Environmental cues regulate epigenetic reprogramming of airway-resident memory CD8⁺ T cells

Abstract

Tissue-resident memory T cells (T_RM cells) are critical for cellular immunity to respiratory pathogens and reside in both the airways and the interstitium. In the present study, we found that the airway environment drove transcriptional and epigenetic changes that specifically regulated the cytolytic functions of airway T_RM cells and promoted apoptosis due to amino acid starvation and activation of the integrated stress response. Comparison of airway T_RM cells and splenic effector-memory T cells transferred into the airways indicated that the environment was necessary to activate these pathways, but did not induce T_RM cell lineage reprogramming. Importantly, activation of the integrated stress response was reversed in airway T_RM cells placed in a nutrient-rich environment. Our data defined the genetic programs of distinct lung T_RM cell populations and show that local environmental cues altered airway T_RM cells to limit cytolytic function and promote cell death, which ultimately leads to fewer T_RM cells in the lung.

Main

T_RM cells act as sentinels of the immune system within peripheral tissues and are uniquely positioned to rapidly recognize and respond to invading pathogens^1,2,3. T_RM cells share many properties with effector-memory T cells (T_EM cells), including genetic architecture poised for cytokine production and robust cytolytic activity^2,4. Despite these functional similarities, T_RM cells are defined by a unique core transcriptional signature that supports long-term tissue residence through the regulation of cell-trafficking molecules and adaptations that enable survival in the tissue microenvironment^5,6. These adaptations allow for long-term protection by T_RM cells in sites such as the skin and gut, where they can provide almost sterilizing immunity when present in sufficient numbers^1,3. In contrast, the efficacy of cellular immunity against respiratory pathogens gradually wanes, and this decline is associated with the progressive loss of virus-specific T_RM cells in the lung^7,8. However, the mechanisms driving the transient nature of lung T_RM cells are not well defined.

After clearance of a primary influenza infection, virus-specific memory CD8⁺ T cells are localized to secondary lymphoid organs and peripheral tissues, primarily the lung interstitium and the lung airways^9,10,11. Evidence from animal models and humans indicates that memory CD8⁺ T cells confer protective immunity to respiratory viruses by substantially decreasing viral loads, limiting immunopathology and lowering disease burden^7,8,12,13,14. The lung CD8⁺ T_RM cell pool comprises two distinct populations: airway T_RM (A-T_RM) cells and interstitial T_RM (I-T_RM) cells, with unique functional properties. A-T_RM cells are poorly cytolytic compared with I-T_RM cells, yet are sufficient to protect against influenza challenge through the rapid production of antiviral cytokines¹⁵. In addition, the number of A-T_RM cells correlates with the efficacy of cellular immune protection in the lung⁸. These findings have raised questions about the requirements for differentiation and maintenance of these distinct populations of lung T_RM cells, and about how environmental niches shape the function and lifespan of these cells, but the molecular underpinnings of differences between A-T_RM and I-T_RM cells have not been explored to date.

Lung T_RM cells are gradually lost under steady-state conditions¹⁶. However, T_RM cells in the lung and other barrier tissues have a consistent transcriptional profile, raising questions as to why the lifespan of T_RM cells would vary between tissues^6,17,18,19. One potential explanation for these conflicting findings is that lung T_RM cells have often been investigated as a single population, without separation into A-T_RM and I-T_RM cell subsets. Given the functional differences between these subsets and the distinct environments where they reside, a detailed comparison of A-T_RM and I-T_RM cells could be informative about the mechanisms that control their biology and regulate the decline of lung T_RM cells.

In the present study, we examined the decline of cellular immunity to the influenza virus over time, with a focus on comparing flu-specific lung T_RM cells in the airways and interstitium. We observed that A-T_RM and I-T_RM subsets were gradually lost due to apoptosis in the tissue, and transcriptome and chromatin accessibility analysis (ATAC) revealed an enrichment of genes in A-T_RM cells associated with the integrated stress response (ISR), notably the amino acid starvation pathway. These stress-related programs were due to the airway environment, whereas core T_RM signature genes were regulated during the initial differentiation of A-T_RM cells after infection. Overall, these findings provide new insight into the role of environmental cues in controlling the differential functions and lifespan of A-T_RM and I-T_RM cells, and identify pathways that may be manipulated to improve the longevity of cellular immunity against respiratory pathogens.

Results

A-T_RM cells rapidly decline after influenza infection

To confirm that cellular immune protection against heterologous influenza challenge is rapidly lost^7,8, C57Bl/6J wild-type mice infected with influenza A/HKx31 (x31, H3N2) were challenged with influenza A/PR8 (PR8, H1N1) at 1, 3 or 8 months after infection with x31. The PR8-challenged mice showed progressively more weight loss (Fig. 1a) and decreased survival (Fig. 1b) as time from the initial infection with x31 increased. To determine the kinetics of the loss of flu-specific A-T_RM and I-T_RM cells in the lung, we analyzed the number of influenza nucleoprotein (FluNP)⁺ and acid polymerase (FluPA)⁺ CD8⁺ T cells on days 8, 10, 14, 21, 35, 60, 90 and 180 post-infection in the lung airways (bronchoalveolar lavage (BAL)), lung interstitium and spleen using intravital labeling, which allows for the identification of extravascular T cells in the tissue by gating on IV⁻CD8⁺ T cells (negative for staining with the intravital antibody). Tissue-resident, FluNP⁺, memory CD8⁺ T cells were present in both the lung airways and the interstitium 35 d post-x31 infection (Fig. 1c). Regardless of specificity, the number of CD8⁺IV⁻ A-T_RM cells and CD8⁺IV⁻ I-T_RM cells gradually declined from day 35 to day 180 post-infection, whereas the number of splenic effector, memory CD8⁺CD62L⁻ T cells (S-T_EM) remained mostly unchanged (Fig. 1d,e). We observed a similar trend after infection with Sendai virus (see Supplementary Fig. 1), indicating that the effect was not specific to influenza virus. A-T_RM cells were lost much more rapidly than I-T_RM cells, declining almost 100-fold by 6 months post-infection compared with 40-fold for I-T_RM cells (Fig. 1f,g). These data indicated that A-T_RM and I-T_RM cells were lost over 6 months post-infection, with a more accelerated decline of the A-T_RM cells.

