Extricating human tumor-unique immune alterations from non-malignant tissue inflammation


 Immunotherapies to treat cancer have achieved remarkable successes, but major challenges persist. An inherent weakness of current treatment approaches is that therapeutically targeted pathways are not only found in tumors, but also in tissue microenvironments, particularly inflamed tissues. This confounding overlap complicates treatment as well as predictions of treatment outcome. In an effort to identify potential tumor-unique immunotherapeutic targets that are distinct from general tissue inflammation, we used complementary single-cell analysis approaches to interrogate immune cell alterations and interactions in human squamous cell carcinomas and site-matched non-malignant, inflamed tissues. We found that a distinct population of intratumoral regulatory T cells (Tregs) received T cell receptor (TCR) signals from antigen-presenting cells and this Treg population was uniquely identified by co-expression of ICOS and IL-1 receptor type 1 (IL-1R1). Intratumoral IL-1R+ Tregs appeared activated and a TCR signal was sufficient to convert IL-1R1- Tregs to IL-1R1+ Tregs ex vivo. Overall, our work identifies an intratumoral Treg population that recognizes antigen in the tumor microenvironment and two biomarkers that allow for specific depletion of these Tregs. Finally, our approach also provides a blueprint for extricating tumor-unique therapeutic targets distinct from general inflammatory patterns in other tumors.


Abstract: 1
Immunotherapies to treat cancer have achieved remarkable successes, but major 2 challenges persist 1,2 . An inherent weakness of current treatment approaches is that 3 therapeutically targeted pathways are not only found in tumors, but also in tissue 4 microenvironments, particularly inflamed tissues. This confounding overlap complicates 5 treatment as well as predictions of treatment outcome 3,4 . In an effort to identify potential 6 tumor-unique immunotherapeutic targets that are distinct from general tissue 7 inflammation, we used complementary single-cell analysis approaches to interrogate 8 immune cell alterations and interactions in human squamous cell carcinomas and site-9 matched non-malignant, inflamed tissues. We found that a distinct population of 10 intratumoral regulatory T cells (Tregs) received T cell receptor (TCR) signals from antigen-11 presenting cells and this Treg population was uniquely identified by co-expression of 12 ICOS and IL-1 receptor type 1 (IL-1R1). Intratumoral IL-1R + Tregs appeared activated 13 and a TCR signal was sufficient to convert IL-1R1 -Tregs to IL-1R1 + Tregs ex vivo. Overall, 14 our work identifies an intratumoral Treg population that recognizes antigen in the tumor 15 microenvironment and two biomarkers that allow for specific depletion of these Tregs. 16 Finally, our approach also provides a blueprint for extricating tumor-unique therapeutic 17 targets distinct from general inflammatory patterns in other tumors. 18 Page 3

