Lung tissue–resident memory T cells optimize protection by IL-10 regulation of innate immunity

Respiratory viruses such as SARS-CoV-2 and endemic influenza A viruses (IAV) pose a constant threat to human health due to their high mutation rate, allowing for the emergence of increasingly infective and pathogenic strains. Both SARS-CoV-2 and IAV typically infect non-immune cells such as epithelial cells, which subsequently activate alveolar macrophages to produce inflammatory cytokines (Manicassamy et al., 2010; Flerlage et al., 2021). This inflammatory response is essential for immune cell recruitment and activation of adaptive immunity to resolve the infection; however, pathogenic strains or dysregulated immunity can trigger severe complications such as acute respiratory distress syndrome and systemic inflammation, also referred to as a cytokine storm (Clementi et al., 2021; Herold et al., 2015; Wei et al., 2023). Defining mechanisms involved in optimizing protection—i.e., promoting rapid viral clearance while minimizing lung damage—is therefore important for preventing and treating multiple types of respiratory viruses that may seed future pandemics.

T cell activation to respiratory virus infection occurs in the lung-associated lymph nodes (LN) and results in differentiation of virus-specific effector CD4⁺ and CD8⁺ T cells, which play important roles in lung viral clearance. CD4⁺ T cells mediate heterogeneous helper and effector roles in promoting B cell responses, CD8⁺ T cell differentiation, and viral clearance in the lung via direct effector function (Swain et al., 2012; Zens and Farber, 2014). CD8⁺ T cells that are recruited to the lung during a primary infection not only mediate protection through direct cytolytic and effector functions, but also can dampen lung inflammation through production of the regulatory cytokine, IL-10 (Sun et al., 2009). In primary responses to SARS-CoV-2, T cell recruitment to the lung was further associated with reduced lung damage and mortality in COVID-19 (Moss, 2022; Sette et al., 2023; Szabo et al., 2021). However, our understanding of mechanisms for control of lung inflammation during infection in mouse models is largely based on primary responses to new viruses, while human immune responses more often reflect repeated viral challenges and recall responses by memory T cells. In this context, the extent of lung inflammation and mechanisms for controlling recurrent respiratory infections are less clear.

Memory T cells generated to respiratory infections are heterogeneous and comprise circulating effector memory T cells (T_EM), which surveil through blood and tissues, and non-circulating tissue-resident memory T cells (T_RM), retained in the lungs and associated lymphoid organs in mice and humans (Poon et al., 2021a; Teijaro et al., 2011). In mice, lung T_RM mediate robust cross-strain protection to different IAV strains through multiple mechanisms including rapid effector cytokine production and enhanced immune cell recruitment from the periphery and local LN (Paik and Farber, 2021a; Paik and Farber, 2021b; Turner et al., 2014; Wu et al., 2014; McMaster et al., 2015). Importantly, lung T_RM mediate enhanced viral clearance with less morbidity compared with circulating T_EM cells (Teijaro et al., 2011; Zens et al., 2016), though the mechanisms for the enhanced protection by lung T_RM remain unclear.

Here, we use a mouse model of heterosubtypic IAV infection to elucidate the lung immune response to secondary challenge and the role of T_RM in modulating lung inflammation and promoting tissue protection. We show that at early time points (1–3 days) after infection, primary responses are marked by rapid induction of IFN-stimulated and innate cytokine pathways in the lung, while secondary responses had an attenuated innate response coincident with upregulation of genes for multiple T cell–mediated effector cytokines and the immunoregulatory cytokine IL-10. Using IL-10-GFP reporter mice, we identify lung T_RM as the principal producers of IL-10 in secondary infections at early time points and that lung macrophages express the highest levels of IL-10R. Inhibition of IL-10 signaling in secondary infections results in enhanced macrophage activation and inflammation, and decreased CD8⁺ T cell recruitment and effector function. We further show that human influenza-specific T_RM from lungs and associated LN are the predominant producers of IL-10, which is associated with enhanced effector function. Together, our findings reveal a conserved function for lung T_RM as dual producers of effector and regulatory cytokines to coordinate optimal protection and limit collateral tissue damage.

To investigate T cell–mediated, lung-localized secondary responses to respiratory viruses, we used a heterosubtypic model of influenza infection (Liang et al., 1994) in which mice infected previously with one serotype of influenza (X31, H3N2) are challenged 5–6 wk later as “memory” mice with a disparate (heterosubtypic) strain of IAV (PR8, H1N1). Primary infection controls were naïve mice infected with PR8 (Fig. S1 A). In the secondary response, memory mice exhibited enhanced protection to PR8 infection relative to the primary response manifested by increased survival and reduced weight loss morbidity (Fig. 1, A and B), along with enhanced lung viral clearance consistent with previous studies (Paik and Farber, 2021a; Teijaro et al., 2010; Wu et al., 2002; Wu et al., 2014).

We hypothesized that events occurring early in the lungs during the primary and secondary response were driving the distinct morbidity and recovery at later times. For an unbiased assessment of lung-localized immune responses, we performed RNA-seq on whole mouse lungs from naïve and memory mice at baseline (day 0) and 1–3 days after infection (dpi)—time points prior to the markedly divergent weight loss in primary versus secondary infections (Fig. 1 B). Principal component analysis (PCA) of the RNA-seq data revealed distinct clustering between samples obtained from naïve and memory mice at baseline and 1 dpi compared with later times 2–3 dpi along PC1, and between the respective primary and secondary responses along PC2 (Fig. 1 C). Genes varying along the infection axis (PC1) include interferon-stimulated genes (ISGs) (Mx1, Mx2, Oas2, and Ifit2) and inflammatory genes (Tnf, Ifnl2, Ifnl3, Il6, Tlr7, and Btk) (Fig. 1 C, Fig. S1 B, and Table S1), while genes distinguishing primary and secondary responses (PC2) are associated with adaptive immunity, including multiple immunoglobulin genes and genes associated with T cell activation (Ctla4, Ifng, Runx3, Eomes, Cd3e) as determined by gene ontology (GO) analysis (Fig. 1 C, Fig. S1 B, and Table S1). These results indicate that key sources of variation in the RNA-seq results were due to the time point of infection and the nature of the immune response as primary or secondary.

