The human tissue-resident CCR5⁺ T cell compartment maintains protective and functional properties during inflammation

CD4 and CD8 CD69⁺ cells are abundant in healthy oral mucosal tissue

We first broadly characterized CD8 and CD4 T cells in blood and healthy mucosal tissue of humans obtained from routine periodontic surgeries (Fig. 1A). We chose to use oral mucosal tissue, because this mucosal barrier tissue is readily available in both healthy and inflamed states in the absence of drug treatment that would interfere with inflammatory processes from patients undergoing routine oral surgery. To assess the health status of the samples, blinded scoring was performed on H&E slides by a pathologist (score of 1 to 5) and scores 1–2 were included in healthy analysis (fig. S1 and table S1). In contrast to the donor-matched blood, almost all T cells in the mucosa had an effector memory (T_EM) cell phenotype (CD45RA⁻CCR7⁻) (Fig. 1A). We next examined markers indicative of transient versus tissue-resident T cell (T_RM) subsets. T_RM are often defined by expression of the biomarkers CD69 and/or CD103, each thought to have a unique contribution towards establishing residency and displaying different expression patterns in human non-lymphoid tissues (16). Similarly to what has been reported for other human non-lymphoid tissues (15), CD8 T cells expressed CD69 and CD103 in the oral mucosa (Fig. 1A and 1B). Using matched blood as a gating reference, we found that half of the CD4 T cells in healthy oral mucosa were CD69⁺ and a portion were CD103⁺, with very few of either population in the blood (Fig. 1A and 1C). The CD103⁺CD69⁺ population for both CD4 and CD8 T cells was exclusively found in the mucosa, but not blood (Fig. 1B and 1C). We found that PD-1 was highly expressed on mucosal CD69⁺ cells (Fig. S2A) and all CD4 T cells in the tissue had higher expression of CTLA4 than those in the blood (Fig. S2B). Further, in 89% of patients, CTLA4 expression was higher on the CD69⁺ population compared to the CD69⁻ population (mucosal T_reg shown for comparison). We also determined the tissue localization of the CD103 (Fig. S2C and S2E) and CD69 (Fig. S2D and S2E) CD4 T cell subsets by immunofluorescence (IF) and found that neither were associated with preferred localization in the tissue. To determine spatial localization and assess cell subset recovery from the tissue digestion process (26), we combined IF staining with a digital pathology (IFDP, indicated by blue graphs in figures) analysis approach to increase the number of cells that can be analyzed and remove any subjective bias when enumerating cells. Specifically, we quantified cells in three relevant compartments: epithelium, sub-epithelium (300μm area below the epithelium) and stroma (Fig. 1D). The sub-epithelial layer was defined after initial analysis indicated an enrichment of T cells in this area. For each donor, 176–14,734 CD4 T cells (mean = 2,102) and 23,379–210,390 nucleated cells (mean = 70,926) were quantified.

We next examined the overall distribution of CD4 and CD8 T cells (Fig. 1E). In healthy oral mucosa we found a 1:1 ratio of CD4 and CD8 T cells in the epithelium whereas this ratio averaged 3:1 and 2:1 in the sub-epithelium and stroma, respectively (Fig. 1F). We also assessed CD4 and CD8 T cells as a percent of nucleated cells, a measurement indicative of the density of cells they would be surveying. By this metric, CD4 and CD8 T cells made up 6% and 2.7%, respectively, of nucleated cells in the sub-epithelium compared to the epithelium where they averaged 0.98% and 0.77% (Fig. S3). It is noteworthy that the nuclei per mm² density is higher in the epithelial compartment (Fig. S3). Thus, CD4 T cells are most abundant in a 300μm area below the epithelium.

CCR5 is expressed on resident and transient CD4 T_conv and T_reg populations in healthy oral mucosa

