Pan-cancer Landscape of Programmed Death Ligand-1 and Programmed Death Ligand-2 Structural Variations

PURPOSE

Programmed cell death protein-1 (PD-1) receptor and ligand interactions are the target of immunotherapies for more than 20 cancer types. Biomarkers that predict response to immunotherapy are microsatellite instability, tumor mutational burden, and programmed death ligand-1 (PD-L1) immunohistochemistry. Structural variations (SVs) in PD-L1 (CD274) and PD-L2 (PDCD1LG2) have been observed in cancer, but the comprehensive landscape is unknown. Here, we describe the genomic landscape of PD-L1 and PD-L2 SVs, their potential impact on the tumor microenvironment, and evidence that patients with these alterations can benefit from immunotherapy.

METHODS

We analyzed sequencing data from cancer cases with PD-L1 and PD-L2 SVs across 22 publications and four data sets, including Foundation Medicine Inc, The Cancer Genome Atlas, International Cancer Genome Consortium, and the Oncology Research Information Exchange Network. We leveraged RNA sequencing to evaluate immune signatures. We curated literature reporting clinical outcomes of patients harboring PD-L1 or PD-L2 SVs.

RESULTS

Using data sets encompassing 300,000 tumors, we curated 486 cases with SVs in PD-L1 and PD-L2 and observed consistent breakpoint patterns, or hotspots. Leveraging The Cancer Genome Atlas, we observed significant upregulation in PD-L1 expression and signatures for interferon signaling, macrophages, T cells, and immune cell proliferation in samples harboring PD-L1 or PD-L2 SVs. Retrospective review of 12 studies that identified patients with SVs in PD-L1 or PD-L2 revealed > 50% (52/71) response rate to PD-1 immunotherapy with durable responses.

CONCLUSION

Our findings show that the 3′-UTR is frequently affected, and that SVs are associated with increased expression of ligands and immune signatures. Retrospective evidence from curated studies suggests this genomic alteration could help identify candidates for PD-1/PD-L1 immunotherapy. We expect these findings will better define PD-L1 and PD-L2 SVs in cancer and lend support for prospective clinical trials to target these alterations.

INTRODUCTION

Over the past decade, immune checkpoint blockade has emerged as an effective therapy across multiple cancer types, in some instances leading to durable, complete remission. For example, the immune checkpoint programmed cell death protein-1 (PD-1) attenuates T-cell activity by binding to their respective ligands, an essential interaction in maintaining balance in the immune system.^1,2 Ligands for the PD-1 receptor include programmed death ligand-1 (PD-L1) and programmed death ligand-2 (PD-L2). Cancer cells and their immune microenvironments can use these mechanisms to escape immunosurveillance through high expression of PD-L1.^3-5 Antibodies to block either PD-1 receptor or PD-L1 have emerged as effective therapies for more than 20 cancer types but are only effective in a subset of patients. PD-L1 expression via immunohistochemistry (IHC) is the most studied predictive biomarker for immunotherapy response, followed by tumor mutational burden and microsatellite instability (MSI). However, studies have demonstrated PD-L1 IHC positivity is only 20%-30% predictive for response to PD-1/PD-L1 immunotherapy in several cancers.^6-8 Therefore, there is a need to further understand the regulatory processes affecting PD-1 ligand expression in cancer.

CONTEXT

• Key Objective

• Are programmed death ligand-1 (PD-L1) and programmed death ligand-2 (PD-L2) structural variations (SVs) in tumors a candidate predictive biomarker for immunotherapy?

• Knowledge Generated

• Using data sets encompassing 300,000 tumors, we curated 486 cases with SVs in PD-L1 and PD-L2 and observed consistent breakpoint patterns, or hotspots. We observed upregulation of PD-L1, an inflamed phenotype in PD-L1 and PD-L2 SV-positive cases, and a > 50% (52/72) retrospective response rate to programmed cell death protein-1 immunotherapy with durable responses.

• Relevance

• Our findings support evidence that PD-L1 or PD-L2 SVs are candidate-predictive biomarkers and that patients with these alterations may be more likely to benefit from programmed cell death protein-1 immunotherapy than patients who do not have these alterations.