We performed differential expression analysis for the RNA-seq results for each infection time point compared with uninfected naïve and memory lung samples, revealing 5 distinct gene clusters (C1–C5) that are differentially expressed between primary and secondary responses over time (Fig. 1, D–F). Cluster 1 (“antiviral response”) contains multiple ISGs (Mx1, Ifit2, and Oas2), type I and III interferons (Ifna1, Ifnb1, Ifnl2), and inflammatory cytokines (Il1a, Il6) that are upregulated more rapidly and to a greater extent at early times in the primary compared with the secondary response (Fig. 1, D–F). Cluster 2 (“innate immunity”) comprises genes enriched for innate cell activation (Klri2, Il18, Cd84, Ccr1) and inflammasome activation (Casp, Cgas) that are upregulated to a greater extent at 1 and 3 dpi in the primary compared with the secondary response. Cluster 3 (“T cell activation”) has genes associated with activated T cells (Ifng, Il10, Il2rg, Btk, Ikzf1, Tcf7, and Lck) that are specifically upregulated in the secondary but not primary response (Fig. 1, D–F). Cluster 4 and 5 genes are associated with metabolism/mitochondrial function (Nd1, Nd2, Cox1, Cox2, Cytb) and cell structure/tissue organization (Dnai2, Dnah9, Cfap43, Ttl11, Invs), respectively, and are downregulated following infection in both primary and secondary responses, consistent with infection-induced structural changes to the lung (Bonnafe et al., 2004; Chen et al., 2015; Tilley et al., 2015).

To assess whether differences in immune activation at these early time points resulted from different viral loads, we quantified viral gene expression from the RNA-seq data (see Materials and methods). There were no significant differences in the level of virus gene transcripts in the lung at 1–2 dpi for primary versus secondary infection, though we observed slightly lower viral transcripts encoding polymerase genes (PB1,2) at 3 dpi in the secondary response (Fig. 1 G). These results are consistent with enhanced viral clearance in the secondary compared to primary response showing maximal difference in viral loads at 4–6 dpi (Paik and Farber, 2021a; Teijaro et al., 2010; Teijaro et al., 2011; McKinstry et al., 2012). Therefore, the major transcriptomic differences between the primary and secondary immune response at 1–2 dpi cannot be attributed to viral burden. Together, this integrated gene expression analysis of lung responses reveals more rapid induction and elevated innate immunity in the primary compared with the secondary response, while the secondary response is associated with enhanced T cell activation and attenuated innate immunity.

We further examined the major pathways that differed between the primary and secondary immune responses, consisting mainly of genes encoding cytokines and chemokines and their signaling pathways (Fig. 2 A). Genes with greater expression in the primary compared with the secondary response included multiple type I and III IFNs (Ifna1, Ifnb1, Ifnl2, Ifnl3) and key macrophage-derived cytokines and chemokines (Il1b, Il6, Ifnb1) (Mosser and Edwards, 2008), showing peak expression at 2 dpi and decreasing by 3 dpi (Fig. 2, A and B). In contrast, genes upregulated in the secondary relative to the primary response also exhibited peak expression at 2 dpi but were associated with T cell activation and included cytokines produced by T helper 2 (Th2) and T follicular helper (T_FH) cells (Il13, Il21, Cxcl9), Th1 cells (Il2, Ifng, Lta), and those with immunoregulatory functions (Il10) (Fig. 2, A and C). We assessed cytokine content in bronchoalveolar lavage (BAL) samples at these same time points, confirming the distinct cytokine profiles and kinetics revealed by RNA-seq (Fig. 2, D and E). The primary response showed elevated levels in BAL of type I IFNs, IL-1α, IL-6, TNF-α, and GM-CSF compared with the secondary response, which in turn had enhanced production of T cell–derived cytokines (IFN-γ, IL-13, IL-22) compared with the primary response (Fig. 2, D and E). These results show striking differences in immune mediator expression in the lungs during a primary and secondary response, including a significantly reduced innate immune response in previously infected lungs.

We hypothesized that the immunoregulatory cytokine IL-10, which was exclusively upregulated in the secondary response at early times after infection, may play a role in dampening innate immunity and inflammation during the secondary response. However, IL-10 protein, known to be secreted in small amounts (Sun et al., 2009), was not detected in the BAL, requiring a more targeted approach to assess its expression and role in infection. We therefore used an IL-10 GFP reporter mouse strain (IL-10 IRES GFP-enhanced reporter[tiger]) (Kamanaka et al., 2006), which enables in vivo identification of IL-10–producing cells based on GFP expression. Consistent with our RNA-seq results, we saw a marked increase in IL-10 production by CD45⁺ cells in mouse lungs during the secondary compared with the primary response (Fig. 3 A). T cells were the primary source of early IL-10 at 2 and 3 dpi in the secondary response, with other immune cells producing significantly lower amounts (Fig. 3 B; see Fig. S2 A for gating strategy). While both CD4⁺ and CD8⁺ T cells produced IL-10 in secondary responses, CD8⁺ exhibited a greater upregulation relative to baseline (∼26.5-fold) compared with CD4⁺ (∼4-fold) T cells (Fig. 3 C).

