The clinical trial was conducted as a phase I/II non-randomized trial and was planned to include up to 12 patients. Patients were treated at the National Center for Cancer Immune Therapy and Department of Oncology, Herlev University Hospital (Herlev, Denmark, ClinicalTrials.gov identifier: NCT02354690).

The primary objective was to evaluate the safety and feasibility of the study treatment. Secondary objectives were to determine ORR by response evaluation criteria in solid tumors (RECIST) V.1.1,25 progression-free survival (PFS), OS and immunological responses.

Eligible patients were between 18–70 years of age, had histologically confirmed diagnosis of MM harboring a mutation in the BRAF gene, American Joint Committee of Cancer stage III or IV, with tumor available for surgical removal and at least one additional measurable lesion according to RECIST V.1.1, had an Eastern Cooperative Oncology Group performance status of 0–1,26 had a life expectancy >3 months, and adequate organ function. Main exclusion criteria were uveal melanoma, untreated or symptomatic brain metastasis, a history of autoimmune disease or chronic infections, and prior treatment with BRAFi.

After screening and inclusion, patients started vem 960 mg orally two times a day 7 days before surgery and until hospitalization. During hospitalization patients received cyclophosphamide 60 mg/kg/day (day −7 and −6) and fludarabine 25 mg/m2/day (day −5 to −1) as previously described.6 27 TILs were infused on day zero followed by continuous intravenous infusion of IL-2 6–8 hours later. IL-2 was administered according to the decrescendo regimen; 18 MIU/m² over 6 hours, 18 MIU/m² over 12 hours, 18 MIU/m² over 24 hours, and 4.5 MIU/m² over 24 hours repeated three times. Maximum dose of IL-2 administered was 135 MIU as 2 m² was set as the maximum body surface area. The treatment and monitoring schedule is shown in online supplemental figure S1.

Prophylactic antibiotics, antiemetics, granulocyte colony stimulating factor and other supportive treatment were administered as previously described.6 Dose reductions of vem and IL-2 followed standard guidelines.

Toxicity was assessed using Common Terminology Criteria for Adverse Events version 4.0. Assessments were performed continuously, and as minimum at baseline, 4 weeks after vem treatment initiation, daily during hospitalization and at subsequent follow-up visits.

Clinical efficacy was assessed by Flour-18-deoxyglucose-positron emission tomography/CT (FDG-PET) scans at baseline, before hospitalization, at week 6 and 12 after TIL infusion and every third month thereafter until disease progression. Objective responses were evaluated according to RECIST V.1.1.

Blood samples for immune monitoring were collected at baseline, before hospitalization, at discharge and at following evaluation visits. Blood was collected in heparinized tubes and was kept for a maximum of 4 hours until handling according to standard operating procedure. In brief, peripheral blood mononuclear cells (PBMCs) were separated using centrifugation on a Lymphoprep (Takeda, Roskilde, Denmark) density gradient. Vials of PBMCs were cryopreserved in medium containing 90% heat inactivated human AB serum (HS; Sigma Aldrich) and 10% dimethyl sulfoxide using controlled-rate freezing (Cool-Cell, Biocision) in a −80°C freezer and the next day moved to a −140°C freezer until further processing.

Serum samples were collected before TIL infusion, 2 hours after TIL infusion and every second day until discharge. Within maximum 4 hours, serum tubes were spun at 3000g for 10 min. Serum was aliquoted and immediately transferred to a −80°C freezer and subsequently stored at −140°C until further processing.

Surgically removed tumors were minced mechanically into 1–3 mm³ fragments and placed in 24-well culture plates (Nunc, Roskilde, Denmark) containing 2 mL culture medium consisting of 90% RPMI 1640 (Thermo Fisher Scientific), 10% HS, penicillin/streptomycin and fungizone (Bristol-Meyers Squibb) and 6000 IU/mL IL-2 (Proleukin, Novartis, Bazel, Switzerland). The plates were incubated in 37°C humidified atmosphere with 5% CO₂. Half the medium was replaced at day 5 and subsequently three times weekly. TILs were propagated following the minimally cultured or “young TIL” method as previously described.12 Initial expansion was successful when pooled cultures reached >50×10⁶ total cells. The pooled TILs were either cryopreserved or directly propagated following the rapid expansion protocol (REP). In the REP 20×10⁶ cells were cultured in 80% RPMI-1640 with 10% human serum/20% AIM-V medium containing 6000 IU IL-2 and irradiated (40 Gy) allogeneic feeder cells (PBMCs from at least three different buffy coats) in a ratio of 1:200. The REP was initiated in static culture flasks and transferred to the dynamic Wave bioreactor system (GE healthcare) as previously described.28 At day 14 of the REP, TILs were harvested, transferred to an infusion bag and immediately administered to the patient. Sterility testing and microbiological control were performed on all TIL cultures prior to REP and infusion.

