Brain-resident memory T cells represent an autonomous cytotoxic barrier to viral infection

To study the generation and function of bT_RM, we infected mice intracerebrally (i.c.) with a genetically engineered, attenuated lymphocytic choriomeningitis virus (LCMV) variant (rLCMV; Pinschewer et al., 2003). As previously shown (Pinschewer et al., 2010), i.c. administered rLCMV rapidly drained to the periphery and triggered a potent peripheral antiviral CD8⁺ T cell response, which resolved the brain infection within ten days (Fig. 1 A). Thereafter, T cell numbers in the brain contracted, but a population of T_M persisted in stable number and frequency from 6 wk after infection onwards (Fig. 1, B and C). At that time point, most persisting T_M cells expressed CD69, a prototypic T_RM marker. CD69⁺ brain T_M expressed high Bcl-2 levels, and many were Granzyme B–positive (59 ± 3%), whereas most of them were specific for the immunodominant LCMV epitope NP_396-404 (82 ± 5%). However, only about half of these cells (50 ± 10%) expressed CD103 (Fig. 1 D and Fig. S1). Both CD103⁺ and CD103⁻ CD69⁺ T_M subsets persisted in stable relative proportions over time (Fig. 1 E). Analogously to the endogenous CD8⁺ T cell pool, both CD103⁺ and CD103⁻ brain T_M subsets were found in adoptively transferred CD8⁺ TCR transgenic P14 or OT-1 cells, respectively, after i.c. infection with rLCMV variants encoding the respective cells’ cognate epitopes (Fig. 1 F). Furthermore, CD103⁺ and CD103⁻ OT-1 brain T_M were generated after intrathecal infection with a modified vaccinia virus (MVA; Rothhammer et al., 2014) expressing OVA (Fig. 1 G). CD69 expression correlated with exclusion from staining with i.v. administered fluorescent antibody (Anderson et al., 2014; Fig. 1 H) and extravascular CD8⁺ T cells identified by immunostaining on brain sections expressed CD69 (95 ± 3%; n = 3 mice). Furthermore, most brain CD69⁺ T_M cells were refractory to depletion by peripherally administered depleting anti-CD8 antibody (Fig. 1 I). Altogether, this demonstrated that the vast majority of CD69⁺ T_M cells in brain were bona fide T_RM (from now on referred to as bT_RM).

To investigate the protective role of bT_RM upon brain reinfection, we generated rLCMV memory mice by either i.v. or co-infection (i.c. + i.v.; Fig. 2 A). Co-infection was used to assure similar generation of circulating T_M. Indeed, both infection routes led to similar frequencies of D^b-NP_396-404 LCMV-specific T_M in peripheral blood (Fig. 2 B). However, in contrast to bT_RM generated upon i.c. infection, only limited numbers of T_M cells were present in the brain after i.v. infection with rLCMV (Fig. 2, C and D). These latter cells were susceptible to depletion by peripherally administered anti-CD8a antibody (unpublished data) demonstrating that T_M detectable in the brain after i.v. rLCMV infection are in equilibrium with the circulation and do not represent bona fide bT_RM. In addition, brain circulating T_M generated by i.v. infection failed to up-regulate CD103 and contained a lower percentage of D^b-NP_396-404–specific cells than bT_RM generated by i.c. infection (Fig. 2 E). Next, rLCMV memory mice with bT_RM (i.c. + i.v. rLCMV co-infection) or without bT_RM (i.v. rLCMV infection) were i.c. challenged with pathogenic WT LCMV (LCMVwt), which shared the immunodominant NP_396-404 epitope with rLCMV (Fig. 2 F). When analyzed 3 d later, rLCMV memory mice with bT_RM exhibited significantly lower LCMV virus load in the brain than rLCMV memory mice with only circulating T_M. This indicated that the generation of bT_RM provided an advantage during reinfection.

