Parity predisposes breasts to the oncogenic action of PAPP-A and activation of the collagen receptor DDR2

Histological slides were prepared as 4 μm formalin-fixed sections embedded in paraffin and stained with H&E by the Oncological Sciences Department Histology core facility at Icahn School of Medicine at Mount Sinai. For immunostaining of paraffin-embedded sections, the samples were deparaffinized in xylene and rehydrated in water. Antigen retrieval was performed using citrate buffer pH 6 for 30 min at 95 °C. Peroxidase blocking was completed using endogenous peroxidase solution (Dako Dual Endogenous Enzyme Block #S20003) at room temperature for 15 min. Antibody Diluent (MP Biomedicals Normal Antibody Diluent #980631) was used for serum blocking for 30 min at room temperature. All buffer washes between incubations were completed with TBS. Primary antibody (1:100 Snail, Cell Signaling Technology L70G2 #3895) incubation was performed overnight at 4 °C in an enclosed moist chamber. Biotinylated mouse antibody (Jackson ImmunoResearch #115-065-205) diluted 1:200 in TBS for 1 h at room temperature was used for the secondary antibody incubation. Washes between antibody incubations were done with TBS with .04% Tween. Streptavidin peroxidase (Vector Labs #SA-5004) was incubated for 30 min at room temperature, followed by a TBS with .04% Tween and TBS wash. The slides were developed using AEC (Vector Labs #SK-4200) for 15 min at room temperature and mounted using VectaMount (Vector Labs #H-5501). Images were acquired using a Zeiss AX10 light microscope and scored according to the parameters shown.

SHG images of the mammary ducts and tumors (acquired as described in the “Second-harmonic generation microscopy” section) were imported into the curvelet-based alignment analysis software CurveAlign (version 2.2 R2012a), available through the Laboratory of Optical and Computational Instrumentation at the University of Wisconsin-Madison. The borders of the mammary ducts and tumors were manually indicated, and the CurveAlign software measured the angle and quantity of collagen fibers with respect to the designated borders within a preset pixel distance. The individual curvelet angles were reported as an output value between 0 and 90. Based on prior literature, angles between 60 and 90 were characterized as a positive TACS3 region, angles between 30 and 60 as TACS2, and angles 0–30 as TACS1. The number of individual collagen fibers having an angle of interaction within each of the three signatures was counted and normalized to the total number of collagen fibers assessed within an ROI by the software.

The sections were deparaffinized in xylene and rehydrated in water. Masson’s trichrome staining kit (DAKO, Cat#AR173) was performed following the manufacturer’s procedure protocol and mounted using Permount toluene solution (Thermo Fisher Scientific). Triplicate images of the stained sections were collected for representative areas of 2560 × 1920 pixels (0.44 μm/pxls) using a Zeiss AX10 light microscope. The positive blue stain was converted to black using Adobe Photoshop CS5’s (version 12.0 × 64) magic wand tool. The images were imported into ImageJ (version 1.46r), and the black selections were quantified as percent area positive for collagen. The images were normalized against the periphery of the duct epithelia in which the final ratio represents collagen/epithelia.

MCF-7^PAPP-A and MCF-7^PAPP-AE483Q cells were generated as previously described (Takabatake et al). MCF-7, MCF-7^PAPP-A, MCF-7^{PAPP-A/DDR2 KO}, and MCF-7^PAPP-AE483Q cells were grown in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin at 37 °C/5% CO₂. rIGF-1 (Preprotech, #100-11) treatments were done in serum-free media at 10 nM for indicated time points. Imatinib (R&D Systems, #59-061-00) and PQ401 (Enzo Life Sciences ALX-270-459-M005) treatments were performed at indicated concentrations in complete media for 48 h, and untreated samples were treated with vehicle control (DMSO) for 48 h. PAPP-A- and PAPP-A E483Q-conditioned media treatments were performed at 10 ng/mL in complete media for 24 h. Cells treated with conditioned media were seeded on collagen I-coated plates 24 h prior to treatment. Preparation of collagen-coated plates was performed as follows: type I rat tail collagen (Corning, #354236) was diluted at 50 μg/mL in 0.1 nM HCl and incubated on plates at room temperature for 1 h followed by two PBS washes.