Main Text: 20
Antigen-presenting cells (APCs) and T cells residing in non-lymphoid tissues adapt 21 distinct phenotypic and functional properties relative to their circulating counterparts in 22 the peripheral blood 5,6 . These immune cells respond to tissue damage or invading 23 pathogens with a tightly regulated effector program 6,7 and are also present in many 24 solid tumors types, where they are thought to be critical determinants of tumor 25 development and disease outcome 1,8 . One hallmark of immune-infiltrated human 26 tumor tissues is the presence of an inflammatory microenvironment, which has been 27 extensively scrutinized during the past decade 3,9 . However, despite these efforts it 28 remains unclear which immune cell subsets and signaling pathways in the human 29 tumor microenvironment are distinct from general inflammatory processes that occur in 30 other tissues. 31 One of the best studied immune populations in tumor tissues are functionally 32 exhausted (dysfunctional) T cells and regulatory T cells (Tregs), both of which are 33 considered pivotal factors for inefficient anti-tumor immune responses 10,11 . These T cell 34 subsets express immuno-inhibitory molecules such as programmed death 1 (PD-1) or 35 cytotoxic T-lymphocyte-associated antigen 4 (CTLA-4), which are the focus of various 36 immunotherapeutic approaches 12 . However, expression of PD-1 and CTLA-4 is not 37 limited to tumor-infiltrating T cells, but also found on T cells in non-malignant tissues 38 during homeostasis and inflammation 6,13 . Importantly, the effector program of T cells 39 and their expression of immuno-regulatory molecules is closely linked to the function of We hypothesized that comparing the human tumor microenvironment to non-47 malignant, inflamed tissues could reveal truly tumor-unique immune alterations that 48 could help reveal why some tumors do not respond to immune checkpoint inhibitors 49 and could even lead to the identification of novel therapeutic targets. Head and neck 50 squamous cell carcinomas (HNSCC) have a large immune infiltrate, but most patients 51 do not respond to treatment with immune checkpoint inhibitors 18 . We thus combined 52 several single-cell analysis pipelines to generate a comprehensive immune landscape 53 of human HNSCC with site matched non-malignant inflamed tissues. Our data revealed 54 substantial congruence of the immune phenotypes between these tissue groups, but 55 computational analysis approaches identified tumor-unique subsets of activated APCs 56 and Tregs. Predicted interaction patterns of APCs and Tregs included MHC-TCR and 57 IL-1-IL-1R signaling. Ex vivo experiments confirmed these computational predictions. 58 IL-1R1 + Tregs showed hallmarks of increased immunosuppressive function and recent 59 TCR stimulation. Finally, these tumor-unique Tregs could be identified among all 60 hematopoietic cells by the combined expression of IL-1R1 and ICOS, thus allowing for 61 specific intratumoral depletion of these Tregs by bispecific antibodies or logic-gated 62 CAR T cells, which could ultimately help restore anti-tumor immune responses. 63 Page 9 to identify potential tumor-unique cross-talk between T cells and APCs using NicheNet 156 35 . For this, we leveraged the ability of NicheNet to predict ligand-receptor interactions 157 based only on differentially expressed genes in the HNSCC vs. OM-derived T cells 158 (workflow outlined in Fig. 3a). We set our scRNA-seq derived APC clusters (excluding 159 pDCs and mast cells) as the sender population, and the CD4 + T cell, CD8 + T cell, and 160 CD4 + Treg clusters as separate receiver populations. For each T cell subset, we 161 focused our analysis on the top 20 ligand-receptor pairs identified by NicheNet and 162 visualized these interactions as circos plots (Fig. 3b). We arbitrarily sub-divided these 163 interactions into 3 groups (unique/interesting, cytokine/co-receptor and other) for the 164 sole purpose of helping with visualization. We further highlighted some ligand-receptor 165 pairs of interest (bold, underlined) across all T cell subsets. Remarkably, NicheNet 166 predicted that four ligand-receptor interactions were unique between the APC and 167 Treg population in HNSCC: ICOS ligand (ICOSLG) via ICOS, the cytokines IL-15 168 through the IL2RG and IL2RB receptor complex, the pro-inflammatory cytokines IL-18 169 via the IL18-R1 and IL-1B via the IL-1 receptors type 1 and type 2 (Fig. 3b, right panel). 170 Furthermore, Nichenet also predicted TCR signaling in the Treg compartment, via the 171 TCRzeta-chain (CD247) and CD3G (Figure 3b, right panel). 172 We were most intrigued by the predicted TCR and IL-1 signaling events given that the 173 Treg population in human HNSCC is expanded and expansion of the Treg 174 compartment has been associated with transient immune checkpoint inhibitor 175 treatment effects in an HNSCC mouse model 36 . Thus, we further interrogated the 176 accuracy and relevance of these interactions in a series of ex vivo experiments. We first 177 asked if IL-1 could be present in the tumor. After ex-vivo culture in the presence of