To determine whether IL-10 derived from circulating or tissue-resident T cells in the lung, we intravenously (IV) administered mice fluorescently labeled anti-CD45 antibody to distinguish circulating T cells, which bind antibody (CD45(IV)⁺), from those resident in the lung, which are protected from labeling (CD45(IV)⁻) (Anderson et al., 2014; Teijaro et al., 2011; Turner et al., 2014). IL-10 was predominantly produced in the secondary response (compared with negligible IL-10 detected in the primary response) by CD45(IV)⁻ CD4⁺ and CD8⁺ T cells (Fig. 3, D and E), indicating their localization in the lung niche (Turner et al., 2014). Moreover, these IL-10–producing lung T cells expressed multiple T_RM markers, including CD69, CD44, and CXCR6 (Fig. 3 F), the latter marker being associated with lung homing (Wein et al., 2019). CD103, a canonical marker expressed by CD8⁺ T_RM in epithelial sites including lungs (Szabo et al., 2019; Turner et al., 2014; Masopust and Soerens, 2019), was expressed by a low proportion of IL-10–producing CD8⁺ (and CD4⁺) T cells (Fig. 3 F), suggesting that IL-10 is produced by a subset of T_RM. In addition, only a small portion (∼10%) of IL-10–producing CD4⁺ T cells expressed FoxP3 (Fig. 3 F), a lineage-defining marker for Treg cells (Hori et al., 2003). These findings demonstrate that CD8⁺ and CD4⁺ T_RM (and not Tregs) are the major source of early IL-10 production in lung secondary responses to influenza challenge.

We examined the expression of IL-10 receptor (IL-10R) on lung immune cells to identify potential targets of IL-10 signaling. Importantly, lung macrophages expressed the highest level of IL-10R relative to other immune cell subsets including NK cells and B cells, which expressed low levels of IL-10R (Fig. 4, A and B). Nearly all macrophages expressed IL-10R, accounting for the majority (∼51%) of IL-10R expressed by lung immune cells (Fig. 4, A and B). To assess whether macrophages exhibited distinct functions in primary and secondary responses that could indicate effects of IL-10, we measured macrophage-derived cytokines at early times after infection in primary and secondary responses. In the primary response, macrophages produced substantial amounts of TNF-α at 3 dpi, which was significantly reduced in frequency and magnitude in the secondary response (Fig. 4, C and D). This reduced macrophage inflammation in the secondary compared with the primary response is consistent with lower levels of TNF-α in the BAL in secondary versus primary infection (Fig. 2 D), and that the major inflammatory cytokines in the lungs are associated with macrophage activation. These findings show that IL-10 secretion in the secondary lung response to viral challenge correlates with reduced production of inflammatory cytokines by macrophages, suggesting macrophages as targets for IL-10–mediated regulation.

To specifically interrogate the role of early IL-10 signaling in secondary responses, we administered anti-IL-10R blocking antibody (αIL-10R) or an IgG isotype control to memory mice at the time of heterosubtypic infectious challenge (Fig. 5 A). Mice treated with αIL-10R exhibited increased weight loss morbidity reaching a nadir at 4 dpi and delayed recovery at 5–6 dpi relative to control mice who lost less weight overall and recovered starting at 4 dpi (Fig. 5 B). This morbidity difference was not due to viral load, which was similar at all time points between αIL-10R–treated compared with control mice, including at peak infection time points (4 dpi, Fig. 5 C). We further investigated whether inhibiting IL-10 signaling led to increased lung inflammation by histological analysis of lung samples after the resolution of infection (2–5 weeks after infection [wkpi]). Lungs of influenza-challenged memory mice treated with αIL-10R demonstrated increased immune cell infiltration forming larger inflammatory lesions and a greater total area of inflammation at 2 wkpi (Fig. 5, D and E), which was resolved by 5 wkpi. Taken together, these results demonstrate that early IL-10 signaling in the lung in secondary responses controls inflammation and morbidity while maintaining viral clearance.

To understand how early IL-10 reshapes lung secondary responses, we performed RNA-seq on lungs of mice undergoing secondary challenge with or without αIL-10R blockade at 2 dpi. Inhibition of IL-10 signaling resulted in few significant changes in lung gene expression including downregulation of genes associated with M2-like macrophages (Ccl24, Chil4 (Ym2), Arg1) (Arora et al., 2018; Gordon and Martinez, 2010; Lee et al., 2020) (Fig. 6 A). Consistent with gene expression data, we found reduced expression of Arg1 in lung macrophages by flow cytometry and decreased concentration of CCL24 in the BAL of αIL-10R versus control mice (Fig. 6, B and C). Together, these data suggest that IL-10 signaling alters macrophage polarization toward a regulatory M2-like phenotype.

Concurrently, inhibition of IL-10 signaling caused an early upregulation at 3 dpi of BAL cytokines associated with innate immune activation, such as the neutrophil chemoattractant CXCL5 (LIX), the macrophage-stimulating cytokine M-CSF, and the monocyte chemoattractant MCP-1 (Fig. 6 D). This increase in the monocyte-recruiting and macrophage-stimulating cytokines at 3 dpi preceded an increase in lung macrophages at 4 dpi (Fig. 6 E). The expanded macrophage population exhibited an inflammatory M1-like profile (Murray et al., 2014) including a CD11b⁺Ly6C⁺ phenotype and increased expression of inducible nitric oxide synthase (iNOS) and inflammatory cytokine TNF-α (Fig. 6 F). This enhanced inflammatory profile under IL-10 blockade is consistent with our findings of elevated TNF-α production in the primary (versus secondary) response when IL-10 production is lacking (Fig. 4, C and D). Taken together, these findings indicate that IL-10 signaling in secondary responses can modulate lung inflammation through direct effects on macrophage recruitment, expansion, and/or function.

