APOBEC3B expression in breast cancer reflects cellular proliferation, while a deletion polymorphism is associated with immune activation

Significance

Somatic mutagenesis is fundamental to the development and evolution of cancers. APOBEC3B (A3B) is a cellular deaminase, which is overexpressed in cancers and believed to be an important cause of cancer-associated mutations. The factors responsible for A3B up-regulation are unknown. Interestingly, a germ-line deletion polymorphism exists, such that a significant proportion of the global population does not express A3B protein. Using large human cancer datasets, we show that A3B expression is strongly associated with cellular proliferation. Furthermore, we identify a pattern of immune activation related to hypermutation in tumors arising in A3B deletion carriers suggesting that these patients could respond differently to immune-directed therapies. These results provide important context for the ongoing study of A3B as a therapeutic target or biomarker.

Abstract

Genomic sequencing studies of breast and other cancers have identified patterns of mutations that have been attributed to the endogenous mutator activity of APOBEC3B (A3B), a member of the AID/APOBEC family of cytidine deaminases. A3B gene expression is increased in many cancers, but its upstream drivers remain undefined. Furthermore, there exists a common germ-line deletion polymorphism (A3B^del), which has been associated with a paradoxical increase in breast cancer risk. To examine causes and consequences of A3B expression and its constitutive absence in breast cancer, we analyzed two large clinically annotated genomic datasets [The Cancer Genome Atlas (TCGA) and the Molecular Taxonomy of Breast Cancer International Consortium (METABRIC)]. We confirmed that A3B expression is associated with aggressive clinicopathologic characteristics and adverse outcomes and show that A3B expression is highly correlated with proliferative features (mitosis and cell cycle-related gene expression) in breast and 15 of 16 other solid tumor types. However, breast cancers arising in homozygous A3B^del individuals with A3B absent did not differ in these features, indicating that A3B expression is a reflection rather than a direct cause of increased proliferation. Using gene set enrichment analysis (GSEA), we detected a pattern of immune activation in A3B^del breast cancers, which seems to be related to hypermutation arising in A3B^del carriers. Together, these results provide an explanation for A3B overexpression and its prognostic effect, giving context to additional study of this mutator as a cancer biomarker or putative drug target. In addition, although immune features of A3B^del require additional study, these findings nominate the A3B^del polymorphism as a potential predictor for cancer immunotherapy.

The recent application of next generation sequencing technologies to characterize the landscape of somatic alterations in solid tumors has yielded major insights into the genes and pathways operant in various cancers. In addition, these studies have enabled the identification of distinct patterns of DNA base alterations that reflect underlying mutational processes (1–6). One of the major discoveries of this effort was the high prevalence of C > T transitions occurring in a preferred sequence motif (TCW, thymine/cytosine/adenine or thymine). This pattern is consistent with the deaminase activity of the AID/APOBEC (apolipoprotein B mRNA editing enzyme, catalytic polypeptide-like) family of enzymes, and thus, its identification invoked an endogenous mutator as a significant contributor to the somatic mutational burden across several cancer types (2–9). Based on the sequence motifs, expression levels, and subcellular localization of APOBEC family members, APOBEC3B (A3B) has been implicated as the likely mutator (6–8).

Analyses of cell line and tumor datasets have shown that A3B gene expression is up-regulated in malignant vs. normal tissues and epithelial cell lines and have shown correlations between A3B expression and the presence of certain somatic mutations, particularly in TP53 (7, 10) and PIK3CA (phosphatidylinositol-4,5-bisphosphate 3-kinase, catalytic subunit alpha) (11). Although the factors that cause up-regulation of A3B in cancer cell lines and tumors remain unknown, these observations form the basis for a model where A3B expression contributes to the accumulation of somatic alterations during the process of carcinogenesis and subsequent evolution, and it has been suggested that inhibition of this activity could represent a strategy for cancer prevention or an adjuvant to other therapies (7, 12). A recent report that A3B expression is associated with adverse outcomes in ER+ breast cancer would be consistent with this hypothesis (13).

