Sex as a biological variable. For all experiments, animals (described below) were distributed into different experimental groups with a similar ratio of males and females. No differences in titers or expression of key molecules of interest (i.e., CD69, CD103, cytokines or chemokines) were observed by sex.

Humanized mouse model. NSG mice and NSG-A2 mice were maintained in ventilated, specific pathogen-free conditions at the Institute of Experimental Immunology, University of Zurich. Newborn pups were reconstituted with human CD34+ hematopoietic progenitor cells (HPCs), derived from human fetal liver tissue (HFL) as previously described (45). For each experiment, animals were reconstituted from a single HFL donor and distributed into different experimental groups with similar reconstitution levels of human immune cell populations.

EBV infection and course of experiment. Mice were infected with 1 × 105 Raji Green units (RGU) of B95-8 EBV-expressing luciferase under the control of the latent EBNA2 locus (Luc-EBV) (45, 46). Since patients with IM have been found to shed approximately 1 × 106 to 1 × 107 viral DNA equivalents per mL of saliva for prolonged periods after infection, this dose represents the transfer of approximately 10–100 μL of saliva (79). Luc-EBV producer cells were provided by Wolfgang Hammerschmidt (Helmholtz Zentrum, Munich, Germany). Since initial histological examination of the nasal sinuses of uninfected animals revealed heterogeneous establishment of NALT structures (Supplemental Figure 1, A, B, and D), we pretreated animals with SEB (62.5 ng/μL in PBS) applied i.n. to mimic bacterial colonization of mucosal surfaces driving postnatal NALT development in mice (48, 49). A similar approach (i.n. application of Propionibacterium acnes bacteria) has been used to establish of NALT structures in NALT-deficient CXCR5–/– mice (83). Both lymphocyte recruitment (Supplemental Figure 1, B, C, F, and G) and reproducibility of i.n. infections were significantly improved by SEB pretreatment (Supplemental Figure 1E). EBV infections were performed i.p. or i.n. 2 weeks following SEB pretreatment. For infections, Luc-EBV was applied by 100 μL injection (for i.p. infection) or 20 μL slowly pipetted and divided between both nostrils (for i.n. infection), respectively. Mice infected i.n. were anesthetized with aerosolized isoflurane prior to administration of EBV. Following infection, mice were monitored regularly for 4–6 weeks. Intravascular staining was performed by i.v. injection of 6 μg fluorescent antibody-conjugated CD45 (HI30, BioLegend) diluted in 100 μL PBS 3 to 5 minutes before euthanasia. In each experimental group, 3–6 biological replicates were tested.

In vivo bioluminescence imaging. The progression of EBV infection was monitored longitudinally every week and quantitatively measured by in vivo bioluminescence imaging with the IVIS Spectrum Imaging System (PerkinElmer). Animals were anesthetized by aerosolized isoflurane and injected i.p. with 150 mg/kg D-Luciferin (Promega) diluted in PBS 10 minutes before imaging. Mice were placed inside the IVIS imaging box and imaged dorsally and ventrally. Representative images were acquired 10–15 minutes after injection for each mouse throughout each experiment to illustrate viral spread within the host. Images for quantification were captured at various time points before the luminescent signal reached the saturation intensity and analyzed with Living image 4.3.1 software (PerkinElmer). Regions of interest (ROI) were set to include the regions with luminescent signal in mice and photon flux (p/s) of light emitted per second within the ROI was measured as the readout.

T cell depletion experiments. CD8+ T cells were depleted weekly, via i.p. injection beginning at week 3 after infection, with the monoclonal antibody against human CD8 (clone OKT-8; BioXCell). The initial injection was 75 μg, followed by 2 injections of 50 μg (week 3 after infection) and 3 injections of 50 μg (week 4 after infection). Mice with over 10% residual CD3+CD4– T cells within lymphocytes in blood and spleen were considered not depleted and excluded from analysis. CD103+ cells were targeted by a single i.p. injection at week 3 with 150 μg anti-CD103 (Ber-ACT8, Ultra-LEAF purified, BioLegend).