To establish a direct link between lung T_RM and modulation and macrophage numbers and function, we assessed the effect of lung T_RM depletion on the lung secondary response to influenza challenge. Administration of Thy1.2-depleting antibody resulted in significantly decreased CD4⁺ and CD8⁺ T cells in the lung (Fig. 7 A), including significant reductions in CD45(IV)⁻CD69⁺ T_RM for both CD4⁺ and CD8⁺ T cells (Fig. 7 B). Cohorts of T_RM-depleted and control memory mice were challenged with the heterologous strain PR8, and lungs were analyzed at 4 dpi, similar to the time point analyzed following infection in mice treated with IL-10R blockade (Fig. 5 C). There were substantially greater numbers of lung macrophages in T_RM-depleted compared with control mice after infectious challenge (Fig. 7 C). To examine the relationship between T_RM and macrophages, we performed correlation analyses, revealing a negative correlation trend between T_RM and macrophages (Fig. 7 D) and a significant positive correlation between T_RM and anti-inflammatory Arg1⁺ macrophages (Fig. 7 E). Together, these findings suggest that T_RM constrain macrophage activation and inflammatory functions in secondary responses. The enhanced macrophage responses seen by IL-10 blockade and T_RM depletion suggest that T_RM-derived IL-10 is a major factor controlling local inflammation and macrophage polarization.

In addition to its immunoregulatory functions, IL-10 can also promote effector functions in NK, B, and T cells (Brooks et al., 2010; Groux et al., 1998; Heine et al., 2014; Mocellin et al., 2004). Our results showed increased expression of genes relating to lymphocyte activation in lungs of αIL-10R compared with control mice (Fig. 6 A), prompting further investigation of T cell responses. In αIL-10R–treated mice, there was a marked reduction in the proportion of CD8⁺ T cells compared with control mice at 4 dpi, while the proportion of CD4⁺ T cells and other lymphocyte lineages was unchanged (Fig. 8 A). This reduction in CD8⁺ T cells was due to lower frequencies of CD44^hi memory cells including T_EM (CD44⁺CD62L⁺) and T_RM (CD44⁺CD69⁺) subsets (Fig. 8 B and Fig. S2 B). Lung CD8⁺ T_EM and T_RM also exhibited lower IFN-γ production in αIL-10R–treated versus control mice (Fig. 8 C), also confirmed by the lower levels of secreted IFN-γ in the BAL (Fig. 8 D). The effects on CD8⁺ T cell activation resulted from direct IL-10 signaling, as demonstrated by ex vivo assays showing augmented IFN-γ production by lung CD8⁺ T cells stimulated in the presence of IL-10 (Fig. 8 E). Thus, early IL-10 signaling in secondary infections alters CD8⁺ T cell differentiation, activation, and functional capacity.

Our results above demonstrated in mice that lung T_RM rapidly produce IL-10 during secondary influenza infection, which simultaneously attenuates innate immunity and enhances T cell effector responses. To address whether human influenza–specific T cells in tissues could mediate similar responses as in mice, we evaluated the functional profile of T cells isolated from lungs, lymphoid organs, and blood of human organ donors that we have extensively validated for human immunology studies (Carpenter et al., 2018; Farber, 2021; Poon et al., 2021a; Thome et al., 2014). We identified influenza-reactive T cells in blood, spleen, lung, and lung-draining lymph node (LLN) of multiple donors (Fig. 9 A) using the activation-induced marker (AIM) assay (Grifoni et al., 2020) adapted for tissue T cells (Poon et al., 2021b; Davis-Porada et al., 2024). Antigen reactivity was determined after culture of mononuclear cells from each site with influenza peptide pools by the expression of AIM markers 4-1BB and OX40 for CD4⁺ T cells and 4-1BB and CD25 for CD8⁺ T cells (Fig. 9 B). To fully assess their transcriptional and functional profile, we sorted flu-reactive AIM⁺ T cells for profiling using Cellular Indexing of Transcriptomes and Epitopes (CITE-seq) (Fig. 9 C) as previously described (Davis-Porada et al., 2024). Clustering of the resultant T cells from all 4 sites and 12 donors visualized by Uniform Manifold Approximation and Projection (UMAP) revealed distinctions between CD4⁺ and CD8⁺ T cells, unstimulated and flu-reactive cells, blood and tissue sites, but not across donors (Fig. 9 D).

Gene set enrichment analysis (GSEA) of the human flu-reactive T cell response revealed a significant enrichment for murine genes (identified in Fig. 1, D–F) as upregulated in either the T cell activation cluster or the secondary infection response (Fig. 9 E). Lead genes with particularly high enrichment scores included TNFRSR4, IFNG, LTA, GZMB, TIGIT, CTLA4, and TNF (Table S1). Furthermore, human flu-reactive T cells upregulated similar cytokine genes (Fig. 9 F) as those expressed in the mouse secondary response (Fig. 2, A and C). Certain cytokines (e.g., IL2, IL13) were primarily expressed by flu-reactive T cells in circulation (blood and spleen), while others (e.g., TNF, IFNG) were expressed by flu-reactive T cells in both circulation and tissue (lung and LLN) (Fig. 9 F). In humans, IL10 expression was enriched in human flu-reactive T cells present in tissues, but not in blood, and was significantly higher in human flu-reactive CD8⁺ T cells compared with CD4⁺ T cells (Fig. 9, G and H), consistent with our findings in mice. Moreover, the expression of IL10 was highest on T_EM (CCR7⁻CD45RA⁻) compared with any other subset (Fig. 8 I), including CD4 Tregs, based on a multimodal classifier (see Materials and methods). Flu-reactive CD4⁺ and CD8⁺ T_EM also expressed multiple T_RM genes (CXCR6, PDCD1, ITGA1, ITGAE) (Fig. 9 J) and surface markers (CD69, PD-1, CD49a) (Fig. 9 K). Together, these findings demonstrate that human flu-reactive T_RM are distinct in their production of IL10 relative to circulating memory T cells and other tissue T cell subsets.