We next wanted to specifically focus on defining the CCR5⁺ CD4 T cell population given the critical role of CD4 T cells in maintaining barrier immunity (21). In healthy tissues, CCR5 was expressed by ~65% of the CD4 T_RM cells (Fig. 2A and 2B). The CCR5⁺CD69⁺ population made up about a third of the total CD4 T cell population (Fig. 2C) and could be readily identified by IF (Fig. 2D). Given that T_reg are necessary to maintain tissue homeostasis and orchestrate effector function (27), we next defined their abundance and phenotype. Identification of T_reg in humans is confounded by the fact that Foxp3 and other identifying markers can also be expressed by activated effector T cells (28). We chose a stringent previously described gating method (28) to minimize the possibility of misclassifying recently activated CD4 T cells as T_reg in tissues. This strategy isolated the most suppressive and highest CD25-expressing cells using CD45RA expression as a reference after first gating out CD127⁺ cells (Fig. 2E). The conservative gating strategy yielded a lower frequency of T_reg (as a % of CD4 T cells, Fig. 2E and 2G) in the blood compared to the typically used gating schemes. Mucosal T_reg had increased Foxp3 and CTLA4 expression compared to those in the circulation (Fig. 2F). Of note, both of these markers were also higher on the T_conv population in the tissue compared to blood. T_reg cells were more prevalent in the tissue compared to the blood (Fig. 2G) and expressed T_RM markers CD69 and to a lesser extent CD103 (Fig. 2H). Importantly, T_reg also displayed a pattern of CCR5 expression (Fig. 2I) similar to the T_conv population (Fig. 2B) across blood and tissue subsets.

Inflammation drives spatially-dependent changes in T cell subsets

We next wanted to determine the effect of inflammation on the CCR5⁺ CD4 T cell compartment to understand if functionality and subset distribution are altered. Inflammation involves changes in vasculature, secretion of inflammatory mediators and cellular infiltrate (29) that together alter tissue architecture and the local immune network (30). As expected, large T cell clusters were readily apparent in the stroma of inflamed tissues (score 3–5, Fig. 3A). IFDP analysis revealed that overall CD4 T cell numbers (as a proportion of nucleated cells) were not markedly changed in the epithelium with worsening inflammation while increasing significantly in the sub-epithelium (p=0.005) and stroma (p=0.001, Fig. 3B). The proportion of CD69⁺ CD4 T cells was preserved with increasing inflammation by both flow and IFDP analysis (Fig. 3C and 3D). We found a significant decrease in the proportion of CD103⁺ CD4 cells by flow cytometry (p=0.012, Fig. S4A) and IFDP analysis revealed that this was occurring in the sub-epithelium (p=0.040); there were no statistically significant differences in the stroma (Fig. S4B). The same results were found for CD8 T cells to an even greater degree (Fig. S4C and S4D).

In inflamed tissues, the pattern of CCR5 expression across blood and tissue compartments was maintained on CD4 T cells (Fig. 4A) and we observed no changes in the overall proportion of CCR5-expressing CD4 T cells with increasing inflammation by flow cytometry (Fig. 4B). However, IFDP analysis revealed that the proportion of CCR5⁺ CD4 T cells increased with worsening inflammation, specifically in the sub-epithelium (and considerably in the stroma) (Fig. 4C and 4D), which is also where the majority of the increase in CD4 T cell numbers occurred. Of note, the proportion of CCR5⁺ CD4 T cells detected by IF was uniformly lower compared to flow cytometry (Fig. 4A and 4B versus 4D), which most likely can be attributed to the fact that flow cytometry captures the full range of CCR5 expression and parameters set for HALO analysis exclude the lower range of expression (to distinguish from background signal). Finally, we wanted to determine if inflammation status affected the frequency of the CCR5⁺CD69⁺ population. Inflammation did not significantly increase in the proportion of CCR5⁺CD69⁺ CD4 T cells below the epithelium (sub-epithelium p=0.07 and stroma p=0.20, Fig. S4E). Importantly, we observed that the proportion of T_reg cells was maintained regardless of inflammation status (Fig. 4E). Treg cells reflected the same patterns as the CD4 T_conv cells in regard to CCR5 expression across blood and tissue subsets (Fig. 4F) as well as the CD103 and CD69 populations (Fig. S4F and S4G).

CCR5⁺CD69⁺ CD4 T cells in the mucosa share a universal T_RM transcriptome signature and have a T_H17 profile

