized immune dynamics within infected lymph
   nodes (LNs), particularly the interplay between cellular subsets,
   remain poorly defined. This knowledge gap impedes the development of
   targeted therapies aimed at enhancing bacterial clearance while
   mitigating immunopathology.

   Conventional studies of LNTB have largely focused on bulk
   transcriptional profiling or immunohistochemical characterization,
   revealing upregulated inflammatory pathways and altered lymphocyte
   frequencies in infected nodes [[50]12–[51]15].

   For instance, Elevated type 1 and 17 cytokines and reduced IL-1β at
   infection sites suggest a skewed immune milieu favoring bacterial
   persistence [[52]16]. However, these approaches lack resolution to
   dissect cell-type-specific responses or intercellular crosstalk, which
   is a critical driver of immune coordination. Single-cell RNA sequencing
   (scRNA-seq) has revolutionized immunology by enabling deconvolution of
   heterogeneous tissues into functionally distinct subsets [[53]17].
   However, to date, no study has systematically mapped cell–cell
   communication networks in human LNTB.

   Cell–cell communication mediated by ligand-receptor interactions
   orchestrates immune cells recruitment, activation, and effector
   functions [[54]18]. In TB, dendritic cells (DCs) and macrophages
   deliver Mtb antigens to LNs, priming T cell responses [[55]11], while
   the cytokine cross-talk between myeloid and lymphoid cells determines
   the outcomes of infection. For example, IFN-γ secreted by CD4 + T cells
   enhances the bactericidal activity of macrophage, while excessive IL-1β
   signaling may exacerbate tissue damage. Despite its pivotal role, the
   spatial and functional organization of these interactions in
   Mtb-infected LNs remains poorly understood. Do infected nodes exhibit
   unique signaling hubs that differentiate them from adjacent normal
   nodes? How do specific immune subsets, such as macrophages and CD8 + T
   cells, collaborate to contain Mtb? Addressing these questions requires
   a systems-level analysis of intercellular communication.

   Here, we combined scRNA-seq with computational network modeling to
   comprehensively map the immune landscape and cell–cell communication
   dynamics in paired swollen lymph nodes (SLN) and normal-appearing lymph
   nodes (NLN) from cervical LNTB patients. We identified 30 subsets of
   immune cell and revealed extensive transcriptional reprogramming in
   SLN. Significantly, CellChat analysis uncovered an enhanced global
   interaction strength in SLN, which was driven by the crosstalk between
   macrophage and CD8 + T cell/NKT cells through the IFN-II and IL-1
   pathways. These findings not only delineate the cellular basis of
   localized immune responses in LNTB but also provide a model for
   targeting intercellular networks to improve TB control.

Methods and materials

Isolation of cervical LNs for scRNA-seq

   Five patients diagnosed with cervical LNTB were recruited based on
   clinical manifestations and microbiological confirmation of Mtb
   infection through either culture or Xpert MTB/RIF assay (Cepheid, USA).
   Exclusion criteria included HIV co-infection and prior exposure to
   anti-TB therapy, ensuring no confounding effects from immunocompromised
   status or drug interventions. The detailed demographic and clinical
   characteristics of the cohort are presented in Supplementary Table 1.
   During surgical resection, paired samples were collected from each
   patient: (i) one cervical SLN exhibiting pathological features of LNTB,
   and (ii) one or two adjacent NLN serving as internal controls. NLN were
   defined as nodes meeting the following criteria: (1) normal anatomical
   size (bean-shaped, < 1 cm in diameter), (2) absence of gross
   pathological features (necrosis, suppuration or granuloma formation),
   and (3) microbiological confirmation of Mtb negativity via Xpert
   MTB/RIF assay. This sampling strategy resulted in a total cohort of 5
   SLN and 8 NLN specimens for downstream analyses. Immediately
   post-excision, tissues were transferred to sterile petri dishes and
   gently washed with ice-cold phosphate-buffered saline (PBS) (Solarbio,
   Beijing, China) to remove peripheral blood contamination.

Preparation of single-cell suspensions for scRNA-seq

   LN samples were cut into approximately 1 mm^3 pieces and immersed in
   RPMI-1640 medium (Thermo Fisher Scientific, Massachusetts, USA)
   Supplemented with 10% fetal bovine serum (FBS) (ScienCell, San Diego,
   California, USA). A portion of tissue fragments was reserved for the
   detection of Mtb through the Xpert MTB/RIF assay (Cepheid, USA), while
   the remaining tissue was Subjected to enzymatic digestion in a 37℃
   water bath for 30 min. The resultant cell Suspension was filtered
   through a 40-μm nylon cell strainer and centrifuged at 300 g for 5 min
   at 4℃. Following the removal of the supernatant removal, the pellet was
   resuspended in erythrocyte lysis buffer (Solarbio, Beijing, China) and
   incubated on ice for 5 min to Eliminate residual red blood cells. After
   being washed twice with PBS, the pellet was resuspended in PBS
   containing 2% FBS for the preparation of a single-cell RNA library.