To characterize the functional capacity of IL10-expressing cells, we identified protein-coding genes that are co-expressed with IL10 consistently across donors (see Materials and methods). Transcripts with positive R values, indicating co-expression with IL10 beyond what would be expected by chance, include FABP5 and DUSP6, which are known to mark tissue-resident cells (Pan et al., 2017; Kumar et al., 2017), as well as LAG3, a negative regulator of T cell responses (Fig. 9 L). In addition, IL10 was co-expressed with effector cytokines and chemokine genes including IFNG, GZMB, GZMH, and CSF2 (Fig. 9 L); these genes were lowly expressed in flu-reactive T cells that did not express IL10 but expressed equivalent levels of AIM markers (while unactivated T cells did not express AIM markers or cytokines) (Fig. 9 M). Together, these findings demonstrate that human T_RM-mediated IL-10 production is associated with other types of effector cytokines, such as IFN-γ, consistent with our results with mouse T_RM.

T_RM cells in the lung and other tissues mediate optimal protective immunity manifested by enhanced viral clearance, reduced morbidity, and more rapid recovery (Teijaro et al., 2011; Paik and Farber, 2021a; Paik and Farber, 2021b), though the underlying mechanisms have been unclear. Here, we used an unbiased sequencing approach combined with targeted analysis using cytokine reporter mice, specific cytokine blockade, and T cell depletion to reveal that lung T_RM produce IL-10, which promotes optimal protective immunity by attenuating macrophage-derived inflammation and enhancing CD8⁺ T cell effector responses. Notably, inhibiting IL-10 signaling caused increased morbidity and lung pathology in the secondary response and depleting lung T_RM results in enhanced macrophage responses. Functional analysis of human IAV-specific T cells recapitulates these findings, demonstrating that human T_RM are key producers of both IL-10 and effector cytokines. These results reveal a mechanism by which T_RM coordinate efficacious protection through dual functional roles and a paradigm for vaccine targeting.

Through unbiased transcriptome analysis of the lung response to influenza infection, we show that secondary infection triggers a significantly attenuated innate immune response and macrophage-derived inflammatory profile compared with primary infection, including substantially reduced levels of type I and type III interferons, the latter of which acts specifically on epithelial cells to reduce viral load (Wack et al., 2015). This downregulation of lung macrophage activation could be attributed to tolerization, which occurs through prolonged exposure to activating signals like LPS (O’Carroll et al., 2014; Seeley and Ghosh, 2017); however, this is a transient state and not likely to persist 5 wkpi when the secondary challenge was given. Lung macrophages can also undergo changes following initial infection through a phenomenon known as “trained immunity” in which macrophages can exhibit enhanced clearance of pathogens to a secondary relative to a primary challenge (Affolter and Moore, 2002; Yao et al., 2018), though the differences in the IFN responses between primary versus secondary infection occurred at 48 h after infection when lung viral loads were similar between mouse groups. However, not all innate cytokines were suppressed in secondary infection—transcripts for Il1b, Il1a, and IL6 were upregulated, consistent with a prior analysis of lung cytokines in memory T cell responses to influenza infection (Strutt et al., 2010). It is possible that a prior respiratory infection confers multiple changes in the lung including T_RM development and a trained or modified innate response, which together contribute to targeted pathogen control with decreased inflammation and tissue damage.

Our results reveal a robust and heterogeneous T cell–mediated effector response in lung secondary responses to influenza challenge comprised of Th1- and Th2-like cytokines (IFN-γ, IL-4, IL-13, IL-22), and the T_FH-like cytokine IL-21 along with chemokines such as CXCL9 within 2 dpi. These findings reveal a comprehensive profile of memory T cell–directed immune responses with specific findings consistent with previous studies of in situ activation of T_RM (McMaster et al., 2015; Turner et al., 2014), of a T_FH-like subset of lung T_RM (Swarnalekha et al., 2021; Son et al., 2021), and lung cytokine profiles in flu-challenged mice with influenza-specific memory T cells (Strutt et al., 2010). In addition, we found robust production of IL-10 at early time points in the secondary but not the primary response. While IL-10 production in a primary response to influenza virus is produced at later times after infection by newly recruited lung T cells (Sun et al., 2009), enhanced IL-10 in the secondary response identified here was produced predominantly by T_RM (CD4⁺ and CD8⁺ T_RM) and not by circulating memory T cells.

We show that early IL-10 production in the secondary response had pleiotropic effects on innate and adaptive immunity in the lung. IL-10 directly regulated macrophage function and polarization through upregulation of Arg1, a canonical marker of anti-inflammatory (also referred to as M2-like) macrophages associated with wound healing (Eming et al., 2021; Munoz-Rojas et al., 2021). Abrogation of IL-10 signaling resulted in decreased Arg1⁺ macrophages and increased pro-inflammatory macrophages, similar to findings in inflammatory bowel disease models showing IL-10–mediated control of local inflammation by intestinal macrophages (Shouval et al., 2014a; Shouval et al., 2014b). In bacterial infection models, IL-10 signaling in macrophages can directly suppress clearance (Branchett et al., 2024); however, IL-10 production in the lungs did not inhibit viral clearance, likely due to distinct mechanisms for viral control. Our findings that T_RM are the predominant source of IL-10, that T_RM depletion resulted in enhanced macrophage inflammation, and our observed correlation between lung T_RM and M2-like macrophages support T_RM-derived IL-10 regulating in situ lung inflammation and tissue damage. However, we cannot rule out that other sources of IL-10 may also be contributing to the observed effects.

