Identification of essential genes for cancer immunotherapy

Abstract

a–c, Kaplan–Meier survival plots of patient overall survival with the expression of antigen presentation genes B2M (a), TAP1 (b) and TAP2 (c) after ipilimumab immunotherapy. Patients were categorized into ‘high’ and ‘low’ groups according to the highest and the lowest quartiles of each individual gene expression (RPKM). Reported P values have been adjusted by multiplying the unadjusted P value by 3 to account for the comparison of the two extreme quartiles. Hazard ratio was calculated using Mantel–Haenszel test; with hazard ratio ranges: B2M (0.02–0.31), TAP1 (0.04–0.52) and TAP2 (0.12–1.07). Data are derived from 42 patients with melanoma from the Van Allen et al. cohort³. d, Schematic of the 2CT-CRISPR assay to identify essential genes for EFT. e, NY-ESO-1 antigen-specific lysis of melanoma cells after 24 h of co-culture of ESO T cells with NY-ESO-1^– SK23, NY-ESO-1⁺ SK23 and NY-ESO-1⁺ Mel624 cells (n = 3 biological replicates) at E:T ratio of 1. f, Survival of Mel624 cells modified through lentiviral CRISPR targeting of MHC class I antigen presentation/processing genes after introduction of ESO T cells. CRISPR-modified Mel624 cells were co-cultured with ESO T cells at E:T ratio of 0.5 for 12 h. Live cell survival (%) was calculated from control cells unexposed to T cell selection. Data are from three independent infection replicates. All values are mean ± s.e.m. ***P < 0.001 as determined by Student’s t-test.

The 2CT-CRISPR assay employs a gene-engineered CD8⁺ T cell to specifically target an antigen expressed in an HLA-class I-restricted manner (Extended Data Fig. 2a–d). We used primary human T cells transduced with a recombinant T cell receptor (TCR) specific for the NY-ESO-1 antigen (NY-ESO-1:157–165 epitope) presented in an HLA-A*02-restricted manner (ESO T cells) that we previously reported to mediate tumour regression in patients with melanomas and synovial cell sarcomas^17,18,19 (Fig. 1d, e, Extended Data Fig. 2a–c). We optimized the 2CT assay to control the selection pressure, alloreactivity and bystander killing²⁰ exerted by T cells by modulating the effector to target (E:T) ratio and length of co-incubation (Extended Data Fig. 2e–i). To test whether the loss of antigen presentation genes can directly compromise T-cell-mediated cell lysis of human cancer cells using our 2CT-CRISPR assay, we targeted TAP2 and B2M with three unique single-guide RNAs (sgRNAs) cloned into the lentiCRISPRv2 lentiviral vector in NY-ESO-1⁺ Mel624 melanoma cells. Fluorescence-activated cell sorting (FACS) analysis confirmed that B2M-targeting lentiviral CRISPRs achieved around 95% protein knockout (Extended Data Fig. 3a). Significant resistance was detected in the cells transduced with B2MsgRNAs (72 ± 5%) and with TAP2 sgRNAs (13 ± 2%) against the cytolytic activity of ESO T cells (Fig. 1f, Extended Data Fig. 3b, c). These results show that loss of key MHC class I genes promotes evasion of T-cell-mediated tumour killing in the optimized 2CT-CRISPR assay.

Genome-wide 2CT-CRISPR screen for EFT

To identify the tumour-intrinsic genes essential for EFT on a genome-scale, we transduced Mel624 cells with the genome-scale CRISPR knockout (GeCKOv.2) library at a multiplicity of infection (MOI) <0.3 (Fig. 2a). The GeCKOv.2 library consists of 123,411 sgRNAs that target 19,050 protein-coding genes (6 sgRNAs per gene) and 1,864 microRNAs (4 sgRNAs per microRNA), and also includes around 1,000 ‘non-targeting’ control sgRNAs²¹. We exposed transduced tumour cells to ESO T cells at E:T ratios of 0.3 and 0.5 for 12 h in independent screens that resulted in around 76% and around 90% tumour cell lysis, respectively. Using deep sequencing, we examined the sgRNA library representation in tumour cells before and after T cell co-incubation (Extended Data Fig. 4a, b). We observed that the distribution of the sgRNA reads in T-cell-treated samples versus controls was significantly altered in screens with the higher number of T cells, E:T of 0.5 (Kolmogorov–Smirnov test, P = 7.5 × 10⁻⁵), and not with an E:T of 0.3 (Extended Data Fig. 4b, P = 0.07), indicating that the efficiency of this 2CT-CRISPR assay was dependent on the selection pressure applied by T cells.

 

a, Design of the genome-wide 2CT-CRISPR assay to identify loss-of-function genes conferring resistance to T-cell-mediated cytolysis. b, Scatterplot of the normalized enrichment of the most-enriched sgRNA versus the second-most-enriched sgRNAs for all genes after T-cell-based selection (inset). The top 100 genes by second-most-enriched sgRNA rank are displayed in the enlarged region. c, Identification of top enriched genes using the RIGER analysis. d, Consistency of multiple sgRNA enrichment for the top 20 ranked genes by second-most-enriched sgRNA score. The number of sgRNAs targeting each gene that are found in the top 5% of most-enriched sgRNAs overall is plotted. e, Schematic of MHC class I antigen processing pathway with candidate genes scoring in the top 0.1% of all genes in the library highlighted.

