Vitamin C increases viral mimicry induced by 5-aza-2′-deoxycytidine

Virtually all culture media lack vitamin C. We therefore examined whether daily supplementation of vitamin C at physiological levels in five cancer cell lines would alter the outcome of DNMTi treatment. Previous studies showed that transient (24-h) exposure of proliferating cells to 5-aza-CdR at 20–300 nM leads to long-term growth inhibition that mimics the prolonged patient responses seen in clinical settings (23–25). We tested various concentrations of 5-aza-CdR and vitamin C alone and in combination (SI Appendix, Fig. S1A). With HCT116 colorectal carcinoma cells, 5-aza-CdR alone showed the expected dose-dependent toxicity between 75 and 600 nM, with about 65% inhibition of cell growth at the highest concentration. Vitamin C alone at concentrations up to 57 μM had little effect on cell growth but was toxic at 228 μM (SI Appendix, Fig. S1B), in line with recent studies of high vitamin C concentrations (125–2,000 μM) (26, 27). In our combination approach, vitamin C increased the effects of low doses of 5-aza-CdR, with 57 μM vitamin C almost doubling the growth inhibition. Using the Chou–Talalay method (28), we found that the two compounds indeed acted synergistically, rather than additively, to inhibit cancer cell growth over the physiological ranges of vitamin C in healthy individuals (26–84 μM) (SI Appendix, Fig. S1C).

We selected four other cancer cell lines: HL60 (acute myeloid leukemia), MCF7 (breast carcinoma), and SNU398 and HepG2 (hepatocellular carcinomas) for further analysis. HCT116 and HL60 cells were treated with 300 nM 5-aza-CdR for 24 h, and MCF7, SNU398, and HepG2 cells were treated with three consecutive daily doses of 5-aza-CdR at 100 nM for 72 h to compensate for the fact that incorporation of the drug is dependent on cell-doubling time (5, 24). Prolonged increases in population-doubling times occurred in all cell lines, with peaks at 8–15 d after 5-aza-CdR treatment (Fig. 1A). The daily addition of vitamin C at 57 μM further lengthened the population-doubling times (Fig. 1A). The increased doubling time was correlated with enhanced apoptosis, as determined by annexin V measurements using flow cytometry (Fig. 1B). We observed a 1.5- to 2.2-fold increase in apoptosis with combination treatment relative to 5-aza-CdR treatment alone; daily vitamin C treatment alone increased apoptosis only slightly (Fig. 1B). Further analysis confirmed a synergy between vitamin C and 5-aza-CdR in inducing apoptosis in all these cell lines (SI Appendix, Fig. S1D).

Both 5-aza-CdR and vitamin C at high doses (in the micromolar and millimolar range, respectively) cause cytotoxicity by damaging DNA, as evidenced by a robust increase in phosphorylation of the histone variant H2AX at serine 139 (γ-H2AX) and an increase in p53 (27, 29). However, daily vitamin C alone at 57 μM did not increase the γ-H2AX or p53 levels in HCT116 cells (SI Appendix, Fig. S2), and 5-aza-CdR treatment increased them only slightly. Moreover, the combination treatment did not increase these levels further over 5-aza-CdR alone (SI Appendix, Fig. S2), suggesting that the apoptosis induced by combination treatment was not caused by DNA damage.

We next used gene-expression microarrays to screen for early changes in gene expression at day 5 after treatment, before the peak of apoptosis. Although few genes were down-regulated (SI Appendix, Fig. S3), genes responsive to IFN-α and -γ were consistently up-regulated by combination treatment (relative to 5-aza-CdR alone) in HCT116, HL60, and SNU398 cells (Fig. 2A and SI Appendix, Fig. S3). These up-regulated genes are involved in dsRNA recognition in the cytoplasm, participating in viral recognition and defense, IFN signaling, apoptotic signaling, and antigen presentation (Fig. 2A). We and others have shown that these viral defense-related genes can be up-regulated by 5-aza-CdR treatment and are responsible for inducing apoptosis in cancer cells (12, 13). Similarly, up-regulation of these genes by combination treatment may cause the increased apoptosis seen with combination treatment compared with 5-aza-CdR treatment alone.