Page 10
Brefeldin A only, a majority of CD14 + monocyte/macrophage like cells expressed IL-1b 179 as well as IL-1a protein, as did up to 20% of the pan cDCs (Fig. 3c). As expected, 180 CD123 + pDCs did not express IL-1a/b. To address whether IL-1 is available in the 181 tumor microenvironment, we performed Luminex analysis of flash-frozen whole tumor 182 lysate, which revealed significant levels of IL-1a, IL-1b and IL-18 (Fig. 3d). Together, 183 these data indicate that an intratumoral IL-1 signal is feasible. Next, we tested if the 184 predicted "receiver" receptors are expressed by T cells using an additional 30 185 parameter flow cytometry panel (Suppl. Table 2). We found that IL-1R1 was specifically 186 expressed by tumor-infiltrating Tregs, but neither by tumor infiltrating CD4 + T cells or 187 CD8 + T cells, nor by T cells in the peripheral blood (Fig. 3e). Importantly, up to 60% of 188 the Tregs expressed IL-1R1, while expression of IL-1R2, which is thought to be a 189 decoy receptor for IL-1 signaling, was detectable on less than 3% of cells. We further 190 analyzed the phenotype of IL-1R1 + Tregs and found that nearly all IL-1R1 + Tregs were 191 co-expressing ICOS and HLA-DR, and higher levels of the chemokine receptor CXCR6 192 ( Fig. 3f). We next asked if the combined expression of IL-1R1 and ICOS could uniquely 193 identify Tregs among all hematopoietic (CD45 + ) cells in HNSCC and blood. We found 194 that nearly all of cells in the CD45 + IL-1R1 + ICOS + gate were CD3 + CD4 + CD25 + CD127 -195 Tregs (Fig. 3g). These data suggest that a large fraction of Tregs in the tumor could be 196 directly targeted and depleted by only using these two cell surface-expressed proteins. 197 Thus, we next wanted to determine if these Tregs are a potentially clinically relevant 198 target and further explore a possible link between IL-1R1 expression and the predicted 199 TCR signaling events. 200

IL-1R1-expressing Tregs represent a functionally distinct Treg population with 202 hallmarks of recent TCR stimulation 203
To further define IL-1R1 + and IL-1R1 -Tregs and assess the biological relevance of IL-204 1R1 + Tregs, we used a targeted transcriptomics approach 37 to measure expression of 205 495 pre-selected genes (Suppl. Table 3) on sorted IL-1R1 + and IL-1R1 -Tregs from 206 three HNSCC tumor donors, identifying two transcriptionally distinct populations of 207 regulatory T cells in the tumor that were separate from peripheral blood Tregs (Fig 4a). 208 The cluster corresponding to IL-1R1 + Tregs (orange) was also marked by high 209 expression of TNFRSF18 (Glucocorticoid-induced TNF receptor, GITR) and TNFRSF9 210 (4-1BB), which has been suggested as a pan-cancer Treg target 38 . Furthermore, the 211 IL-1R1 + cluster showed exclusive expression of the chemokine receptors CXCR6 and 212 CCR8 as well as CD39 and the transcription factor ID3, which has been implicating in 213 formation of a tissue-resident Treg program 39 (Fig. 4b). 214 To further determine whether IL-1R1 + Tregs are indeed functional, we performed an ex 215 vivo stimulation experiment using AbSeq as a read-out for changes in transcript and 216 surface protein expression. After data integration with Harmony 29 we identified two 217 main CD4 + T cell clusters and two Treg clusters based on surface protein and 218 transcript profile (Fig. 4c). Of note, an additional cluster of proliferating cells (high in 219 TOP2A and MKI67) aligned with the IL-1R1 + Treg cluster thus indicating a potential 220 close relationship (Fig. 4c). CD25 + CD127 -Tregs responded to PMA-Ionomycin 221 stimulation by robust upregulation of CTLA4 and CD40L transcript indicating that Tregs 222 are functional and respond as expected (Fig. 4d). Importantly, these CTLA4-high Tregs 223 were mostly found in the IL1R1 + ICOS + population (Fig. 4e).