Single-cell RNA library construction and sequencing

   According to the manufacturer's protocol, single-cell capturing and
   downstream Library constructions were carried out using a 10× Chromium
   Single Cell 3' Reagent Kit (V2 Chemistry). Briefly each cell suspension
   was loaded onto a Chromium single-cell chip, partitioning oil, reverse
   transcription (RT) reagents, and a collection of gel beads. The
   captured cells were lysed, and the RNA released was barcoded through RT
   in individual Gel Bead-In-Emulsions (GEMs). RT was performed using a
   C1000 TouchTM Thermal Cycler (Bio-Rad, California, USA). Complementary
   DNA (cDNA) was generated and amplified by PCR. The quality of amplified
   cDNA was assessed using an Agilent 4200 (performed by BGI Technology,
   Shenzhen, China). Single-cell libraries were then constructed and
   normalized based on the manufacturer's standard parameters. The
   libraries were sequenced using DNBSEQ T1 sequencer with a paired-end
   100-bp (PE100) reading strategy (performed by BGI technology, Shenzhen,
   China).

ScRNA-seq data processing and cell annotation

   The raw gene expression matrix was generated from each sample using
   Cell Ranger (v6.1.2) [[56]19]. Pipeline is coupled with human reference
   genome version GRCh38. Filtered matrices were analyzed with the Seurat
   R package (v 3.2.0) [[57]20]. Quality control was performed based on
   the following criteria: 1) Cell filtering: Cells with fewer than 200
   detected genes or > 90% of the maximum gene count per sample were
   excluded. 2) Mitochondrial read filtering: Cells in the top 15% of
   mitochondrial read ratios were removed. 3) Doublet removal: Potential
   doublets were identified and eliminated using Doublet Detection
   (v3.0; 10.5281/zenodo.2678041). Cell cycle phase was assessed via
   the CellCycleScoring function in Seurat. After quality filtering,
   expression matrices were normalized, and highly variable genes (2,000
   genes) were selected for principal component analysis (15 PCs).
   Dimensionality reduction and clustering were performed using Uniform
   Manifold approximation and projection (UMAP) and the Louvain algorithm
   (resolution = 0.01).

   Differentially expressed genes (DEGs) for each cluster were determined
   using the FindAllMarkers function with the following parameters:
   Wilcoxon rank-sum test with Bonferroni-adjusted P < 0.05; Minimum log2
   fold change (|log2FC|) > 0.25; Genes detected in ≥ 25% of cells within
   the cluster and showing a ≥ 0.2-fold difference compared to other
   clusters. To resolve finer cellular heterogeneity, major clusters were
   re-clustered at a higher resolution (resolution = 0.5). Subpopulation
   identities were annotated based on canonical marker genes.

Flow cytometry analysis

   SLN and NLN were Surgically resected from patients diagnosed with
   cervical LNTB. Single-cell Suspensions were prepared by mechanically
   dissociating tissues using a syringe plunger followed by filtration
   through a 70 μm cell filter. Residual erythrocytes were lysed using ACK
   Lysing Buffer (Thermo Fisher, CA, USA). For Flow Cytometry, Fc
   receptors were blocked using CD16/CD32 monoclonal antibody (BD,
   553,141). Cells were stained with antibodies against surface proteins
   CD45-APC-Cy7 (BD, 368,516), CD19-APC(BD, 555,415), CD14-Bv421(BD,
   325,628), CD3-PE-Cy7(BD, 557,851), CD4-Percp (BD, 652,823),
   CD8-FITC(BD, 555,643). Surface staining was performed in 1× FACS Buffer
   for 30 min at 4 °C in the dark. Cells were washed once with PBS (with
   2% FBS) and resuspend in 200 μl FACS Buffer. Stained cells were
   immediately analyzed on BD FACSymphony™A3 (BD Biosciences, CA, USA) and
   Flow-Jo software (v10.0, Tree Star, Inc.) was used for post-acquisition
   analysis. Cell population frequencies were quantified using FlowJo’s
   gating strategies. Gating strategies are shown in Supplementary
   Fig. 1A. Statistical differences between SLN and NLN were evaluated
   using the paired Wilcoxon signed-rank test. P-values were adjusted for
   the false discovery rate (FDR), and a significance threshold of
   P < 0.05 was applied.

DEGs between SLN and NLN

   DEGs from each cell type between SLN and NLN samples were identified
   using the "FindMarkers" function in Seurat with default parameters. A
   Bonferroni-adjusted P < 0.05 and |log2FC|> 0.25 was used to define
   significant DEGs.