To define the transcriptional breadth of CCR5-expressing CD4 T cells, we examined several CD4 T cell subsets by bulk RNA sequencing (RNA-seq). From 10 individuals (with varying degrees of tissue inflammation) three CD4 T cell populations were sorted: blood CCR5⁺CD69⁻, mucosa CCR5⁺CD69⁺ and mucosa CCR5⁺CD69⁻ (Fig. 5A). We first stratified all mucosa samples by inflammation score, treating it as a continuous variable, and found that inflammation had a very modest effect on CCR5+ T cell transcriptomes. Even with a non-stringent false discovery rate (FDR) of 0.25, only 14 genes were differentially expressed (DE) (Fig. 5B). Therefore, although a clear pattern of transcriptional changes is apparent, the functional potential of CD4 T cells in the tissue is remarkably stable in the presence of increasing inflammation. In contrast, a large number of DE genes were found between cells in the circulation compared to both CD69⁻ and CD69⁺ cells in the oral mucosa (913 DE genes with FDR < 0.05, expression fold-change ≥ 2) (Fig. 5C and Table S2). A Search Tool for Retrieval of Interacting Genes/Proteins (STRING) network analysis (31) within these differentially expressed genes revealed that tissue occupancy was associated with an increased functional specialization that included both effector and regulatory genes (Fig. S5A). Our findings indicate that entry into the tissue, independent of CD69 expression, has a more profound impact on the transcriptional profile of CD4 T cells than inflammation status.

We next performed gene set enrichment analysis (GSEA) (32) to compare our CCR5⁺CD69⁺ population with previously defined T_RM populations (33). For this analysis, we ranked genes according to the absolute values of expression log fold changes between the tissue CCR5⁺CD69⁺ and tissue CCR5⁺CD69⁻ samples. Sample label permutations were used to assess significance. GSEA revealed enrichment of the T_RM Core Signature in the genes differentially expressed in mucosa CCR5⁺CD69⁺ CD4 T cells (p <0.001; Fig. 5D), indicating strong similarity to other human Trm populations. To extend the generality of this comparison beyond human CD4 T cells, we also performed GSEA with a well-defined murine CD8 T cell gene signature (34) and obtained a p value of 0.04 (Fig. S5B). This suggests a trend towards enrichment even in the context of a different species and cell type. Therefore, in line with recent literature (35–38) and since the mucosal CCR5⁺CD69⁺ CD4 T cell population maintains a signature indicative of tissue residency, we will refer to these CD69⁺ simply as T_RM cells, although it is important to note that CD69 expression requires careful interpretation due to its transient expression on the cell surface of recently activated T cells. Given their divergent transcriptional profiles, we wanted to define potential functional differences between these tissue-resident and tissue-transient populations. Comparing the T_RM with the CD69⁻ tissue population, there were 19 DE genes (FDR <0.05, fold-change in expression ≥ 2) (Fig. 5E). A striking difference between the transcriptional profile of the two oral mucosal populations was the expression of IL17F, IL17A, IL26, and aryl hydrocarbon receptor (AhR), which suggest an enhanced T_H17 signature in the T_RM population (Fig. 5E). We did not detect differentially expressed genes indicative of recent activation or exhaustion/dysfunction, indicating that CD69 protein expression is mainly due to tissue residence and likely not driven by other events.

T_H17 CD4 T cells with a T_RM signature are a distinct subset within the mucosa CCR5⁺CD69⁺ population

To address whether T_H17 and T_RM transcriptional profiles could be simultaneously detected in individual cells we used 10X Genomics for single-cell RNA sequencing (scRNA-seq). We sorted six defined populations from the blood and mucosa of a single donor (pathology score 4 with some healthy as well as markedly inflamed areas). This nested sort approach retained marker information ensuring that we could link the transcriptional signatures back to previously defined subsets. Three of the populations matched those sorted for bulk RNA-seq (denoted with a *), with the other populations providing reference datasets: Blood Total CD4⁺, *Blood CCR5⁺CD69⁻, Mucosa CCR5⁺, Mucosa CCR5⁻, *Mucosa CCR5⁺CD69⁺ and *Mucosa CCR5⁺CD69⁻ (sort strategy and sorted cell numbers are listed in Fig. S6A).

Visualization of the single-cell transcriptome data using t-distributed stochastic neighbor embedding (tSNE) plots (Fig. 6A) and differential expression between the nonoverlapping populations (Blood CCR5⁺CD69⁻, Mucosa CCR5⁻, Mucosa CCR5⁺CD69⁺, and Mucosa CCR5⁺CD69⁻) confirmed that localization to the tissue had a profound impact on CD4 T cell transcriptional profiles (Fig. 6B). Importantly, CCR5 expression imparted a distinct transcriptional profile in the tissue (Table S5), as CCR5⁺ T cells had higher expression of genes related to effector function and migration reflecting the diverse functional potential in this population (T_H1, T_H17 and T_reg). This further illustrates the importance of focusing on a specific inflammatory-responsive CD4 T cell subset when comparing blood and tissue.

