Cytidine Deaminase Deficiency Reveals New Therapeutic Opportunities against Cancer

This article is featured in Highlights of This Issue, p. 1877

Despite major advances in the development of chemotherapy, many cancers continue to have a poor prognosis, due to the resistance of cancer cells to antineoplastic drugs through intrinsic or acquired mechanisms (1). It is, thus, highly important to identify markers predicting the response to anticancer treatment and new molecular targets for novel anticancer treatments.

Two families of cytidine deaminases exhibit different biological functions, deaminating either the free cytidines, as performed by cytidine deaminase (CDA or CDD; EC3.5.4.5), or deaminating the cytidines incorporated within the DNA or RNA polymers, as performed by the AID/APOBECs (activation-induced deaminase/apolipoprotein B mRNA editing catalytic polypeptide like) proteins (2, 3). We previously reported that CDA deficiency leads to DNA damage and to genetic instability (4, 5), known to be associated with cancer development, suggesting a possible relationship between CDA underexpression and cancer.

CDA is an enzyme of the pyrimidine salvage pathway catalyzing the hydrolytic deamination of cytidine and deoxycytidine to uridine and deoxyuridine, respectively (6). In addition to native nucleotides, CDA also deaminates and inactivates nucleoside analogues, such as gemcitabine and cytosine arabinoside (Ara-C), agents widely used to treat cancer (7). CDA thus plays an important role in the sensitivity/resistance of cancer cells to treatment with cytidine analogues (8, 9). Indeed, early severe toxicity has been reported in cancer patients, with low levels of CDA activity treated with gemcitabine (10, 11). CDA overexpression might thus be a marker for resistance to chemotherapy based on cytidine analogues.

CDA gene polymorphism has been widely explored (11–17), but little is known about its transcriptional and posttranscriptional regulation. One study recently reported that miR-484 downregulates CDA gene expression by targeting its 3′-UTR, sensitizing breast cancer cells overexpressing CDA to gemcitabine (18). CDA overexpression is thus a good marker for resistance to chemotherapy based on cytidine analogues.

CDA overexpression has recently been identified as a potential target for anticancer treatment. Indeed, CDA has been shown to be involved in the deamination of oxidized and epigenetically modified cytidine nucleosides (19, 20). These cytidine analogues are selectively converted to their uracil counterparts and then incorporated into DNA in cancer cells overexpressing CDA, resulting in DNA damage and cell death (19, 20).

In this study, we focused on CDA deficiency rather than CDA overexpression. We found that CDA was downregulated in about 60% of cancer cells and tissues. The CDA gene was mostly inactivated by DNA methylation. We identified a new subgroup of cancers not expressing CDA that were susceptible to the specific toxic effects of drugs such as aminoflavone. Thus, CDA deficiency in cancer cells is a new biomarker of cellular response to anticancer drugs. Our findings open up promising new avenues for the treatment of cancer.

We used 33 cancer cell lines in this study (Supplementary Material S1): 19 breast cancer cell lines from the Translational Research Department of the Curie Institute (Paris, France; ZR75-1, T47D, HCC-1428, BT-474, MCF-7, MDA-MB-361, MDA-MB-468, MDA-MB-231, MDA-MB-436, HCC-38, HCC-70, HCC-1187, HCC-1937, HCC-1143, BT-20, BT-549, HCC-1954, SKBR-3, and HS578T) and two nonmalignant breast cell lines (MCF-12A and 184B5), four lung cancer cell lines (H522, H23, HOP-92, and HOP-62), three ovarian cancer cell lines (IGROV-1, SKOV-3, and OVCAR-8) from the NCI (Rockville, MD; refs. 21, 22), one melanoma cell line (A2058) from Dr. Stephan Vagner's laboratory (UMR3348 CNRS, Curie Institute), and two cervical cancer cell lines (HeLa-Ctrl and HeLa-shCDA) and two Bloom syndrome cell lines (BS-Ctrl and BS-CDA), described previously (5). All the breast cell lines were authenticated at the Translational Research Department of the Curie Institute using standard DNA microsatellite short tandem repeat (STR) method. Lung and ovarian cancer cell lines were authenticated at the NCI using standard DNA microsatellite STR method (21, 22). No specific authentication of the A2058 melanoma cell line was performed. The isogenic HeLa-Ctrl/HeLa-shCDA and BS/BS-CDA cell lines were established in our laboratory and are checked routinely for several phenotypic characteristics, such as sister chromatid exchange and ultrafine anaphase bridge frequencies (4, 5).