Pathway enrichment analysis

   To determine the significantly enriched pathways, the DEGs were mapped
   to the Kyoto Encyclopedia of Genes and Genomes (KEGG) database for
   pathway enrichment analysis using the Dr.Tom network platform of BGI
   ([58]http://report.bgi.com). The calculated P-values were adjusted for
   FDR, with FDR < 0.05, as a threshold. KEGG pathways satisfying this
   condition were defined as significantly enriched pathways.

Gene set enrichment analysis (GSEA)

   To further explore the functional implications of transcriptional
   differences in cell types between the SLN and NLN groups, GSEA was
   performed. Initially, DEGs analysis was conducted between the SLN and
   NLN groups, with a threshold of absolute |log2FC|> 0.1 and
   Bonferroni-adjusted P < 0.05. GSEA was implemented using the R package
   “clusterProfiler” to identify Gene Ontology (GO) Biological Process
   terms that were enriched in either the SLN or NLN groups. The
   significance of enrichment was assessed with 1,000 permutations, and
   gene sets meeting FDR < 0.25 and P < 0.05 were considered statistically
   significant. Visualization of the enriched pathways was performed using
   dot plots and ridge line plots to highlight normalized enrichment
   scores (NES) and leading-edge genes.

Cell − cell communication analysis

   ScRNA-seq data from all LN samples were analyzed to investigate
   intercellular communications. This analysis was conducted using the R
   CellChat package version 2.1.0 [[59]21, [60]22]. The single-cell
   RNA-seq data were processed as follows: Seurat objects containing
   normalized gene expression matrices and cluster annotations for SLN and
   NLN samples were converted into CellChat objects. Ligand-receptor
   interactions were annotated using the CellChatDB database [[61]22].
   Communication probabilities between cell clusters were calculated by
   integrating the gene expression levels of ligands and receptors.
   Interactions were filtered if either the ligand- or receptor-expressing
   cluster contained fewer than 10% cells, ensuring robustness against
   low-count noise. The SLN and NLN CellChat objects were merged to enable
   a direct comparison of interaction networks. The total numbers of
   interactions and interaction strength (summed communication
   probabilities) were compared between SLN and NLN. For each cell
   cluster, differential incoming (receiver) and outgoing (sender)
   interaction strengths were calculated and visualized in two-dimensional
   space using Wilcoxon rank-sum tests (FDR < 0.05). Pathway-specific
   information flow was defined as the sum of communication probabilities
   across all ligand-receptor pairs within a pathway. Upregulated or
   downregulated ligand-receptor pairs in SLN were extracted and mapped to
   their cellular sources and targets.

Statistical analysis

   All statistical analyses in this study were performed using R (v4.3.3)
   or GraphPad Prism (version 10.0). The statistical methods and the
   associated threshold for each analysis were detailed in the preceding
   method sections.

Results

Characterization of the major cell types in cervical LNs

   To investigate the local immune response to Mtb infection in cervical
   LNTB patients, five paired SLN and NLN samples were collected and
   analyzed by scRNA-seq (Fig. [62]1A). Subsequent to quality control,
   38,968 cells, with a median of 1,311 genes per cell, were obtained from
   SLN samples and 45,735 cells, with a median of 1,203 genes per cell,
   were obtained from NLN samples (Supplementary Table 2). Unsupervised
   clustering performed using the Seurat algorithm classified all the
   analyzed cells into three major categories: T cells (CD3D + CD3E +), B
   cells (CD79A + CD79B +), and myeloid cells (CD68 + LYZ +) (Fig. [63]1B
   and C).

Fig. 1.

   [64]Fig. 1
   [65]Open in a new tab

   Characterization of total cell in cervical LNs from LNTB patients.
   A Schematic representation of the experimental workflow: Paired SLN and
   NLN samples were collected from 5 LNTB patients, followed by
   single-cell RNA sequencing (10 × Genomics) and integrative
   bioinformatic analysis. B UMAP visualization of 84,703 high-quality
   cells color-coded by annotated cell types. Major immune populations
   include T cells (red), B cells (green) and myeloid cells (blue).
   C Feature plots displaying expression gradients of lineage-defining
   markers (CD3D and CD3E for T cells, CD79A and CD79B for B cells, CD68
   and LYZ for myeloid cells). Gray-to-blue gradient indicates normalized
   expression levels. D Boxplots comparing the proportions of B, T and
   myeloid cells between the SLN and NLN samples. The x-axis represents
   the cell type, and the y-axis shows the proportion of each cell type.
   Data are presented as mean ± standard deviation, * adjust P < 0.05
   considered statistically significant, “ns” indicates non-significant
   differences

   Within the total cell population, T cells and B cells were the major
   constituents, while myeloid cells accounted for only a small
   proportion. Comparative analysis showed no significant differences in
   the relative frequencies of these three cell types between the SLN and
   NLN samples (Fig. [66]1D). To confirm these findings, we independently
   analyzed SLN and NLN samples from three additional LNTB patients using
   flow cytometry. Consistent with scRNA-seq results, no significant
   differences were observed in the proportions of T cells, B cells, or
   myeloid cells between the SLN and NLN groups (Supplementary Fig. 1A and
   B).

