Deep profiling deconstructs features associated with memory CD8⁺ T cell tissue residence

SUMMARY

Tissue resident memory CD8⁺ T cells (Trm) control infections and cancer and are defined by their lack of recirculation. Because migration is difficult to assess, residence is usually inferred by putative residence-defining phenotypic and gene signature proxies. We assessed the validity and universality of residence proxies by integrating mouse parabiosis, multi-organ sampling, intravascular staining, acute and chronic infection models, dirty mice, and single-cell multi-omics. We report that memory T cells integrate a constellation of inputs— location, stimulation history, antigen persistence, and environment— resulting in myriad differentiation states. Thus, current Trm-defining methodologies have implicit limitations, and a universal residence-specific signature may not exist. However, we define genes and phenotypes that more robustly correlate with tissue residence across the broad range of conditions that we tested. This study reveals broad adaptability of T cells to diverse stimulatory and environmental inputs and provides practical recommendations for evaluating Trm cells.

Graphical Abstract

eTOC BLURB

Scott, et al., combine migration assays, myriad infection models, and multi-omics to describe the broad adaptability of resident memory T (Trm) cells to diverse stimulatory and environmental inputs. They report limitations in common proxy markers for tissue residence and provide practical recommendations for conceptualizing and identifying Trm cells.

INTRODUCTION

CD8⁺ T cells survey host tissue by cell contact, thus their location is essential to their function. CD8⁺ memory T cells are often parsed into subsets based on putative migration properties. Circulating memory T cells (Tcirc) include central memory (Tcm) cells, which recirculate between blood, secondary lymphoid organs, and lymph, and effector memory (Tem) cells which patrol blood and occasionally nonlymphoid tissues.^1–3 Tissue resident memory T cells (Trm) durably reside in various tissues throughout the body without recirculating,^4–8 and play a specialized role in protection against peripheral re-infections and cancer.^9–15

Although there is considerable interest in Trm cells, migration assays to identify these cells— including parabiosis, tissue grafting, and cell transfers¹⁶— are not always practical or even possible. Thus, studies often infer residence based on Trm-associated markers^17–21 or gene expression profiles^22–25 derived from other studies. While the list of markers and genes that may correlate with migration has grown considerably, a comprehensive evaluation of their accuracy to predict T cell residence across diverse tissues and contexts is lacking.

Here we assessed the validity and universality of residence proxy measurements by integrating mouse parabiosis, multi-organ sampling, intravascular staining, acute and chronic priming models, dirty mice,²⁶ and single-cell multi-omics. This work demonstrates that solely relying on tissue location, a single-marker approach, or migration proxies defined in a different tissue or context can result in misinterpretations. Migration assays and intratissue analyses, which account for resident and equilibrating memory populations within tissues, facilitated the identification of more reliable tissue specific markers. A bifurcated analysis of memory CD8⁺ T cells unveiled genes that more accurately correlated with residence and enabled identification of a more refined, albeit imperfect, combination of flow compatible markers to identify Trm cells in various tissues. Despite this improvement, because T cell phenotype is a product of both differentiation state and diverse environmental cues, we propose that a single gene signature or flow cytometry panel may fail to unambiguously identify all Trm cells in all biological settings.

RESULTS

Location, antigen distribution, and antigen persistence influence whether T cells are resident

Memory CD8⁺ T cells can be isolated from essentially every solid organ and barrier site of immune experienced mice. We asked how migration properties varied across organs and tissues, examining 23 anatomic compartments. As local antigen levels influence T cell differentiation and tissue residence, we utilized five distinct priming modalities that represented a spectrum of antigen distribution or duration. To model a situation in which antigen is not present, we first stimulated naïve Thy1.1⁺ or CD45.1⁺ congenically marked P14 T cells (specific for the immunodominant lymphocytic choriomeningitis virus (LCMV) epitope gp33) in vitro, followed by transfer to naïve recipients. Between 50–60 days post-transfer, mice with Thy1.1⁺ P14 cells were joined via parabiosis surgery with those having CD45.1⁺ P14 cells (Figure 1A). Following 21–28 days of rest, parabionts underwent intravascular (i.v.) staining.²⁷ Tissues were then collected to isolate and stain cells, identify host- and partner-derived cells, and calculate the proportion of resident cells, as previously described^28,29 (Figures 1B and 1C).

Figure 1: Anatomic site, antigen distribution, and antigen persistence influence tissue residency.

(A) Experimental schemes for interrogating residence and representative flow plots showing host and partner derived cells from the spleen, lung, SI-IEL, and FRT. (B) Percent of CD8⁺ T cells. (C) Percent resident. Abbreviations: BAL: bronchoalveolar lavage fluid, BL: blood, BM: bone marrow, BR: brain, d: day, FRT: female reproductive tract, IEL: intraepithelial lymphocytes, KID: kidney, I.N.: intranasal, I.P.: intraperitoneal, i.v.: intravascular (antibody staining), I.V.: intravenous, LI: large intestine, LP: lamina propria, LIV: liver, LU: lung, medLN: mediastinal lymph node, mesLN: mesenteric lymph node, PANC: pancreas, PC: peritoneal cavity, p.s.: post-surgery, p.t.: post-transfer, SG: salivary gland, SI: small intestine, SK: skin, SPL: spleen, and THY: thymus. Number of mice (m), as shown in (A) varies by tissue for some models.

To model acute systemic infections, we transferred either naïve OT-I T cells (specific for the immunodominant SIINFEKL epitope within chicken ovalbumin) followed by intravenous (i.v.) vesicular stomatitis virus expressing ovalbumin (VSVova) infection or naïve P14 T cells followed by intraperitoneal (i.p.) LCMV Armstrong (Arm) infection. To model an acute local infection, P14 T cells were transferred to naïve mice prior to intranasal (i.n.) infection with a recombinant mouse-adapted PR8 strain of influenza A virus (IAV) that expresses gp33 (PR8-gp33). Finally, to model persistent antigen, we transferred P14 cells, depleted CD4⁺ T cells, and infected with LCMV Clone 13 (Cl13) i.v., an oft-used model to study T cell exhaustion.^30–33 Congenically distinct VSVova-, LCMV Arm-, PR8-, or LCMV Cl13-immunized mice were surgically conjoined >30 days post-infection (p.i.), rested for 21–28 days, and then cells from tissues were evaluated.

We found that in vitro activation (no in vivo antigen present) was sufficient to establish Trm cells upon transfer to naïve mice, thus providing compelling evidence that local antigen is not required to initiate residence differentiation within the anatomic compartments we analyzed (Figure 1). Whether cells were primed by in vitro stimulation, or activated in vivo by acute replicating systemic infections, the proportion of Trm cells versus equilibrating cells was relatively consistent between models for a given tissue. For example, irrespective of modality, memory CD8⁺ T cells were predominantly resident (>90%) within the salivary glands and the small intestine, whereas in the liver, they were partially equilibrating (<25% resident), and in the spleen, they almost completely equilibrated (<10% resident). Although i.n. PR8, which causes a local respiratory infection, led to a broad distribution of memory CD8⁺ T cells, it also resulted in amplification of Trm cells within the fraction of lung that did not stain with i.v. antibody, the airways (bronchoalveolar lavage (BAL) fluid), and the lung-draining lymph node (medLN) (Figure 1). This supports the conclusion that local antigen, even if not required, can potentiate site-specific Trm cell differentiation. Collectively, these data support previous conclusions that Trm cells form independently of local antigen recognition,^34–40 but are also consistent with reports that local antigen recognition augments Trm cell differentiation and abundance.^41–44 Chronic infection had the most profound influence on T cell migration, resulting in few P14 cells in blood and a predominance of resident cells in all analyzed tissues (Figures 1A–C).

Taken together, resident CD8⁺ T cells are typically dominant within nonlymphoid tissues (NLT) regardless of modality. However, several variables influence the proportion of CD8⁺ T cells that are resident: tissue location, antigen distribution during the early stage of the response, and antigen persistence.

Correlation between expression of one cell surface marker and residence varies by anatomic location and environmental conditions

Trm cells are defined by residence, but migration assays are technically challenging and sometimes unfeasible. Ergo, cell surface markers are often used as proxies for migration status. We evaluated the correlation between the expression of commonly used phenotypic proxies and the proportion of T cells that were resident by a parabiosis assay. First, we focused on single markers and evaluated memory P14 cells from 18 anatomic locations of LCMV Arm immune parabionts (Figure 2A). The goal was to assess how well the percentage of cells expressing a specific marker aligned with the proportion identified as resident in parabiosis assays.

Figure 2: Correlation between cell surface marker expression and residence.

Thy1.1⁺ and CD45.1⁺ P14 LCMV Arm immune chimeric mice were joined via parabiosis surgery. Following 21–28 days of rest, parabionts underwent i.v. antibody labeling and tissues were collected to calculate the (A) percent of cells that were resident and percent of total P14 cells expressing or lacking indicated markers. n=16–32 mice (n varies by tissue). (B) Representative plots of CD103 and CD101 on P14 cells isolated from indicated tissues (C) Percent residence of CD101⁺ and CD101⁻ P14 cells (n=8 mice). (D) Representative CD49a expression (gMFI ± SD; 5 mice). (E) Representative CD103 and CD62L expression on naïve CD8⁺ T cell splenocytes. (F) Expression of host and partner congenic markers, CD69, and CD103 on LCMV Arm specific P14 cells. (G) Characteristic i.v. antibody and CD69 staining and (H) percent resident values for i.v.⁺ vs i.v. ⁻CD69⁺ P14 cells (n=16 mice from two independent experiments). (I) Illustrative CD69 and CD62L expression for bone marrow derived P14 cells. (J) LCMV Arm immune mice were challenged with VSV and (K) CD69 and CD103 expression on P14 cells were evaluated 2 days later. (L) Percent P14 cells expressing CD69 (left axis) and percent resident (right axis). n=10 mice from two independent experiments. (M) P14 cells from cohoused LCMV Arm immune parabionts were evaluated for (N) CD69 and CD103 expression (representative plot from n=10 mice from three independent experiments). (O) CD8⁺ T cells from cohoused parabionts were assessed for expression of CD44, CD62L, and CD69. (P) Frequency of mesLN CD69⁺ cells in indicated subsets (n=22 mice). Significance was determined (E, I, M) by two-way ANOVA with Sidak’s multiple comparison test. ns, not significant, **p < 0.01, ****p < 0.0001. Data indicate individual mice and mean ± SEM. See Figure 1 for abbreviations.

Individual cell surface markers could overestimate or underestimate residency. For example, CXCR6 is expressed by both Trm and non-Trm cells (Figure 2A). In contrast, CD101 expression stringently identified Trm cells, but many did not express CD101 (Figures 2A–C). Furthermore, the staining intensity of some markers, such as CD49a, varied considerably and did not yield bimodal expression patterns, making “percent positive” measurements challenging to determine (Figure 2D). Moreover, CD103 is not universally expressed by Trm cells and is expressed by recirculating naïve (CD44⁻) CD8⁺ T cells (Figures 2E–F).

Consistent with previous studies, CD69 was not expressed by all Trm cells in tissues, including some resident cells in the pancreas (PANC) and female reproductive tract (FRT) (Figure 2F).²⁸ However, in specific pathogen free (SPF) mice, CD69, unlike CD103, was expressed on many Trm cells in all tissues we examined (Figures 2A and 2F). Indeed, CD69 expression identified Trm cells in both the parenchyma and within the vascular contiguous compartments of the liver and spleen (Figures 2G and 2H). However, this correlation was not absolute; for instance, numerous memory P14 cells expressed CD69 even if they had recently migrated (partner-derived) to the bone marrow (Figure 2I). Moreover, T cells elevate CD69 expression upon activation through the T cell receptor (TCR) or by bystander infection cues such as interferons and cytokines.^45–48 Infecting LCMV Arm-immune mice with heterologous VSV led to increased CD69 expression on non-cross-reactive memory P14 cells in blood and secondary lymphoid organs (SLOs), despite the expectation that they would not be durably resident (Figures 2J–L).

Both CD69 and CD103 expression were impacted by host environment (Figures 2M–N). Fewer LCMV Arm-specific P14 Trm cells in the small intestine (SI) expressed CD103 with increased microbial experience due to cohousing with pet shop mice (Figures 2F and 2N). Furthermore, in cohoused mice, CD69 was expressed by many recent P14 cell migrants to tissues (Figure 2N). Endogenous CD69⁺ CD8⁺ T cells also equilibrated between cohoused parabionts, including those with a naïve phenotype (Figures 2O and 2P). Naïve phenotype CD69⁺ T cells are observed in human SLOs.^49–51 These observations collectively raise the possibility that non-resident T cells may express CD69, particularly outside of SPF environments.

Resting Trm cells are transcriptionally distinct

We sought to identify alternative markers that would distinguish resting Trm cells from other CD69⁺ and CD69⁻ CD8⁺ T cells. To this end, we performed bulk RNA-sequencing on 1) resting CD69⁺ Trm cells (presumed to be resident based on previous parabiosis assays), CD69⁻ Tcm cells, and CD69⁻ Tem cells established by LCMV Arm, 2) recently stimulated CD69⁺ P14 cells (sorted from LNs 4 days after LCMV Arm infection), 3) CD69⁺ bystander activated memory cells (Arm-immune P14 cells sorted from LNs 48 hours p.i. VSV), and 4) endogenous CD69⁺ H-2D^b restricted gp33 tetramer⁺ CD8⁺ cells sorted from SLOs 50 days after chronic LCMV Cl13 infection of CD4⁺ T cell depleted mice.

