PD‐L1 overexpression correlates with JAK2‐V617F mutational burden and is associated with 9p uniparental disomy in myeloproliferative neoplasms

Abstract Myeloproliferative neoplasms (MPN) are chronic stem cell disorders characterized by enhanced proliferation of myeloid cells, immune deregulation, and drug resistance. JAK2 somatic mutations drive the disease in 50–60% and CALR mutations in 25–30% of cases. Published data suggest that JAK2‐V617F‐mutated MPN cells express the resistance‐related checkpoint PD‐L1. By applying RNA‐sequencing on granulocytes of 113 MPN patients, we demonstrate that PD‐L1 expression is highest among polycythemia vera patients and that PD‐L1 expression correlates with JAK2‐V617F mutational burden (R = 0.52; p < .0001). Single nucleotide polymorphism (SNP) arrays showed that chromosome 9p uniparental disomy (UPD) covers both PD‐L1 and JAK2 in all MPN patients examined. MPN cells in JAK2‐V617F‐positive patients expressed higher levels of PD‐L1 if 9p UPD was present compared to when it was absent (p < .0001). Moreover, haplotype‐based association analyses provided evidence for germline genetic factors at PD‐L1 locus contributing to MPN susceptibility independently of the previously described GGCC risk haplotype. We also found that PD‐L1 is highly expressed on putative CD34+CD38− disease‐initiating neoplastic stem cells (NSC) in both JAK2 and CALR‐mutated MPN. PD‐L1 overexpression decreased upon exposure to JAK2 blockers and BRD4‐targeting agents, suggesting a role for JAK2‐STAT5‐signaling and BRD4 in PD‐L1 expression. Whether targeting of PD‐L1 can overcome NSC resistance in MPN remains to be elucidated in forthcoming studies.

highly expressed on putative CD34 + CD38 À disease-initiating neoplastic stem cells (NSC) in both JAK2 and CALR-mutated MPN. PD-L1 overexpression decreased upon exposure to JAK2 blockers and BRD4-targeting agents, suggesting a role for JAK2-STAT5-signaling and BRD4 in PD-L1 expression. Whether targeting of PD-L1 can overcome NSC resistance in MPN remains to be elucidated in forthcoming studies.

| INTRODUCTION
Myeloproliferative neoplasms (MPN) are chronic bone marrow (BM) disorders characterized by clonal hematopoiesis, overproduction of myeloid cells, inflammation, and immune deregulation. 1,2 The three classical BCR-ABL1-negative MPN are essential thrombocythemia (ET), polycythemia vera (PV), and primary myelofibrosis (PMF). The JAK2-V617F mutation drives the disease in $95% of PV patients and 50%-60% of all cases with ET and PMF. [3][4][5][6] Somatic mutations affecting CALR are disease drivers in 30%-40% of ET and PMF patients, 7 while MPL mutations account for 5%-10% of all MPN cases. 8,9 MPN patients with unknown disease drivers (no driver mutation detected) are referred to as´triple negative´MPN. Progression to secondary acute myeloid leukemia (sAML) may occur in all three MPN entities, but is more frequently documented in patients with PMF compared to those with ET and PV. 10 Most of the current therapeutic approaches in MPN are not curative, and disease management focuses on amelioration of symptoms. 11,12 The only exception is allogeneic hematopoietic stem cell transplantation, a procedure that is associated with a relatively high mortality risk, and is, therefore, performed only in younger patients with advanced disease or sAML. 11 Lack of curative potential and commonly occurring resistance to drug therapies applied in MPN patients highlight the need for the development of novel therapeutic concepts. One strategy may be to establish drug-based therapies or antibody-based immunotherapies capable of targeting and eradicating the disease-initiating neoplastic stem cells (NSC) in MPN. 13 While immune dysregulation in MPN has been well documented, the mechanisms by which the MPN (stem) cells escape the antitumor immune response remain unclear. 1 Published studies have shown that MPN cells display certain immune checkpoint molecules, including PD-L1, which may contribute to resistance, and that constitutive activation of JAK2 by the V617F mutation causes PD-L1 upregulation in MPN cells. 14,15 PD-L1 and PD-L2 expression by neoplastic cells and binding of these molecules to the PD-1 receptor on T cells inhibits the antigenspecific Tcell-mediated antitumor immune response, and represents a major mechanism of immune escape. 16,17 PD-L1 is not only expressed on cells of many tumors, but also cells within the tumor microenvironment in inflammatory conditions, thus reducing antitumor immunity. 18,19 Upregulation of PD-L1 in cancer cells can be a consequence of PD-L1 gene amplification, activation of oncogenic signaling pathways, or epigenetic regulation, but it can also be induced by certain pro-inflammatory cytokines. 20 Blocking of the interaction between the PD-1 receptor and its two ligands potentiates antitumor T-cell responses and in line with this notion, checkpoint inhibitors targeting the PD-1/PD-L1 interaction have emerged as highly promising anticancer drugs, leading to durable responses in various solid tumors. [21][22][23][24] The use of PD-1/PD-L1 interaction-targeting agents in MPN is currently under investigation, however, it is unclear, which MPN subtypes are most suitable for testing in clinical trials.
The genes encoding the PD-L1 (CD274, alias PD-L1), PD-L2 (CD273, alias PD-L2), and JAK2 proteins are located in close proximity on the short arm of chromosome 9, which is often affected by copy number neutral loss of heterozygosity through the mechanism of acquired uniparental disomy (UPD) in MPN patients, usually leading to an increase in the JAK2-V617F mutational burden. 25 UPD of chromosome 9p is the most common chromosomal aberration found in MPN. It affects up to 80% of PV patients, close to half of PMF and sAML patients, and 6%-18% of cases with ET. 26 However, it is unknown whether chromosome 9p UPD affects PD-L1/2 genes, which have a more centromeric position than JAK2. In particular, it is not known whether chromosome 9p UPD affects PD-L1/2 expression levels in MPN cells.
In the current study, we analyzed PD-L1 expression in neoplastic cells in a well-characterized cohort of MPN patients, for which in-depth genomic data were available. We combined whole-transcriptome data with chromosomal aberration profiles based on SNP arrays and mutational profiles from targeted sequencing using a myeloid gene panel.
We show that PD-L1 overexpression in MPN is associated with chromosome 9p UPD and that it correlates with JAK2-V617F mutational burden in granulocytes. We also found that germline genetic factors at the PD-L1 gene locus contribute to MPN susceptibility, and we demonstrate that PD-L1 is upregulated on the cell surface of phenotypically defined MPN NSC.  Table S1. Diagnostic criteria were applied as described previously. 5

