Integrins Alpha-2 and Beta-1 expression increases through Multiple Generations of the EDW01 Patient-Derived Xenograft Model of Breast Cancer – Markers but not Drivers of Epithelial Mesenchymal Transition

Background: Breast cancers acquire aggressive capabilities via epithelial to mesenchymal transition (EMT), in which various integrins/integrin linked kinase signalling are upregulated. Methods: We investigated this in two patient-derived xenografts (PDXs) developed from breast-to-bone metastases, and it’s functional signicance in a breast cancer cell line system. ED03 and EDW01 PDXs were grown subcutaneously in immunocompromised SCID mice through 11 passages and 7 passages, respectively. Tumour tissue was assessed using immunohistochemistry (IHC) for estrogen receptor (ER)-alpha, E-cadherin, vimentin, Twist1, beta-catenin, P120-RasGAP, CD44, CD24 and Ki67, and RT-qPCR of EMT-related factors (CDH1, VIM, CD44, CD24), integrins beta-1 (ITGB1), alpha-2 (ITGA2) and ILK. Integrin and ILK expression in epidermal growth factor (EGF) induced EMT of the PMC42-ET breast cancer cell line was assessed by RT-qPCR and Western blotting, as were the effects of their transient knockdown via small interfering RNA +/- EGF. Cell migration, changes in cell morphology and adhesion of siRNA-transfected PMC42-ET cells to various extracellular matrix (ECM) substrates was assessed. Results: The ED03 (ER+/PR-/HER2-/lobular) and EDW01 (ER+/PR-/HER2-/ductal) PDXs were both classied as molecular subtype luminal A. ED03 xenografts exhibited mutated E-cadherin with minimal expression, but remained vimentin-negative across all passages. In EDW01, the hypoxic indicator gene CAIX and Twist1 were co-ordinately upregulated at passage 4-5, corresponding with a decrease in E-cadherin. At passages 6-7, vimentin was upregulated along with ITGB1 and ITGA2, consistent with an increasing EMT. The ED03 PDX displayed minimal change over passages in mice, for all genes examined. ILK, ITGB1 and ITGA2 were also increased in the EGF-induced EMT of PMC42-ET cells (in which E-cadherin was downregulated) although siRNA against these targets revealed that this induction was not necessary for the observed EMT. However, their knockdown signicantly reduced EMT-associated adhesion and Transwell migration. Conclusion: Our data suggest that despite an increase in integrins alpha-2 and beta-1 gene expression in the EMT exhibited by EDW01 PDX over multiple generations, this pathway may not necessarily drive the EMT process. scratch-wound assays. Data are expressed as percentage of control and represent the mean ± SD from 3 biological replicates. (C) Effect of ILK inhibitor QLT0267 (used at 6.25 (cid:0) M, in DMSO) on PMC42-ET cell migration in the Boyden Chamber assay. Results represent average cell counts from of 5 random high-power elds from three independent experiments, error bars represent standard deviation. For A, statistical signicance (*, p<0.05, comparison to control siRNA) was determined using two-way ANOVA, with Dunnett’s multiple comparison test. For B, statistical signicance was determined using two-way ANOVA followed by Holm-Sidak's multiple comparisons test. For C, statistical signicance (***, p<0.0001) was determined using unpaired two-tailed t test with Welch’s correction.


