CD44 isoforms are heterogeneously expressed in breast cancer and correlate with tumor subtypes and cancer stem cell markers

The breast cell lines BT-474, HCC1428, HCC1937, MCF7, MCF10A, MDA-MB-231, MDA-MB-361, MDA-MB-436, SK-BR-3 and ZR-75-1 were obtained from American Type Culture Collection (ATCC, Mannanas, VA). JIMT-1 was purchased from the German Collection of Microorganisms and Cell Cultures (DSMZ, Braunschweig, Germany). L56Br-C1 was established at the Department of Oncology, Lund University [16] and PMC42 [17] was received through a generous gift from Dr. Anna Git at the Breast Cancer Functional Genomics Laboratory, Cancer Research UK, Cambridge Research Institute and Department of Oncology, University of Cambridge, UK. The HCC1937, MDA-MB-436, and L56Br-C1 cell lines are from BRCA1 germline mutation carriers [16].

All cell lines were cultured under adherent conditions in RPMI1640 (Invitrogen, Carlsbad, CA) supplemented with 1 mM Na-puruvate, non-essential amino acids, 50 U/ml penicillin, 50 ng/ml streptomycin (all from Invitrogen) and 10% fetal bovine serum (FBS, Hyclone, Logan, UT). MDA-MB-361 was cultured as above except for 20% concentration of FBS. MCF7, BT-474, MDA-MB-436 and L56Br-C1 were additionally supplemented with 10 μg/ml insulin (Invitrogen). HCC1937 was supplemented with 10 μg/ml insulin and 20 ng/ml epidermal growth factor (EGF, Invitrogen). MCF10A was supplemented with 10 μg/ml of insulin, 20 ng/ml EGF, 100 ng/ml cholera toxin (Sigma, St. Louis, CO) and 500 ng/ml hydrocortisone (Sigma). To enrich for stem cell properties, cells were also cultured as non-adherent multicellular spheres (mammospheres) in Mammary Epithelial Growth Medium (Cambrex, Walkersville) including hydrocortisone, insulin and GA-1000) [18]. The media was additionally supplemented with 20 ng/ml of EGF, 20 ng/ml basic fibroblast growth factor (R&D Systems, Minneapolis, MN) and B27 (Invitrogen). Mammospheres were passaged once a week by enzymatic dissociation with Accutase (Innovative Cell Technologies, Inc., San Diego, CA) followed by mechanical dissociation by pipetting.

Fresh frozen tumor tissue from 151 patients diagnosed with stage II primary breast cancer were obtained from the Southern Sweden Breast Cancer Group's tissue bank at the Department of Oncology, Lund University. These patients were treated with 2 or 5 years of adjuvant tamoxifen and were included in two previous randomized trials [19, 20]. Included in the study were also 36 tumors from patients with germline BRCA1 or BRCA2 mutations (32 and 4 tumors, respectively). The tissue microarray used has been described earlier [10]. The study was approved by the regional ethical committee at Lund University (reg. no. LU240-01 and 2009/658), waiving the requirement for informed consent for the study.

Estrogen (ER) and progesterone receptor (PgR) status and S-phase fraction were obtained earlier from enzyme immunoassay and DNA flow cytometry, respectively [19, 21]. Immunohistochemistry (IHC) on tissue microarrays (TMAs) was also previously used for staining of epidermal growth factor receptor (EGFR) (n = 69 tumors) and human epidermal growth factor receptor 2 (HER2) (n = 82 tumors) protein [22, 23], and for double-staining of CD44 (Clone 156-3C11) and CD24 protein (n = 80) [10]. The proportion of CD44⁺, CD24⁺ and CD44⁺/CD24^- tumor cells was used in scoring: 0 = 0% positive tumor cells, 1 = 1-10% positive cells, 2 = 11-50% positive cells, 3 = 51-75% positive cells, 4 = 76-100% positive cells. PTEN protein expression and PIK3CA mutational status were available for 110 and 107 of the tumors, respectively [24]. Microarray gene expression data was available for 157 of the tumors [25‐27] and tumors were subclassified according to Hu et al. [3] as described [10]. Samples with Pearson correlation < 0.2 to all centroids were considered unclassified.