Genomic alterations in cancer, such as copy-number amplification (CNA), can affect PD-1 ligand expression, as observed in lymphoma and in some solid tumors.^9-11 In 2016, two studies in leukemia and lymphoma revealed structural variations (SVs), such as deletions, inversions, and translocations in PD-L1 (CD274) and PD-L2 (PDCD1LG2) genes.^12,13 Although initial observations were in lymphomas and leukemia, there has been little reported on solid tumors harboring this genomic aberration. Kelly et al¹⁴ reported on PD-L1 SVs in 145 solid tumors, which is the largest assessment of PD-L1 SV in solid tumors. Furthermore, retrospective case-based evidence suggests that patients with tumors harboring SVs may be exceptional responders to PD-1 immunotherapy.

However, there remain several unanswered questions about the pan-cancer prevalence of SVs in both PD-L1 and PD-L2. Although studies in leukemia previously demonstrated genomic breakpoints involving the 3′-UTR in PD-L1, the whereabouts of other relevant breakpoints have not been explored. Additionally, alterations in PD-L2 have not been studied across cancers. The extent to which SVs affect mRNA or protein expression warrants further study across cancers. Conventionally, patients being considered for immunotherapy do not undergo evaluation for PD-L2 protein expression. Furthermore, the tumor immune microenvironment of these tumors has not been characterized. Finally, the clinical implications of genomic alterations in PD-L1 and PD-L2 deserve further attention, as there is growing retrospective evidence that these tumors are exceptional responders.

Here, we sought to characterize both PD-L1 and PD-L2 SVs in hematologic and solid tumors by cataloging the locations of genomic breakpoints and determining their impact on the immune microenvironment. We aggregated data across 22 publications and four databases (Foundation Medicine Inc [FMI], The Cancer Genome Atlas [TCGA], International Cancer Genome Consortium [ICGC], and Oncology Research Information Exchange Network [ORIEN]) with approximately 300,000 total tumors assessed by fluorescence in situ hybridization, targeted exome, whole-exome sequencing (WES), transcriptome, or whole-genome sequencing (WGS). We have also assembled retrospective case reports and series for patients who have received PD-1 immunotherapy to highlight a precision oncology approach for a tumor-agnostic biomarker on the basis of these SVs.

METHODS

Cancer Genomic Data

We manually curated 22 studies that reported cases with SVs in PD-L1 or PD-L2 on the basis of whole genome, whole exome, transcriptome, or targeted exome (Data Supplement). To be included, we required that the study provide sequencing breakpoint information (Data Supplement). We also evaluated genomic data from four large data collaborations with FMI, ORIEN, TCGA, and ICGC.¹⁵ As part of a collaboration with ORIEN, we have obtained gene fusion output generated by M2GEN, which uses STAR-Fusion¹⁶ and Arriba¹⁷ to identify rearrangements in RNA sequencing data. We obtained gene fusion output from TCGA data reported by Gao et al,¹⁸ Hu et al,¹⁹ and Zhang et al.²⁰ Additionally, we identified gene fusions using software CICERO (version 1.4.3)²¹ on RNA sequencing data from a subset of the TCGA. We obtained SV output data on ICGC through their publicly available portal¹⁵ and from Zhang et al.²⁰ This included multi-institutional cohorts and ICGC's Pan-Cancer Analysis of Whole Genomes study. Additionally, we manually curated 12 studies that involved patients harboring PD-L1 or PD-L2 SVs who received PD-1 immune checkpoint inhibitor therapy.

RNAseq Fusion Analysis

We evaluated 733 tumors from ORIEN and 2,745 TCGA representing cases deriving from lymphomas, lung, thyroid, brain, head and neck, and bladder cancers. For further analysis of gene fusions in RNAseq data, we used CICERO (version 1.4.3) on the cohorts described above and obtained gene fusion output generated by M2GEN using STAR-Fusion¹⁶ and Arriba¹⁷ from ORIEN samples. To filter CICERO output, we required that fusion calls have > 420 total reads supporting the SV breakpoint. For ORIEN fusion calls, we required > 5 supporting reads. These thresholds were determined by calculating the median number of reads in each respective output list of PD-L1 and PD-L2 fusion calls.