We quantified consistent enrichment of candidate genes by multiple methods: (1) ranking genes by their second-most-enriched sgRNA (Fig. 2b); (2) the RNAi Gene Enrichment Ranking (RIGER) metric²² (Fig. 2c); and (3) the number of sgRNAs for each gene enriched in the top 5% of all sgRNAs in the library (Fig. 2d). All three methods showed a high degree of overlap (Fig. 2b–d, Extended Data Fig. 4c, Supplementary Table 1). Despite the disparity in the enriched sgRNA distributions between screens with E:T of 0.3 and 0.5, several highly-ranked genes and microRNAs are shared between the screens (Extended Data Fig. 4d–f). On the basis of our initial optimization of the 2CT-CRISPR assay, we expected that genes directly associated with MHC class I antigen processing and presentation would be enriched in our screens, and found that HLA-A, B2M, TAP1, TAP2, and TAPBP were among the most highly-enriched genes in our screen (Fig. 2d, e). In addition, many genes without an established connection to the EFT were ranked amongst the top 20 enriched genes in this genome-scale analysis, such as SOX10, CD58, MLANA, PSMB5, RPL23 and APLNR (Fig. 2c, d).

a, Schematic of the array validation strategy for top candidates from the pooled 2CT-CRISPR screen to test generalization across melanomas, MHC class I antigens and other (non-melanoma) cancer. b, Survival of HLA-A*02⁺ NY-ESO-1⁺ A375 cells edited with individual sgRNAs (2–4 per gene) after co-culture with ESO T cells at E:T ratio of 0.5 (n = 3 biological replicates per sgRNA). c, Shared and unique genes essential for EFT validated in Mel624 and A375 melanoma cells. d, Survival of HLA-A*02⁺ NY-ESO-1⁺ renal cell carcinoma 2245R cells edited with individual sgRNAs (4 per gene) after co-culture with ESO T cells at E:T ratio of 0.5 in 2CT assay (n = 4 biological replicates per sgRNA). b–d, Grey dots indicate sgRNAs without a significant change in cell survival. Coloured dots (purple or green) indicate P < 0.05. Pvalue calculated for positively enriched gene-targeting sgRNAs compared to control sgRNA by Student’s t-test. Data representative of at least two independent experiments.

To test whether these validated genes generalize to an unrelated T cell receptor antigen (MART-1), we introduced sgRNAs targeting several of these genes into MART-1/MLANA⁺ Mel624 cells and then co-cultured them with MART-1 TCR-transduced primary human T cells¹⁸ (MART-1 T cells; Extended Data Figs 2d, 6b). MART-1 T cells are highly potent against tumour cells owing to their high avidity TCRs and bystander destruction of cancer cells^18,20. Tumour cells transduced with control non-targeting sgRNAs were completely eradicated upon co-culture with MART-1 T cells, whereas ESO T cells only induced cytolysis of ~60% of tumour cells with the same incubation (Supplementary Fig. 3). MART-1⁺ Mel624 cells with sgRNAs targeting each of the 9 candidate genes showed increased resistance against these T cells (Extended Data Fig. 6b, P < 0.05). We confirmed on-target gene disruption efficiency of sgRNAs by performing deep sequencing analysis at the target Cas9 cut-sites in these melanoma cells (Extended Data Figs 6c–e, 7). The genes that we validated across different melanoma cell lines and different antigen–TCR combinations include cellular cytoskeleton genes, COL17A1(collagen type XVII alpha 1) and TWF1 (twinfilin-1), a microRNA (hsa-mir-101-2) and a 60S ribosomal subunit (RPL23, Supplementary Discussion). These findings underscore the role of these biological processes in tumour cells in modulation of EFT.

Next, we tested whether immune resistance due to the loss of the cellular signalling-related genes APLNR, encoding a G-protein-coupled apelin receptor, and BBS1, encoding Bardet–Biedl syndrome 1 protein, and the adhesion-related gene CD58 can be generalized to other cancer types. To this end, we overexpressed NY-ESO-1 (CTAG1B) in a patient-derived HLA-A*02⁺ renal cell carcinoma (2245R) cell line using a retroviral vector and targeted each gene with four sgRNAs. For each gene, APLNR, BBS1 and CD58, we found that at least two sgRNAs exhibited >50% resistance to ESO T-cell-mediated lysis (Fig. 4d, P < 0.05). Thus, we validated key genes found in our CRISPR screen using multiple cell lines, target antigens and tumour cell types. To better understand the function of a gene previously unknown to have a role in EFT, we focused on APLNR, which was validated across different cell lines and antigens.

In vivo relevance of APLNR

APLNR is a G-protein-coupled receptor that is mutated in several cancers²⁷. To determine whether loss-of-function mutations in APLNRare present in patient tumours treated with T-cell-based immunotherapies, we mined available whole-exome sequencing datasets^3,5,28,29 from patients with metastatic melanoma and lung cancer treated with checkpoint blockade therapies including anti-CTLA4 (ipilimumab) and anti-PD1 (pembrolizumab or nivolumab) antibodies. From these datasets, we identified seven non-synonymous mutations in APLNR, one of which is a nonsense mutation (W261X), resulting in deletion of the seventh transmembrane helix and cytoplasmic C-terminal tail of the receptor (Fig. 5a, Extended Data Fig. 8a). The C terminus of APLNR contains residues critical for membrane association, receptor internalization, receptor dimerization, and interaction with cytosolic proteins³⁰, and thus its loss is likely to be deleterious to protein function. We also carried out whole-exome sequencing of a metastatic lung lesion resected from a patient (SB-4044) with melanoma resistant to both anti-CTLA4 (ipilimumab) and anti-PD1 (nivolumab) immunotherapies. We found two non-synonymous mutations, T44S and C181S, in APLNR (Fig. 5a, Extended Data Fig. 8a) within this melanoma lesion.

