Human lungs maintain tissue-resident memory T cells against a broad spectrum of pathogens

Abstract

Lung tissue-resident memory T (T_RM) cells are critical for frontline immunity, yet they undergo rapid attrition in the mouse lung. Whether this paradigm applies to humans has remained unknown. Here we present a comprehensive analysis of T cells from human lungs, characterizing the prevalence and properties of lung T_RM cells specific to a broad spectrum of pathogens. Using a T cell receptor-guided approach that integrates single-cell transcriptomics with paired T cell receptor repertoire profiling, we mapped the pathogen specificity of more than 87,000 lung T cells across 40 individuals, the majority of whom harbored T_RM cells specific to multiple pathogens. We confirmed that a large fraction of lung T_RM clones persist in the lung for many months to years. Thus, in contrast to those of mice, human lungs retain a stable and varied pool of pathogen-specific T_RM cells, suggesting that strategies to bolster these responses could provide durable protection against severe lung infections.

Unlike in other tissues, lung-resident memory T cell populations are not well maintained in mice. Here the authors show that this is not the case in human lungs, where tissue-resident memory T cells specific to a variety of pathogens can persist for many years.

Main

While serving their primary function in gas exchange, lungs are exposed to a broad spectrum of airborne pathogens capable of causing life-threatening infections¹, with such infections accounting for more than 2 million deaths worldwide each year². Protection from infections is provided by epithelial cells, pneumocytes and macrophages lining the airways and alveoli, which contribute to innate immune responses against inhaled pathogens¹. In addition, adaptive memory immune responses generated during primary infections provide durable protection upon re-exposure. Multiple studies in mouse models of infections at barrier sites, including the lung, have highlighted tissue-resident memory T cells (T_RM) as a key cell type orchestrating protective responses during reinfection and heterologous infections^3–5.

T_RM cells generated in the mouse lung have been reported to decline rapidly over time, in contrast to the long-lived T_RM cells in other organs, including the skin, liver and intestines^4,6,7. Thus, it has been proposed that the lung-draining lymph nodes (LLN) serve as a reservoir of T_RM cells that migrate retrogradely from the lung to maintain lung T_RM cell responses^8,9. The selective decline of T_RM cells in the mouse lung is considered to offer host benefit by balancing better gas-exchange function with organismal fitness against infections⁴. These observations in mouse models have raised the question of whether T_RM cells in human lungs similarly undergo rapid attrition, which could limit their capacity to provide durable memory responses to reinfection. This would also have major implications for vaccination approaches, in which generation of durable lung T_RM cell responses is critical for long-term, effective protection against the targeted pathogen^5,10,11.

Although T_RM cells have been reported in the human lung^12–17, as have pathogen-specific T_RM cells^18–24, previous studies have been limited by small cohort sizes, low cell yields and restricted pathogen coverage, leaving unresolved the question of whether the mouse paradigm of short-lived lung T_RM and LLN-based maintenance applies to humans. Some studies have attempted to enumerate lung T cells and T_RM cells that recognize epitopes from respiratory viruses such as influenza A virus (IAV)^19,21,24, SARS-CoV-2 (refs. ^22,23) and respiratory syncytial virus^18,24 and herpesviruses (namely, the human cytomegalovirus (CMV) and Epstein–Barr virus (EBV)). However, many of these studies employed peptide megapool (MP)-based stimulation coupled with intracellular cytokine staining to assess the frequency of pathogen-specific T cells in the human lung^18,22–24; this approach is likely to be confounded by false positives due to nonspecific or bystander activation of CD8⁺ T cells, which can occur owing to cytokine-mediated activation of memory CD8⁺ T cells²⁵. Thus, such studies may overestimate the frequency of pathogen-specific cells unless there are adequate controls in place, for example, irrelevant peptide pools, addition of cytokine blockers or brefeldin for the entire duration of stimulation, or additional T cell receptor (TCR) sequencing to exclude bystander activation, as clonotypes of such cells will be detected under different peptide stimulation conditions. Moreover, previous efforts to identify such cells in tissue have lacked the throughput to support comparisons of different pathogen-specific T cell populations within an individual^{18–20,22,23}. As a result, our understanding of how pathogen-specific T cell memory is distributed, shaped and maintained at mucosal sites in humans has remained limited. Thus, there is a need for high-resolution, large-scale approaches capable of capturing diverse pathogen-specific T_RM cell populations for comparison within and across individuals.

To overcome these challenges, we profiled large numbers of T cells from the human lung and LLN to characterize the prevalence, clonal sharing and properties of lung T_RM cells specific to a broad range of pathogens. We developed an integrative strategy that leverages the TCR as a molecular barcode of antigen specificity. By probing the lung TCR repertoires with (i) donor-matched pathogen-specific TCRs derived from blood, (ii) publicly available CD8⁺ TCRs and (iii) a newly generated TCR resource for CD4⁺ T cells, we were able to impute the pathogen specificity of a substantial fraction of lung T cells from 40 individuals. This strategy enabled us to characterize a large number of pathogen-specific T cells in the human lung, while also providing an extensive resource of CD4⁺ TCRs specific to common human pathogens. The scale and resolution of our datasets allowed robust enumeration and phenotypic comparisons of pathogen-specific T cells within and across anatomic sites. We observed that respiratory-pathogen-specific T_RM clones persisted in the human lungs over many months to years and displayed limited clonal sharing with T cells in the LLN.

Human lungs harbor an oligoclonal pool of memory T cells

We obtained uninvolved lung tissue and LLN samples from 40 individuals who were clinically free of recent infections and deemed medically fit to undergo surgical lung resection for early-stage lung cancer (hereafter referred to as the lung cohort; Fig. 1a and Extended Data Fig. 1a). The advanced age of the lung cohort (median 73 years, range 61–83; Supplementary Table 1) offered an opportunity to characterize the residual memory T cell populations in the lung that are shaped by lifelong exposure to pathogens and vaccine antigens. As each TCR is highly specific for an MHC-bound peptide epitope, its identity can be used to map the specificities of T cells to particular pathogens or host antigens. We therefore generated full-length paired α and β chain TCR sequences, including the unique complementarity determining region 3 (CDR3) and associated V and J genes—defined as a TCR clonotype—coupled with transcriptomes of ~600,000 single T cells isolated from the lung tissue of 40 individuals (median 17,173 cells) and of ~330,000 T cells from the LLNs of 36 of the 40 individuals (median 8,671 cells; Fig. 1a, Extended Data Fig. 1b,c and Supplementary Tables 2–4).

Fig. 1. Human lungs harbor an oligoclonal pool of memory T cells.

a, Study overview. b, Comparison of ratios of T cells to TCR clonotypes between matched lung and LLN samples, with cell numbers normalized to the minimum across donor-matched samples (n = 36). ****P = 1.9 × 10⁻¹¹ and ****P = 5.8 × 10⁻¹¹ for CD4 and CD8, respectively. c, Percentages of CD4⁺ and CD8⁺ lung T cells (y axis) represented by the top-ranked TCR clonotypes (x axis). Thin lines represent individual donors (n = 40), and the thick black line indicates the LOESS-smoothed consensus. The vertical dashed line marks the 10% of top-ranked clonotypes. d, TCR clonotype sizes in lung CD4⁺ and CD8⁺ T cells. Each bar represents a donor (n = 40), gray indicates nonexpanded clonotypes (clone size = 1), colored lines delineate expanded clonotypes (clone size > 1) stacked on the basis of absolute clonotype frequency, and white dots indicate the total number of TCR clonotypes. Donor-specific metadata are displayed in the accompanying heatmap. e, Comparison of Gini (diversity) indices between lung and LLN TCR repertoires of CD4⁺ and CD8⁺ T cells, with cell numbers normalized to the minimum across donor-matched samples (n = 36); ****P = 2.9 × 10⁻¹¹. f, As in d but for LLN TCR repertoires from 36 donors. g, Overlap of highly expanded lung TCR clonotypes (n ≥ 10 cells) with LLN TCRs. Line plot shows the relative frequency (y axis) of TCR clonotypes in lung (blue) and LLN (peach), ordered on the basis of frequency in the lung (x axis). Bar plots below show the fractions of lung TCR clonotypes that were either unique to the lung or shared with the corresponding LLN. h, Comparison of Morisita–Horn (similarity) indices for donor-matched lung and LLN TCR repertoires (n = 36) in CD4⁺ and CD8⁺ T cells. Reference values (labeled as ref. ctrl.), representing maximal similarity, were obtained by comparing the two halves of a random split of the combined lung and LLN TCR repertoires for each donor, as described in ref. ⁵⁴. ****P < 2 × 10⁻⁷. Box plots in b, e and h represent median, interquartile range (IQR) and 1.5× IQR (whiskers). Each dot represents a donor, and lines connect donor-matched observations. Statistical significance was assessed using two-sided Wilcoxon signed-rank paired test and adjusted for multiple testing in h. BMI, body mass index; B. pertussis-aP, acellular pertussis vaccine epitopes; B. pertussis-R, non-aP pertussis epitopes, where ‘R’ indicates ‘rest’; MPV, metapneumovirus; PIV, parainfluenza virus; RSV, respiratory syncytial virus. Schematic in a created in BioRender; Fajardo-Rosas, V. https://biorender.com/hdh0rpt (2026).

Source data

Extended Data Fig. 1. Human lungs harbor an oligoclonal pool of memory T cells.

a, Schematic of thoracic sampling sites, indicating lung lobes (across 40 samples) and lung-draining lymph nodes (LLN; across 36 samples). Circles denote LLNs, with color indicating topographic classification and numbers denoting the total number of donors per LLN type. b, Violin plots show distribution of indicated quality control (QC) metrics across lung, LLN or blood CD3⁺ T cells in the indicated cohorts (color-coded), stratified by FACS batch. c, Distribution of the number of lung and LLN T cells recovered from donor-matched samples (n = 36) after transcriptome-based lineage deconvolution to distinguish between CD4⁺ (top) and CD8⁺ (bottom) T cells (Methods). ****P < 1.6 × 10⁻⁵. d, Percentage of LLN CD4⁺ (top) and CD8⁺ (bottom) T cells (y-axis) represented by the top-ranked TCR clonotypes (x-axis). Thin lines represent individual donors (n = 36) and the thick black line indicates the LOESS-smoothed consensus. The vertical dashed line marks the 10% of top-ranked clonotypes. e, Comparison of frequency of expanded TCR clonotypes between donor-matched lung and LLN samples (n = 36) in CD4⁺ (top) and CD8⁺ (bottom) T cells. ****P=2.9 × 10⁻¹¹. f, Overlap of highly expanded lung TCR clonotypes (n$\ge$10 cells) with LLN TCRs, considering cell numbers normalized to the minimum across donor-matched samples. Line plot shows the relative frequency (y-axis) of TCR clonotypes in lung (blue) and LLN (peach), ordered based on frequency in the lung (x-axis). Bar plots below show the fractions of lung TCR clonotypes that are either unique to the lung or shared with the corresponding LLN. g, Comparison of Morisita-Horn (similarity) indices for donor-matched lung and LLN TCR repertoires (n = 36) in CD4⁺ and CD8⁺ T cells, considering cell numbers normalized to the minimum across donor-matched samples. Reference values (labelled as ref. ctrl.), representing maximal similarity, were obtained by comparing the two halves of a random split of the combined lung and LLN TCR repertoires for each donor. h, Correlation between Morisita-Horn indices for donor-matched lung and LLN TCR repertoires (n = 36), comparing values obtained using all CD4⁺ (top) or CD8⁺ (bottom) T cells or numbers normalized to the minimum across donor-matched samples. Spearman correlation coefficients are indicated. ****P < 2.2 × 10⁻¹⁶. i,j, Comparison of Morisita-Horn (similarity) indices for donor-matched lung and LLN TCR repertoires (n = 36) in CD4⁺ (top) and CD8⁺ (bottom) T cells, with donors stratified by lung (i) or LLN (j) region. Number of donors considered for each region are indicated. Cell numbers were normalized to the minimum across donor-matched samples. P-values from the two-sided Wilcoxon rank-sum test adjusted for multiple testing. n.s., P>0.09; *P=0.04. Boxplots in (c), (e), (g), (i) and (j) represent median, IQR, and 1.5x IQR whiskers; each dot represents a donor (up to 36). P-values from the two-sided Wilcoxon signed-rank paired test in (c), (e) and (g), and adjusted for multiple testing in (g). ****P < 0.0001. Schematic in a created in BioRender; Ehtram, A. https://biorender.com/icmgre1 (2026).