Introduction
Human breast cancer cell lines have been used extensively in vitro and in mice to dissect the cellular mechanisms associated with tumour aggressiveness and metastasis. However, cell line xenografts typically fail to recapitulate tumour cell heterogeneity. By contrast, human breast tumours engrafted into mice as patient derived xenografts (PDXs) generally show excellent reproducibility of morphological and genetic characteristics of the original tumour with minimal genetic drift and as such are clinically relevant platforms for preclinical studies [1][2][3][4][5].
Aggressive cancers are known to display epithelial-mesenchymal plasticity (EMP), through which they can uctuate between epithelial or mesenchymal states (or degrees of these) to assist survival in changing microenvironmental conditions [6][7][8]. Epithelial to mesenchymal transition (EMT) describes a phenotypic change toward a more mesenchymal state resulting in more motile and invasive cancer cells. EMT has been shown to promote progression in several cancer types including breast [9,10], contributes to chemoresistance [11,12], and is prominent in circulating tumour cells (CTCs) [7,[13][14][15]. Evidence for EMP has been demonstrated in numerous human breast cancer cell line studies and increasingly in breast cancer in vivo models [16][17][18][19] and clinical material (reviewed in [20]).
Invasive lobular carcinoma (ILC) and invasive ductal carcinoma (IDC), (also known as invasive carcinoma of no special type (IC NST)), have distinctive morphological features [21]. ILC is typi ed by single-le epithelial tumour cells, a ner stromal in ltration and often a minimal sclerotic tissue reaction -the combination of which makes self-detection and screen-detection (mammography) more di cult than for the typically more palpable IDC tumours, which grow as masses of epithelial cells within a desmoplastic stroma. These patterns are intimately linked to E-cadherin: ILCs do not express E-cadherin due to the presence of inactivating mutations [22,23] or silencing via methylation [24], or copy loss [25], and hence grow as individual and linear arrays of tumour cells. By contrast, IDCs typically express E-cadherin and hence grow as cohesive tumour nests [21].
A major initiating event in the transcriptional programming of EMT is E-cadherin repression [26][27][28]. IDC expression of E-cadherin is downregulated when they undergo an EMT, which is associated with increased invasiveness [29,30], thus for the purposes of this study, IDC could be considered as "EMTpositive". Interestingly, EMT does not occur spontaneously in ILC cells [31] and they tend to show less vimentin expression than IDC [32], thus ILC could be considered "EMT-negative". Similarly, not all IDCderived cell lines undergo EMT upon E-cadherin silencing, while some do [18].
Integrin switching is prominent in breast cancer EMT and has been linked to tumour aggressiveness [33], and integrins have been shown to play important roles in tumour cell transmigration via EMT (reviewed in [34]). Integrins are a large family of heterodimeric cell surface receptors that play a prominent role in the adhesive interactions between cells and their surrounding extra-cellular matrix (ECM), providing adhesion for stationary cells, as well as traction during cell movement [35][36][37][38]. TGFb-induced EMT in NMuMG mouse mammary cancer cells results in the downregulation of epithelial α6β4 integrin expression (which mediates contact with the basement membrane) through epigenetically silencing of the gene encoding integrin β4 [39]. α3β1 integrin, which binds laminin and also associates with E-cadherin, is required for progression through EMT in lung alveolar epithelial cells, where it integrates beta-catenin and transforming growth factor-β (TGFβ)-SMAD signalling to promote myo broblast formation and lung brosis [40]. Interactions between α5β1 integrin and bronectin have been associated with EMT in Eph4 mouse mammary cancer cells and human lung cancer cell lines [41,42]. In pancreatic carcinoma cells, increased expression of α1β1 or α2β1 integrins and their interactions with type I collagen facilitate the disruption of E-cadherin complexes and the nuclear translocation of beta-catenin, and promote proliferation and motility [43].
Indeed, α2β1 integrin is widely expressed on epithelial cells and its levels are increased in several carcinoma cells from the epithelial origin [44]. Growing evidence indicates that α2β1 integrin can be a key pathway in cancer pathogenesis [43,45,46]. Furthermore, Chen et al (1994) showed that increased expression of α2β1 integrin is positively correlated with increased metastatic ability in human squamous cell lung cancer cells when i.v. inoculated in severe combined immunode ciency (SCID) mice [47].
It has been demonstrated that ILK can induce a complete EMT in various epithelial cell lines, and thus be involved in the initiation of EMT in vivo, and the maintenance of the mesenchymal phenotype and disease progression [48][49][50][51]. ILK can modulate the expression of not only E-cadherin, but also other epithelial markers such as cytokeratin 18 [52] and MUC1 [53], as well as mesenchymal markers such as LEF1 [54] and vimentin [53,54]. Therefore, ILK is able to initiate an EMT. Gain and loss of function strategies have shown that over-expression, and/or constitutive activation of ILK results in oncogenic transformation and progression to invasive and metastatic phenotypes [55].
In the current study, the divergent cancer subytpes of ILC and IDC, which differ in regards to their EMT status, were studied as PDXs in order to examine the relationship between EMT, α2β1 integrin and ILK in cancer progression, through serial passages in mice.
We coupled this with an investigation into the pattern of ILK and integrin expression changes in PMC42 human breast cancer cells induced by epidermal growth factor (EGF) treatment to undergo an EMT in vitro, and assessed whether these integrin changes were necessary for EMT to occur.

Materials And Methods
Patient material and creation of xenografts Establishment of the PDXs used in this study was described previously [56].
The ED03 xenograft was derived from a lobular breast cancer bone metastasis in a 40 year old woman, 3.5yrs after initial diagnosis of her primary tumour. The EDW01 PDX was derived from a bone metastasis of an invasive ductal carcinoma of the breast presenting as clinically overt macrometastatic deposits in a 44 year old woman.
Brie y, the tumour tissue derived initially from the bone metastasis deposits was diced into ~1mm pieces, mixed with Matrigel® (BD Biosciences, Australia) and implanted bilaterally subcutaneously in SCID mice (ARC, Perth, Australia). For each passage, once tumour volumes reached 2000 mm 3 , mice were euthanised, and the tumours were removed. The tumour tissue was again chopped into chunks, mixed with Matrigel®, and implanted into fresh mice. This was repeated 6 times for EDW01 (total of 7 passages) and 10 times for ED-03 (total of 11 passages). Portions were snap frozen for RNA extraction and formalin xed and para n embedded for immunohistochemical analyses at each passage.