All cells were routinely checked for the absence of mycoplasma and were maintained in the recommended media (see Supplementary Material S1) before the extraction of DNA, RNA, and proteins.

For evaluation of the induction of CDA expression, RNA was isolated from cell lines continuously treated with 1 or 2.5 μmol/L of 5-Aza-2′-deoxycytidine (5-Aza-dC; Sigma Aldrich) for 96 hours.

Cell viability was carried out with 3-(4,5-dimethyl-2-thiazolyl)-2,5 diphenyl-2H-tetrazolium bromide (MTT; Life Technologies) in 96-well microplates. We assessed the functionality of CDA by plating HCC-1954 and IGROV-1 cells at densities of 2 × 10³ and 3 × 10³ cells per well, respectively, on the day before pretreatment and at a density of 8 × 10² cells per well for control conditions. Cells were then left untreated or subjected to pretreatment with 1 μmol/L 5-Aza-dC for 96 hours. The cells were washed twice with PBS buffer, placed in fresh medium, and incubated for 72 hours in the presence of various concentrations of gemcitabine, from 0.001 to 1 μmol/L (Sigma Aldrich). The data were normalized to corresponding controls for each condition. We evaluated aminoflavone cytotoxicity after 72 hours of treatment, by plating MCF-7, MDA-MB-468, MDA-MB-231, SKOV-3, and OVCAR-8 cells at a density of 3 × 10³ cells per well and IGROV-1 cells at a density of 4 × 10³ cells per well. Aminoflavone (NSC 686288) was provided by Dr. Yves Pommier (Developmental Therapeutics Branch, NCI).

The 950 base pairs downstream from the translation initiation codon in the CDA promoter region and the four exons were amplified by PCR, with the Phusion Polymerase enzyme (Promega). The reaction was performed with 50 ng of genomic DNA isolated from 12 breast cancer and two normal-like cell lines. The specific primers used for amplification and nucleotide sequencing to base-pair resolution (Eurofins Genomics) are presented in Supplementary Material S2.

The procedure for real-time PCR (qRT-PCR) was as described by Gemble and colleagues (5). In brief, total RNA was extracted from patient-derived xenograft (PDX) tissues and from cell lines with the RNeasy Mini Kit (Qiagen). Reverse transcription was performed on 1 μg of RNA with the GoScript enzyme (Promega). The cDNA obtained was used at a dilution of 1/10 for real-time PCR with the SYBR Green Supermix Reagent (Bio-Rad) in a Bio-Rad CFX96 machine. Each sample was run in triplicate. Relative expression was determined by the 2^−ΔΔC_t method. GAPDH and TBP were used as internal controls. The specific primers used for qRT-PCR analysis are presented in Supplementary Material S2.

For Western blotting, cells were harvested by centrifugation and lysed in 8 mol/L urea, 50 mmol/L Tris-HCl, pH 7.5, and 150 mmol/L β-mercaptoethanol buffer supplemented with protease inhibitor (Thermo Scientific). They were then sonicated and heated. Protein concentration was estimated with the BCA Kit (Pierce), and the equivalent of 20 μg or protein per cell lysate was run on a 4% to 12% Bis-Tris precast gel (Life Technologies). The proteins were then transferred to PVDF membranes, which were probed with the appropriate antibody. Protein bands were visualized with a CCD camera (Bio-Rad). Details of the primary and secondary antibodies used are provided in Supplementary Material S2.