Source data

We first quantified the frequency of T cells with identical TCR clonotypes to assess their clonal abundance. This analysis provided a measure of antigen-specific clonal accumulation within the lung and LLN. The LLN compartment, as well as capturing cells that are trafficked from the lungs, largely serves as a surrogate for the pool of recirculating naive and memory T cells from the blood, thereby allowing comparison between lung tissue-resident and circulating memory populations. In contrast to the LLN, we observed a substantially higher ratio of T cells to TCR clonotypes in the lung—that is, clonal accumulation (Fig. 1b)—with the top 10% of the clonotypes accounting for 51% of CD4⁺ and 70% of CD8⁺ T cells in the lung (median across individuals; Fig. 1c and Extended Data Fig. 1d). In some individuals, as few as ten TCR clonotypes dominated the lung TCR repertoire (Fig. 1d), indicating preferential accumulation of certain antigen-specific memory T clones in the human lung. This selective clonal expansion of a restricted pool of memory T cells in the lung was supported by a high Gini index²⁶ (Fig. 1e), consistent with pronounced oligoclonality and significantly reduced TCR repertoire diversity of lung T cells relative to LLN T cells (Fig. 1d–f and Extended Data Fig. 1e). Notably, a substantial fraction of highly expanded TCR clonotypes (clone size ≥ 10 cells) identified in the lung were not detected in the LLN TCR repertoires (Fig. 1g and Extended Data Fig. 1f), suggesting that these clonotypes may correspond to lung-resident T cells that do not recirculate to the LLN. The Morisita–Horn similarity index (0, no overlap; 1, complete overlap), which accounts for both number and relative abundance of shared clonotypes in lung and LLN TCR repertoires²⁷, remained low even when equal numbers of T cells were analyzed in each compartment and across all lung regions or LLNs examined (Fig. 1h and Extended Data Fig. 1g–j). Our reanalysis of the TCR repertoire from lung and LLN from an independent cohort of individuals without cancer showed patterns concordant with our results for clonal expansion and TCR sharing¹⁷ (Supplementary Fig. 1). Overall, these findings indicate that the human lung harbors a highly expanded and oligoclonal pool of memory T cells, even in the absence of recent infections.

Selective clonal expansion of T_RM cells in human lungs

If T_RM cells with highly expanded, low-diversity TCR repertoires were abundant in the lung, this would imply that multiple antigen-specific T_RM populations are preferentially retained in the human lung, in contrast to observations in mice, in which pathogen-specific lung T_RM cells have been reported to decline rapidly over time^4,6,7. Based on transcriptional profiles, we quantified different T cell subsets and found that T_RM cells were the dominant CD4⁺ and CD8⁺ T cell subset in the human lung, regardless of the lung region analyzed (Fig. 2a, Extended Data Fig. 2a–e and Supplementary Table 3). These lung T_RM cells displayed a high degree of clonal expansion and a high Gini index (Fig. 2b and Extended Data Fig. 2f); notably, the top 10% of their TCR clonotypes comprised more than 46% and 50% of T cells in the CD4⁺ and CD8⁺ T_RM subsets, respectively (median across individuals) (Fig. 2c). This indicates selective retention of antigen-specific lung T_RM clones in the lung.

Fig. 2. Selective clonal expansion of T_RM cells in human lungs.

a, Single-cell transcriptomes of lung CD4⁺ (top) and CD8⁺ (bottom) T cells displayed by UMAP. Unbiased clustering of 370,233 CD4⁺ and 227,800 CD8⁺ T cells from 40 donors, colored based on cluster; labels indicate assigned T cell subset. Bar charts (right) display the proportions of each T cell subset. b, Comparison of frequency of expanded TCR clonotypes (top) and Gini indices (bottom) between T_RM and T_CM subsets in the lung. Each dot represents a donor (n = 40); lines connect paired T_RM and T_CM observations. ****P < 1.5 × 10⁻¹¹. c, Percentages of cells in the T_RM and T_CM subsets (y axis) represented by the top-ranked TCR clonotypes in the lung (x axis). Thin lines represent individual donors (n = 40), and the thick black line indicates the LOESS-smoothed consensus. The vertical dashed line marks the 10% of top-ranked clonotypes. d, Heatmaps (left) show the relative frequency of highly expanded (≥10 cells) clonotypes (columns) in the indicated T cell subsets (rows). Bar charts (right) display the proportions (x axis) of clonotypes that were phenotypically biased to each lung T cell subset (y axis) (Supplementary Methods). Data represent cells pooled across all 40 donors. e, As in a, but for 198,890 CD4⁺ and 129,737 CD8⁺ LLN T cells from 36 donors. f, As in b, but for the LLN TCR repertoires from 36 donors. ****P < 9.6 × 10⁻⁷. g, Overlap of highly expanded (n ≥ 10 cells) lung TCR clonotypes phenotypically biased to the indicated subsets (T_RM, T_EM and GZMK⁺) with LLN T cells. Line plot shows the relative frequency (y axis) of TCR clonotypes in lung (blue) and LLN (peach), ordered according to the frequency in the lung (x axis). Bar plots below show the fractions of lung TCR clonotypes in each subset that were either unique to the lung or shared with the corresponding LLN. h, For one representative donor (L010), CD4⁺ (left) and CD8⁺ (right) T cells expressing clonotypes biased to the indicated lung cell subsets are shown on lung and LLN UMAPs, with point size indicating relative frequency of the indicated clones; the size scale is consistent across anatomic sites. i, Comparison of Morisita–Horn (similarity) indices for donor-matched lung and LLN TCR repertoires (n = 36) in the indicated CD8⁺ T cell subsets, considering only subset-biased clonotypes. Cell numbers were normalized to the minimum across donor-matched samples. Reference values, representing maximal similarity, were obtained by comparing the two halves of a random split of the combined lung and LLN CD8⁺ TCR repertoires for each donor. ****P < 1.3 × 10⁻⁸. Box plots in b, f and i represent median, IQR and 1.5× IQR (whiskers). Statistical significance was assessed using two-sided Wilcoxon signed-rank paired test and adjusted for multiple testing in i. Act., activated; interm., intermediate; MAIT, mucosal-associated invariant T cell; prolif., proliferating; T_FH, T follicular helper cell; T_reg, regulatory T cell; NS, not significant.

Source data

Extended Data Fig. 2. Annotation of T cell subsets in human lungs.

a, Density on UMAP of CD4⁺ (left) and CD8⁺ (right) lung T cells separated by major lung lobes. Less frequent sites (present in <20% of donors) are grouped as “Others”. Rightmost plot shows the fraction of cells in each T cell subset by lobe. b, Dot plots showing Seurat-normalized mean expression (color scale) and the percentage of cells (dot size) expressing the indicated markers across cells in each subset of lung CD4⁺ (left) and CD8⁺ (right) T cells. Data are pooled across donors (n = 40). c, UMAP shows IDEIS-inferred expression levels of PTPRC isoforms (CD45RO and CD45RA) in CD4⁺ (left) and CD8⁺ (right) lung T cells. d, GSEA with the indicated gene signatures (PMIDs shown), comparing each CD4⁺ (left) or CD8⁺ (right) lung T cell subset to the remaining cells. Heatmap summarizes normalized enrichment scores (NES) for significant comparisons (adjusted P < 0.05), with gray indicating non-significant results. e, Proportion of total lung CD4⁺ (top) or CD8⁺ (bottom) T cells represented by each subset. Each bar represents a donor from the study cohort (n = 40). f, Distribution across donors (up to 40) of the percentage of lung T cells expressing expanded clonotypes (that is, those observed in more than one cell) and the Gini indices, separately in lung CD4⁺ (left) and CD8⁺ (right) T cell subsets. g, TCR clonal overlap between the T_RM subset and others, for highly expanded ($\ge$10 cells) CD4⁺ (left) and CD8⁺ (right) TCR clonotypes. h, Percentage of lung CD4⁺ (left) and CD8⁺ (right) T cells (y-axis) biased to the indicated phenotypes represented by the top-ranked TCR clonotypes (x-axis). Thin lines represent individual donors (n = 40) and the thick black line indicates the LOESS-smoothed consensus. The vertical dashed line marks the 10% of top-ranked clonotypes. i, Plots show the percentage of lung T cells expressing expanded clonotypes and the Gini indices, separately in lung samples from CD4⁺ (n = 40; top) and CD8⁺ (n = 38; bottom) T cells segregated according to phenotype bias. P-values from the two-sided Wilcoxon signed-rank paired test. ****P < 2.4 × 10⁻⁶. Boxplots in (f) and (i) represent median, IQR, and 1.5x IQR whiskers; each dot represents a donor (n = 40).

Source data

To assess whether recirculating T cell subsets in the lung, such as central memory T (T_CM) or effector memory T (T_EM) cells, contributed to the pool of antigen-specific T_RM cells, we assessed clonal sharing of the highly expanded TCR clonotypes (clone size ≥ 10 cells) among different lung T cell subsets. Although most highly expanded TCR clonotypes of T_RM cells were shared with other lung T cell subsets (Extended Data Fig. 2g), a substantial fraction (~50%) exhibited phenotypic bias, with preferential expansion (>2-fold) in the T_RM subset relative to other subsets (Fig. 2d). Notably, this phenotypic bias of TCR clonotypes was also observed for all other lung T cells subsets (Fig. 2d), indicating that TCR clonotypes segregate according to T cell phenotype, as previously reported²⁸. Even among the TCR clonotypes biased to the T_RM phenotype, we observed significant clonal expansion and restricted diversity (Extended Data Fig. 2h,i). This suggests that a subset of highly expanded T_RM clones are largely lung resident and independent of the T_EM and T_CM subsets, which can recirculate and potentially intermix with the circulating memory T cell pool²⁸.

Mouse models have supported a role of the LLN as a major reservoir of pathogen-specific T_RM cells that otherwise undergo rapid attrition in the mouse lung^8,9. To assess whether T_RM cells in the LLN were also a major reservoir of T_RM cells in the human lung, we compared TCR clonotypes of lung T_RM cells with T cells in their corresponding LLN. First, as reported in mouse models⁸, we observed both CD4⁺ and CD8⁺ T cells with tissue-residency features (that is, T_RM cells) in all the LLN stations analyzed (Fig. 2e, Extended Data Fig. 3a–e and Supplementary Table 1). Although these LLN T_RM cells represented a relatively small fraction of both CD4⁺ and CD8⁺ T cells (Fig. 2e and Extended Data Fig. 3e), their TCRs displayed high clonal expansion and their TCR repertoires displayed low diversity compared to other LLN T cell subsets (Fig. 2f and Extended Data Fig. 3f). Compared to other LLN T cell subsets, a large fraction (median ≥ 50% across donors for both CD4⁺ and CD8⁺ T cells) of LLN T_RM TCR clonotypes were clonally related to lung T cells (Extended Data Fig. 4a,b), indicating potential trafficking of T_RM cells between these two compartments. However, the converse was not true, with a major fraction (median ≥80% across donors for both CD4⁺ and CD8⁺ T cells) of the lung T_RM TCR clonotypes not detected in the LLN T cells (Extended Data Fig. 4a). Furthermore, among the highly expanded TCR clonotypes shared between the lung and LLN, those biased toward the T_RM phenotype were significantly more abundant in the lung than in the LLN (Fig. 2g,h and Extended Data Fig. 4c–e).