We explored this possibility further by quantitation with RNA-sequencing (RNA-seq) in HCT116 cells. 5-Aza-CdR treatment alone up-regulated the viral-defense genes four- to 64-fold in HCT116 cells (Fig. 2B and SI Appendix, Fig. S4), as previously reported (12, 13, 24, 30). Daily vitamin C treatment alone increased the expression of these genes two- to fourfold relative to untreated cells (Fig. 2B and SI Appendix, Fig. S4), whereas the transcription of most housekeeping genes remained stable (SI Appendix, Fig. S5A). All the viral-defense genes were up-regulated further (up to 128-fold) by the combination treatment (Fig. 2B and SI Appendix, Fig. S4), but the expression of housekeeping genes was comparable to that under 5-aza-CdR treatment alone (SI Appendix, Fig. S5A). However, cytosolic sensors for DNA and Toll-like receptors that recognize viral DNA/RNA in endosomes were not up-regulated by any treatment (SI Appendix, Fig. S5B). Vitamin C and 5-aza-CdR therefore synergistically up-regulate the cytosolic viral RNA defense pathways, strengthening the viral mimicry state produced by DNMTi treatment alone (12, 13).

It is still unknown to what extent DNMTis up-regulate ERV transcripts genome-wide, which is key for triggering a viral-defense response. We analyzed the transcripts mapping uniquely to ERV loci (which are not covered by the expression array) in two replicates of the RNA-seq data for HCT116 cells after vitamin C, 5-aza-CdR, or combination treatment. The percentage of ERV transcripts relative to total reads did not change after vitamin C treatment but increased in cells treated with 5-aza-CdR and increased further with combination treatment (SI Appendix, Fig. S6A). The total reads for the ERV-1, ERVL, and ERVL-MaLR (mammalian apparent LTR retroposon) subfamilies were increased by both 5-aza-CdR and combination treatment, whereas the ERVK family was down-regulated relative to PBS controls (SI Appendix, Fig. S6B). Expression of other retrotransposons, such as the long interspersed nuclear elements (LINEs) and short interspersed nuclear elements (SINEs), was also increased by 5-aza-CdR and combination treatment; the expression levels of low-complexity repeats, simple repeats, tRNA, and small cytoplasmic RNAs remained stable (SI Appendix, Fig. S6A).

We then asked whether the up-regulated ERVs whose LTRs possess bidirectional promoter activities (31, 32) could form dsRNA secondary structures. Using directional RNA sequencing methods, we detected some unidirectionally transcribed ERVs (e.g., HERV-Fc1, HERV-Fc2, and MLT1E1A) as well as some transcribed bidirectionally (e.g., LTR5, LTR26, PRIMA41, and LTR12C) (Fig. 3A and SI Appendix, Fig. S6C). To test whether bidirectionally transcribed ERVs physically formed dsRNA secondary structures, we used the J2 antibody (33) specifically to pull down transcripts containing dsRNAs (Fig. 3 B–D). Remarkably, the combination treatment greatly increased the expression of ERVs as dsRNAs relative to 5-aza-CdR or vitamin C treatment alone (Fig. 3 B and C). In addition, the 50 most abundant bidirectionally transcribed ERVs largely overlapped with the ERVs that formed dsRNAs (Fig. 3D). LTR12C, which has more than 2,500 copies present in the human genome, was the most abundant dsRNA up-regulated by combination treatment in HCT116 cells (Fig. 3 A and C). Finally, transfection of polyinosinic:polycytidylic acid [poly(I:C)], which mimics the presence of dsRNA, up-regulated viral-defense genes in HCT116 cells to a similar extent as the combination treatment (SI Appendix, Fig. S7). Thus ERVs up-regulated as dsRNA by the combination treatment may be responsible for triggering a strong IFN response and for inducing apoptosis in cancer cells.

Because vitamin C is a cofactor for TET proteins (18, 22), we performed dot blot analysis with a 5hmC antibody to test whether vitamin C or combination treatment altered the global levels of 5hmC. Vitamin C treatment increased 5hmC in all five cell lines, with the biggest response seen in MCF7 cells (Fig. 4A). Combination treatment further increased 5hmC by about twofold relative to vitamin C treatment in all but MCF7 cells (Fig. 4A), in which combination treatment resulted in 5hmC levels comparable to those of vitamin C treatment alone (Fig. 4A). This result might be explained by the very high expression of TET2 and TET3 in the MCF7 cell line (SI Appendix, Fig. S8). In general, in most cell lines the expression of the three TET genes was unaffected by vitamin C, 5-aza-CdR, or the combination treatment (SI Appendix, Figs. S5C and S8). Therefore increased 5hmC by vitamin C and combination treatments was most likely caused by increased TET enzyme activity.