IHC was carried out as described by Baldeyron and colleagues (23). Briefly, paraffin-embedded tissue blocks obtained at the initial diagnosis were retrieved from the archives of the Biopathology Department of Curie Institute Hospital (Paris, France). Sections (3-μm thick) were cut with a microtome from the paraffin-embedded tissue blocks. Tissue sections were dewaxed and rehydrated through a series of xylene and ethanol washes. A primary anti-CDA Ab was used (Supplementary Material S2). The sections were processed with a Dako machine for immunostaining. The specificity of the CDA Ab was confirmed by applying the same protocol to paraffin-embedded human tissue sections and cell block sections. The sections were rehydrated by incubation in PBS for 5 minutes and then incubated with anti-CDA antibody for 1 hour. Antibody binding was detected by incubation with a secondary antibody coupled to a peroxidase-conjugated polymer (Dako Envision+) after treatment with DAB solution (Dako K3468) for 5 minutes and Mayer hematoxylin for 1 minute. The sections were then mounted in resin. We evaluated CDA immunostaining on histologic sections from 19 normal human tissues (20 samples per tissue) and from 6 primary tumor tissues (50 samples per cancer type). For each section, we evaluated two immunohistologic scores: intensity score: score 0, no staining; score 1+, weak staining; score 2+, moderate staining; score 3+, intense staining; and frequency score: score 0, no staining; score 1+, 1% to 33% stained cells; score 2+, 34% to 67% stained cells; score 3+, 68% to 100% stained cells. Then, we defined a final score (H-score = frequency score × intensity score). This H-score was equal to 1 in normal colon tissue, and 1.5 in lung, breast, melanoma, ovary, and endometrium normal tissues. It means that the expression of CDA in normal tissues is between ≥1 and <2.

Thus, the cut-off value of CDA expression in tumor tissues was defined as CDA underexpression by H-score between 0 and 1 (CDA low), and CDA overexpression by H-score between 2 and 3 (CDA high). Thus, the data are presented as a combination of the percentage of CDA-positive cells and intensity scores. The analysis was carried out by two independent pathologists.

The PDX models used here were established as described by Marangoni and colleagues (24). Briefly, breast cancer fragments were obtained from patients at the time of surgery, with prior written informed consent of the patients. Fragments (30–60 mm³) were grafted subcutaneously into the interscapular fat pad of 8- to 12-week-old female Swiss nude mice, under Avertin anesthesia. Mice were maintained in specific pathogen-free animal housing (Curie Institute) and received estrogen (17 mg/mL) in their drinking water. Xenografts appeared at the graft site 2 to 8 months after grafting. They were subsequently transplanted from mouse to mouse and stored frozen in DMSO–FCS solution or dry frozen in liquid nitrogen for RNA isolation. The experimental protocol was performed in accordance with French regulations.

This assay was performed as described by Gemble and colleagues (5). In brief, cells were plated on glass slides in the presence of 10 μmol/L 5-bromodeoxyuridine (Sigma Aldrich). After two divisions, colchicine (Sigma Aldrich) was added (0.1 μg/mL), and the cells were incubated for 1 hour. Cells were then incubated in a hypotonic solution [1:5 (v/v) FCS distilled water] and fixed with a 3:1 (v/v) mixture of methanol and acetic acid. They were then stained by incubation with 10 μg/mL Hoechst 33258 (Sigma Aldrich) in distilled water for 20 minutes. The slides were rinsed with 2× SSC (Euromedex) and exposed to UV light at a wavelength of 365 nm and a distance of 10 cm for 105 minutes. The slides were then rinsed in water, stained with 2% Giemsa (VWR) for 16 minutes, rinsed in water, dried, and mounted in EUKITT (Sigma Aldrich). Metaphases were captured and chromosomes were visualized under a Leica DMRB microscope at a magnification of ×100. We determined the number of sister chromatid exchanges (SCE) per chromosome.