Identification of 10 T cell subsets in cervical LNs

   To explore the T cell Subsets in cervical LNs from LNTB patients, we
   re-clustered all the T cell cluster. As a result, we identified 11
   clusters based on the distinct expression of associated markers
   (Fig. [67]2A). While most clusters expressed canonical T cell markers
   CD3D and CD3E, Cluster 10 was identified as NK cells based on cytotoxic
   genes (GNLY, TRDC, KLRB1) and anti-inflammatory SPINK2 [[68]23]
   (Fig. [69]2B, C, Supplementary Figure S2A-B). Cluster 9 represented NKT
   cells with dual cytotoxicity (GZMA/B, NKG7). A distinct exhausted
   CD4 + T cell subset (Cluster 8), marked by immune checkpoint molecules
   (CTLA4, PDCD1) and lymphoid chemokine CXCL13 (Fig. [70]2B, C),
   indicative of chronic antigen exposure in Mtb infection. Two T
   sub-clusters exhibiting high expression of CD8A and CD8B were
   respectively categorized as naïve CD8 + T cells (CCR7 + SELL +) and
   effector CD8 + T cells (GZMA + GZMK + NKG7 +). Interestingly, six
   subsets (Clusters 1–5, 7) displayed ambiguous CD4 and CD8 expression
   profiles yet were classified into two types of naïve T cells
   (CCR7 + SELL +): XIST + naïve T cells and RPL21 + naïve T cells; three
   central memory T cells (S100A4 + GPR183 +): S100A4 + Tcm, NR4A1 + Tcm,
   and GADD45G + Tcm; and regulatory T cells (Tregs) (FOXP3 + IL2RA +)
   (Fig. [71]2B, C, Supplementary Figure S2A-B). These findings underscore
   the functional diversification of T cells in response to Mtb,
   emphasizing transcriptional reprogramming rather than changes in subset
   frequencies.

Fig. 2.

   [72]Fig. 2
   [73]Open in a new tab

   Characterization of T cell subsets between the SLN and NLN samples.
   A UMAP plot displaying T cell clusters identified across five paired
   SLN and NLN samples. The left panel represents all cell types. The
   right panel focuses on the distribution of samples, with each dot
   representing an individual cell and colors indicating different
   samples. B Dot plot of canonical marker gene expression (columns)
   across clusters (rows). Dot size reflects the percentage of expressing
   cells; color intensity denotes mean normalized expression (log2 scale).
   C Violin plots highlighting differential expression of canonically cell
   marker genes across clusters. The x-axis represents the T cell subsets,
   and the y-axis shows the normalized expression levels. D The GSEA
   analysis highlighting biological processes enriched in SLN samples
   compared to NLN samples. The x-axis represents the running enrichment
   score, and the y-axis indicates the rank in the ordered dataset.
   E Boxplots comparing the proportions of various T cell subsets between
   the SLN and NLN samples. The x-axis represents the T cell subsets, and
   the y-axis shows the proportion of each subset. Data are presented as
   mean ± standard deviation, * adjust P < 0.05 considered statistically
   significant, “ns” indicates non-significant differences. F Boxplots
   showing the number of up-regulated and down-regulated DEGs in each T
   cell subset between the SLN and NLN samples. G KEGG pathway analysis
   with DEGs for each T cell subset. Adjust P < 0.05 is considered to be
   statistical significance

T cell subsets exhibit functional activation in cervical SLN

   To interrogate functional differences in T cells between SLN and NLN,
   we conducted GSEA on DEGs. The DEGs in SLN were significantly enriched
   in biological process related to T cell activation and
   differentiation (Fig. [74]2D), suggesting enhanced functional activity
   within the SLN. Notably, the analysis of T cell subset proportions
   revealed a markedly low abundance of CD4 + T cells, a finding
   corroborated by independent flow cytometry validation (Supplementary
   Figs. 1A and B). Although the frequencies of major T cell subsets were
   comparable between SLN and NLN (Fig. [75]2E), within each T cell
   subset, distinct transcriptional differences were observed. In the SLN
   group, the number of upregulated genes across subsets ranged from 38 to
   119, In contrast, the number of downregulated genes was Substantially
   fewer, with only 10–20 genes showing reduced expression (Fig. [76]2F
   and Supplementary Tables S3-S12). Notably, KEGG pathway analysis
   revealed that the upregulated DEGs were associated with apoptosis,
   TNF/IL-17 signaling, and MAPK activation (Fig. [77]2G), indicating a
   strong inflammatory environment in the SLN. Collectively, these
   findings suggest that Mtb infection induces transcriptional
   reprogramming in T cells, prioritizing effector functions and
   inflammatory signaling over changes in subset frequency, thereby
   orchestrating localized immune responses in LNTB.