We next asked whether blood CCR5⁺CD69⁻, Mucosa CCR5⁺CD69⁻ and Mucosa CCR5⁺CD69⁺ cells co-expressed genes from the T_RM Core Signature, indicative of T_RM, and a T_H17 signature (gene signature list is in Table S3). The mean gene set expression of each signature revealed individual cells expressing both signatures specifically within the CCR5⁺CD69⁺ population, but not in the CCR5⁺CD69⁻ populations (blood or mucosa) expressing both signatures (Fig. 6C). We confirmed that these signatures are not a transcript abundance-related artifact due to capture of multiple cells, as across the three samples, the cells co-expressing these signatures do not show an unusually high number of unique molecular identifiers (UMIs) per cell (UMIs) per cell (Fig. S6B).

CD4 T_RM in inflamed tissue respond rapidly to TCR stimulation ex vivo to produce cytokines

Since we had seen by both bulk and single-cell RNA-seq a strong T_H17 signature in the CCR5⁺ CD4 T_RM population, we wanted to determine whether this actually resulted in enhanced IL-17 protein production. We therefore chose short re-stimulation assays (minimizing possible indirect effects from longer culture(39)) to determine the direct ex vivo TCR-responsiveness of blood and mucosal CD4 T cells. Activating cells for 6 hours was brief enough to avoid substantially changing the percent of CD69⁺ cells, however, it altered CCR5 expression as previously reported (40). The stimulation experiments echoed the transcriptional data as a higher percent of the oral mucosa CD69⁺ population (that typically consists of ~65% CCR5⁺ T cells, Fig. 2B) produced IL-17A/F than the CD69⁻ population (Fig. 7A and 7B). In contrast, IFNg expression was increased in both oral mucosa populations compared to blood, regardless of CD69 expression (Fig. 7A and 7C), also mirroring the transcriptional data (Fig. S7A). Similar results were obtained following PMA/Ionomycin re-stimulation, with overall increased cytokine production (Fig. S7B). Thus, the data strongly suggest the existence of a CCR5⁺CD69⁺ CD4 T cell population with T_H17 properties that is found predominantly in this tissue-resident population.

CCR5⁺ CD4 T_RM are stably maintained during treatment with the CCR5 antagonist Maraviroc

We next wanted to directly test if the CCR5⁺ CD4 T cell population is maintained in mucosal tissue during treatment with Maraviroc, a CCR5 antagonist. To address this question a healthy study cohort is needed with available tissue biopsies pre- and post-treatment in a well-defined time-span. The CHARM-03 clinical trial met all these criteria and included an elegant cross-over study design that even allowed us to compare effects of oral versus topical Maraviroc treatment for one week (with a washout period between the treatment arms) (Fig. 8A and Materials and Methods). To determine if the CD4 T cell compartment changes following oral (systemic) or topical (rectal) Maraviroc treatment, we examined rectal biopsies taken pre- and post-treatment by flow cytometry and IF.

The frequency of the CCR5⁺CD69⁺ population among CD4 T cells was comparable between oral and rectal mucosa (Fig. 2A, C and Fig. 8B, C), however CCR5 expression is minimal in the rectal mucosal CD69⁻ population (Fig. 8B). Importantly, the CCR5⁺ T_RM population remained stable with either oral or topical treatment (Fig. 8C). To help interpret these data, we also assessed the frequency of CD3⁺ cells (% of total lymphocytes) and the frequency of CD4⁺ cells (% of total CD3⁺ cells) and did not observe significant changes with treatment (Control vs. Oral Tx p=0.78, Control vs. Topical Tx p=0.90, Fig. S8A). We did however find alterations in the frequency of CCR5 expression on the CD69⁻ population that resulted in a significant increase in the proportion of CD69⁻CCR5⁺ (Control vs. Oral Tx p=0.0008, Control vs. Topical Tx p=0.0035, Fig. S8B), which has been observed previously (41) and further confirmed that the treatment was working as expected. To complement the flow cytometry data, we examined whether Maraviroc treatment caused changes in the spatial distribution of CD4 T cells in the tissue. These tissue biopsies were taken with large-cup biopsy forceps and thus the orientation of these sections was different compared to our other IF images (Fig. S8C). In the acquired epithelial layer, we observed heterogeneity in the overall CD4 T cell distribution regardless of treatment status and readily detected clusters of CD4 T cells expressing CCR5 pre- and post-treatment (Fig. 8D). Likely due to the different initial processing of these tissues we were unable to simultaneously stain for CD69, CCR5, and CD4. We therefore decided to stain for CCR5 and CD4, since CCR5⁺ CD4 T cells were almost all found within the CD69⁺ fraction by flow cytometry. Together, the flow cytometry and IF data suggest that a stable population of CCR5⁺ CD4 T_RM is maintained in human mucosal tissue capable in the absence of CCR5-directed recruitment.