We next analyzed the spatial distribution of bT_RM on brain sections by immunohistochemistry. bT_RM were preferentially detected at brain surface structures, such as meninges and choroid plexus, around the ventricles or at anatomical borders between different brain regions (Fig. 3 A). Only a small fraction of bT_RM was in direct contact with vascular endothelium (14 ± 3%; Fig. 3 A), but most bT_RM localized in close proximity to vessels and brain surface structures, or both (Fig. 3 B). Quantification of CD103⁺ and CD103⁻ bT_RM in the analyzed areas did not reveal a correlation of CD103 expression with their localization (Fig. 3 C). Interestingly, a consistent fraction of bT_RM expressed proliferation marker Ki-67 (9 ± 1%) long after virus clearance (Fig. 3 D). bT_RM at brain surface structures expressed Ki-67 more frequently than bT_RM localized in brain parenchyma (Fig. 3 E). Previous studies have shown that IL-7– and IL-15–dependent longevity and homeostatic proliferation are maintaining central memory T cells by Stat5 signaling (Burchill et al., 2003; Willinger et al., 2005; Hand et al., 2010). Indeed, a fraction of brain T cells showed Stat5 phosphorylation (26 ± 7%). Similarly to Ki-67⁺ T cells, we observed an increased proportion of P-Stat5⁺ T cells at brain surface structures (Fig. 3 F). Using histocytometry (Fig. S2), we could confirm that meninges contained a higher frequency of P-Stat5⁺ T cells than internal brain structures (Fig. 3 G). Altogether, this indicates that Stat5-driven homeostatic proliferation contributes to the maintenance of the long-lived bT_RM population preferentially at brain surface structures.

To assess the contribution of circulating CD8⁺ T cells to virus control in the brain, we treated a group of rLCMV memory mice with anti-CD8 depleting antibody before LCMVwt i.c. challenge (Fig. 4 A). rLCMV memory mice with bT_RM and circulating T_M contained significantly more T cells in the brain after challenge than mice with only circulating T_M (Fig. 4, B and C). As expected (Schenkel et al., 2013), the numbers of brain-infiltrating T cells in the former group were significantly reduced when circulating T_M were depleted. This indicated that about half of the brain CD8⁺ T cells after LCMVwt challenge of nondepleted mice originated from a recruitment of circulating T_M, whereas the other half represented bT_RM and their progeny. Furthermore, 47 ± 13% of bT_RM expressed proliferation marker Ki-67 after LCMVwt i.c. challenge (Fig. 4 D), indicating that a substantial fraction of bT_RM entered the cell cycle upon antigen reencounter in the CNS. During LCMV i.c. challenge, both CD103⁺ and CD103⁻ bT_RM expanded in similar proportions and showed equal potential for Ki-67 up-regulation (Fig. 4, E and F). Intriguingly, rLCMV memory mice with bT_RM but lacking circulating T_M, cleared virus from the brain equally efficiently as mice with both bT_RM and circulating T_M (Fig. 4, C, G, and H). This suggested that recruitment of circulating T_M was not required for bT_RM-mediated virus clearance from the brain. Conflicting reports exist about T_RM dependency on CD4⁺ T help in different organs (Jiang et al., 2012; Laidlaw et al., 2014). To address the role of CD4⁺ T cells for the generation and function of bT_RM, we depleted CD4⁺ T cells either during primary i.c. infection with rLCMV or performed a combined depletion of circulating CD4⁺ and CD8⁺ T cells during LCMVwt i.c. challenge. bT_RM numbers, as well as bT_RM-mediated rapid virus clearance during LCMVwt i.c. challenge, was not affected in the absence of CD4⁺ T cell help during bT_RM generation or challenge with LCMVwt (Fig. 5, A–L). Furthermore, to exclude that recruited NK cells contribute to virus clearance, we depleted NK cells in addition to circulating T cells during LCMVwt i.c. challenge. However, virus clearance was unimpaired in animals containing bT_RM but not circulating T and NK cells (Fig. 6, A–E), demonstrating that LCMVwt control relied on bT_RM.

Next, we aimed to investigate if presentation of cognate antigen is necessary or if bystander activation and expression of antiviral genes in surrounding tissue is sufficient for bT_RM-mediated viral clearance. Therefore, we challenged rLCMV memory mice with a genetically engineered LCMV variant (LCMV-NP_N400S), in which the MHC-I anchor residue of the immunodominant epitope NP_396-404 was mutated to abrogate MHC binding (Johnson et al., 2015; Fig. 7 A), thus preventing MHC-I–restricted recognition of this virus by the vast majority of bT_RM (Fig. 1 D and Fig. 2 E). Accordingly, bT_RM expansion was reduced when LCMV-NP_N400S was used for challenge instead of LCMVwt (Fig. 7, B and C). In further contrast to LCMVwt i.c. challenge, bT_RM failed to clear the brain from LCMV-NP_N400S within 3 d. This was evident in increased numbers of infected cells (Fig. 7, B and D) and heightened LCMV RNA loads (Fig. 7 E). Interestingly, antiviral interferon-induced transmembrane protein 3 (Ifitm3), a marker of T_RM-induced antiviral state (Ariotti et al., 2014), was similarly induced after infection with LCMVwt or LCMV-NP_N400S, respectively (Fig. 7 F). Furthermore, infection with either of the two viruses induced up-regulation of Granzyme B (Fig. 7, G–H), as well as Ki-67 expression (Fig. 7, G and I) in bT_RM. This indicated that the capacity of bT_RM to autonomously clear brain virus infection depended on epitope-specific target recognition. Conversely, the virtually uniform activation of bT_RM after viral challenge occurred in a largely antigen nonspecific fashion, as D^b-NP396-404-tetramer⁺ bT_RM up-regulated GzmB (96 ± 1%) and Ki-67 (42 ± 3%) during i.c. challenge with LCMV-NP_N400S (unpublished data).