In contrast to these findings, a germ-line deletion polymorphism in A3B (A3B^del), which is present at allele frequencies approaching 100% in some populations (14), has been associated with an increased risk of breast cancer (15, 16). This seemingly paradoxical finding suggests that, if A3B is contributing cancer-causing mutations, it may also play some other function in restraining tumor development. A more recent analysis of the association between the A3B^del polymorphism and the APOBEC somatic mutational signature identified an increased mutational burden among A3B^del carriers, the etiology of which remains unknown (17).

We sought to address the biologic and clinical questions raised by these observations using two large clinically annotated breast cancer datasets with gene expression, copy number, and mutational data (1, 18) with the chief aim of resolving the incongruence between gene expression and polymorphism associations. Taking advantage of the natural occurrence of A3B-null individuals (homozygous for the A3B^del allele), we assess the direction of causality for A3B gene expression associations and the biological consequences of the germ-line A3B^del polymorphism. These analyses have identified proliferation as the upstream feature associated with both increased A3B expression and adverse outcomes and revealed a distinct pattern of immune-related gene expression in tumors arising in A3B^del carriers compared with WT individuals.

These findings will enable more focused study of A3B as a potential biomarker and suggest the possibility that the A3B^del polymorphism could be used to identify patients with immune-activated tumors who may have differential benefit from cancer immunotherapies.

Materials and Methods

Datasets.

We collected the two largest publicly available breast cancer datasets with both genomic and transcriptomic data, namely the Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) (18) and The Cancer Genome Atlas (TCGA) (1). For METABRIC, we downloaded the normalized gene expression profiles generated using the Illumina HT 12 IDATS platform and the clinical information from the European genome–phenome archive (EGAS00000000083). The mutation calls for TP53 generated by Sanger sequencing of coding exons used were from the work by Silwal-Pandit et al. (10). A3B^del polymorphism genotypes were identified using Nexus Copy Number 7.0 software (BioDiscovery Inc.) to process Affymetrix-generated CEL files (EGAD00010000164). Altogether, we collected genomic, transcriptomic, and clinical information for 1,992 breast cancer patients from METABRIC.

For TCGA, we downloaded the normalized gene expression data generated using the Illumina RNA-seq platform and the clinical information from the cBioPortal (19) through the cgdsr R package (version 1.1.30). Gene expression values were transformed as $X^{\prime }={\text{log}}_2\left(X+1\right)$, where X represents the normalized fragments per kilobase transcript per million mapped reads values. The point mutation calls for TP53, PIK3CA, and other recurrently mutated genes were also downloaded from cBioPortal. A3B^del polymorphism genotype calls determined by analysis of next generation sequencing data as well as the hypermutation calls used were from the work by Nik-Zainal et al. (17). Altogether, we collected genomic, transcriptomic, and clinical information for 1,006 breast cancer patients from TCGA.

For coexpression analysis, in 59 breast cancer cell lines from the Cancer Cell Line Encyclopedia (CCLE) (20) and 16 other solid tumor types from TCGA, we used gene expression data (microarray for CCLE and RNA-seq for TCGA) and the coexpression function implemented on cBioPortal (19).

Molecular Subtyping.

We used our robust subtype classification model SCMOD2 (21) implemented in the genefu R/Bioconductor package (version 1.16.0) to classify breast tumors into well-established molecular subtypes (21): basal (ER−/HER2−), HER2+, and luminal (ER+/HER2−), which can be further classified into luminal B (ER+/HER2− highly proliferative tumors) and luminal A (ER+/HER2− lowly proliferative tumors).

Statistical Analysis.

Association.

We used Spearman’s rank-based correlation coefficient (22) to estimate the strength and significance of the association between two continuous variables, such as gene expression. To estimate the significance of the association between a continuous variable and a binary variable, we used the Wilcoxon rank sum test (23); we used the Kruskal–Wallis test (24) for discrete variables with more than two levels.