Evaluation of Genomic Breakpoints

We obtained breakpoint information, as reported in respective publications or tumor atlas studies, where each case originated. Breakpoint types included translocations, deletions, inversions, duplications, and unclassified structural variations. SVs defined as unclassified consisted of events harboring breakpoints in PD-L1 or PD-L2, but the alteration type was not provided or could not be determined. Breakpoints from cases that were reported using hg38 alignment were converted to hg19 to evaluate and plot the breakpoints in a common genomic space. We plotted PD-L1 and PD-L2 SV events from 22 studies and four data sets on genomic space using Python library matplotlib. The histogram was created using a bin size of 250 bases.

Tumor Microenvironment Analysis

Immune signature abundance was evaluated using R packages ImSig (R package version 1.1.3) and CIBERSORT (jar version 1.05), which both take, as input, a gene expression profile matrix.^22,23 We downloaded gene expression data from 9,449 RNAseq samples from the TCGA using the R package TCGABiolinks.²⁴ PD-L1 copy-number status was obtained from an annotated version of TCGA data reported in cBioPortal.²⁵ We selected 5,256 tumors that were copy-number neutral for both PD-L1 and PD-L2. Significance of differential gene expression and immune signature abundances were evaluated using Mann-Whitney tests, applying an adjusted P value of .005 (.05/10 signatures tested) using a Bonferroni correction for ImSig and .002 (.05/25 signatures tested) for CIBERSORT analyses. Violin plots were generated using the Python package Seaborn.²⁶

Clinical End Points From Retrospective Literature Review

We compiled literature describing case series or single case reports from patients with any cancer type found to have a PD-L1 or PD-L2 SV and received some form of PD-1 immunotherapy (Data Supplement). For evaluation of retrospective outcomes, we also included patients who had PD-L1 or PD-L2 CNAs, which we define as copy number greater than or equal to overall specimen ploidy + 4. Patients reported as having regression of tumor size were considered responders to therapy. Patients who continued therapy more than 12 months were considered durable responders.

RESULTS

Structural Variations in PD-L1 and PD-L2 Are Genomically Diverse

We analyzed sequencing data from cancer cases with PD-L1 (CD274) and PD-L2 (PDCD1LG2) SVs across 22 studies and four datasets, including large pan-cancer sources: FMI, TCGA, ORIEN, and ICGC (Fig 1A). These data represent a variety of sequencing formats, including transcriptome, targeted exome, WES, and WGS (Fig 1B). SVs included translocations, deletions, inversions, duplications, and some that were unclassified (Fig 1C). From these sources, we have curated 405 cases harboring SVs in PD-L1 or PD-L2 (Data Supplement). To determine the diversity and breakpoint consistencies of SV in PD-L1 and PD-L2, we plotted these cases across genomic space (Fig 2A). We observed diverse SV types in the region encompassing PD-L1 and PD-L2, including duplications (n = 108), deletions (n = 109), inversions (n = 59), translocations (n = 100), and unclassified variations (n = 29; Figs 2A-2C). Interestingly, we identified breakpoint hotspots in the introns 4, 5, and 6, accounting for 10%, 17%, and 36% of all cases, respectively (Fig 2A). Of the curated translocations, gene fusion partners with PD-L1 and PD-L2 were observed across the genome; these cases did not appear to have high recurrence for specific partner genes (Fig 2B). An additional 81 cases harbored multiple SV calls. Because we were unable to determine the dominant SV in these samples (Appendix Fig A1), we exempted these cases from our main analysis.

FIG 1.

