Differential expression of αVβ3 and αVβ6 integrins in prostate cancer progression

Neuroendocrine prostate cancer (NEPrCa) arises de novo or after accumulation of genomic alterations in pre-existing adenocarcinoma tumors in response to androgen deprivation therapies. We have provided evidence that small extracellular vesicles released by PrCa cells and containing the αVβ3 integrin promote neuroendocrine differentiation of PrCa in vivo and in vitro. Here, we examined αVβ3 integrin expression in three murine models carrying a deletion of PTEN (SKO), PTEN and RB1 (DKO), or PTEN, RB1 and TRP53 (TKO) genes in the prostatic epithelium; of these three models, the DKO and TKO tumors develop NEPrCa with a gene signature comparable to those of human NEPrCa. Immunostaining analysis of SKO, DKO and TKO tumors shows that αVβ3 integrin expression is increased in DKO and TKO primary tumors and metastatic lesions, but absent in SKO primary tumors. On the other hand, SKO tumors show higher levels of a different αV integrin, αVβ6, as compared to DKO and TKO tumors. These results are confirmed by RNA-sequencing analysis. Moreover, TRAMP mice, which carry NEPrCa and adenocarcinoma of the prostate, also have increased levels of αVβ3 in their NEPrCa primary tumors. In contrast, the αVβ6 integrin is only detectable in the adenocarcinoma areas. Finally, analysis of 42 LuCaP patient-derived xenografts and primary adenocarcinoma samples shows a positive correlation between αVβ3, but not αVβ6, and the neuronal marker synaptophysin; it also demonstrates that αVβ3 is absent in prostatic adenocarcinomas. In summary, we demonstrate that αVβ3 integrin is upregulated in NEPrCa primary and metastatic lesions; in contrast, the αVβ6 integrin is confined to adenocarcinoma of the prostate. Our findings suggest that the αVβ3 integrin, but not αVβ6, may promote a shift in lineage plasticity towards a NE phenotype and might serve as an informative biomarker for the early detection of NE differentiation in prostate cancer.

provided support in the form of a salary for PW. The specific roles of this author are articulated in the 'author contributions' section. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests:
The authors of this study have read the journal's policy and have the following competing interests: PW is an employee and a shareholder of Biogen Inc. Biogen holds patents covering avb6 antibodies and their uses for therapeutic purposes. However, this paper does not deal with the use of these antibodies for therapeutic purposes; these antibodies have been used just for immunoblotting in Fig 5. This does not alter our adherence to PLOS ONE policies on sharing data and materials.
pathways that might promote lineage plasticity among PrCa subtypes for which there is no established therapeutic approach. The differential expression of these two lineage-restricted integrins might also serve as a useful biomarker to predict neuroendocrine differentiation and facilitate patient stratification in PrCa.

Generation of mice carrying prostate-specific TRP53 and RB1 gene deletions
Mice of genotype PB-Cre4 PTEN loxP/loxP , PB-Cre4 PTEN loxP/loxP RB1 loxP/loxP , or PB-Cre4 PTEN loxP/loxP RB1 loxP/loxP TRP53 loxP/loxP were generated as previously described [36,37]. Briefly, mice carrying different combinations of the PTEN loxP , RB1 loxP , and TRP53 loxP alleles were interbred, with the ARR2PB-Cre transgene from the PB-Cre4 line always carried through males. Mice used in this analysis are on a C57BL/6 and 129SVJ mixed genetic backgrounds. Mice were backcrossed to the C57BL/6 strain for at least 5 generations. Genotypes were designated as SKO (single PTEN knock-out), DKO (double PTEN:RB1 knock-out), and TKO (triple PTEN:RB1:TRP53 knock-out). Non-recombinant littermates were used as a control. The mice were euthanized using CO 2 and cervical dislocation when the tumor length was approximately 2 cm. All of these mice were maintained following guidelines of the Institutional Animal Care and Use Committee (IACUC), and were bred and kept at Roswell Park Comprehensive Cancer Center (Buffalo, NY, USA).

