This single arm proof-of-concept phase II study was approved by the Institutional Review Board of the Center for Cancer Research (CCR), National Cancer Institute (NCI), Bethesda, Maryland, USA, and written informed consent was obtained from participating patients. Comprehensive enrollment criteria have previously been published.19 Briefly, eligible patients had recurrent HGSOC, an Eastern Cooperative Oncology Group (ECOG) Performance Status of 0 to 2 and good end-organ function. The BRCAwt HGSOC cohort completed enrollment in November 2016 for which details were published and data from that cohort are used for this pilot project.19 The BRCA mutation HGSOC cohort is currently enrolling the patients (NCT02203513). There was no limit on number of prior therapies, however, patients had to have at least 4 weeks wash-out period from their last treatment. Clinical study details, including drug administration, safety, adverse events, and tumor response in the BRCAwt HGSOC cohort have been reported19 and the schema for treatment is shown in figure 1A. Paired blood samples, taken before treatment was started at baseline (cycle 1 day 1 (C1D1)) and 6 to 24 hours post second dose on day 15 of cycle 1 (C1D15), were collected from all patients. Pre-treatment and C1D15 percutaneous core biopsies were obtained by interventional radiologists under CT or ultrasound guidance using local anesthesia. All core biopsy samples were processed immediately in real time into optimal cutting temperature compound and stored at −80°C prior to analysis. Paraffin-embedded archival tissue samples from the time of original diagnosis were obtained at trial enrollment for immunohistochemistry as fresh frozen biopsy samples did not yield evaluable immunohistochemistry findings.

Cells were incubated with Fc receptor blocking agent (Miltenyi Biotec, Bergisch Gladbach, Germany) and stained with monoclonal antibodies for 20 min at 4°C. For intracellular staining for Foxp3 and Ki67 expression, cells were fixed and permeabilized using a Fix/Perm buffer (eBioscience, San Diego, California, USA) according to the manufacturer’s instructions, then stained with anti-Foxp3 or anti-Ki67 antibody. Live cells were discriminated by means of LIVE/DEAD Fixable Aqua Dead Cell Stain (Life Technologies, New York, USA) and dead cells were excluded from the analysis. All analyzes were performed using multiparametric flow cytometry (MACSQuant; Miltenyi Biotec, Bergisch Gladbach, Germany) and data were analyzed using FlowJo software ​V.10.0.7 (FlowJo, LLC, Ashland, Oregon, USA). Gating was first performed on peripheral blood mononuclear cells (PBMCs) with doublet exclusion and then on viable cells. CD8+ or CD4+T cells were further gated for functional markers PD-1, glucocorticoid-induced TNFR-related protein (GITR), Human Leukocyte Antigen DR antigen (HLA-DR), inducible T-cell costimulatory (ICOS) or Ki-67. Monocytic myeloid-derived suppressor cells (M-MDSCs) were defined as CD11b+CD14+HLA-DR-CD15-, early-stage MDSCs (E-MDSCs) as CD11b+CD3-CD14-CD15-CD19-CD56-HLA-DR-CD33+ and polymorphonuclear MSDCs (PMN-MDSCs) as CD11b+CD14 CD15+. For monocytes, total monocytes were subdivided by CD14 and CD16 expression: classical monocytes (CD14 ++CD16−); non-classical monocytes (CD14 +CD16++); and intermediate monocytes (CD14 ++CD16+). Representative figures describing detailed flow cytometry gating strategies can be found in online supplemental figure 1. The monoclonal antibodies used (all from BioLegend, San Diego, California, USA) are listed in online supplemental table 1.