A second effect of IL-10 signaling in the lungs is augmenting CD8⁺ T cell numbers and their effector functions. While IL-10 produced by intestinal T cells in a colitis model has been shown to suppress T cell effector function in the intestines (Chaudhry et al., 2011; Huber et al., 2011), IL-10 can enhance T cell survival and effector function in the central nervous system (CNS) in the mouse experimental autoimmune encephalomyelitis model of CNS inflammation (Yogev et al., 2022)—an effect similar to our results in the lung. Importantly, human T_RM also produce IL-10 as part of the core T_RM signature (Kumar et al., 2017) and we show here that influenza-specific human lung and lymphoid T_RM produced IL-10, which is associated with robust effector functions. In addition to regulating infection responses, steady-state production of IL-10 is crucial in tissue homeostasis as mice with targeted Il10 deletions spontaneously develop colitis, which is reversed by IL-10 supplementation (Asseman et al., 1999; Carlini et al., 2023; Ranatunga et al., 2012). IL-10 production has also been associated with cytokine storms and suppressive tumor microenvironments, enabling tumor growth and metastasis (Ravi et al., 2022; Sato et al., 2011), suggesting context-dependent effects. We propose that limiting IL-10 production largely to T_RM at the infection site provides transient downregulation of inflammation while avoiding sustained signaling.

T_RM are generated by and mediate protection to diverse respiratory pathogens—including different viruses (Teijaro et al., 2011; Turner et al., 2014; Luangrath et al., 2021; Zhao et al., 2016) and bacteria (Sakai et al., 2014; Wilk et al., 2017). Human lung T_RM can likewise be generated to respiratory pathogens such as influenza and SARS-CoV-2 (Poon et al., 2021a; Poon et al., 2021b; Davis-Porada et al., 2024). While T_RM can be established from site-specific vaccination in mouse models (Zens et al., 2016; Allen et al., 2018; Rotrosen and Kupper, 2023), targeting protective T_RM is not achieved by current vaccines. Human T cell–mediated protection in vaccines correlates to the production of multiple pro-inflammatory cytokines as assessed in blood (Seder et al., 2008; Darrah et al., 2007). However, our findings indicate that production of regulatory cytokines in situ is an essential component of protection. The dual functionality of lung T_RM as elucidated here orchestrates a precise balance of protection between viral clearance and maintaining tissue homeostasis can provide a new functional target for vaccines and immunomodulators.

The findings presented here show a major role of IL-10 in controlling lung inflammation in the secondary response, that IL-10 is produced by lung T_RM, which also regulate macrophage responses. We were not able to specifically knock out IL-10 in lung T_RM, and the human data do not show a connection of influenza-specific lung T_RM with macrophage responses.

Wild-type (WT) female C57BL/6J mice (strain 000664) were purchased from the Jackson Laboratory. IL-10-GFP reporter mice (B6.129S6-Il10^tm1Flv/J; strain 008379) were purchased from the Jackson Laboratory and bred under specific pathogen-free conditions in a BSL2 biocontainment room at Columbia University Irving Medical Center Animal Facility. IL-10-GFP reporter mice were genotyped according to a PCR protocol from the Jackson Laboratories. WT female C57BL/6J mice were used for all experiments where WT mice are indicated, and mixed male and female mice were equally distributed in experimental groups for all experiments with IL-10-GFP reporter mice. All animal studies were approved by Columbia University Institutional Animal Care and Use Committee.

Influenza strains A/HKx31 (X31, H3N2) and A/Puerto Rico/8/34 (PR8, H1N1) were grown in eggs and titered as previously described (Teijaro et al., 2010). C57BL/6J (WT) or IL-10-GFP reporter mice were anesthetized with isoflurane and infected intranasally with either X31 or PR8. Memory mice were generated by infecting naïve mice with a sublethal dose of X31 (5000TCID₅₀), monitored daily for weight loss and infection signs, and maintained beyond recovery for 4–6 wkpi. The resultant memory mice were infected with heterosubtypic strain PR8 (H1N1) at 1000 TCID₅₀. For experiments comparing primary and secondary infections, naïve and memory mice were infected with the same dose of PR8. All influenza infections were performed in Biosafety Level 2 biocontainment animal facilities.

Lung lobes (left) were dissected from mice and stored in RNAlater (Thermo Fisher Scientific) at 4°C until RNA extraction. Whole lung lobes were dissociated with the TissueLyser II (Qiagen) according to the manufacturer’s protocol, and total RNA was purified with the QIAsymphony SP (Qiagen). RNA concentration and quality were determined by Bioanalyzer (Agilent) or Tapestation 4200 RNA ScreenTape Assay (Agilent). For RT-qPCR, RNA was isolated with RNeasy Mini Kit (Qiagen) according to the manufacturer’s protocol.

Total RNA isolated from lung samples was enriched for mRNA using a poly-A pulldown, and RNA-seq libraries were prepared with Illumina TruSeq chemistry. Libraries were then sequenced by the Illumina NovaSeq 6000 by the Columbia Genome Center. Pseudoalignment of the sequencing data was performed with Kallisto (v.0.44.0) (Bray et al., 2016) with a minimum fragment length of 100 to mouse reference genome build GRCm39. Differential gene expression analysis was performed with DESeq2 (v.1.32.0) (Love et al., 2014). Genes with an adjusted P value (P_adj) ≤0.05 and a fold change ≥2 were considered significantly differentially expressed. PCA plots were generated with DESeq2. Principal component genes were obtained through PCAexplorer (v.2.18.0) (Marini and Binder, 2019). Heatmaps were generated with pheatmap (v.1.0.12) (Kolde, 2012) and show row-normalized rlog-transformed counts. Clustering was performed by k-means clustering. Biological replicates were averaged for each condition and time point. GO overrepresentation analysis was performed with ClusterProfiler (v.4.0.5) (Yu et al., 2012) and mapped to the GO Biological Processes database (Ashburner et al., 2000; Gene Ontology Consortium et al., 2023). The fold change of each cluster was determined by calculating the fold change of the sum of all genes in a cluster and normalizing each time point to baseline.