Extended Data Fig. 3. Annotation of T cell subsets in human lung-draining LNs.

a, Density on UMAP of CD4⁺ (left) and CD8⁺ (right) LLN T cells separated by major nodal regions. Less frequent sites (present in <20% of donors) are grouped as “Others”. Rightmost plot shows the fraction of cells in each T cell subset by nodal region. b, Dot plots showing Seurat-normalized mean expression (color scale) and the percentage of cells (dot size) expressing the indicated markers across cells in each subset of LLN CD4⁺ (left) and CD8⁺ (right) T cells. Data are pooled across donors (n = 36). c, UMAP shows IDEIS-inferred expression levels of PTPRC isoforms (CD45RO and CD45RA) in CD4⁺ (left) and CD8⁺ (right) LLN T cells. d, GSEA with the indicated gene signatures (PMIDs shown), comparing each CD4⁺ (left) or CD8⁺ (right) LLN T cell subset to the remaining cells. Heatmap summarizes normalized enrichment scores (NES) for significant comparisons (adjusted P < 0.05), with gray indicating non-significant results. e, Proportion of total LLN CD4⁺ (top) or CD8⁺ (bottom) T cells represented by each subset. Each bar represents a donor from the study cohort (n = 36). f, Percentage of LLN T cells expressing expanded clonotypes (that is, those observed in more than one cell) and the Gini indices, separately in LLN CD4⁺ (left) and CD8⁺ (right) T cell subsets. Boxplots represent median, IQR, and 1.5x IQR whiskers; each dot represents a donor (up to 36 per box).

Source data

Extended Data Fig. 4. Limited clonal sharing between lung T_RM cells and LLN T cells.

a, Distributions across donors of the fractions of TCR clonotypes shared between anatomic sites, calculated for each T cell subset from either the lung (top) or the LLN (bottom). Distributions are shown for CD4⁺ (top panels) and CD8⁺ (bottom panels) TCR clonotypes. b, Comparison of Morisita-Horn (similarity) indices for donor-matched lung and LLN TCR repertoires in the indicated CD4⁺ (n = 32 donors; top) and CD8⁺ (n = 34 donors; bottom) LLN T cell subsets, considering cell numbers normalized to the minimum across donor-matched samples. Reference values (labelled as ref. ctrl.), representing maximal similarity, were obtained by comparing the two halves of a random split of the combined lung and LLN CD4⁺ or CD8⁺ TCR repertoires for each donor. P-values from the two-sided Wilcoxon signed-rank paired test comparing the distributions and adjusted for multiple testing within each panel. n.s., P>0.8; ****P < 2.5 × 10⁻⁶. c, Overlap of highly expanded (n$\ge$10 cells) lung TCR clonotypes phenotypically biased to the indicated subsets—T_RM, T_EM and GZMK⁺—with LLN T cells, considering cell numbers normalized to the minimum across donor-matched samples. Line plot shows the relative frequency (y-axis) of TCR clonotypes in lung (blue) and LLN (peach), ordered based on the frequency in the lung (x-axis). Bar plots below show the fractions of lung TCR clonotypes in each subset that are either unique to the lung or shared with the corresponding LLN. d, Comparison of the relative frequency ratio (lung vs. LLN) for shared TCR clonotypes among the indicated CD8⁺ lung T cell subsets. e, For one representative donor (L10), CD4⁺ (left) and CD8⁺ (right) T cells expressing selected clonotypes (color-coded) biased to the indicated lung cell subsets are shown on lung and LLN UMAPs, with point size indicating relative frequency of the indicated clones; the size scale is consistent across anatomic sites. f,g, Comparison of Morisita-Horn (similarity) indices for donor-matched lung and LLN TCR repertoires in the indicated CD8⁺ T cell subsets, considering cell numbers normalized to the minimum across donor-matched samples and for subset-biased clonotypes only. Comparisons stratified by lung (f) or LLN (g) sampling site. Number of donors considered for each site are indicated. P-values from the two-sided Wilcoxon rank-sum test comparing the distributions and adjusted for multiple testing within each panel. n.s., P>0.09; *P=0.017. h, Plot comparing the frequency of the top five expanded lung T_RM-biased clonotypes between lung and LLN CD8⁺ T cells. Donor ID and clonotype rank (according to lung frequency) are indicated by color and shape, respectively. Axis spacing reflects a pseudo-logarithmic transformation (log₁₀(x + 1)). Boxplots in (a), (b), (d), (f) and (g) represent median, IQR, and 1.5x IQR whiskers; each dot represents a donor (up to 36 in each box), except for (d) where each dot denotes individual TCR clonotypes shared between lung and LLN across 36 donors.

Source data

We showed that TCR clonotypes biased to recirculating non-T_RM subsets in the lung (as positive controls), especially those biased to the GZMK⁺ subset, displayed substantial sharing with CD8⁺ T cells in the LLN (Fig. 2g–i and Extended Data Fig. 4c–g). These results suggest that despite sufficient sample size and cell numbers to sensitively detect clonal sharing between the two anatomical compartments, lung T_RM clones display substantially less sharing than GZMK⁺ subset-biased clones (Fig. 2h and Extended Data Fig. 4e). Even when we restricted our analysis to clonal sharing involving T_RM-biased clones in the lung and T_RM cells in the LLN, we identified several instances in which the top two most abundant lung T_RM-biased clones were not shared, whereas less abundant donor-matched clones (for example, those ranked third or lower) were nonetheless shared with LLN cells (Extended Data Fig. 4h and Supplementary Table 4). These results suggest that the small size of the T_RM subset in the LLN does not preclude sensitive detection of clonotype sharing between T_RM cells in the lungs and LLN.

Reanalysis of published single-cell datasets from lung and LLN of individuals without cancer^16,17 showed concordance with the lung T cell subsets identified in our cohort (Supplementary Fig. 2a–c and Supplementary Table 2). However, low lung T cell numbers (median 194 cells; Supplementary Fig. 1a) or small cohort size (n = 2; ref. ¹⁶) in these datasets limited the sensitivity of our analysis to detect significant differences in clonotype sharing between lung T_RM and recirculating non-T_RM subsets with LLN T cells (Supplementary Fig. 2d). Taken together, our results suggest that the LLN is unlikely to represent the major reservoir of antigen-specific T_RM cells in the human lung. Further quantitative studies considering the absolute frequencies and TCR repertoires of T_RM cells across all lung-draining nodal stations will be necessary to confirm the minimal role of LLNs in the maintenance of T_RM cells in human lungs.

TCR-guided imputation of pathogen specificity to lung T cells

We utilized two complementary approaches to impute pathogen specificities to the TCRs expressed by lung T cells. First, to generate a large catalog of donor-matched TCR sequences specific to diverse pathogen epitopes, we employed an in vitro proliferation assay in which peripheral blood mononuclear cells (PBMCs) from each donor were independently stimulated with peptide MPs targeting nine common human pathogens²⁹ (Supplementary Tables 4 and 8 and Supplementary Methods). The proliferating CD4⁺ and CD8⁺ T cells were then sorted for bulk TCR sequencing of α and β chains separately (Fig. 3a,b, Supplementary Fig. 3a and Supplementary Table 4), and pathogen specificity was assigned to these TCRs based on their proliferation and response patterns (Supplementary Fig. 3b–e and Supplementary Methods). To validate this approach, we expressed 13 of the identified TCRs in an established reporter T cell line and demonstrated dose-dependent activation following stimulation with the corresponding pathogen peptide MPs (Fig. 3c, Supplementary Fig. 3f and Supplementary Table 4). By matching the lung TCR clonotypes to the catalog of pathogen-specific TCRs obtained from donor-matched PBMCs (Methods), we imputed pathogen specificity to ~2.7% and ~4.5% among all lung CD4⁺ and CD8⁺ T cells, respectively (range 0–66.4%) (Fig. 3d).

Fig. 3. TCR-guided imputation of pathogen specificities to lung T cells.

a, Experimental design for the in vitro proliferation assay in which PBMCs from each donor (n = 36) were stimulated with peptide MPs independently targeting 9 common human pathogens (Supplementary Methods). b, Bar charts show total numbers of pathogen-specific TCR sequences (unpaired α and β chains combined) from blood CD4⁺ and CD8⁺ T cells of donors in the lung cohort. c, Percentages of TCR-transduced Jurkat cell lines (n = 13) that expressed the NFAT–GFP reporter following stimulation with pathogen-specific peptides and irrelevant peptides at the indicated concentration; NFAT–GFP expression values were background-subtracted from dimethyl sulfoxide-treated conditions. Each line indicates an individual TCR (lung donor-matched) color-coded according to pathogen specificity and symbol-coded (CD4⁺ TCRs, triangle; CD8⁺ TCRs, circle). *P = 0.017; ***P = 0.00037. d, Percentages of lung T cells with assigned pathogen specificities based on the indicated strategies (color-coded); box plots represent median, IQR and 1.5× IQR (whiskers), with each dot representing a donor in the lung cohort (n = 40). e, Bar chart indicating total numbers of CD4⁺ and CD8⁺ TCR clonotypes colored according to the indicated source. f, Experimental design to detect paired TCR sequences of pathogen-reactive CD4⁺ T cells captured by the ex vivo stimulation ARTE assay applied to PBMC samples from donors in the TCR–resource cohort (n = 61). g, Bar charts showing total numbers of TCR clonotypes of CD4⁺ and CD8⁺ T cells specific to each indicated pathogen. h, Percentages of TCR-transduced Jurkat cell lines (n = 18) that expressed the NFAT–GFP reporter as described in c; each line indicates an individual CD4⁺ TCR (from the TCR–resource cohort), color-coded according to pathogen specificity. *P = 0.017; ***P = 0.00037. Symbols (triangles and circles) in c and h indicate median values across duplicate measurements for each condition, with vertical lines indicating the range. Statistical significance was assessed using two-sided Wilcoxon rank-sum test and adjusted for multiple testing. Irr., irrelevant; TCR-seq, TCR repertoire immune profiling. Schematics created in BioRender: a, Ehtram, A. https://biorender.com/m2xn0wz (2026); f, Ehtram, A. https://biorender.com/h4sp2gp (2026).

Source data

Second, we utilized TCR specificity information generated independent of our cohort. We compiled 146,465 TCRs with known pathogen specificity from publicly available databases^30–33, focusing mainly on 6 common human pathogens (Fig. 3e, Supplementary Fig. 4a,b and Supplementary Table 5). As the published TCR datasets almost entirely comprised MHC class I-restricted epitopes recognized by TCRs of CD8⁺ T cells (Fig. 3e and Supplementary Fig. 4c), we experimentally generated an additional large-scale resource of TCRs expressed on CD4⁺ T cells that recognized MHC class II-bound epitopes. Utilizing a large resource of PBMCs collected from an independent cohort of 61 healthy individuals (hereafter, the TCR–resource cohort; Supplementary Table 1), we performed a short-term antigen-reactive T cell enrichment (ARTE) assay³⁴ to capture ex vivo T cells reactive to MPs targeting 10 common human pathogens and vaccine antigens and then generated full-length paired α and β chain TCR sequences for >800,000 putative pathogen-specific T cells (Fig. 3f and Supplementary Table 2). Employing a statistical framework, we assigned pathogen specificities to each TCR sequence, resulting in a curated reference of 45,353 TCRs of CD4⁺ T cells that were specific to common human pathogens (Fig. 3e,g, Supplementary Fig. 4d–f and Supplementary Table 5). More than 99% of these TCRs had not previously been reported in the published databases, and ~96% of the 384 overlapping TCRs showed concordant pathogen specificity (Supplementary Table 5). To experimentally validate this approach, we expressed 18 of these newly reported TCRs in the T reporter cell line and demonstrated dose-dependent activation following stimulation with corresponding pathogen peptide MPs (Fig. 3h, Supplementary Fig. 4g–i and Supplementary Table 5). Finally, we used these datasets of TCRs with known pathogen specificities to infer the pathogens recognized by the lung TCRs that we originally sequenced (integrative approach in Fig. 3d and Methods). Thereby, we imputed the pathogen specificity of ~11% and 15% of all lung CD4⁺ and CD8⁺ T cells, respectively (range 4–70%; Fig. 3d, Extended Data Fig. 5 and Supplementary Table 6). Overall, our findings indicate that a sizable fraction of memory T cells in the human lung display specificity to common human pathogens.