| RNA-sequencing and data analysis
The evaluation of the expression of PD-L1/2 and other selected genes is based on a large RNA-sequencing data set published in Schischlik et al. 27 Further details are available in Appendix S1.

| Targeted DNA-sequencing and microarray analysis
Methods used for targeted DNA sequencing with the TruSight Myeloid Sequencing Panel (Illumina, San Diego, CA) and microarray analysis using Genome-Wide Human SNP 6.0 arrays (Affymetrix, San Diego, CA, USA) are described in the Appendix S1.

| Cell lines and in vitro studies
In vitro studies were performed using HEL, SET-2, and UT-7 cell lines.
UT-7 cells were engineered to express various CALR mutants using CRISPR/Cas9 technology as described recently. 28 Methods related to in vitro studies as well as 3 H-thymidine incorporation assay are described in the Appendix S1.

| Evaluation of expression of PD-L1/2 and PD-1 on primary MPN cells and cell lines
Heparinized BM or PB samples or cell lines were incubated with various combinations of monoclonal antibodies (Table S2) at room temperature in the dark for 15 min. As normal control, we used commercially available BM cells from healthy donors (Lonza, Basel, CH) as previously described. 29 After incubation, erythrocytes were lysed using fluorescence-activated cell sorting (FACS)-Lysing Solution (BD Biosciences, San José, CA, USA). Cells were washed and analyzed by flow cytometry using FACSCanto II (BD Biosciences, San José, CA, USA) essentially as decribed. 30 FlowJo software (version 8.8.7, TreeStar, Ashland, OR, USA) was used for data analysis. Antibodystaining results were controlled by applying isotype-matched control antibodies, and were expressed as either percentage of positive cells or as staining index, which represents the ratio of median fluorescence intensities obtained with the specific monoclonal antibody and the isotype-matched control antibody. Phenotypic NSC were defined as CD34 + CD45 dim CD38 À while progenitors were defined as CD34 + CD45 dim CD38 + cells. 31,32 For detection of T cells, we used monoclonal antibodies against CD45, CD3, CD4, and CD8, while B cells were analyzed using CD19 antibody and NK cells by using a CD56 antibody. Gating strategies were applied as described previously. 30 To assess the effects of drugs on PD-L1 expression, primary

| Linkage analysis, association analysis, and haplotype-based expression analysis
Linkage analysis for the JAK2 and PD-L1 loci was performed using the LDpair and LDmatrix functions as implemented in the LDlink webbased toolset (https://ldlink.nci.nih.gov/). Haplotype-based association analyses were performed using previously published SNP-array data from 272 MPN patients from Vienna, Austria 33 in conjunction with 1620 Bavarian controls from the German Cooperative Health Research in the Region of Augsburg cohort (KORA). 34 Cohort analyses of genetic data, including haplotype determination were performed using PLINK. 35 Association analyses based on haplotype frequency distributions were performed using Fisher's exact test as implemented in R. 36 Haplotype-based comparative expression analysis was performed for MPN patients with both SNP-array and RNA-Seq data available. Statistical evaluation was performed using unpaired t-tests as implemented in Prism (version 8.0.0, GraphPad Software, La Jolla, CA, USA). 3.2 | PD-L1 expression correlates with the JAK2-V617F mutational burden and is higher in patients with UPD of chromosome 9p