Sodium bisulfite treatment is considered the gold standard for measuring DNA methylation, but it cannot distinguish between 5hmC and 5mC and therefore actually measures the sum of both modified bases. We examined the “modification” status (i.e., 5mC+5hmC) of DNA extracted before and after treatment on the Infinium HM450 DNA methylation BeadChip platform that is based on sodium bisulfite treatment. However, at 5 d after treatment we did not observe pronounced cytosine modification changes induced by vitamin C or by combination treatment as compared with 5-aza-CdR treatment (SI Appendix, Fig. S9A).

Because the HM450 BeadChip platform contains few repetitive elements such as ERVs, this result does not preclude localized decreases in cytosine modifications elsewhere in the genome. We next assessed dynamic 5hmC changes at ERV genomic loci, using biotin-based enrichment of 5hmC followed by sequencing on days 1 and 5 after treatment. The levels of 5hmC in untreated and 5-aza-CdR–treated cells were too low to allow successful sequencing. Notably, ERVs up-regulated by combination treatment gained high 5hmC levels on day 1 which were reduced by twofold on day 5 after the treatment (Fig. 4B and SI Appendix, Fig. S10). Vitamin C treatment also transiently increased the presence of 5hmC on day 1, but the levels were lower than seen with combination treatment (Fig. 4B and SI Appendix, Fig. S10). Quantitative PCR showed that the highest 5hmC levels appeared at the LTRs of HERV-Fc1 and HERVW1 on day 1 and gradually decreased on days 3 and 5 after combination treatment (Fig. 4C), whereas the expression of ERVs increased from day 1 to day 5 after the treatment (Fig. 4D). Reduction of 5hmC on day 5 may be caused by the loss of 5mC substrates after inhibition of DNMTs. These data also confirmed the global analysis shown in Fig. 4B. In contrast, 5hmC levels were increased only slightly by vitamin C treatment but did not change with 5-aza-CdR treatment (Fig. 4C). We next examined cytosine modification patterns in the HERV-Fc1 5′ LTR by bisulfite sequencing on day 5 after treatment (SI Appendix, Fig. S9B). No changes were induced by vitamin C or by combination treatment as compared with cells treated with 5-aza-CdR (SI Appendix, Fig. S9B). Further loss of cytosine modifications by combination treatment was not seen, suggesting that the presence of 5hmC at the ERV LTRs may directly facilitate the transient up-regulation of these ERVs.

It is worth noting that the effects of vitamin C on 5hmC production and ERV up-regulation could also be achieved by treatment with dehydroascorbic acid (DHA) (Fig. 5), the oxidized form of vitamin C. Therefore, as previously reported (26), it is possible that vitamin C was oxidized to DHA before it was transported into the cells. On the other hand, other antioxidants such as glutathione (GSH) or DTT were unable to elicit the same effects (Fig. 5).

Because TETs are the only known enzymes that oxidize 5mC to 5hmC, we reasoned that the effects of vitamin C may be mediated by TET proteins. To test this hypothesis, we used a TET2-KO cell line established by CRISPR/Cas9 technology in A2780 human ovarian carcinoma cells (SI Appendix, Fig. S11). The global levels of 5hmC were about four- to eightfold lower in TET2-KO cells than in the wild-type A2780 cells after vitamin C or combination treatment (Fig. 6A), indicating that knocking out TET2 substantially decreases 5hmC levels in A2780 cells. Furthermore, we observed a significant reduction of HERV-Fc1 and LTR12C expression in TET2-KO cells compared with wild-type A2780 cells after combination treatment (Fig. 6B). Therefore the effects of vitamin C on ERV expression in A2780 cells are mediated in part by TET2 protein. Other TET proteins also might compensate for the loss of TET2, because an up-regulation of 5hmC and ERV expression was observed after combination treatment in the KO cells, although the effect was less than that seen in the wild-type A2780 cells (Fig. 6B). Interestingly, knocking out TET2 also reduces the expression of LTR12C after 5-aza-CdR treatment (Fig. 6B). It is possible that TET2 may still have weak activity in wild-type A2780 cells cultured in McCoy’s 5A medium with a low vitamin C concentration (2.9 μM) and may partly contribute to the effects of 5-aza-CdR alone.

