Circulating mucosal associated invariant T cells identify patients responding to anti-PD1 therapy

Immune checkpoint inhibitors that maintain anti-tumor T cell response are used for treating patients with metastatic melanoma. Since the response to treatment is extremely variable, biomarkers are 35 urgently needed to identify patients who could benefit from such therapy. We combined single-cell 36 RNA-sequencing and multiparameter flow cytometry to determine changes in circulating CD8 + T 37 cells in patients with metastatic melanoma. A total of 28 patients starting anti-PD1 therapy were 38 enrolled and followed for 6 months: 17 responded to therapy, whilst 11 did not. The proportion of 39 activated and proliferating CD8 + T cells and of mucosal associated invariant T (MAIT) cells was 40 significantly higher in responders before starting therapy and was maintained over time. MAIT cells 41 expressed higher level of CXCR4 and produced more granzyme B; in silico analysis revealed that 42 they are present in the tumor microenvironment. Finally, patients with higher levels of MAIT 43 showed a better response to treatment. single-cell RNA sequencing (scRNA-seq) and high-dimensional flow cytometry, we identify mucosal associated invariant T (MAIT) cells as possible biomarker of response to anti-PD-1 therapy in patients with metastatic melanoma. mediating tumor regression that can be further amplified by targeting PD-1 or alternate immune checkpoints.


48
CD8 + T cells can drive adaptive immune responses against several types of human 49 malignancies, in particular those with higher mutational burden and neoantigen load 1 . These cells 50 are activated by tumoral antigens, undergo expansion, and can localize and kill infected or cancer 51 cells. However, prolonged exposure to cognate antigens often contracts the effector capacity of 52 T cells and attenuates their therapeutic potential. This process, collectively known as T cell 53 exhaustion, is characterized by limited proliferation, cytokine production and effector capacity, 54 metabolic rearrangement, increased inhibitory receptors expression and genome-wide accumulation 55 of epigenetic modifications at effector and memory-related gene loci 2 . Among inhibitory receptors, 56 programmed death 1 (PD-1) has been extensively studied, and is now targeted by therapies with 57 monoclonal antibodies that are capable to reinvigorate T cells in several cancer settings. However, 58 immune checkpoint inhibitors (ICI) mediate tumor regression only in a subset of patients, and the 59 mechanisms at the basis of therapeutic resistance are poorly known 3 . A number of studies have 60 initially focused on the mutational load of the tumor as well as on quality of the cells infiltrating the 61 tumor microenvironment, and revealed that increased mutational burden and the presence of CD8 + 62 T cells with stem-like qualities 4, 5 , among others, can predict the response to ICI 6-10 . However, 63 tumoral tissue may not be always accessible, thereby making the quest of circulating biomarkers an 64 absolute need. In this regard, recent studies have shown that responding patients have more large 65 clones (those occupying >0.5% of repertoire) post-treatment than non-responding patients or 66 controls, and this correlates with effector memory T cell percentage 11 suggesting that peripheral T 67 cell expansion could predict tumour infiltration and clinical response 12 . 68 Over the last decade, a pressing need to deeply interrogate immune cells either in the tumour 69 microenvironment and/or in blood has led investigators to integrate data obtained from traditional 70 approaches with those obtained with new, more advanced, single-cell technologies, capable to 71 define characteristics of immune cells at an unprecedented degree of resolution 13 . Using single-cell 72 RNA sequencing (scRNA-seq) and high-dimensional flow cytometry, we identify mucosal 73 associated invariant T (MAIT) cells as possible biomarker of response to anti-PD-1 therapy in 74 patients with metastatic melanoma. 75 76 C16, identifying highly proliferating Ki-67+ CD71+ effector cells equipped for cytotoxicity 112 (GNLY + ), whose relative frequency was higher in responder before starting therapy (p<0.001). This 113 difference remained stable also after treatment (p<0.01) ( Figure 1B). 114

