This study was conducted according to the Good Clinical Practice Guidelines and the Declaration of Helsinki. The study was approved by the Institutional Review Boards of Istituto Dermopatico dell’Immacolata (IDI)-IRCCS (510/3, April 2018). All patients enrolled in the study provided written informed consent. This study included patients with unresectable metastatic melanoma, stage IIIc or IV based on American Joint Committee on Cancer (AJCC, version 7) staging (15), enrolled for treatment with anti-PD-1 inhibitors at IDI-IRCCS. Peripheral blood samples were collected before therapy and at every therapy administration up to one-year, disease progression or vitiligo appearance. Nivolumab (Opdivo^®) was given at the dose of 480 mg every 4 weeks, pembrolizumab (Keytruda^®) at the dose of 200 mg every 3 weeks. Patients underwent physical examination and assessment of biochemical parameters monthly, whereas investigator-determined objective response was assessed radiologically with computed tomography scans approximately every 12 weeks after treatment initiation. Tumor response was classified as immune complete response (iCR), partial response (iPR), or stable disease (iSD), according to the immune response evaluation criteria in solid tumors (iRECIST 1.1) (16, 17). Therapy efficacy evaluation was based on best overall response (iOR) determined as best time point response according to iRECIST.

Whole blood samples were collected into vacutainer sodium citrate tubes (cat. no. 367704, BD Biosciences, Plymouth, UK) and PBMCs were isolated by Ficoll gradient centrifugation (GE Healthcare, Little Chalfont, UK). Cryopreserved PBMCs were stained with the following antibodies, as previously described (14): Panel 1- anti-human CD4 FITC (1:100) (cat. no. 130-114-531, Miltenyi Biotech, Auburn, CA, USA) (1:100), anti-human CRTh2-PE (1:150) (cat. no. 130-114-128, Miltenyi Biotech), anti-human CD161-PE/Dazzle594 (1:50) (cat. no. 339939, Biolegend, San Diego, CA, USA), anti-human CD3 PercP-Cy5.5 (1:300) (cat. no. 300327, Beckman Coulter, Brea, CA, USA), anti-human CXCR3-APC Alexa647 (1:40) (cat. no. 353711, Biolegend), anti-human CD8-APC Alexa700 (1:120) (cat. no. A66332, Beckman Coulter), anti-human CCR6-BV421 (1:30) (cat. no. 353407, Biolegend), anti-human PD1-BV650 (1:30) (cat. no. 564104, BD Biosciences), and LIVE/DEAD™ Fixable Aqua Dead Cell Stain Kit (1:200) (cat. no. l34957, Invitrogen, Waltham, MA, USA). Panel 2: anti-human CD4 FITC (1:100) (cat. no. 130-114-531, Miltenyi Biotech), anti-human CD3-ECD (1:100) (cat. no. IM2705U, Beckman Coulter), anti-human CD127-APC Alexa700 (1:200) (cat. no. 351343, Beckman Coulter), anti-human CD25-BV421 (1:60) (cat. no. 564033, BD Biosciences), anti-human PD1-BV650 (1:30) (cat. no. 564104, BD Biosciences), and LIVE/DEAD™ Fixable Aqua Dead Cell Stain Kit (1:200) (cat. no. l34957, Invitrogen). Samples were acquired using Cytoflex cytometer (Beckman Coulter) and analyzed using FlowJo-10 software version 10.3.0. Gating strategy for discrimination of different cell populations is described in Supplementary Figure 1 . Briefly, CD3 (CD3⁺); CD4 (CD3⁺, CD4⁺); CD8 (CD3⁺, CD8⁺); CD8-MAIT (CD3⁺, CD8⁺, CD161^high) as previously reported (18, 19); NK cells (CD3^-, CD56^dim); NK bright cells (CD3^-, CD56^high); TCR-γδ (CD3⁺, TCR-γδ⁺); B cells (CD3^-, CD19⁺); naïve CD4 T cells (CD3⁺, CD4⁺, CD45RA^high), memory CD4 T cells (CD3⁺, CD4⁺, CD45RA^-). For CD4 T cells we performed analysis as previously reported (20, 21): Th1 (CD3⁺, CD4⁺, CRTH2^-, CXCR3⁺, CCR6^-); Th17 (CD3⁺, CD4⁺, CRTH2^-, CXCR3^-, CD161⁺, CCR6⁺); Th1/17 (CD3⁺, CD4⁺, CRTH2^-, CXCR3⁺, CD161⁺, CCR6⁺); Treg (CD3⁺, CD4⁺, CD127^-, CD25^high). All cell populations were analyzed within alive cells, excluding debris and doublets.