Quantification of EBV DNA genome in blood and tissue. Total DNA from whole blood was extracted using NucliSENS easyMag (Biomerieux), while solid organs were processed with DNeasy Blood & Tissue Kit (QIAGEN), according to manufacturer’s instructions. TaqMan (Applied Biosystems) real-time PCR was used to quantify EBV DNA as previously described with modified primers for the BamH1 W fragment (5′-CTTCTCAGTCCAGCGCGTTT-3′ and 5′-CAGTGGTCCCCCTCCCTAGA-3′) and a fluorogenic probe (5′-FAM CGTAAGCCAGACAGCAGCCAATTGTCAG-TAMRA-3′). Viral titer concentration was calculated using a standard curve of the international WHO EBV standard, resulting in titers measured in EBV IU. All samples were performed in duplicates and measured on either ViiATM 7 Real-Time PCR System (Thermo Fisher Scientific) or ABI Prism 7300 Sequence Detector (Applied Biosystems). Samples below the lower limit of quantification (LLOQ) of 173 IU/mL were defined as negative for EBV DNA.

Normalization of viral loads was performed to adjust for differences in EBV infection levels between experiments with humanized mice that were reconstituted from different HPC donors.

Humanized mouse sample collection. Peripheral blood cells were obtained from animals by tail vein bleeding or terminal heart puncture. Whole blood was lysed twice with 1× ACK lysis buffer for 5 minutes, followed by washing with PBS. Splenocytes were prepared as described above with one time lysis only. NALT, salivary gland, lung, and liver tissues were mechanically disrupted into small pieces and enzymatically digested in 2 mL of digestion buffer (1 mg Collagenase D [Roche] and 0.2 mg DNase I [Roche] in 2 mL RPMI or DMEM) at 37°C for 30 minutes with agitation. Dissociated tissues were then passed through a 70 μm cell strainer. NALT was lysed once with 1× ACK lysis buffer for 3 minutes. Other tissues were partly centrifuged in a discontinuous Ficoll gradient (salivary gland [SG], lung) or Percoll gradient (liver, 40% and 70%; Sigma-Aldrich) for 20 minutes at 1,000g using a Sorvall ST 40R centrifuge (Thermo Fisher Scientific). Cells aggregating at the interface between gradient layers were harvested and washed twice with PBS. Bone marrow cells were flushed out of the femur by 5 second pulse centrifugation at 6,000g. Cells were washed with PBS and passed through a 70 μm cell strainer if necessary. Cells from blood and spleen were counted using the AcT diff Analyzer (Beckman Coulter) to aliquot the optimal number of cells for staining and calculation of the total cell numbers for different experimental purposes. Calculation of total cell numbers for NALT samples was done using Accucheck counting beads (Thermo Fisher Scientific) as previously described (84).

Human sample collection. Human tonsil samples were received from anonymous organ donors in the New York Presbyterian Hospital (New York, New York, USA); 6 tonsils from 4 different individuals were collected immediately after surgery from patients undergoing tonsillectomy for chronic inflammation. Further tonsil samples were received from tonsillectomies performed at the University Hospital of Zurich. Samples were frozen and stored in liquid nitrogen until flow cytometric phenotyping. Samples were rested overnight before stimulation experiments. For comparison of resident versus nonresident populations, tonsil samples were divided into CD69+ and CD69– subsamples using biotinylated anti–human CD69 antibody (FN40, BioLegend), and separation was performed using anti-Biotin Microbeads (Miltenyi Biotec, 130-090-485) according to manufacturer’s protocol. Gating during later flow cytometric analyses was chosen according to unstimulated controls from the same samples (Supplemental Figure 6).

Flow cytometry and antibody and MHC pentamer labeling. For surface staining, cells were incubated with anti–human FcR-block (Miltenyi Biotec, 130-059-901), live/dead (Zombie NIR/Aqua, BioLegend, 423106 and 423102), and chemokine-receptor antibodies (listed below) for 15 minutes at room temperature. Cells were subsequently incubated with surface marker antibodies for 30 minutes at 4°C. For intracellular and intranuclear labeling, surface staining was followed by fixation and permeabilization with the Foxp3/Transcription Factor Staining Buffer Set (eBioscience, 00-5523-00) according to manufacturer instructions. Antibodies for intracellular targets were incubated for 1 hour at room temperature. All samples were acquired using DIVA software (BD Biosciences) on LSRFortessa/FACSymphony (BD Biosciences) instruments, and analysis was performed using FlowJo software (Tree Star Inc.).