MCF-7^PAPP-A and MCF-7^PAPP-AE483Q cells were grown to 80% confluence in complete media. Media were then switched to serum-free and collected 24 h later, and PAPP-A protein was concentrated from media using 50 kDa cutoff Amicon Ultra-15 Centrifugal Filter Units (EMD Millipore, #UFC905024) spun at 5000g for 20 min at 4 °C. Concentrated condition media were then stored at − 20 °C, used for rIGFBP-5 protease assay, or quantified by ELISA. PAPP-A concentration was measured using Quantikine PAPP-A ELISA (R&D Systems, # DPPA00) following the manufacturer’s protocol adapted with a primary incubation for 18 h at 4 °C and PAPP-A conjugate incubation for 6 h at room temperature.

Cell-free protease assays were performed using PAPP-A protein secreted in 24 h serum-free culture media from MCF-7 PA and MCF-7 PA E483Q cells. Culture media from parental MCF-7 were used as a negative control for PAPP-A protein in cell media. Media were collected and concentrated as described above. Fifty nanograms of rIGFBP-5 (Abcam, #ab49835) was co-incubated with 15 μL of concentrated media and 15 μL of DMEM for 3 h at 37 °C. The reaction was inactivated by the addition of 1× Laemmli sample buffer to 40 μL final volume and boiling for 5 min. Degradation of IGFBP-5 in samples was then detected by western blot.

Cell lysates were prepared in RIPA lysis buffer (25 mM Tris HCl, pH 7.6, 150 mM NaCl, 1% NP-40, 0.1% SDS, 1% sodium deoxycholate) supplemented with protease and phosphatase inhibitors. Thirty micrograms total protein in 1× Laemmli buffer per sample was separated on a 10% SDS-glycine polyacrylamide gels ran at 80 V for 30 min and 200 V for 45 min. Proteins were transferred to nitrocellulose membranes (GE Healthcare) for 1.5 h at 85 V. Membranes were blocked in 5% milk in TBS-T and incubated on a rotator overnight at 4 °C in the following primary antibodies: DDR2 1:250 (Cell Signaling Technology #12133S), p-DDR2 Y740 1:600 (R&D Systems, #MAB25382), Snail 1:1000 (Cell Signaling Technology L70G2 #3895), Larp6 1:600 (Abnova, #H00055323-B01P), IGFBP-5 1:1000 (EMD Millipore, #06-110), PAPP-A 1:200 (Santa Cruz Biotechnology, # sc-50518), phospho-Akt (Ser473) 1:1000 (Cell Signaling Technology 587F11 #4051), Akt 1:10,000 (Cell Signaling Technology #9272), and actin 1:10,000 (EMB Millipore #MAB1501R). After three washes in TBS-T, and 10-min 10% milk/TBS-T incubation, the membranes were incubated in secondary anti-rabbit or anti-mouse antibodies conjugated to horseradish peroxidase (Thermo Fisher Scientific or KwikQuant) in 10% milk/TBS-T for 1 h at room temperature. Signal was detected using ECL (GE Healthcare, #RPN2106 or KwikQuant) following the manufacturer’s protocols. TBS-T was prepared with 0.1% Tween.

Total RNA was extracted from cell lines using TRIzol (Invitrogen) following the manufacturer’s protocol. One hundred nanograms of each triplicate sample was used in real-time RT-PCR using One-Step SYBR PrimeScript RT-PCR kit (Takara, Cat#RR086A) following the manufacturer’s protocols. Thermal cycle program for DDR2 primers: 50 °C 2 min, 95 °C 20 s, 40 cycles of 95 °C 3 s, 60 °C 30 s, and 72 °C 5 s, and Col1a1 primers: 42 °C 5 min, 94 °C 5 min, 30 cycles of 94 °C 12 s, 60 °C 8 s, 72 °C 8 s, both followed by a melt curve of 65 °C to 95 °C in 0.4 °C increments. Experiments were carried out in triplicate for each data point, and relative expression was determined using 2^−ΔΔCτ method.

Cell permeable membrane inserts (Falcon, #08-771-21) were coated with 200 μL of diluted 1:100 growth factor-reduced Matrigel basement membrane matrix (Corning, #356230) in cold PBS. Inserts were placed in 24-well plates (Falcon), and artificial extracellular matrix was allowed to polymerize for 2 h at room temperature. Following polymerization incubation, excess Matrigel solution was removed by pipette. Fifty thousand cells in 500 μL of serum-free DMEM were seeded atop the polymerized Matrigel in the inserts. Five hundred microliters of complete DMEM containing 10% FBS was added to each lower chamber. Invasion assays were incubated at 37 °C/5% CO₂. After 48 h, invaded cells were fixed and stained using the Hema 3 Manual Staining Stat Pack (Thermo Fisher Scientific, #23-123-869) according to the manufacturer’s protocol and mounted using Permount toluene solution (Thermo Fisher Scientific). All experiments were done in technical triplicates.