Immunohistochemistry (IHC)
A tissue microarray was created of randomised duplicate 2mm diameter cores of tumour blocks corresponding to various passage numbers through mice of the ED03 and EDW01 PDX model systems.
IHC was performed using the Ventana Discovery Ultra Automated Slide Preparation system. Details of antibodies used in this study can be found in Table 1. The membrane associated proteins (E-cadherin, beta-catenin, P120-RasGAP, CD24, CD44, carbonic anhydrase IX (CAIX)) were scored as cytoplasmic or membranous, and whether they were heterogeneous or homogeneous in these areas. Vimentin staining was scored as positive if present in the cytoplasm of cells, whereas for Twist1 and Ki67, the proportion of positive nuclei was recorded.
Quanti cation of IHC data shown in Supplementary Figures 1 and 3 was determined using ImageJ, where DAB brown positive nuclei were separated from blue (total) nuclei, and using the color threshold tool, a threshold applied to only select positive nuclei that was visible by eye. Using this threshold, the area taken up by positive nuclei was quanti ed. The ratio of "relative intensity per cell" was obtained by dividing the overall area of positivity for the IHC target (either pink or brown) by overall nuclear area. In gure 4C, per cent Ki67 nuclear positivity was determined using ImageJ, data processed in the same manner as for IHC quanti cation, with the results expressed as percentage positive cells.
Necrosis quanti cation and determination of cellular area in phase contrast images using ImageJ Whole core images (4x magni cation) were imported into ImageJ (v.1.52a, National Institutes of Health, USA). Necrotic area in E-cadherin/vimentin stained cores was outlined and expressed as a percentage of the total core area. Since cores were sampled from the blocked tumour in a random, unbiased manner, necrotic area in the core was considered representative of the whole tumour. Assessment of Vimentin and E-cadherin positive cellular areas were determined using ImageJ after separating out the relative colors, adjusting the thresholding such that maximal positive cellular area was selected in the absence of background, then area was determined using the "Analyse particles" tool and expressed as a percentage of the total core area.
Reverse transcriptase-quantitative PCR RNA was extracted using the Qiagen RNeasy Mini prep Kit (Qiagen, Doncaster, Vic, Australia). cDNA synthesis and reverse transcriptase -quantitative PCR (RT-qPCR) was performed as previously described, using a speci c reverse transcriptase (RT) primer in the cDNA synthesis step [57,58]. Expression levels indicated by raw cycle thresholds (CTs) of the genes of interest were subtracted from the raw CT of the ribosomal protein L32 (RPL32) mRNA, and plotted as dCT. RPL32 CT values were observed to be unchanged by passage number in mice or PDX. Human-speci c primers for various genes examined in this study are detailed in Table 2.
siRNA mediated knockdown Small interfering RNA (siRNA)-mediated knockdown of ITGB1, ITGA2, and ILK (Horizon, [formerly Dharmacon], Melbourne, Australia) was performed in PMC42-ET cells. These cells display a molecular phenotype of Basal B (E Tomaskovic-Crook and T Blick, unpublished observation), based on clustering of a limited number of the Basal B discriminator genes [19] showing reliable data in an Affymetrix U133A analysis kindly performed by the laboratory of Joe Gray, Lawrence Berkeley National Laboratory, Berkeley, California [59]. These IDC-derived cells express E-cadherin mRNA and protein but do not assemble it at the cell membrane [27]. PMC42-ET were grown in RPMI with 10% foetal bovine serum (FBS, Thermo Fisher Scienti c, Australia) at 37 o C with 5% CO 2 .
The siRNA target sequences for ITGB1, ITGA2, and ILK are presented in Table 3

Cell matrix adhesion assay
Wells of 24-well plates (polystyrene, non-tissue culture treated; Nunc Inc., Naperville, IL) were coated with 100 µg/ml collagen-I, 100 µg/ml collagen-IV, 20 µg/ml bronectin or 50 µg/ml laminin. Proteins were allowed to bind to the cells overnight at room temperature under the laminar ow hood, before the wells were rinsed with phosphate buffered saline (PBS) and non-speci c interactions were blocked for 1 hr at 37°C with 3% bovine serum albumin (BSA) in PBS, pH 7.4. PMC42-ET cells were transfected with siRNA.
Eight hours later, cells were either left unstimulated, or stimulated with EGF for 72 hrs, after which the cells were detached using 0.25% trypsin and allowed to attach to the various (ECM substrate-coated plates for 1hr. Cells attached to ECM was estimated by the average of cell counts from ve random highpower elds under light microscopy (counted in situ on the substrate after washing). Collagen 1, collagen IV, laminin (from Engelbreth-Holm Swarm murine sarcoma), and bronectin (from bovine plasma) were all purchased from Sigma-Aldrich (St. Louis, MO, USA).

Monolayer wound healing assay
This was performed as previously described [57]. Brie y, PMC42-ET cells were plated in a 6-well plate set up in triplicate and incubated at 37°C for 24 hours to allow the formation of a con uent monolayer. The cells were then wounded by using a P200 pipette tip. The wounded monolayers were washed with complete media to remove detached cells. Images of the wounds were taken at 0, 24, and 48 hours. Wound areas at each time point were analysed and quantitated using ImageJ software.

Boyden chamber migration assay
Boyden chamber migration assays were performed as previously described [60]. Brie y, transmigration culture assays were performed using 8 mm-pore Transwell chambers (Corning, USA). Polycarbonate For use of the ILK inhibitor QLT0267 in Boyden Chambers, sub-con uent monolayers of the PMC42-ET cells were either left untreated or treated for 24hrs with this inhibitor (QLT, Inc., Vancouver, Canada) at a nal concentration of 6.25μM prior to assay. Invasion of PMC42-ET cells in vitro was assessed by the invasion of the cells through Collagen-I-coated Transwell inserts. The inhibitor was dissolved in DMSO (0.1%), which was used as the vehicle control.

