Induction of ADAM10 by Radiation Therapy Drives Fibrosis, Resistance, and Epithelial-to-Mesenchyal Transition in Pancreatic Cancer

Pancreatic adenocarcinoma (PDAC) is a deadly malignancy with poor outcomes despite advances in surgery, radiation, and chemotherapy (1). It is characterized by hypocellular tumors with a desmoplastic stroma containing a complex milieu of extracellular matrix (ECM), cancer-associated fibroblasts, pro- and anti-inflammatory immune cells, and tumor vasculature (2). This stromal microenvironment has been shown to have a major impact on patient survival, with multiple groups discovering stromal gene signatures that portend a poor prognosis in PDAC patients (3–5). The stromal component of PDACs leads to increased tissue tension, which has been shown to activate cellular pathways in tumor cells, leading to more aggressive tumor behavior and resistance to chemo- and radiotherapy (6, 7). Additionally, extracellular collagens deposited during fibrosis have been shown to activate pathways driving epithelial-to-mesenchymal transition (EMT) and metastasis (8).

Despite trials with older adjuvant radiation therapy (RT) methods demonstrating a survival disadvantage using adjuvant RT for PDAC (9), neoadjuvant stereotactic body radiation therapy (SBRT) is frequently used to facilitate surgery in patients with borderline resectable pancreatic cancer, where surgery remains the only curative option (10). Though its optimal role, technique, and efficacy are still being investigated, SBRT has been used in the locally advanced pancreatic setting with favorable local control and overall survival outcomes, particularly when higher of doses of radiation can be delivered to the tumor using advanced techniques (11). The ALLIANCE A021501 trial just recently reported inferior outcomes with neoadjuvant SBRT in addition to neoadjuvant chemotherapy (12, 13), raising the concern that even with modern delivery, the aggressive biology of PDAC may overcome the cytotoxic effect of SBRT, with radiation even stimulating more aggressive behavior. This stands in contrast though with smaller series delivering much higher doses of RT using new image-guided adaptive techniques, which have achieved preliminary outcomes far surpassing historical controls (14) and suggest its ideal implementation is still being optimized. Fibrosis is a known sequela of RT (15), and in PDAC where fibrosis contributes to aggressive behavior and therapeutic resistance, this early cytotoxic benefit may be partially offset by an increase in resistance and EMT in residual tumor clonogens. Fibrosis has an impact on immune cell infiltration, which alters the efficacy of RT-mediated tumor killing. Late tissue fibrosis after RT can also increase the technical difficulty of surgery, and is known to contribute to normal tissue toxicity in other cancer types (16). Targeting pathways of fibrosis common to PDAC tumors and the effects of RT presents the possibility of enhancing the efficacy of SBRT and improving oncologic outcomes while reducing treatment toxicity.

The ephrin receptor proteins (EPH) are the largest family of receptor tyrosine kinases that, along with their membrane bound ligands, the ephrins, mediate a large variety of developmental processes and have been implicated in carcinogenesis, particularly angiogenesis, but also fibrosis, EMT, and immune infiltration (17). We have shown that ephrinB2 expression confers poor prognosis in pancreatic cancer (18), and that inhibition of the interaction between ephrinB2 and its cognate receptor, EphB4, reduces RT-induced fibrosis and enhances tumor killing by RT in PDAC xenograft tumors (18). Recent studies have shown that cleavage of the ephrin-B2 ectodomain by the a disintegrin and metalloproteinase domain-containing protein 10 (ADAM10) results in the generation of a profibrotic, soluble ephrin-B2 ectodomain that drives fibroblast activation in lung, skin, and cardiac fibrosis (19, 20). Previous studies have shown that ADAM10 is involved in EMT and metastasis in colorectal (21) and breast cancers (22). In addition, ADAM10 has been shown to be increased in vascular endothelium in response to RT (23). These together suggest a role for ADAM10 in driving fibrogenesis and EMT following RT, prompting us to investigate this pathway in the context of SBRT-induced tumor fibrosis in PDAC. Here, we investigate the role of the ADAM10–ephrinB2 pathway in driving pancreatic tumor fibrosis following SBRT and the therapeutic potential of targeting ADAM10 to inhibit fibrosis and tumor progression.