MCF-7 and MCF-7^PAPP-A cells were seeded at 50% confluence and treated with imatinib and PQ401 at the indicated time points in triplicate samples for 48 h. Following treatment, complete media with MTT at 0.5 mg/mL was added to each well and incubated at 37 °C for 3 h protected from light. After incubation, and DMSO was used as MTT solvent and each will be mixed thoroughly to homogenize the color and subsequently read at absorbance OD = 570 nm. The samples were averaged and normalized to the vehicle control sample.

Two-month-old virgin female FvBN mice were bred and then housed separately post-fertilization. Pregnant mice were checked daily for litter drop and then underwent a lactation period of 2 days. Mice were injected with met1 cells at a time point of post-partum 2 months post-lactation. Met1 cells were kindly provided by Dr. Emily Gallagher at Mount Sinai. Two hundred fifty thousand met1 cells in 100 μL of control MCF-7- or PAPP-A-conditioned media at 0.1 ng/μL were injected into the upper right and left inguinal mammary fat pads (n = 5 mice, 6 tumors per group). The lower left and right inguinal mammary fat pads received injections of MCF-7-concentrated media or concentrated PAPP-A from MCF-7^PAPP-A media at 0.1 ng/μL alone (n = 5 mice, 6 mammary glands per group, PAPP-A concentrated as described above). Subsequently, every 2 days for 3 weeks following the initial injection, 100 μL of control media or 0.1 ng/μL PAPP-A containing media was injected into the upper and lower left and right inguinal mammary fat pads or Met1 xenografts. All injections were done using a 25-G needle. Final tumor volume was recorded by weight, and all tumors and mammary glands were harvested for histological analysis. The lungs were collected for histological analysis following perfusions with sterile PBS and 10% formalin. Clusters of cancer cells were detected by H&E and counted as micrometastases.

Publicly available primary breast cancer gene expression and phenotype data published by Kao et al. (GSE20685) was downloaded from Gene Expression Omnibus [27]. PAPP-A/SNAI1/COL1A1 gene signature was composed of the probes for each gene with the highest average expression across all samples. The probes used for each gene are as follows: PAPP-A (201981_at), SNAI1 (219480_at), and COL1A1 (202310_s_at). The secondary validation dataset in Additional file 4: Figure S4 was collected from Chanrion et al. (GSE9893), and the probes used were PAPP-A: H200006224, COL1A1: H200008097, and SNAI1: H200004861 [28]. Normalized expressions of PAPP-A/SNAI1/COL1A1 were averaged to calculate an expression score for each patient (n = 327). The top 50% of patients with the highest expression scores were designated as PAPP-A/SNAI1/COL1A1 high (n = 163), and the bottom 50% were designated as PAPP-A/SNAI1/COL1A1 low (n = 164). Differential gene expression between the two groups of patients was calculated using by LIMMA [29]. Genes with a false discovery rate (FDR) less than 0.05 were ranked by Log₂ fold change and used for pre-ranked gene set enrichment analysis (GSEA) using GenePattern [30]. Enrichment scores were determined as a running sum statistic at maximum deviation from zero.

For the outcome analysis, we constructed Kaplan-Meier curves for time to distant metastasis as the endpoint. This was defined as the time from primary disease diagnosis to the detection of distant metastasis. Differences between patients with a high and low score of our predefined PAPP-A/SNAI1/COL1A1 signature were evaluated using the log-rank test. Patients who died without metastasis were censored at the time of death. The analysis was performed using R Packages Survival, Survminer, and RMS [31].

For lentiviral vector production, HEK293T cells were used in standard cell culture conditions with DMEM supplemented with 10% FBS. Cells at 80% confluence were co-transfected using Lipofectamine 2000 (Invitrogen #11668-019) following the manufacturer’s protocol with 12 μg of lentiviral vector, 9 μg of psPAX2, and 3 μg of pMD2.G. Fresh media (DMEM supplemented with 10% FBS and penicillin/streptomycin) were exchanged 16 h post-transfection, and supernatants were collected 48 h post-transfection and sterile filtered (.45 μm). Target cells (MCF-7 and MCF-7^PAPP-A) were infected with the lentiCRISPRv2 (Addgene #52961) plasmid lentiviral supernatant described above and polybrene (8 μg/mL) for 24 h. Cells were selected with puromycin for 2 weeks prior to clonal isolation and expansion. gRNAs (see oligo sequence below) targeting DDR2 were designed using the publicly available MIT CRISPR Design program.