Autologous tumor digests (TDs) were established from the same tumor lesion from which the TILs were generated. TD was obtained from fresh tumor fragments after overnight incubation with enzyme cocktails containing 1 mg/mL collagenase type IV (Sigma Aldrich) and 0.0125 mg/mL dornase alpha (Pulmozyme, Roche). The obtained single-cell suspension was passed through 70 µm strainers and cryopreserved immediately without further analysis.

For phenotype analysis, cryopreserved TILs were thawed and rested overnight in RPMI 1640 containing 10% HS. TILs were harvested, washed and resuspended in phosphate-buffered saline (PBS; Lonza, Basel, Switzerland) containing 0.5% bovine serum albumin (BSA; Sigma-Aldrich). TILs were stained in the dark at 4°C for 30 min, washed and resuspended in PBS and analyzed immediately. Antibodies used were the following: CD3-HV510, CD4-PerCP, CD8-BV421 (all from BD Biosciences), and Near Infra-Red Live/dead marker (NIR) (Invitrogen).

Data acquisition was performed using a FACS Canto II flow cytometer and analyzed using FACS Diva software (both BD Biosciences).

Assessment of anti-tumor reactivity was performed in a flow cytometry based assay staining for CD107a, intracellular interferon-γ (IFNγ) and tumor necrosis factor (TNF) as previously described.12 Briefly, in the presence of GolgiPlug (BD Biosciences, dilution 1:1000) and CD107a detection antibody, TILs were cocultured at 37°C for 5 hours with autologous TD at an effector to target ratio of 3:1, or alone (unstimulated sample). After coculture, cells were stained with surface antibodies and intracellular antibodies as described for phenotype analyses. In positive control wells, Staphylococcal enterotoxin B (Sigma-Aldrich) was added. The following antibodies were used: CD3-FITC, CD56-PE, CD8-PerCP, IFNγ-PE-Cy7, TNF-APC, CD107a-BV421, CD4-HV510 (all from BD Biosciences), and NIR (Invitrogen).

Tumor reactive cells were defined as cells staining double positive for any combination of CD107a, IFNγ and TNF. A positive anti-tumor response was defined as more than double the level of activity in paired unstimulated samples (background; TILs alone), at least 0.1% difference from background and a minimum of 50 positive events after subtraction of double positive events in paired unstimulated samples. In selected samples, digests alone were tested under the same coculture conditions, and we did not observe any significant spontaneous reactivity. TILs in the digests were smaller than the cultured TILs used in tumor-recognition assays and were easily gated out from the analyses; hence only cultured TILs were gated to calculate the proportion of tumor-reactive TILs.

Cytokines were measured in serum samples using a 13-plex Bio-Plex Pro Human cytokine kit (Bio-Rad Laboratories, Copenhagen, Denmark) including the cytokines IL-1β, IL-2, IL-4, IL-5, IL-6, IL-8, IL-10, IL-12(p70), IL-13, IL-17A, granulocyte-macrophage colony-stimulating factor, IFNγ and TNF. The assay was performed according to manufacturer’s instructions using a Luminex 200 and Bio-Plex manager software V.6.1 (Luminex Corporation, Austin, Texas, USA). Data analysis was performed with the STarStation V.2.3 software (Applied Cytometry Systems, Sheffield, UK).