Total RNA was isolated from pre-treatment and available C1D15 fresh-frozen core biopsies using the RNeasy microkit (Qiagen). RNA quality was evaluated using the Agilent Bioanalyzer 2100 and RNA integrity number (RIN) values were ensured to be >8.0. For total RNA sequencing (RNAseq), each sample (20 to 100 ng) was preprocessed with NEBnext rDNA depletion kit (New England Biolabs, Ipswich, Massachusetts, USA) to remove ribosomal RNA, then barcoded and pooled to ensure at least 100 million reads per sample on a HiSeq3000 sequencing system (Illumina, San Diego, California, USA). The human reference genome Hg38 was used to align reads and gene expression data were generated as counts per million mapped reads (CPM) values. Quality check of sample and sequencing outputs were performed by the CCR sequencing facility and CCR collaborative bioinformatics resource at NCI, Bethesda, Maryland, USA. As previously described,19 RNAseq data of normal ovarian tissues were obtained from the Genotype-Tissue Expression project (GTEx project, NIH; https://www.gtexportal.org/home/, last accessed on January 13, 2019) as Reads Per Kilobase of transcript per Million mapped reads (RPKM) values and then converted to CPM values using the formula RPKM=CPM/L, where L is the exonic length in kilobases. Log2 of CPM values were used throughout. Biopsies determined to be derived from normal tissue were excluded from analysis. HGSOC subtype was determined using a validated single sample Gene Set Enrichment Analysis (ssGSEA) signature as has been previously described.20 21 IFN-γ and expanded immune gene signature scores (online supplemental table 2) were calculated by averaging the normalized log2 expression levels of the included genes, as described by Ayers et al.22 CIBERSORT, a computational algorithm for quantifying immune cell fractions using gene expression profiles, was performed to determine tumorous immune infiltration as previously described.23

Pre-treatment formalin-fixed paraffin-embedded (FFPE) tissue samples from 24 patients were sent to Johns Hopkins Hospital for pathologic correlative studies; the tissue from one patient (Study ID 40) was inadequate for correlative studies due to tissue preservation. H&E-stained slides of the tissue samples from the remaining 23 patients were evaluated to confirm the pathologic diagnosis of high-grade ovarian carcinoma and to score the presence of tumor-infiltrating lymphocytes (TILs) and tumor-associated macrophages (TAMs). TIL were scored as the percentage (0% to 100%) of tumor stroma area occupied by mononuclear inflammatory cells, as previously described.24 TILs were not scored in tumor specimens lacking stroma, that is, those specimens consisting entirely of detached fragments of carcinoma cells without interface to adjacent stroma. The presence of TAM was recorded as: score 0 (absent), score 1 (focal, largely confined to papillary cores), score 2 (non-focal, involving both papillary cores and tumorous stroma), and score 3 (diffuse, involving papillary cores, tumorous stroma, and carcinoma cell nests). Immunohistochemistry for CD68 (Clone KP1, Dako; Bond RX, Bond Polymer Refine Detection) was performed on all cases to confirm the histologic impression of TAM.25

Immunohistochemistry for PD-L1 was performed manually on unstained slides (PD-L1 clone SP142, 0.096 µg/mL concentration; Spring Bioscience, Pleasanton, California, USA), as described.25 Specifically, the Spring Bioscience PD-L1 clone SP142 was used with a laboratory-developed assay validated against the labeling results of comparable assays.25 Briefly, slides were deparaffinized and rehydrated using standard methods. Antigen retrieval was performed in pH 6.0 CB buffer in a decloaking chamber (Biocare Medical, Pacheco, CA, USA), and slides were treated with peroxidase, protein, avidin, and biotin blocking, and incubated with the primary antibody overnight at 4°C. Slides were then treated with a biotin-labeled anti-rabbit secondary antibody (1 µg/µl concentration). The signal was developed with horseradish peroxidase using the VECTASTAIN Elite ABC Kit (Vector Laboratories, Burlingame, California, USA), amplified using a Tyramide Signal Amplification PLUS Biotin KIT (dilution 1:50; Perkin Elmer, Waltham, Massachusetts, USA), and visualized by 3,3'-Diaminobenzidine (DAB). PD-L1 labeling of carcinoma cells was evaluated in all cases and was scored as percentage (0% to 100%) of malignant cells with membranous labeling, with a cut-off of >1% carcinoma cell labeling considered positive.

Comparison of correlative markers were performed using a non-parametric Mann-Whitney test (for unpaired analyses) or a non-parametric Wilcoxon matched-pairs signed-rank test (for paired analyses) using GraphPad Prism, V.6. Immunohistochemistry data analyses were performed using Microsoft Excel (t-test) or GraphPad Online (Fisher’s exact test). The significance of difference between Kaplan-Meier survival curves was compared with a Mantel-Cox log-rank test. All statistical tests used two-sided significance with a cut-off of p=0.05 and are reported without adjustment for multiple comparisons due to the small study cohort and exploratory nature of this analysis.

The trial was approved by the Institutional Review Board of the CCR, NCI, Bethesda, Maryland, USA. All patients provided written informed consent before enrollment.