Page 12
Based on the NicheNet predictions ( Fig. 3b) we considered that IL-1R1 + Tregs in the 225 tumor could represent a population receiving TCR signals. To determine a possible link 226 between TCR signaling and IL-1R1 expression, we sort-purified Tregs from peripheral 227 blood of healthy donors, and IL-1R1and IL-1R1 + Tregs isolated from HNSCC. These 228 sorted Treg populations were then stimulated with anti-CD3/CD28 beads. We 229 observed robust upregulation of IL-1R1 surface expression on all IL-1R1 -Tregs, 230 including IL-1R1 -Tregs isolated from HNSCC tissues (Fig. 4f). These data indicate that 231 a TCR signal is sufficient to elicit IL-1R1 expression by human Tregs and also suggest 232 that IL-1R1 expression on Tregs can be an indicator for recent or active TCR signaling. 233 In contrast, IL-1R2 expression was more limited (Fig. 4f) suggesting that the decoy 234 receptor is unlikely to interfere with IL-1 signals. 235 236

Discussion 237
Overall, our data revealed that many immune phenotypes typically associated with the 238 human tumor microenvironment were also found in non-malignant, inflamed tissues. 239 The expression pattern of PD-1, a key checkpoint inhibitory molecule that is the target 240 of many therapeutic strategies was essentially identical on T cells in tumor tissues and 241 non-malignant, inflamed tissues, which could offer an explanation for the at times 242 severe side-effects of systemic anti-PD-1 treatment 2,40 . Of note, PD-1 expression is 243 typically considered to be driven by T cell receptor signals, but also upregulated by 244 pro-inflammatory cytokines 41 , which may explain the high expression levels in inflamed 245 tissues. Similarly, our data indicated that recently described mregDCs 32 are not tumor-246 unique, given their presence in non-malignant, inflamed tissues, with minimal Page 13 transcriptional changes between these tissues. Overall, this highlights that studying 248 inflamed human tissues can provide a critical reference point for extricating tumor-249 unique changes from general inflammatory immune adaptation. Tregs occur as well. Of note, CXCL16 was among the chemokine transcripts enriched 265 in HNSCC-infiltrating DC3s (Fig. 2h). CXCR6 has previously been suggested to 266 regulate migration of TRM cells 47 . While ex vivo chemotaxis experiments were not 267 feasible, it is still tempting to speculate that these chemokine-receptor pairs could also 268 regulate Treg migration into the tumor microenvironment based on the increased 269 transcript expression of the corresponding ligand CXCL16 in DC3s.

Page 14
Our ex vivo restimulation experiments indicate that a TCR signal is sufficient to convert 271 intratumoral IL-1R1 -Tregs into IL-1R1 + Tregs, and that IL-1R1 + Tregs represent a Treg 272 population receiving TCR stimulation inside the tumor microenvironment. We observed 273 an actively proliferating Treg cluster in the tumor microenvironment which 274 phenotypically aligned with IL-1R1 + Tregs. This could suggest that some of these IL-275 1R1 + Tregs proliferate in situ after receiving a TCR signal from APCs. IL-1 has been 276 shown to enhance CD4 T cell proliferation in a mouse model system 48 , but if IL-1 plays 277 a role in expanding Tregs in the tumor in situ is still unclear. 278 Our findings also have important implications for the design of future therapeutic 279 strategies, since the specific depletion or functional modification of tumor-infiltrating 280 Tregs is considered a promising anti-tumor therapy to reverse the immunosuppressive 281 environment 49,50,51 . We show that the co-expression of IL-1R1 and ICOS is unique to 282 this population of tumor-infiltrating Tregs and co-expression is not found on any other 283 hematopoietically-derived (CD45 + ) cells in the tumor or peripheral blood. Thus, our data 284 also highlight a possible pathway for tumor-specific depletion of a large intratumoral 285 Treg population using bi-specific antibodies or logic-gated chimeric antigen receptor 286 (CAR) T cells 52 . Our data suggest that this depletion would be rather tumor-specific, 287 given that the IL-1 and TCR signaling axis was highly enriched in HNSCC versus 288 inflamed oral mucosa and given the lack of IL-1R1 + Tregs in the periphery. 289 All summary graphs are represented as mean ± SD (n=12 for OM and n=13 for HNSCC 473 samples for T cell data, n = 16 for OM and HNSCC for APC data). Statistical analyses 474 were performed using one-way ANOVA with Tukey's multiple comparisons test. Representative plots showing that within total CD45+ live hematopoietic cells in 506 HNSCC, the majority of the ICOS+ IL1R1+ cell fraction falls within the CD4+ CD25+ 507 CD127-Treg gate. 508 All summary graphs are represented as mean ± SD, n=4 for Luminex data, and n=6 for 509 flow cytometry data. in the presence of anti-CD3/28 beads for 2 days. 525 All summary graphs are represented as mean ± SD (total n=5 for scRNA-seq data, n=3 526 for stimulation assay) Statistical analyses were performed using one-way ANOVA with 527 Tukey's multiple comparisons test. 528