a, Non-synonymous mutations detected in APLNR in human tumours refractory to indicated immunotherapies. Ipi, ipilimumab; nivo, nivolumab; pembro, pembrolizumab; CS, cytoplasmic signalling domain; EC, extracellular domain. b, Mutations encoding G349E and T44S in APLNR resist the restoration of T-cell-mediated cytolysis in APLNR-perturbed tumour cells (n = 3 biological replicates). c, Second-most-enriched sgRNA score in CRISPR screens for 96 APLNR-interacting proteins from the BioGrid database³¹. d, Western blot probed for APLNR after immunoprecipitation pull-down of JAK1 from A375 cell lysates. Results were similar in an independent replicate experiment (not shown). e, Quantitative PCR analysis of JAK1–STAT1 pathway induced genes in wild-type and APLNR-edited cells at 0, 8 and 24 h post treatment with 1 μg ml⁻¹ IFNγ (n = 3 biological replicates). f, Induction of surface expression of β2M on APLNR-edited cells upon co-culture with ESO T cells as measured by FACS (n = 3 biological replicates). g, IFNγ secretion from ESO T cells after overnight co-culture with CRISPR edited A375 cells determined using ELISA (n = 3 biological replicates). h, Schematic of in vivo experiments to test the response of B2m and Aplnr knockout tumours to ACT in immunocompetent mice. i, j, Subcutaneous tumour growth in mice receiving ACT of Pmel-1 T cells. Tumour area (i) and overall survival (j) are shown. Significance for tumour growth kinetics were calculated by Wilcoxon rank-sum test. Survival significance was assessed by a log-rank Mantel–Cox test (n = 5 mice per ‘no treatment’ groups). For ‘Pmel ACT’ groups, n = 9 mice in control group, n = 10 mice per B2m-sg and Aplnr-sg groups. All values are mean ± s.e.m. ****P < 0.0001, **P < 0.01, *P < 0.05. Data are representative of two independent experiments.

We selected four non-synonymous mutations (T44S, C181S, P292L, G349E) from the Van Allen et al. cohort³ and patient SB-4044 to determine whether these mutations can limit EFT. In APLNR-knockout A375 cells, we re-introduced mutated or wild-type APLNR using lentivirus (Extended Data Fig. 8b). Although re-introduction of wild type APLNR, or APLNR with C181S or P292L mutations rescued the sensitivity of APLNR-knockout cells to T-cell-mediated cytolysis, T44S and G349E mutations only resulted in partial rescue, indicating that these mutations in APLNR reduce EFT (Fig. 5b). The presence of these loss-of-function mutations in non-responding tumour lesions suggests that the functional loss of APLNR in tumours could be associated with immunosuppression in vivo.

To investigate potential mechanisms by which APLNR regulates EFT, we analysed the transcriptome of APLNR-knockout cells using RNA-sequencing. We did not find any substantial differences in mRNA transcript levels of genes involved in antigen presentation, T cell inhibition or co-stimulation (Supplementary Fig. 4). We thus examined whether APLNR regulates EFT by modulating protein signalling. APLNR has been reported to interact with 96 different intracellular proteins (BioGRID interaction database³¹), and in genome-scale CRISPR screens, targeting multiple genes in the same pathway can result in similar phenotypes. Of these 96 previously reported binding partners of APLNR, JAK1 was the most enriched in our screen (enriched in the top 0.5% of all genes, Fig. 5c). Using immunoprecipitation pull-down, we confirmed that APLNR binds to JAK1 in A375 and HEK293T cells (Fig. 5d, Extended Data Fig. 9a). Retroviral overexpression of APLNR coincided with an increase in JAK1 protein levels (Extended Data Fig. 9b). In addition, this overexpression of APLNR significantly increased the sensitivity of tumour cells to EFT (Extended Data Fig. 9c, increase in cytolysis from 71.7 ± 1% to 81.1 ± 0.9%).

IFNγ-driven phosphorylation of JAK1 stimulates the JAK–STAT signalling cascade to augment antigen processing and presentation in tumours, which in turn enhances recognition and cytolysis by T cells³². First, we determined whether apelin treatment of tumour cells can directly induce phosphorylation of JAK1. We did not observe any changes in JAK1 phosphorylation with apelin treatments (Extended Data Fig. 9d). We also tested whether activation of APLNR using peptide ligands ([Pyr¹]apelin 13, apelin 17 and apelin 36) or a chemical agonist (ML233) on tumour cells alters EFT. We did not observe any significant effect of these treatments, suggesting that APLNR may regulate EFT independent of its canonical G-protein-coupled receptor signalling. Next, we tested whether loss of APLNR alters the responsiveness of tumour cells to IFNγ treatment. We found that APLNR-knockout cells exhibited reduced induction of JAK–STAT signalling upon IFNγ treatment, as measured by the phosphorylation status of JAK1 and STAT1 (Extended Data Fig. 9d, e) and induction of specific IFNγ response gene transcripts (Fig. 5e, Extended Data Fig. 9f). Upon co-culture with ESO T cells for 6 h, APLNR-knockout cells induced less β2M expression on the cell surface compared to unedited cells (Fig. 5f, Extended Data Fig. 9g). In agreement with the reduced sensitivity of APLNR-knockout cells to IFNγ-mediated upregulation of antigen processing and presentation genes, we noted a significant decrease in the recognition of APLNR-knockout tumour cells by ESO T cells (Fig. 5g, Extended Data Fig. 9h, Supplementary Fig. 5). Taken together, these data suggest that APLNR augments EFT by modulating IFNγ signalling in tumour cells.