Extended Data Fig. 5. An integrative strategy to assign pathogen specificity to lung and LLN T cells.

a, Area under the receiver operating characteristic (ROC) curve (AUC; y-axis) determined by the comparison framework for each evaluated tool (color-coded) when applied to the indicated groups of pathogen-specific clonotypes (quadrants). TCR clonotype pools were subsetted by the indicated relative frequency percentiles (x-axis). The accompanying table lists the tool rank inferred based on performance (AUC). b, Schematic overview of the integrative strategy developed to assign pathogen specificities to lung and LLN T cells on the basis of TCR data. c, Percentage of TCR chains with conflicting specificity between pathogens pairs, assessed separately for α and β chains (Methods). Boxplots represent median, IQR, and 1.5x IQR whiskers; each dot represents a combination of two different pathogens in the CD4⁺ TCR clonotype resource (n = 90). P-values from the two-sided Wilcoxon signed-rank paired test. ****P=2.83 × 10⁻¹⁶. d, For the reference match approach (Methods), bars indicate the aggregated numbers across donors in the study cohort (n = 40) of lung CD4⁺ (top) and CD8⁺ (bottom) TCR clonotypes intersected with reference TCRs in the indicated fashions. e,f, Aggregated numbers of lung TCR clonotypes (k) or T cells (l) (CD4, top; CD8, bottom) with antigen specificity inferred by the indicated approaches (Methods).

Source data

Pathogen-specific CD4⁺ T_RM cells are prevalent in human lungs

Based on the TCR sequences of lung and LLN CD4⁺ T cells, we estimated the frequency of a broad range of pathogen-specific T cells. Several of the pathogen-specific TCR clonotypes showed substantial clonal expansion in the lungs compared with the LLN, confirming the presence of expanded populations of pathogen-specific CD4⁺ memory T cells in the human lung (Fig. 4a and Extended Data Fig. 6a,b). For all five of the common respiratory viruses examined (SARS-CoV-2, IAV, parainfluenza virus, respiratory syncytial virus and metapneumovirus), we detected memory CD4⁺ T cells in the majority of individuals in our cohort, with frequencies ranging between ~0.1% and ~56% among all lung CD4⁺ T cells across donors and pathogens (median ~0.7%; Fig. 4a and Extended Data Fig. 6b). Comparable frequencies (median ~0.9%, range 0.1–21%) were also observed for CD4⁺ T cells specific to the common herpesviruses (CMV and EBV). We then focused on lung CD4⁺ T cells specific to the Bordetella pertussis bacterium, which causes whooping cough, differentiating vaccine and nonvaccine antigens using two separate peptide MPs³⁵: one targeted antigens included in the acellular pertussis (aP) vaccine, and the other encompassed the top 170 immunodominant non-aP vaccine epitopes. Lung memory CD4⁺ T cells specific to both types of antigen were detected, with relatively higher frequencies for those targeting vaccine antigens (Extended Data Fig. 6c). We also found TCR clonotypes among lung CD4⁺ T cells that were specific to antigens from a common fungus, Aspergillus fumigatus, which is known to cause life-threatening infections in immunocompromised patients³⁶.

Fig. 4. Pathogen-specific CD4⁺ T_RM cells are prevalent in human lungs.

a, Top: percentages across 40 donors of lung CD4⁺ T cells specific to the indicated pathogens, color-coded by pathogen specificity. Bottom: bar plot showing total numbers of pathogen-specific CD4⁺ T cells in the lung, with each bar representing a donor (n = 40). Bars are subdivided into stacked segments based on TCR clonotype frequency, which are color-coded according to pathogen specificity. b, Fractions of pathogen-specific CD4⁺ T cells corresponding to the indicated lung CD4⁺ T cell subsets (color-coded) across donors. c, Heatmap (middle) displaying the percentage of pathogen-specific CD4⁺ T_RM cells for each donor (rows) and pathogen (columns). Only donor–pathogen combinations with at least ten assigned CD4⁺ T cells were considered; donors with fewer than ten CD4⁺ T cells for a given pathogen are shown in black and were excluded from the summary analysis. Box plots (top) summarize these fractions across donors. Left: total number of pathogen-specific T_RM cells per donor, color-coded by pathogen; right: percentage of T_RM cells among all CD4⁺ T cells per donor. d, Fraction across donors (n = 38) of lung CD4⁺ T_RM and T_EM (CTL) cells corresponding to the indicated pathogen specificities. e, Correlations between the frequency of pathogen-specific lung CD4⁺ T_RM cells (x axis) and selected donor variables (y axis), including proportions of lung CD4⁺ T cell subsets and demographic and clinical characteristics. Dots indicate significant correlations (P < 0.01), colored by Spearman coefficient values. f, Principal component analysis of the transcriptomes of pathogen-specific CD4⁺ T_RM cells in the lung, pooled for each donor and pathogen according to the average expression of highly variable genes (Supplementary Methods). Each dot represents a T_RM subset from a given donor (colored symbols; top) that was specific to a pathogen (color-coded; bottom). g, Seurat-normalized mean expression (color scale) and percentages of cells (dot size) expressing transcripts encoding effector cytokines IFNγ and TNF in CD4⁺ T_RM cells specific to the indicated pathogens. Box plots in a and c show the median, IQR and 1.5× IQR (whiskers). Vax., vaccine.

Source data

Extended Data Fig. 6. Pathogen-specific CD4⁺ T_RM cells are prevalent in human lungs.

a, (Top) Percentages across 36 donors of LLN CD4⁺ T cells specific to the indicated pathogens; color-coded based on pathogen specificity. (Bottom) Bar plot shows the total number of pathogen-specific CD4⁺ T cells in the LLN, with each bar representing a donor (n = 36). Bars are subdivided into stacked segments based on TCR clonotype frequency; clonotypes are color-coded based on pathogen specificity. b, Fractions of lung CD4⁺ T cells comprised by each pathogen specificity, as indicated by column colors. A color upper bound was set based on the 95 percentile. White indicates absence of T cells specific to a given pathogen. c, Comparison of B. pertussis-specific lung CD4⁺ T cell percentages between acellular pertussis vaccine (aP) and non-aP (R, for “Rest”) antigens (n = 40 donors). P-values from the two-sided Wilcoxon signed-rank paired test. ***P=0.00017. d, Box plots show the percentage of lung T_RM cells among pathogen-specific CD4⁺ T cells across donors, stratified by sampling site (lobe; numbers indicate maximum count per site). Donors with <10 CD4⁺ T cells for a given pathogen were excluded from this analysis. e, Morisita-Horn (similarity) index for donor-matched TCR repertoires (n = 35), comparing pathogen-specific T_RM-biased lung CD4⁺ T cells and LLN CD4⁺ T cells. Reference values (labelled as ref. ctrl.), representing maximal similarity, were obtained by comparing the two halves of a random split of the combined lung and LLN CD4⁺ TCR repertoires for each donor. f, UMAPs of lung and LLN CD4⁺ T cells showing selected pathogen-specific clones (color-coded) with point size indicating relative frequency of the indicated clones; the size scale is consistent across anatomic sites. g, Density on UMAP of CD4⁺ T cells in the lung, either across the board or segregated based on pathogen-specificity. h, Heatmap shows adjusted P values from two-sided Wilcoxon signed-rank tests comparing pathogen-specific lung CD4⁺ T_RM cell fractions between groups defined by the indicated categorical donor variables; all values shown in the plot are non-significant (P>0.05). i, Box plots show proportion of lung CD4⁺ T_RM cells among total lung CD4⁺ T cells in donors positive (n-18) or negative (n = 22) for the HLA-DQB1*03:01 allele. P-value from the two-sided Wilcoxon rank-sum test; n.s., adjusted P=0.091 (Methods). j, UMAP plot showing the exact localization of functionally validated CD4⁺ T cell clones specific to the indicated pathogens (color-coded). k, Composite panel includes (left) a table listing details of functionally validated lung CD4⁺ TCR clonotypes, and (right) a bar plot displaying the number of lung CD4⁺ T cells expressing each clonotype, stacked by subset identity. Boxplots in (a), (c), (d), (e) and (i) represent median, IQR, and 1.5x IQR whiskers; each dot represents a donor (up to 40 in each panel).

Source data

Phenotypic analysis of pathogen-specific lung CD4⁺ T cells revealed that a large fraction exhibited features of T_RM cells (Fig. 4b,c), regardless of the lung region under assessment (Extended Data Fig. 6d). As expected, several highly expanded respiratory-pathogen-specific T_RM-biased clones in the lungs showed limited or no sharing with the LLN (Extended Data Fig. 6e,f). The frequencies of pathogen-specific T_RM cells varied substantially across donors (Fig. 4c), with no preferential enrichment for any particular pathogen (Fig. 4c,d and Extended Data Fig. 6g) and no significant associations detected with age, sex, body mass index, smoking or vaccination status (Fig. 4e and Extended Data Fig. 6h). However, the prevalence of pathogen-specific CD4⁺ T_RM cells was strongly correlated with the overall abundance of CD4⁺ T_RM cells (Fig. 4e). Although the sample size (40 donors) limited the statistical power of the analysis to definitively assess associations with human leukocyte antigen (HLA) types, donors carrying the HLA-DQB1*03:01 allele showed a trend toward higher frequencies of lung CD4⁺ T_RM cells compared to other donors (Extended Data Fig. 6i and Supplementary Table 7). Together, these findings suggest that donor-intrinsic factors governing T_RM cell generation and retention may have a key role in shaping the abundance of pathogen-specific CD4⁺ T_RM cells in the lung.

Although pathogen specificity did not influence the preferential acquisition of a T_RM phenotype, it did affect the generation of CD4⁺ T_EM cells (cytotoxic T lymphocytes; CTLs), indicating that certain pathogens, such as CMV, preferentially drive CD4⁺ CTL accumulation in the lungs (Fig. 4d and Extended Data Fig. 6f,g). Notably, the experimentally validated CMV-specific TCRs comprising more than 250 lung CD4⁺ T cells also showed the CD4⁺ CTL phenotype in the lung (Extended Data Fig. 6j,k), consistent with previous reports describing this as the predominant phenotype among circulating CMV-specific CD4⁺ T cells^37,38.

Pathogen-specific T_RM cells did not cluster according to pathogen specificity; instead, many instances of donor-specific clusters of cells were observed (Fig. 4f). These results suggest that donor-intrinsic factors have a dominant role in shaping the transcriptional properties of pathogen-specific T_RM cells, as well as in shaping their abundance. Transcripts encoding key effector cytokines TNF and IFNγ, which are known to be constitutively expressed by CD4⁺ T_RM cells³⁹, also showed no major differences among distinct pathogen-specific T_RM cells (Fig. 4g). Finally, we reasoned that if pathogen-specific T_RM cells were to decline rapidly after infection, we would find T_RM cell responses only to those from recent exposures. Contrary to this expectation, many individuals harbored substantial numbers of T_RM cells specific to multiple respiratory viruses (Fig. 4c,d and Extended Data Fig. 6b), all of which are unlikely to reflect recent infections. Taken together, these findings suggest that the human lung serves as a long-term reservoir of pathogen-specific CD4⁺ T_RM cells generated by past infections.