Fig. 1: Decline of A-T_RM and I-T_RM cells over time.

a, Morbidity curve examining decline of protection in x31-immune mice after heterologous PR8 challenge (n = 20 per group, combined from two experiments). b, Mortality curve examining decline of protection from heterologous challenge over time after a ×10 LD₅₀ PR8 challenge (n = 20 per group, combined from two experiments). c, Intravital (IV) staining showing FluNP⁺CD8⁺ T cells from BAL, lung interstitium and spleen (n = 5, representing three experiments). d,e, The number of FluNP⁺ (d) and FluPA⁺ (e) CD8⁺ S-T_EM, A-T_RM or I-T_RM cells in wild-type mice at various times post-x31 infection (n = 15, combined from three experiments). Data are represented as mean ± s.e.m. f,g, Fold decrease in FluNP⁺CD8⁺ T cells (f) and FluPA⁺CD8⁺ T cells (g) from wild-type mice at various times post-x31 infection compared with the number of FluNP⁺CD8⁺ T cells or FluPA⁺CD8⁺ T cells at day 35 post-infection (n = 15, combined from three experiments). Data are represented as the mean ± s.e.m.

A-T_RM and I-T_RM cells do not recirculate and undergo apoptosis

To investigate the mechanisms that contributed to the gradual decline of lung T_RM cells, we first examined whether their loss was limited to a subset of the T_RM population based on expression of the residency markers CD69 and CD103. The number of CD69⁻CD103⁻, CD69⁺CD103⁻ and CD69⁺CD103⁺ FluNP⁺ I-T_RM cells all declined over 6 months post-influenza infection (Fig. 2a,b). To test whether the gradual loss of circulating CD62L⁻ T_EM cells contributed to the loss of lung T_RM cells, we measured the number of S-T_EM cells after x31 influenza infection. There was a modest decrease in the number of FluNP⁺ S-T_EM cells between 1 and 6 months post-infection in wild-type mice (see Supplementary Fig. 2), similar to previous reports²⁰. CD69⁺CD103⁻ and CD69⁺CD103⁺ FluNP⁺ I-T_RM cells declined notably faster than FluNP⁺ S-T_EM cells between 1 and 6 months post-infection (see Supplementary Fig. 2c). In contrast, CD69⁻CD103⁻CD8⁺ FluNP⁺IV⁻ T cells in the lung, which were probably transiting T_EM cells, declined at the same rate as S-T_EM cells (see Supplementary Fig. 2c), indicating that the decline of lung T_RM cells did not coincide with the gradual loss of S-T_EM cells.

Fig. 2: In situ apoptosis drives lung T_RM cell decline.

a, Expression of CD69 and CD103 on FluNP⁺ I-T_RM cells from wild-type mice on days 35, 60, 90 or 180 post-x31 infection. b, Number and frequency of CD69⁻CD103⁻, CD69⁺CD103⁻ and CD69⁺CD103⁺ FluNP⁺ I-T_RM cell subsets from mice infected as in a (n = 15, combined from three experiments). Data are represented as the mean ± s.e.m. c, Expression of CD69 and CD103 on CD45.2⁺ or CD45.1⁺ FluNP⁺IV⁻CD8⁺ lung T_RM cells from one parabiont pair 49 d post-x31 infection and 21 d post-parabiosis surgery. d, Frequency of CD45.2⁺ or CD45.1⁺ FluNP⁺CD8⁺ T cells in the BAL, lung interstitium, lung vasculature and spleen of parabiosis partners as in c (n = 6 parabiont pairs, representing two experiments). Data are represented as the mean ± s.e.m. The significance was determined using the paired Student’s t-test. e, Frequency of CD45.2⁺ or CD45.1⁺ FluNP⁺ I-T_RM cells from parabiosis partners as in c, based on CD69 and CD103 expression (n = 6 parabiont pairs, representing two experiments). Data are represented as the mean ± s.e.m. The significance was determined using the paired Student’s t-test. f, Expression of annexin V and CD69 on FluNP⁺ A-T_RM, I-T_RM or S-T_EM cells from wild-type mice 35 d post-infection with x31 (n = 14, combined from three experiments). g, Frequency of AnnexinV⁺FluNP⁺ A-T_RM, I-T_RM or S-T_EM cells from wild-type mice 35 d post-infection with x31 (n = 14, combined from three experiments). Data are represented as the mean ± s.e.m. The significance was determined using the unpaired Student’s t-test. h, Frequency of AnnexinV⁺ cells among FluNP⁺CD69⁺ A-T_RM or I-T_RM cells as in g (n = 10, combined from two experiments). Data are represented as the mean ± s.e.m. The significance was determined using the unpaired Student’s t-test. P values are as follows: *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

To address whether A-T_RM and I-T_RM cells were truly resident populations, or whether they were maintained by replenishment from the circulating memory T cell pool, we performed parabiosis in congenic mice, 4 weeks post-infection with x31. Intravascular labeling after 3 weeks of parabiosis indicated that 90% of FluNP⁺ I-T_RM cells were host derived and expressed the tissue residency markers CD69 and CD103 (Fig. 2c), whereas the few partner-derived FluNP⁺IV⁻CD8⁺ T cells were CD69⁻CD103⁻ (Fig. 2c), suggesting that they were transiting T_EM cells. Comparison across tissues indicated that FluNP⁺CD8⁺ T cells from the host and partner were in equilibrium in the spleen and the lung vasculature, but were notably enriched for host-derived cells among FluNP⁺ A-T_RM and I-T_RM cells (Fig. 2d). Analysis of CD69 and CD103 expression on FluNP⁺ I-T_RM cells found that >95% of CD69⁺CD103⁻ and CD69⁺CD103⁺ cells were host derived (Fig. 2e). These observations indicated that lung T_RM cells did not exit the tissue and were not replenished by circulating S-T_EM cells.

To investigate the possibility that lung T_RM cells were lost due to death in the tissue, we examined expression of the proapoptotic marker annexin V on FluNP⁺ A-T_RM, I-T_RM and S-T_EM cells on day 35 post-infection with x31. A-T_RM and I-T_RM cells had increased annexin V staining compared with S-T_EM cells (Fig. 2f,g), indicating increased apoptosis. We also observed an important increase in AnnexinV⁺ cells among CD69⁺ A-T_RM cells compared with CD69⁺ I-T_RM cells (Fig. 2h). Thus, the gradual loss of lung T_RM cells was due to increased apoptosis within the tissue, and A-T_RM cells had a higher rate of cell death compared with I-T_RM cells, possibly due to unique microenvironmental effects within the lung on the survival of T_RM cells.