The treatment schema and correlative studies are shown in figure 1A. Characteristics of the 24 evaluable BRCAwt HGSOC patients enrolled in this phase II trial have previously been described in detail19 and are summarized in table 1. Briefly, in this group of heavily pre-treated patients (median 4.5 prior treatments (1 to 13)), there were eight (33%) partial responses (PR) and six (25%) patients with stable disease (SD) ≥6 months for a clinical benefit rate (CBR; PR +SD ≥6 months) of 58%. For this post hoc exploratory biomarker study, not all patients had available samples from each translational endpoint but 100% of patients had paired pre-treatment and post-treatment blood among which 96% (23/24) were evaluable for flow cytometry, 88% (21/24) had evaluable archival tissue from primary diagnosis, and 38% (9/24) had both paired blood and fresh core biopsy samples from this trial; table 1 also includes the number of patients within each of the substudy elements.

Our findings from paired pre-treatment and post-treatment PBMCs showed an increase in the number of lymphocytes with γ-H2AX staining by flow cytometry, a marker of double-stranded DNA breaks.26 Specifically, we observed increases in median fluorescent intensities (MFIs) of γ-H2AX among CD3⁺ T (median 1.63 vs 1.74, p=0.03), CD19⁺ B (median 1.69 vs 1.76, p=0.04), and CD56+Natural Killer (NK) lymphocytes (median 1.82 vs 2.05, p=0.02), suggestive of increased DNA damage in patients treated with CHK1i (figure 1B). Moreover, the observed increase in γ-H2AX corresponded to treatment-induced lymphodepletion as nearly 80% of patients experienced grade 3 or 4 reduced white blood cell counts,19 consistent with previous studies correlating double-stranded DNA damage with peripheral lymphocyte apoptosis.27 28 Change in γ-H2AX was not associated with clinical response or progression-free survival (PFS).

To evaluate changes in the peripheral innate immune response, immunostimulatory cell (ie, monocyte) and immunosuppressive cell (ie, myeloid-derived suppressor cell (MDSC)), subpopulations were compared in paired PBMC samples (online supplemental table 3). There was an increase in percent monocytes per total viable cells (median 31.6% (pre) vs 45.6% (post), p=0.005, figure 2A) after CHK1i treatment, although this did not correlate with clinical response or PFS. However, an increase in the immunocompetence functional marker HLA-DR on total monocytes, seen in 12 of 23 patients (MFI fold-increase 1.15 (1.01 to 2.76)), was associated with improved PFS (9.25 vs 3.5 months, p=0.019, figure 2B) suggesting innate immune activation of monocytes may contribute to the clinical efficacy of CHK1i. Similarly, an increase in HLA-DR on classical and non-classical monocytes, seen in 9/23 (39%) and 10/23 (43%) of patients, respectively, correlated with improved PFS (9.5 vs 3.75 months, p=0.04; 11.5 vs 4 months, p=0.002, figure 2C and D). No differences were seen after treatment in MDSC or other immunosuppressive cell subtypes. We further performed differential gene analysis of RNAseq data set from paired tissue samples to identify prexasertib-induced tumorous innate immune activation. We found no significant changes in the RNA expression levels of several cytokines known to play a critical role in the innate immune response, including IL6, IL8, IL12, TNFα and IFNγ.29 We also investigated changes in mRNA expression levels of key STING pathway genes, including cGAS, STING, TANK-binding kinase 1 (TBK1) and IFN regulatory factor 3 (IRF-3),13 30 31 and observed increased expression of TBK1 among six of nine evaluable patients with paired biopsy samples (median fold-increase 1.14 (1.02 to 1.33)). No other genes were significantly changed. TBK1 leads to phosphorylation and dimerization of IRF-3 and ultimately induces transcription of type I IFN genes,31 and increased expression was associated with improved PFS (9 vs 3 months, p=0.003, figure 2E) suggesting STING pathway activation may contribute partly to response to treatment as demonstrated preclinically.13 Downstream expression of IFNβ was undetectable and expression of the chemokines CXCL10 and CCL5, which are direct targets of IRF-3 and have been shown to increase after STING activation,32 were not significantly different after CHK1i treatment.

To characterize tumorous adaptive immune responses to prexasertib treatment, CIBERSORT, an algorithmic tool that estimates proportions of tumor-infiltrating immune cells using RNAseq data,23 was utilized on paired tumor biopsies. We found increased proportions of naïve B-cells (0.5% vs 14%, p=0.003) and resting CD4 memory T-cells (19% vs 25%, p=0.03), as well as decreased resting mast cells after CHK1i (5% vs 0.8%, p=0.03) (figure 3A). However, none of these changes were associated with clinical response or PFS, and there was no significant change in activated T-cells. We also noted greater tumorous infiltration by T-regs (median=2.9%, (range 0% to 6.3%)) was associated with a significantly worse PFS (9.25 vs 3.5 months for above versus below median, p=0.007; figure 3B), suggesting cross-talk between cell cycle checkpoint inhibition and the adaptive immune response influences clinical activity of prexasertib.

