Epigenetic silencing of CREB3L1 by DNA methylation is associated with high-grade metastatic breast cancers with poor prognosis and is prevalent in triple negative breast cancers

A panel containing 40 breast cancer cell lines (and 4 non-tumorigenic breast cell lines) was obtained from the American Type Culture Collection (ATCC, Manassas, Virginia, USA 30-4500 K). Cells were cultured according to ATCC recommendations for fewer than 6 months from the time of resuscitation. All cell lines were authenticated by the supplier (http://​www.​ATCC.​org).

To examine the impact of DNA methylation and/or histone acetylation on CREB3L1 expression, the human breast cancer cell lines BT20, HCC1806 and MDA-MB-468 cells were treated with a DNA methyltransferase inhibitor, 5-aza-2′-deoxycytidine (DAC) (Sigma Aldrich, Oakville, ON, Canada), and/or a histone deacetylase inhibitor, trichostatin A (TSA) (Sigma Aldrich, Oakville, ON, Canada). Cells were grown to 60–70 % confluency and treated with DAC (1 μM) for 96 hours (changing to fresh DAC-containing media every 24 hours), with or without TSA (1 μM), for the last 18 hours as previously reported [26‐28]. Cytotoxicity measurements were carried out using a Cytotox Glo Cytotoxicity assay (Promega, Madison, WI, USA G9290) according to their instructions, and no cytotoxicity was observed at 1 μM TSA. Three independent experiments were performed with triplicate samples, with one set used to prepare DNA, one for RNA and the other lysed for western blot analysis of CREB3L1 protein levels, as detailed below. In some instances, cells were treated with the proteasomal inhibitor, MG132 (Sigma Aldrich, Oakville, ON, Canada), at a concentration of 3 μM for the last 18 hours prior to lysis, to prevent the degradation of CREB3L1 protein and better enable visualization on western blots.

Sections from 216 human primary breast tumors and corresponding de-identified clinical data were obtained from the Manitoba Breast Tumor Bank (Winnipeg, MB, Canada). Four different tumor types were obtained including: infiltrating ductile, infiltrating colloid, infiltrating lobular and infiltrating papillary. For the purposes of assessing the possible correlation between low CREB3L1 expression and more advanced or aggressive tumor type, we have ordered these tumor types from least to most aggressive (colloid, lobular, ductile, papillary) based on several sources [29‐33]. Tumor samples were graded by pathologists at the time of diagnosis based on mitotic count, nuclear pleomorphisms and tubule formation, from low (grade 4) to high (grade 9) according to the Nottingham derivation of the Scarff Bloom Richardson system, in which grades 4 to 5 are low, 6 to 7 are medium, and 8 to 9 are high [34]. The estrogen and progesterone receptor status of the tumors was also provided by the Manitoba Breast Tumor Bank. No data were available for the human epidermal growth factor receptor 2 (HER2) status of these samples. Each section consisted of 40–70 % invasive tissue, with the remainder of the tissue being composed of stroma and fat. Samples were accessed and handled according to approved ethics committee guidelines at both the University of Saskatchewan and the Manitoba Tumor Bank (ethics approval number 14-37).

CREB3L1 protein expression in breast cancer cell lines was quantified by western blot analysis as previously described [35]. Briefly, SDS-PAGE was performed using 50 μg total protein, unless otherwise stated, as determined by Lowry (Sigma Aldrich, Oakville, ON, Canada TP0300). Samples were transferred to nitrocellulose and were probed with CREB3L1 (11235-2-AP from Protein Tech, Rosemont, IL, USA; rabbit, 1:500) or ß-actin (C-4 from Santa Cruz Biotechnology, Dallas, TX, USA; mouse, 1:500) primary antibodies, followed by infrared 680 nm or 800 nm dye-tagged secondary antibodies (LI-COR Biosciences, Lincoln, NE, USA; 200 ng/ml). Blots were imaged with the Odyssey Infrared Imaging System (LI-COR Biosciences, Lincoln, NE, USA), quantified, normalized to a ß-actin loading control, and reported relative to that in MDA-kB2 cells.