MAIT cells are more abundant in responders as revealed by scRNA-seq 116
To further define the dynamics of T cells potentially involved in therapeutic response, we 117 performed scRNA-seq of isolated CD3 + ,CD8 + T cells from a total of 20 patients at T0, T1 and T2 118 after anti-PD-1. After quality control, 55,200 cells were deemed suitable for analysis. 119 Contaminating 3,498 cells NK cells, expressing TYROBP, FCGR3A, KLRB1 were removed from 120 the analysis. We obtained a total of 51,702 purified CD8 + T cells. Using a cTP-net, a deep neural 121 network trained on multi-omics data, we imputed surface protein abundances within the scRNA-seq 122 data to confirm T cell phenotype 21 (Supplementary Figure 2). homing properties were identified they expressed high level of KLRB1, SLC4A10,MAF and 135 CXCR4 10,24 . 136 Pseudotime analysis revealed that the differentiation process started from naïve T cells 137 towards terminally differentiated T cells passing through activated naïve T cell, transitional effector 138 memory T cells and effector memory T cells (Supplementary Figure 3). In this process, the 139 transcriptionally distinct MAIT cells belong to a different branch of the Pseudotime trajectory 140 compared to the rest of the T cells, albeit mapping close to effector memory T cells, in line with 141 their shared phenotypic identity 25 . 142 No main differences were found between R and NR in the amount of naïve, cytotoxic 143 terminally differentiated and activated naïve T cells, both before and after therapy, as revealed by 144 analysis of gene expression profiles by scRNA-seq (Supplementary Figure 4). The proportion of 145 activated effector memory T cells, reminiscent of C16 as defined by flow cytometry, was higher 146 after two cycles of therapy in R compared to NR ( Figure 2C, left panel). At the same time, 147 activated effector memory T cells from R expressed higher levels of genes indicating activation 148 (FOS, DUSP1, FGFBP2, HLAC) and cytotoxic behaviour (GNLY, GZMH), thereby suggesting 149 heightened functional capacity in R ( Figure 2C, right panels). 150 The proportion of MAIT cells was higher in R before therapy and after the first cycle of 151 therapy ( Figure 2D, left panel). This trend was visible also after the second cycle of therapy. 152 Similarly to EM T cells, also MAIT cells showed overexpression of genes related to cell activation 153 in R compared to NR before (TNFAIP3,NKG7,NFKBIA,JUND,ZNF331,RGCC)  proportion of activated MAIT compared to NR not only before therapy, but also after the first and 166 the second cycle ( Figure 3C). 167 We next used polychromatic flow cytometry to confirm these findings also at the protein 168 level. In this regard, we analysed the percentage and phenotype of MAIT cells, identified as 169 CD3 + CD8 + T cells that expressed TCRa7.2 and CD161 ( Figure 4A, left), and found marked 170 expansion of these cells in the circulation of R patients when compared to NR before therapy 171 ( Figure 4A, right). This difference waned after therapy introduction in line with scRNA-seq data. 172 Moreover, we found that the percentage of MAIT cells expressing the homing receptor CXCR4 173 increased after two cycles of therapy in R, but not in NR, that had a relevant variability 174