T cell–mediated viral clearance from the brain can be achieved by cytolytic, i.e., perforin or Fas-dependent mechanisms, as well as by noncytolytic, cytokine-dependent pathways (Griffin, 2003; Pinschewer et al., 2010; Herz et al., 2015). To investigate cytokine production and cytotoxic potential of bT_RM, we depleted memory mice of circulating CD8⁺ T cells and challenged them with LCMVwt i.c. (Fig. 8 A). We analyzed CD107a surface exposure as indicator of degranulation of cytotoxic granule release and production of IL-2, IFN-γ, and TNF upon in vitro peptide stimulation (Fig. 8 B and not depicted). In the absence of LCMVwt challenge and without peptide restimulation, bT_RM did not spontaneously produce detectable amounts of cytokines and did not degranulate. After LCMVwt i.c. challenge, bT_RM exhibitedIFN-γ production (9.7 ± 1.5%) and surface exposure of CD107a (9.0 ± 0.5%) in the absence of further ex vivo stimulation. After peptide stimulation, most bT_RM produced high amounts of IFN-γ in combination with surface exposure of CD107a, which was significantly higher in bT_RM from LCMVwt i.c. challenged animals (59.0 ± 2.4%) than from unchallenged controls (36.4 ± 3.1%; Fig. 8 B). In contrast, we detected comparably few bT_RM that coproduced IL-2 or TNF, and we failed to detect expression of FasL on bT_RM (unpublished data). Thus our analysis identified IFN-γ–producing bT_RM with cytotoxic granule release as main effector cell population after LCMVwt i.c. challenge.

Considering IFN-γ as the main effector cytokine expressed by bT_RM, we next tested bT_RM function in IFN-γ knockout (GKO) mice. Of note, IFN-γ is dispensable for the clearance of primary infection with attenuated rLCMV (Pinschewer et al., 2010) and bT_RM generation was not impaired in GKO mice (Fig. 8 C). We depleted circulating CD8⁺ T cells in IFN-γ–deficient (GKO) and IFN-γ competent (control) rLCMV memory mice and subsequently challenged them with LCMVwt i.c. (Fig. 8, D–G). Before LCMVwt i.c. challenge, GKO rLCMV memory mice displayed slightly higher bT_RM numbers, but the fold expansion of bT_RM after virus challenge was similar in GKO and control animals (Fig. 8, D and E). This indicated that the cell-intrinsic reactivation capacity of bT_RM was unimpaired in GKO mice. Conversely, the number of infected cells (Fig. 8, D and F) and LCMV RNA loads (Fig. 8 G) 3 d after LCMVwt i.c. challenge were significantly higher in GKO memory mice than in control memory mice.

Based on the notion that bT_RM up-regulated the cytotoxic effector molecule Granzyme B after LCMV i.c. challenge (Fig. 7, G and H), we next investigated whether cytolytic effector functions (Kägi et al., 1994) contributed to bT_RM-mediated virus clearance. We thus generated perforin-deficient (PKO) rLCMV memory mice and depleted circulating CD8⁺ T cells (Fig. 8, H–M), as before (Fig. 8 A). Perforin was dispensable for the generation of bT_RM (Fig. 8 H). Furthermore, after LCMVwt i.c. challenge, perforin-deficient bT_RM showed unimpaired up-regulation of Granzyme B (Fig. 8 I) and secondary expansion (Fig. 8, J and K). Nevertheless, PKO memory mice exhibited significantly higher numbers of infected cells (Fig. 8, J and L) and increased viral RNA loads (Fig. 8 M) than control memory mice. This demonstrated that bT_RM cells exert rapid IFN-γ production and perforin-dependent cytotoxicity, both of which made essential contributions to the rapid control of viral infection in the brain.

