This was a single institution, multicohort, investigator-initiated phase II trial of an oral HMA azacitidine (CC-486) and durvalumab in advanced solid tumors (METADUR). The study recruited patients between September 20 2016 and July 16 2018. The data cut-off date was April 30 2019. Primary endpoint was ORR, defined as the proportion of patients with complete response (CR) or partial response (PR) based on Response Evaluation Criteria in Solid Tumors V.1.1 (RECIST 1.1). Secondary endpoints included safety and tolerability, disease control rate (DCR) defined as CR, PR or stable disease (SD) for ≥16 weeks, progression-free survival (PFS) and overall survival (OS).

Patients ≥18 years of age, with histologically or cytologically confirmed advanced MSS-CRC, irrespective of RAS or RAF mutational status (MSS-CRC); platinum resistant OC; or ER+HER2- BC who had progressed on or were intolerant of prior standard therapy were eligible for enrolment. Patients could not have received prior treatment with an ICI or epigenetic modifier such as HMAs or histone deacetylase inhibitors. Other key eligibility requirements were Eastern Cooperative Oncology Group (ECOG) performance status of 0 or 1, adequate organ and bone marrow function and a life expectancy of ≥12 weeks. Patients were required to have RECIST measurable disease and at least one tumor lesion safely accessible for biopsy. Those with uncontrolled or symptomatic central nervous system metastases were excluded.

The initial treatment regimen included oral CC-486 300 mg once daily days 1–14 (cycles 1–3 only) in combination with durvalumab 1500 mg intravenous day 15, in 28 day-cycles (regimen A). However, 300 mg nce daily of CC-486 was poorly tolerated due to gastrointestinal (diarrhea, vomiting and nausea) and hematological toxicity (neutropenia), which are known to be associated with oral CC-486 (table 1).32 33 In preclinical models, lower doses of CC-486 at nanomolar concentrations are sufficient to induce the observed viral mimicry.11 Additionally, recent data suggest addition of vitamin C to HMA treatment protocols can enhance DNA demethylation and viral mimicry through the activation of TET enzymes.17 Therefore, a protocol amendment after the first 19 patients were enrolled included a reduced dose, increased duration, and altered schedule of CC-486 at 100 mg daily days 1–21 (cycle 1 and beyond), supplemented by continuous daily oral vitamin C 500 mg and maintained durvalumab 1500 mg intravenous day 15, in 28 day cycles (regimen B) (online supplementary figure 1). To improve tolerability, patients also received premedication with oral ondansetron 4–8 mg on days of CC-486 dosing in both regimens.

Adverse events (AEs) were assessed by Common Terminology Criteria for Adverse Events index V.4.03 and tumor response by RECIST 1.1 every two cycles. To evaluate pharmacodynamic effects, mandatory paired tumor biopsies were carried out at baseline (within 28 days of study treatment) and on treatment at cycle 2 day 12 (±2 days). Serial PBMCs were collected at baseline, day 1 and 15 of each cycle and at the end of treatment for immune-profiling and epigenetic analysis. All blood samples were collected prior to dosing. PD-L1 expression in tumor cells was assessed by immunohistochemistry using VENTANA SP263 assay (Roche Diagnostics), with positivity defined as staining in tumor cells >25%. Immune profiling to assess peripheral immune subsets was performed with multiparametric flow cytometry on serial PBMC samples following CC-486 treatment on cycle 1, day 15 (C1D15) and durvalumab cycle 2, day 15 (C2D15) treatment. DNA methylation status of tumor tissue and PBMCs was assessed using EPIC Methylation Array and LINE-1 methylation assay, respectively.

All tumor core biopsies and whole blood collected from patients were processed immediately after collection. Fresh core biopsies were pooled and minced into 2–4 mm² fragments and subjected to enzymatic digestion using the human tumor dissociation kit, as per manufacturer’s direction (Miltenyi, catalog no.130-095-929) and gentle MACS dissociator (Miltenyi, catalog no. 130-093-235). PBMCs were isolated from whole blood by Ficoll density gradient centrifugation. Cell pellets were stored in −80 ºC prior to nucleic acid isolation. Both RNA and DNA were isolated from this frozen material using the AllPrep DNA/RNA Mini Kit (Qiagen, catalog no. 80204), allowing for simultaneous extraction of DNA and RNA from PBMCs or tumor biopsies.