Prognosis.

We used the Cox proportional hazards regression model (25) with disease-free survival to assess the prognostic value of gene expression as a continuous variable. In particular, we used the D index (26) as a robust, unit-free estimate of the hazard ratio (HR). Significance of the D index was computed using the log-rank test (27). We estimated the prognostic value of discrete variables, such as gene expression, divided into binary or ternary groups based on quantiles or A3B allele calls using Kaplan–Meier survival curves (28). We used the log-rank test to test for differences between survival curves (27).

Gene Ontology/Pathway Analysis.

To identify functional clusters of genes coexpressed with A3B, we performed Functional Annotation Clustering using DAVID (29) on the top 100 genes most strongly correlated with A3B expression in each dataset/tissue type.

Gene Set Enrichment Analysis.

To identify the pathways that are significantly enriched in genes associated with the A3B^del polymorphism, we computed the association of each gene with A3B^del calls using the following logistic regression model: $C={\beta }_0+{\beta }_{1}X$, where C is the variable representing the A3B^del calls coded as 0 for homozygous and heterozygous deletions (A3B^del/del and A3B^del/WT, respectively) and 1 for WT (A3B^WT/WT) as in ref. 17, and X is the continuous expression value of each gene in the gene expression profiling platform. The strength and significance of association between each gene and A3B^del calls are given by the coefficient ${\beta }_1$ and its corresponding P value (two-tailed Student t test). Based on the genome-wide ranking of genes based on the significance of their association with A3B^del calls, we used preranked gene set enrichment analysis (GSEA; version 2–2.1.0) (30) to identify gene sets and pathways significantly enriched in genes highly positively or negatively correlated to A3B^del calls. Gene sets were defined by all gene ontology (GO) (31) terms extracted from class c5 in MSigDB (32).

We used a similar approach to identify gene sets enriched in genes associated with the hypermutation calls defined by Nik-Zainal et al. (17). More specifically, we used the same logistic regression model but replaced the A3B^del calls (C) with the hypermutation status (H). To assess the effect of the A3B^del polymorphism beyond the hypermutation phenotype, we updated the model to explicitly control for hypermutation when ranking the genes based on their association with A3B^del calls: $C={\beta }_0+{\beta }_{1}X+{\beta }_{2}H$. As before, the genes are ranked based on the significance of β₁, and preranked GSEA is performed to identify the enriched pathways and gene sets.

We performed the GSEA for A3B^del in each of TCGA and the METABRIC datasets (separately) and for hypermutation status only in TCGA. In each case, we performed the GSEAs for the global population of breast cancer patients as well as each subtype separately. Additional details are provided in SI Appendix, Supplementary Methods.

Reproducibility.

To ensure full reproducibility, this work complies with the guidelines proposed by Gentleman (33) in terms of the availability of the code and the reproducibility of results and figures. The procedure to properly set up the software environment and run our analysis pipeline is provided in SI Appendix.

All analyses were performed in both the METABRIC and TCGA datasets where corresponding data were available. Results from METABRIC, which include more patients and have more complete clinical annotation, are shown in the figures for analyses involving clinicopathologic or survival end points. Results from TCGA are shown in the figures for coexpression and gene signature analyses, because the RNA-seq data in this dataset provide a more quantitative measurement (34). To show the high consistency of our results across datasets, the corresponding figures using the second study are included in SI Appendix.

Results

A3B Expression Correlates with Adverse Prognostic Features and Is Associated with Outcome in Breast Cancer.