ALDH1 protein expression was determined by IHC on TMA. Antigen retrieval was achieved by placing the slides in citrate buffer (Dako S1699) (DAKO, Glostrup, Denmark) at 125°C in a 2100 Retriever (PickCell Laboratories, Amsterdam, the Netherlands) for 5 minutes and ALDH1 was detected with a mouse monoclonal primary antibody (Clone 44, BD Biosciences, San Jose, CA, USA) followed by EnVision™ on an Autostainer (DAKO). Cytoplasmic staining was recorded as negative (< 1% positive tumor cells), weakly or strongly positive. In total 210 breast tumors were analyzed, of which 73 were also analyzed for CD44 isoforms.

Cells were washed once with phosphate-buffered saline and harvested using Accutase. Detached cells were labeled with fluorochrome-conjugated monoclonal antibodies obtained from BD Biosciences Pharmingen (San Diego, CA) against human CD44 (FITC; clone G44-26) and human CD24 (PE; clone ML5). Appropriate isotype controls were used to set the threshold for CD44 and CD24 positive cells. The labeled cells were analyzed on a FACSCalibur (BD Biosciences). Dead cells were excluded by staining with 7-aminoactinomycin D (Sigma-Aldrich, St. Louis, MO).

Approximately 100 mg of frozen tumor material was pulverized using a microdismembrator immediately followed by homogenization in Trizol reagent (Invitrogen, Carlsbad, CA) and total RNA was isolated according to manufacturer's instructions. A second round of purification was performed using the RNeasy kit (Qiagen, Hilden, Germany) according to manufacturer's instructions. The total RNA from breast cell lines was extracted using the RNeasy kit. The concentration of total RNA was measured by a ND-1000 NanoDrop spectrophotometer (NanoDrop Technologies, Wilmington, DE) and integrity of RNA was verified using the 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA).

350 ng RNA per sample was amplified and biotinylated using Illumina TotalPrep 96 RNA Amplification Kit (Ambion, Austin, TX). 750 ng cRNA per sample was hybridized to Illumina Human-12 v3 Expression Bead Chip (Illumina, San Diego, CA) using Whole-Genome Expression Direct Hybridisation kit (Illumina) and scanned with the Illumina BeadArray reader according to manufacturer's instructions. The mean signal of the data was uploaded to BASE2 [28]. The data was quantile normalized in BASE2 and then exported into MeV [29]. Log2 transformation and median center was performed across genes. Gene expression data was available for a panel of 24 cell lines including all 13 cell lines selected for q-RT-PCR analysis. Like Neve et al. [30] we discovered 3 groups when performing hierarchical clustering based on the most varying genes and we also assigned cell lines to luminal, basal A or basal B. 21 of our 24 cell lines were included in Neve et al. and we obtained a similar clustering pattern using our data [30].

Different mRNA transcripts of CD44 were obtained from The National Center for Biotechnology Information (NCBI) Reference Sequence (RefSeq) database. Inventoried exon-exon spanning TaqMan gene expression assays (Applied Biosystems, Foster City, CA) were used for detection of CD44; Hs00153304_m1 (CD44 total, NM_001001389.1, NM_001001390.1, NM_001001391.1, NM_001001392.1, NM_000610.3), Hs01081480_m1 (CD44v3-v10, NM_001001389.1), Hs01081475_m1 (CD44v8-v10, NM_001001390.1), Hs01081473_m1 (CD44S, NM_001001391.1) and Hs01075866_m1 (CD44v2-v10, NM_000610.3). The RefSeqs; NM_001202555.1, NM_001202556.1, NM_001202557.1 recently added to the NCBI database were not included in our study. Three different endogenous controls were used: GAPDH (4333764T), ACTB (Hs99999903_m1), and PUM1 (Hs00206469_m1) (Applied Biosystems). TaqMan Gene Expression Master Mix (Applied Biosystems) was used and all reactions were run in triplicates. A CAS1200 instrument and a Rotor-Gene instrument (Corbett Life Science, Sydney, Australia) was used for automated PCR setup of the q-RT-PCR reactions. The Rotor-Gene 6.0 software was used for calculations of average C_t-values for each sample and the data was exported to Microsoft Excel for further analysis. The average C_t-value for all three endogenous controls was calculated for each sample. To calculate the relative expression of the CD44 transcripts detected by each gene expression assay in respective sample the delta-delta C_t-method was used [31]. Either one of the analyzed tumors or cell lines were arbitrarily set as a calibrator. The calculated relative expression values were log2-transformed.