To test whether the expression of PAPP-A affects the amount, distribution, or orientation of collagen in post-partum mammary glands, we harvested the mammary glands from parous PAPP-A transgenic or non-transgenic mice during involution (12 days post-weaning) or post-partum (2 months post-weaning). We previously shown that due to the use of a truncated MMTV promoter, where the hormonal responsive element is deleted, the expression of PAPP-A in our mouse model is constitutive but is not enhanced by pregnancy [22]. We showed that PAPP-A levels are similar in virgin, pregnant, lactating, involuting, and post-partum mammary glands [22].

First, the amount of collagen in each group was analyzed using Masson’s trichrome staining. As internal controls, we confirmed that the amount of collagen in the non-transgenic females is increased during involution (Fig. 1a, b) and is maintained in the post-partum mammary glands (Fig. 1a, b) [8]. Further, in the PAPP-A transgenic mice, we confirmed that while the level of collagen in virgin females is not significantly different from that in non-transgenic virgin females, collagen during involution is significantly higher in PAPP-A transgenic mice compared to non-transgenic mice (Fig. 1a, b) [22]. We then analyzed the level of collagen in the post-partum glands in the PAPP-A transgenic mice. We found that the level of collagen remains higher in the post-partum mammary glands in PAPP-A transgenic mice (Fig. 1a, b).

Second, we then performed a second-harmonic generation to determine the distribution and orientation of collagen (Fig. 1c). Again as internal controls, we confirmed that in the non-transgenic mice, the collagen during involution shows a wide distribution as indicated by the presence of collagen across longer distances from the ducts and that in contrast, collagen is closer to the mammary duct borders, which is a characteristic of TACS1, in the post-partum mammary glands (Fig. 1d) [8]. We also confirmed that in the PAPP-A transgenic mice, collagen spreads over a wider range of distances from the ducts during involution compared to the non-transgenic mice (Fig. 1e) [22]. We then analyzed the distribution of collagen in the post-partum glands in the PAPP-A transgenic mice. We found that in sharp contrast to the non-transgenic mice, the same wide distribution was maintained in the post-partum glands in transgenic mice (Fig. 1e).

Third, we scored the angle of collagen fiber orientation. TACS3 regions are quantified as percent number of individual collagen fibers (curvelets) at an angle of 60–90° relative to the ductal border [13]. The total curvelets were analyzed using the CurveAlign software, and regions of interest showing an angle of collagen at more than 60° relative to the duct border were scored as TACS3. We found that in the non-transgenic mice, the levels of TACS3 increases during involution but are decreased in the post-partum glands (Fig. 1f). In contrast, in the PAPP-A transgenic mice, TACS3 collagen was increased during involution and was maintained at similar levels in the post-partum glands (Fig. 1g).

We concluded that the constitutive expression of PAPP-A in the mammary gland maintains the elevation in the amount, distribution, and orientation of collagen in the post-partum mammary gland. These alterations result in the persistence of involution-like collagen into the post-partum mammary glands. Considering that the collagen of involution is pro-tumorigenic, this result raises the possibility that the expression of PAPP-A, after involution is completed, in a post-partum mammary gland may contribute to the development of PABC.

In contrast to the PAPP-A transgenic mice, in human breast cancers, rather than being constitutive, the expression of PAPP-A is sporadic and can occur at any time before or after pregnancy. Notably, we show that PAPP-A is a transcriptional target of mutant p53 suggesting that mutations in p53 represent one mechanism by which PAPP-A is abnormally overexpressed in breast cancer [32‐34]. Having established that constitutive expression of PAPP-A in the PAPP-A transgenic mice maintains an involution-like collagen in the post-partum glands (Fig. 1), we next investigated whether sporadic expression of PAPP-A in an established post-partum mammary gland of non-transgenic mice is able to convert the post-partum collagen architecture into an involution-like collagen architecture. To test this possibility, we established a cohort of non-transgenic post-partum mice (2 months post-weaning). In parallel, we collected media from MCF-7 cells overexpressing PAPP-A (MCF-7^PAPP-A), which we previously characterized and shown to secrete proteolytically active PAPP-A [22]. We then concentrated the PAPP-A from the media and injected PAPP-A (PAPP-A media) or control media concentrated from MCF-7 cells into the post-partum mammary glands of the non-transgenic mice every 2 days for 3 weeks.