i.c. infection of naive immunocompetent mice with LCMVwt results in fatal choriomeningitis, which is the result of antiviral CD8⁺ T cell–mediated immunopathology (Cole et al., 1972). T_RM have been shown to amplify immune responses in nonlymphoid tissues during rechallenge with cognate antigen (Schenkel et al., 2013, 2014a). Here, we observed analogous findings in the brain (Fig. 4 B), raising the possibility that bT_RM could enhance immunopathological consequences of brain virus infection (Oehen et al., 1991). We therefore investigated how the presence of bT_RM impacts the disease after LCMVwt i.c. challenge. For that, we depleted LCMV-specific circulating T_M in rLCMV memory mice using anti-CD8 antibody 6 wk before LCMVwt i.c. challenge (Fig. 9 A). This timing enabled the reconstitution of the peripheral compartment with naive CD8⁺ T cells before LCMVwt challenge (unpublished data). Analogous to the experiment in which anti-CD8 depletion and challenge had been performed simultaneously (Fig. 4, B and D), the presence of bT_RM resulted in enhanced T cell numbers in the brain when circulating memory CD8⁺ T cells were present (Fig. 9, B and C). However, irrespective of the presence or absence of circulating memory CD8⁺ T cells, brain LCMV loads as assessed 3 d after LCMVwt i.c. challenge were significantly reduced in mice with bT_RM compared with mice with only circulating T_M (Fig. 9, B, D, and E). As expected (Oehen et al., 1991), mice with only circulating T_M did not develop disease, demonstrating that the observed peripheral virus-specific T_M recall response can afford antiviral protection (Fig. 9 F). Furthermore, circulating T_M-depleted rLCMV memory mice without bT_RM developed fatal choriomenigitis, which had kinetics comparable to unimmunized control mice. Most importantly, six out of seven rLCMV memory mice with bT_RM were protected from disease even though circulating T_M cells had been depleted. This underscored the self-sufficient nature of bT_RM-mediated antiviral protection. Next, we investigated whether bystander activation of bT_RM (Fig. 7) was sufficient to protect animals from fatal immunopathology. We therefore depleted rLCMV memory mice of circulating T_M and challenged them with either LCMVwt or LCMV-NP_N400S i.c. after reconstitution of circulating CD8⁺ T cells. As expected, mice challenged with LCMV-NP_N400S i.c. did not mount a D^b-NP_396-404–specific CD8⁺ T cell response, but developed a CD8⁺ T cell response specific for the D^b-GP_33-41 epitope (data not shown), to which they had not been previously exposed (rLCMV does not encode for LCMV-GP_33-41). Unlike rLCMV memory mice challenged with LCMVwt i.c., challenge with LCMV-NP_N400S i.c. resulted in the development of fatal choriomeningitis with disease kinetics comparable to unimmunized control mice (Fig. 9 G). Altogether, this demonstrated that bT_RM immunity was antigen-specific and provided rapid autonomous elimination of virus-infected cells from the brain and thereby protected from fatal immunopathological disease.

The identification of T_RM as a separate lineage of long-lived memory T cells has fundamentally reshaped our concepts of cell-mediated immunity to infection. With this study, we establish bT_RM as an autonomous first line of adaptive immune defense against viral infection of the brain. bT_RM were found after acute infection with rLCMV or a replication-deficient modified vaccinia virus strain in mice, attesting to the general validity of our findings. bT_RM showed strong proliferative potential and protected against fatal viral infection in an organ-autonomous manner, independent of the recruitment of circulating CD8⁺ memory T cells.

Similar to other nonlymphoid organs (Steinert et al., 2015), CD69⁺ bT_RM represented 90% of CD8⁺ T cells in the brain. Contrary to brain infection with vesicular stomatitis virus (Wakim et al., 2010), we found in two independent models that CD103 was expressed only on a proportion of bT_RM. This may be a result of differences in isolation and quantification methods, as has been described for other organs (Steinert et al., 2015). We do not exclude the possibility that CD103⁺ and CD103⁻ bT_RM may be functionally different; however both subsets persisted in stable proportions over time, exhibited an analogous preference for retention at surface-lining structures of the brain, and showed equal proliferative capacity upon reactivation. Although bT_RM show very limited survival and proliferative capacity in vitro (Wakim et al., 2010), they seem to possess self-renewing capacity in vivo and exhibit high proliferative capacity after reactivation. Similar to central memory T cells (Burchill et al., 2003; Willinger et al., 2005; Hand et al., 2010), the population of long-lived bT_RM seems to be maintained by Stat5-driven homeostatic proliferation, preferentially at brain surface structures, such as meninges and periventricular areas. However, the exact nature of a supposed T_RM maintaining microenvironment near tissue surfaces that was also observed in other organs, i.e., in the skin (Gebhardt et al., 2009), remains to be investigated.