Antibodies used in this study include the following: CD3 (UCHT1, BD Biosciences, BUV661), CD4 (SK3, BD Biosciences, BUV469 / S3.5, Thermo Fisher Scientific, PE-Cy5.5), CD8 (SK1, BioLegend, BV650), CD19 (HIB19, BioLegend, PE-Cy5), CD27 (LG.3A10, BioLegend, BV650), CD39 (A1, BioLegend, BV711), CD45 (HI30, BD Biosciences, BUV395), CD45RA (HI100, BioLegend, APC-Fire750/BV785), CD62L (DREG-56, BioLegend, PE-Cy7/SK11, BD Biosciences, BV510), CD69 (FN50, BioLegend, BV421), CD103 (Ber-ACT8, BioLegend, BV711), CD107a (H4A3, BD Biosciences, FITC), CD127 (IL-7R) (A019D5, BioLegend, Alexa Fluor 700/PE-Dazzle 594), CD279 (PD-1) (EH12.1, BD Biosciences, BUV737), CD335 (NKp46) (9E2, BioLegend, BV510), CD56 (B159, BD Biosciences, APC), Blimp-1 (IC36081R, R&D, Alexa Fluor 647/6D3, BD Biosciences, PE-CF594), CCL5 (VL1, BioLegend, PE), granzyme B (GB11, BioLegend, PE-CF594/Alexa Fluor 700), HLA-DR (G46-6, BD Biosciences, PE-CF594/BV605), IFN-γ (4S.B3, BD Biosciences, BV786), KLRG1 (13F12F2, eBioscience, Alexa Fluor 488), Perforin (dG9, BD Biosciences, PerCP-Cy5.5), TNF-α (Mab11, BD Biosciences, PE-Cy7), anti–mouse TCR-β (H57-597, BioLegend, BV510/605)

EBV-pentamers specific for BMLF1 (GLCTLVAML, HLA-A*02:01, PE/APC) and LMP2 (CLGGLLTMV, HLA-A*02:01, APC/PE) were purchased from Proimmune. Pentamers were added 15 minutes prior to surface antibody labeling during FcR-blocking and live/dead staining.

Human TCR-Vβ repertoire analysis using flow cytometry. Human TCR-Vβ repertoire analysis was performed using antibodies against the most frequent TCR-Vβ variants (covering about 70% normal human TCR-Vβ repertoire, as reported by Beckman Coulter Beta Mark TCR vBeta Repertoire Kit; catalog IM3497) and following previously published nomenclature (85). The antibodies were provided by Roland Martin (University of Zurich, Zurich, Switzerland). In brief, frozen splenocytes were stained as described above using surface antibodies against following TCR-Vβ segments. The following were used: VB3 (CH92, Beckman Coulter, FITC), VB4 (WJF24, Beckman Coulter, PE), VB5.2 (36213, Beckman Coulter, FITC), VB5.3 (3D11, Beckman Coulter, PE), VB7.1 (ZOE, Beckman Coulter, PE), VB7.2 (REA677, Miltenyi Biotec, APC), VB9 (FIN9, Beckman Coulter, PE), VB12 (VER2.32.1.1, Beckman Coulter, PE), VB13.1 (IMMU222, Beckman Coulter, PE), VB13.6 (JU74.33, Beckman Coulter, FITC), VB14 (CAS1.1.3, Beckman Coulter, PE), VB16 (TAMAYA1.2, Beckman Coulter, FITC), VB18 (BA62.6, Beckman Coulter, PE), VB20 (ELL1.4, Beckman Coulter, PE), VB21.3 (IG125, Beckman Coulter, FITC), VB22 (IMMU546, Beckman Coulter, FITC), and VB23 (AF23, Beckman Coulter, PE)

After initial analysis, we identified the top 6 TCR-Vβ clones expanded in splenocyte-derived CD8+ TEMs in 3 individuals. These were chosen to stain splenocytes and lymphocytes from the NALT of the same mouse for the following TCR-Vβ segments to identify any overlap of clonotypes: VB1 (BL37.2, Beckman Coulter, PE), VB2 (RE654, Miltenyi, PE-Vio770), VB5.1 (LC4, eBioscience, APC), VB8 (56C5.2, Beckman Coulter, FITC), VB11 (RE559, Miltenyi, PE-Vio770), and VB17 (E17.5F3.15.13, Beckman Coulter, FITC).