Genomic data sources used for curation of structural variations in PD‐L1 and PD‐L2. (A) Sequencing data from Foundation Medicine Inc, the Oncology Research Information Exchange Network, the International Cancer Genome Consortium, The Cancer Genome Atlas, and curated publications were evaluated for structural variations involving PD‐L1 and PD‐L2 genes. There were 486 tumors identified from sequencing data, including targeted exome panels, RNAseq, whole genome, whole exome, SNP arrays, and BAC sequencing. (B) Pie chart demonstrates type of data used to identify 486 cases. (C) Pie chart demonstrates type of structural variations observed in PD‐L1 and PD‐L2, representing the 405 cases used in the main analysis. BAC, bacterial artificial chromosome; FMI, Foundation Medicine Inc; ICGC, International Cancer Genome Consortium; ORIEN, Oncology Research Information Exchange Nettwork; SNP, single nucleotide polymorphism; TCGA, The Cancer Genome Atlas; WES, whole-exome sequencing; WGS, whole-genome sequencing.

FIG 2.

Diversity of PD‐L1 and PD‐L2 structural variations. (A) Landscape of structural variations (deletions, inversions, duplications, and unclassified structural variations) involving PD‐L1 and PD‐L2, color coded by alteration type, representing 305 cases. Blue histogram (top) represents density of breakpoints across genomic space. (B) Circos plot of intra- and inter‐chromosomal translocations involving PD‐L1 and PD‐L2 representing 100 cases. Key: red ‐ observed gene partners with PD‐L1; blue ‐ observed gene partners with PD‐L2; green ‐ observed gene partners with PD‐L1 and PD‐L2.

PD-L1 and PD-L2 Structural Variations Occur in Diverse Cancer Types

We interrogated this multistudy cohort and determined the cancer types harboring these structural variant events. Of these cases, 274 were from solid tumors and 212 were from lymphomas (Figs 3A and 3B). For solid tumors, squamous cell carcinoma histologies were more frequently represented than expected (Fig 3C). For hematologic malignancies, T-cell and natural killer (NK) cell subtypes were also more frequently represented than expected (Fig 3D). The most frequently affected cancer types included diffuse large B-cell lymphoma (n = 48), NK/T-cell lymphoma (n = 35), lung cancer (n = 54), breast cancer (n = 22), and head and neck cancer (n = 20; Fig 3).

FIG 3.

Cancer distribution of PD-L1 and PD-L2 structural variations. (A) Distribution of solid tumor types in patients harboring PD‐L1 and PD‐L2 structural variations, color coded by the altered gene(s). (B) Pie chart of histological types in solid tumor samples. (C) Distribution of hematologic malignancies in patients harboring PD‐L1 and PD‐L2 structural variations. (D) Pie chart of histological types in hematologic malignancies. NK, natural killer.

PD-L1 and PD-L2 Structural Variation Is Associated With Increased Ligand Expression and an Altered Immune Microenvironment

Previous studies showed that PD-L1 SVs were associated with upregulated PD-L1 gene expression.^12,14 To further evaluate expression, we compared PD-L1 gene expression in 38 tumors with either PD-L1 or PD-L2 SV to approximately 5,000 tumor samples from TCGA that were diploid PD-L1 and PD-L2 copy numbers.^12,20 We observed a significant upregulation in PD-L1 gene expression in these cases harboring SVs (Fig 4A), as well as upregulation in PD-L2 gene expression (Appendix Fig A2). Similarly, we used RNA-based software ImSig and CIBERSORT to evaluate gene expression signatures for immune cell subsets or pathways. From ImSig, we observed a significantly higher abundance of interferon and proliferation signaling pathways in PD-L1 structurally altered tumors compared with PD-L1 copy-neutral tumors. Applying CIBERSORT, which measures cell type proportions in each sample, we observed that PD-L1 altered tumors displayed a significant enrichment in CD8+ T cells, regulatory T cells (Tregs), resting NK cells, activated CD4 T cells, and M1 macrophages compared with diploid tumors (Fig 4C). Furthermore, several cell type signatures, including M2 macrophages, resting mast cells, and monocytes, were significantly depleted (Fig 4C).

FIG 4.