Changes in the peripheral adaptive immune response, including regulatory T-cells (T-regs) and effector T-cells, were compared between pre-treatment and post-treatment PBMCs using flow cytometry (online supplemental table 3). Consistent with the decreased white blood cell count observed in 82% of patients on this trial,19 CHK1i induced peripheral lymphodepletion, including an overall decrease in both CD4+ and CD8+T cells (25% vs 19.5%, p=0.008% and 9.5% vs 7.6%, p=0.005, respectively; figure 3C). There was no significant change in the overall proportion of immunosuppressive T-regs or effector T-regs (eT-regs), defined as active FOXP3^highCD45RA⁻ cells,33 among total viable cells. However, a slight increase in the proportion of T-regs and eT-regs was observed among CD4+T cells (4.8% vs 6.8%, p=0.01; 1.5% vs 2.5%, p=0.009, respectively; figure 3D), suggestive of an immunosuppressive environment. No significant changes in Ki67+CD4+ or CD8+effector T-cells were observed after treatment.

Next, using the paired tissue RNAseq data set, we evaluated changes in pre-treatment versus post-treatment expression of an INF-γ and expanded immune gene signature incorporating key genes involved in the adaptive immune response22 (online supplemental table 2). No significant changes were observed. Finally, as CHK1i has been demonstrated to modify PD-L1 expression preclinically,11–13 changes in expression levels of genes involved in immunosuppression, including PD-1, PD-L1, and cytotoxic T-lymphocyte-associated protein 4 (CTLA-4), were evaluated using RNAseq from nine paired pre-treatment and post-treatment biopsies. No significant changes were observed.

To evaluate whether the tumorous immune environment is prognostic and potentially predictive of response to CHK1i, archival tissue biopsies from the time of initial diagnosis (n=21) or recurrence (n=2) available from 96% (23/24) of patients were examined via H&E and immunohistochemistry for the presence of TILs. TILs were evaluable in 21 patients, with two excluded due to tissue fragmentation and lack of stroma, and contained an average of 20% TIL (range 1% to 90% stromal TIL; online supplemental table 4). In line with reports demonstrating an association between TILs and survival in ovarian cancer patients,34 PFS was significantly improved among patients with >30% TIL (13.7 vs 5.5 months, p=0.05). Moreover, the presence of >30% TIL was seen only in patients who experienced a clinical benefit (25% vs 0%, p=0.22) and the tumors of patients who experienced a clinical benefit from therapy showed a trend towards greater TIL compared with those who lacked clinical benefit (27% vs 11%, p=0.07), suggesting TIL levels may serve as both a prognostic and possibly predictive biomarker, requiring further validation (online supplemental figure 2). Additionally, no association was observed between archival tissue TAMs or PD-L1 carcinoma labeling (positive versus negative) and response to therapy (online supplemental table 4).

As the archival tissue immune environment was associated with response to CHK1i, we next assessed whether this was similarly true of pre-treatment tumor biopsies at the time of enrollment. Eighteen available pre-treatment biopsies were classified into four validated transcriptomic molecular ovarian cancer subtypes, which have been shown to carry prognostic and possible therapeutic relevance20 21 35 (figure 4). As with previous studies, individual tumor samples expressed multiple signatures at different levels of activation.35 Thirty-nine percent (7/18) expressed high levels of the immunoreactive subtype gene signature, associated with immune response and characterized by enhanced cytokine expression, T-cell activation, and TILs.20 35 The immunoreactive subtype did not correlate with PFS or response to CHK1i, possibly due to small numbers.

In addition, the pre-treatment biopsies of the patients with immunoreactive subtype signatures contained more stromal TILs (mean 26% and median 25%) than those with non-immunoreactive subtype signatures (mean 17% and median 10%; p=0.37). Similarly, a higher proportion of pre-treatment biopsies of the immunoreactive subtype signature contained PD-L1⁺ carcinoma cells (68%) than those with non-immunoreactive subtype signatures (38%; p=0.39). These features are indicative of an active immune response.