Finally, to test the relevance of APLNR for immunotherapy in an in vivosetting, we targeted Aplnr in mouse B16 melanoma tumours with multiple sgRNAs (Fig. 5h). We also included B2m-targeting sgRNAs as a positive control. For each target gene, we performed western blot analysis on the cells to select the sgRNAs with the highest protein reduction (Supplementary Fig. 6). Upon subcutaneous implantation of these cells in immunocompetent C57BL/6 mice, we did not observe any significant changes in tumour growth kinetics between untreated control cells and CRISPR-modified cells (Fig. 5i). However, after treatment of these tumours with adoptive transfer of tumour-specific Pmel-1 TCR transgenic T cells (which target melanoma antigen gp100)³³, B2m knockout in these tumours significantly diminished the efficacy of adoptive cell transfer (ACT) treatment as measured by tumour growth (Wilcoxon rank-sum test, P = 0.0002, Fig. 5i) and host survival (log-rank test, P < 0.0001, Fig. 5j). Similarly, for Aplnr-knockout tumour cells, we observed a significant reduction in the tumour clearance (P = 0.0042, Fig. 5i) and host survival (P = 0.019, Fig. 5j) after ACT. As an orthogonal method to our genome-edited B16 tumours, we confirmed that short hairpin RNA (shRNA)-mediated Aplnr knockdown in B16 using two independent hairpins also reduced ACT efficacy (Extended Data Fig. 10). In addition, we found that Aplnr-knockout in B2905 mouse melanoma reduced the efficiency of anti-CTLA4 blockade in vivo with 0 out of 10 mice experiencing a complete regression with sg-Aplnr-transduced cells versus 5 out of 10 mice with cells transduced with control sgRNA (Supplementary Fig. 7). Taken together, these data demonstrate that APLNR loss reduces the effectiveness of T-cell-based cancer therapies, including immune checkpoint blockade and ACT.

Discussion

To capture the genes necessary for a complex phenomenon such as the EFT in cancer immunotherapy, we developed a 2CT-CRISPR screening system and used it to identify several genes capable of modulating melanoma growth when targeted by T cells. Given the association of these genes with intratumoral cytolytic activity in pan-cancer TCGA datasets, this resource provides a rich trove of new targets to investigate resistance to T-cell-based therapies. Reverting or bypassing the functional loss of such genes in tumours using drugs such as epigenetic modulators^34,35 may allow development of new combinatorial treatment strategies to improve clinical responses with immunotherapies.

Our data reveals a novel role of APLNR in regulating the anti-tumour response of T cells via modulation of the JAK–STAT signalling in target cells. It has been shown that angiotensin receptor, a G-protein-coupled receptor closely related to APLNR, also directly regulates the JAK–STAT pathway³⁶. A recent study has shown that T cell effector cytokines IFNγ and TNFα exert an anti-tumour effect by altering endothelial cells of blood vessels (where APLNR is highly expressed³⁷) to induce ischaemia in the tumour stroma³⁸. Given our finding that APLNR interacts with JAK1 to augment IFNγ response, APLNR can further increase the sensitivity of tumour blood vessels to IFNγ and thus improve anti-tumour efficacy of T cells. Hence, we speculate that APLNR might have a role beyond direct tumour cell recognition in vivo, and its expression on tumour blood vessels and stromal cells should be investigated in future studies.

In summary, we have used a two-cell-type CRISPR screen to discover both well-established and novel genes in cancer cells that regulate EFT. Our findings have direct clinical implications as these data may serve as a functional blueprint to study the emergence of tumour resistance to T-cell-based cancer therapies. Recent clinical trials have raised the question of why most patients fail to experience complete regressions of their cancers⁷. Our study provides a comprehensive list of genes that can contribute to resistance of human tumours to immunotherapy and have also identified loss-of-function mutations shared with those observed in patients that fail to respond to immunotherapy (Supplementary Discussion, Supplementary Fig. 8). A careful evaluation and validation of mutations in these genes on a personalized basis in immunotherapy-resistant patients may allow identification of novel mechanisms of immune escape and speed the development of new drugs that circumvent these escape mechanisms.

Methods

Data reporting

Sample size for the mouse experiments was determined on the basis of our previous experience with B16 and B2905 tumour models. Tumours were randomized during treatments. The measurements for implanted tumours were performed in double-blinded fashion.

Human specimens

Peripheral blood mononuclear cells (PBMCs) were isolated from healthy donors and tumour samples were isolated from patients with melanoma. All human specimens were collected with informed consent and procedures approved by the institutional review board (IRB) of the National Cancer Institute (NCI).

Mice

Animal experiments were approved by the Institutional Animal Care and Use Committees of the NCI and performed in accordance with NIH guidelines. C57BL/6J were purchased from The Jackson Laboratory. Female mice aged 6–8 weeks were used for tumour implantation experiments.

Cell culture

The melanoma cell lines HLA-A*02⁺/MART-1⁺/NY-ESO-1⁺ (Mel624.38, Mel1300), HLA-A*02⁻ (Mel938) and HLA-A*02⁺/NY-ESO-1⁻ (Mel526) were isolated from surgically resected metastases as previously described¹⁷, and were cultured in RPMI 1640 (Invitrogen) medium supplemented with 10% fetal bovine serum (FBS, Hyclone, Logan, UT). The A375 (HLA-A*02⁺/NY-ESO-1⁺) cell line was obtained from the American Type Culture Collection (Manassas, VA). The SK23 cell line transduced with retrovirus containing NY-ESO-1 expressing vector (SK23 NY-ESO-1⁺) and HLA-A*02⁺ renal cell carcinoma (2245R) cells was obtained from Ken-ichi Hanada (Surgery Branch, NCI) and cultured in RPMI 1640 medium supplemented with 10% FBS. A375 cells were cultured in D10 medium containing DMEM supplemented 10% FBS, 2 mM L-glutamine, and 1% penicillin–streptomycin (Life Technologies). Cell lines were confirmed mycoplasma negative using Mycoplasma detection kit (Biotool #B3903). All PBMCs and lymphocytes used for transduction and as feeder cells were obtained from aphereses of NCI Surgery Branch patients on IRB-approved protocols. They were cultured in T cell medium, which is: AIM-V medium (Invitrogen) supplemented with 5% human AB serum (Valley Biomedical, Winchester, VA), 100 U ml⁻¹ penicillin and 100 μg ml⁻¹ streptomycin, 2 mM L-glutamine and 12.5 mM HEPES (Life Technologies).