Biological impact of PD-L1 and PD-L2 structural variations in The Cancer Genome Atlas. Our analysis of 9890 tumors demonstrated upregulated expression of PD-L1 and immune microenvironment enrichment in tumors harboring PD-L1 structural variation. (A) Violin plot of PD-L1 gene expression in PD-L1 and PD-L2 structural variation positive (n = 37) and PD-L1 and PD-L2 diploid, or wildtype, tumors (n = 5085). (B) Violin plot of ImSig immune signature scores in PD-L1 and PD-L2 structural variation positive and PD-L1 and PD-L2 diploid, or wildtype, tumors. Violin plot of immune cell subtype proportions using software CIBERSORT in PD-L1 and PD-L2 structural variation positive and PD-L1 and PD-L2 diploid, or wildtype, tumors. *** denotes P < 0.01 and **** denotes P < 0.001. For all tests, power > 0.8. NK, natural killer; RSEM, RNA‐seq by Expectation‐Maximization.

Retrospective Clinical Outcomes for Patients With PD-L1 or PD-L2 SV Receiving PD-1 Immunotherapy

Since tumors with SVs may have increased expression of ligand and may lead to immune suppression, we sought to determine if these patients might have improved outcomes with PD-1 immunotherapy. We assessed the literature for patients with any cancer type harboring deletions, partial duplications, translocations, or inversions (n = 9), as well as CNAs (n = 22), involving either PD-L1 only or both PD-L1 and PD-L2 (Data Supplement). We manually curated 12 studies involving 71 patients with genomic alterations who received PD-1 therapies. Multiple tumor types were represented in these studies; 49 of these patients had lymphoma, while the remaining 22 patients had solid tumors. All studies achieved a response rate > 50%, with an overall > 73% (52/71) response rate to immunotherapy combining all studies. For assessing retrospective clinical outcomes, we included CNAs, an additional class of SV, in our analysis as a class of SVs, but our focus is the less common and less studied SVs, such as deletions, partial duplications, translocations, inversions. To assess the SVs as a clinical biomarker, we evaluated the depth and durability of responses. From the 41 cases with available response data, 14 achieved a complete response, 21 had a partial response, and six had stable disease (Table 1).^8,10,27‐33 From 32 cases with duration of therapy data, we found an average of 20.58 months and a maximum duration of 49.6 months.

TABLE 1.

PD‐1 Treatment Duration Responses of Patients Harboring PD-L1 and PD-L2 Structural Variations

DISCUSSION

Through this comprehensive pan-cancer study involving a combined total of approximately 300,000 tumors, we have identified and highlighted recurrent genomic alterations in both PD-L1 and PD-L2 that drive ligand expression and are potential actionable targets for precision immunotherapy. Although previous studies have largely focused on lymphoma, we have curated data that demonstrate that these genomic alterations occur in both blood cancers and various solid tumors. Prior studies have also focused on PD-L1, whereas we bring attention to 106 cases involving PD-L2. For solid tumors, there appears to be higher-than-expected incidence of SVs in cancers with squamous histology. Furthermore, although SVs appear to affect similar genomic positions, they are quite diverse in structure and size, including translocations, deletions, inversions, and duplications. The nature of these diverse structures and breakpoint locations provides an opportunity for further investigation via sequencing of the noncoding portion of the genome. Current methods provide a good understanding of exonic breakpoints and intronic breakpoints near exonic regions. Intronic breakpoints warrant further investigation through novel intronic sequencing methods. We anticipate that this strategy will help improve our understanding of the true prevalence of PD-L1 and PD-L2 SVs in cancer. Finally, we have collected retrospective evidence demonstrating durable exceptional responses in patients harboring PD-L1 and PD-L2 SVs.