CD8⁺ T_RM cells with enhanced tissue-residency properties

Given that the MHC class I-restricted epitopes recognized by CD8⁺ TCRs found in publicly available databases were predominantly derived from CMV, EBV, SARS-CoV-2 and IAV (Fig. 3g), we focused on lung CD8⁺ T cells that were specific to these viruses. Similar to our observations in CD4⁺ T cells, memory CD8⁺ T cells specific to these viruses were detected in the lung and LLN (Fig. 5a and Extended Data Fig. 7a,b). Pathogen-specific CD8⁺ T cells in the lung belonged to three major populations: T_RM cells, T_EM cells, and a recently described population of T cells expressing high levels of GZMK (GZMK⁺ T cells)⁴⁰ (Fig. 5b,c and Extended Data Fig. 7c,d). CMV-specific CD8⁺ T cells were enriched for the T_EM phenotype and EBV-specific CD8⁺ T cells for the GZMK⁺ phenotype, whereas T_RM cells showed comparable representation of all four viruses (Fig. 5b,c and Extended Data Fig. 7e–h). Although highly expanded respiratory-pathogen-specific T_RM-biased clones in the lungs showed limited or no sharing with the LLN repertoire, several recirculating non-T_RM-biased clones, including EBV-specific GZMK⁺ subset-biased clones, showed significant sharing with the LLN TCR repertoire (Extended Data Fig. 7h,i). As for CD4⁺ T_RM cells, the frequency of pathogen-specific CD8⁺ T_RM cells was correlated with overall abundance of CD8⁺ T_RM cells (Fig. 5d), and their transcriptomes did not cluster according to pathogen specificity (Extended Data Fig. 7j), suggesting that donor-intrinsic factors rather than pathogen-specific cues shape the T_RM phenotype in the lung.

Fig. 5. CD8⁺ T_RM cells with enhanced tissue-residency properties.

a, Top: percentages across 40 donors of lung CD8⁺ T cells specific to the indicated pathogens, color-coded by pathogen specificity. Bottom: bar plot showing total numbers of pathogen-specific CD8⁺ T cells in the lung, with each bar representing a donor (n = 40). Bars are subdivided into stacked segments based on TCR clonotype frequency, which are color-coded according to pathogen specificity. b, Density on UMAP of CD8⁺ T cells in the lung, either across the board or segregated by pathogen-specificity. c, Heatmap (middle) displaying the percentage of pathogen-specific CD8⁺ T cells with the indicated phenotypes for each donor (rows) and pathogen (columns). Only donor–pathogen combinations with at least ten assigned CD8⁺ T cells were considered; donors with fewer than ten CD8⁺ T cells for a given pathogen are shown in black and were excluded from the summary analyses. Box plots (top) summarize these fractions across donors. Left: total number of pathogen-specific T cells per donor, color-coded by pathogen; right: percentages of cells with the indicated phenotypes (color-coded) among all CD8⁺ T cells per donor. d, Correlations between the frequency of pathogen-specific lung CD8⁺ cells within the indicated subsets (x axis) and selected donor variables (y axis), including the proportions of lung CD8⁺ T cell and T_RM subsets and demographic and clinical characteristics. Dots indicate significant correlations (P < 0.01), colored by Spearman coefficient values. e, Single-cell transcriptomes of CD8⁺ T_RM cells in the lung displayed by UMAP. Unbiased clustering (top) of 77,490 CD8⁺ T_RM cells from 40 donors, colored by cluster; labels indicate assigned T_RM subsets. UMAP (bottom) shows Seurat-normalized expression levels of HLA-DRB1 in CD8⁺ T_RM cells. f, Seurat-normalized mean expression (color scale) and percentages of cells (dot size) expressing selected transcripts significantly enriched (FDR ≤ 0.05) in HLA-DR^hi CD8⁺ T_RM cells compared to the remainder of T_RM cells. g, Box plots summarizing across donors the percentage of HLA-DR^hi CD8⁺ T_RM cells among all CD8⁺ T_RM cells and among CD8⁺ T_RM cells specific to the indicated pathogens. Box plots in a, c and g show the median, IQR and 1.5× IQR (whiskers).

Source data

Extended Data Fig. 7. CD8⁺ T_RM cells with enhanced tissue-residency properties.

a, (Top) Percentages across 36 donors of LLN CD8⁺ T cells specific to the indicated pathogens; color-coded based on pathogen specificity. (Bottom) Bar plot shows the total number of pathogen-specific CD8⁺ T cells in the LLN, with each bar representing a donor (n = 36). Bars are subdivided into stacked segments based on TCR clonotype frequency; clonotypes are color-coded based on pathogen specificity. b, Fractions of lung CD8⁺ T cells comprised by each pathogen specificity, as indicated by column colors. A color upper bound was set based on the 95 percentile. White indicates absence of T cells specific to a given pathogen. c, Summary across donors of the fractions of pathogen-specific CD8⁺ T cells corresponding to each lung CD8⁺ T cell subset. d, Box plots summarize across donors the percentage of lung T cells in the indicated subsets among pathogen-specific CD8⁺ T cells, stratified by sampling site (lobe; numbers indicate maximum count per site). Donors with <10 CD8⁺ T cells for a given pathogen were excluded from this analysis. e, Fraction across donors (n = 40) of lung CD8⁺ T_RM, T_EM and GZMK⁺ T cells corresponding to the indicated pathogen specificities. f, UMAP plot showing the exact localization of functionally validated CD8⁺ T cell clones specific to the indicated pathogens (color-coded). g, Composite panel includes (left) a table listing details of functionally validated lung CD8⁺ TCR clonotypes, and (right) a bar plot displaying the number of lung CD8⁺ T cells expressing each clonotype, stacked by subset identity. h, UMAPs of lung and LLN CD8⁺ T cells showing selected pathogen-specific clonotypes (color-coded) with point size indicating relative frequency of the indicated clones; the size scale is consistent across anatomic sites. i, Comparison of Morisita-Horn (similarity) indices for donor-matched lung and LLN TCR repertoires in the indicated pathogen-specific CD8⁺ T cell subsets (31 for T_RM; 30 for T_EM; 33 for GZMK⁺). Reference values (labelled as ref. ctrl.), representing maximal similarity, were obtained by comparing the two halves of a random split of the combined lung and LLN pathogen-specific CD8⁺ TCR repertoires for each donor (n = 36). P-values from the two-sided Wilcoxon rank-sum test and adjusted for multiple testing. **P < 0.007; ****P < 3.7 × 10⁻⁶. j, Principal component analysis of the transcriptomes of pathogen-specific CD8⁺ T_RM cells in the lung, pooled for each donor and pathogen based on the average expression of highly variable genes (Supplementary Methods). Each dot represents a T_RM subset from a given donor (colored symbols; left) that is specific to a pathogen (color-coded; right). k, Gene set enrichment analysis (GSEA) plot comparing the enrichment of the lung T_RM cell signature between cells in the lung CD8⁺ HLA-DR^hi T_RM subset against the rest of the lung CD8⁺ T_RM cells. Normalized enrichment score (NES) and P-value are shown. ***P < 0.001. Boxplots in (a), (d) and (i) represent median, IQR, and 1.5x IQR whiskers.

Source data

Subclustering of CD8⁺ T_RM cells identified a subset enriched for expression of transcripts encoding HLA-DR and other class II MHC receptors (Fig. 5e), a feature typically associated with recently activated effector T cells⁴¹. As individuals in our cohort reported no recent infections, and 39 of the 40 had not received a vaccination in the preceding 90 days (Supplementary Table 1), this finding suggests that the HLA-DR^hi CD8⁺ T_RM subset may accumulate in the lung independent of recent TCR activation, consistent with previous observations in circulating HLA-DR⁺ CD8⁺ T cells from both HIV-1-infected and HIV-1-uninfected individuals^42,43. The HLA-DR^hi CD8⁺ T_RM subset was enriched for transcripts encoding transcription factors BATF and CSL (encoded by RBPJ) (Fig. 5f), which have been linked to the development and maintenance of T_RM cells^44–46. Notably, Notch signaling also contributes to constitutive expression of IFNγ by T_RM cells⁴⁴ and, accordingly, IFNG transcripts were significantly increased in the HLA-DR^hi CD8⁺ T_RM subset compared to the other CD8⁺ T_RM cells (Fig. 5f). Other T_RM-associated molecules whose transcripts were upregulated in the HLA-DR^hi CD8⁺ T_RM subset included the CD103 integrin α-subunit (encoded by ITGAE)^3,44 and fatty acid-binding protein 5 (FABP5)⁴⁷ (Fig. 5f). Furthermore, gene set enrichment analysis revealed significant enrichment of a lung T_RM gene signature in pathogen-specific HLA-DR^hi CD8⁺ T_RM cells, suggesting enrichment of tissue-residency features (Extended Data Fig. 7k).

The frequency of these pathogen-specific HLA-DR^hi CD8⁺ T_RM cells varied substantially across donors, with no preferential enrichment for any particular pathogen (Fig. 5g) and no significant associations detected with the assessed host factors except age, which was positively associated with the frequency of SARS-CoV-2-specific HLA-DR^hi T_RM cells (Fig. 5d). This suggests that other as-yet-unidentified donor factors may shape interindividual variability in this T_RM subset. Overall, these results indicate that the human lung harbors a population of pathogen-specific HLA-DR^hi CD8⁺ T_RM cells that exhibit several features associated with enhanced tissue residency and functional features.

Persistence of lung T_RM cells in humans

To definitively establish the long-term persistence of lung T_RM cells in humans, we performed longitudinal analyses of T cells isolated from distinct regions of uninvolved lungs across 6 individuals who required either completion surgery or resection of a second primary cancer several months to years later (range 7–304 weeks) (Extended Data Fig. 8). A large fraction of lung T cells, including lung T_RM cells (median ≥50% across donors for both CD4⁺ and CD8⁺ T_RM cells), shared TCR clonotypes between longitudinal samples from the same individuals (Fig. 6a and Extended Data Fig. 9a). As expected, T_RM cells with highly expanded TCR clonotypes (clone size ≥10 cells) showed more sharing across longitudinal samples, indicating long-term persistence of certain antigen-specific lung T_RM clones over years (Fig. 6b and Extended Data Fig. 9b). For example, in the donor with the longest sampling interval (>5 years, X06), more than 50% of highly expanded T_RM clones at baseline were detected in lung samples collected 5 years later (Fig. 6b,c). Although TCRs of T_EM-biased clones in the lung were shared with memory T cells in the blood, the majority of the T_RM-biased clones were not shared with the blood, suggesting that T_RM persistence in the lungs is largely independent of trafficking from the circulation (Fig. 6c,d and Extended Data Figs. 8f and 9f).

Extended Data Fig. 8. Annotation of T cell subsets from longitudinal lung samples.

a, Single-cell transcriptomes of lung CD4⁺ (left) and CD8⁺ (right) T cells displayed by uniform manifold approximation and projection (UMAP). Unbiased clustering of 75,567 CD4⁺ and 81,803 CD8⁺ T cells from longitudinal lung samples (n = 12 across 6 donors), colored by cluster; labels indicate assigned T cell subset. Bar charts (right) display the proportions of each T cell subset. b, Dot plots showing Seurat-normalized mean expression (color scale) and the percentage of cells (dot size) expressing the indicated markers across cells in each subset of lung CD4⁺ (left) and CD8⁺ (right) T cells. Data are pooled across samples (n = 12). c, GSEA with the indicated gene signatures (PMIDs shown), comparing each CD4⁺ (left) or CD8⁺ (right) lung T cell subset to the remaining cells. Heatmap summarizes normalized enrichment scores (NES) for significant comparisons (adjusted P < 0.05), with gray indicating non-significant results. d, Proportion of total lung CD4⁺ (top) or CD8⁺ (bottom) T cells represented by each subset. Each bar represents the cell composition in each donor from the study cohort (n = 12 across 6 donors), with square color indicating sampling time point. e, Distribution across donors (up to 6) in the indicated time points of the percentage of lung T cells expressing expanded clonotypes (that is, those observed in more than one cell) and the Gini indices, separately in lung CD4⁺ (top) and CD8⁺ (bottom) T cell subsets. Boxplots represent median, IQR, and 1.5x IQR whiskers. f, Single-cell transcriptomes of CD4⁺ (top) and CD8⁺ (bottom) T cells in blood displayed by UMAP. Unbiased clustering (left subpanels) of 23,748 CD4⁺ and 23,615 CD8⁺ T cells from 6 donors (samples corresponding to the baseline time point), colored by cluster. UMAPs (right subpanels) show Seurat-normalized expression of GZMB in blood T cells, which is enriched in T_EM cells.

