Introduction

Two-pore domain potassium (K2p) channels enable background leak of potassium (K+) ions and the K2p-family has 15 members [1, 2]. K2p-channels are important for baseline cellular activity at rest including membrane-potential, calcium homeostasis and cell-volume regulation [3]. Evidence from laboratory studies supports the hypothesis that alterations in expression/function of K2p-channels may play a role in cancer development/progression [3,4,5]. The role of K2p-channels in breast cancer (BC) is emerging and recent reviews suggest potential clinical utility [4, 6, 7]. Williams et al. examined K2p-channel expression in a microarray database study, and reported that all but five members showed altered expression in BC [3].

Preliminary reports suggest that some K2p-channels may be novel molecular markers with differential expression related to biologically-aggressive triple-negative (TN) type BC. In estrogen-receptor-alpha positive (ER+) cell lines, KCNK5 expression appears to be under regulatory control of ER-alpha, and estrogen-response-elements (ERE) are found in the enhancer region of KCNK5 [8]. Clarke et al. suggest that upregulation of KCNK5 is associated with poor outcome for TNBC-related basal-like subtype [9]. Conway et al. suggest that KCNK4 may be differentially methylated according to race, with relatively higher median DNA-methylation observed for non-Hispanic (nH) black vs. nH-white BC [10]. Other studies similarly suggest that differential methylation of KCNK9 may be related to nH-black race and TNBC [11,12,13,14]. Mu et al. first described the KCNK9 gene on 8q24.3 as a potential proto-oncogene; gene-amplification (10%) and protein-overexpression (44%) were detected in BC, but not in normal tissue-controls [15]. DNA copy-number loss has been observed for KCNK12 in a small clinical study of tubular BC [16].

K2p-channels also hold promise for innovative oncologic therapeutic/prevention strategies [7, 17, 18]. Current molecular-epidemiologic data characterizing K2p-channels in clinical BC are limited. Our study goal was to systematically evaluate associations between K2p-gene-family mRNA-expression and DNA-methylation with TNBC in The-Cancer-Genome-Atlas (TCGA), which contains genomic information from DNA-methylation and mRNA-expression arrays for almost 1100 human BC.

Main text

Materials and methods

Sample and procedure

Publicly available BC data was downloaded/aggregated (KAD) from TCGA-Data-Matrix and cBioPortal-for-Cancer-Genomics (provisional dataset) web-sites (February 2016) [19, 20]. TCGA molecular BC dataset was generated by TCGA-Research-Network and details have been previously published [21, 22]. The assembled dataset of invasive BC (using TCGA sample-code suffix −01 for primary solid-tumors) included information on: (1) clinical-pathological factors: age/stage at diagnosis, menopausal status, lymph-node status, estrogen/progesterone receptor (ER/PR) status, human-epidermal-growth-factor-receptor-2 (HER2) status, and race/ethnicity; and (2) K2p-gene-markers including: (a) Level 3 DNA-methylation probes (Illumina-Infinium-Human-Methylation-450K-BeadChip beta-values (450K array), n = 767) [23]; and (b) Level 3 RNA-Seq gene-expression data (Illumina-HiSeq-platform, gene level RNA-Seq by Expectation Maximization (RSEM)-normalized and log2-transformed values, n = 959) [24]. Cytosine–guanine (GpG) methylation loci within 25 kb from either end of the genes of interest were included for examination (UCSC-Genome-Browser [25], hg37); 724 GpG loci were included, but was reduced to 608 after exclusion of probes with null-reads. Analytic dataset consisted of 1040 women with information on stage, and all women with available data were included in analyses. The study was approved by UIC IRB.

Breast cancer subtype

The primary study-endpoint was a binary variable of combined TN-status, developed by aggregating data from immunohistochemical (IHC) and intrinsic subtype assays (PAM50) with a referent of luminal-A status (ER/PR+/HER2− status). For sensitivity analyses, data on Intrinsic Subtype classification was obtained from supplementary publication data based on PAM50 assay [21], and data on Integrative Cluster (IC) classification based on copy-number-change was obtained from Cancer-Research-UK-Cambridge-Institute (Oscar Rueda and Carlos Caldas) [26,27,28]. Secondary endpoints in sensitivity analyses included: ER/PR− status (IHC; referent ER/PR+); basal-subtype (PAM50; ER/PR−/HER2−, and cytokeratin 5/6+ and/or HER1+; referent luminal-A); and IC10-type (referent IC3). Gene-expression patterns were also compared for RSEM and Agilent arrays.

Statistical methods

For categorical variables, Chi square test-of-association was used to examine significance of relationships. For continuous variables, t-test was used to compare means as appropriate to distribution characteristics; and for non-normally distributed data, nonparametric Wilcoxon rank-sum test was used.