We found that injections of PAPP-A in the post-partum mammary glands of non-transgenic mice induce a remarkable increase in proliferation, indicated by the significantly higher number of ducts per mammary glands (Fig. 2a, b). Further, the post-partum mammary glands injected with PAPP-A had significantly higher collagen abundance (Fig. 2c, d). SHG imaging of collagen orientation and CurveAlign analysis revealed that post-partum mice injected with PAPP-A adopted a TACS3 phenotype, compared to the mammary glands injected with control media, which exhibited predominantly TACS1 collagen (Fig. 2e, f). As a control, we also injected PAPP-A media into the mammary glands of virgin non-transgenic mice. However, injections of PAPP-A had no significant effect on the proliferation or collagen in the mammary glands of virgin mice (Additional file 1: Figure S1A, B). These results suggest that the sporadic presence of PAPP-A is sufficient to convert post-partum collagen into a high TACS3 involution-like collagen architecture.

Considering that involution-collagen is pro-tumorigenic, we then tested whether the effect of sporadic expression of PAPP-A affects tumor growth. To address this possibility, we used Met1 breast epithelial cancer cells, however, as these cells are known to form aggressive metastatic tumors in mice, we injected only 250,000 cells in order to increase the probability to detect differences between the groups. Met1 cells were injected into the post-partum mammary glands of non-transgenic mice along with control media or media containing PAPP-A followed by injections of the respective media every 2 days for 3 weeks. We found that tumors injected with PAPP-A are on average larger compared to the control group, but this difference did not reach statistical significance (Fig. 2g). In addition, CurveAlign analysis of the collagen orientation on these tumors revealed that the mammary tumors injected with PAPP-A had a significantly higher frequency of TACS3 regions compared to the control group (Fig. 2h, i).

As TACS3 is associated with increased metastasis [13], we analyzed these mice for lung metastases. Consistent with the higher TACS3 phenotype, mice injected with PAPP-A had a substantial increase in the frequency of micro-metastases to the lungs compared to the control group within this short period of time (Fig. 2j, k).

Taken together, these results indicate that sporadic expression of PAPP-A between pregnancies is able to convert the elevated levels of collagen of a post-partum gland to provide a pro-tumorigenic microenvironment. These results therefore imply that the abnormal gain of the overexpression of PAPP-A at any time after a first pregnancy may be sufficient to promote the development of PABC. Further, they indicate that PAPP-A also enhances the metastatic potential of cancer cells.

In addition to a structural role of collagen in metastasis [13, 35], collagen activates the discoidin domain receptors DDR1 and DDR2 [36‐38]. DDR1 and 2 are frequently overexpressed in breast cancers and have recently been implicated in breast cancer metastasis [36, 39‐41]. DDR2 has been linked with the invasive TACS3 phenotype, while the loss of DDR2 is associated with a non-invasive collagen signature, TACS2 [40, 41].

A recent report described a novel signaling role for DDR2 in breast cancer metastasis through the stabilization of the epithelial to mesenchymal transition (EMT) transcription factor Snail (Fig. 3a) [39]. In addition to a critical role in activating EMT, Snail may also promote cancer stem-like subpopulations, drug resistance, and tumor cell invasion [42, 43]. Since we found that PAPP-A leads to an increase in collagen [22] (Fig. 1a, b) and promotes metastasis (Fig. 2j), we sought to address whether the DDR2/Snail axis of metastasis is active in PAPP-A-driven PABC (Fig. 3a).

To address this possibility, we performed immunohistochemistry of Snail on the mammary glands from non-transgenic or PAPP-A transgenic mice during involution or late post-partum. This analysis revealed a significant increase in nuclear expression of Snail, which has been reported to be the most aggressive form (Fig. 3b) [44, 45]. To confirm that the upregulation of Snail in cells overexpressing PAPP-A confers an increase invasion capacity, we first determine the level of Snail in MCF-7 and MCF-7^PAPPA cells co-cultured with and without collagen I to mimic the collagen-rich environment of involution and post-partum mammary glands. We found a significant elevation in Snail in the MCF-7^PAPP-A cells (Fig. 3c). While the levels of phosphorylated DDR2 (p-DDR2) could not be determined by immunohistochemistry due to the lack of anti-p-DDR2 antibody that recognizes mouse DDR2, we next aimed at determining the level of p-DDR2 in MCF-7 and MCF-7^PAPPA cells. MCF-7^PAPPA cells showed higher phosphorylated DDR2 (p-DDR2) and Snail expression co-cultured with and without collagen I (Fig. 3c). Therefore, these observations confirm that the DDR2/Snail axis is activated following the overexpression of PAPP-A. Since Snail is associated with invasion, we also measured the invasion capacity of MCF-7 and MCF-7^PAPPA cells using Transwell invasion assay. We found that MCF-7^PAPPA cells have a significantly higher cell invasive capacity (Fig. 3d).

