Most immune cells function locally3,4. For example, T cells require contact with neighbouring cells to sense and eliminate infections or tumours5,6,7. To systematically evaluate the longevity of memory T cells in different locations, we transferred CD8+ T cells from P14 transgenic mice (hereafter, P14 CD8+ T cells) that were also positive for CD45.1 or Thy1.1 into naive female C57Bl/6J mice, followed by infection with the Armstrong strain of lymphocytic choriomeningitis virus (LCMV) (hereafter referred to as P14-immune chimeric mice). Nonlymphoid tissues are surveyed by resident memory T (TRM) cells that are vulnerable to cell death upon tissue digestion, which makes enumeration by this method inadequate8,9. To mitigate this issue, we visually enumerated P14 CD8+ T cells by quantitative immunofluorescent microscopy between 5 and 450 days after infection with LCMV in the tissues of 80–96 mice (Fig. 1a–e). The response to LCMV peaked within 5–9 days, before contracting. A stable population of memory T cells was established in the lymph node by 16 days after infection. We found considerable variation in the durability of nonlymphoid TRM cells, which were stable in salivary gland, decayed modestly in the small intestine (which was best modelled by biphasic decay) and underwent continued attrition in the uterus. We particularly noted attrition of TRM cells within the uterine endometrium, which underwent vacuolation and glandular atrophy in aged mice (Extended Data Fig. 1a–c) that coincided with infertility. Ageing was also associated with the development of prominent peripheral node addressin-expressing tertiary lymphoid organs in salivary glands (excluded from the numbers of TRM cells in Fig. 1c, Extended Data Fig. 1d, e). These data indicate that the longevity of memory CD8+ T cells varies by location, but that—in some compartments—these cells can persist indefinitely.

Fig. 1: Residence sustains organism-wide autonomous immune surveillance by T cells.
figure 1

ae, Using quantitative immunofluorescent microscopy (representative images shown in a), we enumerated P14 memory CD8+ T cells (P14) at 5–450 days after infection with LCMV in lymph node (b), salivary gland (c), uterus (d) and small intestine (e) of n = 80–96 biologically independent mice from 6 independent experiments. fj, Secondary LCMV-specific Tracfl/fl and wild-type (WT) memory T cells were established (Methods, Extended Data Fig. 2a) and mice were treated with tamoxifen to selectively ablate the TCR, as evaluated for peripheral blood leukocytes 12 days after treatment (f, g). h, To assess the effect of TCR ablation on T cell survival, the ratio of Tracfl/fl to wild-type CD8+ T cells was quantified 60 days after treatment with tamoxifen and normalized to pre-tamoxifen values from blood. i, j, Expression of residence markers CD69 (i) and CD103 (j) was compared between wild-type and TCRβ Tracfl/fl TRM cells. Data pooled from four independent experiments for a total of n = 10 mice, except FRT, n = 8 mice (g–j). FRT, female reproductive tract; IEL, intestinal epithelial lymphocytes; PBL, peripheral blood leukocytes; SG, salivary gland; SI, small intestine. ko, Congenically distinct parabiotic P14-immune chimeric mice were surgically separated (k) and, 260 days after infection with LCMV, n = 5–10 mice (n varies by tissue) were examined by quantitative immunofluorescent microscopy (QIM) to evaluate the durability of residence among extravascular P14 memory CD8+ T cells (l). After separation, the peripheral blood of n = 7 parabiotic P14-immune chimeric mice was serially monitored for reemergence of disequilibrium among P14 memory CD8+ T cells, which were subdivided on the basis of CD62L and KLRG1 expression (m, n). Three hundred and thirty-five days after infection with LCMV, CD62LKLRG1 donor and host P14 memory CD8+ T cells in blood were evaluated for CD43–1B11 (o) (representative flow cytometry, left; disequilibrium among CD62LKLRG1CD43+ P14 CD8+ T cells, right). TLO, tertiary lymphoid organ. Scale bars, 100 μm (a, k). Statistical significance was determined by two-tailed Wilcoxon matched-pairs signed-rank test (g, i, j) or two-tailed one-sample Wilcoxon test using 0 (h) or 50% (n) as a hypothetical mean. Data are mean ± s.e.m. or box plots showing median, interquartile range (IQR) and extremes.