DNA and RNA were extracted from baseline tumors, relapse samples (patients #09 and #13), PBMCs and TILs using AllPrep DNA/RNA Mini Kit (Qiagen). Whole exome libraries were constructed from tumor and PBMC DNA as described previously29 using SureSelectXT Clinical Research exome (Agilent) and sequenced on NextSeq500 to mean target depth 38x-176x (median 150x) in tumors and 69x-87x (median 75x) in MNC samples. For T-cell receptor (TCR) sequencing, TIL and tumor RNA were treated with DNAse I (Thermo Scientific) as per Illumina recommendation, and libraries were constructed from 500 ng of DNAse treated RNA using AmpliSeq Immune Repertoire Panel (Illumina) and sequenced on NextSeq500. RNA-seq was performed as previously reported.29 WES data analysis: Alignment, postalignment processing, and variant calling was performed as described30 using SAREK workflow.31 Somatic mutations were derived using Varscan as in30 and Strelka2 (V.2.8.232;). Single nucleotide variants detected by both callers constituted the data set, while insertions and deletions from varscan were used. Mutations were annotated using Annovar33 and only mutations in protein coding regions and at splice sites were retained. RNA-seq data analysis: RNA sequencing data from bulk tumor samples were processed using Tophat234 and Cufflinks35 as described previously.36 Isoform FPKM values of each gene were summed up and protein-coding genes were extracted. The data were then quantile-normalized using37 and log-transformed as log2(data+1). TCR-seq data analysis: Demultiplexed data were used to derive T cell clonotypes using MiXCR38 39 with the command “mixcr analyze amplicon” and the following settings: “--starting-material rna --adapters no-adapters --receptor-type TRB --region-of-interest CDR3 -−5-end no-v-primers -−3-end c-primers”. Clonotypes that were supported by fewer than 10 read counts were discarded; these constituted 10%–78% (median 54%) of all clonotypes per sample but accounted for only 0.1%–14% (median 2%) of total abundance. MiXCR output was further used to derive clonotypes metrics using VDJtools40 as described in.41 VDJtools FilterNonFunctional command was used to remove non-productive CDRs and CalcDiversityStats command was used to derive statistical metrics. For between-sample comparisons, metrics based on resampled data from VDJtools were used to account for different depth of sequencing between samples. Following metrics were used: “Observed Diversity” (the normalized number of clonotypes, named “Richness” throughout the manuscript for clarity and consistency with previous publications); “Normalized Shannon-Wiener index” (reflects distribution of the clonotypes within the TCR repertoire, “Evenness” throughout the manuscript); “Shannon-Wiener index” (a parameter that combines richness and evenness, “Diversity” throughout the manuscript). For comparative analyses between samples, nucleotide sequences of TCR were used. For overlap analysis of TIL and TUM clonotypes from each patient, VDJtools OverlapPair command was used to derive Morisita-Horn index. Venn diagrams were drawn using R package VennDiagram.42

PFS and OS was defined as time from vem treatment initiation until progression, death of any cause, or the date of data cut-off (December 14 2018). Survival was estimated using the Kaplan-Meier method and comparisons were performed with the Mantel-Cox test. Cytokine levels at baseline were compared between groups with PFS longer or shorter than median PFS using a t-test. Comparisons of TCR data were performed using a Mann-Whitney test either paired (intrapatient comparisons) or unpaired (patient group comparisons). Correlations were described using the Pearson and Spearman tests. All statistics were computed by GraphPad Prism V.5.0 or statistical software R.

A total of 13 patients were enrolled in the trial between November 2014 and April 2018. All patients had tumors removed and an initial TIL expansion culture was established. A brain metastasis measuring 25 mm was discovered in one patient on magnetic resonance scan before hospitalization for TIL infusion. Despite surgical and local stereotactic radiotherapy, the metastasis was not controlled, and it was deemed unsafe to treat with TIL therapy (due to high doses of IL-2). The patient was excluded from the trial and is not included in the data analyses.

Patient characteristics are summarized in table 1. Ten of twelve patients treated with TILs had stage M1c at inclusion, one had M1b and one had M1a. Patients had received a median of two prior systemic therapies, including one or more of the immune checkpoint inhibitors ipilimumab, pembrolizumab, or nivolumab administered alone or in combination with an anti-LAG3 antibody.

Patients received vem for a median of 41 days (range 26–61; online supplemental table S1). Overall patient PS improved following treatment with vem. Two patients (patients 07 and 11) had worse PS due to side effects from vem treatment (most notably arthralgia and swollen joints, and nausea, respectively). Treatment-related toxicity is summarized in table 2. No unexpected toxicity was observed, and all adverse reactions were manageable following standard guidelines.