A-T_RM cells are transcriptionally distinct from I-T_RM cells

Next we used RNA sequencing (RNA-seq) to compare the transcriptional profiles of FluNP⁺IV⁻CD8⁺ A-T_RM cells, FluNP⁺IV⁻CD8⁺ I-T_RM cells and FluNP⁺CD8⁺ S-T_EM cells isolated by cell sorting (Fig. 3a) from wild-type mice 35 d post-infection with x31. Principal component analysis (PCA) of the 9,362 genes detected indicated that I-T_RM cells more closely aligned with S-T_EM cells than with A-T_RM cells (Fig. 3b). Differentially expressed genes (DEGs) (false discovery rate (FDR) < 0.05, absolute log₂(fold change) (log₂(FC)) > 1) indicated that each population had a distinct expression profile, with A-T_RM cells being the most transcriptionally distinct (Fig. 3c). We correlated the change in expression for genes differentially expressed between A-T_RM and S-T_EM cells, or between A-T_RM and I-T_RM cells, and identified gene expression changes unique to each comparison (for example, Itgae, Ctla4), as well as shared differences (for example, Klrg1, Asns) (Fig. 3d). DEGs upregulated in airway T_RM cells included genes involved in amino acid transport (Slc1a4, Slc7a5) and amino acid synthesis (Asns, Lars), whereas DEGs shared by A-T_RM and I-T_RM cells included genes associated with programming of T_RM cells (Itga1, Itgae, Ahr) (Fig. 3d). Gene set enrichment analysis (GSEA)²¹ identified a notable positive enrichment in genes associated with the unfolded protein response in A-T_RM cells compared with both I-T_RM and S-T_EM cells (Fig. 3e), suggesting stresses unique to the airway and not present in the lung interstitium or spleen. A-T_RM cells were also negatively enriched for genes involved in the cytotoxic T lymphocyte (CTL) pathway, including Prf1, Gzma, Gzmb and Gzmk (Fig. 3e), which supported previous reports that A-T_RM cells are poorly cytolytic¹⁵. In addition, A-T_RM cells showed altered expression of DEGs related to intrinsic cell death, maintenance of cell survival under cell stress and activation of the ISR, including Dusp1, Bax, Bcl2, Bbc3, Pim2 and the proapoptotic transcription factor Ddit3 (Fig. 3f). However, A-T_RM and I-T_RM cells shared the expression of a core set of known T_RM genes, including Itgae, Cdh1, Ahr, Cxcr6, Klf2 and S1p1r^6,18 (Fig. 3f). To confirm the RNA-seq findings, we performed a second analysis using FluNP⁺CD8⁺ T cells sorted from three independent cohorts of wild-type mice 35 d after x31 infection, which included FluNP⁺IV⁺CD8⁺ T_EM cells from the lung vasculature, in addition to FluNP⁺ A-T_RM, I-T_RM and S-T_EM cells, to address whether processing of the lung tissue impacted the genetic signature of I-T_RM cells. Data from this additional analysis showed a similar pattern of gene expression to the original analysis, and indicated that FluNP⁺ I-T_RM cells and lung vascular T_EM cells had unique transcriptional signatures (see Supplementary Fig. 3). In addition, staining for BCL2 protein indicated that BCL2 was notably upregulated in A-T_RM cells compared with I-T_RM cells (see Supplementary Fig. 4). These data indicated that A-T_RM cells had a distinct transcriptional profile, characterized by cellular stress and balancing of pro- and antiapoptotic signals compared with I-T_RM cells,_, and suggested that the local microenvironment had a critical role in regulating T_RM cell biology.

Fig. 3: Influenza-specific lung A-T_RM and I-T_RM cells have distinct transcriptional profiles.

a, Sorting strategy of FluNP⁺ A-T_RM, I-T_RM or S-T_EM cells from wild-type mice 35 d post-x31 infection for RNA-seq. b, PCA of 9,362 detected genes from cells described in a. The circles represent 99% confidence intervals. BAL (A-T_RM, n = 3), spleen (S-T_EM, n = 3) and lung interstitium (I-T_RM, n = 2). c, Heatmap of 1,622 DEGs from A-T_RM, I-T_RM or S-T_EM cells. d, Scatterplot of correlation of DEGs between A-T_RM and S-T_EM cells and between A-T_RM and I-T_RM cells isolated as in a. Black, DEGs in A-T_RM cells compared with S-T_EM and I-T_RM cells; red, DEGs between only A-T_RM and S-T_EM cells; blue, DEGs between I-T_RM and S-T_EM cells. e, GSEA of the unfolded protein response and CTL pathway gene sets comparing A-T_RM cells with I-T_RM and S-T_EM cells as in a. f, Heatmap of selected genes derived from c, related to a core T_RM cell signature, cytotoxic function and cell survival, as in a.

A-T_RM cell survival and homeostasis are epigenetically regulated

The immediate microenvironment can impact the epigenetic programming and function of immune cells²², including those in the lung²³. To determine the effect of T_RM cell location on epigenetic programming, we used ATAC-sequencing (ATAC-seq)^24,25 to assess the chromatin accessibility landscape of FluNP⁺CD8⁺ A-T_RM, I-T_RM and S-T_EM cells 35 d post-influenza infection. The PCA of 47,683 accessible loci indicated that A-T_RM cells clustered separately from I-T_RM and S-T_EM cells (Fig. 4a). These differences were not due to tissue processing and were consistent in three independent cohorts of wild-type mice that had been infected with x31 35 d earlier (see Supplementary Fig. 3). We identified differentially accessible regions (DARs) (FDR < 0.05, absolute log₂(FC) > 1) between each T_RM cell subset (Fig. 4b), and integrated them with the RNA-seq data to determine coordinated changes in the epigenome and transcriptome for each T_RM cell subset²⁶. Using k-means clustering, we observed three distinct patterns (k1–k3) in the data that could be mapped to each of the cell types: k1 in S-T_EM cells, k2 in I-T_RM cells and k3 in A-T_RM cells (Fig. 4c). This analysis indicated that A-T_RM cells had a unique chromatin accessibility landscape compared with I-T_RM and S-T_EM cells, which resulted in coordinated changes in gene expression and suggested that the epigenetic architecture of A-T_RM cells may be impacted by their environment.

Fig. 4: Chromatin accessibility reveals a distinct epigenetic programming of lung A-T_RM and I-T_RM populations.

a, PCA of 47,683 accessible loci in A-T_RM, I-T_RM or S-T_EM cells from wild-type mice 35 d post-x31 infection. Circles represent 99% confidence intervals. BAL (A-T_RM, n = 4), spleen (S-T_EM, n = 6) and lung interstitium (I-T_RM, n = 3). b, Heatmap of 8,772 DARs across A-T_RM, I-T_RM or S-T_EM cells as in a. c, Integrative analysis of DEGs from Fig. 3c and DARs from b using k-means clustering. Bar plots showing the expression (top) and accessibility (bottom) for A-T_RM, I-T_RM or S-T_EM cells in each of the k patterns. The data represent the mean ± s.d. d, Heatmap displaying the enrichment of transcription factor motifs for DARs within each of the k patterns from c. The P value was determined by HOMER using a binomial distribution. e, Histogram (left) and boxplot (right) of accessibility surrounding the indicated motif for A-T_RM, I-T_RM or S-T_EM cells. The boxplot center line indicates data median, lower and upper boundaries of boxes the 1st and 3rd quartile ranges, and whiskers the upper and lower ranges of the data. The data represent the mean accessibility for A-T_RM, I-T_RM or S-T_EM cells. The significance was determined using the two-tailed Student’s t-test (P value: ****P < 0.0001; NS, not significant). f, Radar plot of GO pathway enrichment for each of the k patterns from c. The significance was determined using Fisher’s exact test. g, Genome plot (left) showing the accessibility pattern at the indicated loci for A-T_RM, I-T_RM or S-T_EM cells as in a. For each gene, the direction of transcription and the location of DARs are annotated. A barplot showing the expression of each gene is also depicted (right). The data represent the mean ± s.d. Chr, chromosome.