To quantify viral reads, the IAV (A/Puerto Rico/8/34 [H1N1]) genome was concatenated to the mouse reference genome (GRCm39) and a pseudoalignment was performed with Kallisto (v.0.44.0) with a minimum base fragment length of 100. Differential gene expression analysis was performed with DESeq2 (v.1.32.0) between mice undergoing primary and secondary infection, and genes were considered significantly differentially expressed if P_adj ≤ 0.05 between primary and secondary infections at the same time point.

BAL samples were isolated from naïve and memory mice at 0–5 dpi by flushing mouse lungs with 5 ml of PBS. BAL samples were concentrated by centrifugation and quantified by Eve Technologies using the Luminex xMAP technology for multiplexed cytokine quantification.

A standard curve was developed by extracting RNA from PR8 viral stock using QIAamp Viral RNA Mini Kit (Qiagen) according to the manufacturer’s protocol. Reverse transcription was performed from RNA using the Superscript III First Strand Synthesis kit (Invitrogen) according to the manufacturer’s protocol. Viral titer was determined by qPCR for M1/M2 PR8 RNA (FW: 5′-GGA​CTG​CAG​CGT​TAG​ACG​CTT-3′, RV: 5′-CAT​CCT​GTT​GTA​TAT​GAG​GCC​CAT-3′).

Mice were injected IV with 2.0 μg of anti-CD45 antibody conjugated to Alexa Fluor 700 (clone 30-F11; BioLegend), and after 4 min, mice were euthanized and lungs were dissected. Lungs were mechanically dissociated in gentleMACS tubes (Miltenyi) in RPMI supplemented with collagenase D (Sigma-Aldrich), DNase (Sigma-Aldrich), and trypsin inhibitor (Sigma-Aldrich) as done previously (Paik and Farber, 2021a). Samples were incubated at 37°C on a shaker for 45 min and then passed through a 70-μM filter. Cells were stained for 30 min on ice in PBS containing 5% fetal bovine serum (FBS) using antibodies targeting extracellular markers (Table S2). For IL-10 (GFP) quantification, samples were fixed with BD Cytofix Fixation Buffer diluted to 1% formaldehyde (wt/wt) with PBS at RT for 25 min and then directly acquired. For intracellular staining, samples were permeabilized with FoxP3 Fixation/Permeabilization Kit (Tonbo Biosciences) for 30 min at RT after initial fixation with BD Cytofix. Intracellular staining was performed for 30 min on ice in PBS supplemented with 5% FBS. All samples were acquired on a 5-laser Cytek Aurora with SpectroFlo (Cytek), and data were analyzed using FlowJo (v.10.10.0).

Mice were injected intraperitoneally (IP) with 1 mg of either InVivoMAb anti-mouse IL-10R (clone 1B1.3A; BioXCell) or anti-horseradish peroxidase IgG1 isotype control (clone HRPN; BioXCell) at the same time as infection.

Left lobes of mice were dissected at the indicated time points and fixed in 4% PFA overnight. Samples were processed by the Columbia University Irvine Medical Center Molecular Pathology Core using the Tissue-Tek VIP 5 Tissue Processor (Tissue-Tek) and embedded in paraffin with the Tissue-Tek Embedding Center (Tissue-Tek). Tissue was sectioned with the Microtome (Leica), and hematoxylin and eosin (H&E) staining was performed using Tissue-Tek Slide Stainer (Tissue-Tek). Slides were scanned under 40× resolution with the Leica Aperio AT2 (Leica).

HALO-AI software and Densely Connected Convolutional Networks (DenseNet) (Huang et al., 2017) were employed for image analysis and development of the learning model. DenseNet, which combines any layer to all subsequent layers (such that the output layer receives inputs from the features of all previous levels), is composed of convolutional layers, pooling, batch normalization, activation function, transition layers, dense blocks, and a classification layer.

H&E-stained slides (scanned at 40×) were manually annotated for two classes. The first class was inflammation, and the second class was background (no inflammatory cells). Slides (3 each) were randomly assigned for training and validation (243 annotations), and 3 slides (187 annotations) were used for testing utilizing the DenseNet model; there was no duplication of slides. The algorithm was trained and evaluated for classification with a resolution value of 2 μm/pixel. The model was evaluated using a threefold cross-validation with statistically powered measure of F1 score: defined as 2 × TP/(2 × TP + FP + FN), where TP is true positives (samples correctly classified as positive), FP is false positives (samples wrongly classified as positive), and FN is false negatives (samples wrongly classified as negative). The HALO-AI operator cross-entropy rates were 0.006 for 4,480 ± 80 iterations, and the DenseNet classification F1 scores were overall 0.920 (95% CI 0.915–0.925), inflammation 0.875 (95% CI 0.868–0.881), and background 0.965 (95% CI 0.969–0.962).

The DenseNet model randomly analyzed all cases and assigned inflammation area to each slide. In addition, the entire tissue area for each case was computed using HALO-AI.

To deplete T_RM from mice, WT mice were infected with X31 influenza virus as above, and 3 wkpi, the resultant memory mice were administered 10 μg/g of Thy1.2-depleting antibody (BE0066; BioXCell) by IP injection four times over the course of 1 wk. Influenza challenge was done 2 wk after the last T cell injection to allow for partial recovery of circulating endogenous T cells.