DNA was extracted from breast tumor sections and cell cultures using the QIAamp DNA Mini Kit (Qiagen, Toronto, ON, Canada). Sodium bisulfite treatment was performed using the EpiTect Bisulfite kit (Qiagen, Toronto, ON, Canada) to convert unmethylated cytosine residues to uracil according to the supplier’s instructions. CpGenome Human Methylated DNA Standard, (Cedarlane, Burlington, ON, Canada) and Epitect Control unmethylated DNA (Qiagen, Toronto, ON, Canada) were used as positive and negative controls, respectively.

Primers were designed using Primer3 software [36] to amplify (from bisulfite-treated DNA) two overlapping fragments of CREB3L1 spanning base pair −492 to +290 relative to the transcription start site (Additional file 1: Table S1). Our preliminary data indicated that DNA methylation was concentrated within the beginning of the coding region, thus, a set of primers that amplify a fragment from −51 to +258 to target this methylation-rich region was used. PCR was performed with 100 ng bisulfite-treated DNA in a 50-μl reaction with the TaKaRa EpiTaq HS kit (Cedarlane, Burlington, ON, Canada) according to the supplier’s instructions. The thermocycling protocol consisted of 45 cycles of 10 seconds at 98 °C, 30 seconds at annealing temperature (Additional file 1: Table S1), and 1 minute at 72 °C. QIAquick PCR Purification Kit (Qiagen, Toronto, ON, Canada) was used to purify PCR products prior to sequencing. Sanger sequencing was performed by the Plant Biotechnology Institute (Saskatoon, SK, Canada) and results were visualized with MacVector version 12.5 software (MacVector, Inc., Apex, NC, USA). Only sequences containing efficient C to T conversions, indicative of effective sodium bisulfite conversion at the non-CpG sites (that would not be methylated) were used for methylation analyses. Sequences were assessed for the presence of methylated cytosine residues at CpG dinucleotide motifs by using a qualitative assessment at each possible methylated position. Methylation was scored as low, but present and given a value of 1, if the C peak was above background noise even if some T was also present. Methylation was scored as high and given a value of 2, if the C peak was the tallest peak observed at that position. Initially sequences were analyzed from −429 to +259, relative to the translational start site, a region that includes 60 CpG sites. As most methylations were concentrated between −15 and +259, subsequent analyses focused on these 24 CpG sites.

Total RNA was extracted from cell lines using the RNeasy kit (Qiagen, Toronto, ON, Canada) and from breast tumor sections using the PicoPure RNA Isolation kit (Life Technologies). RNA was reverse-transcribed to cDNA using Superscript II Reverse Transcriptase and oligo-dT primers (Invitrogen) according to the supplier’s instructions. CREB3L1 expression was measured by quantitative real-time PCR performed using TaqMan probes (assay ID Hs00999642_m1, Life Technologies, Waltham, MA, USA) and TaqMan Gene Expression Master Mix (Life Technologies, Waltham, MA, USA) according to manufacturer’s protocols. The sequence of these primers is proprietary, but it amplifies a 103-bp fragment at the junction of exons 8 and 9 such that it will only detect the full-length transcript. Relative expression was calculated using expression of glyceraldehyde-3-phosphate dehydrogenase (GAPDH) (assay ID Hs99999905_m1, Life Technologies, Waltham, MA, USA) as a reference gene. Samples were analyzed in triplicate per reaction using the StepOnePlus Real-Time PCR System (Applied Biosystems, Waltham, MA, USA). Results are the mean of two independent reactions and reported as the relative change in expression compared to the MDA-kB2 cell line (for the cell line analysis), or control normal breast tissue sample (Manitoba Breast Tumor Bank, Winnipeg, MB, Canada) for the tumor sample analysis. Where samples were available, the HER2 status of the tumor samples was determined by qPCR as above, but using HER2-specific TaqMan probes (assay ID Hs1001580_m1, Life Technologies, Waltham, MA, USA), relative to a GAPDH reference gene. HER2 status was reported as compared to the MDA-kB2 cell line (HER2-positive, but not amplified) as follows (negative <1, positive = 1–9, amplified >9).

CREB3L1 gene copy number was assessed using a digital droplet PCR assay [37]. The number of target fragments in the original sample is calculated using a Poisson distribution and the copy number of the gene of interest is calculated by normalizing it to a reference gene (AP3B1) to adjust for global ploidy changes [38].