Ten B cell subsets identified by scRNA-seq in cervical LNs

   The total B cells were re-clustered into 12 distinct clusters based on
   the expression of associated markers, including ten B cell subsets
   (CD79A +/CD79B +) alongside monocytic cells (cluster 11: LYZ +/CST3 +)
   and cycling T cells (cluster 10: CD3D +/CD3E +/MKI67 +) (Fig. [78]3A-C
   and Supplementary Figure S3A-B). Naïve B cell subsets (clusters
   1/3/4/7/8) were characterized by a high level of co-expression of MS4A1
   and IGHD and were further stratified into five functional subsets based
   on the specific enrichment of marker genes: IGHD + naïve
   (IGHD/TCL1A/IGHM/PLPP5), CD69 + naïve (CD69/ZNF331/CD83/NR4A1),
   PSME2 + naïve (PSME2/NME1/NME2), ISG15 + naïve
   (ISG15/IFI6/IFIT3/IFIT1), and NEAT1 + naïve (NEAT1). Memory B cells
   (clusters 0/2, MS4A1 +/CD27 +) diverged into AIM2 + memory B subset
   (AIM2/TNFRSF13B) and HSPA1A + memory B subset (HSPA1A). While germinal
   center (GC) B cells (clusters 5/6, MS4A1 +/CD27 +/CD38 +) comprised
   RGS13 + GC B subset (CCDC144A/RGS13/MEF2B/LMO2) and HMGB2 + GC B subset
   (HMGB2/STMN1/HIST1H4C/TUBA1B/HMGN2). The unique plasma B subset
   (cluster 9, MZB1 +/CD38 +/IGHG4 +) was specifically enriched with
   antibodies including such as IGHG4, IGHG1, IGHG3, IGLC1, and IGHA1.

Fig. 3.

   [79]Fig. 3
   [80]Open in a new tab

   Characterization of B cell subsets between the SLN and NLN samples.
   A UMAP plot displaying B cell subsets identified across five paired SLN
   and NLN samples. The left panel represents all cell types. The right
   panel focuses on the distribution of samples, with each dot
   representing an individual cell and colors indicating different
   samples. B Dot plot of canonical marker gene expression (columns)
   across clusters (rows). Dot size reflects the percentage of expressing
   cells; color intensity denotes mean normalized expression (log2 scale).
   C Violin plots highlighting differential expression of canonically cell
   marker genes across clusters. The x-axis represents the cell subsets,
   and the y-axis shows the normalized expression levels. D The GSEA
   analysis highlighting biological processes enriched in SLN samples
   compared to NLN samples. The x-axis represents the running enrichment
   score, and the y-axis indicates the rank in the ordered dataset.
   E Boxplots comparing the proportions of various B cell subsets between
   the SLN and NLN samples. The x-axis represents the B cell subsets, and
   the y-axis shows the proportion of each subset. Data are presented as
   mean ± standard deviation, * adjust P < 0.05 considered statistically
   significant, “ns” indicates non-significant differences (F) Boxplots
   showing the number of up-regulated and down-regulated DEGs in each B
   cell subset between the SLN and NLN samples. G KEGG pathway analysis
   with DEGs for major B cell subsets. Adjust P < 0.05 is considered to be
   statistical significance

Functional divergence of B cell subsets in cervical LNs

   GSEA of DEGs in total B cells between SLN and NLN revealed significant
   enrichment of pathways related to adaptive immunity, lymphocyte
   activation regulation, and T cell priming in SLN (Fig. [81]3D),
   underscoring enhanced functional engagement of B cells during Mtb
   infection. As shown in Fig. [82]3E, four major B cell subsets were
   identified: AIM2 + memory B cells, IGHD + naïve B cells,
   HSPA1A + memory B cells, and CD69 + naïve B cells, which together
   accounted for nearly 80% of the total B cell population. Analogous to
   the observations in T cells, no significant differences were observed
   in the frequency of these B cell subsets between the SLN and the NLN
   (Fig. [83]3E). Transcriptomic analysis identified a greater number of
   upregulated DEGs in the SLN when compared to the NLN, while only a
   relatively small number of genes showed significant downregulation
   (Fig. [84]3F and Supplementary Tables S13–S21). Notably, GC B cell
   subsets exhibited minimal transcriptional differences between the two
   groups (Fig. [85]3F and Supplementary Tables S13–S21). KEGG pathway
   analysis further linked the upregulated DEGs to apoptosis, IL-17/TNF
   signaling, and MAPK activation (Fig. [86]3G), suggesting a
   pro-inflammatory B cell phenotype in SLN. These findings suggest that B
   cells may contribute to the amplification of local inflammatory
   responses during Mtb infection through the modulation of cytokine
   signaling and survival pathways, despite the maintenance of stable
   subset frequencies.