Source data

TRM cell longevity is TCR-independent

Naive and central memory T cells recirculate between the blood and lymph nodes. Naive T cells depend on constitutive T cell receptor (TCR) signalling for survival, whereas central memory T cells do not10. Unlike central memory T cells, TRM cells share many properties with T cells that have undergone recent TCR stimulation11,12. To test whether the persistence of TRM cells depends on the TCR, we took advantage of UBC–CreERT2 × Tracfl/fl mice (hereafter, Tracfl/fl mice) so that we could genetically ablate the TCR by tamoxifen treatment in established memory T cells (Methods, Extended Data Figs. 2, 3). We compared wild-type Thy1.1+ and Tracfl/fl Thy1.1 cells within the same mice, and confirmed TCR ablation by staining for TCRβ and a failure to produce cytokines in response to ex vivo peptide stimulation (Fig. 1f, g, Extended Data Figs. 2a–e, 3a–c). No significant difference was observed between the longevity of wild-type and TCR-ablated TRM cells in all of the tissues we examined, nor was expression of CD69 and CD103 reduced in TCR-ablated memory T cells (Fig. 1h–j, Extended Data Figs. 2f–h, 3d). Thus, CD8+ TRM cells are not maintained by persistent viral antigen, self-peptide–MHC I complexes or cross-reactivity with microbial TCR ligands.

Durability of TRM cell organ surveillance

The defining characteristic of TRM cells is the absence of migration. However, migration experiments are typically short-term, which raises the question of whether TRM cell populations are gradually replenished by central memory T cells13 or slowly recirculate14,15,16. To address this issue, we generated mice containing CD45.1+ or Thy1.1+ memory P14 CD8+ T cells by infection with LCMV. Thirty days later, we conjoined mice by parabiosis, which resulted in the equilibration of circulating memory P14 CD8+ T cells in the blood of paired mice. Parabiotic mice were separated 30 days after conjoining, which allowed for the prolonged monitoring of residence (Fig. 1k). Two hundred and sixty days after infection with LCMV (200 days after separation), memory T cells were of host origin in most nonlymphoid tissues, which indicates that TRM cells autonomously dominate immunosurveillance for the lifespan of the mouse (Fig. 1l). Organized lymphoid structures—including tertiary lymphoid organs in salivary glands and Peyer’s patches in the small intestine—were surveyed by both recirculating memory T cells and TRM cells. However, extravascular memory T cells in the lung parenchyma demonstrated near-complete equilibration, regardless of CD69 expression (Fig. 1l, Extended Data Fig. 4a, b). Although TRM cells seed the lung shortly after infection, these data indicate that long-term pulmonary surveillance may depend on circulating memory T cells that enter the tissue, which has ramifications for the durability of T-cell-dependent protection against respiratory infections17,18. Circulating memory T cells also eventually made substantial contributions to the maintenance of memory in liver and kidney (Fig. 1l). Although these observations offer support for a centralized source for nonlymphoid memory in some locations13, for most tissues, residence is responsible for long-term surveillance after clearance of primary infections in mice housed under specific-pathogen-free laboratory conditions.

TRM cells gradually originate memory in blood

Primary infections induce CD62LKLRG1 long-lived effector memory T cells that patrol blood without entering lymph nodes. After surgical separation, we observed a gradual disequilibrium in blood in favour of host-derived P14 CD8+ T cells (Fig. 1l–n, Extended Data Fig. 5a, b). These data indicate that T cells that failed to equilibrate during parabiosis (that is, TRM cells) slowly join the blood circulation. Over time, these ex-TRM cells came to comprise between 15 and 30% of all memory P14 T cells in peripheral blood (Extended Data Fig. 5a, b), and were significantly enriched within the KLRG1 effector memory T cell subset (Fig. 1n). Additional profiling revealed that expression of the CD43 glycoform recognized by the 1B11 antibody clone further distinguished ex-TRM cells, consistent with expression of this glycoform on tissue-resident memory cells (Fig. 1o, Extended Data Figs. 5c, 6a, b). Thus, although TRM cells wane in some nonlymphoid tissues, they contemporaneously give rise to blood-borne effector memory T cells.