We analyzed 482,422 CpGs in the NCI-60 cell lines with Illumina Infinium Human Methylation 450 BeadChips. The DNA methylation datasets are available under accession number GSE66872. The methylation values are presented from 0 to 1. The data were normalized and analyzed as described by Nagales and colleagues (25).

The negative correlations between CDA promoter methylation and CDA expression on The Cancer Genome Atlas (TCGA) samples (26) were generated through the Broad Institute FireBrowse portal (27, 28) and the cBioPortal for Cancer Genomics database (29–31), all the cBioPortal data (expression, mutation, copy number, significance analyses) being loaded directly from FireBrowse. The only promoter CpG site presenting a high significant negative correlation with CDA expression in both NCI-60 cell lines and TCGA samples was selected.

A collection of 40 human breast tumor cell lines (mostly from ATCC) was established in the Translational Research Department of the Curie Institute. Gene expression profiles were generated with the Affymetrix Exon Array and GenoSplice algorithms to summarize multiprobe measurements as single mRNA levels.

CDA expression levels were extracted from various transcriptomic datasets: breast tumor cell lines of the Curie Institute collection (32), NCI-60 (CellMiner tools; ref. 33), Cancer Cell Line Encyclopedia (CCLE; ref. 34), Gene Expression Across Normal and Tumor Tissue database (35, 36), the TCGA portal, and the Gene Expression Omnibus database (GEO; ref. 37). All these data are publicly accessible.

All data analyses and processing were performed with GraphPad Prism 6 software. Pearson correlation analysis was used to assess the association between two variables. P values for sister SCEs were calculated by Mann–Whitney tests. CDA mRNA levels in normal and cancerous tissues were compared in two-tailed unpaired t tests. Differences in the induction of CDA expression by 5-Aza-dC, as assessed by qRT-PCR, were evaluated in two-tailed paired t tests. Survival curves were compared in paired t tests for HeLa-shCDA versus HeLa-Ctrl cells treated with aminoflavone and HCC-1954 and IGROV-1 cells with and without 5-Aza-dC pretreatment. Unpaired t tests were used for the other cell lines. Differences were considered statistically significant if P < 0.05.

We first analyzed CDA expression in cancer cell lines by combining in silico analyses on cancer cell lines from the Curie Institute, the NCI, and the CCLE (Broad-Novartis Cancer Cell Line Encyclopedia), and experimental approaches using qRT-PCR and Western blotting on a set of 26 representative cancer cell lines from the Curie Institute and the NCI. We found that CDA was expressed weakly or not at all in 25 of 34 (73%) breast cancer cell lines from the Curie Institute and 44 of 60 (73%) cancer cell lines derived from nine different organs and tissues from the NCI (Fig. 1A, left and right). Similarly, about 60% (700) of the 1,036 cancer cell lines from 24 different cancer tissues from the CCLE database did not express CDA (Supplementary Fig. S1A). We validated these results, by qRT-PCR and Western blotting on a set of 26 representative cancer cell lines from the Curie Institute and the NCI cancer cell line collections (19 breast cancer cell lines, 4 lung cancer cell lines, and 3 ovarian cancer cell lines; Fig. 1B).

We investigated whether the absence of detectable CDA expression observed in the majority of cancer cell lines also applied to primary tumor tissues, by performing qRT-PCR to analyze CDA mRNA levels in human primary breast tumors xenografted into nude mice (PDXs). As control, we analyzed Cda mouse analogue in parallel, to ensure that mouse stroma was not interfering with human tumor data. This approach made it possible to avoid the contamination of primary tumor tissues with normal cells from the stroma (usually up to 30%). We found that 56 of the 66 (∼84%) human primary breast tumors studied had no significant CDA expression and did not find any correlation between the levels of mouse Cda expression and human CDA expression, indicating that mouse stroma did not contribute to the xenograft data (Fig. 1C). We also analyzed CDA protein levels in six types of primary cancer tissues (50 per type) and in 19 normal human tissues (20 per tissue) by IHC with an anti-CDA antibody validated by IHC on isogenic Bloom syndrome–derived cells not expressing CDA (BS-Ctrl) or expressing exogenous CDA (BS-CDA; see the Materials and Methods section and Supplementary Fig. S1B; ref. 5). According to CDA staining intensity in normal tissues (Supplementary Fig. S1C, top), we stratified CDA expression into two groups on the basis of staining intensity scores: CDA low (scores of 0 and 1) and CDA high (scores of 2 and 3; Fig. 1D). About 50% (endometrium) to 88% (breast triple-negative, ovary, and colon) of cancer tissues displayed very low levels of CDA expression.