Study design

This study was exploratory with the purpose of understanding CD4 T cell dynamics in human oral mucosal tissues. Human oral mucosa was obtained from patients undergoing oral surgery where gingival tissue was removed as a normal part of the procedure. All participants providing oral mucosal samples signed a written informed consent prior to inclusion in the study and the protocols were approved by the institutional review board (IRB) at the Fred Hutchinson Cancer Research Center (8335). Surgical procedures included gingivectomy/gingivoplasty, osseous surgery, implant uncovering and tooth extractions. Samples were collected between 11am and 4pm. Participants were between 14–83 years old (mean = 52), in line with previously published donor populations (49). See Table S1 for more information. Samples sizes were based on availability and because there was no intervention, no blinding or randomization were used. The number of patients included or represented is indicated for each figure. Human rectal tissue was obtained from healthy, HIV-uninfected participants (n=5 male and n=5 female) as part of the CHARM-03 study (ClinicalTrials.gov; NCT02346084). All CHARM-03 participants signed a written informed consent prior to inclusion in the study and the protocol was approved by the IRB (REN18050213) at the University of Pittsburgh. At the time of enrollment, pre-treatment biopsies were collected with large-cup biopsy forceps. Participants then took oral Maraviroc (300mg) for 8 days. 24 hours after the last dose, rectal biopsies were taken. Participants then had a 14–21-day recovery period before beginning 8 days of topical rectal Maraviroc (1% gel) and 24 hours after the last dose rectal biopsies were taken. For n=3 of the female participants, the rectal topical treatment followed a washout period after vaginal topical use – these samples are indicated in the figures. Primary data are reported in data file S1.

Human oral mucosa processing

Fresh tissue was placed immediately into a 50ml conical tube with complete media (RPMI 1640 supplemented with penicillin at 100 U/ml, streptomycin sulfate at 100 μg/ml, and fetal bovine serum (FBS) at 10%). Blood was collected by flushing the exposed surgical site after elevation of the gingiva with sterile saline. Two sterile 5ml syringes with 10-gauge blunt tips were used to collect approximately 10ml of saliva and blood from the oral cavity. Both tissue and blood were kept on wet ice or at 4°C until being processed within 2–4 hours. Cells from mucosal tissue were extracted using an adapted version of a previously published method (50). Briefly, tissues underwent two rounds of enzymatic digestion with collagenase II (Sigma-Aldrich) at 37°C, each followed by mechanical digestion with 30cc syringe and blunt 16-gauge needle. Oral blood was treated with ACK lysing buffer (Gibco) to remove red blood cells.

Human rectal biopsy processing

At each timepoint from each subject, n=6 rectal biopsies were taken and immediately processed for flow cytometry. N=2 biopsies were flash frozen and archived. For the purposed of this study, one of those flash frozen biopsies for each participant at 3 timepoints (pre-treatment, post-oral treatment, and post-topical treatment) was embedded in Tissue-Tek optimal cutting temperature O.C.T.) compound (Sakura Finetek, Thermo Fisher).

Bulk RNA-seq library generation

We performed RNA-seq on 250–1000 sorted CD4⁺CCR5⁺ cells from blood or tissue samples as previously described (51). In total, 28 samples were sequenced: 9 from CD69⁻ blood samples, 9 from CD69⁻ tissue samples and 10 from CD69⁺ tissue samples from a total of 10 subjects. Briefly, cells were sorted directly into SMARTer v4 lysis reagents (Clontech). Cells were lysed and cDNA was synthesized. After amplification, sequencing libraries were prepared using the Nextera XT DNA Library Preparation Kit (Illumina). Barcoded single-cell libraries were pooled and quantified using a Qubit Fluorometer (Life Technologies).