The TRM cell niche is expansible

Cell populations are often numerically fixed in size by cell-extrinsic regulators19. Cytokine (for example, IL-15) abundance, metabolite availability and constitutive TCR signalling all function to restrict T cell abundance in blood, but these factors do not axiomatically control T cell survival in tissues20,21,22. We hypothesized that the resident memory pool may circumvent the numerical constraints imposed on blood-borne T cells. Here we apply a reductionist approach that was previously used to test whether the circulating memory CD8+ T cell population in lymphoid organs as expansible or rigidly defined23 to the analysis of nonlymphoid tissues dominated by TRM cells (Fig. 1). We established P14 CD8+ T cell memory by infection with LCMV, as in Fig. 1a–e. Sixty days after infection with LCMV, we initiated a potent heterologous prime–boost vaccination regimen that established a large and broadly distributed memory CD8+ T cell population specific for the immunodominant nucleoprotein (N) epitope of vesicular stomatitis virus (VSV) (Fig. 2a, b). Concomitant with the expansion of N-specific memory T cells, there was a reduction in the relative frequency of pre-existing P14 CD8+ T cells as a fraction of total CD8+ T cells (Fig. 2c). However, the enumeration of P14 CD8+ T cells using quantitative immunofluorescent microscopy demonstrated that TRM cells were numerically preserved (Fig. 2d). In contrast to a previous study24, we found that new immunizations did not compromise the capacity of pre-existing memory CD8+ T cells to produce antiviral cytokines (Fig. 2e, Extended Data Fig. 7a). These data indicate that the immunological ‘space’ or carrying capacity for TRM cells is malleable, enabling the coexistence of established immunity and de novo immune responses.

Fig. 2: The CD8+ T cell compartment expands to accommodate new and pre-existing resident memory.
figure 2

ae, Sixty days after infection with LCMV, P14-immune chimeric mice were subjected to heterologous prime–boost (HPB) to generate a robust CD8+ T cell response against the N epitope of VSV (a, b). VSV-IND, Indiana serotype of VSV; VSV-NJ, New Jersey serotype of VSV; VV-N, recombinant vaccinia virus expressing VSV-N. Numerical abundancy, evaluated by flow cytometry (c) and quantitative immunofluorescent microscopy (d), and ex vivo functionality (e) of memory P14 CD8+ T cells in various tissues was compared and found to be not significantly different (P > 0.05) (d, e) between mice that received heterologous prime–boost and age-matched control mice. Data shown from 1 of 2 independent experiments with similar results, with n = 4 or 5 mice per group per experiment (c, e). Data pooled from 2 independent experiments for a total of n = 8–10 mice per group (d). fi, Sixty days after infection with LCMV, P14-immune chimeric mice were cohoused with mice obtained from pet shops (f) to facilitate microbial exposure. Memory P14 CD8+ T cells in nonlymphoid (n = 7–10, 16 mice) (h) and lymphoid (n = 5, 9 mice) (i) tissues were enumerated by quantitative immunofluorescent microscopy in 2 independent experiments; a representative example of small intestine is shown in g. CHD, cohoused; SPF, specific-pathogen free. j, P14-immune chimeric mice were cohoused with mice obtained from pet shops for >60 days and conjoined via parabiosis for 30 days, at which point the disequilibrium of memory P14 CD8+ T cells was assessed in n = 9 or 10 mice. LN, lymph node. Scale bars, 100 μm. Statistical significance was determined by two-tailed Mann–Whitney U test. *P = 0.0172, **P = 0.0028, ***P = 0.0002 (h); *P = 0.0159, **P = 0.0040 (i). Data are mean ± s.e.m., or box plots showing median, IQR and extremes.