We then compared CDA mRNA levels between healthy and cancerous tissues of different origins, by replotting the CDA mRNA data downloaded from GEO found in different genomic data sources (Nextbio, Oncomine). Tumor tissues are often contaminated with normal tissues that might express CDA, leading to inappropriate interpretations of CDA expression in some tumor tissues. Nevertheless, CDA expression levels were significantly lower in several tumors than in healthy tissues (Fig. 1E and Supplementary Fig. S1D and S1E). These results (summarized in Supplementary Table S1) reveal that CDA is overexpressed in some tumor tissues, such as those of pancreas, stomach, thyroid, and bladder cancers, as previously reported (19, 35), but underexpressed in other tumor tissues, such as those of liver, cervix, colon, and esophagus cancers. We confirmed these results by qRT-PCR on a small in-house cohort of colon tissues. We found that CDA expression levels were significantly lower (P = 0.0128) in tumor tissues (n = 10) than in healthy tissues (n = 5; Fig. 1F).

Finally, analysis of a recently published gene expression dataset used to determine the molecular mechanism of cervical cancer progression (38) revealed that CDA expression decreased considerably with cervical cancer progression (Supplementary Fig. S1F). This result is consistent with the data presented in Fig. 1E, showing lower levels of CDA expression in cervical cancer tissues than in noncancerous tissues. Overall, these results suggest that CDA expression tends to be lost during carcinogenesis, at least in some tissues, such as the cervix.

We investigated the mechanism underlying the downregulation of the CDA gene in tumor cells by first analyzing CDA copy number in the DNA of the CCLE and NCI60 cell lines. No significant correlation was found between CDA mRNA levels and CDA gene copy number (Supplementary Fig. S2A). The downregulation of CDA levels cannot therefore be attributed to genetic deletions in tumor cells.

We then carried out sequencing analysis to determine whether CDA (promoter and exons) was mutated in 11 breast cancer cell lines that did not express CDA, through comparison with two breast cancer cell lines expressing high levels of CDA (HCC-1143 and MDA-MB-231, see Fig. 1B) and breast cell lines derived from healthy tissues with strong or weak CDA expression (MCF-12A and 184B5, respectively; Supplementary Fig. S2B). No genetic mutation likely to lead to CDA inactivation was identified (Supplementary Table S2A). However, we found several SNPs that had previously been identified and listed in the Single Nucleotide Polymorphism Database (11, 15, 39, 40). These results are consistent with the CDA gene sequencing results for the NCI-60 cell lines (Supplementary Table S2B) and demonstrate that CDA is not inactivated through genetic alterations in cancer cells.

We then explored the possible role of epigenetic regulation of CDA gene expression. We mapped the CpG methylation sites in the CDA gene. Using the NCI-60 methylation datasets, we mapped the CpG methylation sites in the CDA gene and identified 8 CpG methylation sites, three in the body of the gene, two within 1,500 bp of the transcription start site (TSS), one in TSS200, one in 5′-UTR, and one is 3′-UTR (Supplementary Fig. S2C). We then analyzed the levels of methylation of the CDA promoter using the dataset for the methylation of NCI60 cell lines (25). We calculated Pearson correlation coefficients for the relationships between methylation at the various CpG methylation sites and CDA expression. We found a highly significant negative correlation between CDA transcript and methylation levels at CpG two sites, cg04087271 (TSS200) and cg00784581 (5′-UTR; Pearson r = −0.4184, P = 0.009; Fig. 2A), the six other exhibiting no significant negative correlation with CDA expression. High levels of CDA methylation were found in 42% of the CDA-deficient cell lines (19/45), such as MCF-7 and IGROV-1, and no methylation was detected in cell lines overexpressing CDA, such as MDA-MB-231 and HOP-92, except for the LOXIMVI melanoma cell line.

