APOBEC3B is an enzymatic source of mutation in breast cancer

Most humans encode a total of eleven polynucleotide cytosine deaminase family members that could contribute to mutation in cancer – APOBEC1, activation-induced deaminase (AID), APOBEC2, APOBEC3s (A3A, A3B, A3C, A3D, A3F, A3G, and A3H), and APOBEC4. APOBEC2 and APOBEC4 have not shown activity. APOBEC1 and AID are expressed tissue specifically and implicated in cancers of those tissues, hepatocytes and B lymphocytes, respectively^11,12. We therefore hypothesized that one or more of the seven APOBEC3s may be responsible for C-to-T mutation in other human cancers. This possibility is consistent with hybridization¹³ and expression studies¹⁴ (Fig. S1).

To identify the contributing APOBEC3, we quantified mRNA levels for each of the 11 family members in breast cancer cell lines (Fig. S2). Surprisingly, only A3B mRNA trended toward up-regulation. This analysis was expanded to include a total of 38 independent breast cancer cell lines. A3B was up-regulated by ≥3 s.d. relative to controls in 28/38 lines, with levels exceeding 10-fold in 12/38 lines (Fig. 1a & Table S1). MDA-MB-453, MDA-MB-468, and HCC1569, representative lines used below, showed 20-, 21-, and 61-fold up-regulation, respectively. These results correlate with cell line microarray data (Fig. S3, Tables S2-S9 & Supplementary Discussion). A3B up-regulation is most likely due to an upstream signal transduction event because it is not a frequent site of rearrangement or copy number variation (http://dbCRID.biolead.org) and sequencing failed to reveal promoter activating mutations or CpG islands indicative of epigenetic regulation (Fig. S4).

Epitope-tagged A3B localizes to the nucleus of several transfected cell types¹⁵. To ask whether this is also a property of breast cancer lines, an A3B-eGFP construct was transfected into MDA-MB-453, MDA-MB-468, and HCC1569. Live cell images of A3B-eGFP showed nuclear localization, in contrast to the cytoplasmic A3F-eGFP (Fig. 1b & S5). Corroborating data were obtained for HA-tagged proteins (Fig. S5). To study endogenous A3B subcellular compartmentalization and activity, we used a fluorescence-based DNA C-to-U assay. We first found that nuclear fractions of several breast cancer cell lines contain a robust DNA editing activity, which could be ablated by A3B knockdown (Fig. 1c, S6 & S7). Similar results were obtained with an independent knockdown construct (not shown). Protein extracts were then used to assess endogenous A3B’s local dinucleotide deamination preference. Similar to retroviral hypermutation signatures caused by A3B over-expression¹⁶, endogenous A3B showed a strong preference for editing cytosines in the TC dinucleotide context (Fig. 1d & S6). No deaminase activity was observed for extracts from MCF10A (A3B^low) or SK-BR-3 (A3B^null), although it could be conferred by transient A3B transfection (Fig. S8). Both A3B-HA and A3A-HA could elicit measurable TC-to-TU activity in lysates from transfected HEK293T cells (Fig. S9). However, because A3A is myeloid lineage-specific¹⁷ and non-detectable in breast cancer cell lines (Fig. S1 & S2), our expression and activity studies indicated that A3B may be the only enzyme poised to deaminate breast cancer genomic DNA.

To address whether endogenous A3B damages genomic DNA, we employed a combination of biophysical and genetic assays. We first used a mass spectrometry-based approach to quantify levels of genomic uracil in MDA-MB-453 and HCC1569 with high levels of endogenous A3B vs. knockdown levels of A3B (shControl vs. shA3B) (Fig. 2a & S10). Genomic uracil loads decreased by 30% in HCC1569 expressing shA3B and by 70% in MDA-MB-453, where knockdown was stronger (Fig. 2b & S10). Although these relative differences may seem modest, 10 and 20 uracils per Mbp, respectively, this equates to 30,000 and 60,000 A3B-dependent uracils per haploid genome. The actual number of pro-mutagenic uracils may be even higher because several repair pathways may concurrently function to limit this damage.

Second, we used a thymidine kinase-positive (TK^plus) to TK^minus fluctuation analysis¹⁷ to determine whether up-regulated A3B and elevated uracil loads lead to higher levels of mutation. MDA-MB-453 and HCC1569 cells were engineered to express TK, which confers sensitivity to the drug ganciclovir. TK^plus lines were transduced with shA3B or shControl constructs and limiting dilution was used to generate single cell sub-clones. Expanded sub-clones were subjected to ganciclovir selection and resistant cells were grown to visible colonies, which revealed that cells with up-regulated A3B accumulate 3-to-5-fold more mutations (Fig. 2c & S10).

Third, 3D-PCR^17,18 was used to ask whether C-to-T transition mutations accumulate differentially at three genomic loci in A3B^low and A3B^high pools of HCC1569 cells. This technique enables qualitative estimates of genomic mutation within a population of cells because DNA sequences with higher A/T content amplify at lower denaturation temperatures than parental sequences. Lower temperature amplicons were observed for TP53 and c-MYC, but not CDKN2B (Fig. 2d & S10). These amplicons were cloned and sequenced and more C-to-T transition mutations were observed in A3B^high versus A3B^low samples (Fig. 2d & S10). TP53 and c-MYC appeared more mutable than CDKN2B suggesting that all genomic regions are not equally susceptible to enzymatic deamination. Other base substitution mutations were rare, and some C-to-T transitions were still evident in the A3B^low samples possibly due to residual deaminase activity and/or amplification of spontaneous events.