In naive hosts, the task of rapidly detecting invading pathogens and alerting the adaptive immune defense generally falls to the innate immune system (Iwasaki and Medzhitov, 2015). The resulting secretion of antiviral effector cytokines, such as type I interferons, is crucial to limiting virus spread until adaptive immune responses are induced and can accomplish viral clearance (McNab et al., 2015). T_RM from peripheral organs other than the brain have been described to subserve a similar immunological function, i.e., to alert circulating T cell subsets to invading pathogens and to induce antiviral genes in the tissue surrounding the site of initial infection (Ariotti et al., 2014; Schenkel et al., 2014a). These functions were reported to be largely dependent on inflammatory cytokines, such as IFN-γ. Similar to T_RM from other organs, we show here that bT_RM share the capacity to secrete IFN-γ upon antigen reencounter. However, they largely do that in combination with the release of cytotoxic granules, and perforin-mediated cytotoxicity was essential for rapid bT_RM-mediated virus control. One may thus speculate that IFN-γ–deficient bT_RM fail to provide sufficient noncytolytic antiviral function, or fail to prepare the site of infection for lytic effector function as a result of impaired up-regulation of MHC I expression on infected cells (Bergmann et al., 2003). Whereas both possibilities are not mutually exclusive and can contribute to the observed outcome, our data strongly suggest that a direct interaction of bT_RM with infected cells rather than a generalized antiinflammatory state is necessary for bT_RM-mediated rapid viral elimination. This conclusion is based on our observation that bT_RM can be activated in a bystander manner, i.e., in the absence of cognate antigen and solely as the result of the inflammatory environment (Kohlmeier et al., 2010; Soudja et al., 2012; Raué et al., 2013), yet under such circumstances fail to rapidly control the virus. This does not, however, exclude the possibility that bystander activation may contribute to a viral load reduction in the brain. Furthermore, depending on the pathogen type and infection scenario, bT_RM-driven induction of IFN-γ response genes may be sufficient to limit early pathogen spread or even to cleanse viruses from nonrenewable cells such as neurons (Griffin, 2003; Herz et al., 2015).

Neurotropic viruses of hematogenous origin frequently infect the CNS secondary to systemic infection. Neuroinvasion typically occurs through the brains outer or inner linings (McGavern and Kang, 2011), namely the meninges or the ependyma, to subsequently spread into the brain parenchyma. The resulting inflammation manifests as meningitis or even encephalitis and can be life-threatening (Gilden et al., 2007). Our model of intracranial rLCMV immunization recapitulates the spread from the brain linings into the parenchyma albeit it does not reproduce neuroinvasion secondary to systemic infection. Conversely, its attenuated behavior enabled us to study solid bT_RM immunity without the caveats of life-threatening complications from primary brain infection.

Although we noted a protective function of bT_RM in infection, T_RM–mediated activation and recruitment of inflammatory cells to the site of infection can, however, also have detrimental effects such as those described in the context of skin contact hypersensitivity (Gaide et al., 2015). In organs with limited regenerative capacity, such as the CNS, excessive inflammation can induce collateral damage (immunopathology). In this regard, severe immunopathological disease is one of the most dangerous complications of viral infections in the brain, and depending on pathogen loads, circulating T_M can even accentuate disease rather than prevent it (Oehen et al., 1991). Therefore, not only infections of the CNS but also the resulting immune responses must be tightly controlled. The functioning of bT_RM as a border patrol, which rapidly detects and autonomously eliminates the first few virus-infected cells, thus bypasses the need for antigen drainage to lymph nodes and reactivation of T_M cells inside secondary lymphoid organs. It has been shown for other organs that T_RM cells use a continuous migratory behavior, enabling few cells to survey large organ structures and readily detect their cognate antigen (Ariotti et al., 2012). The positioning and preferential maintenance of bT_RM at surface-associated brain structures and in proximity to vessels as described in this study therefore matches the specific immunological challenge and need of the brain to limit viral spread from the circulation and brain linings into the parenchyma and to rapidly eliminate the initially infected cells. Thereby, bT_RM can balance pathogen control against excessive inflammation.