Cells were lysed in RIPA buffer containing protease inhibitor (Roche, Basel, Swizerland) and mixed with 2× NuPage sample buffer. Samples were run on Bis Tris NuPage Gel 4-12% and then transferred to PVDF membranes (Invitrogen). Membranes were blocked with 5% non-fat dry-milk in Tris-buffered saline with 0.1% Tween (TBST) overnight at 4°C. Membranes were incubated with CD44 antibody (Clone 156-3C11) for 2 hours at RT followed by horse-radish-peroxidase conjugated goat anti-rabbit IgG (GE Healthcare, Chalfont, St. Giles, UK) for 1 hour at RT in 5% dry-milk with TBST. Visualization was performed using ECL Plus (GE Healthcare).

Spearman correlation coefficient was used to assess the association between two continuous variables, Mann-Whitney test was used for two-group comparisons of continuous variables and Kruskal-Wallis test was used to compare expression of continuous variables in multiple subgroups. Fisher's exact test was used to analyze contingency tables. Cox regression analysis was used for analysis of time-to-event data and Kaplan-Meier plots to illustrate the results. Schoenfeld's test was used to test proportional hazards assumptions. All statistical analyses were carried out in Stata 10.0 (StataCorp LP, College Station, TX). All tests were two-sided and P-values < 0.05 were considered as significant.

Hierarchical clustering using Pearson correlation with average linkage was performed in MeV 4.6.1 [29]. Log2 ratios were adjusted by median centering of samples followed by median centering of genes before clustering.

To investigate the presence of different CD44 splice variants in breast cell lines we analyzed 13 cell lines by q-RT-PCR. The cell lines were assigned to luminal, basal A or basal B according to Neve et al [30]. The total expression of CD44 transcripts was first compared to flow cytometry data of CD44 and a relatively good agreement between mRNA and protein expression could be seen. However, the MDA-MB-361 cell line has 79.6% of CD44 positive cells despite a moderate expression level of total CD44 (Table 1). Possibly, the relatively high amount of synthesized CD44 protein in MDA-MB-361 can be explained by for example post-transcriptional modifications and translational regulation specific only for this cell line. Exon-exon spanning q-RT-PCR assays were further used to illustrate a very heterogeneous expression pattern of different alternatively spliced CD44 transcripts (Figure 1). In general, the basal A and B cell lines showed a much higher total expression of CD44 than the luminal cell lines. We also found that basal A and basal B cell lines had almost mutually exclusive expression of different CD44 transcripts. For instance, the MDA-MB-231 and HCC1937 cells had similar total CD44 protein and RNA expression levels, however, while this was primarily due to expression of CD44S in the former cells, this isoform was virtually undetectable in the latter cells. Of interest, PMC42 cells, described as having stem cell-like properties [17], showed exclusive expression of the CD44S isoform.

Four of the breast cancer cell lines (MDA-MB-231, MCF7, JIMT-1 and L56Br-C1) could be readily cultured under conditions where they form non-adherent mammospheres, a state thought to enrich for stem cell-like features [18]. Flow cytometric analysis showed that the proportion of CD44⁺/CD24^- cells in MDA-MB-231 was very high (> 95%) in both monolayer and non-adherent cultures while in MCF7 there was no clear increase in the initially very low CD44⁺/CD24^- cell proportion (< 0.3%). The CD44⁺/CD24^- proportion of JIMT-1 and L56Br-C1 cells increased from 45.5% to 82.6% and 1.2% to 7.2%, respectively, when grown as mammospheres compared to monolayers.