Retroviral vectors and transduction of human T cells

Retroviral vectors for TCRs recognizing the HLA-A*02-restricted melanoma antigens NY-ESO-1 (NY-ESO-1:157–165 epitope) and MART-1 (MART-1:27–35 epitope, DMF5) were generated as previously described^17,18. Clinical grade cGMP-quality retroviral supernatants were produced by the National Gene Vector Laboratories at Indiana University. For transduction, peripheral blood lymphocytes (2 × 10⁶ cells per ml) were stimulated with IL-2 (300 IU ml⁻¹) and anti-CD3 antibody OKT3 (300 IU ml⁻¹) on day 0. Non-tissue culture treated six-well plates were coated with 2 ml per well of 10 μg ml⁻¹ RetroNectin (Takara Bio) on day 1 and stored overnight at 4 °C. Vector supernatant (4 ml per well, diluted with D10 media) were applied to plates on day 2 followed by centrifugation at 2,000g for 2 h at 32 °C. Half the volume was aspirated and peripheral blood lymphocytes were added (0.25–0.5 × 10⁶ cells per ml, 4 ml per well), centrifuged for 10 min at 1,000g, and incubated at 37 °C/5% CO₂. A second transduction on day 3 was performed as described above. Cells were maintained in culture at 0.7–1.0 × 10⁶ cells per ml. After collection, cells underwent a rapid expansion protocol in the presence of soluble OKT3 (300 IU ml⁻¹), IL-2 (6,000 IU ml⁻¹) and irradiated feeders as previously described¹⁸. After day 5 of the rapid expansion protocol, cells were frozen down for co-culture later.

Lentivirus production and purification

To generate lentivirus, HEK293FT cells (Invitrogen) were cultured in D10 medium. One day before transfection, HEK293FT cells were seeded in T-225 flasks at 60% confluency. One to two hours before transfection, DMEM media was replaced with 13 ml of pre-warmed serum-free OptiMEM media (Invitrogen). Cells were transfected using Lipofectamine 2000 and Plus reagent (Invitrogen). For each flask, 4 ml of OptiMEM was mixed with 200 μl of Plus reagent, 20 μg of lentiCRISPRv2 plasmid or pooled plasmid human GeCKOv.2 (Genome-scale CRISPR Knockout) library, 15 μg psPAX2 (Addgene) and 10 μg pMD2.G (Addgene). 100 μl Lipofectamine 2000 was diluted with 4 ml of OptiMEM and was combined to the mixture of plasmids and Plus reagent. This transfection mixture was incubated for 20 min and then added dropwise to the cells. 6–8 h after transfection, the media was replaced to 20 ml of DMEM supplemented with 10% FBS and 1% BSA (Sigma). Virus containing media was collected 48 h post-transfection. Viral titer was assayed with Lenti-X GoStix (Clontech). Cell debris were removed by centrifugation of media at 3,000 r.c.f. and 4 °C for 10 min followed by filtration of the supernatant through a 0.45 μm low-protein binding membrane (Millipore Steriflip HV/PVDF). For individual lentiCRISPRv2 plasmids, viral supernatants were frozen in aliquots at −80 °C. For pooled library plasmids, viral supernatants were concentrated by centrifugation at 4,000 r.c.f. and 4 °C for 35 min in Amicon Ultra-15 filters (Millipore Ultracel-100K). Concentrated viral supernatants were stored in aliquots at −80 °C.

2CT T cell and tumour cell co-culture assay

NY-ESO-1 and MART-1 T cells were used for co-culture assays. Two days before co-culture, cells were thawed in T cell media containing 3 U ml⁻¹ DNase (Genentech Inc.) overnight. Tumour cells were seeded at specific density on this day in the same media as the T cells. T cells were then cultured in T cell media containing 300 IU ml⁻¹ IL-2 for 24 h. T cells were co-cultured with tumour cells at various effector:target (E:T ratios) for varying time periods. To reduce T cell killing activity and enrich for resistant tumour cells during the recovery phase, T cells were removed by careful 2× phosphate-buffered saline (PBS) washes following the addition of D10 media without IL-2. At the end of recovery phase of co-culture, tumour cells were detached using trypsin (Invitrogen) and washed twice with PBS. Tumour cells were then stained with fixable Live/Dead dye (Invitrogen) followed by human anti-CD3 antibody (clone SK7, BD) in FACS staining buffer (PBS + 0.2% BSA). Cell counts were measured using CountBright Absolute Counting Beads (Invitrogen) by FACS.

2CT GeCKOv.2 screens and genomic DNA extractions

Cell–cell interaction genome-wide screens were performed using Mel624 cells transduced independently with both A and B GeCKOv.2 libraries²¹. For one screen, we split two groups of 5 × 10⁷ transduced Mel624 cells. One group was co-cultured with 1.67 × 10⁷ patient-derived NY-ESO-1 T cells (E:T ratio of 1:3 or 0.3) for each library. A second (control) group were cultured under the same density and conditions but without T cells. The co-culture phase was maintained for 12 h after which the T cells were removed as described above. The recovery phase was maintained for another 48 h. In this initial screen, NY-ESO-1 T cells lysed ~76% of tumour cells, and the surviving cells were frozen to evaluate sgRNA enrichment later.

For a second screen, we similarly transduced Mel624 cells with the GeCKOv.2 library and split them into two groups of 5 × 10⁷ transduced cells. For one group, we increased the E:T ratio to 0.5 by co-culture of 2.5 × 10⁷ NY-ESO-1 T cells with 5 × 10⁷ transduced Mel624 cells for each library while keeping all other conditions similar. As before, the second group of Mel624 cells were used as controls, which were cultured under the same density and conditions, but without T cells. By increasing the selection pressure, we observed that T cells killed ~90% of library-transduced Mel624 cells.