Source data

To assess whether pre-existing TRM cells were also durably maintained after physiological exposure to diverse microorganisms, we took advantage of a recently developed cohousing model that effects marked changes in innate and adaptive immunity. Cohousing specific-pathogen-free mice with mice bought from pet shops results in the transmission of bacteria, viruses, fungi and helminths, and induces extensive immune activation and adoption of an immune system that is more similar to that of humans25. Sixty days after infection with LCMV, P14-immune chimeric mice were cohoused with mice obtained from pet shops, for two months (Fig. 2f, g). In most tissues, we observed little attrition of P14 CD8+ T cells, which indicates that circulating memory CD8+ T cells and CD8+ TRM cells can persist in the face of substantial exposure to microorganisms (Fig. 2h, i). Of note, the resident property of memory P14 CD8+ T cells in tissues was also preserved in mice that were subjected to cohousing with mice obtained from pet shops (Fig. 2j), which suggests that residence itself is not an artefact of artificially clean mouse husbandry. Small-intestinal TRM cells constituted an exception, and were nearly sixfold-less abundant after cohousing. The attrition of intestinal TRM cells may reflect the saturation of anatomical space, heightened damage-associated molecular patterns that induce cellular toxicity9 or alterations to the microbiome that modulate local survival cues. In two infection models that generate either lung- or skin-resident memory CD8+ T cells, pre-existing memory was preserved after cohousing (Extended Data Fig. 8a–g). Thus, in most tissues, heterologous prime–boost vaccination or physiological exposures to microorganisms did not induce erosion of TRM cells—instead, tissues accommodated more resident cells.

The immune system is an expansible tissue component

Homeostasis balances cellular self-renewal with cell death, and maintains organ size in adult organisms. This process preserves organ function while preventing cancer. The expansible nature of the T cell compartment raised questions about whether the immune system in toto is of a fixed size or exhibits pliancy in proportion to microbial experience. Cohousing induced a durable increase in nearly all types of leukocytes in blood, as well as persistent enlargement of lymph nodes and spleen (Fig. 3a–c). Both memory CD4+ and CD8+ T cells increased after cohousing, and even the so-called ‘innate’ populations of the immune system demonstrated an ‘adaptive’ ability to durably expand in size (Fig. 3a–c, Extended Data Fig. 9a, b). Extensive analysis using quantitative immunofluorescent microscopy revealed that 5–20% of most nonlymphoid tissues of specific-pathogen-free mice were composed of white blood cells. However, the frequency of immune cells was significantly and durably expanded after cohousing, which indicates that the immune system occupies a considerable fraction of visceral and mucosal organs, is flexible in size and is capable of long-lived adaptation in relation to exposure to microorganisms (Fig. 3d, e).

Fig. 3: Tissue pliancy enables immune expansion after microbial conditioning.
figure 3

a, b, The abundance of total leukocytes (a) and immune subpopulations (b) in peripheral blood was compared between n = 21 B6 mice cohoused for 90 days with mice obtained from pet shops and n = 21 B6 mice housed in conventional SPF conditions. NK, natural killer. c, Representative lymphoid organs (lymph node, left; spleen, right) from SPF and CHD mice. d, e, The frequency of immune cells contributing to total organ cellularity (e; quantified by the enumeration of total nucleated cells) in SPF and CHD mice was evaluated by quantitative immunofluorescent microscopy (d). CHD mice were cohoused for 90–120 days (ce). Data are pooled from 4 independent experiments, n = 5–16 mice per group, except pancreas (n = 3–5 mice) (d, e). LI, large intestine. Scale bars, 1 cm (c), 100 μm (d). EpCAM staining shown for uterus and colon only; DAPI staining was excluded for uterus (d). Statistical significance was determined by two-tailed Mann–Whitney U test. *P = 0.0209 (naive CD8), *P = 0.0336 (B cells), **P = 0.0022, ****P < 0.0001 (a, b); *P = 0.0418, **P = 0.0047 (kidney), **P = 0.0043 (heart) or as shown in graph (e). Data are mean ± s.e.m., or box plots showing median, IQR and extremes.