During hospitalization, expected hematological toxicities attributable to the lymphodepleting chemotherapy regimen were seen (anemia, leukopenia and thrombocytopenia). The number of red blood cell and platelet transfusions, and days with neutrophils<0.5 × 10⁹/L were similar to previously reported data.6 All patients suffered from fatigue and/or dyspnoea and nausea at some point either during treatment with chemotherapy and/or IL-2. All patients developed neutropenic fever and were treated with antibiotics. In five patients, an infection was verified by microbial culture; one had bacteraemia and four had local infections at the site of the central venous catheter (n=2), in the urinary tract (n=1), or at the site of surgery (n=1).

IL-2 related toxicity was manageable and intensive care unit intervention was not needed. A median of 96% (range 86–100) of the planned IL-2 dose was administered (online supplemental table S1).

Six of 12 patients obtained a partial response (PR) already during the short-term treatment with vem (patients 02, 05, 06, 09, 10 and 12; figure 1A and table 3). The remaining six had minor tumor regression within the criteria of stable disease (SD). Six patients had further tumor regression fulfilling the criteria for objective response after TIL infusion (patients 03, 04, 05, 06, 07 and 09; marked in bold in table 3), and these patients had longer PFS and OS compared with TIL non-responders (p<0.002 and p=0.006; data not shown). In total, one patient (8.3%; patient 03) achieved a CR, seven patients (58.3%) achieved confirmed PR (one ongoing; patient 07), one had an unconfirmed PR and three patients (25%) had SD as best overall response, as depicted in figure 1B and noted in table 3.

The median PFS was 4.8 months, with two patients having ongoing responses for 43 and 23 months (patients 03 and 07, respectively; figure 1C and table 3). In both patients, remaining tumor lesions visible on a CT scan turned FDG-negative after TIL therapy. In three other patients (patients 04, 05 and 09), singular progressive lesions were treated with stereotactic radiotherapy (patient 04) or surgery (patients 05 and 09), and these patients were followed without active systemic treatment for an additional 9, 5 and 6 months until progression in other lesions.

The median OS was 15.2 months (figure 1D). This does also reflect that patients were included before BRAFi treatment and consequently were eligible for treatment with BRAFi alone or in combination with a MEK inhibitor after progression. At the time of data cut-off, five patients were still alive.

The young TIL cultures were established in a median of 23 days (range 13–34). A median of 86.6×10⁹ TILs (range 23.8–125) were infused on day 0, corresponding to a REP expansion fold of median 4213 (range 2200–6160; table 3). Young TILs were cryopreserved, and subsequently thawed, in four patients before being further propagated in the REP.

The TIL infusion product consisted of a median of 99.1% CD3⁺ cells (range 95.8–99.8). Of those, a median of 36.8% were CD4⁺ and 59.9% were CD8⁺ with high patient variability (table 3). TILs from the infusion product were cocultured with autologous TD. In 10 of 12 patients, a CD8⁺ T cell tumor-specific response was found (figure 2A), and in 5 of 12 a CD4⁺ response was found (figure 2B). The presence of a detectable response in vitro did not correlate with clinical response or PFS (data not shown). However, in vitro response was found in all patients with an objective response to TIL.

A multiplexed Luminex assay was performed on serum samples taken at baseline in an exploratory search for cytokine markers predicting response to treatment. Mean baseline levels of IL-6, IL-8, IL-10 and TNF were all numerically lower in patients with a PFS longer than 6 months (after TIL infusion) compared with patients with PFS shorter than 6 months (after TIL infusion; online supplemental figure S3A–D); however, the difference did not reach statistical significance when cytokines were evaluated individually (IL-6: P 0.25; IL-8: P 0.17; IL-10: P 0.09; TNF: P 0.24).

A series of biomarker profiles including whole exome-, RNA and TCR sequencing were performed using available patient samples. Nine of the 12 patients had baseline samples obtained after 7 days on vem inhibition with sufficient tumor content to obtain whole-exome sequencing data. Next to the known BRAF mutations, individual samples also harbored mutations in the melanoma driver genes PTEN, CDKN2A and TP53 (figure 3A). Overall, tumors had an average number of 283 (range 88–1564) somatic mutations; however, no difference in tumor mutational burden was found between patients with further tumor regression after TIL infusion (responders) and patients without further regression (non-responders) (p=0.41, figure 3B). Indeed, the tumor sample obtained from the patient achieving CR only harbored 170 somatic mutations. Next, we found no correlation between CD8 T cell tumor-specific response and tumor mutational burden (figure 3C). This suggests that intrinsic tumor genetic alterations are not predictive of response to vem primed TIL therapy. Previous studies have demonstrated that treatment with vem confers an increase in TILs.18 19 Our study was not designed to confirm this effect of vem treatment, but the tumors obtained for TIL production was stained with CD3, CD8 and SOX10 antibodies to assess T cell infiltration (figure 3D for representative staining). All tumors had some degree of T cell infiltration, but the infiltration degree was not correlated to therapy response (data not shown).