We evaluated the association between A3B gene expression [detected by microarray (METABRIC) or RNA-seq (TCGA)] and established clinicopathologic features, focusing on the METABRIC dataset, where clinical annotation is more complete. We found that presence of lymph node involvement, histologic grade, lack of ER expression, and overexpression/amplification of HER2 are each associated with increased expression of A3B (Fig. 1). No significant correlation was observed between A3B expression and tumor size at diagnosis. Although A3B expression levels do vary by intrinsic subtype (SI Appendix, Fig. S1), the association between A3B expression and tumor grade is present across all tumors as well as within each subtype (Fig. 1). Consistent with these observations, in univariable analysis, A3B expression as a continuous variable is associated with recurrence [HR = 1.54; 95% confidence interval (95% CI) = 1.34–1.77; P = 1.2 × 10⁻⁹], with its prognostic value being limited to the luminal tumor subtype (HR = 1.60; 95% CI = 1.34–1.93; P = 4.1 × 10⁻⁷). To take into account the imprecision of HR estimates within subtype, we included an interaction term between A3B expression and luminal status in the Cox model. This interaction term was highly significant (P = 2.7 × 10⁻⁵), suggesting that the observed difference of A3B prognostic value with respect to luminal status is statistically significant. This adverse prognosis associated with elevated A3B expression is similarly evident when patients are binarized into high and low A3B-expressing groups. Although this association holds across the entire METABRIC population, it is driven by the luminal subgroups (Fig. 2).

Fig. 1.

A3B gene expression is associated with clinicopathologic features, including lymph node involvement (P = 3.2 × 10⁻⁴), histologic grade (P = 1.4 × 10⁻⁴⁵), lack of ER expression (P = 4.8 × 10⁻⁴³), and HER2 overexpression/amplification (P = 7.6 × 10⁻⁵), in METABRIC breast cancers. P values are for the global dataset.

Fig. 2.

High A3B gene expression is associated with recurrence after treatment for luminal breast cancer in METABRIC. Kaplan–Meier plots are dichotomized by median A3B expression for all breast cancers (global) or within each intrinsic subtype. P < 0.001 for global, luminals, luminal A, and luminal B. DFS, disease-free survival.

Several gene expression-based prognostic signatures have been developed to predict breast cancer outcomes. We evaluated whether A3B expression is associated with scores calculated with these classifiers and observed strong correlation between A3B expression and several well-established signatures, including the Genomic Grade Index (35) (Fig. 3 and SI Appendix, Fig. S2), Chromosomal Instability (36), MammaPrint (also referred to GENE70) (37), and Oncotype DX (38). These associations were present in both datasets, shared across subtypes, and either specific to or much stronger for A3B compared with other A3 family members. Proliferation has been identified as a common underlying feature driving these prognostic signatures (39). We directly assessed the association between A3B expression and key proliferation-associated genes and observed strong correlation [for example, AURKA (aurora A kinase) (Fig. 4 and SI Appendix, Fig. S3), MKI67, and CCNB1]. Indeed, coexpression analysis in both METABRIC and TCGA identified strong enrichment of mitosis and cell cycle-related functional ontologies associated with A3B expression across all cancers and within subtypes (SI Appendix, Table S1). We extended this analysis to 16 other solid tumor types using DAVID with RNA-seq gene expression data from TCGA and confirmed that A3B expression is strongly associated with mitosis and cell cycle in 16 of 17 examined, with the exception lung squamous cell carcinoma (SI Appendix, File S1). As in the breast cancer datasets, the correlations between AURKA and A3B gene expression in these other tumor types provide support for the results of the GO analyses (SI Appendix, Fig. S4).

Fig. 3.

A3B gene expression correlates with the Genomic Grade Index (GGI) score in all breast cancers (global: ⍴ = 0.51, P < 1 × 10⁻¹⁶) and within individual intrinsic subtypes (basal: ⍴ = 0.23, P = 3.1 × 10⁻³; Her2: ⍴ = 0.33, P = 1.1 × 10⁻⁴; luminals: ⍴ = 0.55, P < 1 × 10⁻¹⁶; luminal A: ⍴ = 0.39, P = 1.4 × 10⁻¹⁴; luminal B: ⍴ = 0.53, P < 1 × 10⁻¹⁶) in TCGA.