Lung cells were isolated as previously described, and 10⁶ cells were co-cultured with 25 µL of Dynabeads Mouse T-Activator CD3/CD28 beads (Gibco) in RPMI supplemented with 10% FBS and 1% penicillin/streptomycin for 2 days. IL-2 (30 U/ml) and IL-10 (50 U/ml) were supplemented where indicated.

Mononuclear cells were isolated from blood of living subjects collected through an approved Columbia University Institutional Review Board protocol and blood, spleen, lung, and LLN of organ donors in collaboration with LiveOnNY, the organ procurement organization for the NY metropolitan area. Cells were rested overnight and stimulated with Flu peptide megapools or control DMSO for 24 h, and resting and flu-reactive CD4⁺ and CD8⁺ T cells were sorted, as previously described (Poon et al., 2021a, 2021b; Davis-Porada et al., 2024). Sorted single-cell suspensions from six healthy subjects and six organ donors were prepared for CITE-seq using the 10x Genomics pipeline as described previously (Davis-Porada et al., 2024), and alignment, quality control, cell classification using MultiModal Classifier Hierarchy (Caron et al., 2025), and clustering and UMAP generation were performed as previously described (Davis-Porada et al., 2024). Human protein-coding genes with murine orthologs (using BioMart [Smedley et al., 2009]) were converted from AnnData to SingleCellExperiment using zellkonverter v1.10.1, and raw counts were pseudobulked for each sample (donor, tissue, and stimulation condition) with at least 25 cells using Dreamlet v1.3.1 aggretateToPseudoBulk. Differential expression of genes with at least five total counts in at least 40% of samples was performed between unstimulated T cells and flu-reactive T cells using linear mixed modeling, with donor identity and tissue source as random effects; precision weights for the model were estimated using dreamlet::processAssays. Differentially expressed genes were ranked by log₂(fold change), and GSEA (GSEAPY v1.1.3) was used to analyze the statistical enrichment of mouse secondary response profiles among the human flu-reactive T cell responses. The top 500 genes with human homologs differentially expressed between day 0 and day 3 from the entire secondary response or from the T cell activation cluster were used as gene sets for GSEA. To analyze the co-expression and mutual exclusivity of IL10 and other genes in AIM⁺ cells, we calculated the log-transformed ratio of the joint probability of detecting the two genes in the same cell to the product of their marginal detection probabilities, as previously described for transcription factors (Mizrak et al., 2019). R values were calculated within individual donors with >10 IL10-expressing cells (n = 4) for all protein-coding genes expressed in >80 cells across the entire AIM⁺ dataset. Transcripts with positive R values across all four comparisons indicate co-expression, and transcripts with negative R values across all four comparisons indicate mutual exclusivity beyond what would be expected by chance. To display this analysis, we plotted the log-normalized expression (log₂[counts*1,000/total_counts + 1]) of each gene versus R for all genes where all four comparisons resulted in an R value with the same sign. Visualization was performed using ScanPy v1.9.3 and bioinfokit v2.1.3 functions, and statistics were performed using Scipy.stats v1.11.1 in Python v3.9.12.

Statistical analyses were performed with GraphPad Prism (v.9.3.1) unless otherwise specified. Statistical tests and significance cutoffs are specified in each figure legend.

Fig. S1 shows experimental design for RNA-seq performed in Figs. 1 and 2 and analysis of the genes comprising principal component loadings in PCA of RNA-seq data in Fig. 1 C. Fig. S2 shows gating strategy for flow cytometry analysis. Table S1 lists genes from Fig. 1 C (PCA plot), Fig. 1 D (heatmap), Fig. 9 E (GSEA), and Fig. 9 L (volcano plot). Table S2 lists flow cytometry antibodies used in this study.

The mouse lung RNA-seq data presented in Figs. 1 and 2 are available from NCBI Gene Expression Omnibus (GEO) under the accession number GSE284691. The human CITE-seq data in Fig. 9 are available from GEO under the accession number GSE261278. The remaining data are available from the corresponding author upon reasonable request.

The authors wish to acknowledge the Molecular Pathology Shared Resource of the Herbert Comprehensive Cancer Center at Columbia University, supported by National Institutes of Health (NIH) grant P30 CA013696 for RNA extraction, quality assessment, histological processing, staining, and scanning. We would also like to thank Daniel Caron for advising on technical aspects of sequencing analysis, and Joanna DeBarros Martin for help with the T cell depletion experiments.

This study was supported by NIH grants AI150680 and AI168634 awarded to Donna L. Farber and AI128949 awarded to Donna L. Farber and Peter A. Sims. Julia Davis-Porada was supported by F30AI174785 and T32GM145440. Work in this study was also supported by NIH grants S10OD020056 and S10OD030282, awarded to the Columbia University Flow Cytometry Core.

Author contributions: Alexander Y. Yang: conceptualization, data curation, formal analysis, investigation, methodology, project administration, software, supervision, validation, visualization, and writing—original draft, review, and editing. Julia Davis-Porada: formal analysis, investigation, visualization, and writing—original draft, review, and editing. Daniel H. Paik: conceptualization, investigation, and writing—review and editing. Alex B. George: data curation, formal analysis, investigation, and software. Brea H. Brown: investigation and writing—review and editing. Paige L. Ruschke: investigation. Peter A. Sims: data curation, formal analysis, funding acquisition, and writing—review and editing. Ziv Frankenstein: conceptualization, data curation, formal analysis, investigation, methodology, project administration, resources, software, validation, visualization, and writing—original draft, review, and editing. Anjali Saqi: formal analysis, investigation, methodology, supervision, validation, and writing—original draft. Donna L. Farber: conceptualization, data curation, formal analysis, funding acquisition, investigation, project administration, resources, supervision, validation, visualization, and writing—original draft, review, and editing.