Next we examined the enrichment of transcription factor DNA-binding motifs within the DARs in patterns k1, k2 and k3. Chromatin accessibility patterns for A-T_RM, I-T_RM and S-T_EM cells were all enriched for motifs for ETS and RUNX (Fig. 4d), which are common to memory CD8⁺ T cells²⁷. The pattern k3, which was enriched in A-T_RM cells, was enriched in binding motifs for STAT5 and DDIT3 (Fig. 4d). A-T_RM and I-T_RM cells shared enrichment for AP-1-binding motifs, whereas I-T_RM cells were uniquely enriched for FOS and CREM motifs (Fig. 4d). Analysis of the unique accessibility footprint surrounding CREM, STAT5 and DDIT3 motifs indicated that CREM motifs were highly accessible in I-T_RM cells compared with S-T_EM and A-T_RM cells (Fig. 4e). Similarly, STAT5 and DDIT3 were notably more accessible in A-T_RM cells compared with S-T_EM and I-T_RM cells (Fig. 4e).

To complement the motif analysis, we performed gene ontology (GO) analysis to determine the functional enrichment of the genes in each pattern. The S-T_EM pattern (k1) was enriched for genes associated with T cell differentiation and activation (Fig. 4f). The A-T_RM pattern (k3) was enriched for genes implicated in cellular homeostasis, endoplasmic reticulum (ER) stress, hypoxia and glucose starvation (Fig. 4f). The extrinsic apoptosis pathway was uniquely enriched in I-T_RM cells (k2, Fig. 4f), which might explain the increased apoptosis in these cells compared with S-T_EM cells. Slc7a5 and Asns, which encode an amino acid transporter and asparagine synthetase, respectively, contained DARs with increased accessibility and were more highly expressed in A-T_RM cells compared with I-T_RM and S-T_EM cells (Fig. 4g), suggesting differential epigenetic and transcriptional programming of pathways related to cell stress and apoptosis in A-T_RM cells. These data suggest that the local environments of the airways and interstitium contribute to differential epigenetic programming in lung T_RM cell subsets.

Amino acid starvation controls the lifespan of A-T_RM cells

The ISR is activated by several triggers of cellular stress, including viral infection, ER stress and amino acid starvation, with the goal of inhibiting protein translation, and can induce apoptosis if the stress is not resolved²⁸. Different stressors activate unique pathways to restore cellular homeostasis²⁸. To determine the pathway responsible for the gradual loss of A-T_RM cells, we analyzed genes known to be involved in responses to cellular stress that were differentially expressed in A-T_RM cells. Most transcripts upregulated in A-T_RM cells were involved in amino acid transport (Slc25a22, Slc1a4, Slc7a5), amino acid synthesis (Mars, Lars, Sars, Aars, Asns), recognition of uncharged transfer RNAs (Eif2ak4), cell cycle arrest (Cdkn1a) and proapoptotic transcriptional regulation in response to stress (Ddit3) (Fig. 5a), suggesting that amino acid starvation could be the primary trigger for activation of the ISR in A-T_RM cells²⁸. GSEA indicated that gene sets involved in intrinsic apoptotic signaling, in response to ER stress and amino acid transport, were highly enriched in A-T_RM cells compared with I-T_RM and S-T_EM cells (Fig. 5b), suggesting ISR activation and amino acid starvation in these cells. To further investigate the impact of the airway environment in the amino acid stress response, we assessed the expression of the neutral amino acid transporter CD98, a heterodimer of Slc7a5 and Slc3a2 (ref. ²⁹), in combination with CD11a, which is highly expressed on memory CD8⁺ T cells, but is downregulated within 36 h of entry into the airways³⁰, on FluNP⁺ A-T_RM, I-T_RM and S-T_EM cells 35 d post-influenza infection. A-T_RM cells expressed more CD98 compared with I-T_RM and S-T_EM cells (Fig. 5c,d). Furthermore, increased expression of CD98 on A-T_RM cells was limited to CD11a^lo cells (Fig. 5d), indicating that upregulation of CD98 occurred after entry into the airway environment. To address whether the airway environment was sufficient to drive upregulation of CD98 on memory CD8⁺ T cells, we transferred, intratracheally, S-T_EM cells, from wild-type mice infected with x31 30 d earlier, into naive, congenic, wild-type recipient mice. Expression of CD98 on transferred S-T_EM cells was notably increased 4 d post-intratracheal (i.t.) transfer compared with S-T_EM cells pre-transfer (Fig. 5e). In addition, expression of CD98 on S-T_EM cells transferred into the airways steadily increased between day 1 and day 8 post-i.t. transfer (Fig. 5f), suggesting that CD98 expression correlated with the length of time that memory CD8⁺ T cells were in the airway environment.

Fig. 5: Exposure to the airway environment drives activation of the ISR.

a, Heatmap for selected ISR genes associated with the amino acid starvation response in A-T_RM, I-T_RM or S-T_EM cells from wild-type mice 35 d post-x31 infection. b, GSEA of intrinsic apoptotic signaling in response to ER stress and amino acid transmembrane transport gene sets, comparing A-T_RM cells with I-T_RM cells and A-T_RM cells with S-T_EM cells as in a. c, CD11a and CD98 staining on FluNP⁺ A-T_RM, I-T_RM and S-T_EM cells from wild-type mice 35 d post-x31 infection. d, Geometric mean fluorescence intensity (gMFI) of CD98 measured on FluNP⁺CD69⁺ A-T_RM, I-T_RM, and S-T_EM cells, and on FluNP⁺CD11a^hi or FluNP⁺CD11a^lo A-T_RM cells, from wild-type mice 35 d after x31 infection (n = 7, representing two experiments). The significance was determined using the two-tailed Student’s t-test. e, The gMFI of CD98 on S-T_EM cells isolated from wild-type mice 35 d post-x31 infection, either pre-transfer or 4 d post-i.t. transfer (n = 8, combined from two experiments). The significance was determined using the Mann–Whitney U-test. f, The gMFI of CD98 on S-T_EM cells as in e from days 1–8 post-i.t. transfer (n = 8–12, combined from two experiments). g, Expression of Ddit3 by RNA-seq in A-T_RM, I-T_RM and S-T_EM cells as in a, shown as FPKM. The data represent the mean ± s.d. h, The ratio of wild-type (WT) to Ddit3^−/− FluNP⁺ A-T_RM and I-T_RM cells normalized to the ratio of FluNP⁺ S-T_EM cells from mixed bone marrow chimeras infected with x31 on day 10 or day 60 post-infection (n = 9 mice at day 10, 10 mice at day 60, representing two experiments). i, Frequency of CD45.1⁺ wild-type and Ddit3^−/− S-T_EM cells sorted from the spleen on day 35 post-x31 infection before (day 0) or 3 d after i.t. transfer into CD45.1⁺CD45.2⁺ recipient mice. The data are represented as mean ± s.d. The significance was determined using the paired Student’s t-test. P values are as follows: *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