To understand the transcriptional relatedness between these populations, we first performed unsupervised multidimensional scaling of their gene expression profiles. Resting Trm populations were distinct from Tcirc populations (Figure 3A). However, SLO-derived Cl13-specific CD8⁺ T cells, which are predominantly resident via migration assays (Figure 1), were not closely aligned with any LCMV Arm-generated Trm populations (Figure 3A).

Figure 3. Resting CD69⁺ Trm cells are transcriptionally distinct.

RNA sequencing analysis of 1) resting Trm and Tcirc LCMV Arm-specific P14 cells, 2) d4 p.i. LCMV Arm P14 cells, 3) bystander activated P14 cells, and 4) LCMV Cl13 specific H-2D^b restricted gp33 tetramer⁺ cells. (A) Two-dimensional scatterplot of indicated populations. (B) Top: Venn diagram with numbers of unique and overlapping DEG. Bottom: Expression heatmap of DEG common between Trm populations and Cl13 responding CD8⁺ T cells, in comparison to Tcirc populations. (C) Volcano plot of DE analysis results. (D) Results of enrichment analysis. (E) Volcano plot of DE analysis results. See Table S1 for DEG lists.

Relative to Tcirc populations, there were only 114 differentially expressed genes (DEG) shared between LCMV Cl13- and Arm-generated resident populations, including known residence and exhaustion associated genes, such as S1pr1 and Tox, respectively (Figure 3B; Table S1). Thus, although resident CD69⁺ Cl13-responding populations may adapt some features associated with resting CD69⁺ Trm cells, their gene expression profiles shared much in common with recently stimulated CD8⁺ T cell populations, namely d4 p.i. or bystander activated cells. Indeed, increased expression of proliferation-associated genes was a common feature of chronically, recently, and bystander activated CD69⁺ CD8⁺ T cells (Figures 3C and 3D).

Finally, we grouped all resting memory Trm populations— regardless of tissue of origin— and compared them to Tcirc and recently stimulated populations. Genes elevated in resting Trm populations included Cd38, P2rx7, Cd160, and Itga1. Genes reduced in Trm populations included Klf2, Klrg1, Il18r1, and Sell (Figure 3E; Table S1). Itgae (encoding CD103) and Cd101 were also increased in resting Trm cells, although expression was predominantly confined to a subset of Trm cells (Figure 2) in this analysis. Furthermore, Cxcr6, which is expressed by many Trm cells but, unlike CD69, is not induced on memory CD8⁺ T cells by inflammation,⁵² was not differentially expressed. These findings highlight the need for higher-resolution single-cell analyses of memory CD8⁺ T cells across a broader range of tissues and contexts to pinpoint potential universal features unique to Trm cells.

In summary, even though Cl13-specific CD8⁺ T cells are predominantly resident, they share a limited transcriptional profile with resting Trm cells. Moreover, while some genes were unique to resting resident populations, they may not necessarily represent universal residency features.

High dimensional analysis better discriminates migration properties

Thus far, our findings collectively demonstrate that 1) residence can exist across a variety of contexts (including chronic infection), 2) phenotype varies by location, and 3) recent stimulation and ephemeral environmental perturbations can transiently induce expression of markers often used to infer residence. These observations, and the myriad parameters that influence T cell differentiation, exemplify the challenges associated with attempts to identify unequivocal residence-defining molecular readouts, including cell surface markers. To address these challenges, we focused on a single viral infection model in SPF mice where we tested how the expression of various combinations of cell surface markers correlated with residence across diverse anatomic locations.

To this end, 70 days after LCMV Arm infection, congenically distinct P14 immune chimeras underwent parabiosis for 40 days (Figure 4A). After in vivo intravascular labeling, lymphocytes were isolated from tissues and co-stained with antibodies for putative Trm cell markers as well as markers that defined cell lineage, parabiont origin, and identified intra- and extravascular populations. Host and partner derived P14 cells from all mice and tissues were merged for the analysis (Figure 4B). The PhenoGraph algorithm⁵³ identified 16 distinct subpopulations based on the expression of 12 cell surface markers (Figure 4C–H).

Figure 4: High dimensional flow cytometry identifies Trm cells.

(A) Experimental scheme. (B) Data cleaning and gating was performed prior to merging P14 cells across mice (n=8 mice, one of four independent experiments) and tissues. (C) PhenoGraph defined clusters. (D) Host and partner derived cells for each tissue. Percent (E) parabiont and (F) tissue origin for each cluster. (G) Percentage of cells from each tissue in the Trm cell-dominated clusters (1–8, as shown in (E)). (H) Expression histograms for clustering input markers. (I) Characteristic expression plots and (J) percent host and partner for indicated subsets for bone marrow derived P14 cells (n=14 mice, three independent experiments). (K) Typical expression plot for salivary gland (SG) derived P14 cells. Significance was determined by two-way ANOVA with Sidak’s multiple comparison test. **p < 0.01, ****p < 0.0001. Data indicate individual mice and mean ± SEM. See Figure 1 for abbreviations.

We next established the tissue and parabiont origin of each subpopulation. PhenoGraph definitively separated some tissue resident (clusters 1–6; Figures 4D and 4E) and equilibrating (approximately equal number of host- and partner-derived cells) populations (clusters 9–16; Figures 4D and 4E). Some Trm cell phenotypes were tissue-specific, particularly within the salivary glands, whereas some (e.g., cluster 6; CD69⁺P2RX7⁺CD49a⁺CXCR6⁺CD62L⁻) were common to many tissues, including the spleen (Figures 4D–F). However, Trm cells in a single tissue can exhibit extensive heterogeneity (Figure 4G). Moreover, there were clusters (for instance, 7 and 8) that included both resident and equilibrating cells. Thus, in the context of LCMV Arm infection in SPF mice, these 12 markers were helpful but not entirely sufficient to discriminate durable Trm cells from partner-derived cells that arrived in the tissue during the parabiosis equilibration phase (referred to as “newcomers”). Nevertheless, additional markers are better than CD69 alone. For example, P2RX7 and CD62L further resolved resident from equilibrating CD69⁺ cells in the bone marrow (Figures 4I and 4J). Yet, P2RX7 was not expressed on many Trm cells in the salivary gland (Figure 4K).

Taken together, multi-dimensional flow cytometric analyses improved the ability to distinguish Trm from equilibrating memory CD8⁺ T cells. The CD69⁺P2RX7⁺CD49a⁺CXCR6⁺CD62L⁻ phenotype was common to Trm cells isolated from many diverse tissues including the spleen. However, extensive phenotypic heterogeneity exists both across and within tissues, and newcomer cells may also express Trm cell-associated markers. Thus, unequivocal identification of Trm cells remained challenging without migration assays.

Infection, route, and priming conditions influence Trm cell phenotypes

We next varied the transgenic T cell specificity, viral infections or priming conditions, and infection routes. We first more thoroughly examined the phenotypes of memory OT-I CD8⁺ T cells after in vivo intravenous VSVova infection (Figure S1A). Based on the same 12 cell surface markers used to identify LCMV Arm-specific clusters in Figure 4, the PhenoGraph algorithm reduced memory OT-I cells from 14 tissues and blood into 17 subpopulations (Figure S1B). Clustering results were similar to i.p. LCMV Arm infection, with comparable tissue specific Trm cell phenotypes. For example, many salivary gland Trm cells lacked expression of P2RX7 (cluster 2; Figures S1C–F). However, once again, definitively identifying Trm cells remained challenging because newcomer cells could adopt a Trm cell-like phenotype. For example, partner derived cells in the skin expressed CD69 and other Trm cell associated markers (Figure S1G).

We also transferred in vitro activated P14 cells to naïve mice to test differentiation in the absence of local pathogen replication or antigen. At day 34 post-transfer, P14 cells were isolated from various tissues, with CD69 and CD103 expression comparable to LCMV Arm specific memory populations (Figure S2A). PhenoGraph analysis of in vitro-activated P14 cells from 16 tissues, performed 83 days post-transfer to naïve mice, indicated once more that cell phenotypes were primarily associated with anatomic site and migration property (data not shown).

We next considered whether local infection would impact memory CD8⁺ T cell phenotypes. Thus, we examined the phenotypes of memory P14 cells established by intranasal PR8-gp33 (PR8) infection (Figure S2B). As observed with memory populations established after acute systemic infections, some Trm cell phenotypes were tissue-specific (Figures S2C–G). Trm cells in the airways (BAL) expressed CD69, yet unlike lung Trm cells, they lacked expression of CXCR6⁵⁴ and CD38 (Figures S2C–G). Moreover, infection history impacted the phenotypes of memory CD8⁺ T cells in the lungs⁵⁵ and other tissues. For instance, PR8-specific, but not LCMV Arm-specific, Trm cells in the FRT expressed CD160 (Figures S2C–G versus Figure 4).

To directly investigate the impact of local versus systemic viral infection on memory CD8⁺ T cell phenotypes, we paired immune mice from i.n. PR8 and i.p. LCMV Arm infections and allowed the parabionts to rest for 29 days before phenotyping analyses (Figure 5A). PhenoGraph clustering of P14 cells from 12 diverse anatomic locations, both extra-and intravascular, revealed 14 subpopulations (Figure 5A–F).

Figure 5: Infection history influences Trm cell phenotypes.

(A) Experimental scheme. For each infection, P14 cells derived from both host and partner in a specific tissue were concatenated across replicates (n=4 mice, one of three independent experiments). Then, P14 cells for each infection and from all tissues were concatenated. PhenoGraph was used to define clusters. (B) Host and partner derived cells for each tissue and infection. Percent (C) tissue, (D) parabiont, and (E) infection history from the most (top) to least resident (bottom) cluster. (F) Expression histograms for clustering input markers. (G) Representative expression histograms for indicated markers. (H) Percent of CD69⁺ P14 cells expressing indicated markers (n=22 mice, three independent experiments). Statistical significance was determined by two-way ANOVA with Sidak’s multiple comparison test. ns, not significant, ***p < 0.001, ****p < 0.0001. Data indicate individual mice and mean ± SEM. See Figure 1 for abbreviations. See also Figure S2.

While some Trm cell phenotypes were tissue specific, others (clusters 5 and 6) were common to many tissues (Figures 5B–F). Several Trm cell phenotypes were largely infection-specific, with cluster 2 (CD69⁺CD103⁺P2RX7⁻) predominantly associated with PR8 and cluster 4 (CD69⁺CD103⁻P2RX7⁺) mainly derived from LCMV Arm (Figures 5B–F). Although CD69 expression was a consistent feature of resident cells (Figures 5D and 5F), equilibrating cells in cluster 11 were CD69⁺. The expression of CD103 and P2RX7 differed in response to local PR8 compared to systemic LCMV Arm infection (Figures 5G and 5H). However, the expression of CD49a coupled with CD69 remained consistent among Trm cells across infections (Figure 5H).

Next, to test how phenotype correlated with migration and anatomic site among chronically stimulated CD8⁺ T cells, we utilized chronic LCMV Cl13 infection (Figure S3A). PhenoGraph clustering of chronically stimulated P14 cells from 14 tissues and blood identified a single equilibrating cell cluster lacking CD69 expression (Figures S3B–F). Unlike acute infection, regardless of the marker combinations employed, clustering by phenotype was not predominantly driven by anatomic site (Figures S3E and S3F). One exception was that Ly108⁺CD101⁻ resident CD8⁺ T cells (corresponding to progenitor exhausted CD8⁺ T cells, Tpex^56,57) were enriched in spleen and thymus and scarcely distributed among other tissues (LN was not included in this analysis). Chronically stimulated P14 cells from a single anatomic site, for instance, the large intestine, exhibited phenotypic heterogeneity, which was unrelated to mouse source or migration property (Figures S3G and S3H). Furthermore, in contrast to acute infection, many chronically stimulated P14 cells expressed CD101 and PD1 and exhibited diminished expression of other resident-associated markers including CD103, P2RX7, and CD49a (Figures S3I and S3J versus Figures 2A). For example, very few chronically stimulated CD8⁺ T cells resident within the bone marrow expressed P2RX7, while many expressed CD101 (Figure S3K versus Figures 4). Thus, antigen persistence pronouncedly impacts tissue resident CD8⁺ T cell phenotypes, further adding to the complexity of resolving marker-based migration properties.

Anatomic location, infection history, and migration property influence CD8⁺ T cell transcriptomes

Thus far, we have shown that cell surface marker expression by memory CD8⁺ T cells is influenced by many variables, including location, infection history, migration properties, and environment. This could expose limitations in interpreting previous transcriptome studies attempting to define signatures of Trm cells without incorporating migration assays, but instead relying on anatomic location or proxy markers derived from different contexts (other species, priming modalities, tissues, etc.). To address this gap and define features unique to Trm cells, we first established parabionts containing either LCMV Arm- or PR8-induced memory P14 cells (Figure 6A). We then performed in vivo intravascular staining, isolated lymphocytes from several tissues, and leveraged Cellular Indexing of Transcriptomes and Epitopes sequencing (CITE-seq).

Figure 6: Experimentally discerning tissue- from migration-associated genes.

