Identification of PADI2 as a potential breast cancer biomarker and therapeutic target

The MCF10AT cell line series (MCF10A, MCF10AT1kC1.2, MCF10DCIS.com, and MCF10CA1aC1.1) was obtained from Dr. Fred Miller (Barbara Ann Karmanos Cancer Institute, Detroit, MI, USA). This biological system has been extensively reviewed [16, 17] and culture conditions described [18‐20]. The MCF7, BT-474, SK-BR-3, and MDA-MB-231 cell lines were from obtained from ATCC (Manassas, VA, USA) and cultured according to manufacturer’s directions. All cells were maintained in a humidified atmosphere of 5% CO₂ at 37°C. For the experimental treatment of cell lines with Cl-amidine, cells were seeded in 6-well plates (2 × 10⁴) and collected by trypsinization 5d post-treatment. Counts were performed using a Coulter counter (Beckman Coulter, Fullerton, CA, USA) and are represented as mean fold difference in cell number after treatment. Cl-amidine was synthesized as previously described [21].

Tissues from the MMTV-neu mouse were a generous gift from Dr. Robert S. Weiss, Cornell University, and the MMTV-Wnt-1 hyperplastic mammary glands and tumors were a gift of Dr. Louise R. Howe, Weill Cornell Medical College. MCF10DCIS xenograft tumors were generated by injecting 1 × 10⁶ cells in 0.1 mL Matrigel (1:1) (BD Biosciences, San Jose, CA, USA) subcutaneously near the nipple of gland #3 in 6-week old female nude (nu/nu) mice (Taconic, Germantown, NY, USA). When the tumors reached ~200 mm³, intraperitoneal injections of Cl-amidine (50 mg/kg/day) or vehicle control (PBS) were initiated and carried out for 14 days. Tumor volume was calculated by the formula: (mm³) = (d² × D)/2, where “d” and “D” are the shortest and longest diameters of the tumor, respectively. Tumor volume was measured weekly by digital caliper, and the differences between tumor volumes were evaluated by the non-parametric Mann–Whitney–Wilcoxon (MWW) test. Results are reported as mean ± SD. After 14 days, tumors were removed and either snap-frozen, placed in RNAlater (Qiagen Inc., Valencia, CA, USA), or added to 10% buffered formalin. Seven mice per group were used for each treatment. All mouse experiments were reviewed and approved by the Institutional Animal Care and Use Committees (IACUC) at Cornell University. Multicellular tumor spheroids were generated using the liquid overlay technique as previously described [22‐24]. The spheroids were allowed to form over 48h and maintained up to 6–10 days for morphological analysis, then collected, rinsed with phosphate buffered saline, and fixed in 10% buffered formalin.

Cell lines were assayed for PADI activity as previously described [25, 26]. Briefly, citrulline levels were determined using BAEE (α-N-benzoyl-L-arginine ethyl ester) as a substrate. After incubating lysates for 1h at 50°C with BAEE substrate mixture, the reaction was stopped by the addition of perchloric acid. The perchloric acid-soluble fraction was subjected to a colorimetric reaction with citrulline used as a standard and absorbance measured at 464 nm.

IHC and IF experiments were carried out using a standard protocol as previously described [27]. Primary antibodies are as follows: anti-PADI2 1:100 (ProteinTech, Chicago, IL, USA), anti-ERBB2 (A0485) 1:100 (Dako, Carpentaria, CA, USA), anti-Cytokeratin 1:100 (Dako), and anti-p63 1:100 (Abcam, Cambridge, MA, USA). Sections prepared for IHC were incubated in DAB chromagen solution (Vector Laboratories, Burlingame, CA, USA) according to the manufacturer’s protocol, washed, and then counterstained with hematoxylin. The IF slides were incubated in streptavidin conjugated-488 (Invitrogen, Carlsbad, CA, USA), washed, and then mounted using Vectashield containing DAPI (Vector Laboratories). Negative controls for both IHC and IF experiments were either rabbit or mouse IgG antibody at the appropriate concentrations. Tumor sections were examined for general morphological differences after hematoxylin and eosin (H&E) staining. Basement membrane integrity was determined using periodic acid-Schiff (PAS) stained slides, and was scored by SM on a scale of 0–3: 0- continuous with no breaching, 1- a few small interruptions, 2- several interruptions with breaching by tumor cells, 3- extensive loss of basement membrane with invasion of tumor cells over the breached area; observations were performed under 10X magnification.