Formalin-fixed, paraffin-embedded (FFPE) sample blocks of normal skin (2 patients), primary melanomas (10 patients), metastases (5 samples from 3 patients), or vitiligo in patients without concomitant melanoma (10 patients), were collected from the archive of the IDI-IRCCS Histopathology Unit. Biopsies from melanoma patients who developed vitiligo during therapy (10 patients) were surgically taken at the lesion margin, fixed in 10% formalin, and embedded in paraffin. Four-µm sections were obtained from each sample, dewaxed, and rehydrated. After quenching endogenous peroxidase, achieving antigen retrieval, and blocking non-specific binding sites, sections were incubated with the following antibodies: mouse monoclonal antibody anti-CD25 (cod. LSBio-B7396-50; LS Bio, Seattle, WA; 1:5); mouse monoclonal antibody anti-interleukin (IL)-17A (cod. AF-317-NA, R&D Systems, Minneapolis, MSP; 1:30); monoclonal antibody anti-CD56 (cod. NCL-CD56, Novocastra Scytek, Wetzlar, Germany; 1:20), mouse monoclonal antibody anti-CD3 (cod. A0452, Dako, Santa Clara, CA; 1:100), mouse monoclonal antibody anti-CD8 (cod. CM154A, Biocare Medical, Concord, CA; 1:75) overnight at 4°C in a humid chamber. Secondary biotinylated polyclonal Abs and staining kits were obtained from Vector Laboratories (Burlingame, CA). Immunoreactivity was visualized with peroxidase reaction using 3-amino-9-ethylcarbazole (AEC) in H₂O₂ and specimen counterstained with hematoxylin. As a negative control, primary Abs was omitted. Stained sections were analyzed with the AxioCam digital camera coupled to the Axioplan 2 microscope (Carl Zeiss AG, Oberkochen, Germany). Staining intensity was evaluated by quantitative analysis (Image J color deconvolution) in five adjacent fields of each section by two independent observers, blinded to the status of the specimens. For immunofluorescence, 4-µm sections were dewaxed, rehydrated, and after quenching endogenous peroxidase with 3% bovine serum albumin in 1x phosphate buffer, achieving antigen retrieval and blocking non-specific binding sites, sections were incubated with the monoclonal antibody anti-TCR V alpha 7.2 FITC conjugated (cod.130-123-929, Miltenyi Biotech; 1:50) and incubated for 1 hour at 37°C in a humid chamber. After a few washes, sections were closed using mounting medium with DAPI for stained nuclei (Antifade Mounting Medium, Vectashield, Vector Laboratories). Images were acquired with the ApoTome System (Zeiss) connected with an Axiovert200 inverted microscope (Zeiss); image analysis was performed with ZEN software (Zeiss). Fluorescence was evaluated by quantitative analysis (Image J color deconvolution) in three adjacent fields of each section by two independent observers, blinded to the status of the specimens.

Sample preparation. Five to 10 slices of 5 µm thickness from vitiligo or primary/metastatic melanoma biopsies from the same patient were obtained from FFPE samples and used for DNA extraction. For tissue samples, QIAamp DNA FFPE Tissue Kit (Qiagen, Valencia, CA) was used in conjunction with an incubation at 55°C for 4 hours to complete tissue lysis. Next, samples were incubated at 90°C for 1 hour to reverse formaldehyde modification of nucleic acids. For blood samples, DNA extraction from PBMCs was performed following the DNeasy Blood & Tissue Kit (Qiagen). Proteinase K was used for digestion and extraction of DNA following a blood/cell protocol with RNase treatment and using spin-column method. After isolation by QIAamp MinElute column (Qiagen), variable amounts of buffer were added to each column to elute the DNA. Samples were quantified using Dropsense96 and diluted for library preparation in DEPC water.