DNA extracted from human breast tumors was digested with HaeIII (New England Biolabs, Whitby, ON, Canada) prior to analysis, as per supplier’s instructions. To measure CREB3L1 gene copy number, custom-designed CREB3L1 primers and FAM-labeled TaqMan probe (forward 5′- GATCCAGCTTCCTGGACTTG-3′, reverse 5′-GTAAGATGAAGGGTCTCCGTTC-3′, probe 5′-ACGAGTCGGACTTCCTCAACAATGC-3′; Bio-Rad, Mississaugo, ON, Canada) were combined in a duplex reaction with pre-validated AP3B1 primers and HEX-labeled TaqMan probe (assay ID dHsaCP1000483, Bio-Rad, Mississaugo, ON, Canada) and ddPCR Supermix for Probes (Bio-Rad, Mississaugo, ON, Canada). Droplets were generated using the QX100 Droplet Generator (Bio-Rad, Mississaugo, ON, Canada) and thermocycled to the completion of the following protocol: 10 minutes at 95 °C; 40 cycles of 30 seconds at 94 °C and 1 minute at 60 °C; 10 minutes at 98 °C. Droplets were analyzed using the QX100 Droplet Reader (Bio-Rad, Mississaugo, ON, Canada) and copy numbers were computed with QuantaSoft software (Bio-Rad, Mississaugo, ON, Canada) after normalization to AP3B1. Results are reported as the average copy number of CREB3L1 genes per cell.

Immunohistochemical staining of CREB3L1 protein was performed on formalin-fixed paraffin-embedded human breast tumor samples obtained from the Manitoba Tumor Bank, according to the manufacturer’s instructions using Pierce Peroxidase Detection Kit (Thermo Scientific, Burlington, ON, Canada). Briefly, sections were de-paraffinized and heated in citrate buffer (pH = 6), followed by an overnight incubation at 4 °C with a rabbit polyclonal anti-CREB3L1 antibody (1:100; Protein Tech, Rosemont, IL, USA 11235-2-AP). Subsequently, slides were incubated with a horseradish peroxidase goat anti-rabbit secondary antibody (1:1000; Abcam, Toronto, ON, Canada ab6721) for 2 hours at room temperature and reacted with 3,3′-Diaminobenzidine (DAB) for 15 minutes. Sections were counterstained with hematoxylin. CREB3L1 staining was evaluated by a pathologist using two criteria, staining intensity (absent = 0, weak = 1, moderate = 2, strong = 3, very strong = 4) and % cells staining positive (0–5 % = 0, 6–49 % = 1, 50–69 % = 2, 70–89 % = 3, 90–100 % = 4). Scores were added together and described as little or no CREB3L1 staining (combined score 0–1), low (scores 2–3), medium (scores 4–5) and high (scores 6–8) CREB3L1 expression, analogous to the Allred system [39]. In addition, the subcellular location of CREB3L1 was evaluated as nuclear or cytoplasmic.

Publically available RNA-Seq Version 2 containing normalized gene expression datasets for 24 different cancer types were downloaded from the online database, The Cancer Genome Atlas (TCGA; http://​tcga-data.​nci.​nih.​gov). These data contained expression profile, clinical information, tumor stage, and immunohistochemical (IHC) results for the determination of breast cancer subtype for each breast tissue sample. The RNA-seq by expectation-maximization (RSEM) algorithm-normalized gene expression profile was downloaded from TCGA. Although the microarray-based dataset is also available in this database, to exclude problems that could be caused by combining different platforms, measurement types, and normalization procedures, we used only the RNAseqV2 dataset for all our gene expression analyses. To identify different subtypes of breast cancer, we used the IHC annotations available within our downloaded TCGA dataset. This approach was further verified by an additional independent analysis, where the TNBC population was defined as the group of samples with the lowest 10 % of estrogen receptor (ER), progesterone receptor (PR) and HER2 expression. Analysis of these populations confirmed that the same patients matched the IHC category. Similarly, for stage-specific classification, the annotation of each patient from the downloaded TCGA dataset was used. The subtype classifications or stage-specific classifications were analyzed using python scripts with pylab and the scipy stat built-in libraries to generate the graphs showing relationships with methylation and expression. The non-parametric Mann–Whitney U test was used to compare two groups. The methylation data were also downloaded from TCGA (Human Methylation 450 K data). Methylation probes were deconvolved for every single region across the promoter, and the intronic and exonic regions. Normalized values were downloaded onto Gene-E to generate the heatmap. Methylation data of either selected regions as identified in Li et al. [40] or as pre-selected by cBioPortal were used to generate correlation analyses between methylation and gene expression.