The earliest studies describing SVs in PD-1 ligands involve lymphoma. Fluorescence in situ hybridization (FISH) identified copy-number alterations at the 9p24 region in classical Hodgkin lymphoma, including coamplifications of PD-L1, PD-L2, and JAK2 genes.⁹ Clinical trials would show promising response rates and durable responses and lead to the US Food and Drug Administration–approval of PD-1 immunotherapy in Hodgkin lymphoma.^27,34 Overcoming limitations of FISH, Steidl et al³⁵ used next-generation sequencing to identify gene fusions in classical Hodgkin lymphoma. Subsequent studies identified SVs in subsets of non-Hodgkin lymphoma, including primary mediastinal B-cell lymphoma (2014),³⁶ diffuse large B-cell lymphoma (2016),³⁷ primary testicular lymphoma (2016),³⁸ and adult T-cell leukemia (2016).¹² Clinical trials of single-agent PD-1 immunotherapy across various B-cell lymphomas have included all patients, without molecular selection, and have not been as successful as classical Hodgkin lymphoma, with overall response rates < 20%. By contrast, primary mediastinal B-cell lymphoma has the highest prevalence of PD-L1 and PD-L2 alterations with impressive overall response rates of 40% with durable responses. Pembrolizumab was US Food and Drug Administration–approved in 2018 for primary mediastinal B-cell lymphoma without molecular selection. Current clinical trials of PD-1 immunotherapy in lymphoma are largely combination therapy trials with chemotherapy or other immune modulators. However, through our retrospective evaluation of case studies and series (Data Supplement), we have observed that there are exceptional responders in lymphoma who have genomic alterations on the basis of FISH or sequencing. More impressive is that patients have long-lasting durable responses (Table 1). We propose that the prospective selection of patients with non-Hodgkin lymphoma on the basis of SVs may identify a molecular subset of patients who can benefit from PD-1 immunotherapy. It is currently not standard of care or routine for patients with lymphoma to undergo genomic evaluation of PD-1 ligands.

Through various sequencing and analysis strategies, our study revealed SVs in both PD-L1 and PD-L2 across cancers. We observed that these strategies varied in detection rates. For example, in one approach, Kataoka et al¹² analyzed 10,210 tumors from TCGA and showed that RNAseq could detect events that include an expressed breakpoint or impact gene expression. They applied a fusion calling approach and an exon-ratio strategy to identify SVs and truncations and identified 32 cases harboring this type of alteration in PD-L1 (0.31% prevalence). By contrast, in a pan-cancer analysis of WGS data from TCGA, Zhang et al²⁰ evaluated 1,220 cancer cases and using a two-consent algorithm strategy to detect alterations, identified 28 cases (2.3% prevalence) that harbored SV in PD-L1 or PD-L2. Specifically, 14 of those cases harbored an SV in PD-L1 (1.15% prevalence). Furthermore, Chong et al¹³ and Chapuy et al³⁸ used custom probes to comprehensively capture PD-L1 and PD-L2 regions and detected SVs at a higher rate compared with prior WES studies for lymphoma.¹² In our meta-data analysis, we observed overall a higher prevalence of SVs from WGS studies compared with exome or RNAseq studies (Data Supplement). Furthermore, a total of 145 breakpoints were identified in the breakpoint hotspot region located in introns 5 and 6 from this study. This concentration of breakpoints may warrant further study of intronic regions that could regulate expression of immune checkpoints.

Tumors harboring PD-L1 and PD-L2 SVs appear to have increased mRNA expression and protein expression. We presented findings on tumors with paired DNA and RNA sequencing that demonstrate increased expression of ligands compared with wild-type (Fig 4). Similarly, Kelly et al, reported on 43 SV-positive solid tumors with available IHC for PD-L1. Of these cases, 72% (31/43) had scores > 50% compared with wild-type tumors, while only 15% had scores > 50%.¹⁴ For comparison, a study of real-world PD-L1 IHC testing for nearly 5,000 lung cancers shows that 30%-32% of tumors have PD-L1 scores > 50%, 30% are 1%-49% and 40% are < 1%.³⁹ In light of a subset of tumors harboring PD-L1 or PD-L2 SVs, there may be limitations on current approaches for IHC assessment. First, the convention is to evaluate PD-L1 status by IHC with one of several antibodies against either the N-terminus (22C3, 28-8) or C-terminus (SP142) of PD-L1 protein. For tumors with SVs in PD-L1 that truncate the open reading frame and C-terminus, antibodies against the C-terminus (SP142) will miss protein expression with 3′ truncations, whereas N-terminus antibodies can still detect the ligand.^12,29 Therefore, tumors tested with any N-terminus antibodies, including SP142, SP263, and E1L3N, will have poor sensitivity for tumors that may have the highest ligand expression. Second, since IHC assessment focuses on PD-L1, patients who may harbor PD-L2 SVs will be unrecognized as candidates for immunotherapy. Although prior studies demonstrated PD-L2 SVs in hematologic malignancies, here we show that solid tumors also harbor PD-L2 SVs (Fig 3A). Prognostic outcome studies after surgical resection often correlate PD-L1 expression with increased risk of relapse. Similarly, PD-L2 expression correlates with increased risk of relapse further supporting the notion that PD-L2 SVs leading to overexpression are clinically significant.^40,41 Responders to PD-1 immunotherapy with zero or low-level PD-L1 expression could therefore be explained by PD-L1 truncations missed by current IHC or PD-L2 SVs that are not assessed by IHC.