Next, we obtained gene expression data from tumor specimens from 8 patients using RNA sequencing. Comparing tumors from responder and non-responder samples displayed a high correlation in mean mRNA levels, with only a few genes having a mean log-fold change above two (online supplemental S4A), suggesting that expression levels are not fundamentally different between responders and non-responders. We then extracted genes involved in antigen presentation, interferon-gamma signaling, immune checkpoint molecules, T cell phenotype- and melanoma state markers, and the microenvironment cell population signatures (figure 3E); however, no significant difference was observed. CD8A expression values from tumors were not correlated with levels of tumor-reactive CD8 T cells (online supplemental figure S4B), indicating that the majority of present CD8⁺ cells are part of an unspecific immune response. We then hypothesized that an intrinsic melanoma cell state indicated by MITF mRNA levels and CD8⁺ T cell infiltration in combination is essential for predicting clinical response. When plotting MITF against CD8 mRNA levels we found three tumor specimens clustered on the left side of the plot. These three cases had a mean PFS and OS that were shorter than the samples clustered to the right of the plot. However, the number of cases was too low to reach statistical significance (figure 3F). This is in line with a recent study demonstrating that increase of melanocyte antigens improved response to immune checkpoint blockade.43

To study the TCR repertoire, we derived TCR sequencing data from tumors removed during vem treatment (n=12) and their corresponding TIL products (n=12). We detected a variable number of clonotypes per sample (ie, richness normalized by the depth of sequencing), defined by different nucleotide TCR sequences: median 12,775.5, range 2223–27 006. We then determined TCR evenness and diversity (figure 4A). There was no correlation between the richness of baseline tumors and TIL products (Pearson 0.05; Spearman 0.03), and there was no significant difference between these two groups (figure 4A). However, intriguingly, evenness was significantly lower in TIL products as compared with tumors (Wilcoxon paired test, p<0.001). We then compared TCR richness, evenness and diversity in TILs and tumor from responders to TILs and tumors from non-responders (figure 4B). Although not significant, the most considerable difference was observed in tumor TCR richness between responders and non-responders with responders having a lower value (figure 4B). Comparison of TCR repertoires of the baseline tumors to their corresponding TIL infusion product revealed low degree of overlap between the repertoires in all patients (online supplemental table S2). In particular, many of the most abundant clonotypes in TIL samples were utterly absent or present at only low frequencies in the corresponding tumors. These data suggest skewing of TCR repertoire during in vitro expansion, promoting certain clonotypes and corroborating a recent study.44 In contrast, many of the most abundant T cell clonotypes in tumors were not detected in TIL cultures, suggesting their inefficiency to expand in culture.

To explore the impact of vem primed TIL therapy on tumor evolution, we obtained tumor exome sequencing data during vem treatment and at relapse on TIL therapy from two patients. For both patients, only a limited number of new mutations were detected in the relapse samples (n=4 and n=5 missense mutations for patients 09 and 13, respectively), none of them affecting a known melanoma driver gene (figure 5A). We then compared the transcriptome using predefined signatures for immune cell subsets.45 46 In patient 09, the myeloid lineage displayed upregulation in the relapse while in patient 13, a general downregulation of all immune cell subsets in the relapse was found (figure 5B,C). TCR sequencing showed an increase in the number of clonotypes from baseline tumor (12,818) to relapse sample (19,443) with a decrease in evenness (from 0.84 to 0.59) in patient 09, and decrease (from 19 143 to 11 313 clonotypes, with unchanged evenness) in patient 13. The baseline and the relapse tumors from each patient shared few clonotypes, with most of the clonotypes being unique to the particular tumor (figure 5D). Overall, these data show few tumor mutations occurring during resistance development, while TCR repertoire evolves rapidly during evolution.