Level of MAIT cells before therapy identifies responder patients. 193
We next evaluated the prognostic significance of the levels of MAIT cells in the circulation 194 as predictive biomarker of the response to anti-PD-1 therapy. Flow cytometric analysis revealed 195 that, within CD8 + T cells, the median level of MAIT in the population of patients with metastatic 196 melanoma was 1.7%, thus this value was used as a cut-off to stratify patients. The main finding of our study is that patients who respond to ICI are characterized by a 205 different composition of T cell subpopulations compared to those who do not respond, that are 206 detectable before therapy initiation. The most relevant of these differences is at the level of MAIT 207 cells, an innate population of CD3 + T cells previously involved in early immunity against infection 208 in peripheral tissue. Although the direct role of MAIT in mediating anti-tumor immune responses in 209 melanoma is still under scrutiny, our data suggest that investigating MAIT cell frequency in the 210 peripheral blood could be considered a possible predictive marker of successful therapy. Following 211 introduction of ICI, R show differential dynamics of T cells compared to NR, involving the 212 expansion of activated effector memory cells showing features of immune activation, proliferation 213 and effector differentiation, as previously reported by other groups 27 . 214 During the last decade, the immune response mediated by T cells in cancer patients 215 assuming ICI has been deeply investigated by analysing both tumor-infiltrating lymphocytes and 216 circulating T cells. Patients with melanoma or non-small cell lung cancer are characterized by an 217 exhausted T cell phenotype along with impaired proliferation and low metabolic activation, and a 218 high oligoclonal repertoire 28-30 . Activation of CD8 + T cells has been considered a hallmark of 219 response to therapy, and indeed after 1 cycle of therapy, Ki67 (a marker of cell proliferation) was 220 found increased among effector memory cells 27, 31 . 221 We show here that even if before treatment R and NR were characterized by similar clinical 222 characteristic in terms of tumour burden and LDH level, activated effector memory T cells were 223 more abundant in R, which can reflect a more activated CD8+ T cell compartment. This was 224 particularly evident in MAIT cells. Circulating MAIT cells are a pro-inflammatory and cytotoxic 225 population within effector memory T cells 32 and can represent up to 10% of peripheral CD8 + T 226 cells. They recognize microbial proteins presented by MR1 and display homing properties, as they 227 express different homing and cytokine receptors. Furthermore, MAIT cells are deeply involved in 228 patrolling mucosae and orchestrating the immune response in this environment 33 . 229 The role of MAIT cells in cancer has been widely investigated. However, few studies have 230 investigated their role during therapy with ICI. It was found that MAIT cells were decreased in 231 blood and displayed an altered cytokine production in patients with cervical, colorectal, gastric, 232 hepatocellular carcinoma, lung cancer and multiple myeloma. Moreover, controversial data exist on 233 the prognostic benefit of MAIT cells in the tumour microenvironment, as it has been shown for 234 instance in hepatocellular carcinoma 33 . Recent studies also show that MAIT cells promote tumour 235 initiation, growth and metastases via tumour MR1 34 . 236 To the best of our knowledge, these are the first data that characterize MAIT cells in the 237 peripheral blood of patients treated with anti-PD-1. We found that in R compared to NR, at baseline 238 and after therapy introduction, i) the percentage of MAIT cells was higher; ii) MAIT cells displayed 239 enhanced expression of genes related to immune activation and effector functions; iii) the 240 percentage of MAIT cells expressing CXCR4 was higher in R after two cycles of therapy. 241 CXCR4-CXCL12 axis plays an important role in the interactions between cancer cells and 242 their microenvironment. This axis modulates the traffic of tumor cells to metastasis, and mediates 243 invasiveness, vasculogenesis and angiogenesis. However, pre-clinical melanoma models reported 244 that this pathway can be influenced by anti-cancer treatments 35 . 245 Hence, it is possible to hypothesize that, among other activities, the increased expression of 246 CXCR4 on MAIT cells induced by anti-PD1 therapy could facilitate their migration towards 247 metastases, where they could exert a pro-inflammatory and cytotoxic activity. To support this 248 hypothesis, we observed that MAIT cells from R expressed CD69, which is not only an activation 249 marker, but also a constitutively expressed marker of tissue residency. In immunotherapy-naive 250 melanoma patients, the intratumor presence of CD8 + ,CD103 + ,CD69 + T cells that are able to 251 significantly increase during anti-PD-1 therapy has been associated with improved survival 36 . 252 Very recently a population of MHC class-I-related molecule-restricted T cells belonging to 253 the family of MAIT cells (defined "MR1" T cells) has been described as a rare population able to 254 respond to a variety of tumor cells of different tissue origin, but not to microbial antigens 37 . Thanks 255 to its ability to kill several cancer cell lines expressing low levels of MR1 while remaining inert to 256 noncancerous cells, this population represents a subset with a great potential for cell therapy 257 approaches in several malignancies 38,39 . 258 We are well aware that this study has some limitations. The first is represented by the 259 relatively low number of patients enrolled in the study, the second by the lack of data regarding the 260 characterization of MAIT in the tumour microenvironment, and the analysis of a possible 261 mechanism responsible of a better prognosis. Thus, further studies are needed not only to confirm 262 the utility of MAIT as biomarkers, but also to demonstrate their therapeutic potential or to provide 263 actionable information about tumour's biology, which together holds great promise with respect to 264 realising "personalized" treatment of melanoma. 265 In conclusion, we provide evidence of the association between the frequency and the  Methods 451

Patients 452
The study was conducted on 28 patients with metastatic melanoma treated with standard-of-care 453 nivolumab or pembrolizumab. According to the RECIST, responders (n=17) were defined as 454 patients with complete response (CR), partial response (PR), stable disease (SD), or mixed response 455 (MR) of greater than 6 months with no progression, and non-responders (n=11) as patients with 456 progressive disease (PD). In particular, among responders, 41.2% had CR, 35.3% had a PR, 17.6% 457 had SD, and 5.9% (which corresponds to one patient) had a MR. The clinicopathologic 458 characteristics of patients are reported in Table 1 previously titrated to define the optimal concentration. Chemokine receptors were stained for 20 505 min at 37°C, whereas all the other markers were stained for 20 min at room temperature. 506 Intracellular detection of Ki-67 and granulysin was performed following fixation of cells with the 507 FoxP3/ transcription factor staining buffer set (eBioscience, ThermoFisher) according to 508 manufacturer's instructions and by incubating with specific mAbs for 30 min at 4°C. Samples were 509 acquired on a Cytoflex LX flow cytometer (Beckman Coulter, Hialeah, FL) equipped with six 510 lasers (UV,355 nm;violet,405 nm;blue,488;yellow/green,561 nm;red,638 nm;IR,808nm) and 511 capable to detect 21 parameters. Flow cytometry data were compensated in FlowJo by using single 512 stained controls, as above 41 . Gating strategy is shown in Supplementary Figure 5. 513 In parallel, thawed PBMC were rested for 4 hours at 37°C and then in vitro stimulated with 514 anti-CD3/CD28 (1µg/ml) (Miltenyi, Bergisch Gladbach, Germany) and suboptimal concentration 515 of IL-12 (2 ng/mL) (Miltenyi) and IL-18 (50 ng/mL) (R&D System, Minneapolis, MN) and a 516 combination of those 25 . A 11 parameter/10-color flow cytometer panel was optimized to identify 517 MAIT cells producing Granzyme (GRZM) A, GRZM B, TNF-a and IFN-g that were detected after 518 Table 3). For the quantification of intracellular cytokines, 519 cells were fixed with BD Cytofix/Cytoperm Fixation/Permeabilization Solution kit (BD 520