(A) Experimental approach. UMAP plot of the scVI (RNA only) latent space, showing (B) clusters, (C) tissues, (D) vascular localization, and (E) and parabiont origin. (F) Pdcd1 and Cxcr6 gene expression in lung Trm cells. (G) Infection history for each cell. (H) Enrichment scores for genes that were previously determined to have increased expression (‘up’) or decreased expression (‘down’) in putative Trm cells were identified by comparing cells of different phenotypes from different tissues. CD69⁺ vs. CD69⁻ genes were derived from the analysis performed in Figure 3E (Table S1) and the CD103⁺ vs. CD103⁻ genes were derived from a previous study.²¹⁶³ (I) PR8-generated Trm and Teq cells were compared by a scVI differential expression test to identify genes more abundantly expressed by Trm cells. (J) Enrichment scores for genes identified in (I). (K) Two DE tests. (L) DEG from two DE tests (K) were compared to previously published mouse⁶³ and human⁷⁶ Trm signatures. (M) Relative expression of genes. (N) Tissue origin of human CD3⁺ T cells⁷⁶ and the relative expression of lung-associated genes. Abbreviations: d: day, DE: differential expression, DEG: differentially expressed genes, i.v.: intravascular, LU: lung, SPL: spleen, and Teq: tissue equilibrating. See also Figure S4A.

For each infection, we first jointly analyzed the RNA and surface protein expression for each cell using Total Variational Inference (totalVI)⁵⁸ (Figure S4A). Expression of the congenic markers and the i.v. antibody was used to determine the parabiont origin and the i.v. antibody status for each cell, respectively, on a per-tissue basis. We then focused our analysis on P14 cells isolated from the lungs and spleen, as previous studies^20,21,59 defined signatures of Trm cells by comparing memory populations exhibiting distinct phenotypes from these tissues.

To evaluate how anatomic location, infection history, and migration property influenced the transcriptomes of memory CD8⁺ T cells derived from the lungs and spleen, we used single-cell VI (scVI)⁶⁰ to perform dimension reduction on the RNA only and identify Leiden⁶¹ cell clusters (Figure S1A). The Uniform Manifold Approximation and Projection (UMAP) representation of the Leiden clusters stratified memory CD8⁺ T cells primarily according to tissue of origin (Figures 6B and 6C). Yet, cells from each tissue were partitioned into several distinct clusters.

While one lung-specific cluster (cluster 3) was almost exclusively extravascular, there were no discernable differences in i.v. antibody staining observed with the spleen-specific clusters (Figure 6D). We next assessed parabiont origin and infection history. Clusters 3, 6, and 9 were Trm cells, as evidenced by the paucity of partner-derived cells (Figure 6E). The remaining clusters were tissue equilibrating (Teq) cells as they showed a mix of host- and partner-derived cells. However, a few partner-derived cells were evident in the Trm cell clusters, and some lung-derived newcomers expressed some genes such as Pdcd1 and Cxcr6 that have previously been associated with Trm cells^20,23,54 (Figure 6F). Trm cells were separated by tissue, with spleen-derived Trm cells predominantly found in cluster 6 and lung-derived Trm cells in cluster 3 (Figures 6C–E). Teq cells also largely segregated into transcriptionally distinct clusters based on tissue origin (Figures 6C–E; Table S2).

Furthermore, within each tissue, gene expression differences were evident after PR8 or LCMV Arm infection (Table S3). Collectively, aside from migration properties, various factors influence the gene expression profiles of memory CD8⁺ T cells, including vascular localization, infection history, and the organ of isolation.

Discriminating tissue- from migration-associated genes

Endeavors to define core or universally applicable gene signatures of Trm cells have often relied on tissue location,⁵⁹ sometimes with CD69 expression²⁰ or intravascular antibody staining,⁶² as proxies for migration assays. We sought to distinguish transcriptional features that simply correlated with tissue location rather than migration property. First, to evaluate the reliability of signatures identified through phenotypic proxies, we computed enrichment scores with the resting CD69⁺ Trm cell genes defined in Figure 3H and a reported Trm cell signature derived by contrasting CD103⁺ lung-, skin-, and SI-derived CD8⁺ memory T cells (putative Trm cells) to CD103⁻ CD8⁺ memory T cell populations from the spleen (putative non-Trm cells).^21,63 Genes that were increased in Trm cells based on these previous analyses were enriched in all P14 cells isolated from lungs, irrespective of migration property, including Teq cells (clusters 5 and 2; Figure 3H). However, these genes were not enriched in spleen Trm cells (cluster 6; Figure 3H).

We next tested if comparing all migration assay-defined Trm cells to Teq cells would yield an inclusive and reliable set of unambiguous migration-associated markers (Figure 6I). Unfortunately, we found that inferring migration based on differentially expressed genes identified when memory T cells are categorized into only two classes, resident and equilibrating, posed a challenge (Figure 6J). These genes were enriched in all lung-derived cells, including Teq cells, but were not enriched in spleen Trm cells. Thus, the challenge of identifying a universal Trm cell signature likely arises due to the marked influence of tissue of origin on gene expression.

To directly evaluate if previous Trm cell signatures were “contaminated” with tissue-associated, non-migration-related genes, we performed two separate differential expression tests (Figure 6K). First, spleen and lung Teq cells were compared (inter-tissue test) to identify genes associated with tissue location. Second, lung Trm and Teq cells were compared (intra-tissue test) to determine genes associated with being resident in the lungs (Table S4). Next, we examined the intersections of the genes identified from these tests with previously reported mouse^21,63 and human lung⁵⁹ presumed signatures of Trm cells (Figure 6L; Table S4). Numerous previously reported Trm cell associated genes, including Vps37b, Bhlhe40, and Dusp5 were not differentially expressed between lung Trm and Teq cells.^20,21,64,65 However, this approach validated some Trm cell associated genes, including Itgae, Rgs1, Xcl1, and low Klf2 and S1pr1 expression (Figures 6L and 6M; Table S4).

To answer whether the mouse lung-associated genes similarly differentiated between human lung and spleen T cells, we analyzed published single-cell transcriptome profiles of human CD3⁺ T cells.⁵⁹ Genes associated with being a cell isolated from the lungs (irrespective of migration) in mouse were recapitulated in cells from human lungs (Figure 6N).

In summary, memory T cells express some genes related to their location and others associated with migration. Using tissue of origin as a migration proxy to generate Trm cell gene signatures is thus problematic.

Two transcriptionally distinct clades of Trm cells

The primary, albeit inconvenient, take-home message drawn from our study is that T cell biology is influenced by a multitude of variables. Therefore, any attempt to identify core Trm cell genes based on comparisons limited to a few tissues, a single infection model, or lacking migration information is likely to encompass attributes that do not consistently correlate with residency. Residence may be a characteristic of various T cell differentiation states or ontogenies. Nonetheless, we conducted a more thorough investigation to determine approaches that could yield more reliable Trm-associated genes, even if they remained imperfect.

First, we expanded the CITE-seq analysis to LCMV Arm and PR8-specific P14 cells isolated from the blood and 9 tissues and performed dimensionality reduction, clustering, and differential expression analyses based on the RNA only (Figure S1A). The UMAP representation of the Leiden clusters stratified memory CD8⁺ T cells primarily according to their tissue of origin and migration property (Figures S4B–J). Clusters 3, 8, and 9 were Trm cells and the remaining clusters were Teq cells (Figures S4H–J). Consistent with Figures 2G and 2H, i.v. antibody positive cells isolated from the liver and spleen were resident (cluster 8; Figures S4L–M).

The findings in Figure 6L highlighted the value of intra-tissue testing to define migration-associated genes. Accordingly, migration-associated gene signatures were derived from the results of differential expression tests comparing Trm and Teq cells within each tissue (Figure S5A; Table S5). Enrichment scores of DEGs, particularly those increased in Trm cells, correlated with residence within the tissue in which they were defined, but not necessarily in other tissues (Figure S5B). For example, genes that were statistically increased by SI-LP Trm cells (cluster 9) were highly enriched in lung Teq cells in cluster 2 (Figures S5A and S8B). To evaluate gene co-occurrences among Trm cells from diverse tissues, we calculated Jaccard similarities of the tissue-specific Trm cell gene signatures (Figure S5C; Table S6). Because Trm cells were separated into three distinct clusters (clusters 3, 8, and 9; Figures S4B–J) we also assessed transcriptional relatedness among the clusters by computing Pearson correlation coefficients (Figure S5D).

These analyses unveiled two major transcriptional clades in which Trm cells were dispersed, designated as “Icos^hi” and “Icos^lo” based on the observed differential expression of Icos, (Figures S5C–H) which is reportedly higher in resident versus circulating memory CD8⁺ T cells^20,21,64 and required for the generation of CD8⁺ Trm cells in the SI, salivary glands, and kidneys.⁶⁶ Although Itgae has historically been used to sub-classify Trm cells, it was not differentially expressed between the two clades (Table S5). Hypoxia response and adipogenesis genes were differentially enriched among the clades suggesting that some of the transcriptional diversity among Trm cells may reflect differences in local physiology (Figures S5I and S5J).

Bifurcating Trm cells better resolves migration-associated genes

We then explored whether a combined analysis of cell surface protein (Ly6C, CD62L, CD103, CD69, CD127, KLRG1, and the i.v. antibody) and RNA expression would better unveil unique features shared by all Trm cells in the dataset (Figure S1A). After performing dimensionality reduction and clustering based on the joint totalVI latent space representation of the data, and considering tissue origin, migration status, and transcriptional relatedness between the Leiden clusters⁶¹, we observed once again that Trm cells were largely dispersed into two clades, exhibiting differential expression of Icos but not CD103 protein (Figures 7A–D; Table S6). Hence, to mitigate the overriding influence of tissue origin and uncover features unique to the property of residence, we segregated the memory CD8⁺ T cells according to their clade designations (Figures 7B and 7E).

Figure 7: Features and strategies to more accurately discern Trm cells.

(A) UMAP plots of the totalVI latent space (both RNA and protein) from CITE-seq data of LCMV Arm and PR8-specific memory CD8⁺ T cells. See Figure S4A for analysis details. (B) Pearson correlation coefficients and hierarchical clustering of cell clusters. totalVI expression for indicated (C) genes and (D) proteins. (E) UMAP plots depicting clade designations as shown in (A-B) and further supported by the analyses in Figure S5. (F) DE proteins and genes were ranked by a gini importance score obtained from generating a random forest classifier model for each clade. The top 32 features with increased expression (‘up’) or decreased expression (‘down’) in Icos^hi and Icos^lo Trm cells are shown. (G) Analysis scheme. Signed pan-Trm gene signature scores were used as input for logistic regression and probability values from the logistic regression served as predictor scores for generating ROC curves. (H) UMAP plots of enrichment scores for signed signature genes identified in Figure 3E (CD69 proxy) and the pan-Trm cell signatures from the joint (protein and RNA) and RNA only analyses. (I) ROC curve plots of the true positive rate (TPR) against the false positive rate (FPR). (J) P14 cells from mice d4 after LCMV Arm infection and naïve CD8⁺ T cells from uninfected mice were evaluated for expression of the indicated markers (one representative mouse from three independent experiments, n=14 mice). (K) Memory CD8⁺ T cells from cohoused (CoH) parabionts were evaluated for the indicated markers (n=22 mice). (L) Gating strategy to identify Trm cells in various tissues. (M) Flow phenotyping strategy summary for evaluated tissues. (N) Cells with phenotype shown in (M) were identified and percent resident was calculated and compared to the percent resident for CD8⁺CD44^hiCD69⁺ T cells in the same tissue. n=10 mice from 2 separate experiments. Statistical significance was determined by two-way ANOVA with Sidak’s multiple comparison test. **p < 0.01, ****p < 0.0001. Data indicate (K) individual mice and mean ± SEM or (M) mean ± SEM. See Figure 1 for abbreviations. See also Figures S4–S6.

For each clade, random forest classification gini importance scores were utilized to rank differentially expressed proteins and genes based on how well they predicted if a cell was resident (Figures 7F; Table S6). Despite the differences between the clades, both clades shared 18 increased and 6 decreased Trm cell predictive genes, collectively referred to as “pan-Trm genes” (Table S6). The predictive accuracy of CD69 protein expression (but not the Cd69 gene) for determining cell residency varied across tissues (Figure 7F; Table S6). CD69 expression exhibited lower predictive accuracy for Icos^lo compared to Icos^hi Trm cells. Furthermore, the absence of i.v. antibody staining was highly predictive for Icos^hi Trm (including lung derived), but not for Icos^lo Trm cells (including liver and spleen derived) (Figure 7F).

Since gene expression signatures of Trm cells have not historically been identified through combined protein and RNA assays, we wished to compare the predictive genes from our joint analysis with those identified solely through an RNA analysis. Therefore, we employed scVI tools to generate a latent space and perform dimensionality reduction based exclusively on the RNA data for each clade. We then performed differential expression testing and random forest classification to identify highly predictive Trm cell-associated genes. Despite transcriptional heterogeneity within and between clades (Figures S6A–D), both clades shared 23 genes with increased expression and 15 with decreased expression that were predictive of Trm cells (Figures S6E–G). For both clades, reduced Klf2 expression had the highest random forest feature importance value when RNA alone was relied upon to predict cell residency (Figures S6E and S6F; Table S7).