Based on the present observations, we propose to adapt the concept of immunological privilege of the brain. With identification of the bT_RM compartment, the brain undoubtedly maintains powerful immunological forces to fight recurrent microbial intruders. This observation stands in stark contrast to the several layers of immunomodulatory mechanisms, including constitutively low expression of MHC molecules (Neumann et al., 1995), which are in place to limit inflammation in the brain (Galea et al., 2007). Thus, pathogens that have previously invaded the brain appear to be specifically excluded from this immunological privilege and are instead rapidly and efficiently eliminated by bT_RM. It is tempting to speculate that bT_RM immunity must be particularly potent because the blood–brain barrier prevents antibody access to the brain, and thus is largely excluding this organ from protection by the humoral arm of immunological memory.

C57BL/6 WT, IFN-γ deficient (GKO; Dalton et al., 1993), perforin-deficient (PKO; Kägi et al., 1994), OT-1 TCR transgenic (Hogquist et al., 1994), and P14 TCR (Pircher et al., 1989) transgenic mice were bred and lodged under P2 conditions in the animal facilities of the University Medical Centre of Geneva and Munich. Sex- and age-matched mice between 6 and 12 wk of age were used for experiments. All animal experiments were authorized by the cantonal veterinary office of Geneva and performed in agreement with the Swiss law for animal protection.

Recombinant LCMV strains were generated according to established methods (Pinschewer et al., 2003). The following virus strains were used: recombinant LCMV encoding for the glycoprotein of vesicular stomatitis virus (VSV) instead of its own glycoprotein (rLCMV/INDG, abbreviated rLCMV throughout); rLCMV encoding for a chimeric glycoprotein of VSV containing the leader sequence of LCMV glycoprotein with the GP33 epitope of LCMV (rLCMV-GP33); and recombinant LCMV encoding for OVA (rLCMV-OVA). Genetically engineered WT LCMV (LCMVwt, Armstrong strain) and LCMV mutant virus (LCMV-NP_N400S) lacking the CTL-specific epitope NP_396-404 were generated as described previously (Johnson et al., 2015). Viruses were produced, titrated, and administered to mice as previously described (Flatz et al., 2006). For transient virus infection in the brain, 10⁴ PFU rLCMV was administered i.c. i.v. infected control animals were injected i.c. with vehicle. Memory and naive mice were challenged with 10³ PFU LCMVwt (Armstrong strain) i.c. Incidence of choriomenigitis was determined by occurrence of general malaise and hind limb cramps, and severely diseased animals were immediately sacrificed. Brain virus load was analyzed by quantitative RT-PCR for LCMV-NP (McCausland and Crotty, 2008). Brain total RNA was isolated from deparaffinized tissue sections using RNeasy Mini kit (QIAGEN), cDNA was synthesized using iScript cDNA synthesis kit (Bio-Rad Laboratories), and relative expression of LCMV S segment was determined against GAPDH as housekeeping gene (QIAGEN) using iQ SYBR Green Supermix (Bio-Rad Laboratories).

Replication-deficient MVA-OVA was generated as described previously (Staib et al., 2004). 2 × 10⁶ IU MVA-OVA was administered intrathecally (i.th.) as previously described (Rothhammer et al., 2014). In some experiments, OT-1 or P14 TCR transgenic CD8⁺ T cells were adoptively transferred into naive C57BL/6 mice 24 h before i.c. or i.th. infection.

For depletion of peripheral CD4⁺ T cells, CD8⁺ T cells, or NK cells, mice were treated i.p. twice over a 3-d interval with 250 µg anti-CD4 antibody (YTS191), 450 µg anti-CD8a depleting antibody (clone YTS169.4), or 500 µg anti-NK1.1 depleting antibody (clone PK136), respectively. Depletion of T cells or NK cells was confirmed by flow cytometry showing T cell or NK cell frequencies <0.1% during LCMVwt i.c. challenge. After antibody-mediated CD8⁺ T cell depletion and subsequent reconstitution by naive CD8⁺ T cells, frequencies of D^b-NP_396-404–specific CD8⁺ T cells in rLCMV memory mice were reduced from 1.4 ± 0.3% to 0.06 ± 0.02%, which was not significantly above technical background in uninfected naive control mice (0.005 ± 0.004%).