We found considerable and variable changes in the expression of CD44 isoforms during these experiments (Figure 2A). Expression of CD44S decreased in three of four cell lines, the CD44v3-v10 isoform was upregulated in all four cell lines investigated and the expression of CD44v2-v10 and CD44v8-v10 were higher in mammospheres as compared to adherent cultures for three of four cell lines. For instance, MDA-MB-231 cells dramatically decreased their expression of CD44S and increased CD44v2-v10 and CD44v3-v10 expression, while the L56Br-C1 cells showed the opposite pattern. Changes in the pattern of different CD44 protein variants were also indicated by Western blot for JIMT-1 (clone 156-3C11) (Figure 2B).

RNA expression of CD44 isoforms was investigated in a set of 187 primary breast tumors. As shown in Table 2, the CD44v3-v10 isoform was most strongly associated to the total CD44 mRNA expression, while CD44S showed the lowest correlation. A subset of these tumors has previously [10] been stained for CD44 (and CD24) by IHC. When comparing CD44 RNA and protein expression we found that all analyzed isoforms except CD44S were positively correlated to the protein level of CD44 (Table 2). Similar results were obtained when CD44 RNA expression levels were compared with the presence of cells with the putative cancer stem cell phenotype CD44⁺/CD24^- (Table 3).

We further analyzed the correlation of CD44 isoforms to expression of ALDH1, another suggested marker of normal and malignant stem cells [5]. ALDH1 staining was detected in 31% of all 210 tumors, with a strong staining in 9%. Strong ALDH1 staining was significantly associated with negative ER (P = 0.008) and PgR status (P = 0.015), strong HER2 expression (P = 0.025) and with a high S-phase fraction (P = 0.005). ALDH1 status did not correlate with presence of CD44⁺/CD24^- tumor cells. We had overlapping ALDH1 IHC and CD44 q-RT-PCR data for 73 tumors, 12 (16%) of which were strongly ALDH1 positive. Surprisingly, expression of CD44S was positively associated to ALDH1+ tumors (Table 3). This was confirmed by microarray ALDH1 expression data, which showed correlation to CD44S (Spearman ρ = 0.49, P < 0.00001; n = 131) but not to other CD44 isoforms or total CD44 RNA expression (Figure 3A-B).

The expression of different CD44 isoforms was significantly associated to certain patient and tumor characteristics (Table 4). A high expression of CD44v2-v10 was correlated to positive ER and PgR status, low S-phase fraction and postmenopausal age. High expression of CD44v3-v10 correlated to positive steroid receptor status and low proliferation, but also to negative/low HER2 status. Strong EGFR expression and negative/low HER2 status were correlated to expression of CD44v8-v10 but showed no association to steroid receptor status or proliferation rate. Interestingly, a correlation was found between the expression of CD44S and strong HER2 expression, as well as to smaller tumor size. Lymph node status could not be associated to any of the isoforms. We found a correlation of CD44v2-v10 expression to the presence of PIK3CA mutations. However, none of the CD44 isoforms were associated to PTEN status. 10-year overall survival was significantly better for patients with tumors showing high expression of CD44v2-v10 (HR = 0.64, 95% CI = 0.27-1.21, P = 0.007).