While distinct predictive genes were identified from the joint (Figure 7; Table S6) and RNA only (Figure S6; Table S7) latent space representations of the data, there were shared genes exhibiting predictive accuracy for identifying both Icos^hi and Icos^lo Trm cells. Specifically, both analyses revealed that increased expression of Rgs1, Chn2, Xc11, Gpr34, Igflr1, Itga1, Ckb, P2rx7, Cxcr6, Fgl2, along with lower Klf2, S1pr1, Sell, Aff3, and Sidt1 were highly predictive of a cell being resident. The pan-Trm genes identified through both analyses demonstrated better performance in classifying a cell as either resident or equilibrating compared to using genes identified in Figure 3, which relied on CD69 protein expression as a proxy for residence (Figures 7G–I).

When we applied the Trm cell signature identified by the RNA only analysis (Figure S6G) to published human CD4⁺ and CD8⁺ T cells,⁵⁹ those originating from the jejunum and skin were predominately identified as being resident (Figure S7A). A fraction of cells from the bone marrow, lymph nodes, lung, and spleen also were enriched for Trm, but not Teq cell genes (Figure S7B). Inferring migration properties is improved by considering both genes with increased and decreased expressed in Trm cells. For example, many cells derived from the lungs exhibited enrichment of Trm cell genes. However, a fraction of these cells also expressed characteristic Teq cell genes such as KLF2, S1PR1, KLRG1, and CX3CR1 (cluster 4; Figure 6M and Figure S4). Moreover, these findings revealed transcriptional heterogeneity among Trm cells isolated from the lymph nodes, particularly the mesLN, and the spleen, with some expressing CCR9 and CD160, while others did not (Figures S7B and S7C). Also, these results suggested that previous claims of central memory (CD62L⁺CCR7⁺) CD8⁺ T cells inferred to be resident (based on CD69 expression) were likely Teq cells (Figure S7B and S7C; see LNs ‘up’ gene scores vs. SELL and CCR7 expression) which may invite reinterpretation of claims suggesting that naïve CD8⁺ T cells become resident in human LNs.^49–51

We observed that about 20% of blood cells exhibited enrichment for Trm cell genes alongside Teq cell genes including KLF2 and KLRG1 (Figures S7D and S7E); as Trm cells can sometimes forego residence and re-enter circulation,^67–70 this could explain why some memory CD8⁺ T cells in the vasculature may express residency-associated genes (Figure 6).

Altogether, by utilizing migration assays and categorizing memory CD8⁺ T cells into two clades, we identified features unique to Trm cells across diverse tissues. Although CD69 protein expression is not equally effective in determining residence across all tissues, combining its expression with all CITE-seq features in our analysis enabled us to identify genes that more accurately correlate with residence compared to genes identified when CD69 was used as a proxy for residence. We also demonstrate the application of these genes beyond mouse models by evaluating their expression in a subset of memory CD3⁺ human T cells from various tissues.

Flow cytometry strategy to discern residence

We next tested whether we could apply our findings to develop a flow cytometry strategy that better correlates with residence than CD69 alone. Thus far, we observed that many Trm cells express CXCR6, CD38, and CD49a proteins and genes. While CXCR6 and CD38 are highly expressed on recently activated CD8⁺ T cells, including those in the spleen and blood, CD49a is not, suggesting it may be a more reliable marker for identifying Trm cells in non-SPF contexts (Figure 7J).

To explore potentially more reliable flow strategies for identifying Trm cells, we utilized co-housed mouse parabionts as our experimental (non-SPF) model to evaluate the migration property and phenotypes of diverse CD8⁺ T cells. Our analyses of CD8⁺ T cells from co-housed mouse parabionts were intentionally agnostic of antigen-specificity and intravascular antibody staining status to better relate to scenarios where tetramer and intravascular staining are not performed (e.g., many human studies).

We first evaluated the expression of the pan-Trm cell markers (Figures 7A–F and S6G), with available flow-compatible antibodies. With individual markers we observed varying degrees of equilibration in tissues (Figure 7K). Thus, we next utilized multi-dimensional flow cytometry to examine pan-Trm cell surface and intracellular marker combinations within and across 14 diverse anatomic locations. Through this work, we propose a cell surface marker combination and analysis strategy that identifies Trm cells across the tissues we evaluated. This strategy entails identifying memory (CD44^high) CD8⁺ T cells that lack CX₃CR1 and CD62L and express CD69 (Figures 7L and 7M). To reduce conflation with recently activated CD8⁺ T cells, CD49a or CD103 were also applied, even though these markers varied based on location (Figure 7M). The CD62L⁻CX₃CR1⁻CD69⁺CD49a⁺ phenotype outperformed CD69 expression alone in identifying CD8⁺ Trm cells in SLOs, lungs, and bone marrow (Figure 7N). While this panel increased the accuracy in excluding Teq cells, it also led to the exclusion of some migration-assay-defined Trm cells in NLTs such as the pancreas, FRT, and stomach where some bona fide Trm cells do not express CD69 (Figures 2F and 7L).

In summary, a multi-parameter flow panel improves the identification of Trm cells across a range of conditions, however it did not fully substitute for migration assays.

DISCUSSION

This study integrated migration assays, intravascular staining, and single-cell intra-tissue analyses to identify migration associated genes within diverse anatomic locations. While the field has pursued unequivocal proxies for T cell residence, this study illustrates that residency exists across various contexts and that markers and techniques for assessing residency in one context may not work in another. Cells equilibrating within the vasculature can also adopt tissue-associated gene expression patterns. Although it is uncertain whether this phenomenon is a result of tissue isolation methods, it nevertheless poses a practical impediment (perhaps addressable with spatial transcriptomics). Even CD69, found on most tissue resident CD8⁺ T cells (as measured by protein expression, gene expression is less informative), can also be induced by bystander cues such as cytokines or expressed by recent migrants to tissues.

Nevertheless, we identified an improved strategy that was applicable to the experimental conditions we tested. Intra-tissue analyses to define migration-associated genes were helpful in tissues where non-Trm cells are prevalent (e.g., lungs) and for removing “contaminating” genes associated with anatomic location. Previous work documented genes impacted by enzymatic tissue digestion in CD8⁺ T⁶² and neuronal⁷¹ cells. By employing intra-tissue analyses in this study, we exclude digestion-associated genes from our Trm cell signatures and genes that are elevated in specific tissues irrespective of migration status. In our RNA only analyses, decreased expression of the transcription factor Klf2 (which promotes S1p1 expression) was the most predictive feature for T cell migration property. Unfortunately, a practical and accessible method to evaluate KLF2 protein expression is currently unavailable. Even if reagents were available for assessment of KLF2 or S1PR1 protein expression, they may not distinguish between CD69⁺ Trm cells and CD69⁺ non-Trm cells that were recently stimulated via the TCR activation or cytokines.⁷² In our study, the most universal phenotype for identifying CD8⁺ Trm cells with commonly available antibodies is CX₃CR1⁻CD62L⁻CD69⁺ and CD49a⁺ (SLOs, bone marrow, and non-gastrointestinal NLTs) or CD103⁺ (gastrointestinal tissues). This phenotyping strategy outperformed CD69 alone in multiple tissues of co-housed mice with diverse microbial experiences.

There are caveats to this strategy for identifying resident cells. CX₃CR1 may be cleaved and therefore reduced by enzymatic digestion.⁷¹ Expression of KLRG1, which correlates with CX₃CR1 expression and emerged as a top predictive marker for Icos^hi Trm cells, may be useful in scenarios requiring enzymatic digestion, though this correlation could vary depending on the infection model.³ Some CD8⁺ Trm cells do not express CD69,²⁸ CD49a,^73,74 or CD103 so this strategy may underestimate Trm cells. Also, CD49a and CD103 are often absent on chronically stimulated tissue resident cells.

While i.v antibody staining is a method sometimes used to assess residency, our data indicate that in the liver and spleen, the percentage of resident CD8⁺ T cells is comparable between i.v.⁺ and i.v.⁻ fractions. Moreover, substantive recirculating populations, naïve and central memory CD8⁺ T cells in LNs and the white pulp of the spleen for example, do not stain with i.v. antibody. Thus, i.v staining is not a faithful proxy for inferring residency, but rather informs anatomic location. Tissue of origin as a migration proxy is also problematic; both SLOs and nonlymphoid tissues can contain Tcirc and resident CD8⁺ T cells. We speculate that persistent TCR stimulation in settings of chronic antigen exposure may promote sustained transcriptional repression of Klf2 and S1pr1, thus impairing T cell egress from tissues and resulting in residence through a distinct developmental pathway compared to residence following acute infections.^72,75

In conclusion, CD8⁺ T cell phenotype reflects not only migration properties, but past and present input signals including priming history, location, environmental perturbations, and persistence of antigen. While our study provides expanded criteria that may inform more accurate identification of durably resident CD8⁺ T cells and be of immediate practical utility, it also reveals the potentially insurmountable challenge of identifying universally applicable resident-exclusive markers.

Limitations of the Study

We identified a multi-parameter strategy that better, but still imperfectly, identified resident CD8⁺ T cells. However, this strategy was not evaluated in a vast array of conditions (e.g. autoimmunity, cancer, bacterial or parasitic infections, graft rejection, etc.). Perhaps an improved strategy will be identified in the future. That said, identifying a “universal” and precise ‘Trm’ cell signature could remain elusive because residence is a property that can be shared by diverse lymphocytes (functional, exhausted, regulatory, Th1, Th2, chronically activated, etc.), that exhibit distinct functional specializations and varied stimulation histories, ontogenies, anatomic distributions, and environmental exposures. The property of residence can be relatively durable and antigen independent (perhaps these qualities should be considered defining features of ‘Trm’ cells), transiently induced by cytokines, or contingent upon recent or chronic TCR stimulation (perhaps these cells should not even be considered memory T cells, yet current nomenclature practices remain inconsistently ambiguous on this issue). Nevertheless, cells in these varying scenarios may utilize partially overlapping mechanisms (e.g., downregulation of KLF2) to achieve residence. Because T cells require cell-contact for antigen-sensing, their surveillance strategies are fundamental to their role in the immune system. Since its discovery, residence has proven to be a remarkably common and important mechanism. So much so that it is shared by a diversity of T cells.

RESOURCE AVAILABILITY

Lead Contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, David Masopust (masopust@umn.edu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

The raw and processed RNA-sequencing and CITE-seq data files are deposited at GEO (accession numbers GSE276767 and GSE277081). Any additional information required to reanalyze data reported in this paper will be available from the Lead Contact, David Masopust (masopust@umn.edu), upon request.

STAR METHODS

EXPERIMENTAL MODELS ND STUDY PARTICIPANT DETAILS

Mice

Female 6-week-old mice, including C57BL/6J (CD45.2⁺ B6) and B6.SJL-Ptprca Pepcb/BoyJ (CD45.1⁺ B6), were purchased from The Jackson Laboratory. They were maintained under specific-pathogen-free conditions at the University of Minnesota. We bred and housed CD45.1⁺/Thy1.2⁺ and Thy1.1⁺/Thy1.2⁺ OT-I and CD45.2⁺/Thy1.1⁺ and CD45.1⁺/Thy1.1⁺/Thy1.2⁺ P14 transgenic mice in-house. For cohousing experiments to phenotype endogenous CD8⁺ populations, we purchased female CD45.2⁺ and CD45.1⁺ B6 mice from Charles River Laboratories. Cohousing was performed as described²⁶, within the University of Minnesota Animal Biosafety Level 3 facility. Female pet store mice were purchased from shops in the Minneapolis–St. Paul metropolitan area. Mice were handled in accordance with the guidelines set forth by the Institutional Animal Care and Use Committees at the University of Minnesota. The following housing conditions were regulated: temperature (20.0–23.3 °C), humidity (30–70%) and light/dark cycling (14-h on/ 10-h off)

Adoptive cell transfers and infections

For acute infections, 5 × 10⁴ naïve OT-I or P14 CD8⁺ T cells were adoptively transferred into naïve C57BL/6J mice followed by i.v infection with 1 × 10⁶ plaque forming units (PFU) VSVova, or i.p infection with 2×10⁵ PFU LCMV Armstrong (Arm), or i.n. infection with 500 PFU PR8-gp33 1 day later. To establish chronic infection and assess migration, C57BL/6J mice received 200 μg of anti-CD4 antibody (BioCell, clone GK1.5) one day prior to the adoptive transfer of 2.5 × 10³ P14 cells, followed by i.v. infection with 2 × 10⁶ PFU LCMV Cl13 the next day, and subsequently received anti-CD4 antibody treatment again one day after infection.^30,32,33 To establish chronic infection and sequence endogenous gp33⁺ CD8⁺ T cells, C57BL/6J mice received 200 μg of anti-CD4 antibody one day prior to i.v. infection with 2 × 10⁶ PFU LCMV Cl13 followed by anti-CD4 antibody treatment the next day.³¹ For RNA-Seq of chronically stimulated endogenous gp33 tetramer⁺ CD8⁺ T cells, female C57BL/6J were CD4 T cell depleted one day prior and one day after LCMV Cl13 infection (2×10⁶ PFU via i.v. injection). To achieve bystander activation, LCMV Arm P14 immune chimeras were infected i.v. with 1×10⁶ PFU VSV-NJ. For all experiments, donor OT-I or P14 CD8⁺ T cells were derived from 6–14-week-old transgenic C57BL/6J female mice and recipient female C57BL/6J mice were 6–10 weeks old.