Source data

Residency typifies tissue immunity

Memory T cells achieve durable immune surveillance of the entire organism through prolonged residence in most nonlymphoid tissues (Fig. 1). Tissue residency has previously been demonstrated for several types of immune cell, primarily in immunologically naive or single-infection mouse models26,27. We asked whether residence is a common mechanism used by the immune system to achieve broad immunological surveillance, by performing parabiosis surgery of cohoused B6 mice that have diverse microbial experiences (Extended Data Fig. 10a). Leukocytes (distinguished by CD45) equilibrated completely within peripheral blood of parabiotic mice (Extended Data Fig. 10b), but tissue-resident immune populations occupied—and often dominated—most organs (Fig. 4a). Over one month, many CD4+ and CD8+ T cells of the adaptive immune system, as well as macrophages, innate lymphoid cells and natural killer cells of the innate immune system, stably occupied nonlymphoid tissues (Fig. 4b, Extended Data Fig. 10c–f). Consistent with their rapid turnover, granulocytes in tissues relied on continuous replenishment from blood (Fig. 4b, Extended Data Fig. 10g). B cells, which differentiate into the antibody-secreting cells of the immune system, largely equilibrated between parabionts (Fig. 4b, Extended Data Fig. 10h). Overall, residence was a shared feature that was exhibited by many adaptive and innate immune cell types in mice that were exposed to diverse microorganisms.

Fig. 4: Tissue residence typifies organism-wide immune surveillance.
figure 4

a, b, Congenically distinct B6 mice were cohoused with mice obtained from pet shops. After >60 days, mice were conjoined by parabiosis for 28–32 days. The equilibration of total leukocytes (a) and immune populations (b) was evaluated in peripheral blood and tissues. Lung and kidney populations exclude intravascular leukocytes. In the heat map (b), lung macrophages are divided into alveolar (top left) and interstitial (bottom right) populations; intestinal populations were isolated from lamina propria (LP). Grey denotes populations that were not detectable. The heat map shows the mean percentage value of host-derived cells. Data are pooled from four independent experiments. Typically, n = 8–14 mice with exceptions for some tissues or cell types (Extended Data Fig. 10) (a, b). B, B cells; CD4 and CD8, memory T cells (CD44+ PD1); eos, eosinophils; ILC, innate lymphoid cells; Mo/Ma, monocytes and macrophages; neut, neutrophils. Data are mean ± s.e.m.

Source data


The immune system has defied classification as a commonly recognized organ system, but is partially captured by the skeletal system (which includes the bone marrow), the cardiovascular system (which includes blood cells) and the lymphatic system (which includes the secondary lymphoid organs)28,29. This framework excludes many immune cells in the body, particularly those that are most responsible for regional immune surveillance and effector functions. Our study generalizes two features of immunity outside of dedicated lymphoid organs: (1) most immune cells stably reside in—rather than recirculate through—tissues and (2) the size of the immune system durably adapts to microbial experience to accommodate additional cells. Although parenchymal cell populations are numerically constrained to maintain organ homeostasis, the adaptive flexibility of the immune system allows leukocytes to comprise a considerable fraction of the body—up to a quarter of visceral organs in microbially experienced mice. Analyses that are limited to blood and lymphoid organs have emphasized the renewable and migratory aspects of the immune system30,31,32,33. Our study, which initially focused reductively on T cells, shows that resident T cell immunity can autonomously endure over time, without displacement by subsequent inflammation, infection or competition for a fixed immunological niche. Given that most infections and tumours develop in nonlymphoid organs, these are relevant findings for immunization strategies that seek to harness T cell immunity34,35. In contrast to expectations, we observed that, after a primary infection, most long-lived blood-borne effector memory T cells are ex-TRM cells (Fig. 1); recent reports indicate that reactivated TRM cells can contribute even further to circulating memory populations36,37. The phenomenon of expansible residence that we observed for T cells extended to most populations of immune cells. Although immune responses develop in centralized sites, our data indicate that nonlymphoid organs provide a flexible reservoir for the long-term preservation of adaptive and innate immunity. Given these findings, it may be reasonable to conceptualize the immune system as its own organ system, albeit one that consists of a diverse network of motile sensory cells that are durably integrated throughout the entire body, and that is permissive to considerable plasticity in cell number, composition and distribution.