The relationship between CREB3L1 methylation and expression (Fig. 6c) was analyzed using cBioPortal (http://​www.​cbioportal.​org/​index.​do) [41, 42]. The dataset analyzed was the breast invasive carcinoma (TCGA Provisional) containing 737 cases, and was accessed on 12 February 2015. Kaplan–Meier survival analysis was carried out using KM-plotter (http://​kmplot.​com/​analysis/​) [43]. Gene expression data and relapse-free survival information were downloaded from Gene Expression Omnibus (GEO) (Affymetrix microarrays only), European Genome-phenome Archive (EGA) and TCGA. The database is handled by a PostgreSQL server, which integrates gene expression and clinical data simultaneously. To analyze the prognostic value of a particular gene, the patient samples were split into two groups at the median of the proposed biomarker. The two patient cohorts were compared using a Kaplan–Meier survival plot, and the hazard ratio with 95 % confidence intervals and log-rank p value were calculated. The database was accessed on 22 February 2015.

Spearman correlations were determined using free online software [44]. Statistical analyses were performed using SAS version 9.3 (SAS Institute Inc., Cary, NC, USA) software. Significance was set at p <0.05 and error reported as plus or minus the standard deviation (SD). The non-parametric Mann–Whitney U test was used to compare two groups. The Kruskal–Wallis test was used to compare four or more groups of sample data, using SPSS Statistics 23. Provided that significant differences were detected by the Kruskal–Wallis test, a post-hoc test was performed using pairwise comparisons. Survival analysis was performed using the Kaplan–Meier estimator with the non-parametric log-rank test to measure the equity of strata.

We analyzed CREB3L1 mRNA expression in a large panel of 40 breast cancer cell lines using quantitative real-time PCR (qPCR; Fig. 1a). In parallel, we measured CREB3L1 protein levels using a quantitative western blot analysis (Additional file 2: Figure S1 and Fig. 1a). CREB3L1 protein was low in four non-tumorigenic normal breast cell lines (184B5, MCF10A, MCF10F and MCF12A), and also in TNBC cell lines, as compared to luminal and HER2 amplified cell lines (Fig. 1b). A positive correlation between CREB3L1 protein and mRNA expression was observed (Spearman coefficient 0.315; p = 0.048) (Fig. 1c).

We analyzed the methylation status of the CREB3L1 promoter region, as one possible mechanism that could regulate CREB3L1 expression in breast cancer cells. A large CpG island (672 nucleotides) was identified in the promoter region of the CREB3L1 gene, using the University of California, Santa Cruz (UCSC) genome browser [45] (Additional file 3: Figure S2 and Additional file 4: Figure S3a). This CpG island contains 51 CpG sites and extends from −119 nucleotides upstream of the translational start site to 554 nucleotides into the coding region of the gene. In addition, there are 27 CpG sites between the transcriptional start site at −451 and the start of the CpG island (Additional file 3: Figure S2). We decided to analyze a fairly large region using sodium bisulphite-treated DNA and Sanger sequencing [46‐48]. This provided semiquantitative methylation data and importantly allowed us to assess a relatively large region including 60 CpG sites between −429 and +259 (Additional file 4: Figure S3b, c). We noted a strong preferential methylation at the 3′ end of this region at CpG sites 238 and 259. Samples with more extensive methylation had additional methylations extending towards the 5′ end of this region (Additional file 5: Table S2).