METHODS DETAILS

Naïve CD8⁺ T cell in vitro programming

Naïve (CD44^low) CD8⁺ T cells were activated and programmed in vitro as previously described.^77,78 Pooled lymph nodes (including axillary, brachial, cervical, inguinal, and mesenteric) and spleen from a naïve P14 mouse were dissociated and CD8⁺ T cells obtained using negative selection (STEMCELL technologies). Purity was consistently >94% CD8⁺ and >99% CD44^low. Flat bottom 24-well tissue culture dishes were precoated overnight at 4°C with 10 μg/ml anti-CD3 (BioXCell, 145–2C11) and 0.8 μg/ml B7–1/Fc Chimera (RnD Systems, 740-B1–100). 0.5 × 10⁶ CD44^low P14 cells in 2 ml of T cell media (RPM1 1640 (Cytiva) containing 10% Fetal Bovine Serum (FBS), L-glutamine (Gibco), and non-essential amino acids (Gibco), and 0.04% BME (Gibco)) were added to each well in the presence of 100U/ml rIL-2 (NIH) and 2.5 ng/ml murine rIL-12 (RnD Systems). Activated P14 cells were harvested at 72 hours by pipetting and washed twice with ice-cold PBS. 1×10⁶ activated P14 cells were adoptively transferred i.v. into naïve C57BL/6J mice.

Parabiosis and percent resident calculations

Parabiosis surgeries were performed on age matched mice as described.⁷⁹ Briefly, mice were anesthetized with avertin (mg kg⁻¹). Lateral skin was shaved and disinfected before matching incisions were created from the olecranon process to knee joint of each mouse. Continuous surgical staples were placed to join the skin from each mouse. Equilibration in the peripheral blood was confirmed, and residence was calculated in all instances as described²⁹ except for two scenarios. For the LCMV Cl13 infection and in vitro activated transferred cell models, residence was calculated as described.²⁸ Samples with fewer than 60 cells of interest were excluded from percent residence calculations.

NICD-protector and intravascular staining

Mice were intravenously injected with ARTC2.2-blocking nanobody S+16a (Treg-protector, BioLegend) prior to organ harvest as described.⁸⁰ Intravascular staining was performed to differentiate cells present in the vasculature from cells in the parenchyma as explained.²⁷ Briefly, 3μg of biotinylated- or fluorophore-conjugated anti-CD8a (53–6.7) antibody was injected retro-orbitally into mice 3 minutes before sacrifice. Mice were bled via cheek vein puncture 30 seconds before sacrifice to confirm intravascular antibody staining.

Cell isolation from tissues

Spleen, thymus, bone marrow, and LNs

Published protocols were followed for cell isolation from bone marrow⁸¹, thymus⁸², and secondary lymphoid organs (SLOs)⁸³ with slight modification as necessary. Concisely, cells were released from the spleen, thymus, and LNs by mechanical disruption followed by filtration (70-μm filter) and then washed in RPMI 1640 containing 5% fetal bovine serum (FBS). Spleen RBCs were lysed with ACK buffer²⁷ and washed once. For isolation of cells from the bone marrow, one femur was cut on one end and then centrifuged for 10 seconds at 14,000 g prior to negatively selecting for CD8⁺ T cells (STEMCELL technologies).

Large intestine, small intestine, and stomach

Lymphocytes from the large intestine, small intestine and stomach were isolated as described with modifications.^28,83

For isolation of small intestinal (SI) intraepithelial lymphocytes (IEL), Peyer’s patches and contents were first removed. The SI was cut longitudinally and then mucus and remaining contents were gently scraped off the intestine to clean and expose the inner side (mucosal surface) of the intestine. The SI was then cut laterally into small pieces. For isolation of large intestine (LI) IEL, contents were removed, and the LI was cut longitudinally and then laterally into small pieces. For isolation of stomach (ST) IEL, the organ was immersed in CMF solution, cut longitudinally to quickly remove the contents, cut into smaller pieces. SI, LI, and ST pieces were washed in CMF solution and then incubated for 30 minutes with stirring at 37 °C with 0.154 mg/ml dithioerythritol (DTE, Sigma-Aldrich) in 10% Hank’s Balanced Salt Solution (HBSS, without sodium bicarbonate, calcium, and magnesium from Corning). After incubation for 30 minutes, SI, LI, and ST pieces were vortexed on high speed to release IEL and the supernatant was collected and enriched on a Percoll gradient. The SI, LI, and ST pieces were washed in room temperature PBS, put in RPMI 1640 containing 5% FBS, 2 mM MgCl², 2 mM CaCl² (digestion buffer) and 0.5mg/mL 100 U/mL Collagenase type 1 (Worthington Biochemical), and incubated for 60 minutes (SPF mice) or at 45 minutes (CoH mice) at 37 °C with stirring to obtain lamina propria (LP) lymphocytes. After enzymatic digestion, the remaining tissue pieces were mechanically disrupted using a gentleMACs dissociator (Miltenyi Biotec) and homogenates were passed through a 70-μm filter. Lymphocytes were purified on a 44/67% Percoll gradient (800xg at 23°C for 20 minutes).

Kidney, lungs, and liver

For isolation of cells from the kidneys, both kidneys were excised, and the capsules were removed. Lung cells were isolated as described.⁸⁴ Bronchoalveolar lavage (BAL) fluid was collected in three 1-ml cold PBS lavages followed by removal of the medLN and accompanying lymphatic vessels before the lungs were excised.

After mincing, kidney and lung tissues were digested in Collagenase type I for 60 minutes at 37°C with constant shaking. After 1 hour, the remaining tissue pieces were gentleMACs dissociated. Kidney tissues were dissociated once, while lung tissues underwent two dissociation cycles. Subsequently, homogenates were filtered through a 70-μm mesh filter, and lymphocytes were purified on a 44/67% Percoll gradient (800xg at 23°C for 20 minutes).

Liver was physically dissociated using the plunger of a 3-mL syringe and filtered through 70-μm mesh filter. Liver RBCs were lysed with ACK buffer.²⁷ Cells were then washed once, and lymphocytes were purified on a 44/67% Percoll gradient (800xg at 23°C for 20 minutes).

FRT, pancreas, and salivary glands

After being excised, the FRT (female reproductive tract) was cleaned of any fat, cut longitudinally. FRT, pancreas (PANC) and salivary gland (SG) tissues were cut into small pieces (not minced). Tissue pieces were enzymatically digested in digestion buffer containing 0.5 mg/ml Collagenase type IV (Sigma-Aldrich) for FRT or 100 U/ml Collagenase type I for PANC and SG. FRT tissues were incubated for 60 minutes, SG tissues for 45 minutes, and PANC tissues for 20 minutes, all at 37°C with shaking. After enzymatic digestion, the remaining tissue pieces underwent gentleMACs dissociation, followed by homogenates being filtered twice through a 70-μm mesh filter. FRT tissue pieces underwent two rounds of dissociation, while PANC and SG tissues underwent one round each.

Brain

After being excised, the brain was cut into small pieces and enzymatically digested in digestion buffer containing 100 U/mL Collagenase type 1 for 45 minutes at 37°C with constant shaking. The brain pieces were then physically dissociated using the plunger of a 3-mL syringe, filtered through 70-μm mesh filter. Lymphocytes were purified on a 44/67% Percoll gradient (800xg at 23°C for 20 minutes).

Skin

Cells were isolated from the skin as described with modifications.⁸⁵ Briefly, a 1–2 cm² area of shaved belly skin was harvested, chopped into small pieces, and incubated with digestion buffer containing type IV collagenase (1 mg/ml) and DNase I (2 mg/ml) at 37°C with constant shaking for 60 minutes then further gentleMACs dissociated and filtered twice through a 70-μm mesh filter. Lymphocytes were purified on a 44/67% Percoll gradient (800xg at 23°C for 20 minutes).

Flow cytometry data acquisition and analyses

Enriched lymphocytes were surface stained with monoclonal anti-mouse antibodies indicated in the KEY RESOURCE TABLE. Cell viability was determined using either Ghost Dye Red 780 (Tonbo Biosciences) or Zombie NIR dye (BioLegend). Flow cytometric data was acquired on LSR Fortessa (BD Biosciences) or Cytek Aurora (Cytek Biosciences) instruments and analyzed using FlowJo^™ v10 software (Treestar). Previously published protocols and considerations for high-dimensional panel design and data analyses^86,87 were followed to avoid technical artifacts in the clustering analysis. Data cleaning and gating was performed prior to merging host and partner derived cells across mice experiments and tissues (illustrated in Figure 4B). Specifically, contaminants in the data, such as dead cells, doublets, and cells stained by fluorochrome aggregates, were excluded from the analysis using a series of relevant gating strategies.⁸⁶ Clustering analysis was performed in FlowJo^™ v10 software with the PhenoGraph⁵³ plugin.