Primary cells: 532
The head and neck squamous cell carcinoma (HNSCC) tissue samples were obtained 533 after informed consent from otherwise treatment-naïve patients undergoing surgical 534 resection of their primary tumor, ensuring that the immune infiltrate was not influenced 535 by prior therapeutic interventions such as radiotherapy. Inflamed oral tissue biopsies 536 were obtained from individuals undergoing routine dental surgeries for a variety of 537 inflammatory conditions such as periimplantitis, periodontitis or osseous surgery. 538 Matched peripheral blood samples were collected from each tissue donor. All study 539 participants signed a written informed consent before inclusion in the study, and the

Isolation of leukocytes from solid human tissues and peripheral blood: 549
After surgical procedures, fresh tissue samples were placed immediately into a 50ml 550 conical tube with complete media (RPMI1640 supplemented with Penicillin, 551 Streptomcyin and 10% Fetal Bovine Serum (FBS)) and kept at 4°C. Samples were Page 28 processed within 1-4 hours after collection based on optimized protocols adapted from 553 (Leelatian et al., 2017). Briefly, tissue pieces were minced using a scalpel into small 554 pieces and incubated with Collagenase II (Sigma-Aldrich, 0.7 mg/ml) and DNAse 555 (50000 Units/ml) in RPMI1640 with 7.5% FBS for 30-45 minutes depending on sample 556 size. Subsequently, any remaining tissue pieces were mechanically disrupted by 557 repeated resuspension with a 30ml syringe with a large bore tip (16x1 ½ blunt). The cell 558 suspension was filtered using a 70um cell strainer, washed in RPMI1640 and 559 immediately used for downstream procedures. Eppendorf tubes containing 500-1000 µL of complete RPMI, washed once in PBS and 622 immediately used for subsequent processing. 623