Since we found that collagen is required in vivo for the activation of PAPP-A [17], we were surprised by the observation that in MCF-7^PAPP-A cells, DDR2 is activated in the absence of extraneous collagen. We hypothesized that this observation may be due to the ability of PAPP-A to promote the production of collagen in these cells. To test this possibility, we measured the level of collagen 1 mRNA in MCF-7 and MCF-7^PAPPA cells. We found an elevation in collagen mRNA expression MCF-7^PAPPA cells (Fig. 3e). Further, since we found that injecting PAPP-A promotes collagen deposition in vivo (Fig. 2c), we also measured collagen mRNA levels in MCF-7 cells treated with media containing PAPP-A. We found higher mRNA transcripts of col1a1 under these conditions (Fig. 3f).

La ribonucleoprotein domain family member 6 (LARP6) is a post-transcriptional modifier of col1a1 that enhances col1a1 transcripts’ translation by binding to a specific 5′ UTR stem-loop of col1a1 thereby both stabilizing col1a1 transcripts and aiding in their transport from the nucleus to the cytoplasm [46, 47]. IGF signaling has been previously reported to increase collagen production specifically by upregulating LARP6 [48, 49]. Therefore, since PAPP-A increases IGF signaling, we analyzed the level of LARP6 in MCF-7 cells treated with PAPP-A. We found that MCF-7 cells treated with PAPP-A resulted in an upregulation of LARP6 protein (Fig. 3g). To verify if this observation holds true in PAPP-A transgenic mice, we analyzed the level of LARP6 in non-transgenic and PAPP-A transgenic mammary glands during involution and in post-partum glands. We found that LARP6 protein is upregulated in vivo during involution but more significantly in post-partum glands (Fig. 3h). These results indicate that in addition to its known role in promoting IGF signaling, PAPP-A also promotes metastasis through the activation of DDR2 and promotes the stabilization of collagen through activation of LARP6.

PAPP-A is a large protease with multiple uncharacterized domains. Therefore, it remains possible that the action of PAPP-A on DDR2 is independent of its proteolytic function. To test this possibility, we treated MCF-7 cells with media containing wild-type PAPP-A and media containing a proteoloytic-dead mutant form of PAPP-A (E483Q). We confirmed that the E483Q mutant is inactive and unable to cleave recombinant IGFBP-5 in vitro (Fig. 3i). We found that wild-type but not inactive PAPP-A activates DDR2 and Snail in MCF-7 cells in vitro (Fig. 3j). Since PAPP-A activates IGF signaling, we then asked if IGF-1 downstream of PAPP-A is able to activate DDR2. To investigate this possibility, we tested the effect of exogenous IGF-1 on DDR2 in vitro. We found that IGF-1 activates DDR2 in MCF-7 cells, but this effect was markedly increased in MCF-7^PAPP-A cells (Fig. 3k). Further, MCF-7^PAPP-A cells expressing inactive PAPP-A also show some activation although the effect was markedly decreased compared to MCF-7^PAPP-A cells (Additional file 2: Figure S2).

These results suggest that through its ability to increase IGF signaling and the elevation in collagen, PAPP-A leads to the increased activation of DDR2 signaling.

To further validate the role of DDR2 in the increased invasion of cells overexpressing PAPP-A, we used CRISPR to knock out DDR2 in MCF-7 and MCF-7^PAPP-A cells. First, we confirmed that DDR2 expression is abolished in the MCF-7^PAPPA-DDR2KO cells (Fig. 4a). Inhibition of DDR2 resulted in undetectable levels of Snail by immunoblot (Fig. 4a). Further, the decrease in Snail correlated with a marked decrease in cell invasive capacity (Fig. 4b). We confirmed that inhibition of DDR2 abolishes the activation of Snail by PAPP-A using MCF-7 cells and MCF-7 cells where DDR2 was inhibited by CRISPR (MCF-7^DDR2KO) despite treatment with PAPP-A containing media (Fig. 4c).