To evaluate sgRNA enrichment, cells with and without T cell co-culture (along with an early time point collected after puromycin selection) were collected and frozen. For gDNA extraction from collected tumour cells, we used our previously optimized ammonium acetate and alcohol precipitation procedure to isolate gDNA¹⁴ but substituted AL buffer (Qiagen) for the initial cell lysis step.

2CT-CRISPR pooled screen readout and data analysis

To determine sgRNA abundance as the readout of library screens, two-step PCR amplification were performed on gDNA using Takara Ex-Taq polymerase (Clontech). The first PCR step (PCR1) included amplification of the region containing sgRNA cassette using v2Adaptor_F and v2Adaptor_R primers, and the second step PCR (PCR2) included amplification using uniquely barcoded P7 and P5 adaptor-containing primers to allow multiplexing of samples in a single HiSeq run¹². All PCR1 and PCR2 primer sequences, including full barcodes, are listed on the Addgene website (https://www.addgene.org/pooled-library/zhang-human-gecko-v2/). Assuming 6.6 pg of gDNA per cell, 150 μg of gDNA was used per sample (~350-fold sgRNA representation), in 15 PCR1 reactions. For each PCR1 reaction, 10 μg gDNA were used in a 100 μl reaction performed under cycling conditions: 95 °C for 5 min, 18 cycles of (95 °C for 30 s, 62 °C for 30 s, 72 °C for 30 s), and 72 °C for 3 min. PCR1 products for each sample were pooled and used for amplification with barcoded second step PCR primers. For each sample, we performed 7 PCR2 reactions using 5 μl of the pooled PCR1 product per PCR2 reaction. PCR2 products were pooled and then normalized within each biological sample before combining uniquely barcoded separate biological samples. The pooled product was then gel-purified from a 2% E-gel EX (Life Technologies) using the QiaQuick gel extraction kit (Qiagen). The purified, pooled library was then quantified with Tapestation 4200 (Agilent Technologies). Diluted libraries with 5–20% PhiX were sequenced on a HiSeq 2000.

Demultiplexed reads were trimmed by cutadapt using 12 bp flanking sequences around the 20 bp guide sequence³⁹. Trimmed reads were aligned using Bowtie⁴⁰ to the GeCKOv.2 indexes created from library CSV files downloaded from the Addgene website (https://www.addgene.org/pooled-library/zhang-human-gecko-v2/). Read alignment was performed with parameters ‘-m 1 -v 1 –norc’, which allows up to one mismatch and discards any reads that do not align in the forward orientation or that have multiple possible alignments. Aligned counts of library sgRNAs were imported into R/RStudio. Counts were normalized by the total reads for each sample and then log-transformed. A gene ranking was computed using the second-most-enriched sgRNA for each gene (second-most-enriched sgRNA score) and the RIGER weighted-sum method²².

Gene pathway enrichment analysis

For pathway analysis, we constructed a set of enriched candidate genes by taking advantage of the 1,000 control (non-targeting) sgRNAs embedded in our pooled library with at least one sgRNA enriched above the most-enriched non-targeting control (Fig. 3a, FDR <0.1%, second-most-enriched sgRNA score >0.5). This yielded a list of 554 genes. In this list, we looked for gene category over-representation using Ingenuity Pathway Analysis (QIAGEN Redwood City, http://www.qiagen.com/ingenuity). The analysis criteria were set as follows: (1) querying for molecules with Ingenuity Knowledge Base as a reference set; (2) restricted to human species; and (3) experimentally observed findings as a confidence level. Fisher’s exact test (P < 0.05) was used to compute significance for over-representation of genes in a particular pathway or biological process.

The Cancer Genome Atlas (TCGA) correlation and mutation analysis

TCGA RNA-seq data in form of normalized RNA-seq by Expectation-Maximization (RSEM) values from multiple cancer datasets was downloaded from the Firehose Broad GDAC (http://gdac.broadinstitute.org/, DOI for data release: 10.7908/C11G0KM9) using the TCGA2STAT package for R⁴¹, and used to find overlap between TCGA gene expression indicative of cytolytic activity and genes from our pooled screen where loss-of-function confers resistance to T cell killing. We first identified the genes correlated with a previously identified cytolytic activity signature (CYT), namely the geometric mean of granzyme A (GZMA) and perforin 1 (PRF1) expression¹⁵. To identify these genes in the TCGA data, we calculated the geometric mean of GZMA and PRF1 in each dataset and searched for any genes with a positive correlation to this quantity across patients (Pearson’s r > 0, P < 0.05). We then examined the intersection between genes whose expression was correlated with cytolytic activity (TCGA datasets) and the enriched genes found in the CRISPR screen (554 genes). Individual heatmaps for the two sets of clustered genes were regenerated for each cancer type and can be found at https://bioinformatics.cancer.gov/publications/restifo.

For the top 20 ranked gene candidates from any of the screens, we obtained patient mutation data from the TCGA database using cBioPortal^42,43. For mutation frequency counts, tumours containing likely loss-of-function genetic aberrations (defined as homozygous deletion, missense, nonsense, frame-shift, truncated or splice-site mutations) were included in the analysis.