Our analysis of the biological significance of PD-L1 and PD-L2 SVs revealed upregulated ligand expression and an altered tumor microenvironment. From TCGA, we found an enrichment in PD-L1 mRNA expression in tumors harboring PD-L1 or PD-L2 SV compared with wild-type PD-L1 and PD-L2 copy-neutral tumors (Fig 4). This finding is consistent with other studies that showed an association between PD-L1 alteration and PD-L1 expression.^12,20 To our knowledge, we are the first to analyze the association between PD-L1 and PD-L2 SV and the tumor microenvironment. Our analysis of the tumor microenvironment demonstrates enrichment in inflammatory phenotypes. First, using ImSig, we observed enrichment in signaling for both interferon signaling and immune cell proliferation. Upregulated interferon signaling in PD-L1 rearranged tumors is of particular interest, as interferon exposure is known to drive PD-L1 and PD-L2 expression⁴² and secrete proinflammatory cytokines in cancer.⁴³ Analysis of cell subtypes revealed interesting patterns in PD-L1 and PD-L2 altered tumors. In these cases, we observed higher proportions of M1 macrophages, which has been found to induce PD-L1 expression in hepatocellular cells.⁴⁴ Second, M1 macrophages elicit a proinflammatory response, whereas M2 macrophages, of which we observed significant depletion in our test cohort, are considered anti-inflammatory.^44,45 Finally, tumors with PD-L1 or PD-L2 alteration also had significantly higher proportions of CD8+ T cells than their wild-type counterparts. CD8+ cells also contribute to inflammation, as they produce proinflammatory cytokines and chemokines and recruit lymphocytes and leukocytes to the site of inflammation.⁴³ Interestingly, one study found that among patients who harbored MSI, enriched CD8+ cells were associated with a durable response to immune checkpoint inhibitor therapy.⁴⁶ Overall, our findings demonstrate an altered immune microenvironment is associated with PD-L1 and PD-L2 altered tumors. There is evidence that PD-L1 expression alone is not a reliable biomarker to predict patient response to immunotherapy. Berry et al proposed that additional protein markers, including CD8+ T cells, FoxP3, and CD163, could be used for more accurate prediction of patient response to PD-1 inhibitors in melanoma.⁴⁷