Statistical analyses
Gene expression data across PDX passages were analysed using the two-tailed Mann-Whitney test (nonparametric) and Pearson's correlation co-e cient, and the IHC intensity changes across PDX passages were analysed using Ordinary One-Way ANOVA. Cell matrix adhesion and Boyden chamber migration assay results were analysed using two-way ANOVA with Dunnett's multiple comparison test. All statistical analyses were performed using GraphPad Prism v7 (GraphPad Software, San Diego, USA).

Results
Histological comparison of the ED03 and EDW01 xenografts with increasing passages through mice The ED03 PDXs was serially passaged up to passage (p) 11, whereas the EDW01 PDX was passaged up to p7. Histologically, the ED03 xenografts displayed a diffuse growth pattern with minimal visible tumour stroma, often growing in long cords of cells, consistent of ILC. By contrast, EDW01 PDX revealed histology consistent with IDC, with clearly visible stromal septae separating growing tumour islands ( Figure 1). In ED03 the stromal collagen was evident only under higher magni cation as it was ner and more pericellular compared with EDW01, in which the thicker stromal cords separated lobules of tumour.
This stroma was of murine origin, as it did not stain with human speci c vimentin antibody (Figure 1).

Assessment of estrogen receptor (ERa) across serial passages.
We investigated the expression of ERa in the ED03 and EDW01 PDX models. As shown in Figure 2 (low power in part A, higher power shown in part B), immuno-reactivity to ERa in the ED03 PDX was approximately 99% tumour cells, and this level of staining was maintained across passages 2-7. The EDW01 PDX also displayed almost 100% positivity for ERa at p3, however this progressively declined to approximately 40% in passage 6 and 7. These relative changes are plotted in Figure 2C. Progesterone receptor (PR) expression was negative in ED03 in all passages and was weak (<15%) in EDW01 at p3, disappearing by p4. HER-2 in both PDXs was negative (data not shown). The clinical approximated subtypes of breast cancer (de ned according to the 2011 St Gallen International Breast Cancer Conference) classi es both ED03 and EDW01 as Luminal A, since Ki67 is less than 14% in both PDXs ( Figure 3) [61].

Immunohistochemistry (IHC) and RT-qPCR quanti cation of EMT markers
To assess any changes in EMT status over sequential passaging, key effector molecules implicated in the EMT process (vimentin/E-cadherin, Twist1, beta-catenin, P120-RasGTPase activating protein [P120-RasGAP], CD24/CD44) and the proliferative marker Ki67 were screened in the ED03 and EDW01 PDX models across the series of passages using IHC (Figure 3, Supplementary Figure 2) and human-speci c RT-qPCR ( Figure 4). E-cadherin immunostaining was almost completely absent in ED03 original patient material, consistent with its lobular carcinoma derivation [23]. We subsequently con rmed a putative somatic missense variant (p.His128Asn, data not shown). Consistent with this, less than 1% of cells expressed E-cadherin in any ED03 PDX passage in mice (Figure 3, top left panel). This was re ected in a relative E-cadherin staining intensity index per cell score of the range 0-0.2 compared with 0.2-0.9 for EDW01 (Supplementary Figure 1). Similarly, beta-catenin was not detectable within the ED03 PDXs, p120-RasGAP was aberrant, with staining observed to be mostly cytoplasmic. By contrast, EDW01 PDXs displayed strong E-cadherin immunostaining (Figure 3, top right panel) and readily detectable RNA levels ( Figure 4). However, this was accompanied by a progressive increase in human-speci c vimentin mRNA expression with each passage (Figure 4): passage 6 material displayed an approximate 8-fold increase (P=0.008), and passage 7 material displayed an approximate 12-fold increase (P=0.024) in comparison to passage 3 material. This was consistent with a signi cant (p>0.05) increase in vimentin protein intensity per cell index across the passages for EDW01, derived from IHC (Supplementary Figure 1). E-cadherin positive tumour cells transitioning to vimentin positivity (possibly remaining E-cadherin positive, see high magni cation inset, Figure 3) were observed in EDW01 PDXs whereas vimentin mRNA and protein in ED03 were almost negligible (Figure 3, 4B, Supplementary Figure 1). Figure 3, positive nuclear Twist1 expression was seen only in EDW01 xenograft tumours, adjacent to regions of necrosis (as indicated by black arrows), however nuclear Twist1 positivity did not increase in abundance across the passages (Supplementary Figure 1). Beta-catenin and P120-RasGAP was also mainly observed in EDW01 at the cell membrane, and corresponded with E-cadherin staining.