Tertiles were used for categorization of gene-expression, except where limited by small sample-signals, then binary mRNA-overexpression was defined as a z-score > +1 standard deviation (SD) [29]. Association of all K2p gene-expression and DNA-methylation with TN-subtype was evaluated using age (years) and race (nH-white, nH-black) adjusted generalized-linear regression-models (glm) with binomial-distribution and log-link to estimate prevalence risk-ratios (RRs) and their corresponding 95% confidence-intervals (CIs). The significance threshold for gene-expression and DNA-methylation was p < 0.05 and Bonferroni-corrected p-values were used where appropriate (Bonferroni-adjusted threshold for expression was p < 0.0033; and for methylation p < 8.22 × 10−8).

Methylation glm results were sorted on p-values (smallest to largest) and all loci associated with TN-subtype were selected for reporting and further analysis; loci were then evaluated for functionality related to expression using Spearman’s correlation. Methylation was deemed significantly correlated with expression if the Spearman’s Rho-value was either ≤−0.2 or ≥+0.2 and associated with a p-value <0.05 [10].

Results

Characteristics of the study sample are presented in Table 1. Overall, most of the sample was age >50 years (73%), postmenopausal (76%), nH-white (86%), stage-II (58%), and ER/PR+ (66%) and HER2− (89%); 631 patients had data on TN (n = 159) vs. luminal-A (n = 472) status.

Table 1 Selected characteristics of study sample
Full size table

All K2p gene-expression was non-normally distributed (Additional file 1: Table S1). Table 2 demonstrates that overexpression of KCNK5/KCNK9/KCNK12, and underexpression of KCNK6/KCNK15, were associated with TN-subtype in age/race adjusted models (all p < 0.0033). In fully-adjusted models, KCNK5/KCNK9/KCNK12/KCNK15 remained associated with TN-subtype.

Table 2 Association of K2p gene expression and triple negative subtype
Full size table

Gene-expression sensitivity analyses (Additional file 2: Table S2) show similar K2p RSEM-expression patterns for ER/PR-, basal and IC10 type tumors; however, overexpression of KCNK7 was also seen with basal-subtype, and expression of KCNK9 and KCNK5 was not associated with basal and IC10 types, respectively. Comparative analyses using the Agilent expression-array (n = 590) (Additional file 2: Table S2) shows similar expression patterns for KCNK5/KCNK6/KCNK12/KCNK15, but additionally shows consistent underexpression of KCNK4 across subtype, and that KCNK9 was not associated with any subtype for this array.

Of 608 CpG loci, 195 (114 hypomethylated and 81 hypermethylated) loci were found to be associated with TN-subtype (p < 8.22 × 10−8, Additional file 3: Table S3). Significantly-associated negatively-correlated expression patterns differentially observed in TN vs. luminal-A subtype (Table 3; and Additional file 4: Table S4) were demonstrated for: KCNK2 (gene-body: cg04923840, cg13916421), KCNK5 (gene-body: cg05255811, cg18705155, cg09130674, cg21388745, cg00859574) and KCNK9 (TSS1500: cg21415530, cg12175729; KCNK9/TRAPPC9 intergenic: cg17336929, cg25900813, cg03919980). Loci listed for KCNK5/KCNK9 all showed relative hypomethylation for probability of TN vs. luminal-A subtype.

Table 3 Annotation and expression correlation for select K2p methylation loci associated with triple negative subtype
Full size table

Discussion

K2p RSEM-expression analyses show that specific genes are associated with TN-subtype (overexpression of KCNK5/KCNK9/KCNK12, and underexpression of KCNK6/KCNK15). Sensitivity analyses demonstrate consistency; similar RSEM-expression patterns of association were largely evident for ER/PR-, basal and IC10 related subtypes, although KCNK9 and KCNK5 overexpression failed to meet significance criteria for basal and IC10 types, respectively. Comparative analyses with the Agilent-array supports RSEM findings for KCNK5/KCNK6/KCNK12/KCNK15.

Regarding specificity for association of K2p-expression with tumor vs. normal sample-type, we used the MEXPRESS web tool for visualizing expression, DNA methylation and clinical TCGA data [30], and found that KCNK9 (p = 2.58 × 10−9) and KCNK12 (p = 4.63 × 10−10) overexpression appeared to be associated with tumor type, while KCNK5 overexpression appeared to have marginal association (p = 0.0757). KCNK6 and KCNK15 overexpression (associated in our study with luminal-A subtype) also appeared to be associated with tumor sample-type (both p < 2.2 × 10−16).