DNA collected from tumor biopsies was used to determine genome wide DNA demethylation measured by Infinium Methylation EPIC BeadChip Kit (Illumina, catalog no. WG-317–1001). DNA collected from PBMCs was used to measure global DNA demethylation induced by CC-486 treatment by LINE-1 methylation assay by pyrosequencing. RNA collected from tumor biopsies was used for total RNA-sequencing to determine global gene expression changes.

Quantification of DNA methylation levels of global LINE-1 retroelements was performed using pyrosequencing as previously described.34 In brief, a total of 500 ng DNA input was used for each sample with the PyroMark Q24 CpG LINE-1 kit (Qiagen, catalog no. 970012) to quantify methylation level of three CpG sites in positions 331–318 of LINE-1 (GenBank accession number X58075). Demethylation level of each of the three CpG sites were normalized against baseline methylation levels. The mean of all three CpG sites was calculated and plotted (along with the SD) as previously shown.35

A total of 250 ng of DNA was bisulfite converted using EZ DNA Methylation Kit (Zymo, catalog no. D5001), following the Illumina manufacturer’s protocol. The purified bisulfite converted DNA was then used as starting material for the Infinium MethylationEPIC bead chip, carried out by the Princess Margaret Genomics Centre, following manufacturer’s protocol. IDAT files for Illumina EPIC BeadChip data were processed using minfi R-package.36 Single sample Noob normalization was performed to yield normalized beta values that were used for further analysis. Limma R-package was used for statistical comparisons.37

Processed data were obtained from EBI array express for a previously published dataset of cell lines that were grown in the presence of azacitidine and profiled at multiple time-points on the Illumina 450 k platform.13 We computed beta-values for samples that had been grown in the presence of azacitidine for 7 days (beta-value >0.3, false discovery rate (FDR) <0.01). On the tumor biopsies, we identified all probes that were highly methylated (beta-value >0.8) at baseline and estimated beta-values for the same probes post-CC-486 treatment on the tumor biopsies.

Library preparation and sequencing: for all samples, total RNA was quantified by Qubit RNA kit (Thermo Fisher Scientific, catalog no. Q32852) and quality assessed by Agilent Bioanalyzer instrument. All libraries (rRNA depleted, total, single stranded RNA) were prepared using TruSeq Stranded Total RNA Library Prep Gold (Illumina, catalog no. 20020599) following manufacturer’s protocol. All libraries were normalized, pooled together and loaded at 1.4 pM onto an Illumina NextSeq cartridge for cluster generation. Library preparation and sequencing was performed by the Princess Margaret Genomics Center on an Illumina Nextseq500 instrument using paired-end 75 bp protocol to achieve ~80 million paired-reads per sample.

RNA-seq data were processed using Kallisto 38 to yield quantitative estimates of Gencode transcripts and lncRNAs as well as consensus sequence repeats for RepBase repeats. Data were processed using the limma-trend 39 framework to perform all statistical comparisons. Signature scores were summarized using single sample Gene Set Enrichment Analysis scores through the GSVA R-package. Gene set enrichment analysis was performed using test statistics with the fgsea R-package.

Deconvolution was performed using CIBERSORTx 40 with unlogged counts per million whole transcriptome data. The LM22 reference matrix was used to estimate the abundance of 22 cell types, both as a relative fraction and as an absolute fraction, wherein the averages of marker genes for all infiltrating cell types and other cell types are taken into account. Linear modeling was used to examine whether there were concordant shifts from baseline to post-therapy samples for both the values themselves. Further analysis was performed by examining whether the post-therapy versus baseline fold changes varied by regimen to ascertain whether the therapeutic regimen made a difference to any changes in the TME.

Fresh PBMCs were stained with immune markers, and data were acquired using a 5-laser LSR Fortessa X-20 (BD, Mississauga, Ontario, Canada). Flow cytometry data were analyzed using FlowJo software V.10.6.1 (Treestar, Ashland, Oregon, USA). For flow cytometry panels, please refer to online supplementary table 2.

Descriptive statistics, such as the mean, median, counts, frequency and proportion, were used to summarize the patient characteristics and AEs. Kaplan-Meier method was used for PFS and OS analysis. Estimated median and its 95% CIs were reported. For flow cytometry data, both paired t-test and non-parametric method were used. Results were consistent between the two methods; paired t-test results are indicated. Statistical analyzes were preformed using R V.3.5.2 (R Core Team 2014, Vienna, Austria).

The original planned sample size was 58 patients; 20 patients in the MSS-CRC cohort, 19 for both PR-OC and ER+HER2- BC cohorts. This sample would allow detection of a 20% difference in ORR from 0.05 to 0.25 in MSS-CRC and from 0.10 to 0.30 in both PR-OC and ER+HER2- BC with greater than 85% power at approximately 0.1 significance level. At least three responders would be required to claim a positive study in MSS-CRC, with at least four responders needed in PR-OC and ER+HER2- BC. Lastly, we observed zero response out of total of 28 patients, the probability of response rate being 0.25 or higher (alternative hypothesis) was calculated to be less than 0.001. Therefore, despite the fact that an interim futility analysis was not preplanned, the lack of antitumor activity seen after 19 patients on regimen A and 9 patients on regimen B led to early termination of this study.

In total, 28 patients were enrolled and treated in the METADUR trial. Nineteen patients were treated on regimen A and nine patients were treated in regimen B. Fifteen patients had MSS-CRC, four PR-OC, and nine ER+HER2- BC. Median age was 56 (range 36–78), 68% were female and 75% ECOG 1. All patients had received ≥3 prior lines of therapy, with a median of 7 (range 3–12) prior systemic therapies, 100% and 64% of patients underwent prior surgery and radiotherapy, respectively. Patient baseline characteristics are shown in table 2 and online supplementary table 1.

The most common treatment-related grade 1–2 AEs observed among all patients were fatigue (54%), diarrhea (46%), nausea (46%), vomiting (46%), anorexia (29%) and neutropenia (29%) (table 1). Toxicity was generally comparable between regimens, however the reduced dose and prolonged administration of CC-486 with vitamin C supplementation was associated with reduced hematological toxicity (neutropenia (11% vs 37%) and anemia (0% vs 16%), respectively). Grade ≥3 neutropenia was observed in 18% of patients across all cycles. Only one patient experienced febrile neutropenia. Other grade ≥3 AEs were rare. In regimen A, one patient experienced G3 AST/ALT increase and another patient experienced G3 anemia. In regimen B, one patient had asymptomatic G3 amylase and lipase elevation and two patients had G3 hyponatremia. Twenty-seven patients discontinued treatment due to disease progression, one patient discontinued due to an AE (G4 neutropenia on cycle 1, day 28). No patients required dose reductions of CC-486 in either regimen, however, on regimen A, 5% of patients experienced a 1-week delay on two occasions, 5% had three doses omitted and 16% did not receive durvalumab cycle 2, day 15. On regimen B, one patient did not receive cycle 1, day 15 durvalumab.

The median follow-up time was 4.7 months, median PFS was 1.9 (95% CI 1.5 to 2.3) months, and median OS was 5 (95% CI 4.5 to 10) months. No objective responses were observed as most patients experienced disease progression (online supplementary figure 2A), with a DCR of 7.1% (defined as SD for ≥16 weeks). For the platinum-resistant OC cohort, we measured cancer antigen 125 (CA125) in the blood, and all assessed patients experienced rise in CA125 levels during treatment (online supplementary figure 2B).

To address whether the lack of clinical response was due to low pharmacodynamic activity, we first sought to evaluate whether oral CC-486 in both of our treatment regimens were able to induce systemic DNA hypomethylation in PBMCs. Previous studies have reported correlation between global demethylation in whole blood and clinical response to DNA demethylating agents in myelodysplastic syndromes (MDS) and acute myeloid leukemia (AML) patients treated with HMAs.35 We performed global LINE-1 elements methylation assessment as a surrogate for genome-wide changes in DNA methylation.41 42

In most patients, CC-486 induced less than 10% demethylation of genome-wide LINE-1 repetitive sequences (figure 1). Regimen A was able to induce greater LINE-1 demethylation when compared with regimen B. Among the patients treated on regimen A, the maximum LINE-1 demethylation observed was 19.9% for ER+HER2- BC, 13.8% for MSS-CRC, and 9.2% for PR-OC. Among the patients treated on regimen B, the maximum LINE-1 demethylation observed was 6.8% for ER+HER2- BC, 6.6% for MSS-CRC, and 7.2% for PR-OC. Four patients in regimen A and none in regimen B achieved peak demethylation of at least 10% (figure 1). However, these patients did not experience clinical benefit, thus suggesting poor systemic pharmacodynamic activity of CC-486 in our study.

To directly evaluate whether the two oral CC-486 regimens in METADUR were able to induce DNA demethylation in tumor samples, Illumina EPIC methylation array platform was used that surveys around 850 000 CpG sites genome-wide in matched pretreatment/post-treatment pairs. A total of four pairs for regimen A and seven pairs for regimen B was profiled (n=11 patient-matched samples). Most samples showed no or very weak global DNA demethylation (figure 2A, online supplementary figure 3). Altogether, these results suggest that the evaluated regimens of oral CC-486 were not able to induce comparable tumor demethylation as preclinical models, explaining the discrepancy between clinical and preclinical responses.

To directly compare the levels of DNA demethylation in preclinical models and the tumor demethylation observed in the METADUR trial, we obtained in vitro demethylation estimates from a large panel of cancer cell lines (n=25 cell lines) grown in the presence of azacitidine and profiled in relation to mock-treated controls from a published source.13 A linear model fit on the delta-beta distribution from the patient biopsies reveal that tumor biopsies from regimen A or regimen B do not exhibit significant demethylation after controlling for baseline methylation levels of each patient, in sharp contrast to the preclinical models (figure 2A, p<2.2e-16).

Next, we sought to rule out the possibility that despite a lack of robust demethylation in tumor biopsies, CC-486 could still induce the inflammatory ‘viral mimicry’ response. Therefore, we performed total RNA-sequencing on all assessable paired-tumor biopsies from regimen A (n=10 pairs) and regimen B (n=6 pairs). Then, we examined the change in expression of repetitive elements that may play a role in inducing viral mimicry response on preclinical HMA treatment.10 11 43 A simple comparison of baseline versus post-CC-486 treatment transcriptomes revealed there were no consistent changes in the expression of these putative immunogenic repetitive elements (figure 2B). These patterns were highly variable, and no repetitive elements were differentially expressed between the groups or individual patients. Moreover, we observed neither significant activation of anti-viral pathways nor interferon response as previously reported in preclinical models and in clinical samples,9 11 13 44 consistent with our results that tumor DNA demethylation was not achieved in this study (figure 2A, online supplementary figure 3).

Finally, we took advantage of the available data and available biospecimen to evaluate whether the combination of CC-486 and durvalumab had an impact on the TME and on peripheral immune cell subsets. Using the same RNA-sequencing data from paired tumor biopsies, we applied CIBERSORT to score the infiltration of 22 immune cell types.40 There were no large-scale changes in either regimen A or Regimen B, with no cell type reaching significance at FDR <0.1 (figure 2C). In addition, the fold change in infiltration scores comparing post-CC-486 samples versus matched baseline samples showed low variability (figure 2D).

We also performed immunophenotyping by multiparametric flow cytometry (online supplementary table 2) in PBMC samples at baseline, cycle 1 day 15 (C1D15) and cycle 2 day 15 (C2D15) following CC-486 and durvalumab treatment. All evaluable patient samples, across all cohorts and treatment regimens were combined for our analysis. In comparison to baseline and C1D15, we observed a reduction in circulating monocytes (CD33⁺HLA-DR⁺CD14⁺) at C2D15 (p=0.09; vs baseline) (figure 3A), although not statistically significant. Similarly, when compared with baseline and C1D15, we observed a reduction in circulating CD3⁺ T cells at C2D15 (p=0.01, and p<0.01, respectively) (figure 3B). We did not observe changes in other peripheral immune subsets including CD3⁺CD4⁺ T cells, CD3⁺CD8⁺ T cells, regulatory T cells (Tregs; CD4⁺FOXP3⁺CD25⁺CD127^-), γδ T cells and NK cells following CC-486 and durvalumab treatment online supplementary figure 4. Ki67, a marker of cell proliferation, was assessed in CD4⁺ and CD8⁺ T cells and in Tregs. Between C1D15 and C2D15, we observed a minimum 1.5-fold increase of Ki67 expression (figure 3C) in CD4⁺ T cells in 8/17 patients, in CD8⁺ T cells in 12/17 patients, and in Tregs of 8/17 patients.