Source data

Fig. 6. Persistence of lung T_RM cells in humans.

a, Schematic showing the lung region (blue color with dotted outline) and time interval between longitudinally collected samples (n = 12; 6 for each time point). Pie charts show the fraction of CD4⁺ (top) and CD8⁺ (bottom) T_RM cells with shared TCR clonotypes (light green) between baseline (left) and follow-up (right) samples. Rightmost bar plots display total numbers of CD4⁺ and CD8⁺ T cells analyzed per donor and time point. b, Fractions of CD4⁺ and CD8⁺ T_RM cells with highly expanded (n ≥ 10 cells; purple) or other (n < 10) TCR clonotypes (gray) that were shared between time points for each donor (n = 6); shape denotes the reference time point. c, For selected representative donors, CD4⁺ (left) and CD8⁺ (right) T cells expressing lung T_RM-biased clonotypes are shown on blood and lung UMAPs, with lung samples stratified by time point. Point size indicates the relative frequency of the indicated clones. d, Comparison of Morisita–Horn (similarity) indices for donor-matched baseline lung versus blood TCR repertoires (n = 6; top) or donor-matched baseline and follow-up lung TCR repertoires (n = 4, CD4; n= 6, CD8; bottom), stratified by the indicated T cell subsets. Cell numbers were normalized to the minimum across donor-matched samples. Reference values, representing maximal similarity, were obtained by comparing the two halves of a random split of the combined TCR repertoires of the compared condition for each donor. P values are from two-sided Wilcoxon signed-rank paired tests and adjusted for multiple testing: NS, P > 0.35; *P < 0.03, **P < 0.009. e, Correlations of relative frequency for shared lung T_RM-biased clones between time points in each donor. Significant correlations are indicated (P < 0.01; gray for nonsignificant correlation), with color denoting the Spearman coefficient. f, Log-transformed ratios of frequencies of T_RM-biased TCR clonotypes shared across time points. Each dot represents a clonotype, with size indicating relative frequency at baseline. g, Heatmaps displaying percentages of pathogen-specific CD4⁺ (left) and CD8⁺ (right) T cells with the indicated phenotypes, for each donor (columns) and pathogen (rows). Only donor–pathogen combinations with at least 10 assigned T cells were considered; donors with <10 cells for a given pathogen are shown in black and were excluded from the summary analyses. Box plots (right) summarize these fractions across donors. Box plots in d, f and g represent median, IQR and 1.5x IQR (whiskers). Schematic in a created in BioRender; Fajardo-Rosas, V. https://biorender.com/ph9ekyc (2026).

Source data

Extended Data Fig. 9. Persistence of lung T_RM cells in humans.

a, Schematic showing the lung region (blue color with dotted outline) and time interval between longitudinally collected samples (n = 12; 6 for each time point). Pie charts show the fraction of CD4⁺ (top) and CD8⁺ (bottom) T cells with shared TCR clonotypes between baseline (left) and follow-up (right) samples. b, Plot indicates the fraction of CD4⁺ (left) and CD8⁺ (right) T cells in the indicated subsets (color-coded) with shared TCR clonotypes between time points; shape denotes the reference time point. c, For selected representative donor, CD8⁺ T cells expressing lung TCR clonotypes biased to the indicated cell subsets are shown on blood and lung UMAPs, with lung samples stratified by time point. Point size indicates relative frequency of indicated clones, with a consistent scale across anatomic sites, time points and subsets. d, Scatter plots show correlations of relative frequencies for shared lung CD4⁺ (left) and CD8⁺ (right) TCR clonotypes between time points, across donors. Spearman correlation coefficients are indicated. ****P < 2.2 × 10⁻¹⁶. e, Dots denote individual T_RM-biased TCR clonotypes at baseline (size indicates relative frequency of clones) that are also detected at follow-up; the y-axis shows the fraction of cells in the T_RM subset at the later time point. f, Volcano plot showing differential gene expression between cells belonging to persistent and non-persistent clonotypes (Supplementary Methods) within the T_RM subset of a representative donor at baseline. g, Percentage of lung CD4⁺ and CD8⁺ T cells specific to the indicated pathogens; color-coded based on pathogen specificity. Boxplots represent median, IQR, and 1.5x IQR whiskers; each dot represents an individual sample (n = 12). h, Scatter plots show correlations of frequencies of pathogen-specific lung CD4⁺ (left) and CD8⁺ (right) T cells between time points, across donors. Each dot represents an individual donor-pathogen pair. Spearman correlation coefficients are indicated. ****P < 8.3 × 10⁻⁶. i, Plot indicates the fraction of CD4⁺ and CD8⁺ pathogen-specific T_RM cells with highly expanded (n>=10 cells; purple) or other (n<10) TCR clonotypes (gray) that are shared between time points; shape denotes the reference time point. j. For two representative donors (X04 and X06), CD4⁺ T cells expressing selected clonotypes specific to respiratory pathogens (color-coded as in (g)) are shown on lung and blood UMAPs, with point size indicating relative frequency of the indicated clones; the size scale is consistent across anatomic sites. Schematic in a created in BioRender; Fajardo-Rosas, V. https://biorender.com/ph9ekyc (2026).

Source data

Lung T cells with T_RM-biased TCR clonotypes at baseline largely maintained their tissue-resident phenotype and showed significant correlations in relative abundance across longitudinal samples, suggesting cell-intrinsic properties that sustain long-term phenotypic and clonal stability (Fig. 6e,f and Extended Data Fig. 9d,e). Notably, the expression of genes downstream of TCR signaling (TNFRSF9 and ENTPD1) did not differ significantly between the persistent and nonpersistent T_RM clones (Extended Data Fig. 9f and Supplementary Table 7), suggesting that ongoing TCR signaling is unlikely to account for their long-term persistence.

Consistent with our observations in the cross-sectional lung samples (Figs. 4b,c and 5c), many individuals in the longitudinal cohort harbored substantial numbers of respiratory-pathogen-specific T_RM cells (Fig. 6g, Extended Data Fig. 9g and Supplementary Table 6). Most importantly, we demonstrated that a large fraction of these pathogen-specific T_RM cells shared TCR clonotypes across longitudinal samples from the same individuals, indicating persistence over months to years (Extended Data Fig. 9h,i). For example, in donor X04 and X06, more than 50% of highly expanded pathogen-specific T_RM clones at baseline, including CD4⁺ T_RM clones specific to multiple respiratory viruses, were detected in longitudinally collected paired lung samples (Extended Data Fig. 9i,j and Supplementary Table 6). Overall, our analysis of longitudinal samples confirmed that T_RM clones, including those specific to common human pathogens, persist in the human lung.

Discussion

Our study supports the concept that the human lung serves as a reservoir of pathogen-specific T_RM cells. Memory T cells in the lung exhibited a high degree of clonal expansion compared with the LLN, and, in many individuals, a few TCR clonotypes dominated the repertoire, reflecting preferential accumulation of certain antigen-specific memory T cells. A substantial fraction of these cells were specific to common human pathogens, and highly expanded T_RM-biased TCR clonotypes exhibited restricted diversity, suggesting that they are largely tissue resident and not clonally related to other lung T cell subsets that reflect the circulating T cell pool. Several individuals harbored substantial T_RM populations specific to multiple respiratory viruses even in the absence of recent infection. Finally, longitudinal analysis of surgically resected lung tissue samples from the same donors revealed that a large fraction of T_RM clones persisted for months to years. Overall, these observations indicate that unlike in mouse models, in which lung T_RM cells wane^6,7 and the LLN may serve as a principal reservoir^8,9, the human lung can act as a long-term reservoir for pathogen-specific T_RM cells. This highlights fundamental differences in the maintenance of T_RM populations between species and indicates a need for caution when extrapolating findings from mouse models of lung infection and vaccination to humans. Unlike animal studies, studies involving living humans cannot be fully quantitative, as it is not ethically possible to obtain the entire lung and all its LLNs, nor is it technically possible to extract all cells with 100% efficiency. However, as we analyzed more than one million T cells uniformly processed from a large number of donor-matched lungs and LLNs, this study provides a foundational resource for understanding the properties, trafficking and persistence of a wide range of pathogen-specific lung T_RM cells in a real-world setting.

We speculate that anatomical differences in the bronchial and bronchiolar structure between humans and mice could account for the observed differences in lung T_RM density. This is based on previous studies of human lungs in which the majority of the CD8⁺ T_RM cells (CD103⁺ cells) were observed to be present in the bronchial/bronchiolar tissue (nongas-exchange compartment) and not the alveolar tissue, the gas-exchange compartment^48,49. Therefore, the accumulation and persistence of T_RM cells in the human lungs (bronchial compartment) may not adversely affect the gas-exchange function of the alveolar compartment, which is the claim made for the mouse lung. Notably, anatomical studies have shown that there are substantially more bronchial/bronchiolar divisions in the human lung compared to the mouse lung (>21 divisions versus ~13 divisions), such that there is substantially more bronchial/bronchiolar surface area in the human lungs^50,51. Given that specific-pathogen-free (SPF) mice cohoused with pet-store ‘dirty’ mice showed increased frequency of T_RM cells in the lungs and many other organs compared to SPF mice⁵², it is also likely that the natural microbial exposures in humans may explain part of the observed differences compared with laboratory SPF mice.

Several findings from our study indicate important caveats in extrapolating conclusions from previous studies regarding the persistence of respiratory-pathogen-specific T_RM cells in the human lungs^4,6,7. First, a substantial fraction of lung T cells consists of recirculating non-T_RM subsets. This includes T cells specific to a wide range of pathogens (Figs. 4a and 5a), which implies that the results of studies that fail to discriminate T_RM cells in the total or pathogen-specific lung T cell populations^12,53 are likely to be confounded by recirculating T cell subsets. Second, viruses capable of establishing persistent infections, such as EBV and CMV, give rise to large T_RM clones in the lungs and, in some individuals, comprise a substantial fraction of the pathogen-specific T_RM compartment (Figs. 4c and 5c and Extended Data Figs. 6f and 7h). This finding has important implications for the interpretation of previous studies^13,15–17, as the mere detection of T_RM cells in the lungs, in the absence of antigen-specific resolution, cannot be assumed to reflect durable T_RM populations generated by acute respiratory infections. Rather, persistent viral infections may also shape the composition and apparent longevity of the lung T_RM pool. Third, our imputation-based approach to define the spectrum of virus-specific T cells in the lungs revealed that a single pathogen, such as SARS-CoV-2 or IAV, could generate both T_RM-biased and non-T_RM-biased clones within the same individual (Extended Data Figs. 6f and 7h and Supplementary Table 6). In certain individuals, the responses were skewed toward non-T_RM-biased clones, highlighting substantial intraindividual heterogeneity in tissue residency (Extended Data Fig. 7h). These findings suggest that previous studies^18–21 that have relied on tetramer staining of a single viral epitope may have underestimated the diversity of the T cell repertoire and misrepresented the prevalence of tissue-resident populations, potentially yielding a biased view of pathogen-specific immunity in the lung. Finally, our reanalysis of published datasets^16,17, particularly our comparison of TCR repertoires across different T cell subsets in lung and LLN, demonstrated that the large number of donors and higher T cell counts in our study reduced sampling biases and improved the sensitivity of clonal sharing analyses compared with previous work.