For advanced, metastatic, or refractory solid tumors and lymphomas, PD-L1 and PD-L2 SVs may be a predictive biomarker for exceptional responses to PD-1 immunotherapy. Our evaluation demonstrates that 52 of 71 retrospective cases had sustained partial responses or complete responses > 12 months duration (Data Supplement). The magnitude and duration of these responses is akin to tumors with MSI high+, which can be seen across cancer types.⁴⁸ Further evidence that SVs may be highly predictive is the entity of PD-L1 IHC > 50%, or high, in metastatic non–small-cell lung cancer, where monotherapy with PD-1 immunotherapy is so effective, chemotherapy can be omitted.^49,50 SVs correlate strongly with high IHC staining (> 50%).¹⁴ Conversely, as expected, PD-L1 IHC tumor cell expression > 50% is a poor prognostic feature in patients with surgically resected non–small-cell lung cancer, who did not receive immunotherapy.⁵¹ Across many other cancer types, clinical trials have demonstrated response rates of 10%-15%, with some correlation with tumor mutation burden and PD-L1 positivity.⁵² However, these studies did not evaluate for the presence of SVs and generally used the lower threshold for PD-L1 tumor cell expression of > 1%. There are several limitations to the observed clinical outcomes. There is a positive selection bias to case reports, as clinical trials have not evaluated SVs, and current sequencing strategies are not standardized for accurate detection of PD-L1 and PD-L2 SVs. This may undermine clinical trials for PD-1 immunotherapy, where perceived response rates are considered too low, yet there may be a molecular subset of patients who are exceptional responders. One example is primary mediastinal B-cell lymphoma, which has a high prevalence of SVs, where PD-1 immunotherapy drug approval was based on high response rates in an unselected population. By contrast, PD-1 immunotherapy was considered ineffective as monotherapy in two studies for diffuse large B-cell therapy in patients without molecular profiling.^53,54 Present studies of PD-1 immunotherapy are combination studies and do not select patients for SVs.

In summary, there are diverse SVs that occur in both PD-L1 and PD-L2 across both solid and hematologic malignancies. We showed that a prominent proportion of curated events occur in the 3′-UTR of these genes. Furthermore, these tumors appear to have higher expression by RNAseq or IHC. There is considerable clinical signal in patients who have tumors with SVs and experience durable exceptional responses. These findings support further study of SVs as predictive biomarkers for PD-1 immunotherapy prospectively.

APPENDIX

FIG A1.

Landscape of PD-L1 and PD‐L2 structural variations from samples with multiple breakpoint calls. (A) Landscape of structural variations (deletions, inversions, and duplications, and unclassified) involving PD‐L1 and PD‐L2, color coded by alteration type representing 94 cases and 185 total structural variations. (B) Circos plot of intra‐ and inter‐chromosomal translocations involving PD‐L1 and PD‐L2 (n = 110). Key: red ‐ observed gene partners with PD‐L1; blue ‐ observed gene partners with PD‐L2; green ‐ observed gene partners with PD‐L1 and PD‐L2. (C) Pie chart of the distribution of structural variation types observed in PD‐L1 and PD‐L2, representing a total of 295 cases. SV, structural variation.

FIG A2.

PD-L2 expression in PD-L1 and PD-L2 altered tumors. Violin plot of PD‐L2 gene expression in PD‐L1 and PD‐L2 SV positive (n=37) and PD‐L1 and PD‐L2 diploid, or wildtype, tumors (n=5085). ****P < .001. RSEM, RNA‐seq by Expectation‐Maximization; SV, structural variation.

AUTHOR CONTRIBUTIONS

Conception and design: Emily L. Hoskins, Eric Samorodnitsky, Michele R. Wing, Julie W. Reeser, Karthikeyan Murugesan, Raven Vella, John Carpten, Sameek Roychowdhury

Administrative support: John Carpten, Matthew Reilley

Provision of study materials or patients: John Carpten, Jill Kolesar, Matthew Reilley, Nancy Single

Collection and assembly of data: Emily L. Hoskins, Eric Samorodnitsky, Karthikeyan Murugesan, Leah Stein, Zachary Risch, Serifat Adebola, John Carpten, Jad Chahoud, Stephen Edge, Jill Kolesar, Matthew Reilley, Abhishek Tripathi, Nancy Single, Lee A. Albacker

Data analysis and interpretation: Emily L. Hoskins, Eric Samorodnitsky, Michele R. Wing, Julie W. Reeser, Julia F. Hopkins, Karthikeyan Murugesan, Zheng Kuang, Raven Vella, Leah Stein, Lianbo Yu, Serifat Adebola, Anoosha Paruchuri, Martin McCarter, Kenneth G. Nepple, Matthew Reilley, Courtney Scaife, Abhishek Tripathi, Richard S.P. Huang, Lee A. Albacker

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/po/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).