KEY RESOURCE TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Biotinylated anti-CD8α antibody (53-6.7)	eBioscience	Catalog# 13-0081-82; RRID: AB_466346
Brilliant ultraviolet 395 anti-CD49a antibody (Hα31/8)	BD Biosciences	Catalog# 740262; RRID: AB_2740005
Brilliant ultraviolet 496 anti-CD8β antibody (YTS156.7.7.rMAb)	BD Biosciences	Catalog# 755242; RRID: AB_3099639
Brilliant ultraviolet 496 anti-CD8α antibody (53-6.7)	BD Biosciences	Catalog# 750024; RRID: AB_2874242
Brilliant ultraviolet 563 anti-Ly-6C antibody (HK1.4.rMAb)	BD Biosciences	Catalog# 755198; RRID: RRID:AB_3099650
Brilliant ultraviolet 615 anti-CD62L antibody (MEL-14)	BD Biosciences	Catalog# 752311; RRID: AB_2875828
Brilliant ultraviolet 661 anti-CD49a antibody (Hα31/8)	BD Biosciences	Catalog# 750628; RRID: AB_2874760
Brilliant ultraviolet 737 anti-CD45.1 antibody (A20)	BD Biosciences	Catalog# 612811; RRID: AB_2870136
Brilliant ultraviolet 737 anti-CD45.2 antibody (104)	BD Biosciences	Catalog# 564880; RRID: AB_2738998
Brilliant ultraviolet 737 anti-CD90.1 (Thy-1.1) antibody (HIS51)	BD Biosciences	Catalog# 741774; RRID: AB_2871129
Brilliant ultraviolet 737 anti-CD62L (MEL-14)	BD Biosciences	Catalog# 565213 RRID: AB_2721774
Brilliant ultraviolet 805 anti-CD45.1 antibody (A20)	BD Biosciences	Catalog# 741958; RRID: AB_2871266
Brilliant ultraviolet 805 anti-CD45.2 antibody (104)	BD Biosciences	Catalog# 741957; RRID: AB_2871265
Brilliant ultraviolet 805 anti-CD90.1 (Thy-1.1) antibody (OX-7)	BD Biosciences	Catalog# 741975; RRID: AB_2871279
Brilliant Violet 421 anti-CD69 antibody (H1.2F3)	BioLegend	Catalog# 104528; RRID: AB_2686969
Brilliant Violet 421 anti-CX3CR1 antibody (SA011F11)	BioLegend	Catalog# 149023 RRID: AB_2565706
Pacific Blue anti-CD27 antibody (LG.3A10)	BioLegend	Catalog# 124218; RRID: AB_2561546
Brilliant violet 510 anti-CD103 antibody (M290)	BD Biosciences	Catalog# 563087; RRID: AB_2721775
Brilliant violet 510 anti-CD103 antibody (2E7)	BioLegend	Catalog# 121423 RRID: AB_2562713
Brilliant Violet 510 anti-CD62L antibody (MEL-14)	BioLegend	Catalog# 104441; RRID: AB_2561537
Brilliant violet 605 anti-CD62L antibody (MEL-14)	BioLegend	Catalog# 104438; RRID: AB_2563058
Brilliant Violet 605 anti-PD1 antibody (29F.1A12)	BioLegend	Catalog# 135220; RRID: AB_2562616
Brilliant Violet 650 anti-CD8a antibody (53-6.7)	BioLegend	Catalog# 100742; RRID: AB_2563056
Brilliant violet 711 anti-Ly6C antibody (HK1.4.rMAb)	BioLegend	Catalog# 128037; RRID: AB_2562630
Brilliant violet 711 anti-CXCR6 antibody (SA051D1)	BioLegend	Catalog# 151111; RRID: AB_2721558
Brilliant violet 711 anti-Ly-108 antibody (13G3)	BD Biosciences	Catalog# 740823 RRID: AB_2740481
Brilliant violet 711 anti-CD45.1 (A20)	BioLegend	Catalog# 110739 RRID: AB_2562605
Brilliant violet 785 anti-CD44 antibody (IM7)	BioLegend	Catalog# 103059; RRID: AB_2571953
Brilliant violet 786 anti-CD62L antibody (MEL-14)	BD Horizon	Catalog# 564109 RRID: AB_2738598
FITC anti-CD27 antibody (LG.3A10)	BioLegend	Catalog# 124208; RRID: AB_1236466
FITC anti-KLRG1 antibody (2F1)	BioLegend	Catalog# 138410; RRID: AB_10643582
FITC anti-CD127 antibody (SB/199)	BioLegend	Catalog# 121106; RRID: AB_493503
FITC anti-CD45.1 antibody (A20)	BioLegend	Catalog# 110705; RRID:AB_313494
FITC anti-CD90.1 antibody (S20007C)	BioLegend	Catalog# 109007; RRID:AB_3083262
PerCPCy5.5 anti-CD38 antibody (90)	BioLegend	Catalog# 102722 RRID: AB_2563332
PerCPCy5.5 anti-CD43 antibody (1B11)	BioLegend	Catalog# 121224; RRID: AB_2286556
PE anti-CD160 antibody (7H1)	BioLegend	Catalog# 143004; RRID: AB_10960740
PE anti-P2RX7 antibody (1F11)	BioLegend	Catalog# 148704; RRID: AB_2650951
PE anti-CD101 antibody (Moushi101)	eBioscience	Catalog# 12-1011-82; RRID: AB_1210728
PE anti-CD218 (IL18Ra) antibody (A17071D)	BioLegend	Catalog# 157904; RRID: AB_2860732
PE-Dazzle 594 anti-CD69 antibody (H1.2F3)	BD Biosciences	Catalog# 562455; RRID: AB_11154217
PE-Dazzle 594 anti-CX3CR1 antibody (SA011F11)	BioLegend	Catalog# 149014; RRID: AB_2565697
PE-Dazzle 594 anti-CXCR5 (L138D7)	BioLegend	Catalog# 145522; RRID: AB_2563643
PE-Dazzle 594 anti-GZMB antibody (QA16A02)	BioLegend	Catalog# 372216 RRID: AB_2728383
PE Cy7 anti-CD101 antibody (Moushi101)	eBioscience	Catalog# 25-1011-82; RRID: AB_2573378
PE Cy7 anti-CD38 antibody (90)	BioLegend	Catalog# 102717 RRID: AB_2072892
PE Cy7 anti-P2RX7 (1F11)	BioLegend	Catalog# 148708; RRID: AB_2721685
APC anti-CD160 antibody (7H1)	BioLegend	Catalog# 143012; RRID: AB_2562698
APC anti-P2RX7 (1F11)	BioLegend	Catalog# 148706; RRID: AB_2650954
Alexa fluor 700 anti-CD8β antibody (YTS156.7.7)	BioLegend	Catalog# 126618; RRID: AB_2563949
Alexa fluor 700 anti- CD38 antibody (90)	BioLegend	Catalog# 102742; RRID: AB_2890672
Alexa fluor 700 anti-CD90.2 antibody (53-2.1)	BioLegend	Catalog# 140324; RRID: AB_2566739
PE Fire 700 anti-CX3CR1 antibody (SA011F11)	BioLegend	Catalog# 149052; RRID: AB_2910299
PerCP eFlour 710 anti-CD218(IL18Ra) antibody (P3TUNYA)	eBioscience	Catalog# 46-5183-82; RRID: AB_2573764
anti-CD4 antibody (GK1.5)	Bio × Cell	Catalog# BE0003-1; RRID: AB_1107636
anti-CD3 antibody (145-2C11)	Bio × Cell	Catalog# BE0001-1; RRID: AB_2687679
TotalSeq-A0112 anti-mouse CD62L antibody	BioLegend	Catalog# 104451; RRID: AB_2750364
TotalSeq-A0197 anti-mouse CD69 antibody	BioLegend	Catalog# 104546; RRID: AB_2750539
TotalSeq-A0201 anti-mouse CD103 antibody	BioLegend	Catalog# 121437; RRID: AB_2750349
TotalSeq-A0380 anti-rat CD90/mouse CD90.1 (Thy-1.1) antibody	BioLegend	Catalog# 202547; RRID: AB_2783141
TotalSeq-A0178 anti-mouse CD45.1 antibody	BioLegend	Catalog# 110753; RRID: AB_2800573
TotalSeq-A0013 anti-mouse Ly-6C antibody	BioLegend	Catalog# 128047; RRID: AB_2749961
TotalSeq-A0563 anti-mouse CX3CR1 antibody	BioLegend	Catalog# 149041; RRID: AB_2783121
TotalSeq-A0250 anti-mouse/human KLRG1 (MAFA) antibody	BioLegend	Catalog# 138431; RRID: AB_2800648
TotalSeq-C0198 anti-mouse CD127 (IL-7Rα) antibody	BioLegend	Catalog# 135047; RRID: AB_2819874
TotalSeq-A0595 anti-mouse CD11a antibody	BioLegend	Catalog# 101125; RRID: AB_2783036
TotalSeq-A0301 anti-mouse Hashtag 1 antibody	BioLegend	Catalog# 155801; RRID: AB_2750032
TotalSeq-A0302 anti-mouse Hashtag 2 antibody	BioLegend	Catalog# 155803; RRID: AB_2750033
TotalSeq-A0303 anti-mouse Hashtag 3 antibody	BioLegend	Catalog# 155805; RRID: AB_2750034
TotalSeq-A0304 anti-mouse Hashtag 4 antibody	BioLegend	Catalog# 155807; RRID: AB_2750035
TotalSeq-A0305 anti-mouse Hashtag 5 antibody	BioLegend	Catalog# 155809; RRID: AB_2750036
TotalSeq-A0306 anti-mouse Hashtag 6 antibody	BioLegend	Catalog# 155811; RRID: AB_2750037
TotalSeq-A0307 anti-mouse Hashtag 7 antibody	BioLegend	Catalog# 155813; RRID: AB_2750039
TotalSeq-A0308 anti-mouse Hashtag 8 antibody	BioLegend	Catalog# 155815; RRID: AB_2750040
Bacterial and virus strains
Lymphocytic choriomeningitis virus (LCMV)- Armstrong strain	Dr. R. Ahmed, Emory University	N/A
Lymphocytic choriomeningitis virus (LCMV)- Cl13 strain	Dr. R. Ahmed, Emory University	N/A
Vesicular stomatitis virus (VSV)-Indiana strain expressing ovalbumin (VSVova)	Dr. L. Lefrancois, Univ. of Connecticut	N/A
PR8-gp33	Dr. R.A. Langlois, Univ. of Minnesota	N/A
Chemicals, peptides, and recombinant proteins
Collagenase I	Worthington Biochemicals	Catalog# LS004197
Collagenase IV	Sigma Aldrich	Catalog# C5138-5G
Dnase I	Sigma Aldrich	Catalog# D4513
Dnase I	Roche	Catalog# 11284932001
Live dead Ghost Dye Red 780	TONBO Biosciences	Catalog# 13-0865-T500
Live dead Ghost Dye Violet 510	TONBO Biosciences	Catalog# 13-0870-T100
Zombie NIR Fixable Viability Kit	BioLegend	Catalog# 423106
H-2Db /gp33–41 KAVYNFATM biotinylated monomer	Prepared in house following NIAID tetramer core facility protocol	N/A
Streptavidin-APC (monomer tetramerization)	Life Technologies	Catalog# S868
Streptavidin-PE (monomer tetramerization)	Life Technologies	Catalog# S866
Foxp3/Transcription Factor Staining Buffer Set	eBioscience	Catalog# 00-5523-00
Fetal bovine serum (FBS)	Atlas Biologicals	Catalog# F-0500-AR
HyClone RPMI 1640 medium (2.05 mM L-glutamine)	Cytiva	Catalog# SH30027.02
HyClone Phosphate-buffered saline (without calcium and magnesium)	Cytiva	Catalog# SH30028.03
0.5 mM EDTA (pH 8.0)	Promega	Catalog# V4231
1 M HEPES	Lonza	Catalog# 17-737E
10X Hank’s Balanced Salt Solution	Corning	Catalog# 20-021-CV
Percoll	Cytiva	Catalog# 17-0891-09
Brilliant Violet 650 Streptavidin	BioLegend	Catalog# 405232
TotalSeq-A0951 PE Streptavidin	BioLegend	Catalog# 405251
Nuclease-Free water (not DEPC-Treated)	Invitrogen	Catalog# AM9937
Deionized double distilled water	Bioworld	Catalog# 423000001
Treg-Protector (ARTC2.2-blocking nanobody) (S+16)	BioLegend	Catalog# 149802 RRID: AB_2819899
Recombinant mouse B7-1/CD80 Fc Chimera protein	RnD Systems	Catalog# 740-B1-100
Recombinant human IL-2 protein (Lot# C161500-02)	NIH	Identifier: Ro-23-6019 MeSH identifier: M0373553
Recombinant mouse IL-12 protein	RnD Systems	Catalog# 419-ML-500/CF
L-glutamine	Gibco	Catalog# 25030-081
Non-Essential Amino Acids (NEAA)	Gibco	Catalog# 11140-050
2-Mercaptoethanol (BME)	Gibco	Catalog# 21985023
Critical commercial assays
EasySep Mouse CD8+ T Cell Isolation Kit	STEMCELL Technologies	Catalog# 19853
RNeasy Plus Micro Kit	QIAGEN	Catalog# 74034
QIAshredder columns	QIAGEN	Catalog# 79654
Dual Index Kit TT Set A	10X Genomics	Catalog# 1000215
KAPA HiFi Master Mix	Roche	Catalog# 07958927001
TruSeq Small RNA Sample Prep Kit (ADT Index)	Illumina	Catalog# RS-200-0012
Single Cell 3’v3 Gel Bead and Library Kit	10X Genomics	Catalog# 1000092
Chromium i7 Multiplex kit	10X Genomics	Catalog# 120262
Deposited data
Transcriptional comparisons of CD8+ T cell populations	This paper	NCBI GEO accession code: GSE276767
CITE-seq dataset of LCMV Arm and PR8-specific memory CD8+ T cells	This paper	NCBI GEO accession code: GSE277081
Human memory CD3+ T cell single cell RNA-seq	Poon et al., 2023	NCBI GEO accession code: GSE206507 SRA accession code: PRJNA851108
The Molecular Signatures Database (MSigDB)	Subramanian et al., 2005	https://www.gsea-msigdb.org/gsea/msigdb/index.jsp
MSigDB: Gene Ontology biological process (GOBP) and molecular function (GOMF)	Ashburner et al., 2000	https://www.gsea-msigdb.org/gsea/msigdb/human/collections.jsp#C5
Experimental models: Cell lines
Baby hamster kidney-21 (BHK-21) cells	ATCC	Catalog# CRL-6281 RRID:CVCL_1914
Vero cells	ATCC	Catalog# CCL-81 RRID:CVCL_0059
Experimental models: Organisms/strains
C57BL/6J	The Jackson Laboratory	Catalog# 000664-JAX; RRID: IMSR_JAX:000664
C57BL/6NCrl	Charles River Laboratories	Catalog# 027-CRL; RRID:IMSR_CRL:027
Mouse strain: P14	Dr. R. Ahmed, Emory University	Catalog# 037394-JAX; RRID:MMRRC_037394-JAX
Mouse strain: OT-I	Dr. K. Hogquist, Univ. of Minnesota	Catalog# 003831-JAX; RRID: IMSR_JAX:003831
B6 CD45.1	Charles River Laboratories	Catalog# 564-CRL; RRID: IMSR_CRL:564
B6 Thy1.1	The Jackson Laboratory	Catalog# 000406-JAX; RRID: IMSR_JAX:000406
Pet store mice	Pet stores in the greater Minneapolis-St. Paul metropolitan area	N/A
Software and algorithms
FlowJo (v9 and v10)	TreeStar Inc.	RRID:SCR_008520
Prism (v9.1.2.)	GraphPad Inc.	RRID:SCR_002798
BD FACSDiva Software	BD Biosciences	RRID:SCR_001456
Trimmomatic (v0.33)	Bolger et al., 2014	RRID:SCR_011848
Picard MarkDuplicates (v2.19.0)	http://broadinstitute.github.io/picard/	RRID:SCR_006525
Samtools (v1.5)	Li et al., 2009 http://htslib.org/	RRID:SCR_002105
Cufflinks (v2.2.1)	http://cole-trapnell-lab.github.io/cufflinks/cuffmerge/	RRID:SCR_014597
Hisat2 (v2.0.2)	Kim et al., 2015 http://ccb.jhu.edu/software/hisat2/index.shtml	RRID:SCR_015530
featureCounts (v2.0.1)	Liao et al., 2014	RRID:SCR_012919
Bioconductor (v3.10)	Gentleman et al., 2004	RRID:SCR_006442
edgeR (v3.36.0)	Robinson et al., 2019	RRID:SCR_012802
cellRanger (v7.0.0)	10X Genomics	https://support.10xgenomics.com/single-cell-gene-expression/software/downloads/latest
R (v4.1.2)	The R Project	https://www.r-project.org/
Seurat v4	Satija et al., 2015 Stuart et al., 2019	https://github.com/satijalab/seurat/
Python (v3.5.5)	https://www.python.org/	RRID:SCR_008394
matplotlib (v3.5.2)		https://zenodo.org/record/3984190
numpy (v1.21.6)		https://github.com/numpy/numpy
numba (v0.55.0)		https://numba.pydata.org/
pandas (v1.4.2)	The pandas development team	https://zenodo.org/record/6702671
scanpy (v1.8.2)	Wolf et al., 2018	http://github.com/theislab/scnpy
scikit-learn (sklearn) (v1.0.2)		http://scikit-learn.org
scipy (v1.8.0)		https://zenodo.org/record/3958354
scvi-tools (v0.16.0)	Lopez et al., 2018 Gayoso et al., 2022	https://github.com/scverse/scvi-tools
Seaborn (v0.11.2)	Waskom, 2021	https://seaborn.pydata.org
totalVI	Gayoso et al., 2021	https://github.com/YosefLab/totalVI_reproducibility
Microsoft Excel	Microsoft	RRID:SCR_016137
Microsoft PowerPoint	Microsoft	RRID:SCR_023631
Adobe Illustrator	Adobe
Other
BD FACS Aria III Cell Sorter	BD Biosciences
2100 BioAnalyzer	Agilent	RRID:SCR_018043
Chromium Controller	10X Genomics	Chromium Controller
BD LSRFortessa X-20 Cell Analyzer	BD Biosciences	RRID:SCR_025285