These results show that targeting the cancer DNA methylome by combining low-dose 5-aza-CdR and vitamin C stimulates the expression of ERVs, the induction of a cell-autonomous immune activation response, and increased apoptosis of cancer cells. The lack of vitamin C in a patient’s plasma might cause a poor response to DNMTi treatment in some patients. We therefore measured plasma concentrations of vitamin C in a small number of patients with miscellaneous hematologic malignancies. Strikingly, 58% of patients with hematological neoplasia who were not taking vitamin C supplements had severe vitamin C deficiency (serum concentration <11.4 μM, at which clinical features of scurvy may be manifested) (34), and 33% had vitamin C levels below the normal range (Fig. 7 and SI Appendix, Table S1). In this small sample, no specific pattern was observed with respect to patient age, disease state, or treatment status. In contrast, only 7.1% of the United States population was found to be deficient in vitamin C according to the 2003–2004 National Health and Nutrition Examination Survey (NHANES) (34). Vitamin C deficiency has been seen previously in patients with multiple types of cancer, including hematological malignancies (35–37). We predict that these patients might receive the most benefits from the combination treatment.

Twenty-four patients diagnosed with hematological malignancies from Department of Hematology, Rigshospitalet, Copenhagen University Hospital, Denmark, were randomly selected for measurement of plasma vitamin C levels (details are listed in SI Appendix, Table S1). The study was approved by the Regional Ethical Committee, Capital Region of Denmark, and all participants provided written informed consent before enrollment. Detailed description of the materials and methods used in this study are provided in SI Appendix, SI Materials and Methods.

HCT116 (ATCC no. CCL-247), HL60 (ATCC no. CCL-240), MCF7 (ATCC no. HTB-22), SNU398 (ATCC no. CRL-2233), and HepG2 (ATCC no. HB-8065) cells were obtained from the American Type Culture Collection (ATCC; https://www.atcc.org/). A2780 and TET2-KO cells were obtained from the S.B.B. laboratory. Mutations contained in these cell lines are listed as Dataset S1. Information about cell culture, cell viability assays, and evaluation of combination effects are described in SI Appendix, SI Materials and Methods.

HCT116 and HL60 cells were untreated or treated with a daily dose of 57 μM vitamin C, with 300 nM 5-aza-CdR for 24 h, or with the combination of daily vitamin C at 57 μM and transient exposure to 5-aza-CdR for 24 h. For MCF7, SNU398, and HepG2 cells, 5-aza-CdR treatment was performed with three daily consecutive doses at 100 nM (72-h exposure). Cell-doubling time was calculated as previously reported (13) and is described in SI Appendix, SI Materials and Methods.

Cellular apoptosis was measured by annexin V and propidium iodide (PI) staining using the annexin V-FITC apoptosis detection kit (MBL) according to the manufacturer’s protocol. After staining, the cells were analyzed by a MoFlo Astrios Cell Sorter (Beckman Coulter, Inc.). Annexin V-FITC staining indicated apoptotic cells.

Gene-expression analysis was performed at Sanford–Burnham Medical Institution, La Jolla, CA using the Illumina genome-wide expression BeadChip (HumanHT-12_V4.0_R1) (Illumina). Gene-expression data were normalized using the quantile normalization method with the lumi package, and differential expression analyses were performed using the limma package in R.

For total RNA-seq with rRNA reduction and dsRNA sequencing, libraries were sequenced as single-end 75 bases on a NextSeq 500 instrument (Illumina) at the Van Andel Research Institute Genomics Core. For directional RNA-seq with rRNA reduction and strand specificity retained, libraries were prepared and sequenced as paired-end 50 bases on a HiSeq 2500 instrument (Illumina) at HudsonAlpha Institute for Biotechnology Genomic Services Laboratory. Details of J2 antibody pull-down, RNA-seq library preparation, and data processing are described in SI Appendix, SI Materials and Methods.

Dot blot analysis of 5hmC was performed as described previously (18). Chemical capture of 5hmC was performed using the Hydroxymethyl Collector Kit (Active Motif, 55013) according to the manufacturer’s protocol with minor modifications described in SI Appendix, SI Materials and Methods.