To ask whether A3B triggers additional cancer hallmarks¹⁹, we tried and failed to stably express A3B in several epithelial cell lines. We therefore constructed a panel of HEK293 clones with doxycycline (Dox)-inducible A3B, A3B-E68A-E255Q, A3A, or A3A-E72A eGFP fusions. As measured by flow cytometry, A3-eGFP levels were barely detectable without Dox and induced in nearly 100% of cells with Dox (Fig. S11). A3A over-expression caused rapid S-phase arrest, cytotoxicity, and γ-H2AX focus formation, as reported²⁰ (Fig. 3a-c & S11). In comparison, A3B induction caused a delayed cell cycle arrest, a more pronounced formation of abnormal anucleate and multinucleate cells, and eventual cell death (Fig. 3a-b & S11). A3B induction also caused γ-H2AX focus formation, DNA fragmentation, as evidenced by visible comets, and C-to-T mutations (Fig. 3c-e). A3B catalytic activity, as evidenced by the glutamate mutants, was required for the induction of these cancer phenotypes.

We next asked whether our cell-based results could be extended to primary tumors. First, we quantified mRNA levels for each of the 11 family members in 21 randomly chosen breast tumor specimens, in parallel with matched normal tissue procured simultaneously from an adjacent area or the contralateral breast. Only A3B was expressed preferentially in tumors (p=0.0003) (Fig. S12). We confirmed this analysis by measuring A3B levels in 31 additional tumor/normal matched tissue sets. In total, A3B was up-regulated by ≥3 s.d. in 20/52 tumors in comparison to the patient-matched normal tissue mean, and in 44/52 tumors in comparison to the reduction mammoplasty tissue mean (Fig. 4a, p=7.1×10⁻⁷ and p=2×10⁻⁵; patient information in Table S10). These are underestimates because tumor specimens have varying fractions of non-A3B expressing normal cells. Some of the matched ‘normal’ samples may also be contaminated by tumor cells, as judged by mean levels in mammoplasty samples (Fig. 4a; p=0.002). The related deaminase, A3G, was not expressed differentially in these samples, indicating that these observations are not due to immune cells known to express multiple APOBEC3s¹⁴ (p=0.591). Similar results were obtained by quantifying RNA sequencing data for independent matched tumor and normal pairs²¹, with ~50% showing up-regulated A3B (defined as tumors with A3B levels >3 s.d. above the mean of the normal matched samples; p<0.0001) (Fig. 4b).

Finally, we assessed the impact of A3B on the breast tumor genome by correlating A3B’s deamination signature in vitro and the somatic mutation spectra accumulated during tumor development in vivo. Using a series of single-stranded DNA substrates varying only at the immediate 5’ or 3’ position relative to the _target cytosine, we found that recombinant A3B prefers TC>CC>GC=AC and CA>CT=CC (Fig. S13). These local sequence preferences were then compared to the expected distribution of cytosine in the human genome and the reported C-to-T mutation profiles for melanoma²², liver²³, and breast^8,9,21 tumors. Consistent with a spontaneous origin, the C-to-T frequency is low in liver tumors (~20%) and mutational events appear random (Fig. 4c-d). As expected, C-to-T frequencies are high in melanomas (~80%) and focused at di-pyrimidines consistent with UV-induced lesions and subsequent error-prone lesion bypass synthesis (Fig. 4c-d). Interestingly, the C-to-T frequency was intermediate in 3 independent breast tumor data sets (~40%) and largely focused at trinucleotides that mimic the preferred sites for A3B-dependent DNA deamination in vitro (Fig. 4c-d & S13). The availability of both RNAseq and somatic mutation data²¹ also enabled the establishment of strong positive correlations between A3B expression levels and C-to-T mutation load, overall base substitution mutation load, and TP53 inactivation (Fig. 4e-g). Importantly, tumors expressing high A3B have twice as many mutations (Fig. 4e-f & S14). This equates to 10 C-to-T and 30 total mutations per exome, or approximately 1000 and 3000 mutations per genome, attributable to A3B.

Taken together, we conclude that A3B is a significant mutational source in breast cancer accounting for C-to-T mutation biases and increased mutational loads. Moreover, the disproportional increase in overall base substitutions indicates that some of these other patterns may be due to further processing of U/G mispairs by ‘repair’ enzymes into transitions, transversions, and DNA breaks that could precipitate chromosomal rearrangements (model in Fig. S15 with similarities to AID-dependent antibody diversification mechanisms²⁴). Future work is needed to understand A3B regulation and the potential interplay with other oncogenes and tumor suppressors. For example, although several common breast cancer markers do not correlate with A3B up-regulation, a mechanistic linkage between elevated A3B and inactivated TP53 is evident in primary tumor data and cell lines (Fig. 4g & S16). TP53 inactivation may be required to allow cells to bypass DNA damage checkpoints triggered by A3B.

This is the first study to demonstrate up-regulation of the DNA deaminase A3B in breast cancer and reveal it as a significant source of enzymatic mutation. Conceptually supportive of the original mutator hypothesis²⁵, A3B-catalyzed genomic DNA deamination could provide genetic fuel for cancer development, metastasis, and even therapy resistance. We propose that A3B is a dominant underlying factor that contributes to tumor heterogeneity by broadly affecting multiple pathways and phenotypes. A3B may represent a new marker for breast cancer and a strong candidate for _targeted intervention, especially given its non-essential nature²⁶. A3B inhibition may decrease the rate of tumor evolution and stabilize the _targets of existing therapeutics.