Five myeloid cell subsets identified by scRNA-seq in cervical LNs

   The entire population of myeloid cells were re-clustered into seven
   distinct clusters based on the cell-specific marker genes, including
   three macrophage subsets (CD68 +), two plasmacytoid DC (pDC) subsets
   (CLEC4C + LILRA4 +), and two mesenchymal stem cell (MSC) subsets
   (DCN + COL3A1 +) (Fig. [87]4A-C and Supplementary Figure S4A, B). The
   macrophage subsets were classified as IL1B +, C1QC +, and
   S100B + subset, all of which displayed high expression of the
   M1-polarization marker CD86, indicating a pro-inflammatory phenotype
   within this cell population. Notably, the IL1B + macrophages expressed
   a high level of the pro-inflammatory cytokine IL1B and associated
   mediators (CCL3L1, S100A8,S100A9), suggesting specialized roles in
   inflammatory cascades. Both pDC subsets highly expressed cytotoxic and
   immune-regulatory genes (GZMB, JCHAIN, DUSP5, GPR183, TCL1A, and PTGDS)
   (Fig. [88]4B).

Fig. 4.

   [89]Fig. 4
   [90]Open in a new tab

   Characterization of myeloid cell subsets between the SLN and NLN
   samples. A UMAP plot displaying myeloid cell subsets identified across
   five paired SLN and NLN samples. The left panel represents all cell
   types. The right panel focuses on the distribution of samples, with
   each dot representing an individual cell and colors indicating
   different samples. B Dot plot of canonical marker gene expression
   (columns) across clusters (rows). Dot size reflects the percentage of
   expressing cells; color intensity denotes mean normalized expression
   (log2 scale). C Violin plots highlighting differential expression of
   canonically cell marker genes across clusters. The x-axis represents
   the cell subsets, and the y-axis shows the normalized expression
   levels. D The GSEA analysis highlighting biological processes enriched
   in the SLN samples compared to NLN samples. The x-axis represents the
   running enrichment score, and the y-axis indicates the rank in the
   ordered dataset. E Boxplots comparing the proportions of various
   myeloid cell subsets between SLN and NLN samples. The x-axis represents
   the myeloid cell subsets, and the y-axis shows the proportion of each
   subset. Data are presented as mean ± standard deviation, * adjust
   P < 0.05 considered statistically significant, “ns” indicates
   non-significant differences. F Boxplots showing the number of
   up-regulated and down-regulated DEGs in each myeloid cell subset
   between the SLN and the NLN samples. G KEGG pathway analysis with DEGs
   for IL1β + macrophages and MSC1. Adjust P < 0.05 is considered to be
   statistical significance

Assessment of myeloid cell subset differences between SLN and NLN samples

   To explore the functional heterogeneity among myeloid cells in SLN as
   compared to NLN, GSEA was conducted on DEGs obtained from myeloid
   populations. This analysis revealed that DEGs in SLN were significantly
   enriched in pathways related to embryonic organ morphogenesis, G
   protein-coupled receptor signaling, and inflammatory responses
   (Fig. [91]4D). Cellular composition analysis demonstrated comparable
   frequencies of myeloid subsets between SLN and NLN (Fig. [92]4E), with
   IL1B + macrophages and GZMB + pDCs collectively dominating. Subsequent
   transcriptional profiling uncovered subset-specific gene expression
   changes. In particular, IL1B + macrophages in SLN exhibited 313 DEGs,
   including 249 upregulated and 64 downregulated genes (Fig. [93]4F and
   Supplementary Table S22-23). The upregulated genes were strongly
   associated with pro-inflammatory signaling pathways, including IL-17,
   TNF, NOD-like receptor, and NF-κB signaling (Fig. [94]4G). Strikingly,
   MSC1 (a distinct myeloid subset) displayed 500 DEGs, with 428 genes
   significantly upregulated. These genes were enriched in pathways
   related to oxidative phosphorylation, TNF signaling, and antigen
   presentation machinery (Fig. [95]4G). In contrast, other myeloid
   subsets showed minimal transcriptional divergence between the SLN and
   NLN, suggesting a general functional stability across LN conditions.
   These findings highlight the activation of specific myeloid subsets,
   particularly IL1B + macrophages and MSC1, in driving inflammatory and
   metabolic reprogramming in LNTB.

Enhanced cell–cell communications in SLN samples

   To systematically compare cell–cell communications dynamics between the
   SLN and the NLN, a computational network analysis was carried out using
   CellChat. The SLN exhibited a higher number of interactions (SLN: 9,614
   vs. NLN: 8,275) and a stronger interaction strength intensity
   (SLN = 285.919 vs. NLN = 214.418), indicating a systemic enhancement of
   cellular crosstalk in the infected nodes (Fig. [96]5A). Visualization
   via a heatmap confirmed more extensive connectivity across all cell
   types in the SLN. Among them, C1QC + macrophages, MSC1, and MSC2
   emerged as the predominant receivers of incoming signals, while MSC1
   and HMGB2 + GC B cells showed heightened outgoing signaling activity
   (Fig. [97]5B). Notably, naïve CD8 + T cells, MSC2, and NKT cells ranked
   as the top three cell types with elevated incoming interaction strength
   in SLN, whereas MSC1 and S100B + macrophages dominated outgoing
   interactions. Critically, macrophages in SLN demonstrated intensified
   crosstalk with naïve CD8 + T cells and NKT cells, suggesting their
   central role in coordinating adaptive immunity.