Immunoblotting was carried out as previously described [27]. Primary antibodies were incubated overnight at 4°C using the following concentrations: anti-PADI2 1:1000 (ProteinTech) and anti-ErbB2 1:5000 (Dako). To confirm equal protein loading, membranes were stripped and re-probed with anti-β-actin 1:5000 (Abcam).

RNA was purified using the Qiagen RNAeasy kit, including on-column DNAse treatment to remove genomic DNA. The resulting RNA was reverse transcribed using the ABI High Capacity RNA to cDNA kit according to the manufacturer’s protocol (Applied Biosystems, Foster City, CA, USA). TaqMan Gene Expression Assays (ABI) for human PADI2 (Hs00247108_m1) and GAPDH (4352934E) were used for qRT-PCR. Data were analyzed by the 2 ^{-ΔΔ C(t)} method [28]. Data are shown as means ± SD from three independent experiments, and were separated using Student’s t-test. For the analysis of cell cycle gene expression, cDNA was synthesized (RT² First Strand Kit, Qiagen) and samples analyzed for expression of 84 genes involved in cell cycle regulation by RT² Profiler PCR Cell Cycle Array (PAHS-020A, Qiagen). For data analysis, the RT² Profiler PCR Array software package was used and statistical analyses performed (n = 3). This package uses ΔΔ C_T–based fold change calculations and the Student’s t-test to calculate two-tail, equal variance p-values.

Monolayers of MCF10DCIS and MCF10A cells were seeded into 25 cm² flasks (2 × 10⁶ cells) and treated with either Cl-amidine (200 μM or 400 μM), or 10μg/mL tunicamycin (apoptosis positive control). BT-474, SK-BR-3, and MDA-MB-231 cell lines were treated as previously described for MCF10DCIS and MCF10A; however, they were also treated with 100 μM Cl-amidine. Cells were harvested after 4d using Accutase (Innovative Cell Technologies, Inc, San Diego, CA, USA), fixed, then permeabilized, and blocked in FACS Buffer (0.1M Dulbecco’s phosphate buffered saline, 0.02% sodium azide, 1.0% bovine serum albumin, and 0.1% Triton X-100) containing 10% normal goat serum and stained (except the isotype controls) with rabbit anti-cleaved Caspase-3 antibody (Cell Signaling Technology, Inc, Danvers, MA, USA). Isotype controls were treated with normal rabbit IgG (Vector Laboratories) at 4 μg/mL. All samples were stained with secondary goat anti-rabbit IgG conjugated to Alexa-488 (Invitrogen) and DAPI (Invitrogen) according to the manufacturer’s instructions. Cells were analyzed on a FACS-Calibur (BD Biosciences) or a Gallios (Beckman Coulter) flow-cytometer and data analyzed for percent apoptotic cells (cleaved Caspase-3+) and cell cycle analysis with FlowJo software (TreeStar, Inc, Ashland, OR, USA). Data are shown as means ± SD from three independent experiments, and were separated using Student’s t-test.

Whole transcriptome shotgun sequencing (RNA-seq) was completed on breast cancer cell lines and expression analysis was performed with the ALEXA-seq software package as previously described [29]. Briefly, this approach comprises (i) creation of a database of expression and alternative expression sequence ‘features’ (genes, transcripts, exons, junctions, boundaries, introns, and intergenic sequences) based on Ensembl gene models, (ii) mapping of short paired-end sequence reads to these features, (iii) identification of features that are expressed above background noise while taking into account locus-by-locus noise. RNA-seq data was available for 57 lines (17 basal, 5 basal-NM, 6 claudin-low, 29 luminal). An average of 70.6 million (76bp paired-end) reads passed quality control per sample. Of these, 53.8 million reads mapped to the transcriptome on average, resulting in an average coverage of 48.2 across all known genes. Log2 transformed estimates of gene-level expression were extracted for analysis with corresponding expression status values indicating whether the genes were detected above background level.