For the validation of these methylation data, we treated a set of cancer cell lines derived from breast, lung, ovarian, and melanoma tumors not expressing CDA (Fig. 1B and Supplementary Fig. S2D) with the DNA methyltransferase activity inhibitor (5-Aza-dC), resulting in DNA demethylation (42). We found that 5-Aza-dC induced a strong increase (up to 1,000-fold induction) in CDA mRNA levels (Fig. 2B) without major toxicity (Supplementary Fig. S2E) in the 7 CDA-deficient cell lines analyzed. By contrast, it had little or no effect on CDA transcript levels in the MDA-MB-231, HOP-92, HCC-1143, and HCC-1937 control cell lines, which have constitutively high levels of CDA (∼2-fold induction).

It has been reported that the selection of CDA overexpression in response to prolonged drug exposure is responsible for resistance to gemcitabine (18, 43) and that the ectopic expression of CDA in CDA-deficient cancer cells leads to a significant increase in resistance to gemcitabine (18, 44). We thus evaluated the functionality of the CDA protein produced after 5-Aza-dC treatment by breast and ovarian cancer cells, HCC-1954 and IGROV-1, respectively. The cells were left untreated or were subjected to pretreatment with 5-Aza-dC for 96 hours and then to treatment with various concentrations of gemcitabine over a period of 72 hours. The induction of CDA protein production by 5-Aza-dC led to a significant increase in gemcitabine resistance (Fig. 2C). Our data are consistent with 5-Aza-dC inducing the expression of functional CDA protein in CDA-deficient cancer cells, leading to resistance to gemcitabine.

We then analyzed in silico CDA promoter methylation levels (Supplementary Fig. S2C) on TCGA samples for 16 different tissue cancers, using the CBioPortal for cancer genomics (29, 30) and the Broad Institute FireBrowse portal (27). We found a highly significant correlation between CDA transcripts levels and CDA promoter methylation on two CpG sites (cg04087271 and cg24502330). Methylation of the cg04087271 site was the only one correlating with CDA deficiency in both tumor tissues and NCI-60 cell lines (Fig. 2D). In both cancer cell lines and tumor tissues, CDA promoter methylation levels were significantly higher in samples with low CDA transcript levels (Fig. 2D) than in those with high CDA transcript levels. This correlation was not significant in some other cancer types, but we nevertheless identified a subpopulation of these cancers with low CDA expression and high CDA promoter methylation levels (Supplementary Fig. S2F). Thus, the association of methylation status to gene expression is consistent with the hypothesis that promoter methylation is a principal driver of CDA gene expression in cancer.

We previously reported that CDA deficiency in BS cells or CDA depletion in HeLa cells leads to an increase in SCE frequency, a hallmark of genomic instability (4, 5).

We therefore investigated whether constitutive CDA deficiency in tumor cells was also associated with an increase in SCE frequency by analyzing basal SCE levels in several cancer cell lines derived from breast, lung, and ovary tumors. SCE frequency was significantly higher in the cancer cell lines not expressing CDA than in those expressing CDA (Fig. 3A).

Thus, tumors from the same classically defined groups may display differences in CDA expression status, resulting in contrasting cellular properties, such as SCE levels (e.g., CDA-proficient HCC-1143 cells and CDA-deficient BT-20 cells are both classified as triple-negative breast cancer cells). We thus propose the use of CDA expression status in tumor cells to define two new subgroups: CDA-deficient tumors and CDA-proficient tumors. These new subgroups may differ in their sensitivity to antitumor therapies. The targeting of CDA-deficient tumor cells might therefore open up new possibilities for cancer therapy.