Library preparation. Extracted DNA was used for TCR Vβ analysis using the ImmunoSEQ hsTCRB Kit (Adaptive Biotechnologies, Seattle, WA), according to the manufacturer’s instructions. For PBMCs and metastasis samples, 1 µg of total DNA was used for duplicate, whilst for primary melanoma samples, 200-1000 ng of total DNA was used for duplicates. For FFPE vitiligo samples, the staring gDNA was poor (50-325 ng) and, when possible, sequencing was repeated in order to increase the output. Due to the post-hoc nature of this study, we decided to perform the sequencing as well, at the best of our conditions using at least 50 ng of starting gDNA. The somatically rearranged human CDR3 was amplified from genomic DNA using a two-step, amplification bias-controlled multiplex PCR approach (22–24). The first PCR consists of forward and reverse amplification primers specific for every V and J gene segment and amplifies the hypervariable complementarity-determining region 3 (CDR3) of the immune receptor locus. The second PCR adds a proprietary barcode sequence and Illumina^® adapter sequences (25, 26). Following purification with the Agencourt™ AMPure™ XP Reagent (Beckman Coulter, Inc.), 10 µl of each sample were pooled in equimolar concentration into three final libraries (46 libraries/pool). Final libraries were quantified using KAPA Library Quantification Kits (Roche, Switzerland, CH), diluted to 4 nM, and denatured using 0.2N NaOH.

Sequencing. CDR3 libraries were loaded at 20 pM on an Illumina NextSeq 550 system with 156 paired-end sequencing (Illumina, San Diego, CA). Sample data was generated using the immunoSEQ^® Assay (Adaptive Biotechnologies, Seattle, WA). For the runs, ~146.92M reads where generated and ~128.58M passing filter. The coverage for each sample varied with the highest being 92x. The release of sequencing data and the QC control were performed by Adaptive Biotechnologies technical support.

The starting material (gDNA) was not always compliant to manufacturer’s instructions. When possible, the sequencing was repeated using other specimens, to increase the output. As this was a small cohort, the threshold we established is considered exploratory.

Data Analysis. Raw sequence reads were demultiplexed according to Adaptive’ s proprietary barcode sequences. Demultiplexed reads were then further processed to remove adapter and primer sequences; identify and correct for technical errors introduced through PCR and sequencing; and remove primer dimer, germline, and other contaminant sequences. Data were filtered and clustered using both the relative frequency ratio between similar clones and a modified nearest-neighbor algorithm, to merge closely related sequences. The resulting sequences were sufficient to allow annotation of the V(N)D(N)J genes constituting each unique CDR3 and the translation of the encoded CDR3 amino acid sequence. V, D and J gene definitions were based on annotation in accordance with the IMGT database (www.imgt.org). The set of observed biological human CDR3 sequences were normalized to correct for residual multiplex PCR amplification bias and quantified against a set of synthetic human CDR3 sequence analogues (26, 27). Data was analyzed using the immunoSEQ Analyzer toolset (ImmunoSEQ Analyzer 3.0 https://clients.adaptivebiotech.com/). Libraries were sequenced and organized providing productive and non-productive sequences (CDR3 regions explicitly encoding a premature stop, and those predicted to put the receptor gene out-of-frame downstream of the CDR3 rearrangement, were considered non-productive). Productive TCR-β CDR3 frequencies were used for generating scatterplots. Additional analyses were performed using GraphPad prism (GraphPad Software) and the R graphical library ggplot2. Alluvial flow diagrams (28) were used to describe common TCR sequences that can be viewed as multiple streams that flow smoothly throughout different samples of the same patient (primary melanoma, vitiligo and, if present, metastasis). To generate an alluvial map, common V(D)J rearrangements data from ImmunoSeq Analyzer were used in order to generate a series of networks for shared TCR sequence across different specimens, and these networks were loaded into the alluvial generator (gg alluvial library; https://cran.r-project.org/web/packages/ggalluvial/vignettes/ggalluvial.html).