RNA-seq experiment

Memory P14 CD8⁺ T cell populations specific to lymphocytic choriomeningitis virus (LCMV) strain Armstrong (Arm) were created for RNA-seq and transcriptome analyses. First, 5 × 10⁴ naïve CD45.1⁺ P14 CD8⁺ T cells were adoptively transferred into naïve C57BL/6J mice followed by i.p infection with 2×10⁵ PFU LCMV Arm. At >30 days p.i., i.v. antibody labeling was performed with anti-CD8a 3 minutes prior to euthanizing the mice and removal of tissues. We then created single cell suspensions (as described above) from the excised tissues. Cells were stained with anti-CD8b, anti-CD45.1, anti-CD44, anti-CD69, anti-CD62L, and Ghost Dye Red 780 (Tonbo Biosciences) flow antibodies. Subsequently, LCMV Arm specific Tcm (CD8b⁺CD45.1⁺CD69⁻CD62L⁺), Tem (CD8b⁺CD45.1⁺CD69⁻CD62L⁻), and Trm (CD8b⁺CD45.1⁺CD69⁺CD62L⁻)⁸⁵ cells were sorted from pooled SLO cell suspensions with a FACSAria II (BD Biosciences). Extravascular Trm cells (identified as i.v. antibody negative CD8b⁺CD45.1⁺CD69⁺CD62L⁻ P14 cells) were sorted from single-cell suspensions originating from the small intestine epithelial and lamina propria compartments, as well as the FRT. Extravascular LCMV Arm specific CD69⁺CD62L⁻ P14 cells in these tissues were presumed to be resident based on prior parabiosis studies.

Bystander activated (CD8⁺CD45.1⁺CD69⁺) P14 cells were sorted from pooled SLO of LCMV Arm immune mice 48 hours after i.v. VSV-NJ infection. Recently stimulated (CD8⁺CD45.1⁺CD69⁺) P14 cells were sorted from pooled SLO of mice 4 days after LCMV Arm infection. Endogenous CD69⁺ H-2D^b/gp33 tetramer⁺CD8⁺ T cells were sorted from pooled SLOs of mice 50 days after LCMV Cl13 infection in CD4⁺ T cell depleted mice.

After sorting to obtain cell populations of interest, RNA isolation, integrity assessment, and sequencing was performed at the University of Minnesota’s Genomics Core as previously detailed.⁸⁸ Sequencing libraries were prepared with the Clontech SMARTer Stranded Total RNA-seq Kit v2 (Pico Input Mammalian Kit). 50 base pair single-end sequences were then generated from the with the HiSeq 2500 Illumina.

RNA-sequencing data processing

Sequencing reads were trimmed, assessed for quality, mapped to the mouse reference genome (UCSC version mm10), and then processed for abundance estimates as explained.⁸⁸ Cutadapt was used to remove the first three nucleotides of the sequencing read, which were derived from the Clontech template-switching oligo⁸⁹ and quality control was performed with the FastQC software (version 0.11.5).⁹⁰ Sequences were trimmed with Trimmomatic (version 0.33),⁹¹ mapped to the UCSC version mm10 mouse genome reference (GRCm38; Genome Reference Consortium Mouse Build 38; GCA_000001635.2) with the HISAT2 aligner (version 2.0.2-β).⁹² Insertion size metrics were calculated for each sample using Picard software (version 1.126; http://picard.sourceforge.net). Samtools (version 1.3)⁹³ was used to sort and index the bam files. Gene abundance estimates were generated using the Rsubread featureCounts program.⁹⁴ Genes <200bp and with a mean abundance value <0 across all the samples were removed prior to analyses.

RNA-seq data analyses

All RNA-seq analyses were performed in RStudio (R version 4.1.2). Multidimensional scaling analysis and visualization of RNA-Seq data generated from CD8⁺ T cell populations was accomplished with the plotMDS function. Differential expression comparisons were performed with the Bioconductor⁹⁵ package edgeR⁹⁶ (version 4.1.3. Differentially expressed genes (DEG) were identified as having an FDR p value ≤ 0.05 and absolute fold change ≥ 2.0 between comparison groups. The Bioconductor packages EnhancedVolcano⁹⁷ and ComplexHeatmap⁹⁸ were utilized to visualize DEG between comparison groups. Gene Ontogeny⁹⁹ term enrichment of genes was evaluated with the clusterProfiler¹⁰⁰ Bioconductor package.

CITE-seq experiment

PR8-gp33 and LCMV Arm immune parabionts were created as described earlier and as shown in Figure 1. For both infection models, at >40 days p.i., mice with Thy1.1⁺ P14 cells (chimeric mice) were joined via parabiosis surgery with those having CD45.1⁺ P14 cells. Following 25–28 days of rest, equilibration in the blood was confirmed. For each infection, three pairs of parabionts were used.

For intravascular staining²⁷, each parabiont was injected i.v. with 3μg of biotinylated-conjugated anti-CD8a (53–6.7). Three minutes post-injection, animals were bled, sacrificed, and tissues were harvested and digested (either mechanically or enzymatically) as explained earlier.

Prior to staining, cells from a particular tissue from all three parabiont mice with Thy1.1⁺ host (and CD45.1⁺ partner) memory P14 cells were pooled to constitute one technical replicate (referred to as “Thy1.1 host”) for that tissue. Similarly, cells from a particular tissue from all three parabiont mice with CD45.1⁺ host (and Thy1.1⁺ partner) memory P14 cells were pooled to form the other technical replicate (referred to as “CD45.1 host”).

The CITE-seq experiment was performed following the TotalSeq protocol. Cells were first surface stained for 30 minutes at 4°C with fluorescent-conjugated and oligonucleotide-conjugated TotalSeq-A anti-mouse antibodies (see KEY RESOURCE TABLE) and Ghost Dye 510 (Tonbo Biosciences) cell viability dye. TotalSeq-A PE anti-streptavidin oligonucleotide-conjugated reagent was included in the stain to recognize cells that were labeled with i.v. antibody during both the flow sorting and data analyses steps.

After staining the cells with flow and CITE-seq compatible antibodies, the cells were washed. Then, for multiplexing purposes, cells from each (pooled) tissue were stained for 30 minutes at 4°C with a unique TotalSeq-A anti-mouse oligonucleotide hashtag. This hashtag comprised a mixture of two monoclonal antibodies (against mouse CD45 and MHC class I) conjugated to the same oligonucleotide. Afterwards, cells were washed again and resuspended in Phosphate-buffered saline (PBS, without calcium and magnesium from Cytiva) supplemented with 2% FBS, 2 mM EDTA (Gibco of Thermo Fisher Scientific), and 10 mM HEPES buffer (Gibco of Thermo Fisher Scientific). Cells were sorted using a FACSAria II (BD Biosciences).

After sorting, cells from all tissues for each parabiont (CD45.1 host or Thy1.1 host) were pooled to constitute one technical replicate. We followed the 10X Genomics Chromium Single Cell 3’ v3 protocol to prepare RNA, antibody-derived-tag (ADT), and hashtag oligos (HTO) libraries. The cell suspension was loaded onto a 10X Genomics Chromium Controller at a concentration of 1,200 cells per microliter to target about 40,000 cells per two technical replicates.

RNA, ADT, and HTO libraires were sequenced with an Illumina NovaSeq 6000 S4 at the University of Minnesota’s Genomics Core. Reads were processed with 10X Genomics Cell Ranger v.7.0.0 with feature barcoding, where RNA reads were mapped to the mouse mm10–2.1.0 mouse reference and antibody reads mapped to known barcodes (see KEY RESOURCE TABLE).

CITE-seq data preprocessing

Before analyses, we used Seurat tools to perform preliminary quality control, cell selection, and feature selection on the CITE-seq data (Figure S4A). A centered log-ratio transformation was performed on the HTO data and then the HTODemux function (Seurat v4.0)¹⁰¹ was utilized to demultiplex cells and only select singlet cells for further analyses. Cells with a high percentage of UMIs derived from mitochondrial genes (> 5% of a cell’s total UMI count) were removed. We also removed cells expressing <600 genes and retained cells with an RNA library size between 1,000 and 20,000 UMI counts and a protein library size between 700 and 20,000 UMI counts. A few contaminating B cells (uniquely expressing high levels of Cd79a) were identified and removed from the datasets for both infections.

After filtering, technical replicates (CD45.1 or Thy1.1 parabionts) were integrated with SelectIntegrationFeatures, FindIntegrationAnchors, and IntegrateData functions in Seurat^101,102 with 10,000 anchors. The Seurat Convert function was used to convert the Seurat object into an annotated data (AnnData) object¹⁰³ for further quality control filtering and analyses with Total Variational Inference (totalVI)⁵⁸ and single-cell variational inference (scVI) tools.¹⁰⁴

An initial filter removed genes expressed in fewer than four the cells. After these filtering steps, the concatenated LCMV Arm dataset contained 25,254 cells and 19,092 genes and the concatenated PR8-gp33 dataset contained 21,001 cells and 17,494 genes. Before analyses, the top 6,000 highly variable genes (HVGs) were selected by the Seurat v3 method¹⁰¹ as implemented by single-cell variational inference (scVI).¹⁰⁴

totalVI analysis of CITE-seq data for each infection

For each infection, totalVI was performed with a 20-dimensional latent space and default parameters. Each 10X lane (representing either the CD45.1 or Thy1.1 host parabiont) was treated as a ‘batch’ for generating the totalVI latent space (Figure S4A). The totalVI input consisted of matrices of RNA (for the top 6,000 HVGs) and protein (Thy1.1, CD45.1, Ly6C, CD62L, CD103, CD69, CD127, KLRG1, and the intravascular (i.v.) antibody) unique molecular identifier (UMI) counts. After performing the CITE-seq experiments for both infections, we learned that CX₃CR1 protein may be cleaved by enzymatic digestion.⁷¹ So, for the analyses, CX₃CR1 protein counts were removed. The CITE-seq experiment for PR8-gp33 included protein counts for CD11a; however, this protein was not part of the LCMV Arm CITE-seq experiment, we excluded CD11a counts from the analyses.

To annotate the cells as resident (Trm cells) or equilibrating (Teq cells) based on the totalVI latent space we first computed neighbor distances and connectivities and determined Uniform manifold approximation and projection coordinates. The Scanpy¹⁰⁵ implementation of the Leiden algorithm⁶¹ was used to assign cells to communities. Cells were clustered at a resolution of 1.0.

The totalVI denoised (corrected for background noise) protein expression values for CD45.1 and Thy1.1 were then evaluated to annotate cells by their parabiont of origin. Simply, a cell determined to be expressing CD45.1 from the CD45.1 host parabiont was annotated as “host”, while a cell determined to be expressing CD45.1 from the Thy1.1 host parabiont was annotated as “partner”.

We determined which cells were ‘host’ versus ‘partner’ derived for each tissue separately due to intrinsic and expected (based on flow cytometric analyses) differences in expression of the Thy1.1 and CD45.1 congenic markers across tissues. To this end, totalVI denoised protein counts for the host and partner congenic markers of cells from a given tissue were log-transformed and visualized in biaxial scatter plots with marginal histograms (seaborn.jointplot).¹⁰⁶ The contours of cell densities in these plots and the distribution of the marginal histograms were used to establish “positive” versus “negative” expression of the host and partner congenic markers. For both infections, cells with positive expression for both the host and partner congenic markers were filtered out. These 3,107 (total number from both infections) double positive cells were mostly derived from the spleen for both infections.

For the CITE-seq dataset from each infection, after annotating cells as ‘host’ or ‘partner’, we evaluated their abundance in the cell clusters of the UMAP plots of the totalVI latent space from the CITE-seq data. Host derived cells in clusters with few to no partner derived cells were annotated as Trm cells. Partner derived cells in clusters with few to no partner derived cells were annotated as Teq cells, even though we could not rule out that those cells were differentiating into Trm cells. Cells in clusters with both host- and partner-derived cells were annotated as Teq cells. There were, as expected based on flow cytometry results in this study (and previous studies),^28,79,88 very few partner-derived cells from the FRT, SG, and SI. The partner parabiont derived P14 cells from these tissues were annotated as Teq cells.