Although our dataset was large in terms of cellular resolution and depth, the size of the overall study cohort remained modest (n = 40 donors). Expanding the cohort will be important for exploration of the roles of genetic and host factors that influence the magnitude of pathogen-specific lung T_RM cells. The underlying diagnosis of lung cancer may influence the nature of memory T cell responses in the uninvolved lung tissue, although our reanalysis of the single-cell transcriptomes of lung T cells from a cohort of cancer-free individuals showed concordant results. Another limitation was that the study cohort primarily reflected the baseline state of lung tissue, as most donors were infection-free and had not been recently vaccinated at the time of sample collection. Although study of acutely infected individuals is not feasible owing to the requirement for surgical tissue resection, individuals undergoing surgery after recent vaccination could provide an opportunity to examine vaccine-elicited T cell responses in the lungs. Overall, our study demonstrates the power of TCR-based imputation to uncover pathogen-specific T cell responses in the human lung, establishing a framework to systematically map antigen-specific tissue-resident immunity across the human body, with implications for understanding infection, vaccination and immune-mediated disease in humans.

Methods

Human samples

Ethical approval for this study was obtained from the South Central – Hampshire A Research Ethics Committee (14/SC/0186) and the Ethics Committee of La Jolla Institute for Immunology (LJI). Written informed consent was obtained from all participants before sample collection. Individuals did not receive compensation for participation in this study.

Two cohorts of a total of 46 individuals undergoing standard-of-care surgical resection for primary lung nonsmall-cell lung cancer at Liverpool Heart and Chest NHS Foundation Trust Hospital and University Hospital Southampton NHS Foundation Trust were enrolled. The cross-sectional (lung) cohort (n = 40) included 16 men and 24 women, all of white British ethnicity, with a median age of 73 years (range 61–83). The longitudinal cohort comprised 6 individuals who either required completion surgery or developed recurrence after their initial surgery and therefore required a second lung resection several months to years later (median 35 weeks, range 7–304). The cohort included 3 men and 3 women, all of white British ethnicity, with a median age of 71.5 years (range 67–78). Clinical and demographic metadata are provided in Supplementary Table 1. Lung tissue, along with matched peripheral blood and LLN (the latter was available for the lung cohort only), was collected intraoperatively. Following lobectomy, the surgical team followed a standardized protocol to collect the macroscopically uninvolved, nontumor-bearing regions of the lung parenchyma. To ensure that assessment of the surgical tumor margin was not compromised, the uninvolved lung tissue was typically obtained at least 5 cm away from the tumor margin. Although lung tissues were obtained from different lung lobes (Extended Data Fig. 1a), the lung parenchyma was uniformly resected to ensure consistent representation of small airways and alveoli in all cases. The lung tissue volume obtained was dependent on the tumor size and the resected lobe, with a median volume of 3 cm³ (range 1.2–15 cm³) (Supplementary Table 1). Lung tissues were processed immediately by dicing in <1-mm³ pieces and cryopreserved in multiple cryovials.

The surgical team obtained tumor-uninvolved lymph nodes that drained the resected lung lobe. In this study, LLN refers specifically to the hilar and mediastinal lymph nodes (per International Association for the Study of Lung Cancer nomenclature)that constitute the direct lymphatic drainage pathway of the resected lung lobe. As detailed in Extended Data Fig. 1a and Supplementary Table 1, LLNs were obtained from stations 4, 5, 7, 8, 9, 10 and 11. The specific LLN stations collected were not identical across all donors, as they were anatomically matched to the noninvolved nodes draining the resected lung lobe. In the majority of the patients (n = 23), the subcarinal or hilar nodes (stations 7 and 10, respectively) were obtained. In other cases, station selection depended on the resected lobe and lack of tumor involvement; for example, subaortic nodes (station 5; n = 3) were analyzed for patients undergoing left lower lobe resections, and right paratracheal nodes (station 4; n = 2) were analyzed for some of the patients undergoing right upper lobe resections. The size of resected lymph nodes ranged from 0.2 cm³ to 0.9 cm³ (median 0.4 cm³). All samples were immediately processed by dicing into small pieces and cryopreserved in freezing medium (90% fetal bovine serum (FBS) and 10% dimethyl sulfoxide) before downstream processing.

For the TCR–resource cohort, PBMCs were obtained from a total of 61 healthy donors through the San Diego Blood Bank between December 2021 and February 2022. Each donor provided ~8 ml of leukoreduction system whole-blood collection material. Donor ages ranged from 27 to 79 years (median 65). Of the 61 donors, 47 were male, 12 were female, and 2 had unspecified sex information. No clinical or immunological abnormalities were reported for these donors at the time of collection. PBMCs were isolated and cryopreserved according to standard protocols for downstream immunological assays.

HLA typing was performed for class I (HLA-A, HLA-B and HLA-C) and class II (HLA-DRB1, HLA-DRB3/4/5, HLA-DQA1/DQB1 and HLA-DPB1) alleles in blood samples from the lung cohort. Typing was carried out by an ASHI-accredited laboratory at Murdoch University, as described in ref. ⁵⁵.

Sample processing

Cryopreserved human lung tissue and LLN samples were thawed in a 37 °C water bath for ~5 min until only a small ice fragment remained, followed by washing with FBS and dilution in MACS buffer. The thawed tissue suspension was filtered through a 70-µm strainer; retained tissue pieces were collected and, in the case of lung tissue, finely diced using a scalpel. The minced tissue was transferred to a gentleMACS C tube (Miltenyi Biotec, 130-096-334) and supplemented with RPMI, after which enzymatic digestion was initiated by addition of DNase I (Sigma, D4527-200K) and Liberase (Sigma, 5401127001). The sample was subjected to mechanical and enzymatic dissociation using an OctoDissociator (Miltenyi Biotec) set at 37 °C for 15 min. Following dissociation, the cell suspension was filtered through a MACS SmartStrainer, with gentle pressing of residual tissue to maximize cell yield. For the single-cell analysis of circulating memory T cells (longitudinal cohort), cryopreserved PBMCs (n = 6) were thawed and washed twice with phosphate-buffered saline. The resulting cell suspensions (across all sample types) were centrifuged, and the pellet was resuspended in MACS buffer containing human FcR Blocking Reagent (Miltenyi, 130059901), followed by addition of Cell-hashtag TotalSeq-C antibody (1 μg per condition) to enable hashtag-oligonucleotide-based sample deconvolution at the analysis stage. Cells were then stained at 4 °C for 30 min with a panel of antibodies, as listed in Supplementary Table 8. After washing, the cells were underlaid with FBS and centrifuged again. The final cell pellet was resuspended in cold MACS buffer supplemented with DAPI for viability staining, and cells were sorted for CD3⁺ T cells (from 40 cross-sectional lung and 8 LLN samples) or CD4⁺ and CD8⁺ T cells (from the 6 longitudinal lung and PBMC samples and the 28 LLN samples) on a FACSAria Fusion Cell Sorter (Becton Dickinson) and processed for downstream single-cell sequencing. Flow cytometry data were analyzed using FlowJo software (v.10).

Cell isolation and single-cell RNA sequencing with the 10x platform

Single-cell RNA sequencing (scRNA-seq) library preparation was performed alongside TCR repertoire immune profiling and cell surface protein detection to enable hashtag-oligonucleotide-based sample demultiplexing following the manufacturer’s protocol (10x Genomics).

Single-cell transcriptomic analysis

Sequencing reads from scRNA-seq data were demultiplexed using Cell Ranger (v.7.1.0, 10x Genomics) with the mkfastq module and aligned to the GRCh38 reference genome (prebuilt reference 2020-A) with the count module. This generated unique molecular identifier (UMI)-collapsed gene expression matrices for 79 single-cell libraries (Supplementary Table 2). Independent gene expression libraries were aggregated into combined datasets using the aggr module, yielding four datasets (Supplementary Table 2): (i) lung CD3⁺ T cells from 40 donors (lung cohort), (ii) LLN CD3⁺ T cells from 38 donors (lung cohort), (iii) blood CD3⁺ T cells from 6 donors at baseline (longitudinal cohort) and (iv) lung CD3⁺ T cells from 12 longitudinal samples across 6 donors (longitudinal cohort) and 8 unrelated samples that were pooled in the same 10x libraries. Two donors (L034 and L040) were excluded from downstream analyses because of the very low numbers of LLN T cells (<100) recovered, resulting in an LLN dataset comprising 36 donors. Unrelated samples were removed.

Aggregated datasets were processed in R (v.3.6.1) using Seurat (v.3.2.3)⁵⁶. Genes expressed in fewer than 0.1% of cells were excluded. To minimize doublets and eliminate low-quality transcriptomes while allowing variability in the distribution of UMI and gene counts, we excluded cells that did not meet the following quality control criteria: 500 ≤ UMI count ≤10,000; 400 ≤ gene count ≤3,000; and mitochondrial UMI percentage ≤15%. Data normalization was performed using NormalizeData with default parameters. Highly variable genes were identified with the FindVariableFeatures function using the vst method and default settings; selection was restricted to genes with a mean expression >0.01 and contributing 12% (CD8⁺ LLN dataset), 25% (blood datasets) or 30% (lung and CD4⁺ LLN datasets) of the standardized variance. Gene-wise scaling was performed using ScaleData, with regression of total UMI counts and mitochondrial UMI percentages (vars.to.regress argument), and default settings for other arguments. Although fluorescence-activated cell sorting batches showed similar quality-control-related feature variability within each aggregated dataset (Extended Data Fig. 1b), potential batch effects were further minimized using Harmony⁵⁷ (RunHarmony; v.0.1.1). Dimensionality reduction was performed via principal component analysis, followed by uniform manifold approximation and projection (UMAP) embedding using RunUMAP with the top 40 PCs, 15 neighbors (n.neighbors = 15) and a minimum distance of 0.1 (min.dist = 0.1). Unsupervised clustering was conducted with FindNeighbors and FindClusters, with a defined resolution, 30 principal components specified per iteration and all other parameters left unchanged. For the last iteration, a resolution parameter between 0.1 and 1 (by steps of 0.1) was selected for each dataset, defined as the value that maximized separation of expected T cell subsets while minimizing overclustering.

Subsets of CD4⁺ and CD8⁺ T cells were defined through iterative rounds of clustering and phenotype-guided filtering. In each round, contaminant clusters were reassigned to the appropriate lineage-specific dataset for further refinement. A total of 16, 14 and 6 rounds of deconvolution were performed for lung, LLN and blood T cells, respectively, after which no lineage-contaminating clusters could be discerned.

To ensure consistency in cell subset annotation and UMAP embeddings across lung datasets, CD4⁺ and CD8⁺ lung T cell datasets from the longitudinal cohort (75,567 and 81,803 cells, respectively) were mapped onto larger reference datasets from the lung cohort (370,233 and 227,800 cells for CD4⁺ and CD8⁺, respectively) using Seurat’s reference mapping workflow for scRNA-seq data⁵⁶. To minimize cohort-specific variation in TCR V and J gene usage, TCR genes were excluded from the set of highly variable genes, followed by recalculation of principal components in the reference datasets (RunPCA, using the same parameters as above). The recommended workflow was then applied as follows: (i) transfer anchors were identified using FindTransferAnchors (reduction = ‘pcaproject’, reference.reduction = ‘pca’, dims = 1:40); (ii) UMAP embeddings for the reference datasets were recomputed to reflect the updated highly variable gene set (RunUMAP, using the same parameters as above, with return.model = TRUE); and (iii) longitudinal datasets were mapped and annotated using MapQuery (reference.reduction = ‘pca’, reduction.model = ‘umap’). This procedure was performed separately for CD4⁺ and CD8⁺ T cells following lineage-specific deconvolution, as described above.

Expression levels of PTPRC isoforms (CD45RO and CD45RA) were quantified using IDEIS⁵⁸ on each scRNA-seq library, using the corresponding cell barcodes from Cell Ranger’s cell calling algorithm as a barcode whitelist, with all other parameters left unchanged.