Details on V(D)J usage for each productive sequence in the samples are reported as VDJ usage Supplementary File . For each patient, the nucleotide sequence of CDR3 unique region of shared TCR-β clonotypes and the productive frequency are listed in Supplementary Table II .

For TCR Vβ statistical analysis, a limited number of findings were evaluated for statistical significance. Groups were compared using the Producting Simpson Clonality Index (ImmunoSEQ™ software) and Kruskal-Wallis nonparametric test was used to analyze the difference in TCR clonality between different sample types. Two-stage linear step-up procedure of Benjamini, Krieger and Yekutieli was used for post-hoc analysis.

For flow cytometry data, differences between pairs were analyzed by paired Student’s t-test, and multiple comparisons by two-way ANOVA test.

For tissue analyses, before proceeding with the hypothesis tests, we analyzed the data of each group through a Shapiro-Wilk test which highlighted a non-normal distribution of our observations. Furthermore, given some values ​​far from the mean for some groups and the limited data available, we decided to compare the experimental groups using a Kruskal-Wallis test which is a non-parametric test which highlights the differences between the medians. Statistical significance was set at p<0.05 and all statistical analyzes were conducted using GraphPad Prism software (La Jolla, CA, USA) and R.

To investigate whether the immune profile of melanoma patients undergoing anti-PD-1 immunotherapy was associated to the irAE vitiligo, we performed flow cytometry analyses on patients’ PBMCs. Twelve patients who had a positive response to anti-PD-1 immunotherapy were examined: six developed vitiligo during therapy, while the other six did not. After stratification of patients into two groups matched for sex and age, we compared immune cell frequency considering as the unique variant the development of vitiligo ( Table 1 ). Cells were isolated from blood samples taken before therapy initiation (T0), and on-treatment at each administration up to one year or at vitiligo development. For the comparison, two time points of collection were considered: before therapy (T0) and at the number of treatment when vitiligo onset was observed, for patients who developed vitiligo; before therapy (T0) and at treatment number 4 or 5, for patients who did not develop vitiligo. Control time points were chosen according to the onset of vitiligo, which did not occur earlier than the fourth therapeutic administration in the matching group. Our results showed a reduction of circulating CD8-mucosal associated invariant T (MAIT), T helper (Th)-17, and T-reg cells associated to vitiligo onset. Moreover, a reduction in natural killer (NK) CD56^bright cell frequency was detected during treatment in both patient groups, suggesting its association with a positive clinical response ( Figures 1A, B ).

To analyze immune cell subtypes potentially responding to anti-PD-1 immunotherapy, we examined PD-1 expression in different immune cell subsets. A reduction in the frequency of PD-1 expressing CD3⁺, CD4⁺, CD8⁺, Th1, Th1/17, and Th17 cells was seen in patients not experiencing vitiligo during therapy ( Figure 2A ). A trend of such a reduction was also observed in patients developing vitiligo but was significant only for Th1 cells. Conversely, as therapy cycles increase, increased PD-1 expression of NK CD56^bright cells distinguished patients developing vitiligo ( Figure 2A ). In order to evaluate the activation status of different immune cell subtypes, we examined the expression of CD69, classical early marker of lymphocyte activation due to its rapid appearance on the surface of the plasma membrane after stimulation of different immune cell subsets (29). As shown in Figure 2B , there were neither significant changes in CD69 expression during therapy nor differences in CD69 expression before therapy in patients who developed vitiligo or not. Supplementary Figures 2, 3 report the frequency data obtained for each patient at every time-point of treatment.