Mice were injected i.v. with 3 µg PE-labeled anti-CD8b (YTS156.7.7) according to previously described protocols (Anderson et al., 2014). 3 min later, peripheral blood samples were taken, and mice were anesthetized and perfused with PBS. Brains were dissected and processed immediately (<10 min after antibody injection) for leukocyte isolation (see Flow cytometry).

For immunohistochemical bright field staining of mouse tissue, brains and spleen were prepared and fixed in Hepes-glutamic acid buffer-mediated organic solvent protection effect (HOPE, DCS Innovative) fixative as previously described (Bergthaler et al., 2007) and embedded in paraffin. Upon inactivation of endogenous peroxidases, tissue sections were incubated with rat anti–mouse CD8a (YTS169.4) or rat anti-LCMV NP (VL-4; Battegay et al., 1991). Bound primary antibodies were visualized with biotin-labeled anti–rat antibody and streptavidin-peroxidase staining method using polymerized 3,3′-diaminobenzidine (all reagents from Dako; haemalaun counterstaining of nuclei).

For immunofluorescence staining of mouse tissue, cryosections were fixed with 2% PFA for 5 min and unspecific binding blocked (PBS/1% BSA/2% FCS). Sections were incubated with FITC-labeled Armenian hamster anti–mouse CD103 (2E7) and rat anti–mouse CD8a (YTS169.4), followed by rabbit anti-FITC (Invitrogen) antibody. Bound antibodies were visualized with Alexa Fluor 488–labeled goat anti–rabbit and Cy3-labeled goat anti–rat secondary antibodies (Jackson ImmunoResearch Laboratories). Nuclei were stained with DAPI (Invitrogen).

For immunofluorescence staining of paraffin embedded tissue, mice were transcardially perfused with 4% PFA. After antigen retrieval and unspecific binding blocking (PBS/10% FCS), sections were incubated with rat anti–human CD3 (CD3-12), rat anti–mouse CD8 (clone 4SM15), goat anti–mouse CD69 (AF2386), rabbit anti-VWF (ab6994), rabbit anti–Ki-67 (ab66155), and rabbit anti–phospho-Stat5 (tyr694, C71E5). Bound antibodies were visualized with Alexa Fluor 488 or Alexa Fluor 568 goat anti–rabbit tyramide signal (TSA) amplification (Thermo Fisher Scientific), Cy2- or Cy3-labeled goat-anti–rat, or Cy5-labeled goat anti–rabbit secondary antibodies (Jackson ImmunoResearch Laboratories). Nuclei were stained with DAPI (Invitrogen).

Immunostained sections were scanned using Pannoramic Digital Slide Scanner 250 FLASH II (3DHISTECH) in 200× magnification. Quantifications were performed manually using Pannoramic Viewer software (3DHISTECH) blinded to the experimental setup of histological samples. For representative images, white balance was adjusted and contrast was linearly enhanced using the tools levels, curves, brightness, and contrast in Photoshop CS6 (Adobe). For histocytometry analysis, fluorescence-immunostained brain sections were scanned using a high-dynamic range camera (16 bit).

For histocytometry and density plots, CD3⁺ or CD8⁺ cells, respectively, were detected on scanned slides applying a custom-programmed script in Cognition Network Language (Definiens Developer XD software). For analysis of distance, a grayscale distance map was computed to encode the closest distance of each detected CD3⁺ or CD8⁺ cell to the brain surface and vessels (detected by von-Willebrand Factor staining) and mean values of distance maps were extracted. For histocytometry, each CD3⁺ cell size (area), as well as mean and max grayscale values for CD3, P-Stat5, and DAPI signals, were extracted. The resulting CSV files were converted to .fcs format using an open source program (FlowJo). The converted file was analyzed using FlowJo X. Co-staining of anti-CD3 and anti–phospho-Stat5 was gated based on control staining without primary P-Stat5 antibody.

For staining, the following antibodies were used: anti–Bcl-2 (BCL/10C4), anti-CD3e (145-2C11), anti-CD8a (53–6.7), anti-CD11a (M17/4), anti-CD11b (M1/70), anti-CD44 (IM7), anti-CD45.1 (A20), anti-CD45.2 (104), anti-CD45R/B220 (RA3-6B2), anti-CD49a (HMα1), anti-CD49b (DX5), anti-CD49d (9C10), anti-CD69 (H1.2F3), anti-CD103 (2E7), anti-CD107a (1D4B), anti-CD127 (A7R34), anti–E-Cadherin (36/E-Cadherin), anti-FasL (MFL3), anti–Granzyme B (GB11), anti–IFN-γ (XMG1.2), anti–IL-2 (JES6-5H4), anti-Integrinβ7 (FIB27), anti–Ki-67 (SolA15), anti-KLRG1 (2F1), anti-NK1.1 (PK136), and anti-TNF (MP6-XT22). For detection of virus-specific CD8⁺ T cells, D^b-NP_396-404-tetramer (TCmetrix) or K^b-B8R_20-27-tetramer (provided by D.H. Busch, Institute of Medical Microbiology, Immunology and Hygiene, Technical University Munich, Munich, Germany) were used.