The CellMiner web tool (45) can be used to assess the correlation between gene expression and drug sensitivity/resistance. We searched for drugs with antiproliferative activity significantly correlated with CDA expression levels. We identified 277 such drugs, 94 of which were more toxic to CDA-deficient cells and 183 of which were more active against CDA-proficient cells (Supplementary Table S3). We tested our hypothesis that drugs that do not affect CDA-proficient cells can specifically target CDA-deficient cells by focusing on an aminoflavone derivative (AFP464; NSC 710464), for which we found a highly significant negative correlation (Pearson r = −0.379, P = 0.0031) with CDA deficiency (Supplementary Table S3). Twenty CDA-deficient cell lines of the 43 tested (46.5%), including MCF-7 and IGROV-1, were sensitive to aminoflavone, whereas 13 of the 16 (81.25%) CDA-proficient cell lines, including MDA-MB-231, were resistant to this drug (Fig. 3B).

Among the selected molecules presenting a significant negative correlation with CDA expression, we choose aminoflavone because it has reached phase II clinical trials in the United States for treating solid tumors, including breast cancer (NCT01015521, NCT01233947, NCT00369200, NCT00348699; ref. 46), and because the antitumor efficacy of aminoflavone had been validated in vivo by three independent groups on MCF-7 and MDA-MB-468 xenograft breast cancer models (47–49). No effect of aminoflavone was observed on MDA-MB-231 xenograft breast cancer cells (49). Together with our results showing that MCF-7 and MDA-MB-468 breast cancer cell lines are CDA deficient, and MDA-MB-231 cells are CDA proficient (Fig. 1A and B), these data are consistent with our hypothesis that aminoflavone specifically targets CDA-deficient cancer cells both in vitro and in vivo.

We thus evaluated the causality of the relationship between CDA downregulation and the antiproliferative activity of aminoflavone by shRNA-mediated CDA depletion in HeLa cells (Supplementary Fig. S3). We found that CDA depletion significantly increased sensitivity to aminoflavone treatment (Fig. 3C). We then confirmed the cytotoxicity of aminoflavone in six breast and ovary cancer cell lines, three of which were CDA deficient (MCF-7, MDA-MB-468, and IGROV-1), and the other three being CDA proficient (MDA-MB-231, OVCAR-8, and SKOV-3). The CDA-deficient cell lines were highly sensitive to aminoflavone, whereas the CDA-proficient cell lines were resistant (Fig. 3D, left and right).

Our results show that CDA expression status can be used as a predictor of sensitivity to aminoflavone; CDA deficiency is thus a potential new sensitive biomarker or target for anticancer therapies.

We found that CDA expression was lost in a large proportion of cancer cells and tumor tissues, and our findings identify CDA-deficient tumors as a new subgroup of cancers. SCE frequency was significantly higher in cancer cell lines not expressing CDA than in those expressing CDA. This increase in SCE frequency might reflect a nucleotide pool imbalance due to CDA deficiency that slows down replication fork progression, leading to an increase in replication-associated DNA breaks that could account, at least in part, for SCE formation (4).

The loss of CDA expression is mostly due to DNA methylation, and the treatment of CDA-deficient cells with 5-Aza-dC was sufficient to restore the expression of a functional CDA. This is the first study, to our knowledge, to reveal the extent of CDA inactivation and its epigenetic control in cancer.

DNA methylation may be the predominant mechanism of CDA silencing, but it is clearly not the only one, as some CDA-deficient cell lines present no CDA gene methylation. The other mechanisms involved in regulating CDA gene expression merit further investigation but are beyond the scope of this study.

CDA has already been shown to play a crucial role in the response of cancer cells to widely used nucleoside analogues, such as cytosine arabinoside and gemcitabine, and the dose-limiting toxicity of these drugs (8, 50–53). Our results suggest that IHC assessments of CDA levels could be used to determine the CDA status of tumors, with potential implications for treatment.