We found that more than half the breast cancer cell lines contained CREB3L1 methylated CpG sites; in particular, those of the TNBC subtype of breast cancer were highly methylated (Fig. 1a and d). In addition, four non-tumorigenic normal breast cell lines (184B5, MCF10A, MCF10F and MCF12A) all had many methylated sites in the region analyzed (Additional file 5: Table S2 and Fig. 1d). There was an inverse correlation between CREB3L1 DNA methylation within this region and CREB3L1 mRNA expression (Fig. 1e; Spearman correlation −0.381, p = 0.015). These results suggest that methylation of the CpG sites in the CREB3L1 gene may in some cases negatively regulate CREB3L1 mRNA expression, particularly in TNBC cell lines.

To investigate the role of epigenetic mechanisms in the regulation of CREB3L1 expression, the impact of the DNA methyltransferase inhibitor, 5-aza-2′-deoxycytidine (DAC), and the histone deacetylase inhibitor, TSA, were tested [49‐51]. BT20, HCC1806 and MDA-MB-468 human breast cancer cells were treated with DAC and/or TSA and CREB3L1 mRNA levels were analyzed by qPCR, relative to a GAPDH-specific control (Fig. 2a). TSA strongly induced CREB3L1 mRNA levels in BT20 cells, whereas both TSA and DAC induced CREB3L1 mRNA in HCC1806 and MDA-MB-468 cells. Methylation analysis of CREB3L1 DNA in these samples showed little or no change in BT20 cells, a small reduction in the HCC1806 cells, and a larger reduction in the MDA-MB-468 cells in response to DAC treatment (Fig. 2b). These results were consistent with the effects of DAC (±TSA) on CREB3L1 mRNA levels, suggesting that in BT20 cells, CREB3L1 is not regulated significantly by DNA methylation, but strongly by histone acetylation. In contrast, CREB3L1 mRNA expression in MDA-MB-468, and to a lesser extent HCC1806 cells, are regulated by both DNA methylation and histone acetylation.

Parallel samples were assessed for CREB3L1 protein levels by CREB3L1 immunoblotting (Fig. 2c) and showed only modest increases upon TSA and/or DAC treatment. As CREB3L1 protein levels have been reported to be downregulated via constitutive ubiquitination and proteasomal degradation in mouse embryo fibroblasts and C6 glioma cells [52], the TSA and/or DAC treatments were repeated including the proteasomal inhibitor, MG132. The presence of MG132 increased CREB3L1 protein levels more robustly (Fig. 2c). These results suggest that both epigenetic and post-translational mechanisms can contribute to reduced CREB3L1 in human breast cancer cells.

We further extended these studies from human breast cancer cell lines to human breast cancer tissues. Tumor samples included the associated data for ER and PR status, but the HER2 status was unknown. Tumor grade was provided for each sample and was determined using the Nottingham derivation of the Scarff Bloom Richardson system, where grades 4 to 5 are low, 6 to 7 are medium, and 8 to 9 are high [34]. An IHC analysis was carried out to characterize CREB3L1 protein expression and subcellular localization in 97 human breast tumor sections (Fig. 3). Of the 97 samples stained, 56 (58 %) showed little or no CREB3L1 staining (Fig. 3a–b). The samples with CREB3L1 protein staining were further divided into low (n = 9; Fig. 3c–d), medium (n = 14; Fig. 3e–f) and high (n = 18; Fig. 3g–h) based on the intensity and frequency of CREB3L1 protein expression.

As a stress-activated protein, CREB3L1 resides within the cytoplasm as an endoplasmic reticulum transmembrane protein, which can be processed after trafficking to the Golgi to release an active transcription factor that translocates into the nucleus. As has been noted in bladder cancer tumors [20], the 41 breast tumor samples that expressed CREB3L1 protein showed two main CREB3L1 localization patterns. Approximately half (n = 19) had predominantly nuclear CREB3L1 protein localization (Fig. 3c–f), whereas the other half (n = 22) had intense, mainly cytoplasmic staining (Fig. 3g–h). Importantly, the low-grade breast tumors more frequently had nuclear CREB3L1 protein, in contrast to the high-grade breast tumors in which CREB3L1 was cytoplasmic (p = 0.003, Mann–Whitney U test) (Fig. 3j). These results suggest that when CREB3L1 protein is expressed in breast tumors, the high-grade tumors have lost the nuclear localization of this transcription factor.