Bulk RNA-seq statistical analysis

A quality filter was applied to retain libraries in which the fraction of aligned reads examined compared to total FASTQ reads was >75%, the median coefficient of variation of coverage was less than 0.8, and the library had at least 1 million reads. All sequenced samples passed these quality filters. Non-protein coding genes, mitochondrial genes, and genes expressed at less than 1 count per million in fewer than 10% of samples were filtered out. Expression counts were normalized using the TMM algorithm (52). For differential gene expression analysis, we used the linear models for microarray data (Limma) R package (53, 54) after Voom transformation (55); this approach either outperforms or is highly concordant with other published methods(56, 57). Linear models comparing blood and tissue sample gene expression and comparing CD69⁺ and CD69⁻ cells from tissue samples were used. In both models, donor identity was included as a random effect. Genes with a false discovery rate (FDR) of less than 0.05 and expression fold-change of greater than 2 between two blood and tissue samples were considered differentially expressed. STRING was used to identify protein-protein interactions in differentially expressed genes. STRING connects genes according to known or predicted interactions to form a network and can be used to perform enrichment analysis to identify biological processes associated with these networks (31).

Single-cell RNA-seq library generation and sequencing

Freshly isolated cells from blood and tissue were sorted into complete media and processed using the 10X Genomics Platform with the Single-Cell 3’ Reagent (v2) as described previously (58). Sorted populations and total cell numbers loaded into the Chromium Controller: Blood CD4 T cells (4000), Blood CCR5⁺CD69⁻ CD4 T cells (3000), Blood CD4 T_reg (1000), Tissue CCR5⁺ CD4 T cells (4000), Tissue CCR5⁻ CD4 T cells (4000), Tissue CCR5⁺CD69⁻ CD4 T cells (4000), Tissue CCR5⁺CD69⁺ CD4 T cells (4000) and Tissue CD4 T_reg (2000). According to the manufacturer’s instructions, the target cell number for each population was calculated as half the number loaded.

Single-cell RNA-seq data analysis

Sequencing data were processed to read and unique molecular identifier (UMI) counts per cell using the 10X Cell Ranger software, v. 2.1, using default parameters. Reads were converted from BCL to FASTQ format, demultiplexed, filtered, aligned to the genome, collapsed to UMIs, and assigned to barcodes. Barcodes were called as cells vs. background using default settings in Cell Ranger, with target cell numbers as described above. Data from each sample were aggregated without normalizing, as normalization was conducted during downstream analysis.

Count data were analyzed using the Seurat toolset (59). We removed genes detected in < 3 cells, and removed cells with < 200 or > 1800 genes detected, total UMI counts > 6500, or > 10% mitochondrial reads, based on observed breaks in the cell distributions. We log-transformed the UMI counts and regressed out total UMIs and percent mitochondrial reads per cell. We projected the samples using t-distributed stochastic neighbor embedding (tSNE) on the first 10 principal component (PC) axes from 1030 highly variable genes identified using default cuts. To summarize each cell’s expression of defined gene sets, we calculated the mean normalized expression across the set in each cell; to remove noise around 0 introduced by normalization, we restored 0 values for each gene in each cell with a raw UMI count of 0. We quantified differential expression among populations using the hurdle model implemented in MAST(60), with total UMIs per cell as a covariate; MAST has high precision and low false positive rates, and generally identifies DE genes that overlap with those found by other methods (61).

Pathology assessment and scoring

Tissue was processed as described for IF staining (frozen in O.C.T. and 8μm sections were cut) and slides were stained with hematoxylin and eosin (H&E). Histologic sections of oral mucosa (sampled from each patient) were assessed for quality and orientation and excluded if the majority of the epithelium or a representative portion of the lamina propria were not present. All sections were evaluated blinded and scored according to the following criteria (Fig. S1): severity of inflammation (1–5), location of inflammation (lamina propria; diffuse or perivascular; intraepithelial), type of inflammatory infiltrate (mononuclear; neutrophilic), presence of lamina proprial edema and epithelial lesions (degeneration/necrosis, erosion, or ulceration). From these criteria, each sample of oral mucosa was deemed as healthy (inflammation score of 1 and 2) and unhealthy (inflammation score of 3–5). In cases of more than one section representing the sample/patient, the scores were based on the most severely affected regions of all sections.

Statistical analyses

For flow cytometry and IFDP data, a paired t test, linear regression or repeated measures one-way ANOVA with Tukey’s post-test was performed as described in figure legends. Two-sided P-values >0.05 were considered not significant (ns), and values denoted with (*) symbols reflect significance levels as follows: P ≤ 0.05 (*), P ≤ 0.01 (**), P ≤ 0.001 (***), and P ≤ .0001 (****).