624
Whole Transcriptome single-cell library preparation and sequencing: 625 cDNA libraries were generated using the 10x Genomics Chromium Single Cell 3' 626 Reagent Kits v2 protocol or the v3 protocol (10x Genomics). Briefly, after sorting single 627 cells were isolated into oil emulsion droplets with barcoded gel beads and reverse 628 transcriptase mix using the Chromium controller (10x Genomics). cDNA was generated 629 within these droplets, then the droplets were dissociated. cDNA was purified using 630 DynaBeads MyOne Silane magnetic beads (ThermoFisher, #370002D). cDNA 631 amplification was performed by PCR (10 cycles) using reagents within the Chromium 632 Single Cell 3' Reagent Kit v2 or v3 (10x Genomics) (see Suppl.  Table 3) via PCR (10-11 cycles). PCR 652 products were purified, and mRNA PCR products were separated from Sample-Tag 653 (and AbSeq, where applicable) PCR products with double-sided size selection using 654 SPRIselect magnetic beads (Beckman Coulter). mRNA and Sample Tag products were 655 further amplified using PCR (10 cycles). PCR products were then purified using 656 SPRIselect magnetic beads. Quality of PCR products was determined by using an 657 Agilent 2200 TapeStation with High Sensitivity D5000 ScreenTape (Agilent) in the Fred 658 Hutch Genomics Shared Resource laboratory. Quantity of PCR products was 659 determined by Qubit with Qubit dsDNA HS Assay (#Q32851). Targeted mRNA product 660 was diluted to 2.5 ng/μL, and the Sample Tag and AbSeq PCR products were diluted 661 to 1 ng/μL to prepare final libraries. Final libraries were indexed using PCR (6 cycles). 662 Index PCR products were purified using SPRIselect magnetic beads. Quality of all final 663 libraries was assessed by using Agilent 2200 TapeStation with High Sensitivity D5000 664 ScreenTape and quantified using a Qubit Fluorometer using the Qubit dsDNA HS Kit Page 34 rpm for 5 minutes. The supernatant was collected and immediately flash-frozen on dry 688 ice. Processing for Luminex was performed by the Immunomonitoring Core of the Fred 689 Hutchinson Cancer Research Center. 690 691 Pre-processing for WTA and targeted transcriptomics data: 692 Raw base call (BCL) files were demultiplexed to generate Fastq files using the 693 cellranger mkfastq pipeline within Cell Ranger (10x Genomics). Whole transcriptome 694 Fastq files were processed using the standard cellranger pipeline (10x genomics) 695 within Cell Ranger 2.1.1 or Cell Ranger 3.0.2. Briefly, cellranger count performs read 696 alignment, filtering, barcode and UMI counting, and determination of putative cells. The 697 final output of cellranger (the molecule per cell count matrix) was then analyzed in R 698 using the package Seurat (3.0) as described below. For targeted transcriptomics data, 699 Fastq files were processed via the standard Rhapsody analysis pipeline (BD 700 Biosciences) on Seven Bridges (www.sevenbridges.com). Briefly, after read filtering, 701 reads are aligned to a reference genome and annotated, barcodes and UMIs are 702 counted, followed by determining putative cells. The final output (molecule per cell 703 count matrix) was also analyzed in R using Seurat (version 3.0) as described below. 704 705 Seurat workflow for targeted and WTA data: 706 The R package Seurat 57 was utilized for all downstream analysis, with custom scripts 707 based on the following general guidelines for analysis of scRNA-seq data 58 . 708 Briefly, for whole transcriptome data, only cells that had at least 200 genes (v2 kits) or 709 800 genes (v3 kits), and depending on sample distribution less than 7-15%

Page 35
mitochondrial genes were included in analysis. All acquired samples were merged into 711 a single Seurat object, followed by a natural log normalization using a scale factor of 712 10000, determination of variable genes using the vst method, and a z-score scaling. 713 Principal component analysis (PCA) was used to generate 75 PCs, followed by data 714 integration using Harmony 29 . The dimensionality reduction generated by Harmony was 715 used to calculate UMAP, and graph-based clustering with a resolution between 0.2 716 and 0.6. For cell annotation, we applied SingleR as a purely data-driven approach 31 , 717 and used the expression of typical lineage transcripts to verify the cell label annotation. 718 For all subsequent analysis steps, the integrated Seurat object was separated into two 719 objects containing all T cells or all myeloid cells, respectively, and UMAP calculation as 720 well as clustering steps were repeated. 721 For targeted transcriptomics data, separate cartridges from the same experiment were 722 merged (if applicable), and only cells that had at least 30 genes were included in 723 downstream analysis. After generating a Seurat object, a natural log normalization 724 using a scale factor of 10000 was done, followed by determination of variable genes 725 using the vst method, and a z-score scaling. PCA was used to generate 75 PCs, 30 of 726 which were used for subsequent UMAP calculation and graph-based clustering with 727 tuned resolution. 728 For all differential gene expression analysis we utilized the Seurat implementation of 729 MAST (model-based analysis of single-cell transcriptomes) with the number of UMIs 730 included as a covariate (proxy for cellular detection rate (CDR)) in the model 33 . 731 732 NicheNet workflow: