Patient demographics

A total of 59 patients were enrolled on this study from March 2012 to August 2019 (Table 1). The median patient age was 60 years (range: 32–85 years), and the median number of prior therapies was 5 (range: 0–19). Numerous tumor types were represented, with the most prevalent being colorectal and head and neck and carcinomas (14 and 12 patients, respectively).

Table 1 Patient characteristicsToxicity

The combination of oral FdCyd and THU was well-tolerated. The maximum tolerated dose was found to be 160 mg FdCyd administered orally once daily + 3000 mg THU administered orally once daily on days 1–6 and 8–13 of each 21-day cycle (dose level 8). As observed in the phase 1 study of the intravenously administered combination [7], the most commonly occurring drug-related grade 3/4 adverse events for the orally administered combination were hematological toxicities, followed by gastrointestinal toxicities (Tables 2 and 3). The twice-daily THU dosing schedule (dose levels 10 and 11) yielded greater toxicity than the single THU dose administered on lower total dose levels (Tables 2 and 3). Two patients had DLTs on DL9 (the highest once-daily dose level): 1 patient with grade 3 refractory nausea, vomiting, and diarrhea, and 1 with grade 3 diarrhea (Supplementary Table S2). Although no DLTs were measured in the initial 6 patients accrued to the lowest twice-daily dose level, DL10 (3 were inevaluable), 2 patients had DLTs (grade 3 nausea, vomiting, and diarrhea, and grade 4 thrombocytopenia) on DL11, which was the maximum administered dose. Accrual continued at DL10 but was halted after both of the accrued patients experienced DLTs: grade 3 oral mucositis and grade 4 neutropenia. These DLTs on DL10 and DL11 suggest that extending FdCyd exposure to the second 12-hour period each day may increase the likelihood of high-grade toxicities, even with lower Cmax and total exposure relative to the 160 mg daily administered at DL8. Following accrual of an additional 6 patients to DL8 and the resulting absence of any DLTs (Supplementary Table S2), DL8 was determined to be the MTD and RP2D. The expansion cohort patients enrolled on DL8 were also included in the toxicity analyses for this study (Tables 2 and 3, Supplementary Table S2).

Table 2 Adverse events attributed to oral FdCyd + THUTable 3 Adverse events attributed to oral FdCyd + THUClinical response

The best response to the oral FdCyd-THU combination was prolonged stable disease (Fig. 1). Six patients experienced stable disease for ≥ 8 cycles: 1 patient with head and neck squamous cell carcinoma (HNSCC) on DL5 who was on study for 17 cycles; 2 patients on DL4 (with breast and appendiceal carcinoma and on study for 14 and 8 cycles, respectively); 1 patient with colorectal carcinoma (CRC; 8 cycles) who started on DL9 and was then dose-reduced to DL8 and subsequently DL7 due to grade 3 anemia, abdominal pain, nausea, vomiting, diarrhea, and dehydration during cycle 1 and grade 3 dehydration during cycle 2, respectively; and 2 patients on the MTD/RP2D dose level (DL8), 1 with HCC and 1 with HNSCC on study for 12 and 8 cycles, respectively (Fig. 1).

Fig. 1

Clinical response to oral FdCyd + THU therapy. The number of cycles of treatment completed is shown for each patient on study for > 1 cycle, colored according to starting dose level (A) or tumor type (B). The best response measured on this study was stable disease. Patients who were biopsied for tumor PD analysis are indicated by asterisks

Pharmacokinetic analysis

Plasma pharmacokinetic analysis demonstrated that FdCyd concentrations peaked at approximately 1.2 h post-dose, with an elimination half-life of 1.5 h (Fig. 2A, Supplementary Table S3). FdCyd exhibited a largely dose-proportional exposure profile, though with substantial interpatient variability (Fig. 2B-C), and the target Cmax of 1 µM (245 ng/mL) [28] was reached by individual patients starting at 60 mg (Fig. 2B, Supplementary Fig. S1). Dose-normalized FdUrd concentrations at 2 h did not rise over days 1–3, but were higher with 6000 mg THU twice daily relative to 3000 mg THU once daily (Supplementary Fig. S2B, Supplementary Table S4). THU concentrations peaked at around 3 h post-administration and had a half-life of 8.7 h (Fig. 2D, E; Supplementary Table S5). Interestingly, the doubling of both THU dose (3000 to 6000 mg) and frequency (once to twice daily) resulted in an approximate 10-fold increase in 24 and 48 h trough concentrations as opposed to the expected 4-fold increase, possibly due to saturable and more prolonged absorption, consistent with the decreased dose-normalized Cmax and AUC0 − 6 as dose and schedule were intensified (Supplementary Table S5). The more intense THU exposure appeared to increase the dose-normalized FdCyd Cmax (P = 0.028; Supplementary Table S3; Supplementary Fig. S2A), although this effect could not be detected in the apparent clearance (P = 0.16; Supplementary Table S3). As noted above, 2 h FdCyd concentrations were also increased with the higher dose of THU. Surprisingly, at the higher THU dose, dose-normalized FdUrd Cmax (P = 0.0016) and AUC (P = 0.028) also increased. The FdUrd/FdCyd metabolic ratios of both Cmax (P = 0.93) and AUC (P = 0.82) were not impacted by the increase in THU (Supplementary Table S4). This suggests that additional THU increases the oral bioavailability of FdCyd but does not further reduce subsequent systemic cytidine deaminase-mediated metabolic conversion of FdCyd to FdUrd. Urine excretion data, though limited and at very low concentrations, also suggested increased FdCyd and FdUrd excretion with higher THU dosing (Supplementary Table S6). While there was no statistically significant relationship between week 1 cumulative FdCyd exposure and occurrence of DLT (Supplementary Fig. S3), patients who experienced a DLT had significantly higher THU exposures than those who did not (Fig. 2F); this difference remained statistically significant regardless of whether THU exposure was expressed as AUC0 − 6 (P = 0.007), Cmax (P = 0.005), C24h (P = 0.002), or C48h (P = 0.004).

Fig. 2

Plasma pharmacokinetic analysis of oral FdCyd combined with THU. A, Geometric mean plasma concentrations of FdCyd (⬤), FdUrd (), and FU (ρ) for patients receiving 160 mg FdCyd + 3000 mg THU (dose levels 5–9). Error bars indicate standard deviation; n = 31 patients. B, C, FdCyd exposure (Cmax [B] and AUC [C]) as a function of FdCyd and THU dose. Active target exposure of 245 ng/mL (1 µM) is indicated by thick horizontal black lines. D, E, Plasma concentrations of THU for patients receiving THU on a 3000 mg once daily (QD; D) or 6000 mg twice daily (BID; E) schedule. Plasma THU pharmacokinetic profiles for individual patients (dashed lines), along with geometric means (solid lines and circles), are shown; error bars indicate geometric standard deviations. F, THU exposure-toxicity relationship following oral FdCyd + THU administration. Closed circles represent geometric mean AUC0 − 6 values, while open circles indicate values for individual patients; error bars indicate geometric standard deviations. Samples with values below the lower limit of quantitation (200 ng/mL) were imputed as 100 ng/mL. Statistical significance (P = 0.007) is indicated by asterisks

Pharmacodynamic analysesAnalysis of DNMT1 levels and cell cycle arrest in tumor biopsy specimens

Because covalent trapping of DNMT1 on DNA has been shown to induce DNMT proteasomal degradation [20, 21], we assessed whether oral FdCyd-THU therapy yielded changes in tumor DNMT1 levels via IHC analysis of baseline (pre-dose) and on-treatment (Cycle 1 Week 3; C1W3) biopsy specimens. Of the 7 patients with assessable paired biopsies, none showed appreciable treatment-induced decreases in DNMT1 levels; indeed, 2 patients (5010001 and 1010020, with best responses of PD and 4 cycles SD, respectively) demonstrated increases in DNMT1 levels following FdCyd-THU administration (Supplementary Table S7; Supplementary Fig. S4), consistent with DNMT rebound effects reported for other DNMT-inhibiting agents [22, 23].

DNMT inhibition has also been shown to result in DNA damage and G2/M phase cell cycle arrest [24, 25], and we therefore assessed changes in tumor levels of the mitotic marker serine 10–phosphorylated histone H3 (pHH3) using a previously validated quantitative immunofluorescence microscopy assay for pHH3 [13] in baseline and C1W3 biopsies. Among the 6 patients with assessable paired tumor specimens, there were no appreciable changes in the percentage of pHH3+ cells in response to FdCyd-THU (Supplementary Fig. S5). This lack of treatment-induced reduction in pHH3 is consistent with the paucity of clinical activity for oral FdCyd-THU therapy.

P16 expression in tumor tissue and circulating tumor cells

Because p16 has been demonstrated to be a pharmacodynamic biomarker of FdCyd epigenetic effects [8], we assessed treatment-associated p16 modulation in both tumor and CTCs. Eight patients had baseline and/or on-treatment tumor tissue available for p16 IHC evaluation; tumors from all these patients had some level of p16 expression at baseline (Table 4). Of the 7 patients with paired biopsies of sufficient quality for analysis, just 2 showed qualitative modulation of tumor p16 staining in response to FdCyd-THU treatment: patient 1,010,019 exhibited a slight increase in p16 expression, while patient 1,010,020 exhibited a slight decrease (Table 4; Supplementary Fig. S6). The overall lack of substantial tumor p16 expression modulation is consistent with the general lack of antitumor activity of this regimen in these patients.

Table 4 Therapy-associated changes in tumor and CTC p16 expression in expansion cohort patients receiving oral FdCyd + THUTable 5 Epigenetically regulated genes with FdCyd-THU–induced promoter hypomethylation in patients with a best response of stable disease

To assess the effects of FdCyd on the epigenetic regulation of p16, we also assessed tumor promoter methylation of the p16-encoding gene CDKN2A, as measured in our genome-wide promoter hypomethylation analysis of tumor biopsy cores. A total of 6 expansion cohort patients had both reportable tumor CDKN2A hypomethylation data and qualitative tumor p16 IHC data available. Five of these 6 patients exhibited treatment-associated decreases in tumor CDKN2A methylation, yet only 1 showed increased tumor p16 expression by IHC (Table 4), indicating a potential discrepancy between the changes in transcriptional versus translational level regulation of tumor p16 expression at the examined timepoints.

A total of 30 patients were evaluable for assessment of FdCyd-induced changes in the proportion of p16-expressing (p16+) CK+ CTCs. Given intrapatient baseline variability in CTC numbers [26], as well as the extended timeframe required for DNMT inhibitor-mediated gene expression changes [8, 27], we assessed 3 different “baseline” specimens (C1D1 pre, C1D1 post, and C1D2) and, conservatively, used the highest percentage of p16+CK+ CTCs as the “baseline” value, as described previously [8]. Also as previously established [8], p16 re-expression was defined as a ≥ 3-fold increase in the percentage of p16-expressing CTCs at any post-treatment time point compared to baseline values; for patients with 0 p16+CK+ CTCs at baseline, an increase in the proportion of p16+CK+ CTCs was defined as an increase from 0 to ≥ 2 p16+CK+ cells. Per these criteria, the majority (23 of 30; 77%) of patients evaluable for p16 CTC response demonstrated a treatment-associated increase in the proportion of p16+CK+ CTCs (Fig. 3). Of the 7 patients who did not demonstrate such an increase, 3 had high baseline levels of p16+CK+ CTCs (≥ 20%). These rates of high-baseline p16+CK+ CTC prevalence and FdCyd-induced increases in the proportion of p16+CK+ CTCs are similar to those previously observed in a phase 2 study of intravenously administered FdCyd and THU [8].

Fig. 3

P16 re-expression in CK+ CTCs following oral FdCyd + THU therapy. The % CK+ CTCs positive for p16 over time is shown for patients in which FdCyd-induced CK+ CTC p16 re-expression was measured (A) and those for whom no such p16 re-expression was observed (B). Red asterisks indicate patients with a best response of stable disease. Data are shown for specimens collected during the first 2 cycles of treatment

To explore concordance across tumor and blood p16 measurements and whether CK+ CTCs are an adequate surrogate for tumor in PD assessments of FdCyd-associated changes in p16 expression, we compared CTC p16 expression changes with treatment-induced modulation of tumor p16 expression and CDKN2A hypomethylation. Among the 4 patients evaluable for p16 protein expression changes in both CK+ CTCs and tumor biopsy specimens (patients 3010009, 3010013, 4010014, and 4010016), none showed appreciable modulation of tumor p16 expression by IHC, yet 3 had significant FdCyd-induced increases in the proportion of p16+CK+ CTCs (Table 4). In contrast, there was high concordance between FdCyd-associated changes in tumor CDKN2A hypomethylation and p16 expression in CK+ CTCs among the 6 patients evaluable for both, with 5 of 6 patients exhibiting both increased tumor CDKN2A hypomethylation and CK+ CTC p16 re-expression and the remaining patient exhibiting no increases in either p16 measurement. Together, these data suggest that p16 protein expression in circulating CK+ tumor cells may be more sensitive to modulation by FdCyd relative to lesion-resident tumor cells.

The proportion of CTCs expressing p16 was also measured longitudinally for vimentin-positive (V+), mesenchymal-like phenotype CTCs, given the potential clinical significance of this CTC population [8, 28, 29]. We evaluated p16+V+ CTCs in a pilot study, as the technology for quantitating vimentin expression became available during the course of this study. Increases in the proportion of p16-expressing V+ CTCs were measured in just 2 of the 22 patients evaluable for p16+V+ CTC prevalence (9%)—a substantially lower prevalence than that observed for p16 re-expression in CK+ CTCs and consistent with the lack of p16 expression modulation observed in tumor p16 IHC analysis (Table 4; Supplementary Fig. S7).

FdCyd-mediated changes in genome-wide DNA methylation

To more broadly explore FdCyd-associated changes in DNA methylation patterns, we performed bisulfite sequencing on DNA isolated from paired biopsy specimens and measured genome-wide CpG methylation to identify genomic regions that were differentially hypomethylated in response to oral FdCyd-THU therapy. The overall number of genomic regions that were significantly demethylated at C1W3 relative to baseline ranged from 123 to 1821 (median, 179), though the maximum value of 1821 for patient 1,010,020 appeared to be an outlier that was likely due to the specimens for this patient being sequenced in a later, separate batch for which the sequencing read depth was greater than that of the prior batch (Supplementary Fig. S8A). There was no obvious association between the number of demethylated regions and clinical response (stable disease vs. progressive disease) to oral FdCyd-THU therapy (Supplementary Fig. S8A).

Genes that are epigenetically regulated and have significant FdCyd-induced promoter demethylation in patients with stable disease are of particular interest, as these changes in promoter methylation status may yield transcriptional changes that are associated with FdCyd-THU activity. Therefore, we identified genes with significant treatment-induced DNA hypomethylation within 5 kilobases (kb) upstream of the transcription start site (TSS) for the 3 patients with stable disease for whom these data were available: 1,010,020 (CRC, 4 cycles SD), 4,010,016 (HNSCC, 8 cycles SD), 4,010,017 (HNSCC, 6 cycles SD). There were no genes with significant FdCyd-induced promoter hypomethylation across all 3 of these patients (Supplementary Table S8; Supplementary Fig. S8B), suggesting heterogeneity in any epigenetic modulations that may mediate clinical benefit. We focused further analyses on genes considered to be likely epigenetically regulated and identified those as genes for which RNA expression was negatively correlated with DNA promoter methylation (Pearson r ≤ −0.3) across GDSC (Genomics of Drug Sensitivity in Cancer) cell lines based on CellMinerCDB data and analysis tools [16]. The tumors from patients 4,010,016 and 4,010,017 each had 14 such genes with significant promoter hypomethylation in response to FdCyd-THU treatment, including several genes with known tumor suppressor function: BMPR1B [30], CDH1 [31], FCGRT [32], LRIG3 [33], and MAMDC2 [34] for patient 4,010,016, and RASSF1 [35] and TET1 [36] for patient 4,010,017 (Table 5). Likely due to the greater sequencing read depth, patient 1,010,020 had a substantially greater number of such genes—a total of 187; these likewise included several tumor suppressor genes, such as ANGPTL4 [37], ARHGEF10 [38], MARVELD1 [39], and SQSTM1 [40]. The top most enriched molecular function gene ontology sets represented were transcription regulator activity, sequence-specific DNA binding, and DNA binding/transcription activator activity (Supplementary Table S8, Table 5). Together, these data suggest a potential role for re-expression of tumor suppressor and/or transcriptional regulator genes in mediating oral FdCyd activity.

In addition to these methylation changes within 5 kb of the TSS, we also assessed differentially methylated regions within intronic regions, as such intronic DNA methylation has been demonstrated to regulate cancer cell gene expression and alternative splicing [16, 41, 42]. As expected based on batch-specific differences in sequencing read depth, we found a substantially greater number of genes with FdCyd-induced intronic DNA hypomethylation for patient 1,010,020 compared to patients 4,010,016 and 4,010,017 (2116 vs. 311 and 307, respectively; Supplementary Fig. S8C). A total of 7 genes had significant intronic hypomethylation in response to oral FdCyd-THU in all 3 patients with stable disease, including documented tumor suppressor genes BCYRN1 [43], NLRP12 [44], and RUNX1 [45] (Supplementary Fig. S8C). These data suggest a potential role of intronic hypomethylation in mediating the activity of oral FdCyd-THU therapy.