TRAMP (Transgenic Adenocarcinoma of the Mouse Prostate) mice
Male TRAMP mice were generated as described previously [38]. Twenty-four male TRAMP mice were used. No female mice were analyzed in this study. The mice were euthanized using CO 2 and cervical dislocation when the tumor volume was approximately 10,000 mm 3 . Care of animals was in compliance with standards established by the Office of Laboratory Animal Welfare, NIH, Department of Health and Human Services. All mice were maintained following recommendations of the IACUC. Experimental protocols were approved by IACUC.
tissue acquisition necropsy in a manner which limited warm ischemic time as much as possible (aiming for 4-8 hours after death). A few samples of primary PrCa were obtained from surgical procedures. Harvested tumor tissues were evaluated by pathologists, and viable tumor tissue was macro-dissected to minimize content of stroma, fat, and necrotic tissue. Tumor fragments were implanted subcutaneously in 6-to 8-week-old intact male athymic Nu/Nu (NU-Foxn1nu) or CB-17 severe combined immunodeficient (SCID, CB17/Icr-Prkdcscid/IcrCrl) mice (Charles River Laboratory). Tumor samples were harvested from later passages (>3) and frozen or embedded in paraffin for characterization. LuCaP PDXs are maintained by constant passaging in SCID mice. The levels of SYP in the LuCaP PDX were assessed by IHC analysis.

PLOS ONE
The αVβ3 integrin is a marker of neuroendocrine prostate cancer

LuCaP TMA immunohistochemical assessment and statistical analysis
LuCaP PDX TMA immunostaining was scored by multiplying each staining intensity level ("0" for no stain, "1" for faint stain, and "2" for definitive stain) by the percentage of cells at each staining level. The multiplicands provided a final score for each sample (score range was 0 to 200). The score for each LuCaP core was the average of the scores of each triplicate. Relative detection levels of SYP were provided by Dr. Corey and defined as 0 (-), 1 (+), 2 (++), and 3 (+++). The normalization was performed by assigning to the higher score for each immunostaining (αVβ3, αVβ6, and SYP) a value of 100. Correlation analysis between the integrin scores and the expression levels of SYP and its significance was performed using Spearman correlation (Matlab v.R2016a).

RNA-sequencing (RNA-seq)
RNA-seq was performed as previously reported in [39] and publicly available on GEO Expression Omnibus (accession number: GSE90891). Briefly, RNA-seq was performed on SKO (n = 4), DKO (n = 5), and TKO (n = 4) prostate tumors and on normal prostate (n = 4) by the Roswell Park Cancer Institute Genomics shared resource. Sequencing libraries were prepared with the TruSeq Stranded Total RNA kit (Illumina Inc) from 1 μg total RNA following manufacturer's instructions. After ribosomal RNA depletion, RNA was purified, fragmented, and primed for cDNA synthesis. Fragmented RNA was reverse transcribed into first-strand cDNA using random primers. AMPure XP beads were used to separate the cDNA from the secondstrand reaction mix resulting in blunt-ended cDNA. A single 'A' nucleotide was then added to the 3' ends of the blunt fragments. Multiple indexing adapters, containing a single 'T' nucleotide on the 3' end of the adapter, were ligated to the ends of the cDNA to prepare them for hybridization onto a flow cell. Libraries were purified and validated for the appropriate size on a 2100 Bioanalyzer High Sensitivity DNA chip (Agilent Technologies, Inc.). The DNA library was quantitated using KAPA Biosystems qPCR kit and normalized to 2 nM prior to pooling. Libraries were pooled in an equimolar fashion and diluted to 10 pM. Library pools were clustered and run on a HiSeq2500 rapid mode sequencer according to the manufacturer's recommended protocol (Illumina Inc.).
Raw sequencing reads passing the Illumina RTA quality filter were pre-processed using FASTQC for sequencing base quality control. Reads were mapped to the mouse reference genome (mm9) and RefSeq annotation database using Tophat. A second round of quality control using RSeQC was applied to mapped bam files to identify potential RNA-seq library preparation problems. The number of reads aligning to each gene was calculated using HTSeq, and for each gene, the corresponding RPKM value was calculated.
For differential gene expression analysis, RNA-seq counts were processed to remove genes lacking expression in more than 80% of samples. Scale normalization was done using the Limma package in R. After Voom transformation, data from primary SKO, DKO, and TKO tumors were compared to generate differentially expressed gene lists with P < 0.05 and logFC > 1.5.