To assess whether there was any relationship between different tumor subtypes and either overall CREB3L1 protein expression and/or subcellular localization, we needed to determine their HER2 status. Where sufficient tumor samples were available, we determined the HER2 status using qPCR as detailed in “Methods”. This allowed us to group the tumor samples into molecular subtypes defined as: luminal (ER+ and/or PR+, ± HER2 but not amplified levels), HER2 (HER2 amplified), and TNBC (ER–, PR–, HER2–). In the absence of Ki67 data the luminal subtype was not further stratified into luminal A and luminal B. The amount of some samples was too small to determine the HER2 status, and these were grouped as unknown. There were no TNBC samples in our human tumor specimens, although several of the unknown samples were negative for ER and PR, but unknown for HER2 status.

For each subtype there were similar numbers of samples with and without CREB3L1 staining (Additional file 6: Figure S4a). Although the sample size within each group was small, there was a larger proportion of HER2 amplified tumor samples with cytoplasmic CREB3L1 protein, rather than nuclear CREB3L1 (Additional file 6: Figure S4b). We also compared the distribution of breast cancer subtypes by their nuclear and cytoplasmic localization and levels of CREB3L1 protein expression (Fig. 4a). Luminal breast cancers were more frequently observed to have low-medium levels of CREB3L1, effectively localized to the nucleus, or high levels of CREB3L1 within the cytoplasm. In contrast, HER2 amplified breast cancers contained mainly high levels of cytoplasmic CREB3L1 (Fig. 4a). In parallel, luminal breast cancers with nuclear CREB3L1 were most frequently of low-medium grade, whereas those with cytoplasmic CREB3L1 were often high-grade tumors (Fig. 4b). This was not observed for the HER2 amplified breast tumors. These results suggest that in HER2 amplified breast cancers about half express CREB3L1, and the CREB3L1 is primarily localized in the cytoplasm where its transcriptional regulation would not be active. Further, in luminal breast cancers, again about half express CREB3L1, that in lower grade tumors is typically nuclear, where it could regulate transcriptional targets, and in higher grade tumors is typically cytoplasmic, where it could not.

We measured CREB3L1 mRNA expression levels by qPCR in 213 tumor samples and found that low- and medium-grade tumors had increased CREB3L1 mRNA expression when compared to normal breast tissue samples (Fig. 5a). In contrast, high-grade breast tumors had reduced CREB3L1 expression. The majority of grade 8 (51 %) and grade 9 (73 %) breast tumors lacked CREB3L1 mRNA expression (Fig. 5b). Overall, CREB3L1 mRNA expression was negatively correlated with tumor type (r = −0.325, p <0.00001) and tumor grade (r = −0.342, p <0.00001) (Table 1). In addition, CREB3L1 mRNA expression was positively correlated with age at diagnosis (r = 0.294, p <0.00001), estrogen receptor expression (r = 0.232, p <0.001) and weakly with progesterone receptor expression (r = 0.158, p <0.05). These results suggest that loss of CREB3L1 is more frequently seen in high-grade, more aggressive breast cancers that lack ER and PR expression.

Our breast cancer cell line data suggested that CREB3L1 expression can be regulated in part by DNA methylation. Therefore, we also analyzed CREB3L1 methylation in a large group of breast tumor samples using the same approach we described above for the breast cancer cell lines. Again, this provided qualitative methylation data for 60 CpG sites between −429 and +259 (Additional file 7: Table S3). These tumor samples generally showed less CREB3L1 DNA methylation than the cell lines had, and similar to the cell line data most methylation was clustered within the 3′ end of the region analyzed. Therefore, midway through our analysis we narrowed our focus to the 24 CpG sites between −15 and +259 for about half of these samples (Additional file 7: Table S3). In all we analyzed 201 tumor samples and found 35 contained CREB3L1 methylation. Tumor samples with fewer methylated sites showed site-specific methylation at the most 3′ end including sites 221, 238 and 259 (Additional file 7: Table S3). These results suggest a hierarchy with the sites at the 3′ end of this −15 to +259 nucleotide region being methylated preferentially to those upstream.