To test whether amino acid starvation mediated the loss of A-T_RM cells under homeostatic conditions, we assessed whether deletion of Ddit3, a downstream regulator of apoptosis via amino acid starvation^31,32, which was upregulated in A-T_RM cells compared with I-T_RM and S-T_EM cells (Fig. 5g), increased the survival of A-T_RM cells. Chimeric mice generated by co-injecting wild-type and Ddit3^−/− bone marrow cells into congenic mice were infected with influenza x31 8 weeks post-transfer. The number of FluNP⁺ A-T_RM, I-T_RM and S-T_EM cells was determined at the peak of acute infection (day 10) and after establishment of T cell memory (day 60). The ratio of wild-type to Ddit3^−/− FluNP⁺ A-T_RM and I-T_RM cells (normalized to the ratio of wild-type to Ddit3^−/− FluNP⁺ S-T_EM cells in each mouse) showed no differences (Fig. 5h), indicating that Ddit3 deficiency did not rescue the accumulation of FluNP⁺ A-T_RM cells. However, the dynamic balance between cell death and cell recruitment that regulates the number of A-T_RM cells made it difficult to accurately define whether expression of Ddit3 impacted the survival of memory CD8⁺ T cells in the airway. Thus, to assess cell survival in the absence of cell recruitment to the airways, we sorted S-T_EM cells from the spleens of congenic wild-type CD45.1⁺ and Ddit3^−/− mice infected with x31 35 d earlier, and transferred them intratracheally in equal mixes into wild-type CD45.1⁺CD45.2⁺ recipients infected with x31 35 d earlier; 3 d post-transfer, the ratio of wild-type S-T_EM cells to Ddit3^−/− S-T_EM cells in the airways was notably biased toward Ddit3^−/− S-T_EM cells (Fig. 5i), indicating increased survival of Ddit3^−/− S-T_EM cells in the airways compared with wild-type S-T_EM cells. To address whether deletion of Ddit3 impacted cellular immunity, A-T_RM cells from wild-type or Ddit3^−/− mice, infected with x31 35 d earlier, were transferred intratracheally into the airways of naive hosts before challenge with PR8 influenza 1 d post-transfer. Viral titers were similar between mice that received wild-type or Ddit3^−/− A-T_RM cells on day 4 post-challenge (see Supplementary Fig. 5)_, indicating that expression of Ddit3 did not alter the protective functions of A-T_RM cells. Thus, the airway environment affected T_RM cell biology and suggested that amino acid starvation contributed to the gradual loss of the A-T_RM cells.

The airway environment drives activation of the ISR

To provide a causative link between the microenvironment and the transcriptional signature of A-T_RM cells, we sorted FluNP⁺ S-T_EM cells from wild-type (CD45.2) mice 35 d post-x31 infection, and transferred them intratracheally or intraperitoneally into wild-type CD45.1 mice that had been infected with x31 35 d earlier. RNA-seq on CD45.2⁺ S-T_EM cells isolated from the airways or peritoneum on day 2 post-transfer found 375 transcripts altered in S-T_EM cells post-i.t. transfer compared with S-T_EM cells pre-transfer, whereas only 54 transcripts were differentially expressed in S-T_EM cells post-intraperitoneal (i.p.) transfer compared with S-T_EM cells pre-transfer (Fig. 6a). GSEA comparing S-T_EM cells post-i.t. or post-i.p. transfer with S-T_EM cells pre-transfer indicated an important enrichment for pathways mediating the amino acid starvation response and ER stress in post-i.t. transfer S-T_EM cells, but not post-i.p. transfer S-T_EM cells (Fig. 6b)³³. Published datasets from alveolar and lung interstitial macrophages showed no enrichment of genes involved in amino acid starvation or ER stress (see Supplementary Fig. 6), indicating that these pathways were enriched in S-T_EM cells but not in macrophages exposed to the airway environment.

Fig. 6: The airway environment is sufficient to alter the transcriptional program of S-T_EM cells but does not induce core T_RM programming.

a, Volcano plots of DEGs between pre-transfer S-T_EM (n = 4) and post-i.t. transfer S-T_EM (n = 4) or post-i.p. transfer S-T_EM (n = 4) cells, sorted from the spleen of CD45.2⁺ wild-type mice on day 35 post-x31 infection and transferred into congenic CD45.1⁺ mice that had been infected with x31 35 d earlier. Transferred CD45.2⁺ cells were isolated 2 d post-transfer for RNA-seq analysis. Example genes are highlighted and the number of DEGs are indicated. b, GSEA on post-i.t. transfer S-T_EM and post-i.p. transfer S-T_EM cells as in a using the indicated gene sets. c, Volcano plot of DEGs comparing post-i.t. transfer S-T_EM cells sorted from the spleen of CD45.2⁺ wild-type mice on day 35 post-x31 infection, and transferred into the airways of congenic CD45.1⁺ mice that were infected with Sendai virus 8 d (post-i.t. acute) or 35 d (post-i.t. memory) earlier. Transferred CD45.2⁺ cells were isolated 2 d post-transfer for RNA-seq analysis. Example genes are highlighted and the number of DEGs are indicated (n = 5 for acute transfer samples, n = 4 for memory transfer samples). d, Barplot showing the expression of cell stress genes (Sars, Slc7a5) in pre-transfer S-T_EM, post-i.t. memory S-T_EM, post-i.t. acute S-T_EM and post-i.p. memory S-T_EM cells, as in a and c. The data represent the mean ± s.d. e, CD98 staining on post-i.t. memory S-T_EM, post-i.t. acute S-T_EM and post-i.p. memory S-T_EM cells as in a and c.