Oxidized and epigenetically modified cytidine nucleosides specifically target tumors overexpressing CDA (19, 20). We found that 5-Aza-dC treatment strongly induced the expression of a functional CDA in CDA-deficient tumor cells, with little or no effect on CDA expression in CDA-proficient cells. These findings suggest that DNA-demethylating agents could be assessed as a possible treatment for CDA-deficient tumors to induce CDA overexpression and then sensitize these tumors to treatment with oxidized and epigenetically modified cytidine nucleosides.

Finally, our results suggest that the targeting of CDA deficiency might offer new possibilities for treatment. In silico screening with the NCI CellMiner analysis tool identified aminoflavone as a proof-of-principle candidate for the targeting of CDA-deficient tumor cells. Aminoflavone acts mainly by activating aryl hydrocarbon receptor, which in turn induces the expression of cytochrome P450 CYP1A1/1A2/1B1 that metabolizes aminoflavone to toxic products, inducing replication stress, DNA damage, and apoptosis via a p53/p21-mediated pathway (48, 54–57).

We found that aminoflavone was specifically effective in CDA-deficient tumor cells and that this drug had no effect on CDA-proficient cells, in agreement with the literature (47–49).

Thus, the subgroup of tumors not expressing CDA could be specifically targeted by such treatment, and CDA expression status could be used as a new marker to guide anticancer therapy. Molecules not yet shown to be active against this tumor subgroup will probably be discovered through the systematic screening of CDA-proficient and -deficient cells.

In conclusion, although additional human trials are needed, our results constitute a proof of concept that CDA deficiency may turn out to be a new predictive marker of susceptibility to antitumor drugs that could be used as a new target for anticancer therapies, thus opening up new possibilities for the treatment of cancers.

S. Varma is an employee of HiThru Analytics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: H. Mameri, Y. Pommier, M. Amor-Guéret

Development of methodology: H. Mameri, I. Bièche, E. Marangoni, S. Vacher, Y. Pommier, M. Amor-Guéret

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H. Mameri, I. Bièche, E. Marangoni, G. Buhagiar-Labarchède, A. Nicolas, S. Vacher, R. Onclercq-Delic, W.C. Reinhold, Y. Pommier

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H. Mameri, I. Bièche, D. Meseure, E. Marangoni, G. Buhagiar-Labarchède, S. Vacher, R. Onclercq-Delic, V. Rajapakse, S. Varma, Y. Pommier, M. Amor-Guéret

Writing, review, and/or revision of the manuscript: H. Mameri, I. Bièche, D. Meseure, A. Nicolas, Y. Pommier, M. Amor-Guéret

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): G. Buhagiar-Labarchède, R. Onclercq-Delic, Y. Pommier

Study supervision: M. Amor-Guéret

We thank D. Gentien, E. Henry, B. Albaud, F. Reyal, V. Maire, B. Marty-Provost, T. Dubois, S. Roman-Roman (Translational Research Department of the Institut Curie), F. Radvanyi (UMR144CNRS/Institut Curie), and O. Delattre (INSERM U380, Institut Curie) for providing us with the complete study of the breast cancer cell lines and the TCGA network for making available the high-throughput data. We thank S. Vagner for the melanoma cell lines and for critical reading of the manuscript. We also thank M. Dutreix for stimulating discussions.

This work was supported by the Curie Institute (Translational Cancer Research grant 2014 and PICSysBio), CNRS, Cancéropôle Ile de France, Ligue Nationale contre le Cancer (Comité de l'Essonne), the Association pour la Recherche sur le Cancer, the Agence Nationale de la Recherche, and the Center for Cancer Research (BC 006161), the Intramural Program of the NCI, NIH. H. Mameri was supported by a postdoctoral fellowship from Cancéropôle Ile de France and Association pour la Recherche sur le Cancer.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.