As shown in
We went on to further examine the expression of breast cancer stem cell markers CD44 and CD24, as upregulation of CD44 and downregulation of CD24 is observed in breast cancer cell line EMT [19]. ED03 displayed a homogeneous CD44 IHC pattern, which was relatively consistent throughout the passages at the protein and mRNA level (Figure 3, 4, Supplementary Figure 2-3). By contrast, CD44 protein was heterogeneously expressed in EDW01. Although CD24 was upregulated with increasing passages in both models, EDW01 exhibited higher mRNA abundance overall than ED03 (Figure 4), consistent with the appearance (Figure 3) and quanti cation of protein abundance by IHC (Supplementary Figure 3). Within EDW01, but not ED03, there were regions of tumour cells that appeared to lack both CD24 and CD44 (Figure 3).
A parallel study of integrin expression in the PMC42-ET breast cancer cell line induced to undergo EMT with EGF revealed that ITGA2 and ITGB1, and their downstream regulator ILK, were consistently upregulated (Supplementary Figure 4). Hence ILK and these integrins were examined further in the PDX models. Increases in ITGA2 (p7 signi cantly higher than p3) and ITGB1 (p6 signi cantly higher than p3) were observed in ED03 xenograft material, which were maintained ( Figure 4). However, similar to the increased vimentin seen with each passage in EDW01, the levels of human ITGB1 mRNA in the xenografts increased with successive passage, demonstrating an approximately 26-fold increase (p=0.026) at passage 6 in comparison to passage 3 ( Figure 4). Whilst passage 7 showed approximately a 38-fold increase in ITGB1 mRNA levels when compared to passage 3, this did not reach statistical signi cance. However, ITGA2 mRNA expression in EDW01 xenografts (Figure 4) was signi cantly upregulated by approximately 33-fold at passage 7 material when compared to passage 3 (p=0.024). ILK is activated by integrins including a2b1, and mediates a number of signalling responses in relation to survival and proliferation in addition to induction of EMT [62]. A trend was observed towards upregulation of ILK mRNA expression in both ED03 and EDW01 xenografts (Figure 4).
These ndings illustrate that with serial passage EDW01 has accrued features consistent with EMT. The co-induction of the mRNA levels of ITGB1 and ITGA2 in EDW01 implies that they may be important for the EMT process and/or phenotype, because they track with the indices of EMT (decreased E-cadherin and increased vimentin) clearly observed in this model system.

E-cadherin expressing xenograft tumours exhibit more necrosis
Cores of tumours from the EDW01 series at passage 3 exhibited noticeably more necrosis than any other passage from this line, whereas necrosis was minimal or absent in all PDXs across the various passages in the ED03 line (data not shown). We have previously demonstrated that E-cadherin expression is associated with high proliferative rate and observed an association with E-cadherin expression and the appearance of necrotic tissue in actively growing xenografts [18]. We investigated whether an association existed between E-cadherin expression in the EDW01 line and the proportion of necrosis. As shown in Figure 5, a trend was observed such that in the cores that had detectable necrosis (measured as % total area of the core), E-cadherin was generally expressed at a high level, and more highly expressed than vimentin, measured as % total area of the core.