The αVβ3 integrin is selectively upregulated in NEPrCa murine models
In a recent study, we have shown that the αVβ3 integrin is found in small extracellular vesicles released by cancer cells and that small extracellular vesicles containing αVβ3 have a unique ability to promote NED of PrCa in vivo [7]. Based on these findings, we hypothesized that elevated expression levels of αVβ3 might correlate with NED in PrCa. We tested this hypothesis by analyzing the levels of αVβ3 and αVβ6 integrins in primary tumors, as well as lung metastatic lesions, from NEPrCa mice carrying PTEN, RB1, and TRP53 triple conditional knockouts in the prostatic epithelium (PBCre4 PTEN loxP/loxP RB1 loxP/loxP TRP53 loxP/loxP , TKO). This model has been reported to develop NEPrCa similar to its human counterpart [39]. We compared the TKO model to a double knock-out model lacking PTEN and RB1 in the prostate (PBCre4 PTEN loxP/loxP RB1 loxP/loxP , DKO). In addition, we analyzed a PTEN single conditional knock-out mouse model (PBCre4 PTEN loxP/loxP , SKO) whose gene expression signature has been shown to be comparable to human ADPrCa [39]. The immunostaining analysis reveals high levels of the αVβ3 integrin (Figs 1 and 2, top panels) which correlate with SYP expression (Figs 1 and 2, bottom panels) in the prostate tumors (Fig 1) and lung metastatic lesions (Fig 2) of DKO and TKO mice (n = 5 for each group). The results are consistent in all samples except for one of the DKO samples which does not exhibit detectable αVβ3 integrin expression. In the tumors from the SKO mice, the αVβ3 integrin is not detectable (Fig 1, top panels), whereas the αVβ6 integrin is highly expressed in SKO prostate tumor samples (Fig 1, middle panels), and is low with some patchy positivity in the DKO and TKO primary tumors (Fig 1, middle  panels). Consistent with these results, lung metastatic lesions from DKO and TKO mice show some patchy positivity for the αVβ6 integrin (Fig 2, middle panels) but at a considerably lower level than for αVβ3. We did not observe any metastases in SKO mice.
Consistent with the immunostaining results, RNA sequencing analysis of the publicly available datasets on Geo Expression Omnibus (GSE90891, [39]) reveals higher levels of the β3 integrin subunit (ITGB3) expression in DKO and TKO tumors compared to SKO samples. Moreover, ITGB3 mRNA is upregulated in SKO compared to normal prostate (wild type, WT; Table 1), although our immunostaining analysis does not detect the αVβ3 integrin in the SKO samples analyzed (Fig 1). These results indicate that, although the ITGB3 mRNA is present in SKO tumors, the mRNA is likely to be unstable. In addition, the levels of the αV integrin subunit (ITGAV) and β6 integrin subunit (ITGB6) are lower in DKO and TKO tumors compared to SKO, although noticeably higher in all three knock-out genotypes compared to normal prostate (WT) samples (Table 1).
NEPrCa expresses elevated levels of PARP1 which is a nuclear enzyme involved in DNA repair, DNA replication, inflammation, and chromosome organization [40,41]. Consistent with these previous publications, PARP1 expression is upregulated in DKO and TKO tumors ( Table 1). Although PARP1 mRNA is also upregulated in SKO samples compared to the WT control, the levels of PARP1 mRNA are not as elevated as in DKO and TKO tumors (Table 1). In addition, another gene involved in NED (BRN4 [POU3F4]) [42] is upregulated in DKO and TKO samples but not in SKO (Table 1). These results demonstrate that high expression of ITGB3 and of genes implicated in NED co-occur in DKO and TKO tumors. We also performed immunohistochemical analysis of tumor samples from TRAMP mice to assess the levels of αVβ3 and αVβ6 integrin expression in their tumors. This mouse model, which is known to have RB and p53 inactivated, develops NEPrCa together with ADPrCa [43,44]. Our immunostaining shows that the NE marker chromogranin A (CgA) co-occurs with the αVβ3 integrin in 10 of the 13 TRAMP NE tumor samples analyzed (Fig 3). The αVβ6 integrin, however, is not detected in the NE tumors from the TRAMP model (Fig 3). In contrast, the αVβ6 integrin is detected exclusively in the ADPrCa, NE-negative areas of the TRAMP tumor samples (Fig 3). Our results, from the DKO and TKO NE mouse genetic models as well as the TRAMP mice, taken together, clearly demonstrate a consistent correlation between the high expression of αVβ3 integrin and NEPrCa occurrence. Conversely, ADPrCa tumors are consistently associated with expression of the alternative αVβ6 integrin subtype.