Female, SPF C57BL/6J (CD45.2+ B6) and B6.SJL-PtprcaPepcb/BoyJ (CD45.1+ B6) mice at 6–8 weeks of age were purchased from The Jackson Laboratory. For cohousing experiments, female, SPF CD45.2+ and CD45.1+ B6 mice at 6–8 weeks of age were purchased from Charles River Laboratories. Female pet shop mice (age not provided by vendor) were purchased from pet shops in the greater Minneapolis–St Paul metropolitan area. Cohousing of SPF mice with sex-matched pet shop mice was performed as previously described25, within the University of Minnesota biosafety level 3 facility. The following housing conditions were regulated: ambient temperature (20.0–23.3 °C), humidity (30–70%) and light/dark cycling (14-h on/ 10-h off). Tracfl/fl mice38 were fully backcrossed to UBC–CreERT2 mice39 (JAX stock no. 007001) to generate UBC–CreERT2 Tracfl/fl mice (provided by K. Hogquist). Male and female UBC–CreERT2 Tracfl/fl mice at 6–10 weeks of age were used in tandem with age- and sex-matched Thy1.1+ B6 mice. CD45.1+ sex-matched mice at 6–12 weeks of age were used as recipient mice. P14 CD8+ T-cell-transgenic, OT-1 CD8+ T-cell-transgenic, Thy1.1+ B6 and CD45.1+ B6 mouse strains were maintained in-house. All mice were used in accordance with guidelines established by the Institutional Animal Care and Use Committee at the University of Minnesota. The University of Minnesota Institutional Review Board approved all protocols used.

Adoptive transfers and infections

P14-immune chimeric mice were generated by intravenous adoptive transfer of 5 × 104 P14 CD8+ splenocytes into naive female B6, mice and infection with 2 × 105 plaque-forming units (PFU) of the Armstrong strain of LCMV via intraperitoneal injection on the subsequent day. Alternatively, P14-immune chimeric mice were generated by intravenous adoptive transfer of 5 × 104 P14 CD8+ splenocytes into naive female B6 mice and intranasal infection with 500 PFU PR8-gp33 recombinant influenza virus (provided by R. Langlois) on the subsequent day. OT-1-immune chimeric mice were generated by intravenous transfer of 5 × 104 OT-1 CD8+ splenocytes into naive female B6 mice and infection with 106 PFU VSV–OVA the following day. For experiments using Tracfl/fl mice, Tracfl/fl mice and wild-type Thy1.1+ B6 mice were intraperitoneally infected with 2 × 105 PFU of LCMV. After 30 days, 107 lymphocytes—isolated from secondary lymphoid organs—were intravenously transferred into naive CD45.1+ B6 mice. On the subsequent day, CD45.1+ recipients were intraperitoneally infected with 106 PFU of LCMV. To generate primary polyclonal memory with Tracfl/fl mice, naive Tracfl/fl mice and wild-type Thy1.1+ B6 mice were killed, and CD8+ T cells were enriched from secondary lymphoid organs via negative selection per the manufacturer’s protocol (Stem Cell Technologies). A total of 2 × 106 enriched cells were stimulated per well in flat-bottom 12-well plates with anti-CD3ε (clone 145-2C11, 10 μg ml−1) (Bio X Cell) and rB7-1/Fc chimeric protein (0.8 μg ml−1) (R&D Systems) immobilized on the surface in the presence of 5 U ml−1 IL-2 (R&D Systems) with 10 ng ml−1 mouse recombinant IL-12 (R&D Systems) as previously described40. After 3 days of in vitro activation, 107 CD8+ T cells isolated from Tracfl/fl mice and wild-type Thy1.1+ B6 mice were intravenously co-transferred into naive CD45.1+ B6 mice. For heterologous prime–boost immunization, three viruses were administered by intravenous injection at 30-day intervals in the following order, as previously described23: (1) 106 PFU of VSV, New Jersey serotype; (2) 2 × 106 PFU of recombinant vaccinia expressing the VSV nucleoprotein; and (3) 107 PFU of VSV, Indiana serotype.