In order to investigate PADI2 expression during tumor progression, we first utilized TaqMan quantitative real-time PCR (qRT-PCR) to measure PADI2 mRNA levels in cells from the MCF10AT tumor progression series (Figure 1a). As shown previously, these cell lines closely model the progression from normal (MCF10A), to hyperplastic (MCF10AT), to ductal carcinoma in situ (DCIS) with necrosis (MCF10DCIS.com), and finally to invasive/metastatic (MCF10CA1) breast cancer [16, 17, 30]. Results (Figure 1b) show that PADI2 mRNA expression is elevated in the transformed cell lines, with the highest levels found in the comedo-DCIS MCF10DCIS.com cell line (hereafter MCF10DCIS). Additionally, PADI2 protein levels closely correlated with PADI2 mRNA levels across these lines, with the highest levels of PADI2 protein observed in the MCF10DCIS line. Given the previous microarray studies correlating PADI2 expression with HER2/ERBB2, we also probed this cell line series with a well-characterized HER2/ERBB2 antibody and found that HER2/ERBB2 levels were also elevated in the transformed cell lines compared to the non-tumorigenic "normal" MCF10A line (Figure 1c). We also tested whether the increase in PADI2 expression correlated with PADI2 enzymatic activity, with results (Figure 1d) showing that citrulline levels are, in fact, highest in the MCF10DCIS cell line; therefore, indicating a strong correlation between increased PADI2 expression and enzymatic activity. While these cell lines have been previously classified as basal-like [31], both MCF10A and MCF10DCIS have been shown to possess bipotential progenitor properties [19, 32, 33]. Furthermore, the MCF10AT cells have been reported to show the same multipotent properties [34], but until recently, there has only been one other report showing that HER2/ERBB2 is upregulated in the transformed lines of this series [35]. These data suggest that PADI2 activity may play a role in mammary tumor progression and that PADI2-mediated citrullination may be particularly relevant to comedo-DCIS biology.

To test whether PADI2 displays a restricted expression pattern with respect to breast cancer subtype, we next investigated PADI2 mRNA and protein expression in cell lines representing four common breast cancer subtypes: MCF7 (luminal A), BT-474 (luminal B), SK-BR-3 (HER2/ERBB2+), and MDA-MB-231 (basal). At the protein level, PADI2 was observed in both BT-474 (ER+, PR+, HER2/ERBB2+) and SK-BR-3 (ER-, PR-, and HER2/ERBB2 overexpressing) cell lines. Interestingly, the comparison of PADI2 and HER2/ERBB2 protein levels across these four cell lines supports the hypothesis that these two proteins are coexpressed (Figure 2a). While the PADI2 protein expression is not observed in MCF7 cells in Figure 2a, a longer exposure of this blot finds that PADI2 is weakly expressed in these cells (Additional file 1, Figure S1a). Analysis of PADI2 transcript levels in these cell lines finds that, as expected, PADI2 mRNA is sharply elevated in the BT-474 line (Figure 2b), and is ~2 fold higher that that seen in the MCF10DCIS cells (Additional file 1, Figure S1b) when compared to MCF10A cells. To test whether PADI2 expression is elevated in HER2/ERBB2 expressing cells in vivo, we next measured PADI2 mRNA in normal murine mammary epithelium and in primary mammary tumors collected from MMTV-neu mice. Results indicate PADI2 mRNA levels are ~15-fold higher in the HER2/ERBB2 overexpressing tumors compared to normal mammary tissue from littermate controls (Figure 2c). The ~15-fold increase in PADI2 expression found in our study, compared to the ~4-fold increase found in the previous study [12], may simply reflect technical differences between the studies as we utilized TaqMan qRT-PCR compared to microarray analysis. We also investigated the level of PADI2 mRNA in MMTV-Wnt-1 mice, which is a basal mouse model of breast cancer [36‐38]. The MMTV-Wnt-1 model is unique in that it exhibits discrete steps in mammary tumorigenesis; the mammary glands are first hyperplastic, and then advance to invasive ductal carcinomas, finally culminating in fully malignant carcinomas that undergo metastasis [39]. Interestingly, we see that PADI2 levels are higher in the hyperplastic mammary glands (Figure 2c) when compared to normal mammary glands; however, the levels are less than those seen in the MMTV-neu tumors and are further reduced in the fully malignant MMTV-Wnt-1 tumors.