Cellular stress due to nutrient-poor conditions in the airway may limit the lifespan and effector functions of airway T_RM cells during homeostasis to prevent unnecessary immunopathology. To assess whether an inflamed airway environment had a similar impact on the programming of T_EM cells, FluNP⁺ S-T_EM cells isolated from CD45.2 wild-type mice 35 d post-x31 infection were transferred intratracheally into the airways of CD45.1 wild-type mice 8 d post-infection with Sendai virus (post-i.t. acute) or 35 d post-infection with Sendai virus (post-i.t. memory), based on the fact that influenza and Sendai viruses are antigenically distinct and share no cross-reactive T cell epitopes⁴. Comparison of RNA-seq on CD45.2⁺ S-T_EM cells, isolated on day 2 post-i.t. transfer from post-i.t. acute mice or post-i.t. memory mice, found over 400 DEGs (Fig. 6c). Genes related to amino acid starvation (Sars, Slc7a5) and those related to cell survival (Myc, Bbc3) were notably increased in post-i.t. memory S-T_EM cells compared with post-i.t. acute S-T_EM cells (Fig. 6c). Analysis of gene expression from pre-transfer, post-i.t. memory, post-i.t. acute and post-i.p. S-T_EM cells showed similar expression of Sars and Slc7a5 in pre-transfer, post-i.p. and post-i.t. acute S-T_EM cells (Fig. 6d), indicating that the amino acid starvation response was not induced in S-T_EM cells transferred into acutely infected airways. Expression of the amino acid transporter CD98 was higher on post-i.t. memory S-T_EM cells compared with post-i.p. and post-i.t. acute S-T_EM cells (Fig. 6e), confirming the RNA-seq results. Thus, the airway microenvironment could drive activation of the amino acid stress response pathway in S-T_EM cells at steady state, but did not do so during infection.

The airway environment regulates the A-T_RM cell genetic program

To determine the gene signatures induced by the airway environment alone, compared with those induced during the differentiation of A-T_RM cells after viral clearance, we compared the genes differentially expressed in A-T_RM cells versus S-T_EM cells, isolated from wild-type mice 35 d post-x31 infection, with genes differentially expressed in S-T_EM cells pre-transfer and post-i.t. transfer. A number of genes associated with amino acid starvation, such as Slc7a5, Asns and Myc, were upregulated in both A-T_RM cells and post-i.t. transfer S-T_EM cells, whereas known T_RM genes such as Itgae and Itga1 were upregulated only in A-T_RM cells, but not post-i.t. transfer S-T_EM cells, and independent of the tissue environment (Fig. 7a). Comparisons of select DEGs across A-T_RM, I-T_RM, S-T_EM, and pre- and post-i.t. transfer S-T_EM cells indicated that, although DEGs associated with cell stress (Aars and Slc7a5) or lack of CTL activity (Gzmb) were observed only in A-T_RM cells and post-i.t. transfer S-T_EM cells exposed to the airway environment, known T_RM genes (Itgae, Itga1 and Cxcr6) were similarly differentially expressed in A-T_RM cells and I-T_RM cells compared with S-T_EM cells and pre- and post-i.t. transfer S-T_EM cells (Fig. 7b). These data indicated that the program linked to lung T_RM cell differentiation was shared between A-T_RM and I-T_RM cells, with subsequent differences in gene expression being driven by adaptations to the local environment.

Fig. 7: Restoration of a nutrient-rich environment resolves environmentally driven cellular stress in A-T_RM cells.

a, Scatterplot correlating the DEGs between pre-transfer S-T_EM cells and post-i.t. memory S-T_EM cells as in Fig. 6a versus DEGs between A-T_RM and S-T_EM cells as in Fig. 3a. The location of select genes is indicated. X indicates DEGs in both comparisons. b, Barplot showing the expression of cell stress (Aars, Slc7a5) and T_RM (Itgae, Itga1, Gzmb and Cxcr6) signature genes from A-T_RM, I-T_RM, S-T_EM, pre-transfer S-T_EM and post-i.t. transfer S-T_EM cells. The data represent the mean ± s.d. c, Heatmap of 5,715 DEGs comparing S-T_EM and A-T_RM cells sorted from wild-type mice 35 d post-x31 infection before (pre-in vitro) and 2 d after (post-in vitro) culture (n = 5 for A-T_RM cells, n = 3 for S-T_EM cells). d, GO pathway analysis for three distinct regions of gene expression from A-T_RM pre-in vitro, A-T_RM post-in vitro, S-T_EM pre-in vitro and S-T_EM in vitro cells as in c. The significance was determined using Fisher’s exact test. e, Heatmap for selected ISR genes associated with the amino acid starvation response from A-T_RM pre-in vitro, A-T_RM post-in vitro, S-T_EM pre-in vitro and S-T_EM in vitro cells as in c. f, Barplot showing the expression of amino acid stress (Slc7a5, Slc3a2, Aars), proapoptotic (Ddit3, Bbc3), cell stress (Myc, Pim2) and cell adhesion (Itgae) genes in A-T_RM pre-in vitro, A-T_RM post-in vitro, S-T_EM pre-in vitro and S-T_EM in vitro cells as in c. The data represent the mean ± s.d.

As activation of the ISR was a key adaptation of A-T_RM cells to their environment, we examined the impact of nutrient restoration on A-T_RM cells. After 2 d of in vitro culture without any stimulation, A-T_RM cells from wild-type mice infected with x31 35 d earlier showed 1,698 DEGs (FDR < 0.05, absolute log₂(FC) > 1) compared with A-T_RM cells before in vitro culture, whereas S-T_EM cells cultured in vitro showed only 9 DEGs compared with S-T_EM cells before in vitro culture (Fig. 7c). DEGs from in vitro A-T_RM cells were grouped in three clusters, and GO analyses indicated processes involved in apoptosis, tRNA charging and responses to ER stress being notably altered in A-T_RM cells after in vitro culture (Fig. 7d). Genes related to the ISR (Slc7a5, Slc3a2, Ddit3, Sars, Aars and Bbc3)³⁴ were downregulated in A-T_RM cells cultured in a nutrient-rich environment compared with A-T_RM cells isolated directly from the airways (Fig. 7f,g), indicating that this pathway was reversible and driven by the local microenvironment. These observations defined the key transcriptional programs induced by the differentiation of T_RM cells versus the airway environment, and showed that the ISR driven by amino acid starvation could be rescued by nutrients in the local microenvironment.

Discussion

In the present study, we found that A-T_RM and I-T_RM cells have different transcriptional and epigenetic profiles due to their unique localization within the tissue. A-T_RM cells showed increased apoptosis, decreased expression of cytolytic genes and a transcriptional signature indicative of increased cellular stress due to amino acid starvation. Comparisons of naturally generated A-T_RM cells and S-T_EM cells transferred into the airways indicated that expression of core T_RM genes was largely unaffected by the environment, and was instead regulated during initial T_RM cell differentiation after viral clearance. Thus, the lung CD8⁺ T_RM pool comprised functionally and genetically distinct A-T_RM and I-T_RM cell populations that shared a common T_RM cell differentiation program, but were further shaped by their respective microenvironments.