Specific methylation loci on KCNK2/KCNK5/KCNK9 were found to be significantly-associated with negatively-correlated gene-expression and differentially observed in TN vs. luminal-A subtype, which suggests potential functional significance. All select loci on KCNK5/KCNK9 had negative delta-beta values indicating relative hypomethylation in TN-subtype. The overall levels of negative expression Rho-values for KCNK5 and KCNK9 were more than twice larger for TN than luminal-A subtype. Hence it is plausible that KCNK5/KCNK9 overexpression related to TN-subtype may be functionally related to specific tumor CpG loci hypomethylation.

Prior literature suggests KCNK5 may be under control of ER-alpha signaling and upregulation is associated with worse prognosis for basal-type tumors [8, 9]. Although we found association of KCNK5/KCNK9 overexpression and TN-subtype, we were unable to relate to prognosis. Further, 17-beta-estradiol is known to induce KCNK5 gene-expression via EREs found in the enhancer region of the KCNK5 gene [8]. Hence the observed pattern in our study could be interpreted as evidence of loss of regulatory control of ER-alpha signaling pathways. An alternate explanation could be that KCNK5 gene-expression in TNBC is incompletely understood. Prior studies have implicated KCNK9 as a proto-oncogene in BC and suggest importance in TNBC [12,13,14,15, 31]. KCNK9 is a maternally-imprinted gene with monoallelic-expression predominantly in brain tissue, but expression has been observed in breast tissue, and both are of ectodermal origin [11]. It is theorized that KCNK9 overexpression may occur due to acquired relative hypomethylation and subsequent functionally biallelic-expression which may be equivalent to duplication of an active allele [12]. Our findings regarding the functional relationship of KCNK9 methylation/expression may be consistent with this theory, as we observed increased KCNK9-expression together with relative hypomethylation at functional loci. We also observed a similar relationship for KCNK5. There is scant prior data on KCNK12 expression in BC and Williams et al. suggested no alteration in their study [3]; we found KCNK12 overexpression to be associated with TN subtype. Williams et al. described overexpression of KCNK6/KCNK15; we found that these were specifically overexpressed in luminal-A subtype.

Our findings are consistent with prior reports suggesting that some K2p-channels may be novel molecular-markers with differential expression related to biologically-aggressive TNBC, and may hold promise for innovative oncologic therapeutic/prevention strategies, as experience gained with pharmacological manipulation of K+ channels in other pathologies, might facilitate their use as molecular-targets in precision-medicine [7]. Wallace et al. suggest that K+ channel activity demonstrates promise as a pharmacologic target in BC and may represent a mechanism through which phytoestrogens act (e.g. Genistein) [18]. A recent study showed that inhibition of the intermediate-conductance calcium-activated K+ channel KCNN3 with specific blockers including the antifungal clotrimazole, suppressed cell proliferation, migration and epithelial–mesenchymal transition in TNBC cells, and illustrates the potential of K+ channels as novel targets [17]. Sun et al. demonstrated that a monoclonal antibody (Y4) against KCNK9 extracellular-domain effectively inhibits growth of human lung cancer xenografts and BC metastasis in mice, and suggest that antibody-based KCNK9 targeting is a promising therapeutic strategy in KCNK9 overexpressing malignancies [32]. Skaar et al. have been researching whether loss of KCNK9 imprint-control and/or differential methylation could be adapted in novel approaches for clinical diagnosis and targeted therapy for TNBC [11,12,13,14].

Conclusions

In summary, our results suggest that several K2p-channel methylation/expression markers are related to TNBC. KCNK5 and KCNK9 show overexpression and relative hypomethylation of specific CpG loci, and have negative methylation/gene-expression correlations, which are features that support functional biology. Study strengths include: TCGA is a large, well characterized BC resource; we used the target tumor-tissue of interest, not peripheral blood; and our findings are consistent with, and build on, other reports from TCGA. Our findings are considered preliminary but may lead to further exploration of K2p-channels as potential molecular-targets in precision-medicine, as pharmacological manipulation of K+ channels is currently feasible.

Limitations

TCGA is not a population-based study or clinical trial and may be subject to selection bias. There was missing data on CpG loci with null-reads which may have been excluded due to quality-control protocols. Missing data may cause loss of analytic precision and power and could potentially bias results [33]. Our analysis used specific CpG probes present on the 450K array, but a more comprehensive fine-mapping sequencing approach could facilitate interrogation of additional loci. Potential modeling inadequacies may also result from failure to include molecular-markers not available to us, such as protein-expression, non-coding RNAs and expression-quantitative-trait-loci. K2p copy-number-change may also be important and this analysis is planned as a future study. Regarding survival analysis, specific treatment data is unavailable, and since follow-up time and number of survival events are limited, we were unable to examine prognosis, as our study is largely under-powered to detect differences between groups. Regarding sensitivity analyses, overlap between the RSEM and Agilent expression-arrays was only 499 cases, and as such sampling differences could account for observed variation in expression profiling. These limitations make the present study results preliminary in nature and our findings require validation in additional studies.