Peripheral blood samples and splenocytes were collected in FACS-Buffer (10% FCS, 10 mM EDTA, and 0.01% NaN₃ in PBS). Peripheral blood erythrocytes were lysed using BD FACS lysing solution (BD). For the preparation of brain leukocytes, mice were anesthetized and transcardially perfused with PBS. Brains were minced, digested with Collagenase/DNaseI (Roche), and homogenized using 70-µm cell strainers (BD). Leukocytes were separated using a discontinuous Percoll gradient (30%/70%). Surface staining was performed with directly labeled antibodies and tetramers in FACS buffer. Isolated CD8⁺ T cell numbers were quantified using TruCOUNT tubes (BD) or AccuCheck Counting Beads (Invitrogen) as previously described (Steinbach et al., 2013). Intracellular staining of Ki-67 was performed using FoxP3/Transcription Factor Staining Buffer Set (eBioscience) according to manufacturer’s instructions. For ex vivo staining of Granzyme B, cells were fixed and permeabilized using commercial permeabilization buffer set (BioLegend). To assess degranulation and intracellular cytokine production, brain leukocytes were cultured for 5 h in the presence of 5 µg/ml FITC-labeled anti-CD107a antibody, monensin, and brefeldin A. Cells were stimulated in vitro with 1 µM NP_396-404 peptide or left unstimulated, respectively. Cells were harvested and dead cells were excluded from the analysis using Zombie NIR Fixable Viability kit (BioLegend). Cells were fixed and permeabilized using commercial permeabilization buffer set (BioLegend), followed by intracellular staining for cytokines, respectively. Flow cytometric samples were acquired on a Gallios cytometer (Beckman-Coulter) equipped with three lasers (blue, 488 nm; red, 633 nm; and violet, 405 nm) using appropriate filter sets and compensation controls. Gates were assigned according to appropriate control populations. A CyAn ADP analyzer (Beckman-Coulter) was used for experiments involving MVA-OVA.

Analysis was performed using GraphPad 6.0 (Prism). To assess significant differences between single measurements of two groups, unpaired or paired two-tailed Student’s t test was used, whereas differences between single measurements of more than two groups was assessed by one-way ANOVA, followed by post hoc analyses of selected pairs of datasets using Sidak’s multiple comparisons test or all pairs of datasets using Tukey’s multiple comparisons test to control for multiple comparisons. To assess significant differences between groups with multiple measurements, two-way repeated-measures ANOVA was performed followed by Dunnet’s multiple comparisons test against control group. For comparison of disease incidence, Log rank (Mantel-Cox) test was used. P < 0.05 was considered significant, and P < 0.01 was considered highly significant, whereas P > 0.05 was considered statistically not significant.

Fig. S1 shows the gating strategy used for analysis of brain-derived leukocytes. Fig. S2 shows the histocytometry data processing and the gating strategy.

We thank Prof. D.H. Busch (Technical University Munich) for providing K^b/B8R20 tetramers.

D. Merkler holds stipendiary professorships of the Swiss National Science Foundation (No. PP00P3_152928) and is supported by the Klaus-Tschira Foundation, Helmut Horten Foundation, and Gebert-Rüf Foundation. D. Merkler, K. Steinbach, and N. Page are supported by the Swiss MS Society. T. Korn is funded by SyNergy, ERC consolidator grant, DFG SFB 1054, TR 128. I. Drexler is supported by Deutsche Forschungs Gesellschaft GRK 1949.

The authors declare no competing financial interests.

Author contributions: K. Steinbach and I. Vincenti performed the experiments and analyzed the data. M. Kreutzfeldt, N. Page, I. Wagner, and A. Muschaweckh assisted in mouse experiments, data collection, and analysis. D. Pinschewer and T. Korn were involved in study design. I. Drexler provided viral vector. K. Steinbach and D. Merkler designed the studies. K. Steinbach, D. Pinschewer, and D. Merkler wrote and revised the manuscript. All authors discussed results and commented on the manuscript.