Expression of αVβ3 integrin and synaptophysin correlates in patientderived xenografts
To confirm these results in human specimens, we conducted an immunohistochemical analysis of 42 LuCaP PDXs [31,32]. These PDX models were generated by implanting primary PrCa or metastatic lesion tumor fragments from PrCa patients into immunocompromised mice [32], and the resulting PDX models were subsequently characterized for their expression of NE markers [31]. We assessed the presence of αVβ3 or αVβ6 integrin using immunohistochemical analysis and scored the immunostaining intensity of each LuCaP core in the tumor micro-array (TMA) using the scoring system described in the Materials and Methods section. We observe a positive correlation between the αVβ3 integrin and the NE marker SYP (Fig 4A and 4B, r = 0.42; P = 0.0046). In contrast, the αVβ6 integrin shows no correlation with SYP (Fig 4A and 4B, r = 0.22; P = 0.1622), confirming the results described above obtained for mouse tumor samples.
We further validated the results obtained using the LuCaP PDX TMA by screening PrCa samples from the Department of Pathology at Thomas Jefferson University and the Cooperative Human Tissue Network. Of the 7 ADPrCa primary tumors none expresses αVβ3 (Fig 4C). On the other hand, as previously reported [21], most of the ADPrCa express αVβ6 which was used as positive control. These findings suggest a differential expression of these two αV integrins during PrCa progression, whereby the αVβ3 integrin is specifically expressed in NEPrCa samples, and in contrast, the αVβ6 integrin is specifically expressed in ADPrCa samples lacking NE characteristics.

Loss of androgen receptor signaling does not result in upregulation of αVβ3 or αVβ6 integrin expression in PrCa cell lines
NEPrCa is characterized by the activation of pro-tumorigenic pathways independently from the AR signaling [28]. We hypothesized that loss of AR signaling might induce upregulation of the αVβ3 integrin in LNCaP and C4-2B, two AR positive PrCa cell lines. To test our hypothesis, we downregulated AR expression in LNCaP and C4-2B cells using siRNA. Our results show that downregulation of AR in C4-2B or LNCaP cells does not upregulate αVβ3 (Fig 5A) or αVβ6 integrin (Fig 5B) expression. Thus, it is possible that other factors in the tumor microenvironment contribute to the regulation of αVβ3 integrin and αVβ6 integrin expression after AR signaling loss.

Discussion
Our results demonstrate that increased expression of the αVβ3 integrin correlates with the occurrence of NE markers in human patients' samples and murine models. In contrast, the