To strengthen the hypothesis that PADI2 is primarily expressed in luminal breast cancer cell lines and is coexpressed with HER2/ERBB2, we next investigated PADI2 mRNA levels by querying RNA-seq datasets collected from 57 breast cancer cell lines. A summary of PADI2 expression in these lines is shown in the Additional file 2, Figure S2, with the most significant difference (p = 3.59 × 10^-5) in PADI2 expression across subtypes being found when luminal lines were compared with all non-luminal subtypes (Figure 2d). We then quantified the correlation between PADI2 and HER2/ERBB2 expression across the 57 cell lines. Results show that the correlation between PADI2 and HER2/ERBB2 overexpression is highly significant across the luminal, basal-NM (non-malignant), and claudin-low cell lines (rho = 0.828, p = 2.2 × 10^-16) (Figure 3). Interestingly, a correlation between PADI2 and HER2/ERBB2 expression was not observed across the basal cell lines. In contrast, a significant anti-correlation was observed (rho = −0.495, p = 0.045), suggesting that the expression of these genes may be regulated by different mechanisms in these cell lines. Lastly, we queried the RNA-seq dataset to determine which genes were best correlated with HER2/ERBB2 and PADI2 expression in the luminal, basal-NM, and claudin-low lines to assess the relative strength of their coexpression. Only a single gene (CCL17) was as correlated with PADI2 as HER2/ERBB2 (rho = 0.832, p = 2.2 × 10^-16), and PADI2 represented the 13th most highly correlated gene with HER2/ERBB2 (Table 1), thus suggesting co-regulation between HER2/ERBB2 and PADI2.

To investigate whether PADI2 expression is important for breast cancer cell proliferation, we next tested whether the pharmacological inhibition of PADI2 activity negatively affects the growth of tumor cells in vitro. We utilized the small molecule inhibitor Cl-amidine for this study because we have previously shown that this drug binds irreversibly to the active site of PADIs, thereby blocking activity in vitro and in vivo[40]. Cl-amidine functions as a “pan-PADI” inhibitor as it blocks the activity of all active PADI family members (i.e. PADIs 1–4) with varying degrees of specificity [41]. Cultures from the MCF10AT cell line series were treated with 10 μM, 50 μM, or 200 μM of Cl-amidine, and the effects of the inhibitor on cell proliferation were quantified. Results show a dose-dependent decrease in the growth of all cell lines. Additionally, given that 200 μM Cl-amidine decreased the growth of MCF10DCIS cells by 75% (Figure 4a), this cell line appeared to be particularly affected by the inhibitor. Given the high level of PADI2 expression in the MCF10DCIS line, this finding suggests that PADI2 is likely playing an important role in the growth of MCF10DCIS cells. Importantly, while Cl-amidine also suppressed the growth of MCF10DCIS cells at lower concentrations, these doses did not inhibit the growth of the non-tumorigenic “normal” MCF10A line. These data suggest that Cl-amidine is not generally cytotoxic. In addition, citrulline levels in the Cl-amidine treated MCF10DCIS cells were significantly reduced, suggesting that the inhibitory effect of Cl-amidine was specifically due to the blockade of PADI activity (Figure 4b). In order to test the potential anti-tumor efficacy of Cl-amidine in a physiological model, we investigated the effects of this inhibitor on the growth of MCF10DCIS tumor spheroids. Spheroids grown from this cell line have been shown by others to form acinar-like structures that closely recapitulate the comedo-DCIS lesions that form in MCF10DCIS xenografts [18, 20, 42]. Results from our studies found that Cl-amidine treatment significantly reduces tumor spheroid diameter (Figures 4c). Representative images of the effects of Cl-amidine on the growth of MCF10DCIS monolayers and spheroids are shown in Figure 4d.