We observed one tumor that showed CREB3L1 DNA methylation and had relatively high CREB3L1 mRNA levels, prompting us to examine CREB3L1 gene copy number. According to the Catalogue of Somatic Mutations in Cancer (COSMIC) database, CREB3L1 is rarely mutated in cancers but shows reduced copy number in 18 % (140/782) of breast cancers. In our samples there were only a few breast tumors with alterations in CREB3L1 copy number; 9 % had increased copy number (3–5 copies), 4 % had a decreased copy number of 1 and 86 % had a normal copy number of 2. The tumor with CREB3L1 DNA methylation but high CREB3L1 mRNA levels had a normal gene copy number (data not shown).

We observed that CREB3L1 DNA methylation was found in some samples within all tumor grades (Fig. 5c). The high methylation frequency observed for grade 4 (3/12 = 25 %) and grade 9 (2/10 = 20 %) were likely significantly influenced by the relatively small sample sizes for these two groups. The sample sizes within the remaining tumor grades (5–8) were larger, making these data more robust. Overall, CREB3L1 methylation was weakly, positively correlated with tumor type (r = 0.150, p <0.05) and tumor grade (r = 0.146, p <0.05) (Table 2). In addition, CREB3L1 methylation was weakly, negatively correlated with age of diagnosis (r = −0.166, p <0.05), ER expression (r = −0.168, p <0.05) and PR expression (r = −0.143, p <0.05). Our data also showed weak negative correlation between CREB3L1 mRNA expression and CREB3L1 methylation (r = −0.149, p <0.036) (Fig. 5d). These results suggest that the majority of the breast tumor samples with DNA methylation had reduced CREB3L1 mRNA expression, implicating methylation as a mechanism of CREB3L1 gene silencing.

We then expanded our analysis of CREB3L1 mRNA expression and methylation to the large TCGA dataset. CREB3L1 mRNA levels were increased in breast tumors, particularly for early-stage tumors (Fig. 6a). Similar to what we had observed for our smaller tumor dataset, CREB3L1 mRNA expression was increased in luminal and HER2 breast cancers but was low in TNBC (Fig. 6b). CREB3L1 mRNA expression also inversely correlated with CREB3L1 DNA methylation as noted above (Fig. 6c, cBioPortal: r = −0.464, p <0.0001). TNBCs were found to have both the lowest CREB3L1 mRNA levels and the highest CREB3L1 DNA methylation (Fig. 6d, p <0.0001), suggesting the low CREB3L1 often observed in TNBC cells may in part be due to increased methylation.

The methylation analysis carried out for these samples included assessing 26 50- nucleotide regions, each containing one or more CpG sites within the CREB3L1 gene (Additional file 8: Table S4). Five of these regions did not contain any methylation, and one had methylation but it was unchanged in normal and tumor breast samples. The remaining 20 methylation regions were differentially methylated between normal breast and breast tumor samples (Fig. 6e). We assigned each of these 20 regions a number from 1–20 for easy reference and show the location of each region that was analyzed, relative to the overall gene structure for CREB3L1 (Fig. 6f) and their precise CpG-containing sequences for those in the promoter region (Additional file 3: Figure S2). We determined the relative methylation for each region in normal versus breast tumor samples to assess which regions might be involved in regulating CREB3L1 expression changes (Additional file 9: Figure S5a, b). Many of the tested regions did not change their methylation status appreciably, including those in the 5′ UTR (numbers 14, 1, 4, 6, 5, 7, and 8), some within the 20 kb intron 1 (numbers 15, 17, and 18), those within a second CpG island within intron 1 (numbers 10, 11, 13, and 12), or within intron 3 (number 9). We focused on the remaining five regions (numbers 2, 3, 16, 19, and 20). We found that methylation of two regions (numbers 2 and 3), within the first CpG island, negatively correlated with CREB3L1 mRNA expression (Additional file 9: Figure S5c), and they were less methylated in breast tumor samples (Additional file 9: Figure S5a). The remaining three regions (numbers 16, 19, and 20) are all located in a shore near the second CpG island (Fig. 6f). Methylation in each of these regions positively correlated with CREB3L1 mRNA expression (Additional file 9: Figure S5d) and all were significantly more methylated in breast tumor samples compared to normal breast tissue (Additional file 9: Figure S5b). The inverse relationship between methylation at regions number 2 and 3, coupled with the direct relationship between methylation at regions number 16, 19, and 20 raises the possibility that complex methylation-dependent changes at several regions within the CREB3L1 gene (Additional file 10: Figure S6) could contribute to the regulation of CREB3L1 expression (Fig. 6b). When subdivided into low and high methylation, as compared to the median methylation of the normal breast samples, all five regions showed a significant difference in the relative CREB3L1 mRNA expression for the two groups (Additional file 11: Figure S7). Region number 2 showed the largest significant difference in methylation between normal breast and breast tumors, suggesting that changes in methylation in this region may have the largest influence on CREB3L1 mRNA expression. Region number 2 contains two CpG sites at nucleotides 259 and 238 (Additional file 3: Figure S2), the most frequently methylated within the breast tumor cell lines and breast tumor samples that we analyzed within our datasets (Additional file 5: Tables S2 and Additional file 7: Table S3).