Gene expression microarray data was used to classify tumors into subtypes according to Hu et al. [3]. A Pearson correlation ≥ 0.2 to at least one of the five subtype gene expression centroids was found in 145 tumors, while the remaining tumors was considered as unclassified and excluded from further analysis. As expected from their predominant expression in ER/PgR positive and less proliferating tumors, the CD44v2-v10 and CD44v3-v10 isoforms were associated to the luminal A subtype and were expressed to lesser degree in basal-like tumors (Figure 4A-B). Specifically, CD44v2-v10 expression was significantly higher in luminal A compared to the basal-like (P = 0.0001), luminal B (P = 0.038), normal-like (P = 0.011) and HER2-enriched subtype (borderline significance; P = 0.051). A similar pattern was seen for CD44v3-v10 with higher expression in luminal A tumors compared to basal-like (P < 0.001), luminal B (P < 0.001), normal-like (P = 0.003) and HER2-enriched subtype tumors (P = 0.001). A different pattern was seen for CD44v8-v10, with highest expression in normal-like and basal-like tumors and lowest expression in luminal B (P = 0.018, as compared to normal-like) and HER2-enriched (P = 0.019) classified tumors (Figure 4C). The expression of CD44S varied less between subtypes but, as expected, showed highest median expression in the HER2-enriched subtype, being significant when comparing to luminal A tumors (P = 0.050) (Figure 4D). The total expression of CD44 was significantly higher in the luminal A subgroup compared to basal-like (P = 0.007), luminal B (P = 0.009) and HER2-enriched (P = 0.002) classified tumors while the normal-like subgroup showed borderline significance (P = 0.051) (Figure 4E).

The basal-like subtype showed a better 10-year overall survival for tumors with high expression of total CD44 (HR = 0.22, 95% CI = 0.06 to 0.79, P = 0.021) as well as CD44v2-v10, CD44v3-v10, CD44v8-v10 (HR = 0.44, 95% CI = 0.24 to 0.81, P = 0.008; HR = 0.29, 95% CI 0.14 to 0.61, P = 0.001; HR = 0.41, 95% CI 0.18 to 0.90, P = 0.026, respectively). Similar to basal-like tumors the luminal A subtype showed a better survival at high expression of total CD44 (HR = 0.20, 95% CI = 0.05 to 0.93, P = 0.04).

The expression levels of CD44 isoforms were used in hierarchical clustering to subdivide the 187 tumors into four main clusters (Cluster A-D; Figure 5). Strikingly, while Cluster A had high CD44v2-v10 and low CD44v8-v10 expression, the opposite pattern was seen in Cluster D. Furthermore, Cluster B tumors showed high expression of CD44v2-v10, CD44v3-v10 and low expression of CD44S while the reverse expression pattern was observed in Cluster C. The four clusters were significantly correlated to various biomarkers (Table 5) and to 10-year overall survival (log rank P = 0.05) (Figure 6). Cluster B tumors were associated with best overall survival and was characterized by mostly ER and PgR positive status, negative/low HER2 status, low proliferation and luminal A subtype, but also by higher PIK3CA mutation frequency. Cluster A tumors were mostly ER and PgR positive, but also more often strongly HER2 positive and of luminal B or HER2-enriched subtypes, as well as diagnosed at higher age. Cluster A showed a slightly worse prognosis compared to Cluster B (P = 0.17, HR = 1.7, 95% CI: 0.8 to 3.6). Clusters C and D showed enrichment for the basal-like subtype and were associated to higher proliferation (Figure 5), although Cluster C included tumors with high CD44S expression and of luminal B and HER2-enriched subtype. Cluster D was enriched for tumors of both basal- and normal-like subtypes characterized by high CD44v8-v10 and low CD44S expression. Both Cluster C and Cluster D showed significantly worse prognosis compared to Cluster B (P = 0.02, HR = 2.2, 95% CI = 1.1 to 4.2; P = 0.01, HR = 2.4, 95% CI = 1.2 to 4.6, respectively). Three tumors with high CD44 gene and 11p13 amplification were all assigned to cluster D and they all had a high expression of v8-v10. Presence of tumors with the CD44⁺/CD24^- phenotype varied significantly (P = 0.01), being more common in Cluster B and D (Figure 5). ALDH1+ status was found in all clusters, but less common in Cluster D. Multivariate survival analysis adjusted for age, node status and tumor size showed significantly worse outcome for Cluster C and D compared to Cluster B (HR = 2.1, 95% CI = 1.1 to 4.3, P = 0.03 respective HR = 2.1, 95% CI = 1.1 to 4.4, P = 0.04), respectively, but no significant difference between Cluster A and Cluster B.