The observed effects of Cl-amidine on cell proliferation suggested that this drug might affect tumor growth by altering the expression of genes involved in cell cycle progression. To test this hypothesis, mRNA from the Cl-amidine treated and control MCF10DCIS cells was examined for the expression of cell cycle associated genes using the RT² Profiler PCR Cell Cycle Array via qRT-PCR. Using a threshold value of 2-fold expression change and a statistical significance of p < 0.05, we found that Cl-amidine affected the expression of a subset of genes (for the full unsorted list see Additional file 3, Table S1), with the top 10-upregulated and -downregulated genes presented in Table 2. Importantly, previous studies have shown that increased expression of GADD45α, the second most highly upregulated gene in our study, leads to cell cycle arrest and apoptosis in a range of cell types, including breast cancer cells [43]. This observation suggested that, in addition to affecting cell cycle gene expression (e.g. p21), Cl-amidine might also alter MCF10DCIS cell growth by inducing apoptosis. To test this hypothesis, we next treated MCF10A and MCF10DCIS cells with increasing concentrations of Cl-amidine for 4 days. Cells were fixed and labeled with anti-activated Caspase-3 antibody or DAPI, and then analyzed by flow-cytometry. Results show that Cl-amidine treatment significantly increased the percent of apoptotic MCF10DCIS cells in a dose-dependent manner (Figure 4e). In contrast, the MCF10A cells were largely unaffected. Furthermore, we also show that treatment of MCF10DCIS cells with Cl-amidine appears to induce cell cycle arrest in S-phase (Figure 4f). Lastly, we wanted to see whether the increase in apoptosis occurs earlier after treatment, so we tested the cells again following 2 days of treatment, but were unable to see any effect (Additional file 4, Figure S3a). However, this was not surprising, as the effects of Cl-amidine are most pronounced after 3 days of treatment (data not shown). Taken together, it appears that Cl-amidine treatment after 4 days leads to S-phase coupled apoptosis, which is an intrinsic mechanism that prevents DNA replication of a damaged genome in a mammalian cell [44]. We also tested the effects of Cl-amidine on HER2/ERBB2 overexpressing cell lines BT-474 and SK-BR-3. Again, we see a reduction in cell growth (Figure 5a) and an increase in apoptosis (Figure 5b) that is coupled to S-phase cell cycle arrest (Figure 5c) for both BT-474 and SK-BR-3. These results show that Cl-amidine is effective in inhibiting the growth of luminal-HER2/ERBB2+ cell lines, BT-474 and SK-BR-3, and agree with previously reported data on Cl-amidine inhibition of growth in MCF7 cells [8, 9, 11, 45]. We wanted to test whether there would be any effect on a basal cell line, and chose MDA-MB-231 for comparison. Surprisingly, we see an effect on both cell growth and apoptosis (Additional file 4, Figure S3b and c), albeit a smaller effect on apoptosis than we see in BT-474 and SK-BR-3. While this is interesting, and perhaps suggests the expression of a different PADI family member in this basal cell line, we have focused on PADI2 expressing cancers for this study, which are predominantly luminal and HER2/ERBB2 expressing. Taken together, these results suggest that Cl-amidine blocks the growth of MCF10DCIS cells by inducing cell cycle arrest and apoptosis. This prediction is supported by our previous finding that Cl-amidine can also drive apoptosis in lymphocytic cell lines in vitro[3]. Importantly, the lack of an apoptotic effect in MCF10A cells suggests that Cl-amidine may primarily target tumor cells for killing. Consistent with this possibility is the fact that Cl-amidine did not affect the growth of non-tumorigenic NIH3T3 cells and HL60 granulocytes [11].