Kaplan–Meier survival analysis was performed using KM-plotter [43] with gene expression data and relapse-free survival information accessed from several sources including GEO (Affymetrix microarrays only), EGA and TCGA (Fig. 7). When all breast cancers were analyzed as a group there was a shorter relapse-ree survival time for patients whose tumors contained low levels of CREB3L1 mRNA expression (hazard ratio (HR) = 1.27; p <0.0001) (Fig. 7a). When breast cancers were subdivided into different molecular classifications, CREB3L1 mRNA expression did not significantly influence the relapse-free survival time for luminal B (ER+ and/or PR+, HER+ or HER2– with high Ki67) (Fig. 7c) and HER2 (Fig. 7d) breast cancer patients. In contrast, both luminal A (ER+ and/or PR+, HER2–, low Ki67) (Fig. 7b) and triple negative (Fig. 7e) breast cancer patients with low CREB3L1 mRNA expression had higher HRs (1.49 and 1.33, respectively) that were statistically significant (p <0.00001 and p <0.05, respectively). These results strongly implicate the loss of CREB3L1 expression as an indicator of poor prognosis in breast cancers that are HER2–, such as luminal A and triple negative.

Using the TCGA database, we determined if CREB3L1 mRNA expression was differentially regulated between normal tissues and the corresponding tumor tissues (Fig. 8). Several tumor tissues had little or no change in CREB3L1 mRNA expression as compared to normal tissue, including esophageal cancer, glioblastoma multiforme, head and neck squamous cell carcinoma, lung adenocarcinoma, sarcoma, papillary thyroid carcinoma, thymoma and uterine corpus endometrial carcinoma. A second group of cancers had significantly higher CREB3L1 mRNA expression than their normal counterparts, including breast carcinoma (p = 6.5 × 10⁻³¹), prostate adenocarcinoma (p = 1.5 × 10⁻¹¹), chromophobe renal cell carcinoma (p = 1.9 × 10⁻⁹), cholangiocarcinoma (p = 5.5 × 10⁻⁴), stomach adenocarcinoma (p = 0.0025), liver hepatocellular carcinoma (p = 0.0027) and pancreatic ductal carcinoma (p = 0.019). Although these cancer types (including breast cancer) typically show an increase in CREB3L1 expression, patients with tumors expressing low levels of CREB3L1 may experience a poorer prognosis, as we found for breast cancer (Fig. 7a). A third group of cancers had significantly lower CREB3L1 mRNA expression than the corresponding normal tissues, including: lung squamous cell carcinoma (p = 1.7 × 10⁻¹⁹), clear cell kidney carcinoma (p = 1.2 × 10⁻¹⁰), papillary kidney carcinoma (p = 1.2 × 10⁻¹⁰), bladder urothelial carcinoma (p = 2.8 × 10⁻⁶), colon adenocarcinoma (p = 0.00010), pheochromocytoma and paraganglioma (p = 0.0029), rectal adenocarcinoma (p = 0.0057), cutaneous melanoma (p = 0.043) and cervical squamous cell carcinoma (p = 0.046). Thus, the loss of CREB3L1 is a frequent occurrence and could play an important role in tumor progression and metastasis in many cancer types.