Generation of EBV-specific TCR transgenic T cells. EBV-specific TCR generation and adoptive T cell transfer experiments were performed as previously described (45, 46). Briefly, for each specificity, a total of 200,000 TCR+CD3+ T cells was transferred i.v. into HPC donor–matched recipient mice and monitored longitudinally during the course of EBV infection.

scRNA-Seq. scRNA-Seq of up to 10,000 sorted TEMs was performed using 10X Genomics 3′-kit (v3.1) and Illumina Novaseq S1. Sequencing was performed by the Functional Genomics Center Zurich. Analyses were done with the guidance of NEXUS Zurich and following the OSCA handbook and publication (86). Reads were aligned to a combined reference genome comprising human and EBV reference using the STAR incorporated within the Cell Ranger software (87). Using R 4.0.5, data were quality controlled and normalized to cell cycle genes and within samples before further analysis (88). With corrected values and Poisson residual principal component analysis (PCA), t-distributed stochastic neighbor embedding (t-SNE), and UMAP calculations were done. Subsequent analyses were performed using BioConductor 3.12 unless stated otherwise (86). Unsupervised clustering was performed using the R-implementation of the Phenograph-algorithm (52), and cellular trajectories were calculated with the TSCAN algorithm (53).

Ex vivo stimulation experiments. Ex vivo isolated and rested CD8+ T cells or transduced T cells from splenocytes were incubated with R10 (RPMI 1640 + 10% FBS + 1% penicillin/streptomycin + L-glutamine + 20 U/mL of IL-2; Thermo Fisher Scientific) alone or R10 containing PMA/ionomycin for 2 hours, followed by the addition of Brefeldin A and Monensin and further incubation for an additional 3 hours. Cells were stained for intracellular cytokines and acquired on a BD FACSymphony. For CD107a labeling, this antibody was added at the very start of coculture. Gating was chosen according to unstimulated controls of the same samples (Supplemental Figure 6).

Histology. Whole mouse skulls were fixed using 4% formalin before cutting with a diamond blade. Pieces were decalcified in EDTA at a pH of 9.0 for 20–30 minutes at 100°C and embedded in paraffin. Histological staining was performed by an outside laboratory (Sophistolab AG). In brief, 3 μm sections were processed on a Leica BOND-MAX or Bond-III automated IHC system. Samples were stained with horse radish peroxidase– or alkaline phosphatase–labeled antibodies for 30 minutes at room temperature. The following were used: rabbit α–human CD20 (SP32, Cell Marque), mouse α–human CD20 (L26, Dako), rabbit α-CD3 (SP7, Diagnostic Biosystem), rabbit α–human CD103 (EPR4166[2], Abcam), rabbit α–mouse Lyve-1 (polyclonal [103-PA50AG], RELIATech GmbH), and rat α–mouse/human PNAd (MECA-79, BioLegend). For detection of horse radish peroxidase or alkaline phosphatase, BOND Polymer Refine (DAB) or BOND Polymer Refine Red (Fast Red) were used, respectively (both from Leica Biosystems). EBER in situ hybridization was performed using a Benchmark Ultra automated slide stainer (Ventana). Briefly, tissue sections were pretreated with protease 3 before incubation with an EBER-specific probe (Ventana), which was detected via the ISH iView Blue Detection Kit (Ventana). Images were acquired with the slide imaging PerkinElmer Vectra 3 system. Cell segmentation, total cell count, and phenotype quantification was done using the inform 2.5.1 Tissue Finder Advanced Image Analysis Software from PerkinElmer, as previously described (89). For the final analysis, only cells that the software identified with a confidence level of > 90% were included in the quantification.

Statistics. Data were analyzed and graphed using Prism Software (v9.5.1 or 8.3, GraphPad). If not described otherwise, data sets of 2 were compared using 2-way Mann-Whitney U test, and data sets of 3 were compared by Kruskal Wallis test with Dunn’s correction for multiple comparisons. Data within same individuals or samples were compared using paired Wilcoxon signed-rank test. For all statistical tests, P values less than 0.05 were considered significant. For graphing, either mean ± SD or geometric mean and 95% CI were displayed.

Study approval. All animal work strictly followed the animal protocols ZH159/2017 and ZH041/2020, licensed by the Veterinary Office of the Canton of Zurich (Zurich, Switzerland). Human tonsil samples collected at New York Presbyterian and the University of Zurich (Zurich, Switzerland) were obtained as part of IRB-approved protocols within previous investigations (90). All participants provided informed consent in accordance with the Declaration of Helsinki, and the institutional ethics committees (New York Presbyterian and the University of Zurich) approved all protocols used (90).

Data availability. Underlying data are provided in the Supporting Data Values file. Raw and processed scRNA-Seq data have been uploaded onto the European database (Annotare/ArrayExpress) and are available under the accession no. E-MTAB-13853 (https://www.ebi.ac.uk/biostudies/arrayexpress/studies/E-MTAB-13853).