Given that PADI2 expression is elevated in the MCF10DCIS cell line, we investigated PADI2 expression and localization in primary tumors derived from MCF10DCIS-injected mouse xenografts. Previous studies have shown that when MCF10DCIS cells are injected into the mammary fat-pad of immunodeficient nude (nu/nu) mice, tumors develop within 2–3 weeks. These tumors faithfully recapitulate the human comedo-DCIS condition, with the basement membrane limiting duct-like structure being comprised of an outer myoepithelial layer, an inner layer of luminal epithelial cells, and a central necrotic lumen [18, 19, 46, 47]. We chose to use subcutaneous injections instead of orthotopic or intraductal [48] methods, as previous work by Hu et al. showed that the progression and phenotype of the MCF10DCIS tumors grown subcutaneously in the mammary fat pad were highly similar to human high-grade comedo-DCIS tumors [19]. In our study, we found that PADI2 protein expression was restricted to the luminal epithelium of the duct-like structures in the MCF10DCIS xenografts, and was not observed in the stromal tissue or the necrotic core (Figure 6a, panel I and II). At the subcellular level, PADI2 appears to be expressed in both the cytoplasmic and nuclear compartments of luminal epithelial cells (Figure 6a, panel II). This observation supports our recent findings that PADI2 can be targeted to the nucleus of both human normal mammary tissue and breast cancer cells [49] and regulate gene activity via citrullination [49, 50].

Next, we examined whether the observed correlation between PADI2 and HER2/ERBB2 expression also occurred in vivo. We found that both HER2/ERBB2 and PADI2 were expressed within the luminal epithelium of MCF10DCIS tumors (Figure 6a, panel III and IV). Interestingly, a previous report by Behbod et. al. found low levels of HER2/ERBB2 in MCF10DCIS tumors that were grown intraductally. The disparity between this data and our data may be due to differences in the microenvironment. We then quantified PADI2 mRNA in the MCF10DCIS xenografts by qRT-PCR, and found that PADI2 levels were significantly higher in the tumors when compared to monolayer cultures (Figure 6b). We also carried out immunofluorescence (IF) analysis of these tumors to examine PADI2 intratumoral localization, and found that PADI2 protein expression appears entirely limited to cytokeratin-positive luminal epithelial cells (Figure 6c, panel I and III, and Additional file 5, Figure S4), while no detectable PADI2 signal was observed in the p63 positive myoepithelial cells (Figure 6c, panel IV and VI, and Additional file 5, Figure S4).

Given the inhibitory effects of Cl-amidine on MCF10-DCIS monolayer and spheroid growth, we next tested whether the treatment of mice with this inhibitor would suppress the growth of MCF10DCIS-derived tumors. For this study, mouse fat-pads were injected with MCF10DCIS cells (1 × 10⁶) and the tumors were allowed to establish and grow for ~2 weeks as described previously [18‐20, 46]. Mice were randomly assigned into treatment or control groups and administered daily intra-peritoneal (IP) injections of either Cl-amidine (50 mg/kg/day) or vehicle (PBS). Note, that the choice of dose and route of administration were based on the previous demonstration that Cl-amidine reduces disease severity in the murine collagen induced arthritis model of rheumatoid arthritis [5]. Treatment continued for 14 days, at which point the tumors were harvested. Results from our xenograft study show that Cl-amidine treatment (n = 7/group) caused a significant reduction in the size of the tumors (Figures 7a). Moreover, the analysis of tumor morphology by H&E and PAS staining shows that, while tumors from the sham-injected group displayed an advanced, potentially invasive, tumor phenotype (Figure 7b, panel II), tumors from the Cl-amidine treated group (Figure 7b, panel I) were much more benign in appearance. Furthermore, the basement membrane of Cl-amidine treated tumors remained largely intact (Figure 7c) and had considerably less membrane breaching and leukocyte infiltration compared to the control group. These findings suggest that PADI2 plays an important role in comedo-DCIS progression and that the inhibition of PADI activity can suppress tumor progression in vivo.