Fig. 4.

A3B gene expression correlates with expression of the proliferation-associated gene AURKA across all breast cancers (global: ⍴ = 0.46, P < 1 × 10⁻¹⁶) and within individual intrinsic subtypes (basal: ⍴ = 0.22, P = 3.8 × 10⁻³; Her2: ⍴ = 0.28, P = 9.8 × 10⁻⁴; luminals: ⍴ = 0.50, P < 1 × 10⁻¹⁶; luminal A: ⍴ = 0.31, P = 1.1 × 10⁻⁹; luminal B: ⍴ = 0.45, P = 8.9 × 10⁻¹⁶) in TCGA.

To further evaluate the prognostic importance of A3B expression, we performed multivariate analyses with a model incorporating clinical features (age, grade, tumor size, nodal status, and ER status) and proliferation-associated gene expression (AURKA). When all of these features were considered, A3B expression was no longer a significant predictor of outcome (all tumors: HR = 0.98; 95% CI = 0.85–1.12; P = 0.74; luminals: HR = 1.12; 95% CI = 0.86–1.47; P = 0.39). Borderline significance was observed in luminal A and HER2 subgroups (luminal A: HR = 2.11; 95% CI = 0.99–4.49; P = 0.053; HER2: HR = 1.60; 95% CI = 1.07–2.39; P = 0.02).

Together, these analyses in two large independent breast cancer datasets affirm that A3B expression is associated with adverse prognostic features and outcomes in breast cancer and reveal a link to cellular proliferation.

A3B Expression Is Increased in TP53- but Not PIK3CA-Mutated Breast Cancers.

Because A3B has been implicated as an endogenous mutator in breast and other cancers, we evaluated the association between mutations in the most common recurrently altered genes in breast cancer (TP53 and PIK3CA) (1, 40) and A3B expression. As has been recently shown (10), we found that A3B expression is increased in tumors with TP53 mutations in both the METABRIC and TCGA datasets (Fig. 5 and SI Appendix, Fig. S5A). For PIK3CA, however, the presence of point mutations was, instead, associated with lower A3B expression (Fig. 5) in breast cancers. Because the spectrum of base changes ascribed to A3B activity occurs in a specific sequence context, we considered the PIK3CA hotspot mutations separately: the E542K and E545K alterations conform to the A3B motif, whereas H1047R does not (11). Nevertheless, each point mutation exhibited a similar pattern, where the presence of a PIK3CA alteration was associated with lower A3B expression (Fig. 5 and SI Appendix, Fig. S5B). Systematic analysis of genes recurrently mutated in breast cancer (9) using TCGA breast dataset revealed no remarkable pattern of increased A3B expression in somatically mutated vs. unmutated tumors, although this analysis was limited by low mutation frequency (<10%) for all genes with the exceptions of TP53, PIK3CA, CDH1, and GATA3.

Fig. 5.

A3B gene expression in TCGA breast cancers is increased in tumors with TP53 mutation (vs. WT) but decreased in tumors with any PIK3CA mutation or the E542K or E545K A3B motif mutations (vs. WT). Similar associations are present across all tumors and within individual subtypes. Note that groups with less than three samples were omitted. MUT, mutation.

Association Between A3B Expression and Proliferation Is Retained in Breast Cancer Cell Lines.

To assess whether the observed associations between A3B expression and proliferation are cell-autonomous, we further evaluated the relationship between A3B expression and proliferation in the CCLE breast cancer cell lines dataset. As in the METABRIC and TCGA breast tumor datasets, coexpression analysis identified mitosis and cell cycle-related terms as the most strongly associated GOs (SI Appendix, Table S2).

Germ-Line A3B Deletion Polymorphism Is Associated with A3B Expression but Not Clinical Variables.