Quantitative immunofluorescence microscopy

Collected mouse tissues were fixed in 2% paraformaldehyde for 2 h, followed by overnight cryoprotection in 30% sucrose solution at 4 °C. Sucrose-treated tissue was embedded in optimal cutting temperature (OCT) compound and frozen in a chilled isopentane bath. Alternatively, collected mouse tissues were directly embedded in OCT compound and snap-frozen in a chilled isopentane bath. Studies of the salivary gland focused exclusively on the submandibular gland. Frozen tissue blocks were sectioned at 7 μm in a Leica cryostat. Sections were stained with primary and secondary antibodies, counterstained with DAPI or SYTOX Green to detect nuclei, and immunofluorescence microscopy was performed using a Leica DM6000 B microscope. Monoclonal anti-mouse antibodies, used at a 1:100 dilution unless noted otherwise, were: CD8α (53-6.7), CD8β (YTS156.7.7), CD45 (30-F11), CD45.1 (A20), Thy1.1 (OX-7) (1:1,000), B220 (RA3-6B2), EpCAM (G8.8) (1:500) and PNAd (MECA-79), all from BioLegend. Polyclonal goat anti-mouse collagen type IV antibody (1:200) from MilliporeSigma and secondary bovine anti-goat IgG (H+L) antibody (1:300) from Jackson ImmunoResearch were used. Images were processed using FIJI software41 and cell enumeration was performed manually as previously described8, or using ImageJ scripts developed in house.

Leukocyte isolation and phenotyping of cells

An intravascular staining method was used to discriminate between cells within the vasculature and those within the parenchyma of tissues. Three minutes before being killed, mice were intravenously injected with either 3 μg of biotinylated- or fluorophore-conjugated CD8α (53-6.7) antibody or 2 μg of fluorophore-conjugated CD45 (30-F11) antibody42. Tissues were collected and leukocytes isolated as previously described8. Studies of the salivary gland focused exclusively on the submandibular gland. Isolated leukocytes were surface-stained with the following monoclonal anti-mouse monoclonal antibodies at a 1:100 dilution, unless otherwise noted: CD4 (GK1.5), CD5 (53-7.3), CD8α (53-6.7), CD8β (53-5.8), CD11c (N418), CD43 (1B11), CD44 (IM7), CD45 (30-F11), CD45.1 (A20), CD45.2 (104), CD62L (MEL-14), CD64 (X54-5/7.1), CD69 (H1.2F3), CD103 (2E7), Thy1.1 (OX-7) (1:250), B220 (RA3-6B2), F4/80 (BM8), Ly6C (HK1.4) (1:400), Ly6G (1A8), NKp46 (29A1.4), PD-1 (29F.1A12), CX3CR1 (SA011F11) and CXCR3 (CXCR3-173), from BioLegend; CD11b (M1/70), CD19 (1D3), NK1.1 (PK136), Siglec-F (E50-2440) and TCF-1 (S33-966) (1:50), from BD; CD3ε (145-2C11), CD127 (A7R34), and KLRG1 (2F1) and TCRβ (H57-597) from Tonbo Biosciences. Cell viability was determined using Ghost Dye Violet 510 or Ghost Dye Red 780 (Tonbo Biosciences) (1:300). To identify VSV N-specific CD8+ T cells, leukocytes were stained with H-2Kb/N (MHC class I tetramer) (1:200), conjugated to PE. To identify gp33-specific CD8+ T cells, leukocytes were stained with H-2Db/gp33 (MHC class I tetramer) (1:200), conjugated to APC. Staining for intracellular transcription factors and proteins was performed using a transcription factor staining buffer kit (Tonbo Biosciences) with monoclonal anti-mouse antibodies: T-bet (4B10) from BioLegend; Eomes (Dan11mag), FOXP3 (FJK-16s), GATA-3 (TWAJ), Ki67 (SolA15) (1:400) and, RORγt (B2D) from ThermoFisher Scientific. The stained samples were acquired using LSRII or LSR Fortessa flow cytometers (BD) and analysed with FlowJo software (BD). Neutrophils were distinguished by expression of CD11b and Ly6G. Eosinophils were identified by Siglec-F expression. Innate lymphoid cell, natural killer cell, B cell and monocyte and macrophage populations were distinguished after excluding lineage-positive cells using combinations of CD3, CD5, CD19, B220 and Ly6G, and then using recommended lineage-defining markers as previously described43,44. Monocytes were further subdivided into classical and patrolling populations on the basis of Ly6C expression. Lung macrophages were subdivided into alveolar and interstitial populations on the basis of Siglec-F and CD11b expression.