PLOS ONE
The αVβ3 integrin is a marker of neuroendocrine prostate cancer αVβ6 integrin is expressed in human and murine ADPrCa, suggesting that the αVβ3, but not αVβ6, integrin might serve as a suitable biomarker to characterize NED in the context of PrCa.
Here, we show that these two integrins are differentially expressed in ADPrCa and NE cancers. Specifically, expression of the αVβ3 integrin in primary prostate tumors and metastatic lesions of mice carrying deletions of the PTEN (SKO), RB1 and PTEN (DKO) or RB1, PTEN, and TP53 (TKO) inversely correlates with αVβ6. Expression of the αVβ3 integrin in primary prostate tumors of mice carrying deletions of the PTEN (SKO) is undetectable, while it is significantly increased in DKO or TKO tumors and metastatic lesions. This indicates that RB1 loss, and consequent activation of transcription factors of the E2F family [45][46][47], is sufficient to induce αVβ3 expression in these models. This integrin expression persists in TKO tumors which, in contrast to DKO tumors exhibiting both SYP and AR expression, develop homogenous AR-negative NEPrCa, similar to its human counterpart [39]. It remains to be investigated whether downregulation of αVβ6 and gain of the αVβ3 integrin occur in CRPrCa since RB1 is known to influence integrin expression [48,49], and its loss occurs frequently in human CRPrCa [50,51].
A factor that may influence the processing of the αVβ3 integrin, is the expression of the αV subunit which is required for the heterodimeric complex. The RNA analysis summarized here (Table 1) indicates that the levels of the αV integrin subunit (ITGAV) become limiting and that β3 acts in a dominant fashion over the β6 integrin subunit. We also detect high αVβ3 integrin expression in the NE areas of primary tumors from TRAMP mice that develop NEPrCa together with ADPrCa. In contrast, we detect the related αVβ6 integrin in the ADPrCa areas of the TRAMP tumors. Our findings underline the specificity of the αVβ3 integrin in NEPrCa, nominating this integrin as a potential biomarker for patient stratification in PrCa treatment. Our future studies will benefit from the use of mice carrying deletion of the αVβ3 integrin crossed with the DKO, TKO, or TRAMP mice, in order to shed new light on the mechanism of action of the αVβ3 integrin in NEPrCa development and/or metastatic progression.
Multiple strategies have been developed to target the αVβ3 integrin due to its role in tumor angiogenesis and tumor growth [16]. For example, LM609, an inhibitory antibody against the αVβ3 integrin, reduced angiogenesis and tumor growth in a SCID mouse/human chimeric model for breast cancer [52]. Its humanized counterpart JC-7U IgG1 has been reported to inhibit tumor growth in a Kaposi sarcoma mouse model and was also able to inhibit, in part, the binding of human immunodeficiency virus (HIV-1) Tat protein to αVβ3 integrin, which is necessary to stimulate Kaposi sarcoma growth [16,53]. Previous studies also reported the ability of the αVβ3 integrin to support metastasis in PrCa [54] as well as other cancers [55][56][57][58]. Likewise, the expression of the αVβ3 integrin conceivably facilitates the metastatic behavior of NEPrCa. In support of this idea, our SKO mouse model (PB-Cre4 PTEN loxP/loxP ) does not metastasize and expresses low levels of αVβ3 integrin, whereas DKO and TKO, the two NE models that acquire αVβ3 integrin expression as a consequence of additional RB1 knock-out, develop metastases in the lungs [39]. We can speculate that upon RB1 loss, downregulation of αVβ6 and gain of the αVβ3 integrin are required in the primary tumors in the early stages of NED to confer upon NEPrCa the ability to metastasize in different sites (Fig 6). Upon metastasizing, the αVβ3 integrin expression is sustained as shown here and as previously described [7] in NEPrCa bone metastasis, indicating additional pro-survival functions provided by this integrin.
Whether one or more of the many pathways activated by the αVβ3 integrin is involved in NED remains to be established. For example, the expression of the αVβ3 integrin reportedly stimulates cell migration by activation of the phosphatidylinositol 3-kinase (PI3K)/AKT pathway [59]. Other studies have demonstrated that AKT1 is involved in stabilizing N-MYC [60,61], one of the main promoters of NED in PrCa [62]. Since pAKT is not detectable in TKO prostate tissue [39], we speculate that pAKT activated by αVβ3 primes the cells to stabilize N-MYC but is not required for long-lasting NED. The RNA-seq analysis presented here highlights potential downstream effectors of αVβ3. For example, αVβ3 integrin might be able to induce NED in PrCa by upregulating Trop2 expression, which is known to induce NEPrCa by upregulation of PARP1 [40]. Underlining the importance of targeting this pathway to prevent or delay the most aggressive forms of PrCa, the U.S. Food and Drug Administration has recently approved olaparib, a PARP1 inhibitor [63], for the treatment of metastatic CRPrCa. However, there are as yet no reports on the safety or efficacy of olaparib for the treatment of NEPrCa.
Our previous study demonstrates that the dysregulated expression of the αVβ3 integrin in small extracellular vesicles released by PrCa cells promotes a shift in lineage plasticity towards a NE lineage [7]. Moreover, although our group has reported that the αVβ6 integrin, in small extracellular vesicles released by cancer cells, induces M2 polarization in recipient monocytes [64] and stimulates angiogenesis in endothelial cells during cancer progression [65], is absent in NEPrCa. Here we show that the αVβ3 integrin is upregulated in tumor samples from patients affected by NEPrCa and in corresponding NE murine models. Moreover, our findings demonstrate that conversely, the expression of the αVβ6 integrin is upregulated in ADPrCa samples from humans and mice. It is therefore reasonable to speculate that monitoring the expression of these two integrins during PrCa progression will help to predict the potential for NED in PrCa patients. Moreover, based on our emerging findings that NE metastatic lesions express relatively high levels of the αVβ3 integrin, targeted therapies directed against this integrin might prove to be effective in preventing or delaying plasticity and metastasis in NEPrCa [56].
Supporting information S1 Table. Raw data of the signature score used to generate the heatmap in Fig 4. (TIF) S1 Raw images. (PDF) Fig 6. Schematic representation of the findings described in this study. SKO (PB-Cre4 PTEN loxP/loxP ) cancer cells do not metastasize and express low levels of αVβ3 integrin and high levels of αVβ6 integrin. On the other hand, DKO and TKO tumors (PB-Cre4 PTEN loxP/loxP RB1 loxP/loxP and PB-Cre4 PTEN loxP/loxP RB1 loxP/loxP TRP53 loxP/loxP respectively) express high levels of αVβ3 integrin and low levels of αVβ6 integrin. These αVβ3 positive tumors acquire metastatic behavior and expression of NE markers. https://doi.org/10.1371/journal.pone.0244985.g006