Further investigation of EMT drivers and markers in the EDW01 xenograft model.
Hypoxia is a common driver of EMT in breast cancer, and E-cadherin repressor genes have been implicated in this process [63]. We sought to examine the pattern of Snail1, Snail2, Twist1 and Zeb1/2 expression through the serial passages in mice in the ED03 and EDW01 xenograft models, in comparison to the hypoxic indicator gene carbonic anhydrase 9 (CAIX).
As shown in Figure 6, of the E-cadherin repressor genes examined (Snail1, Snail2, Twist1, Zeb1 and Zeb2), Twist1 was more highly expressed in the EDW01 xenograft compared with ED03 ( Figure 6A, i). Zeb1 and Zeb2 were not expressed at detectable levels in either PDX. Twist1 displayed the highest correlation with CAIX in the EDW01 xenograft model, with both exhibiting a progressive increase from p2 to p5 which was then reduced in p6 to p7 ( Figure 6A, i). Membrane intensity of CAIX ( Figure 6C) aligned with the expression data, suggesting Twist1 may drive hypoxia-induced EMT through consecutive passages of the EDW-01 PDX.
CD24 is an epithelial-associated marker with relevance to breast cancer stem cells, where its expression is reduced in comparison to luminal breast cancer cells [19]; it's expression has been shown to indirectly stimulate cell adhesion to bronectin, collagens I and IV, and laminin through the activation of integrin activity [64]. Interestingly, the expression pattern of CD24 with ITGB1 was signi cantly positively correlated in the ED03 series (R 2 =0.9, p=0.0012) and in the EDW01 series (R 2 =0.96, p=0.023). CD24 was also positively correlated with ITGA2 in the ED03 series (R 2 =0.84, p=0.0035) and this reached near signi cance in the EDW01 series (R 2 =0.76, p=0.054) ( Figure 6C). No signi cant or near signi cant correlations were observed for CD44 with integrins in either PDX systems ( Figure 6B, iii).
Functional assessment of candidate genes ITGB1, ITGA2 and ILK in the PMC42 system As expression of the α2β1 integrin components were associated with the EMT observed in EDW01 xenografts over serial passages through mice, we tested whether they could be a "driver" of the EMT, using the PMC42 EMT model system.
Although already somewhat mesenchymal [27], PMC42-ET cells treated with EGF in vitro undergo a further EMT in which ILK, ITGB1 and ITGA2 are upregulated (supplementary Figure 1), [63]. We examined the effects of siRNA knockdown of ITGA2, ITGB1 and ILK on late stage mesenchymal gene expression, cell adhesion and cell migration.
As shown in Figure 7A, individual siRNA knockdown of ITGB1, ITGA2 or ILK resulted in the expected reduction of expression of the target genes. In addition, ILK inhibition also led to the reduction of ITGB1 protein levels by 80% (0.2 in Figure 7A); inhibition of ITGB1 also led to reduced ITGA2 expression (0.06 in Figure 7A)); and, ILK inhibition led to a reduction in ITGA2 expression by 60% (0.41 in Figure 7A).
Inhibition of ITGB1 and ITGA2 by siRNA did not affect ILK protein levels. These data indicate a complex interplay between these three components To determine whether the cells with suppressed ITGB1, ITGA2 or ILK were able to undergo EMT with EGF treatments, cellular morphology and protein expression was examined. After 72 hours of EGF treatment, cellular morphology ( Figure 7B) revealed a clear acquisition of spindle-shapes and breaking apart of cellular islands, consistent with an EMT. Protein expression of the "classical" indicators of EMT, vimentin, E-cadherin, and N-cadherin were measured by Western immunoblotting ( Figure 7C). E-cadherin was dramatically reduced in the "cells alone" treated with EGF compared to the untreated "cells alone" control, whereas slight increases in vimentin and N-cadherin were observed in these untransfected cells, consistent with EGF-induced EMT of these cells as previously reported [27,63].
No observable differences in the responses of vimentin, N-cadherin or E-cadherin protein levels to 72hrs of EGF treatment were seen following treatment with ITGB1, ITGA2 or ILK siRNA, when compared to treatment with control siRNA ( Figure 7C). Similarly, the mesenchymal morphology caused by EGF treatment was not abrogated by any of the siRNAs ( Figure 7B).
Together, this indicates that these candidates do not directly mediate the EMT induced by EGF in this breast cancer cell line.
We then investigated what effect the stepwise increase in integrin expression, observed in the EDW01 xenografts ( Figure 4) may have had on growth and invasion/motility of these tumours over serial passages in mice, inferred by parallel analyses in the PMC42-ET cells. We focused on cell adhesion to various substrates and migration, properties known to be mediated by integrins and ILK, and these were again examined in PMC42-ET cells.
PMC42-ET cells require ITGB1, ITGA2 and ILK for maximal adherence to collagen I, collagen IV, laminin and bronectin substrates, as knockdown of these molecules signi cantly reduced adhesion in comparison to control siRNA ( Figure 8A-D, p<0.05). When stimulated with EGF, the attachment of the PMC42-ET cells treated with ITGB1, ITGA2, and ILK siRNA was also signi cantly abrogated (p<0.001).
Coordinated regulation of cell adhesion and adhesion complex remodelling are crucial for cell movement. ITGB1, ITGA2, and ILK siRNA-mediated silencing in PMC42-ET cells caused them to be signi cantly less migratory, as shown by the assessment of migration using the Boyden Chamber Assay ( Figure 9A) and the Monolayer Wound Healing assay ( Figure 9B). Inhibition of ILK by a speci c inhibitor, QLT0267, also signi cantly reduced cellular movement in the Boyden Chamber Assay (Figure 9C, p<0.001). EGF treatment caused these cells to increase their migration, whereas ITGB1, ITGA2, and ILK silencing each signi cantly reduced migration in both assays under EGF-stimulated conditions ( Figure 9A, B, p<0.05).