To analyze at the tissue level the role of immune cells differently modulated in melanoma patients developing vitiligo, we performed immunohistochemistry and immunofluorescence analyses on vitiligo specimens collected from ten melanoma patients treated with anti-PD-1 antibodies ( Table 2 ). We compared them with the specimens from ten patients developing vitiligo in the absence of concomitant melanoma or immunotherapy. Two normal skin specimens were used as a control. As shown in Figures 3 , 4 , we observed a trend of increased number of cells positive to an anti-CD25 antibody, that could correspond to T-reg cells or other activated conventional T cells, and a significant increment of cells stained with an anti-IL-17A antibody in vitiligo lesions induced by immunotherapy compared to the other vitiligo samples. Interestingly, this high number of IL-17-positive cells detected in immunotherapy-induced vitiligo lesions may suggest the recruitment of Th17 cells in situ, according to the decrease of circulating Th17 cells observed over treatment in this patient population ( Figure 1 ). Although not significant, we observed a lower amount of TCR V alpha 7.2 positive cells, possibly MAIT cells, in the immunotherapy-derived vitiligo patients compared to vitiligo samples without melanoma.

To further characterize T cells in vitiligo lesions of melanoma patients, we performed sequencing of CDR3 variable region of the β-chain of the T-cell receptor (TCR-seq) on gDNA extracted from samples of vitiligo biopsies. Moreover, from the same patient, we have also carried out a TCR sequencing of PBMCs at the time of vitiligo onset, and on retrospective samples from the same patient of primary melanoma and of metastasis sites, when occurred (n=4). Table 2 summarizes the melanoma patients included in the TCR-Seq, and their clinical features, whereas Supplementary Table I reports the most relevant metrics from the TCR-Seq for all the analyzed samples, including total and productive templates and rearrangements, as well as the productive and sample Simpson clonality of the TCR-seq.

The TCR sequence analysis revealed a differential productive TCR clonality among tissues (peripheral blood: range 0.0081-0.2048; n=10; primary melanoma: range 0.0035-0.2401; n=13; vitiligo tissues: range 0.0053-0.0609; n=10. metastasis: range 0.0078-0.199; n=5) and productive templates as shown in Figure 5B (p value=0.0139).

Notably, we observed common V(D)J TCR rearrangements between samples derived from the same patients at different time points ( Table 2 and Figure 5A ). Clonotypes shared by metastasis, surgically removed during the anti-PD-1 therapeutic period, and vitiligo samples are higher compared to the number of clonotypes shared by primary melanoma and vitiligo specimens. Interestingly, we observed the highest number of shared clonotypes between vitiligo and primary melanoma in the two patients achieving iCR (n=21 and n=15; Table 2 and Figure 5A ). Furthermore, shared TCR clonotypes have been sequenced in primary melanoma and in metastasis of all the four patients, underlying the possibility that tumor-infiltrating T cells sharing the same specificity, are present in both primary and metastatic sites.

The TCR sequences of PBMC samples has been performed at the time of vitiligo occurrence, and as expected PBMCs showed the lowest Simpson clonality respect to the other tissues. It was also noted that several TCRs sequenced in the PBMCs samples were already present in the primary melanoma or in the metastasis, as shown in Table 2 and in Figure 5 .

Flow cytometry analysis showed that frequency of Th17, NK, and CD8-MAIT cells was differently modulated in the blood of patients developing vitiligo compared to those who did not develop it. To investigate whether a correspondence could be seen between circulating immune cell profiles and those present in the tumor tissues, we analyzed by immunohistochemistry and immunofluorescence the primary melanomas of patients who developed vitiligo and primary melanoma of three patients who did not developed vitiligo during therapy, selected from patients listed in Table 1 (IMM-9, IMM-11, and IMM-12), as a tissue reference to T0 circulating samples. Lymph node metastases from patients VIT_02, VIT_04, and VIT_08 ( Table 2 ), who developed vitiligo, were chosen as a reference for on-going therapy reference tissues, in addition to vitiligo skin samples examined in Figures 3 , 4 .

No significant differences were observed for cells stained with anti-CD56 and anti-TCR V alpha 7.2 antibodies. We detected a trend of lower CD25 positive cells in the lymph nodal metastases in respect to primary melanomas ( Figure 6 ). Interestingly, a significative lower frequency of IL-17A positive cells was detected in primary melanomas of patients developing vitiligo during therapy in respect to patients not developing this irAE ( Figure 6 ).