Fig. 5.

   [98]Fig. 5
   [99]Open in a new tab

   Characterization of cell–cell communications among all cell types
   between the SLN and NLN samples. A The number of interactions and
   interaction strength among all cell types between the SLN and NLN. B
   Heatmaps show differential number of interactions (left) and
   differential interaction strength (right) among all cell types in the
   overall signaling patterns between the SLN and NLN. The color bars on
   the top show the total incoming signal for each column, while the bars
   on the right show the total outgoing signal for each row. Red and blue
   color indicates increases and decreases, respectively in signal in the
   SLN compared to the NLN. C Identification and visualization of
   conserved and specific signaling pathways. Orange color indicate
   pathways that were expressed only or higher in the SLN; Green color
   indicate those that were expressed only or higher the NLN. D Heatmaps
   of the overall signaling patterns of the SLN (left) and NLN (right).
   Circle plots show the INF-II (E) and IL-1 (F) signaling networks among
   all cell types between the SLN and NLN

   Comparative pathway analysis identified 18 SLN-specific pathways
   (VISFATIN, TGFb, PTN, CD46, CD80, PERIOSTIN, NOTCH, GAS, TWEAK, XCR,
   OSM, KIT, HSPG, DESMOSOME, EPHB, PROS, RA, and BMP) and 4 NLN-enriched
   pathways (SEMA3, ICOS, CD160, and SELL) (Fig. [100]5C). Shared pathways
   showed SLN-specific hyperactivation, including MK, CCL, and CD40
   signaling (Fig. [101]5D). Cell-type-restricted pathways were also
   identified: PD-L1 signaling localized to monocytes and exhausted
   CD4 + T cells in both groups, while NCAM signaling linked MSC1 and NKT
   cells (Fig. [102]5D).

   Focusing on antimicrobial responses, IFN-II signaling, which is
   critical for the control of Mtb), originated from CD8 + effector T
   cells, NKT cells, and T cell subsets in the SLN. This signaling
   targeted macrophages (C1QC +/IL1B +) and plasma B cells (IGHG4 +)
   (Fig. [103]5E). In contrast, the NLN limited IFN-II signaling to
   NKT → macrophage interactions. Similarly, IL-1 signaling, a key innate
   immune driver, was amplified in the SLN. IL1B + and C1QC + macrophages
   transmitted signals to both NKT cells and MSC1, whereas the NLN
   restricted IL-1 reception to NKT cells (Fig. [104]5F). These findings
   establish macrophages as the central orchestrators of inflammatory
   crosstalk in the SLN, bridging the innate and adaptive immunity through
   spatially coordinated signaling.

Discussion

   Despite its significant clinical implications, EPTB receives far less
   research attention than pulmonary TB due to its minimal infectiousness.
   Following Mtb infection, LNs are the first path of immune cell movement
   and are the most often infected area in EPTB [[105]24, [106]25].
   Understanding their immune microenvironment—specifically whether these
   sites foster protective immunity or drive pathology—is fundamental to
   disease control. Conventional bulk RNA-seq approaches fail to resolve
   cellular heterogeneity, whereas scRNA-seq enables comprehensive
   characterization of distinct cell subsets and cell states [[107]17].
   Notably, within infected individuals, only select LNs develop
   pathological changes while adjacent LNs appear histologically normal,
   presumably reflecting differential Mtb containment or clearance. By
   comparing the immune responses in abnormal lymph nodes to those in
   normal-appearing lymph nodes from individuals with LNTB, a deeper
   insight into the pathogenesis of LNTB can be gained.

   Our single-cell transcriptomic analysis of paired SLN and NLN obtained
   from cervical LNTB patients provides deep insights into the immune
   microenvironment of LNTB. Although previous studies have described
   broad inflammatory signatures in Mtb-infected LNs [[108]12, [109]16],
   our work advances the field by systematically mapping
   cell-type-specific transcriptional reprogramming and intercellular
   communication networks. The enhanced global cell–cell communication in
   SLN, which is characterized by macrophage-centric signaling hubs and
   amplified IFN-II/IL-1 pathways suggests a spatially coordinated immune
   strategy to counter Mtb, albeit with potential trade-offs between
   bacterial containment and immunopathology.