| Statistical analyses
As PD-L1 and JAK2 are both located on the 9p24.1 locus (320 kb apart; Figure 1C), we hypothesized that the presence of chromosome 9p UPD has an effect on the observed PD-L1 expression in MPN cells, which would explain why MPN cells in patients with PV have higher PD-L1 expression levels compared to other disease entities. To test this hypothesis, we analyzed a previously published cohort of MPN patients for whom we had available data from Genome-wide Human Affymetrix 6.0 SNP arrays. 33 We detected chromosome 9p UPD in 195 MPN patients, while 15 patients carried three copies of chromosome 9p (9p gains), out of 408 patients analyzed. In all of the identified MPN cases with chromosome 9p aberrations, we observed that the commonly affected region invariably covers both, the JAK2 and PD-L1 genes ( Figure 1C). As PD-L1 has a more centromeric position, it is possible that this gene represents the second target of chromosome 9p UPD in MPN.
To determine the presence of 9p UPD in our cohort of patients characterized by RNA-sequencing, we analyzed the SNP-array data that was available for 73 JAK2-V617F positive patients, and found that 37 patients carried the 9p UPD, while in 36 patients this aberration was absent. To assess the effect of the 9p UPD on PD-L1 expression in MPN cells, we compared the PD-L1 mRNA expression between JAK2-V617F-positive patients with and without 9p UPD, and observed a significantly higher expression of PD-L1 in MPN cells in patients with 9p UPD (p < .0001; Figure 1D). In addition, we found that the presence of the 9p UPD leads to higher PD-L1 expression in various MPN subsets ( Figure S3). Next, we performed correlation studies, which revealed that PD-L1 levels correlate with the JAK2- JAK2-V617F mutational burden was lost when cases with 9p UPD were excluded from these analyses (p = .9, R = 0.03; Figure 1F).

| Germline genetic variation at the PD-L1 gene locus affects MPN risk possibly through acting on PD-L1 expression
Previous data have shown that a common haplotype referred to as GGCC (also: 46/1) haplotype preferentially acquires JAK2-V617F and thus confers susceptibility for MPN. 37 Our data suggest that the presence of chromosome 9p UPD might be relevant for PD-L1 upregulation also at the more distal PD-L1 locus. Therefore, we evaluated a possible role of germline genetic variation at the PD-L1 locus and its interplay with the GGCC haplotype at the distal JAK2 locus. We performed linkage analysis, including the GGCC haplotype and an SNP within the 3'UTR of PD-L1 (rs4143815), as there were no suitable coding SNPs available. We did not observe a strong linkage disequilibrium (LD) between the JAK2 and PD-L1 loci (D 0 = 0.1171, R 2 = .0112), however, the presence of a haplotype block (genomic region with low recombination rate and low haplotype diversity) spanning both loci at weak LD could not be excluded ( Figure 2A). Next, we determined the frequencies of the possible JAK2/PD-L1 haplotypes in a large SNP-array-typed MPN cohort (n = 272) 27 and a population-matched non-MPN control cohort (n = 1620; Table S3). Haplotype-based association analyses on the observed counts suggested the PD-L1 rs4143815 minor allele as risk factor for MPN susceptibility independent of the JAK2 GGCC haplotype tagged by rs10974944 (Table 1). This was observed on both GGCC risk (rs10974944 minor allele; p = .04) and GGCC protective (rs10974944 major allele; p = .02) backgrounds (Table 1) T A B L E 1 Association analysis for an MPN patient cohort (n = 272) versus a non-MPN control cohort (n = 1620) for haplotypes, including both JAK2 (rs10974944) and PD-L1 (rs4143815); results from homozygous calls at both loci are shown   The ability of BET inhibitors, such as JQ1, to suppress constitutive or IFN-γ induced PD-L1 expression in certain cancer cell lines as well as stem cells of chronic myeloid leukemia patients were previously described. 52 also showed little effects on disease progression to post-MPN sAML. 54,55 It is also of interest to note that BET protein bromodomain targeting agents were previously shown to be highly active against post-MPN sAML cells, and to exert a synergistic effect with ruxolitinib. 56,57 As previous reports also demonstrated synergistic effects of anti-PD-1 antibodies with BET inhibitor JQ1, 52 our data indicates that such applications should be assessed in future studies in the context of MPN.
In conclusion, our data show that PD-L1 is expressed abundantly in MPN cells, including MPN-initiating phenotypically defined CD34 + CD45 dim CD38 À NSC. In addition, we demonstrate that PD-L1 levels are highest in neoplastic cells in patients with PV, correlate with the JAK2-V617F burden and with chromosome 9p UPD, and are promoted by IFN-γ exposure through a BRD4/MYC-dependent pathway.
We also provide first evidence that germline genetic factors at the

DATA AVAILABILITY STATEMENT
The RNA-sequencing and microarray data have been deposited in the following repositories: European Genome-phenome Archive; Accession ID: EGAS00001003486 and Array express; Accession ID: E-MTAB-1845.