For the LCMV Arm CITE-seq dataset, after visualizing the UMAP plots of the totalVI latent space, a cluster of CD103⁺ SI-IEL partner derived cells (n=323 cells) were identified and removed as experimental contaminants. After these contaminating cells were removed, totalVI was run on the LCMV Arm dataset again to annotate cells as host- or partner-derived and establish Trm cell and Teq cell annotations (grounds truths; Figure S4A).

totalVI analysis of combined CITE-seq dataset

The AnnData objects containing CITE-seq data from PR8-gp33 and LCMV Arm immune parabionts were concatenated. After concatenation, the data set was filtered to remove genes that were expressed in less than four cells. The concatenated dataset contained 40,916 cells. Before analyses, the top 6,000 HVGs were selected by the Seurat v3 method¹⁰¹ as implemented by single-cell variational inference (scVI).^60,104

totalVI was performed with a 20-dimensional latent space and default parameters. Infection history (PR8 or LCMV Arm) was used as ‘batch’ (Figure S4A). The totalVI input to generate the latent space consisted of matrices of RNA (for the top 6,000 HVG) and protein (Ly6C, CD62L, CD103, CD69, CD127, KLRG1, and the intravascular (i.v.) antibody) UMI counts. Prior to generating the totalVI latent space for cells from both infections, the protein counts for each infection were scaled to account for differences in sequencing depth across the infections. Specifically, for each cell, the protein count value for each protein was divided by the library size for that cell to get a normalized count value. The normalized count value was then multiplied by the average library size for all cells in the data set (per infection) to get a scalar value which was then rounded up to the nearest integer.

The totalVI latent space was used to compute neighbor distances and connectivities and determine Uniform manifold approximation and projection coordinates. The Scanpy¹⁰⁵ implementation of the Leiden algorithm⁶¹ was used to assign cells to communities. Cells were clustered at a resolution of 1.0.

scVI analysis of CITE-seq data

In all instances where a probabilistic scVI latent space of a subset or all the CITE-seq data was generated (Figure S4A), the matrix of RNA (for the top 6,000 HVGs) UMI counts was used as the input. We ran scVI using a 20-dimensional latent space, dispersion=‘gene’, gene_likelihood = ‘zinb’, and latent_distribution =‘normal’. Infection history (PR8 or LCMV Arm) was used as ‘batch’.

Differential expression testing

We conducted one-vs-all differential expression tests between Leiden algorithm⁶¹ defined clusters to identify cluster defining features. Differential expression tests between PR8-gp33 specific and LCMV Arm tissue resident memory P14 cells were performed to examine how infection history influences gene expression. We also performed differential expression tests between Trm and non-resident (Teq) cells to identify features unique to Trm cells, either within tissues or shared across tissues. To identify cluster specific, infection specific, or Trm cell specific features we filtered by false discovery rate (is_de_fdr_0.05 = TRUE)¹⁰⁷, significance (log(Bayes factor) > 2.0 for protein, log(Bayes factor) >3.0 for genes), effect size (mean log fold change > 1 for both proteins and genes), and the proportion of expressing cells (detected expression in > 10% of the population of interest for genes).

Differentially expressed genes in each cluster were sorted by median log fold change and the top differentially expressed genes for each cluster were used for dot plot visualizations. Dendrograms in the expression dot plots are based on Pearson correlation and ‘complete’ linkage method of denoised gene expression values within each cluster.

Gene enrichment scores

Signed gene set scores— a combined score that accounts for both genes with increased expression (‘up’) and decreased expression (‘down’) in a gene signature (Figures 7H and S7A)— and unsigned gene set scores (representing only ‘up’ or ‘down’ genes in a signature; Figures S5B and S7B) for each cell were computed using visionpy, the Python implementation of the VISION package in R.¹⁰⁸ Depending on the analysis, normalized counts from either the scVI or the totalVI latent space were used for generating enrichment scores. Genes used for scoring were derived from differential analysis testing and random forest classification results or from publicly available gene lists. Mouse hypoxia response genes were retrieved from Molecular Signatures Database (MSigDB) gene set MM3861.^109,110 Mouse adipogenesis genes were downloaded from QIAGEN’s Ingenuity Knowledge base.¹¹¹ Mouse Trm cell genes were obtained from a previous study⁶³ that derived them from through a reanalysis of a published transcriptomics dataset.²¹ Human lung Trm cell genes with a median log fold change of > 2 relative to all other cells in the data set were derived from a previous study⁵⁹ and are listed in Table S4.

Random forest classification

Denoised protein and RNA values were used for random forest classification modeling. For both of the clades of memory CD8⁺ T cells (containing either Icos^hi and Icos^lo Trm), 80% of the cells were used to train a random forest (sklearn.ensemble.RandomForestClassifier) with 100 trees, each at the default max depth setting to reduce overfitting. Once trained, the forest for each clade would predict the Trm or Teq identity and provide a proportion of tress in agreement with the classification. The 20% of the cells that were not used to build the model were used to check the random forest model predictions against actual truths (Trm or Teq cell annotation labels). Gini importance scores for each gene based on the random forest classifier were used to rank DEG identified from differential tests between Trm and Teq cells of each clade.

Logistical regression and ROC curves

Signed signature scores generated by visionpy were used as the input for logistic regression (sklearn.linear_model.LogisticRegression). As with random forest classification, 80% of the cells were used to train the classifier and 20% were used to check the model predictions against actual ground truths (Trm or Teq cell annotation labels). Probability values for how well the logistic regression classifier model (built from the gene scores) predicts Trm versus Teq cells were used as the prediction scores for generating Receiver Operating Characteristics (ROC) curves (sklearn.metrics.roc_curve) and calculating the Area Under the Curve (AUC) (sklearn.metrics.roc_auc_score).

Human single cell sequencing data analyses

Previously published single cell sequencing data of memory CD3⁺ T cells from various tissues and two donors were downloaded from the Sequencing Read Archive (SRA).⁵⁹ As done in the original study,⁷⁶ cells with between 10,000 and 25,000 UMI total counts, between 700 and 4,000 genes, and less than 25% mitochondrial genes were retained for the analysis. An initial filter removed genes expressed in fewer than 0.5% of the cells. The filtered dataset contained 53,140 cells. Prior to scVI modeling, the top 8,000 HVGs were selected by the Seurat v3 method¹⁰¹ as implemented by single-cell variational inference (scVI).¹⁰⁴ For generating a scVI latent space, each donor was treated as a ‘batch’ and a 20-dimensional latent space and default parameters were used. The scVI latent space was used to compute neighbor distances and connectivities, determine Uniform manifold approximation and projection coordinates and identify Leiden cell clusters. Cells were clustered at a resolution of 0.7. Gene enrichment scores for genes defined by mouse migration assays (genes shown in Figure S6G) were computed by visionpy. The percentage of Trm-like cells, based on the expression of Trm cell genes, was calculated for each tissue using UMAP-1 and UMAP-2 coordinates from the scVI representation of the data and the Trm gene scores. The UMAP coordinates were visualized on an XY coordinate graph in GraphPad Prism (GraphPad Software Inc.) to count and compute the percent of Trm-like cells in each tissue.

QUANTIFICATION AND STATISTICAL ANALYSIS

Two-way ANOVA with Sidak’s multiple comparison test was used if the effect of two independent variables were being considered among more than two sample groups. All statistical analysis was done in GraphPad Prism (GraphPad Software Inc.). In all experiments, P < 0.05 was considered significant. Statistical details for experiments are provided in the figure legends. Additionally, for the RNA-seq and CITE-seq data analyses, details regarding the selection of significantly differentially expressed features can be found the relevant METHOD sections.

Supplementary Material

2

Table S1. Differentially expressed genes (DEGs) from edgeR comparisons of CD69⁺ and CD69⁻ CD8⁺ T cell populations, related to Figure 3. (A) DEGs from a comparison of LCMV Arm specific Trm populations (CD69⁺ P14 cells from the spleen, LN, SI-IEL, SI-LP, and FRT) and Tcirc populations (Tcm (CD62L⁺CD69⁻) and Tem (CD62L⁻CD69⁻) P14 cells). (B) DEGs from a comparison of Tcirc cells and SLO-derived Cl13-specific CD8⁺ T cells (CD69⁺ H-2D^b restricted gp33 tetramer⁺ CD8⁺ T cells sorted from SLOs 50 days after chronic LCMV Cl13 infection of CD4⁺ T cell depleted mice). (C) DEGs from a comparison of Trm populations and recently stimulated CD8⁺ T cell populations. (D) DEGs from a comparison of Trm populations and all other populations shown in Figure 3A.

3

Table S2. scVI one versus all cluster differential tests, related to Figure 6. The number in the ‘comparison’ and ‘group 1’ columns represents the cluster number shown in Figure 6B.

4

Table S3. scVI differential tests comparing PR8- and Arm- generated Trm cells derived from either the lungs or spleen, related to Figures 6C–G. (A) scVI differential test results from a comparison of PR8- and Arm- generated Trm cells in cluster 3 which were predominately lung-derived. (B) scVI differential test results from a comparison of PR8- and Arm- generated Trm cells in cluster 6 which were spleen-derived.

5

Table S4: scVI differential intra- and inter-tissue test results and comparisons to published mouse and human Trm cell genes, related to Figures 6K–M. (A) scVI differential inter-tissue test results. (B) List of tissue associated DEGs that are increased (‘up’) in equilibrating (Teq) lung-derived (LU) cells relative to Teq spleen-derived (SPL) cells. Tissue associated DEGs were used in the analysis presented in Figure 6L, with some also depicted in Figure 6M. (C) scVI differential intra-tissue test results. (D) List of migration associated DEGs that are increased (‘up’) in LU-derived resident (Trm) relative to Teq cells. Migration associated DEGs were used in the analysis presented in Figure 6L, with some also depicted in Figure 6M. (E) Previously reported genes with increased expression in mouse-derived Trm cells. These genes were derived by contrasting CD103⁺ lung-, skin-, and SI-derived CD8⁺ memory T cells (putative Trm cells) to CD103⁻ CD8⁺ memory T cell populations from the spleen (putative non-Trm cells). See related Results text and Methods for reference. (F) Previously reported genes with increased expression in human-derived T cells presumed to be Trm cells based on their location (isolated from lung tissue). See related Results text and Methods for reference.

6

Table S5: scVI differential intra-tissue test results for each tissue in the data set (related to Figures S5A and S5B) and scVI differential test results from comparing clade 1 and clade 2 Trm cells (related to Figures S5F–H). (A) scVI differential test results from comparing LCMV Arm specific Trm and Teq cells derived from the female reproductive tract (FRT). (B) scVI differential test results from comparing LCMV Arm specific Trm and Teq cells derived from the liver. (C) scVI differential test results from comparing LCMV Arm and PR8 (combined analysis) specific Trm and Teq cells derived from the lungs. (D) scVI differential test results from comparing PR8 specific Trm and Teq cells derived from the medLN. (E) scVI differential test results from comparing LCMV Arm specific Trm and Teq cells derived from the mesLN. (F) scVI differential test results from comparing PR8 specific Trm and Teq cells derived from the salivary glands (SG). (G) scVI differential test results from comparing LCMV Arm specific Trm and Teq cells derived from the SI-IEL. (H) scVI differential test results from comparing LCMV Arm specific Trm and Teq cells derived from the SI-LP. (I) scVI differential test results from comparing LCMV Arm and PR8 (combined analysis) specific Trm and Teq cells derived from the spleen. (J) scVI differential test results from comparing clade 1 and clade 2 Trm cells (as shown in Figures S5F–H and in Figure 7). See Methods section ‘Differential expression testing’ for how significantly differentially expressed genes were identified from each scVI test.

7

Table S6: Clade specific totalVI differential test results and random forest gini scores, related to analyses in Figures 7A–I. (A) totalVI one versus all cluster differential tests, related to Figure 7. The number in the ‘comparison’ and ‘group 1’ columns represents the cluster number shown in Figure 7A. (B) totalVI differential test results from comparing Icos^hi Trm and Teq cells (as shown in Figure 7E; ‘clade 1’). (C) totalVI differential test results from comparing Icos^lo Trm and Teq cells (as shown in Figure 7E; ‘clade 2’). (D) Gini importance score for each feature (gene and protein) obtained from generating a random forest classifier model for Icos^hi Trm cells. (E) Gini importance score for each feature (gene and protein) obtained from generating a random forest classifier model for Icos^lo Trm cells.

8

Table S7: Clade specific scVI differential test results, random forest gini scores, and Icos^hi and Icos^lo specific Trm cell gene lists (with orthologous human gene names), related to Figure S6 analyses and used for the analysis of human CD3⁺ T cells in Figure S7. (A) scVI differential test results from comparing Icos^hi Trm and Teq cells (as shown in Figure 7E; ‘clade 1’). (B) scVI differential test results from comparing Icos^lo Trm and Teq cells (as shown in Figure 7E; ‘clade 2’). (C) Gini importance score for each feature (gene only) obtained from generating a random forest classifier model for Icos^hi Trm cells. (D) Gini importance score for each feature (gene only) obtained from generating a random forest classifier model for Icos^lo Trm cells. (E) Sixty genes with increased (‘up’) expression and thirty genes with decreased (‘down’) expression in Icos^hi Trm cells, both selected for their high Gini importance scores, along with their corresponding human orthologs. (F) Sixty genes with increased (‘up’) expression and thirty genes with decreased (‘down’) expression in Icos^lo Trm cells, both selected for their high Gini importance scores, along with their corresponding human orthologs.

HIGHLIGHTS.

1. Location, priming history, inflammation, and antigen persistence impact Trm phenotype

2. Migration assays reveal limitations of putative universal Trm cell gene signatures

3. Trm cells express some genes associated with location and others with tissue residence

4. A combination of markers more accurately correlates with residence than CD69 alone