The CD8⁺ T_RM cell core lineage programming shared by both A-T_RM and I-T_RM cells is driven by interactions between the transcription factors Blimp1 and Hobit⁶. However, A-T_RM cells were also enriched in Ddit3 motifs and coordinately upregulated genes in the amino acid starvation response, with distinct accessible chromatin peaks at stress response genes such as Slc7a5 and Asns. Consistent with the concept that T_RM cell molecular programming is both lineage and environmentally determined^22,23, we observed only a partial adoption of the A-T_RM cell program after i.t. transfer of S-T_EM cells into the airways. The amino acid stress response, but not core T_RM genes, was induced after i.t. transfer, indicating that the core transcriptional program of A-T_RM cells was not driven by exposure to the local microenvironment. Thus, these data identify a common lung T_RM signature, distinct from the impact of different microenvironments within the lung.

A-T_RM cells are known to have a limited lifespan^30,35. The waning of the A-T_RM cell pool over time had suggested that the source of these newly recruited cells must also be similarly transient, but whether these cells come predominantly from the circulation or from within the tissue remained unclear. Our parabiosis experiments in mice previously exposed to influenza virus indicated that A-T_RM and I-T_RM cells were not maintained by recruitment of circulating cells into the lung T_RM cell pool, which agrees with recent reports that antigen encounter in the pulmonary environment, and not simply entry into the lung tissue itself, is required for T_RM cell differentiation^14,36,37. These data are in contrast to a study showing that circulating T_EM cells were able to re-seed the lung T_RM cell pool¹⁶. The observation that lung T_RM cells could develop independently of pulmonary antigen encounter under specific inflammatory conditions may explain this discrepancy³⁸. Thus, there may be scenarios where circulating T_EM cells can convert into lung T_RM cells, but further investigation is required to define these antigen-independent mechanisms.

Results in animal models of respiratory viral infections indicate that the poor longevity of A-T_RM and I-T_RM cells would be an impediment to the development of cell-mediated vaccines. However, approaches to maintain lung T_RM cells and protective cellular immunity for at least 1 year have been reported^39,40. Comparison of primary and quaternary CD8⁺ T cell memory indicates that lung T_RM cells generated by repeated influenza infections are resistant to apoptosis and persist in larger numbers³⁹, suggesting that prime-boost strategies may improve lung T_RM cell survival. In addition, vaccination with replication-defective adenoviruses expressing 4-1BBL sustained lung T_RM cells and prolonged heterosubtype immunity⁴⁰. Additional research into the mechanisms behind the longevity of lung T_RM cells is needed to inform the design of efficacious cell-mediated vaccine strategies.

The advantage or utility of relatively short-lived A-T_RM cells remains unclear. Severe influenza infections are accompanied by extensive tissue damage that leads to fluid leakage in the airspaces. Under these conditions, the airways could become an environment rich in nutrients and amino acids, and virus-specific CD8⁺ T cells would thus be able to maintain their cytolytic function and promote viral clearance⁴¹. In contrast, the nutrient-poor conditions in the airway during homeostasis may serve as a brake on CTL activity to avoid unnecessary damage to the epithelium. Notably, this nutrient-poor environment does not influence the secretion of antiviral cytokines such as interferon-γ, which is critical for the protection mediated by A-T_RM cells¹⁵. Therefore, the difference in nutrient availability among distinct microenvironments of the lung may create a division of labor among the lung T_RM cell pool, with A-T_RM cells serving primarily a ‘sensing and alarm’ function that could recruit cytolytic I-T_RM cells and other immune cells to the site of infection².

In summary, our results indicate that the lung T_RM cell pool comprised two distinct subsets—A-T_RM cells and I-T_RM cells—with distinct functions shaped by epigenetic reprogramming in response to environmental cues. Developing a more thorough understanding of how tissue microenvironments in the lung influence the genetic program of T_RM cells, and defining mechanisms by which cells adapt to and possibly overcome these challenging environments, will assist in the rational design of cell-mediated vaccines against respiratory pathogens.

Methods

Mice

C57BL/6J (WT), B6.SJL⁻Ptprca Pepc^b/BoyJ (CD45.1) and B6.129S(Cg)⁻Ddit3^tm2.1Dron/J (Ddit3^−/−) mice were purchased from the Jackson Laboratory and colonies were maintained at Emory University in specific pathogen-free conditions. Mice were aged between 8 and 13 weeks at the time of infection and housed under specific animal biosafety level 2 conditions after infection. All experiments were completed in accordance with the Institutional Animal Care and Use Committee (IACUC) guidelines of Emory University and Kindai University.

Infections

Mice were anesthetized with 300 mg kg⁻¹ of either Avertin (2,2,2-tribromoethanol, Sigma) or isoflurane (Patterson Veterinary) before infection. Mice were infected intranasally with 30,000 EID₅₀ (50% egg infectious doses) influenza A/HKx31 (x31, H3N2) or 3,000 EID₅₀ Sendai virus in a total volume of 30 μl. Heterologous challenge of x31-immune mice was performed with a dose of 10 LD₅₀ (median lethal dose; 3,000 plaque-forming units) of influenza A/PR8 (PR8, H1N1) in 50 μl. Challenged mice were monitored daily for weight loss and humanely euthanized if they fell below 75% of their original weight in accordance with Emory IACUC guidelines.

Intravenous labeling, single cell isolation and staining

Mice were intravenously labeled via tail vein injection under a heat lamp with either CD3e (1.5 µg of fluorophore-conjugated α-CD3ε antibody in 200 μl of phosphate-buffered saline (PBS)) or CD45.2 (2 µg of fluorophore-conjugated α-CD45.2 antibody in 200 μl of PBS). Mice were euthanized, 5 min after an intravenous injection, with Avertin. and exsanguinated before harvest of BAL, lung and other tissues. Lung and other tissues were disassociated as previously described¹⁵. Single cell isolations were Fc blocked using the antibody 2.4G2. Then they were surface stained with tetramers at room temperature for 1 h, followed by surface staining with listed antibodies. Cell viability was determined using either Zombie NIR (BioLegend) or 7AAD. Tetramers were against influenza epitopes NP_366-374 D^b and PA_224-233 D^b. For in vitro cultures, total BAL and CD8⁺ T cell-enriched spleens from influenza-immune mice were plated in round-bottomed plates in R10 (RPMI, 10% fetal bovine saline and 1% penicillin, streptomycin and glutamine) for 48 h in a 5% CO₂ incubator at 37 °C before sorting for RNA isolation and downstream analysis. For RNA-seq samples, CD8⁺CD44^Hi, CD62L⁻ T cells were sorted from the BAL and spleen pre- and post-in vitro culture.