Single-cell TCR sequence data analysis

Reads from TCR repertoire immune profiling data were processed using Cell Ranger vdj (v.7.1.0) with the prebuilt GRCh38 reference (vdj-GRCh38-alts-ensembl-7.1.0) and aggregated following previously described workflows^34,59. For the lung cohort, one aggregated dataset was generated for lung CD3⁺ T cells (n = 40 libraries, one per donor) and another for LLN CD3⁺ T cells (n = 21 libraries spanning 36 donors). For the TCR–resource cohort, 45 libraries derived from pathogen-reactive CD4⁺ T cells were aggregated into a single TCR dataset. For the longitudinal cohort, one aggregated dataset was generated for lung CD3⁺ T cells (n = 16 libraries spanning 12 samples across 6 donors) and another for blood CD3⁺ T cells (n = 2 libraries spanning 6 samples). Metrics from individual TCR libraries are presented in Supplementary Table 2. For single-cell TCR data, a TCR clonotype was defined as a unique set of CDR3 α and β chain sequences and the corresponding V and J gene segments.

TCR repertoire diversity was quantified using the Gini index (repDiversity, immunarch²⁶ v.0.9) based on the TCR clonotype definition above, for each donor. TCR repertoires comprised either all TCR clonotypes or those from a given T cell subset, defined by unbiased clustering or phenotype bias (Supplementary Methods). Indices were provided only for TCR repertoires with ≥50 T cells.

Similarity of pairs of donor-matched TCR repertoires was quantified using the Morisita–Horn index (index_morisita, tabula⁶⁰ v.3.3.2), calculated based on relative frequencies of clonotypes in each repertoire. The union of clonotypes across the two repertoires was used as the universe for each calculation. TCR repertoires were defined as described above, and figure panels specify whether clustering (cell subset) or phenotype bias subset was used as a filtering criterion. Reference values, to be used as positive controls representing maximal similarity between lung and LLN TCR repertoires, were obtained by comparing the two halves of a random split of the combined repertoires for each donor. Cell numbers were normalized to the minimum across donor-matched samples. Reference values were calculated separately for CD4⁺ and CD8⁺ T cells, yielding a CD4 and a CD8 reference control, respectively.

Individual strategies to assign pathogen specificities to T cells

We developed three complementary strategies to assign pathogen specificities to lung T cells on the basis of their TCRs, as follows.

The donor-matched, circulating, antigen-specific TCR-based (‘donor-matched’) approach relied on pathogen specificities derived from the peptide MP-induced in vitro proliferation assay conducted on donor-matched blood samples (‘assay’ TCRs). CDR3 amino acid sequences from lung TCR clonotypes were compared against those from the assay TCRs, matched by chain, donor and T cell lineage. When a lung TCR clonotype matched an assay TCR with either the CDR3 of the α or β chain alone, its pathogen specificity was directly assigned. When both chains matched, we determined whether the two chains yielded concordant or conflicting specificities. Among all lung TCR clonotypes with a specificity provided by the donor-matched approach, ~16% (585 of 3,656) and ~20% (183 of 906) matched both chains for CD4⁺ and CD8⁺ TCRs, respectively. Of these, the majority (~89% (517 of 585) and 96% (176 of 183) of CD4⁺ and CD8⁺ TCRs, respectively) showed concordant specificities between the α and the β chains (Supplementary Fig. 3e and Supplementary Table 6). Given that the α and the β chain sequences were obtained from independent TCR sequencing experiments for the same sample, the very high concordance between the α and β chains linked to a given pathogen-specific TCR, indicated high specificity of our donor-matched assignment approach.

In rare cases of conflicting assignments, which comprised only 68 CD4⁺ and 7 CD8⁺ lung TCR clonotypes (accounting for <1% of all the lung T cells predicted to be reactive to the common pathogens tested), priority was given to the specificity determined by the β chain, ensuring single, unambiguous specificity assignments. Notably, only TCR matches with shared V and J gene segments were considered. To determine the relative specificity when using a single β chain versus an α chain in resolving conflicting assignments, we used our in-house-generated large-scale resource of full-length paired α and β chain TCR sequences specific to ten common human pathogens as a positive control. We estimated the frequency with which single α or β chains from full-length aβ TCRs specific to a given pathogen were also observed in full-length TCRs annotated as specific to other pathogens (nonspecific). For each chain type, we quantified the conflict rate, that is, the proportion of cases in which the same chain appeared in TCRs annotated with different pathogen specificities. The frequency of β chain conflicts was significantly lower than that of α chain conflicts, when TCRs specific to each pathogen were separately tested against each other pathogen (P < 0.0001; Extended Data Fig. 5c). The overall conflict rate for all paired TCRs tested was lower for β chain compared to α chain conflicts (median of 10.3% versus ~17.5%). These results were consistent with those obtained using previously published TCR-based prediction approaches^61,62, which showed that the TCR β chain provided a stronger determinant of antigen specificity without diminishing the predictive contribution of the TCR α chain.

The donor-unmatched, reference-match-based (‘reference match’) approach was based on perfect matches and quasi-matches between the lung TCR clonotypes and TCR clonotypes present in the CD4⁺ and CD8⁺ TCR reference sets described in the Supplementary Methods. A perfect match was defined as an identical CDR3 amino acid sequence and V and J gene segment for either chain (α or β). A quasi-match was defined as an equally long CDR3 sequence containing exactly one amino acid mismatch, with identical V and J gene segments. Matches and quasi-matches were restricted by chain and T cell lineage, such that lung CD4⁺ T cells were compared only against the CD4⁺ TCR reference set and lung CD8⁺ T cells against the CD8⁺ TCR reference set. Quasi-matches were attempted only for TCR clonotypes that already had a perfect match for one TCR chain. This process yielded five types of match between lung TCR clonotypes and reference TCR clonotypes: (i) two-chain perfect match, (ii) β-only perfect match, (iii) α-only perfect match, (iv) perfect β match with α quasi-match and (v) perfect α match with β quasi-match (Extended Data Fig. 5d). Except for the α-only perfect matches, all types were considered when assigning pathogen specificity to lung TCR clonotypes. Notably, two-chain matches (types i, iv and v) could involve only one TCR clonotype (perfect match) or multiple TCR clonotypes in the reference set, and the latter could imply concordant (unambiguous) or conflicting (ambiguous) specificities. In cases of ambiguous assignment, priority was given to the specificity dictated by the β chain to ensure an unambiguous call.

The TCR unbiased clustering model (UCM) approach similarly relied on the TCR reference sets but leveraged another type of nonperfect yet biologically significant match determined by TCR UCMs. Let S_I represent the set of inquiry TCR clonotypes (from lung), S_R the set of reference TCR clonotypes, and M the set of m = 5 TCR UCMs that have been systematically compared (Supplementary Methods), with a column vector R representing their performance ranks. One round of clustering was performed for each TCR UCM in M, using the combined set $S_{\mathrm{I}}{\cup S}_{\mathrm{R}}$ as input. Of note, S_I and S_R could overlap, a feature exploited by the reference match approach described above. The standardized adjacency list was then constructed for each UCM as described in Supplementary Methods, allowing interactions between input TCR clonotypes to be readily retrieved. The goal was to assign unambiguous specificities to each TCR clonotype in S_I.

The proposed solution is as follows. From the UCM standardized outputs, define the following tensors in indicated order.

• A^X, an n × m integer matrix, where n is the number of TCR clonotypes in S_I and m is the number of TCR UCMs in M. Each entry ${\mathit{A}}_{i,j}^X$ denotes the number of interactions between inquiry TCR clonotype i and TCR clonotypes in S_R known to be specific to antigen X, as established by TCR UCM j. The columns of A^X follow the UCM ranking order in R.

• B^X, an n × m numeric matrix, where B^X = A^X × R, that is, ${\mathit{B}}_{i,j}^X$ denotes a weighted score indicating that inquiry TCR clonotype i is specific to antigen X, supported by UCM j.

• C^X, a column vector of size n, where ${\mathrm{C}}_i^X={\mathrm{sum}(\mathit{B}}_i^X)$, that is, the sum across UCMs of the weighted evidence that inquiry TCR clonotype i is specific to antigen X.

• D, an n × o integer matrix constructed by column-wise concatenation of o C^X vectors. Note that column order is arbitrary but must be tracked.

Finally, inquiry TCR clonotype i is predicted to be specific to the antigen(s) corresponding to column j if it meets the criteria in equations 1 and 2:

$$D_{i,j}=\text{max}\left(D_i\right)$$ 1

$$\frac{\text{max}\left(D_i\right)}{\text{max}\left(\text{SetDiff}\right(\text{max}\left(D_i\right),D_i\left)\right)}>t_r.$$ 2

In equation 1, the function max interrogates for the maximum value across columns of D_i. In equation 2, t_r is a tunable threshold. TCR clonotype i is left without a specificity prediction if the criterion of either equation is not met.

This process was run separately with the CD4⁺ and CD8⁺ TCR reference sets to predict specificities for lung CD4⁺ and CD8⁺ T cells, respectively. For stringency, t_r was set to 3.

Integrative strategy to assign pathogen specificities to T cells

This strategy incorporated individual results from the three approaches described above, with priority assigned in the following order: (1) reference match strategy, (2) donor-matched strategy and (3) UCM strategy. Numbers of TCR clonotypes with assignments by each individual approach within the integrative approach are provided in Extended Data Fig. 5f.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41590-026-02592-6.

Author contributions

V.F.-R., C.H.O. and P.V. conceived the work; V.F.-R. analyzed all data and developed the TCR-based analysis workflow for imputation of pathogen specificity under the supervision of B.P. and P.V.; A.E. performed tissue processing, cell sorting, ARTE assays, in vitro T cell proliferation assays and TCR validation experiments under the supervision of P.V.; I.M.-C. performed ARTE assays under the supervision of B.J.S. and P.V.; F.E.C.-C. and Z.R. assisted with data analysis under the supervision of V.F.-R. and P.V.; S.J.C., J.C., M.S., A.A. and E.W. provided clinical oversight and coordinated clinical assessments and patient recruitment, and S.J.C. and L.C. coordinated collection and processing of tissue and PBMC samples under the supervision of C.H.O.; M.M., and C.F.A. performed sequencing library preparation and next-generation sequencing under the supervision of P.V.; A.G. and A.S. provided peptide MPs and critical input for experimental design; E.P. and S.M. performed HLA typing; V.F.-R. wrote the first draft of the manuscript, and this was revised by P.V.

Peer review

Peer review information

Nature Immunology thanks the anonymous reviewers for their contribution to the peer review of this work. Primary Handling Editor: Nick Bernard, in collaboration with the Nature Immunology team. Peer reviewer reports are available.

Funding

P.V., A.S., B.P. and C.H.O. disclose support for this work from the NIH (grant number U19AI118626). P.V. discloses support for this work from the William K. Bowes Jr Foundation. C.H.O. discloses support for this work from the North West and South Central Research Delivery Networks, National Institute of Health Research, UK. The remaining authors declare no relevant funding. The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.

Data availability

Raw sequencing data and associated processed objects have been deposited in the Gene Expression Omnibus under accession code GSE305372. All other data generated or analyzed during this study are included in the published article and its accompanying figures, tables and supplementary materials. Source data are provided with this paper.

Code availability

Code used for data processing and visualization is available at the HIPC_Lung01 repository (https://github.com/vijaybioinfo/HIPC_Lung01). Code for antigen specificity assignment in single-cell datasets is available at the PatSpcImpByTCRs repository (https://github.com/VicenteFR/PatSpcImpByTCRs).

Competing interests

A.S. is a consultant for AstraZeneca, Darwin Health, Desna Therapeutics, Geovax, Link Campus University and RiverVest Venture Partners and Wugen, and he has equity interest in Sanofi. P.V. has previously consulted for Merck and Eli Lilly. The remaining authors declare no competing interests. L.J.I. has filed for patent protection for various aspects of T cell epitope and vaccine design work.

Extended data

is available for this paper at 10.1038/s41590-026-02592-6.

Supplementary information

The online version contains supplementary material available at 10.1038/s41590-026-02592-6.