Discussion
We have shown that the EDW01 PDX model displayed evidence of EMT with progressive passages through mice, which was not seen in ED03. This is consistent with the known EMT status of IDC versus ILC, of which EDW01 and ED03 are examples of, respectively. However a comparison of the progression of such divergent cancer types with regards to EMT status has enabled the discovery of some unique ndings. The partial EMT observed in EDW01 was associated with a rising hypoxia leading to Twist1 expression in early-mid passages, repressing E-cadherin expression and orchestrating vimentin upregulation, and accompanied by upregulation of ITGB1 and ITGA2 expression. The mesenchymal shift appeared to then return to the epithelial direction in later passages of EDW01, however the increased integrin expression persisted. We present cell line data to support the association of integrin a2b1 and the ILK signalling pathway with the observed EMT, but that the EMT was not mediated by these.
Although the EMT-associated ITGB1, ITGA2, and ILK are not essential for EGF-induced EMT in PMC42ET cells (Figure 7) they are necessary for breast cancer cell adhesion to ECM-substrates ( Figure 8) and cellular movement (Figure 9). These molecules were more signi cantly upregulated in increasing EDW01 passages through mice than ED03 (Figure 3,4), thus they may have enabled ECM adhesion in this PDX.
Indeed, previous studies demonstrate that the α2β1 integrin is primarily a receptor for collagen and laminin [65] and expression is also associated with motility, invasiveness, and cellular differentiation of a variety of tumours [66,67]. This is in contrast to studies in which ITGA2 and ITGB1 have been found to suppress metastasis in models of mouse and human cancer [68]. However, Dedhar and Saulnier [69] showed that the expression of α2β1 integrin increased in the chemically transformed human osteosarcoma cells, and this integrin was implicated in tumour progression and metastasis. Similarly, α2β1 integrin expression accelerated either experimental metastasis or tumour dissemination of melanoma [70] and rhabdomyosarcoma [71,72], gastric cancer [73,74] and colon cancer cells [75]. Taken together, our data suggest that a2b1 integrins contribute to the EMT phenotype observed increasingly in EDW01 over serial passages in mice.
The PDX models of ED03 and EDW01 were characterised as luminal A (ER positive, Her2 negative, PR low or absent) whereas the PMC42-ET breast cancer cell line used in this study was of the Basal B molecular phenotype [63]. We acknowledge that this is a limitation of our study which directly in uences the conclusions that we can logically make in regards to the importance of integrins a2b1 in cancer progression. However, despite being typically luminal, the EDW01 displayed features of a cumulative EMT over passages in mice, likening it to the more basal phenotype of PMC42-ET. Consistent with this suggestion, integrin a2b1 increased in expression in EDW01 over passages , as it did in PMC42-ET.
Because of the relative paucity of EMT in luminal systems, we chose to analyse the signi cance of integrin a2b1 in the EMT-positive basal breast cancer system which is PMC42-ET.
Actively growing tumours acquire areas of hypoxia, a product of imperfect angiogenesis coupled with rapid growth, which can facilitate cellular invasion via the induction of EMT [76]. Vimentin and Twist1 positivity was observed in close proximity to necrotic areas in early passages and less commonly found at the centre of tumour 'islands' (Figure 3). Of the E-cadherin repressor genes examined, Twist1, a target of HIF-1a [77][78][79] displayed the strongest correlative pattern of induction with CAIX ( Figure 6B, R 2 =0.81, p=0.04). Induction of Twist1 coincided with the repression of E-cadherin and induction of vimentin mRNA, and therefore may be the instigator of the observed EMT in the EDW01 xenograft model. Indeed, hypoxia has been implicated in inducing EMT-related genes in another PDX model of serial transplantation. Wegner and colleagues [80] demonstrate in their cervical cancer PDX model serially transplanted in mice, that the EMT orchestrating gene Snail1 and stem cell markers were found to be increased in late compared to early passages along with hypoxic CAIX gene expression, accompanied with an increase in tumour aggressiveness and proliferative rate. Their nding, in a different cancer type (cervical) adds further support to our supposition that hypoxia was a major driving force in the observed progressive EMT in the EDW01 PDX model.
However, why did the mesenchymal shift return to epithelial in later passages of EDW01? Tumours in vivo to have been found to adapt to low oxygen environments, such as reprogramming Akt signalling in the mitochondria [81]. This coupled with the well-known ability of tumours to increase angiogenesis [82] contributes to tumour cell survival and progression. Although beyond the scope of this investigation, the EDW01 PDX model provides a means to investigate these phenomena further, with relevance to understanding the progression of breast cancer in women.
We have previously shown that MDA-MB-468 xenografts, which express E-cadherin, exhibit hypoxiarelated necrosis and subsequent EMT, features of which were lost upon E-cadherin knockdown [18]. In the current study we also observed necrosis at p4 in EDW01 but not ED03. This is consistent with EDW01 exhibiting a high grade of ductal tumour, and is rarely observed in ILC, of which ED03 is an example.
Furthermore, the necrosis at p4 in EDW01 was observed to coincide with the induction of the hypoxic indicator gene CAIX ( Figure 5A, 6A, C). IDC, the classi cation of EDW01 PDX, generally proliferate at a higher rate than ILC [83], of which ED03 is classi ed. In our previous study, E-cadherin knockdown MDA-MD-468 xenografts grew slower than their control counterparts [18]. In contrast, EDW01 (IDC) and ED03 (ILC) displayed similar Ki67 expression ( Figure 5C) and we did not nd signi cant differences in implantation to harvest time durations between ED03 and EDW01 PDXs (data not shown). However, changes in daily growth rate cannot be ruled out between EDW01 and ED03, as apoptosis was not monitored. Regardless, the results from the current study reinforce a direct connection between Ecadherin tumoural expression and the appearance of hypoxia.
Although many studies have associated EMT with therapy resistance [11,84], it is important to note that the EDW01 xenograft was not challenged by therapy; the EMT progression was spontaneous.
Interestingly, considerable emphasis is being placed recently on the hybrid state of EMP recently [6,[85][86][87][88], and this phenotype appears to manifest in the EDW01 xenografts (elevated vimentin and apparently only partially lost E-cadherin - Fig 3, 4). A separate analysis of circulating tumour cells (CTCs) in the ED03 model indicate that despite the lack of any evidence of EMT in the primary xenograft tumours, the CTCs are enriched in mesenchymal gene expression, but also in epithelial genes (E-cadherin and CD24), compared to the primary tumour, indicating a dysregulation of this axis and/or possibility of hybrid cells [58].
Although the PDX models examined had varied ERa positivity in early passages, this increased in both models over serial transplantation in mice ( Figure 2). Mice used in this study for serial xenograft propagation were not administered estradiol, which is often used to foster the growth of ERa-positive xenografts, which generally have a lower engraftment rate than ERa negative tumours [89]. ERa positivity is likely to have been maintained/increased because estradiol was not administered, as ERa undergoes ligand-dependent downregulation [90]. Indeed, removal of estrogen for several weeks in PDXs has been found to increase ERa levels [89].
Tumour-stroma crosstalk plays an integral role in EMT in vivo [91], similarly, we observed key changes in murine stroma in the PDX models examined in this study. We have previously demonstrated that EDW01 evoked greater expression of MMPs (-2, -9, -11 and MT1-MMP) in the murine stroma than ED03 [56]. Furthermore, EDW01 displayed MT1-MMP and MMP-13 at the tumour-stromal boundary, but did not express these factors or MMP-2 and -9 within the tumour mass itself. As shown in the current study, the EDW01 tumours grew as islands traversed by thick collagenous stromal bands whereas the ED03 had delicate pericellular stroma dispersed throughout ( Figure 1). This pattern of growth may be directly attributable to the pattern of MMP expression of each of these PDX models -EDW01 lacked the capacity to invade as individual cells, possibly due to the lack of induction of intratumoral MMP-2 and -9.
Furthermore, the murine microenvironment (non-orthotopic) in which the EMT occurred in the EDW01 PDX over successive passages may have been conducive to this change. We found that murine (stromal) ITGB1 expression aligned with human (tumoural) expression of the same integrin, in fact for EDW01, stromal ITGB1 expression was approximately 22-fold higher than tumoural ITGB1 at passage 7 ( Figure  4). This leads to speculation as to whether the murine microenvironment was the instigator of the EMT or a reponder in this process. However, given that an EMT was not observed in the ED03 line, which was passaged through mice of the same genotype (SCID), it could be postulated that drivers of EMT came from within the tumour itself, lending further support to the notion that hypoxia was an initiating event.
Decreased CD44/CD24 expression ratio in later passages in both PDX lines was an unexpected nding, at least in EDW01, as CD24 +/high /CD44 lowphenotype is associated with the epithelial phenotype despite EMT being observed in this xenograft model. However, CD24 expression can also confer adhesive properties enabling invasion. In a meta-analysis of 16 studies of 5,697 breast cancers, CD24 was found to be signi cantly associated with poorer survival [92], presumably due to non EMP functions. In studies on breast cancer cell lines in vivo, CD24 was found to act as a ligand for P-selectin on the lung vascular endothelium [64]. We found that CD24, but not CD44, correlated with ITGA2 and ITGB1 in both PDX models ( Figure 6B ii versus iii), providing further evidence in addition to our PMC42-ET integrin knockdown/EGF studies (Figure 7) that activation of these integrins is not necessarily intimately related to the EMT process.
Posttranslational cleavage of CD44 may explain the discrepancy between CD44 protein expression between the two PDXs -EDW01 CD44 gene expression is comparable with ED03 ( Figure 4) whereas CD44 membranous protein expression is strong and uniform in ED03 but somewhat weaker and more heterogeneous in EDW01 (Figure 3, Supplementary Figure 2-3). CD44 may be shed from the cell by the action of MMPs, namely MMP-9 and MT1-MMP [93,94], resulting in the loss of cell membrane CD44. The CD44 antibody used in this study (clone 156-3C11, abcam) recognizes cell-membrane localized CD44. As previously mentioned in our earlier studies, EDW01 evoked greater expression of MMPs (-2, -9, -11 and MT1-MMP) in the murine stroma than ED03 [56]. As shown in Supplementary Figure 1 [95], which can result in the transcriptional activation of EMT-associated genes [96,97] and induction of stemness [98]. Nuclear CD44 has been shown to occupy the Twist1 promoter [98] therefore CD44 cleavage in EDW01 could have contributed to Twist1 transcriptional upregulation ( Figure 6A).