Parabiosis and separation surgeries

Parabiosis and separation surgeries were performed with age-matched female mice as previously described45, with some modifications. For surgeries, anaesthesia to full muscle relaxation was achieved using avertin (250 mg kg−1) by intraperitoneal injection, and surgical site preparation included betadine application to surgical site in a gradually enlarging circular pattern. For parabiosis, corresponding lateral aspects of mice were thoroughly shaved with electric clippers from about 1 cm superior to the shoulder and about 1 cm inferior to the hip. Excess hair was wiped off with alcohol preparation pads. Following surgical site preparation, matching skin incisions were made about 0.5 cm superior to the shoulder and about 0.5 cm inferior to the hip. Subcutaneous fascia was bluntly dissected to create around 0.5 cm of free skin. Dorsal and ventral skins of adjacent mice were approximated by interrupted horizontal mattress stitches with 3-0 Prolene suture and overlying surgical wound clips. To separate parabiotic mice, pre-existing suture and wound clips were removed. Excess hair was shaved and wiped off with alcohol preparation pads. After surgical site preparation, a longitudinal incision was made with sharp scissors lateral to the initially conjoined skin. Newly formed fascia was gently detached with a pair of curved forceps. The superior and inferior aspects of the skin of each mouse were then approximated by running stitch with a single 4-0 Vicryl suture. For analgesic treatment, mice received preoperative subcutaneous buprenorphine (2 mg kg−1) and postoperative subcutaneous bupivacaine (2 mg kg−1) and carprofen (5 mg kg−1). Mice were kept on heating pads during and after surgery, and their recovery was monitored continuously.

Tamoxifen administration

Tamoxifen was dissolved in corn oil at 37 °C with shaking overnight to a working concentration of 20 mg ml−1. Working stocks were freshly prepared for each experiment. For CreERT2 induction, tamoxifen was administered to mice intraperitoneally at a dose of 75 mg kg−1 every 24 h over 5 consecutive days.

In vitro stimulation assays

Isolated lymphocytes were incubated at 37 °C for 4 h in stimulation media with or without gp33–41 peptide (0.2 μg ml−1) or phorbol myristate acetate and ionomycin (cell stimulation cocktail, ThermoFisher Scientific). Stimulation medium consisted of RPMI 1640, 10% FCS, 2 mM l-glutamine, 100 U ml−1 penicillin, 100 mg ml−1 streptomycin, 50 mM 2-mercaptoethanol and brefeldin A (GolgiPlug, BD). Intracellular staining for cytokines was performed using the Cytofix/Cytoperm kit per the manufacturer’s directions (BD) with anti-mouse antibodies: IFNγ (XMG1.2) from BioLegend and TNF (MP6-XT22) from ThermoFisher Scientific.

Quantification and statistical analysis

No statistical methods were used to predetermine sample size. Mice were randomly assigned to experimental groups and investigators were not blinded to allocation during experiments and outcome assessment. Specific statistical tests, sample size (n) and P values can be found in figure legends. Individual data points represent biological replicates. All statistical tests were two-tailed and, generally, non-parametric tests were used to test for significance (Mann–Whitney U test for unpaired samples and Wilcoxon matched-pairs signed-rank test for paired samples). All statistical analysis was done using Prism (GraphPad). For all experiments, a P value <0.05 was considered significant. Mean and s.e.m. are used to represent the centre and dispersion, unless otherwise stated. Nonlinear regression analysis using Prism (GraphPad) was used to model memory T cell population kinetics data, using data points between 30 and 450 days after infection. Model constraints imposed that decay plateau = 0 and a positive rate constant. An exponential decay model was fit to tissue populations if R2 > 0. Either one-phase or two-phase exponential decay models were selected after comparison using extra sum-of-squares F test and the Akaike information criterion.

Reporting summary

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