hours of incubation (Supplementary
Biosciences) according to the manufacturer's instructions. Samples were acquired on an Attune 521 NxT acoustic flow cytometer (ThermoFisher) equipped with four lasers (violet, 405 nm; blue, 488; 522 yellow/green, 561 nm; red, 640 nm) and capable to detect 14 parameters. Flow cytometry data were 523 compensated in FlowJo by using single stained controls as above. Gating strategy is shown in 524 Supplementary Figure 6. 525 526

High-dimensional flow cytometry data analysis 527
Flow Cytometry Standard (FCS) 3.0 files were analysed using FlowJo software V 9.6. Aggregates 528 and dead cells were removed from the analyses and identify CD3 + CD8 + T cells were gated. 10,000 529 CD8+ T cells per sample were exported and biexponentially transformed in FlowJo V10. Further 530 analyses were performed by a custom-made script that makes use of Bioconductor libraries and R 531 statistical packages 4 . Data were analyzed using the Phenograph algorithm coded in the Cytofkit 532 package (version 1.6.5; 42 ) in R (version 3.3.3). Parameter K was set at 60. Phenograph clusters 533 were visualized using tSNE. Clusters representing <0.5% were not analysed in subsequent analysis. Starting from a total of 74,405 cells, 55,200 were deemed suitable for analysis. Downstream 558 analysis was performed in R using Seurat v3.0 44 . Cells that had less than 10% of mitochondrial 559 genes, read counts of at least 150 genes and less than 1,500 genes were kept for the following 560 analysis. The quality of cells was assessed applying a threshold on the percentage of mitochondrial 561 genes, on number of UMI and gene count. A cluster of 200 cells featuring genes related to the 562 myeloid lineage was excluded from the analysis. Additional 3,010 cells were excluded due to 563 technical artifacts during library preparation. Genes expressed in less than three cells were 564 excluded, then each gene expression measurement was normalized by total expression in the 565 corresponding cell and multiplied by a scaling factor of 10,000 and natural log-transform the result. 566 Previous steps were performed on T0, T1, T2 dataset. Subsequently, all three datasets were 567 integrated yielding an expression matrix of 51,701 cells by 17,745 genes 45 . 568 Principal components were selected using the jackstraw and Elbow methods. The dimensional 569 reduction was performed using Uniform Manifold Approximation and Projection (UMAP) on the 570 previously selected principal components. Unsupervised clustering was performed by finding the 571 nearest neighbors (KNN) and then, to group the cells, a modularity optimization-based algorithm 572 was applied. 573 The resolution was selected using clustree package 46 . Differentially expressed genes were 574 identified using the FindAllMarkers function, and the top 15 genes for each cluster were visualized 575 in a heatmap. Differential expression analysis was performed between each cluster and all other 576 cells using a Wilcoxon rank-sum test. Genes were selected to be significant as logFC>0.3 and 577 adjusted p value <0.05. Cells from a single cluster were selected and re-clustered to identify the 578 presence of subpopulation. Comparative analyses across conditions inside of each cluster was 579 performed using FindMarkers, genes were considered as significant with logFC > 0.3 and adjusted 580 p value <0.05. Furthermore, a random subset was performed on all 51,701 cells selecting 4,000 581 cells and then a trajectory analysis was performed using Monocle v2 47 . 582 583 cTP-net analysis 584 The surface protein imputation was performed using a pre-trained deep neural network (cTP-net) 585 trained on PBMC processed using multi-omics approach (CITE-seq and REAP-seq) 21 . cTP-net 586