Antibodies and flow cytometry

Marker	Clone	Manufacturer
CD8	53-6.7	BioLegend
CD4	RM-45	BioLegend
CD69	H1.2F3	BioLegend
CD103	M290	BD
CD103	2E7	BioLegend
CD11a	M17I4	BD
CD11a	M17I4	Invitrogen
CD44	IM7	Ebioscience
CD44	IM7	BioLegend
CD62L	MEL-14	BioLegend
Zombie NIR	–	BioLegend
7AAD	–	BioLegend
CD3e	145-2C11	BD
CD45.2	104	BioLegend
CD98	RL388	BioLegend
CXCR3	CXCR3-173	BioLegend
CD45.1	A20	BioLegend
Annexin V	–	BioLegend
BCL2	3F11	BD

Tetramers were provided by the National Institutes of Health (NIH) Tetramer Core Facility at Emory (H-2D^b Influenza A NP 366-374 (ASNENMETM) and H-2D^b Influenza A PA 224–233 (SSLENFRAYV)). All samples were run on an LSRII or Fortessa X20 (BD Biosciences) flow cytometer, or sorted on a FACSAria II (BD Biosciences). Flow cytometry data were analyzed using FlowJo v.10 software.

Parabiotic surgery

Parabiotic surgery was performed as described³⁶ with the following modification: each parabiont partner was infected with x31 (30,000 EID₅₀) and allowed to mature to a memory time point (28 d), then stitched together and maintained as parabiont pairs for 3 weeks. Equilibration was confirmed in the peripheral blood before separation, intravital labeling and analysis.

Mixed bone marrow chimeras

Mixed bone marrow chimeras were generated as previously described⁴² and allowed to reconstitute for 8 weeks before infection. CD45.1 and Ddit3^−/− mice were used as bone marrow donors, and CD45.2/CD45.1 heterozygous mice were used as recipients. Mice were irradiated using an RS2000 X-ray irradiator (Rad Source) and received 2 doses of 4.75 Gy, 6 h apart.

Intratracheal and intraperitoneal transfers

Intratracheal and intraperitoneal transfers were performed as described^15,35. Cells were isolated from the spleens of x31-immune mice (35–100 d post-infection) and sorted on the CD8⁺CD44^Hi, CD62L⁻ population to isolate T_EM cells. Between 5 × 10⁴ and 10 × 10⁴ cells were transferred intratracheally or intraperitoneally into congenic, infection-matched recipient mice and cells were collected by BAL or peritoneal lavage 2 d later. Cells were sorted on the basis of congenic marker staining to isolate transferred cells for downstream analysis.

Viral titers

Lung viral titers were measured after PR8 infection of naive congenic mice receiving either wild-type- or Ddit3^−/−-sorted A-T_RM cells as previously decribed ⁴³.

RNA-seq

For each population, 1,000 cells were sorted into RLT lysis buffer (Qiagen) containing 1% BME and total RNA purified using the Quick-RNA Microprep kit (Zymo Research). All resulting RNA was used as an input for complementary DNA synthesis using the SMART-Seq v.4 kit (Takara Bio) and 10 cycles of PCR amplification. Next, 1 ng cDNA was converted to a sequencing library using the NexteraXT DNA Library Prep Kit and NexteraXT indexing primers (Illumina) with 10 additional cycles of PCR. Final libraries were pooled at equimolar ratios and sequenced on a HiSeq2500 using 50-bp paired-end sequencing or a NextSeq500 using 75-bp paired-end sequencing. Raw fastq files were mapped to the mm9 build of the mouse genome using Tophat2 (ref. ⁴⁴) with the mm9 University of California Santa Cruz KnownGene reference transcriptome⁴⁵. The overlap of reads with exons was computed and summarized using the GenomicRanges⁴⁶ package in R/Bioconductor and data normalized to fragments per kilobase per million (FPKM). Genes that were expressed at a minimum of three reads per million (RPM) in all samples for each cell typed were considered to be expressed. DEGs were determined using the glm function in edgeR⁴⁷ using the mouse from which each cell type originated as a covariate. Genes with an FDR < 0.05 and absolute log₂(FC) > 1 were considered to be significant. For GSEA²¹, all detected genes were ranked by multiplying the sign of the fold change by the −log₁₀ of the P value between two cell types. The resulting list was used in a GSEA pre-ranked analysis.

ATAC-seq

For ATAC-seq^24,25, 2,000 cells were isolated by FACS, resuspended in 50 μl of nuclei isolation buffer (10 mM Tris-HCl, pH 7.4, 10 mM NaCl, 3 mM MgCl₂, 0.1% IGEPAL CA-630, molecular grade H₂O, filter sterilized), and centrifuged for 30 min at 500g and 4 °C. Nuclei were then resuspended in 25 μl of Tagmentation reaction buffer (2X TD buffer, 1 μl of Tagmentation enzyme, molecular grade H₂O (Illumina, Inc.)), incubated for 1 h at 37 °C, and DNA isolated by addition of 25 μl of lysis buffer (326 mM NaCl, 109 mM ethylenediaminetetraacetic acid, 0.63% sodium dodecylsulfate) with incubation for 30 min at 40 °C. Low-molecular-mass DNA was purified by SPRI-bead size selection and PCR amplified using Nextera indexing primer (Illumina) and 2x HiFi ReadyMix (KAPA Biosystems). Final libraries were purified by a second size selection and pooled at equimolar ratio for 50-bp paired-end sequencing on a HiSeq2500. Raw fastq reads were mapped to the mm9 build of the mouse genome using Bowtie⁴⁸ with the default settings. For the analysis of accessible regions, the first accessible peaks were identified for each sample using MACS2 (ref. ⁴⁹). Second, all unique peaks were merged and the read depth annotated for each sample and normalized to reads per peak using GenomicRanges⁴⁶ and R/Bioconductor. DARs were determined using edgeR⁴⁷ and those with an FDR < 0.05 and absolute log₂(FC) > 1 were considered to be significant. DARs were annotated to the nearest gene transcription start site using HOMER⁵⁰.

Integrative analysis

To integrate the RNA-seq and ATAC-seq data, we used a normalized, Euclidean distance, k-means clustering pipeline that we have previously described^26,27. First, DARs were annotated to the DEGs using the overlap of Entrez ID, resulting in 1,652 DARs mapping to 704 DEGs. Second, for each DAR–DEG combination, data were aggregated by cell type, variance normalized and a pair-wise Euclidean distance matrix calculated. The resulting matrix was k-means clustered using a k of 3. For each cluster, GO analysis was performed on the DEGs using DAVID⁵¹ and enriched motifs in the DARs were identified with HOMER⁵⁰ using the ‘findMotifsGenome.pl’ script. All other data display was done in R/Bionconductor.

Statistical analysis

Statistical analysis was performed using Prism (GraphPad Software). Each figure legend indicates methods of comparison and corrections.

Reporting Summary

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