Conclusion
This study illustrates a cumulative EMT in a PDX system through several passages in mice, an effect characterized by, but not orchestrated by, ILK signalling via ITGA2 and ITGB1, and possibly ILK.

Declarations
Ethics approval and consent to participate All human tissue samples used in this study and associated approval and consent to participate was obtained in accordance with the standards of the St. Vincent's Hospital Human Ethics Committee.

Consent for publication
Not applicable.

Availability of data and materials
All data generated or analysed during this study are included in this published article and its supplementary information les.

Competing Interests
The authors declare that they have no competing interests. Author's contributions RW generated 50% of the real-time PCR data and all siRNA work including associated Western blots. EDW performed EDW01 PDX generation initiation and harvest, and maintained records. EdeS performed all ED03 generation initiation and harvest, and maintained records. PTS and JRK performed P120, Ki67 and beta-catenin IHC and PTS provided direct guidance in scoring. AER performed ED03 CDH1 mutation analyses. MW assisted in ED03 and EDW01 generation initiation and harvest. CES assisted in hormone receptor IHC interpretation and PDX morphology analysis. TB assisted in all statistical analysis. EWT and HJH designed the study and provided intellectual input as to the direction of the work and HJH performed 50% real-time data, remaining IHC, and drafted the manuscript. All authors reviewed and edited the manuscript.  Table 3 Sequences of siRNA used in the current study.  bronectin, by integrin subunit-speci c and ILK siRNAs. Results are expressed as % of cells adhered and represent the mean ± SD from 4 biological replicates. Results were analysed using two way ANOVA, with Dunnett's multiple comparison test, with P values adjusted for multiple comparisons; * = p<0.01, ** = p<0.001, *** = p<0.0001.

Figure 9
Migration properties of PMC42-ET cells following siRNA knockdown of ITGB1, ITGA2 and ILK without and with stimulation by EGF. Effect on migration was assessed using the (A) Boyden Chamber and (B)