To assess the causality of these A3B associations, we evaluated associations between A3B^del and these same factors, reasoning that, if A3B expression contributes to the development of high-grade, proliferative, and clinically aggressive cancers, the constitutional absence of A3B protein would result in an altered spectrum of tumor features. A3B^del allele frequencies were determined for METABRIC, and previously published values (17) were used for TCGA. As expected, A3B expression was highly associated with the A3B^del genotype (Fig. 6 and SI Appendix, Fig. S6). Importantly, we did not observe reciprocal up-regulation of other A3 family members in the absence of A3B—specifically, expression of A3A (the 3′ UTR of which is replaced by the A3B 3′ UTR in the polymorphic A3B^del allele) was not increased in tumors from A3B^del carriers (Fig. 6 and SI Appendix, Fig. S6).

Fig. 6.

A3B expression is (A) reduced in TCGA breast cancers from patients with germ-line A3B^del polymorphism (B) without a reciprocal increase in A3A. The A3B^del genotype is not associated with (C) expression of AURKA or (D) the Genomic Grade Index (GGI). Similar patterns are present across all tumors and within intrinsic subtypes.

In contrast to the associations observed between A3B expression and clinicopathologic features, we observed no association between the A3B^del germ-line polymorphism and tumor ER or HER2 status, lymph node involvement, tumor grade, intrinsic subtype distribution (Table 1), or somatic TP53 or PIK3CA mutation status (SI Appendix, Table S3). Age at diagnosis was slightly lower in A3B^del carriers. As previously shown, A3B^del is associated with the presence of the hypermutation phenotype defined by Nik-Zainal et al. (17).

Table 1.

Association of A3B^del polymorphism with clinicopathologic features of breast cancers in the METABRIC and TCGA datasets

Genotype	A3Bdel/del	A3Bdel/WT	A3BWT/WT	P value
METABRIC
All patients*	18	199	1,771
Grades 1/2/3	1/7/9	22/71/95	147/695/851	0.7
Age (y) median (SD)	59.8 (10.7)	59.8 (12.8)	62.3 (13.0)	0.04
Size (cm) median  (interquartile range)	2.5 (1.0)	2.1 (1.4)	2.3 (1.3)	0.3
ER+	83.3%	77.8%	77.3%	0.8
HER2+	0.0%	18.8%	18.0%	0.4
Node+	50.0%	46.2%	47.8%	0.9
Subtype				0.1
Luminal A/B	4/9	66/57	609/639
Basal	3	44	343
Her2	2	32	180
TCGA
All patients*	13	86	597
Age (y) median (SD)	54 (12)	54 (12.4)	58 (13.3)	0.07
ER+	69.2%	77.6%	78.9%	0.7
Node+	62.5%	55.2%	49.8%	0.6
Subtype				0.54
Luminal A/B	3/3	25/34	160/240
Basal	2	12	95
Her2	5	12	83
Hypermutated†	30.8%	19.0%	10.7%	0.02

^*For each data type, the total number of cases may differ because of missing or incomplete data.

^†From the work by Nik-Zainal et al. (17).

Similarly, we observed no association (either statistically significant or consistent trend) between A3B^del status and expression of individual proliferation-associated genes or prognostic signature scores (Fig. 6 and SI Appendix, Fig. S7). Finally, we observed no association between the A3B^del polymorphism genotype and breast cancer outcomes (Fig. 7).

Fig. 7.

Germ-line A3B^del genotype is not associated with recurrence after treatment for early breast cancer in METABRIC. Kaplan–Meier plots show disease-free survival (DFS) by A3B^del genotype for all breast cancers (global) and by intrinsic subtypes. P > 0.6 for all.

Germ-Line A3B Deletion Polymorphism Is Associated with Expression of Immune-Related Genes in Breast Tumors.