Therefore, these results suggest that in addition to increasing IGF signaling and collagen production, PAPP-A leads to the activation of a distinct collagen signaling pathway mediated by DDR2 which promotes invasion.

To further explore the potential clinical relevance of DDR2 signaling in PAPP-A-driven PABC, we sought to investigate the effect of DDR2 in vivo by comparing the growth of MCF-7, MCF-7^PAPP-A, and MCF-7^{PAPP-A/DDR2KO} xenografts in the mammary fat pads of nude mice. Additionally, we previously found that using a 1:1 Matrigel-collagen reduces the growth of MCF-7 xenografts compared to Matrigel alone [22]. This observation suggests that Matrigel-collagen mixture mimics the collagen-rich and anti-proliferative environment of a post-partum mammary gland [14, 22]. We therefore performed these xenografts using a 1:1 Matrigel-collagen mixture.

Consistent with our previous reports, we found that MCF-7^PAPP-A cells formed significantly larger tumor than MCF-7 cells under these conditions (Fig. 4d). However, this effect was abolished upon DDR2 depletion (Fig. 4d). Therefore, we conclude that DDR2 is a significant mediator of the accelerated growth of cells overexpressing PAPP-A in vivo.

Further, in agreement with the results obtained in vitro, we found that MCF-7^PAPP-A xenografts have the highest Snail protein expression by IHC, while MCF-7^{PAPP-A/DDR2KO} xenografts exhibit undetectable Snail expression (Fig. 4e).

We next analyzed the orientation of collagen in these xenografts. We found that MCF-7 xenografts exhibit elevated TACS1 but low TACS3, while MCF-7 ^PAPP-A xenografts show elevated TACS3 orientation (Fig. 4f, g). Strikingly, MCF-7^{PAPP-A/DDR2KO} xenografts had significantly less TACS3 regions of collagen and distribution of curvelet angles comparable to that seen in MCF-7 tumors (Fig. 4f, g). These results suggest a critical role for DDR2 in the ability of PAPP-A to promote and maintain collagen into a TACS3 orientation.

Imatinib has been shown to inhibit the activity of DDR2 [50, 51]. Since PAPP-A activates both IGF and DDR2 signaling, we tested the ability of the IGF inhibitor PQ401 alone or in combination with imatinib on the growth of MCF-7^PAPP-A cells. First, we confirmed that imatinib reduces the activation of DDR2 in these cells (Fig. 4h). However, the reduction was enhanced in cells treated with imatinib in combination with PQ410 (Fig. 4h). We found that while imatinib has no effect on the growth of either MCF-7 or MCF-7^PAPP-A cells (Fig. 4i) and PQ401 inhibited cell growth in both cell lines equally (Fig. 4j), the combination selectively inhibited the growth of the MCF-7^PAPP-A cells (Fig. 4k). Mechanistically, we found that MCF-7^PAPP-A cells have higher p-Akt levels than MCF-7 cells (Additional file 3: Figure S3). However, a decrease in p-Akt levels was only observed in MCF-7^PAPP-A cells when treated with the combination treatment of PQ401 and imatinib (Additional file 3: Figure S3). This observation is consistent with the observation that MCF-7^PAPP-A cells have higher DDR2 activation in response to treatment with IGF (Fig. 3k). Further, a recent study reported a positive crosstalk between DDR1 and IGF signaling, therefore raising the possibility that a similar crosstalk exists between DDR2 and IGF signaling [52].

These results indicate that complete inhibition achieved by either deletion of DDR2 by CRISPR or treatment with pharmacological combination is necessary to abolish the growth of the MCF-7^PAPP-A cells, while DDR2 inhibition by imatinib alone does not.

We previously reported that a long lactation of 2 weeks or more prior to involution inhibits the formation of PABC in the PAPP-A transgenic mice [22]. As long lactation represents an alternative method to inhibit the action of PAPP-A upstream of DDR2, we tested the effect of long lactation on the level of collagen deposition in post-partum glands. We harvested mammary glands from parous PAPP-A transgenic mice during involution (12 days post-weaning) or post-partum (2 months post-weaning) as described in Fig. 1 with the exception that mothers were kept with their litters for 2 weeks prior to the initiation of involution. These mammary glands were compared to the mammary glands from involuting and post-partum PAPP-A transgenic mice following a short lactation described in Fig. 1. In non-transgenic mice, the length of lactation had no effect as expected (Fig. 5a, b) and was therefore not pursued further. In PAPP-A transgenic mice, however, we found that long lactation significantly decreased collagen abundance in both involuting and post-partum mammary glands (Fig. 5a, b).