Arrayed validation of 2CT-CRISPR screen genes

Individual lentiCRISPRs were produced as above except that viral supernatants were not concentrated. For each gene, we used 3–4 sgRNA guide sequences as listed in Supplementary Table 7, where 2 sgRNAs sequences were designed de novo and the other 2 sgRNAs were from GeCKOv.2 library. We cloned these sgRNAs into the lentiCRISPRv2 vector (Addgene) as previously described²¹. To produce virus in a high-throughput format, HEK293FT cells were seeded and transfected in six-well plates where each well received a different lentiCRISPR plasmid. The lentiviral production protocol was the same as the one described above for GeCKOv.2 library lentivirus production (scaled to 6-well format). Mel624 and A375 cells with unique gene perturbations were generated by transduction with these viral supernatants. Typically, we used 500 μl of lentiCRISPRv2 virus per 5 × 10⁴ cells for Mel624 and A375. Puromycin selection (1 μg ml⁻¹) was applied to these cells for 5–7 days. Over this period, untransduced cells were killed completely by puromycin.

After taking a subset of cells for analysis of insertion–deletion (indel) mutations, the remainder were normalized to seed 1 × 10⁴ cells per well in 96-well plates. During the arrayed screen, each cell line was co-cultured with appropriate T cells (either NY-ESO-1 or MART-1) in a 96-well plate format at an E:T ratio of 1:3 for 12 h in T cell media. As in the pooled screen, we performed gentle 2× PBS washes to remove the T cells. Mel624 or A375 cells were collected after a recovery phase culture of 48 h for high-throughput FACS analysis. Tumour cell counts were measured using a FACS-based CountBright bead method (Life Technologies). We noticed variability in proliferation and survival rates across cells depending on the sgRNA received. To account for this variability, we calculated a relative per cent change for each sgRNA: (%Δ, 2CT versus noT) in tumour cells co-cultured with T cells (2CT) compared to tumour cell counts without T cells (noT). The normalized cell survival was calculated by the following formula:For co-culture with MART-1 T cells, all parameters were the same, except that the co-culture period duration was 24 h with the same recovery period as with NY-ESO-1 experiments.

Indel mutation fetection for array 2CT validation

Genomic DNA from thawed cell pellets was purified using the Blood & Cell Culture kit (Qiagen). sgRNA target site PCR amplifications were performed for each genomic loci using conditions as described previously¹⁴, pooled together and then sequenced in a single Illumina MiSeq run. All primer sequences for indel detection can be found in Supplementary Table 8.

To analyse the data from the MiSeq run, paired-end reads were trimmed for quality using trimmomatic with parameters SLIDINGWINDOW:5:25⁴⁴. Reads with surviving mate pairs were then aligned to their targeted amplicon sequence using Bowtie2⁴⁵. To determine indel sizes, we calculated the size difference between observed reads and predicted read size based on the genomic reference sequence. If observed read size was equal to the predicted size, these reads were scored as no indels. The size difference was used to detect insertions or deletions.

Flow cytometry and immunoassays

Tumour cells or T cells suspended in FACS staining buffer were stained with fluorochrome-conjugated antibodies against combinations of the following surface human antigens: CD4 (RPA-T4, BD), CD8 (SK1, BD), CD3e (SK7, BD), HLA-A*02/MART-1:27-35 tetramer (DMF5, Beckman Coulter Immunotech); NY-ESO-1 tetramer (NIH Tetramer Core Facility); PD1-L1 (MIH1, eBiosciences); PD1-L2 (24F.10C12, Biolegend); IFNγ (25723.11, BD), CD58 (1C3, BD) and β2M (2M2, Biolegend). Cell viability was determined using propidium iodide exclusion or a fixable Live/Dead kit (Invitrogen). Intracellular staining assay (ICS) on ESO T cells after 5–6 h of co-culture with A375 cells, where T cells were pre-treated with monensin (512092KZ, BD) and brefeldin A (512301KZ, BD), was performed using manufacturer’s instructions for BD Flow cytometry ICS. Flow cytometric data were acquired using either a FACSCanto II or LSRII Fortessa cytometer (BD), and data were analysed with FlowJo version 7.5 software (FlowJo LLC).

The amount of IFNγ release by T cells after co-culture with tumour cells was measured by sandwich ELISA assay using anti-IFNγ (Thermo Scientific M700A) coated 96-well plates, biotin-labelled anti-IFNγ (M701B), horseradish peroxidase (HRP)-conjugated streptavidin (N100) and TMB substrate solution (N301).

Quantitative PCR analysis

Total RNA was extracted from cells using RNeasy Plus Mini Kit (Qiagen). Gene expression was quantified using TaqMan RNA-to-Ct 1-Step Kit (Thermo Fisher) and Taqman assay probes (Invitrogen): Aplnr(Mm00442191_s1), Actb (Mm00607939_s1), APLN (Hs00936329), ACTB(Hs01060665_g1), IRF1 (Hs00971965_m1), IFI30 (Hs00173838_m1), TAP1(Hs00388675_m1), TAP2 (Hs00241060_m1), TAPBP (Hs00917451_g1) and JAK1 (Hs01026983_m1). Relative mRNA expression was determined via the ΔΔC_t method.

Immunoprecipitation and western blot analysis

For immunoprecipitation, cells were lysed on ice in IP lysis buffer (Thermo Fisher #87787) supplemented with protease and phosphatase inhibitors (Thermo Fisher #88668) for 15 min. Lysates were cleared by centrifugation at 16,000g for 10 min at 4 °C and supernatants were incubated with indicated antibodies for 24 h at 4 °C. Following incubation, cells were added to pre-washed Dynabeads coupled to protein G or protein A (Life Technologies) and incubated for 3 h at 4 °C. Immune complexes were purified on magnets with extensive washing with lysis buffer. Purified complexes were mixed with sample buffer and assayed by western blot. Antibodies used included: anti-Jak1 (BD 610231), anti-Jak2 (Cell Signaling 3230), anti-Flag (Cell Signaling #8146). In some cases, immune complexes were isolated with sepharose-conjugated antibodies specific for IgG (Cell Signaling 3420) or Flag (Cell Signaling 5750).