The surprising observation that A3B^del is not associated with any standard clinicopathologic characteristics, despite the ascribed role for A3B as an endogenous mutator and the shown association between A3B^del and breast cancer risk, prompted us to investigate the METABRIC and TCGA gene expression datasets to identify potential distinguishing features of tumors arising in A3B^del carriers. To discover biological processes associated with A3B^del, we applied GSEA (30) in METABRIC and TCGA. Consistent with our observations above, no association between proliferation-associated processes and A3B^del was observed. Instead, in each dataset, a striking association was identified between A3B^del status and enrichment of immune response-related gene sets [Table 2 shows the top-ranked gene set for METABRIC (immune response) and TCGA (cytokine binding)].

Table 2.

GSEA of A3B^del vs. A3B^WT and hypermutated vs. nonhypermutated breast cancers

Comparison and gene set	Normalized ES	FDR q value
METABRIC
A3Bdel vs. A3BWT
Immune response	−3.19	<1 × 10−16
Cytokine binding	−2.68	<1 × 10−16
TCGA breast
A3Bdel vs. A3BWT
Immune response	−2.01	0.006
Cytokine binding	−2.18	0.002
Hypermutated vs. not
Immune response	2.81	<1 × 10−16
Cytokine binding	2.30	<1 × 10−16
A3Bdel vs. A3BWT hypermutated model
Immune response	−1.77	0.096
Cytokine binding	−1.93	0.068

ES, enrichment score; FDR, false discovery rate.

Immune Activation in A3B^del Tumors Is Related to a Hypermutation Phenotype.

Antitumor immune responses have been attributed to neoepitopes arising from somatic mutations in cancer cells (41), and the A3B^del polymorphism has been associated with a hypermutation phenotype in breast cancers (17). To assess whether hypermutation could be responsible for the A3B^del-associated immune signature that we detected, we compared hypermutated with nonhypermutated breast cancers from TCGA (17) using GSEA. Across all tumors as well as within luminal and basal subtypes, we identified a strong signal of immune activation in the hypermutated cancers reminiscent of the expression patterns in A3B^del cancers (Table 2). A3B^del is associated with hypermutation status, particularly in luminal tumors (SI Appendix, Fig. S8). To determine whether the A3B^del polymorphism contributes to immune activation independent of the hypermutation phenotype, we constructed a GSEA linear model comparing tumors based on A3B^del status corrected for hypermutation status. In this model, the immune activation pattern was attenuated (Table 2), suggesting that hypermutation contributes to this phenomenon.

Discussion

Since the initial identification of a pattern of mutations fitting the activity of APOBEC3 enzymes in breast cancer sequencing data (3, 7, 8), numerous studies have gone on to identify a similar pattern of mutations across several human cancer types (2, 5, 6, 8, 42), implicating A3B in somatic mutagenesis and by inference, carcinogenesis. Despite the great interest that these results have attracted, the cause of A3B up-regulation that is observed has remained enigmatic. Our results show that A3B expression is increased in highly proliferative states, consistent with the increased expression in cancers vs. normal tissues (7, 43) and high- vs. low-grade tumors. Our analysis of A3B-null tumors (homozygous for the A3B^del polymorphism), however, revealed that the constitutive absence of A3B has no significant impact on the same clinicopathologic features. These data suggest that A3B expression is regulated in a cell cycle-dependent manner, and gene expression profiling of cell cycle states supports this conclusion (44). An association with proliferation is also supported by the observation that APOBEC pattern mutations predominate later in tumor evolution (42, 45). Although increased A3B expression might result in accumulation of additional oncogenic mutations, resulting in increased proliferation and a positive feedback loop, we show that A3B expression per se is not the primary cause of proliferative or adverse clinical features in breast cancer but rather, a consequence of these features. This finding has implications for the potential clinical use or application of strategies to inhibit A3B.