We next analyzed the orientation of collagen. Consistent with the lack of effect of long lactation on collagen abundance, we found that long lactation had no effect on collagen orientation in involuting mammary glands from non-transgenic mice (Fig. 5d–f). However, long lactation resulted in a distribution of collagen that is closer to the ductal border and decreased TACS3 in involuting mammary glands of PAPP-A transgenic mice (Fig. 5g–i). Importantly, the distribution of collagen and the frequency of TACS3 in PAPP-A transgenic involuting mammary glands after a long lactation mimic those observed in the non-transgenic involuting mammary glands (Fig. 5e, f).

In post-partum mammary glands from PAPP-A transgenic mice, we found that a long lactation also significantly altered collagen orientation, as indicated by a decreased presence of collagen dispersed from the ductal border and decreased TACS3 (Fig. 5j–l). Furthermore, the decrease in TACS3 in PAPP-A transgenic post-partum mammary glands following a long lactation was even more significant than that observed in involuting mammary glands after a long lactation (Fig. 5f, i, l). These results highlight that a longer lactation ablates the ability of PAPP-A to maintain an involution-like collagen in post-partum mammary glands.

Several barriers exist in the proper diagnosis of PABC. One is the lack of reliable diagnostic biomarkers. Despite the identification of PAPP-A as a potentially important biomarker of PABC, PAPP-A alone is inefficient as a biomarker since it is also overexpressed in the vast majority of breast cancer [21]. Our findings that virgin PAPP-A transgenic mice do not develop mammary tumors suggest that, despite its overexpression, PAPP-A does not act as an oncogene in the absence of collagen. Therefore, for PAPP-A to be useful as a diagnostic biomarker, it must be analyzed in the context of high collagen content. However, another difficulty arises from the fact that information regarding breastfeeding is either sparse or entirely missing from medical charts. Therefore, the use of combined expression of PAPP-A and collagen as a biomarker performs poorly as the impact of high expression of both genes can be ablated by long breastfeeding. Having identified the involvement of a DDR2/Snail axis in PAPP-A-driven PABC, we reasoned that Snail in combination with PAPP-A and collagen might represent a more reliable readout for monitoring the activation of the pathway. To test this hypothesis, we investigated the correlation between a three-gene signature of PAPP-A, COL1A1, and SNAI1 and clinical outcomes in breast cancer patients. First, we generated a score based on the average gene expression of PAPP-A/SNAI1/COL1A1 from a publicly available dataset of 327 patients with primary breast cancer (GSE20685) [27]. We next stratified patients based on this score into PAPP-A/SNAI1/COL1A1 high and low (Fig. 6a). We found that the PAPP-A/SNAI1/COL1A1-high patient population develops distant metastases at a significantly higher rate than those patients with low values of the PAPP-A/SNAI1/COL1A1 expression score (P = 0.042, Fig. 6b). This finding was validated in an additional dataset (GSE9893) with the PAPP-A/SNAI1/COL1A1-high patients developing distant metastases at a significantly higher rate (p = .0045, Additional file 4: Figure S4) [28].

To unravel the major molecular differences between patients with high and low values for our PAPP-A/SNAI1/COL1A1 expression score, we next conducted differential gene expression between these patients [29]. We found 78 significantly upregulated genes of interest in the PAPP-A/SNAI1/COL1A1-high patient population (Fig. 6c, Additional file 5: Table S1). Of these 78 genes, highlighted in red is PAPP-A, COL1A1, and SNAI1 as expected in addition to DDR2 and LARP6 (Fig. 6c). These 78 genes are involved in extracellular matrix components, collagen, collagen processing and remodeling, EMT, IGF signaling, Wnt signaling, and cell invasion and migration (Fig. 6c).

All significantly upregulated genes from the differential gene expression data were used for gene set expression analysis (GSEA) to identify significantly activated pathways in the PAPP-A/SNAI1/COL1A1-high patient population [30]. We found 13 statistically significant and relevant pathways active in the PAPP-A/SNAI1/COL1A1-high patient group (Fig. 6d, Additional file 6: Table S2). Of note, EMT, extracellular matrix organization, and collagen formation were the top 3 highly activated pathways in the PAPP-A/SNAI1/COL1A1-high patient group (Fig. 6d). This is consistent with the increased frequency of TACS3 collagen and cell invasion observed in our PAPP-A overexpressing cells in vitro and in vivo.