For immunoblots, total protein was extracted with 1× RIPA lysis buffer (Millipore) with 1× protease inhibitor (Roche). Protein concentration was determined using the BCA assay (Thermo/Pierce). Cell lysates were resolved on 4–20% Tris-Glycine gels (Invitrogen), transferred to PVDF membranes (Millipore), and incubated overnight at 4 °C with the appropriate primary antibodies: anti-Aplnr (Abcam ab214369), anti-β2M (Abcam ab75853) and anti-β-actin (Abcam ab8227) for mouse tumour cells; anti-APLNR (Abcam ab97452), anti-phospho-JAK1 (Cell Signaling 3331) and anti-phospho-STAT1 (Cell Signaling 7649). Signals were detected using HRP-conjugated secondary antibodies (Santa Cruz Biotechnology) and SuperSignal West Femto Chemiluminescent Substrate (Thermo/Pierce). Images were captured using ChemiDoc Touch imaging system (Bio-Rad).

Whole-exome sequencing and transcriptomics analysis

Whole-exome sequencing of patient tumours was performed as previously described⁴⁶. For transcriptome analysis on APLNR-knockout cells, total RNA was extracted using the RNeasy Plus Mini Kit (Qiagen). The quality and quantity of extracted total RNA was assessed using TapeStation 2200 (Agilent). All RNA-sequencing (RNA-seq) analyses were performed using three biological replicates. 200 ng of total RNA was used to prepare RNA-seq library using the TruSeq RNA sample prep kit (Illumina). Paired-end RNA sequencing was performed on a HiSeq 2000 (Illumina). Sequenced reads were aligned to the human genome (hg19) with TopHat (v.2.0.11)⁴⁷ and the mapped data were then processed by Cufflinks. Gene expression values were normalized to obtain RPKM (reads per kilobase exon model per million mapped reads) values using CuffDiff⁴⁸. To define differentially expressed genes, we used a twofold change and P < 0.05 difference between groups.

Characterization of APLNR patient mutations using 2CT assay

APLNR was perturbed in A375 cells using LentiCRISPRv2 encoding APLNR_sg1 (Supplementary Table 7). Cells were selected with puromycin for 5–7 days. For rescue experiments, cells were transduced with a lentiviral (blasticidin-selectable) plasmid encoding codon-shuffled APLNR sequences (either wild-type or with specific patient mutations). APLNR sequences were re-coded to ensure that APLNR_sg1 cannot introduce Cas9-mediated breaks in the APLNR transgene. Transduced A375 cells were selected with 10 μg ml⁻¹ blasticidin for 5 days. Expression of these constructs in cells was verified by western blot analysis. Cells were subject to 2CT-assay with ESO T cells as described above.

Adoptive cell transfer and tumour immunotherapy

For immunotherapy experiments, we used mouse–human chimaeric gp100 antigen overexpressing B16 melanoma (mhgp100-B16) cells that are responsive to adoptive transfer of Pmel-1 TCR transgenic CD8⁺ T cells in an in vivo setting. We generated gene-deleted mhgp100-B16 cells using lentiviruses encoding sgRNAs targeting Aplnr (5′-GATCTTGGTGCCATTTTCCG-3′) or B2m (5′-GAAATCCAAATGCTGAAGAA-3′) loci as described above. C57BL/6 mice were subcutaneously implanted with 5 × 10⁵ mhgp100-B16 cells. After 10 days of tumour implantation, mice (n = 10 for treatment groups) were sub-lethally irradiated (600 cGy), and injected intravenously with 1 × 10⁶Pmel-1 CD8⁺ T cells and received intraperitoneal injections of IL-2 in PBS (6 × 10⁴ IU per 0.5 ml) once daily for 3 consecutive days. All tumour measurements were blinded. Tumour size was measured in a blinded fashion approximately every two days after transfer and tumour area was calculated as length × width of the tumour. Mice with tumours greater than 400 mm² were euthanized. The products of the perpendicular tumour diameters are presented as mean ± s.e.m. at the indicated times after ACT. In another experimental setting, we also used RNA-interference-mediated knock-down of Aplnr in mhgp100-B16 cells to perform adoptive transfer as above. Lentiviral particles encoding shRNAs (Dharmacon, shRNA1- V2LMM_36640; shRNA2- V3LMM_517943) were used to knock-down expression of Aplnr in mhgp100-B16 cells. qPCR analysis was performed on these cells to confirm knock-down efficiencies of these constructs.

B2905 is a melanoma cell line derived from spontaneous tumour induced by UV irradiation of C57BL/6-HGF transgenic mice⁴⁹. We generated gene-deleted B2905 cells using lentiCRISPR encoding AplnrsgRNA as above. We implanted 1 × 10⁶ tumour cells subcutaneously in C57BL/6 mice. Intraperitoneal injections of 250 μg of anti-CTLA4 monoclonal antibody (Bio X cell, BE0131) or IgG control antibody were given on days 10, 13, 16 and 19 post implantations. Tumour size was measured in a blinded fashion approximately every 10 days and tumour area was calculated as length × width of the tumour. Owing to large intra-individual differences in tumour growth rates (Supplementary Fig. 7a), the significance of resistance to anti-CTLA4 treatment was determined by Fisher’s exact test comparison of control sgRNA versus sg-Aplnr groups using the number of progressing tumours and completely regressed tumours in each group.

Statistical analysis

Data between two groups were compared using a two-tailed unpaired Student’s t-test or the Mann–Whitney test as appropriate for the type of data (depending on normality of the distribution). Unless otherwise indicated, a P value less than or equal to 0.05 was considered statistically significant for all analyses, and not corrected for multiple comparisons. To compare multiple groups, we used an analysis of variance (ANOVA) with the Bonferroni correction. All group results are represented as mean ± s.e.m., if not stated otherwise. Prism (GraphPad Software Inc.) was used for these analyses.