   The macrophages in SLN communication networks aligns with their dual
   roles as Mtb reservoirs and immune sentinels [[110]10, [111]11]. We
   observed that IL1B + macrophages, which exhibited marked upregulation
   of inflammation-related genes, engaged extensively with NKT cells via
   IL-1 and TNF signaling. This interaction likely amplifies IFN-γ
   production from cytotoxic lymphocytes, a mechanism critical for
   macrophage activation and intracellular pathogen control
   [[112]26–[113]30]. However, excessive IL-1β signaling may result in
   tissue damage [[114]31, [115]32]. Strikingly, while SLN macrophages
   displayed transcriptional activation, their frequency remained
   unchanged compared to NLN. This paradox mirrors findings in pulmonary
   TB granulomas, where macrophage functional polarization, rather than
   abundance, determines infection outcomes. Our data thus reinforce the
   paradigm that Mtb subverts myeloid cell plasticity to establish chronic
   infection niches. The expanded IFN-II signaling network in SLN—with
   CD8 + T/NKT cells as key IFN-γ sources—contrasts sharply with the
   attenuated responses observed in peripheral blood of TB patients
   [[116]33–[117]36]. This compartmentalization suggests that SLN serve as
   localized hubs for anti-Mtb effector functions, potentially
   compensating for systemic immune exhaustion. IFN-γ is known to enhance
   macrophage autophagy and antigen presentation, while chronic IFN-γ
   exposure may also induce T cell exhaustion. These findings highlight
   the need for therapies that balance localized immunity with systemic
   tolerance.

   Despite constituting nearly 50% of LN immune cells, B cell subsets
   showed minimal frequency changes between SLN and NLN. However,
   transcriptional profiling revealed significant upregulation
   of ISG15 and IFIT1 in naïve B cells. These genes are associated with
   antiviral responses. Conventionally, B cells are regarded as having a
   secondary role in the defense against TB, Nevertheless, our data
   suggest they may participate in Mtb control through non-canonical
   mechanisms, such as cytokine secretion or antigen-independent T cell
   priming. The absence of the Mtb-specific antibody detection in our
   study limits definitive conclusions, but high levels of IGHG1/IGHG4 in
   plasma cells hints at localized humoral responses. Future studies
   integrating BCR sequencing and spatial proteomics could clarify their
   role.

   The current study provides valuable insights into the immune
   microenvironment of cervical LNTB but has several limitations that
   warrant further discussion. The absence of healthy controls limits the
   ability to distinguish between baseline immunological variation and
   Mtb-specific changes. While the study employed rigorous bioinformatic
   pipelines, potential batch effects arising from technical variability
   were not explicitly addressed. Batch effects could confound
   transcriptional comparisons between the SLN and NLN groups,
   particularly given the small cohort size. Future studies should
   incorporate batch correction methods to ensure robust cross-sample
   comparisons. The cohort (n = 5 patients) may underrepresent the
   diversity of cervical LNTB immunopathology, especially given the
   heterogeneity of Mtb strains and host genetic backgrounds.
   Additionally, all samples were sourced from a single geographic region,
   which may limit the global applicability of findings. CellChat inferred
   enhanced cell crosstalk in SLN, these predictions rely on
   ligand-receptor databases and expression probabilities. Functional
   assays are needed to validate these interactions and establish
   causality. Last, the absence of paired peripheral blood mononuclear
   cell (PBMC) data precludes comparisons between local LN immunity and
   systemic immune responses. Such comparisons could elucidate
   compartmentalized immune exhaustion or activation patterns in LNTB.

   In conclusion, our single-cell transcriptomic study uncovers the
   complex immunological architecture of Mtb-infected cervical LN,
   providing the first comprehensive atlas of immune cell heterogeneity in
   cervical LNTB. We identified 10 T cell, 10 B cell, and 6 myeloid cell
   subsets, revealing remarkable functional specialization within each
   lineage. While SLN and NLN exhibited comparable immune subset
   frequencies, SLN demonstrated widespread transcriptional reprogramming
   across cell types, characterized by upregulation of pro-inflammatory
   pathways and metabolic remodeling. This chronic inflammatory milieu,
   coupled with enhanced intercellular communication networks centered on
   macrophage-T cell interactions, likely drives both anti-mycobacterial
   responses and pathological LN enlargement. By mapping the
   ligand-receptor axes governing immune crosstalk, this work paves the
   way for developing precision immunotherapies targeting cellular
   crosstalk in LNTB, while offering a framework to dissect
   compartmentalized immunity in other EPTB manifestations.

Supplementary Information

   [118]Supplementary Material 1.^ (3.1MB, pdf)
   [119]Supplementary Material 2.^ (10.1KB, xlsx)
   [120]Supplementary Material 3.^ (10.2KB, xlsx)
   [121]Supplementary Material 4.^ (92KB, xlsx)
   [122]Supplementary Material 5.^ (83KB, xlsx)
   [123]Supplementary Material 6.^ (72.6KB, xlsx)

Acknowledgements