Although these analyses do not directly address the increased risk of breast cancer development that has been associated with the A3B^del allele (15, 16), we find no evidence of increased APOBEC3A-B hybrid mRNA expression in deletion carriers, which has been suggested as a potential cause of increased mutational burden and cancer risk in carriers (46). However, while performing our analyses, we did observe a remarkable association between expression of each of the APOBEC3 family members with the exception of A3B and IFN/inflammation-associated gene expression (STAT1 signature) (SI Appendix, Fig. S10). In light of this finding, it is conceivable that elevated expression of APOBEC3A occurs in A3B^del carriers during periods of inflammation and results in hit and run mutagenesis but is no longer detectable in the established tumors in the breast cancer datasets.

We found no evidence of systematic differences in traditional clinicopathologic features, subtype distributions, or outcomes between breast cancers based on germ-line A3B^del status, although the relatively low A3B^del allele frequency (and consequent small numbers of A3B^del homozygous individuals) in these predominantly Caucasian patient datasets is a limitation. Future analyses of other ethnic populations with higher A3B^del allele frequency (14) will be important to confirm these findings. Nevertheless, we have identified a reproducible pattern of increased immune response—including cytokine and T-cell activation gene expression programs—in tumors arising in A3B^del carriers. Although the nature and temporal development of the immune processes associated with the observed expression changes detected by GSEA are of great interest, this technique provides limited resolution to characterize the associated phenomena, and additional study is required. However, given the canonical function of APOBEC enzymes in the immune system (47, 48), several potential mechanisms for this observation could be considered arising from either tumor or immune cell intrinsic effects of the A3B^del polymorphism. The previously described association of A3B^del with a somatic hypermutation phenotype (17) and recent interest in the relationship between somatic mutational burden and antitumor immune response prompted us to examine the differences in gene expression profiles between the hypermutated and nonhypermutated breast cancers. Our analysis revealed, to our knowledge, previously undescribed and marked immune activation in the hypermutated cancers. By controlling for the hypermutation status, we observed that the original signal that we identified in A3B^del tumors can be attributed, at least in part, to increased mutational burden (or the process responsible for it). Because hypermutation is likely a continuum but was binarized in these analyses (17), some or all of the residual immune-associated patterns in A3B^del tumors may also be attributable to degrees of hypermutation. Lacking exome- or genome-level sequencing data in METABRIC, confirmatory analyses in our second breast cancer dataset were not possible; however, given the reproducible A3B^del-associated immune gene expression signatures, we believe that this phenomenon is likely generalizable.

In light of the immune activation observed in A3B^del-hypermutated tumors, the increased breast cancer risk found in A3B^del carriers (15, 16) seems, perhaps, counterintuitive. Although our analyses of established tumors are unable to ascertain the presence or temporal development of the observed immune phenomena during tumorigenesis, the increased mutational rate/genomic instability presumably overrides the immune activation and results in tumor formation. We did not find any association between A3B^del and breast cancer outcomes, which one might expect in the setting of an activated immune response. However, recent analyses of tumor-infiltrating lymphocytes in breast cancer have shown this pathologic measure of immune response to have important prognostic or predictive effects only in specific tumor subtypes or clinical contexts (49, 50). Thus, the lack of any detectable effect of A3B^del on prognosis in this analysis is perhaps unsurprising, especially given the relationship with hypermutation that we identified.

Notwithstanding the lack of prognostic effects of the A3B^del polymorphism in patients treated with standard therapies in the METABRIC and TCGA cohorts, the recent and dramatic advances in the development of cancer immunotherapies (51) provide an emerging context where the immune effects that we have observed could be of clinical importance. Recent studies examining factors predictive of response to immune checkpoint inhibitors have identified associations between mutation- and immune-associated features and treatment response (52–54), although they rely on complex tumor analyses, including whole-exome sequencing. It is certainly plausible that the immune features associated with A3B^del cancers (by way of the hypermutated phenotype) could influence response to immunotherapy-based treatment strategies. Genotyping for A3B^del (which can be performed quickly and inexpensively from peripheral blood or tissue samples) could provide a simple selection or stratification factor to identify populations of interest for such treatments, and its exploration as a predictive biomarker is warranted.