We demonstrate upregulation of gene-expression signatures associated with fibrosis and EMT in 29 human patient PDAC tumor samples treated with SBRT. Patients with high expression of ADAM10 and ephrinB2 following SBRT have a worse prognosis than those with lower levels. Mechanistically, we demonstrate that ADAM10 is upregulated in tumor cells following SBRT in PDAC cells in vitro and in vivo, and that pharmacologic inhibition as well as genetic knockout (KO) of ADAM10 abrogates post-radiation fibrosis and delays tumor progression in orthotopic and metastatic mouse models of PDAC. We found that ADAM10 KO prevents proteomic changes consistent with matrisome activation after RT, and ADAM10-deficient cells have a reduction in fibroblast-dependent migration and invasion, which reciprocally is stimulated by the addition of ephrinB2 FC protein. We also detect soluble ephrinB2 in plasma samples collected from our in vivo models, as well as in pancreatic cancer patients treated with SBRT on clinical trial NCT02873598, in a dose-dependent manner. These data suggest a utility for antifibrotic therapy targeting ADAM10 to block RT-induced fibrosis and improve outcomes in PDAC, as well as ephrinB2, a serum marker for RT-induced fibrosis that could direct treatment selection and monitor therapeutic efficacy.

For RNA sequencing (RNA-seq) library preparation, a TempO-Seq FFPE Assay 96 sample kit was used (BioSpyder). FFPE slides of FNA biopsy and post-neoadjuvant patient samples were obtained from the University of Colorado biorepository. All the patients had borderline resectable pancreatic cancer, treated with neoadjuvant chemotherapy followed by restaging, and treatment with 30–33.6 Gy SBRT to pancreatic tumors, followed by pancreaticoduodenectomy and adjuvant chemotherapy. Tumor samples were reviewed by a pathologist, and areas of tumor cellularity were identified and marked. Areas of tumor cellularity were scraped and processed per above BioSpyder standard protocol. Samples were pooled and run in two sequencing lanes using NextSeq high-throughput sequencing instrument (Illumina). Reads were aligned and counts generated using Biospyder TempoSeqr platform. Genes with <1 mean raw counts or <1 mean counts per million (CPM) were removed from the data set. Differential expression was calculated using the limma R package (24). The resulting fold change was used with the fgsea R package (25) to perform gene set enrichment analysis for Hallmark and C2 Curated gene sets (26). Heat map displaying Z-score transformed gene expression was generated using the ComplexHeatmap R package (v1.20.0; ref. 27) and hierarchical clustering using Euclidean distance with complete linkage. CPM-normalized gene expression of post-SBRT samples was used to group patients into high- and low-expression groups using a median split for ADAM10 and EFNB2. Survival was calculated using the survival R package, and significance was calculated with a log-rank test.

Gene-expression data were obtained from The Cancer Genome Atlas (TCGA) for 179 PDAC patients. Expression of ephrinB2 and ADAM10 was plotted and separated into groups having above or below median expression. Overall survival and disease-free survival for these groups were calculated using the Kaplan–Meier method, using log-rank tests for comparisons of groups.

Mouse KPC pancreatic cancer cell lines FC1242 and PK5L1940 were passaged in RPMI-1640 supplemented with 10% FBS. Mouse NIH-3T3 fibroblasts were passaged in RPMI-1640 supplemented with 10% FBS. All cells were passaged every 2–3 days when subconfluent, at a density of 1:3–1:8. Cells were not allowed to grow beyond passage 15. PK5L1940 cells were kindly provided by laboratory of Michael Gough (Earle A. Chiles Research Institute, Providence Cancer Institute, Portland, OR). Mycoplasma testing was performed by University of Colorado core facility every 3 months.

ADAM10 inhibitor GI254023X was obtained from Sigma-Aldrich. Inhibitor was added in in vitro culture experiments at a concentration of 20 nmol/L. In in vivo mouse experiments, inhibitor was delivered by intraperitoneal injection at a concentration of 20 mg/kg, every 72 hours.

The following oligos were used for ADAM10 gRNA: Adam10-For–caccGGTTTCATCAAGACTCGTGG Adam10-Rev–aaacCCACGAGTCTTGATGAAACC. The oligos were annealed and cloned into BbsI-digested PX458 (Addgene Plasmid #48138). The ligations were transformed into NEB Stable Competent E. coli (NEB; cat. #C3040H) and plated on ampicillin-agar plates. Plasmids from colonies were sequenced using the U6-for primer: gagggcctatttcccatgattcc. PX458-gRNA plasmids were transfected into FC1242, PK5L1940, and NIH-3T3 cells and flow sorted for GFP expression. GFP-positive clones were grown in culture, and KO was confirmed by Western blotting.

Informed written consent was obtained for all tumor sample collection, studies were performed in accordance with U.S. Common Rule, and approved by institutional review board. Patient archival tumor samples were identified and obtained from biorepository, collected per COMIRB13-0315. Patients were selected from all borderline resectable pancreatic cancer patients seen through University of Colorado pancreas multidisciplinary clinic between January 2013 and December 2018, who were then treated with neoadjuvant chemotherapy and SBRT. SBRT doses were 30 to 33 Gy in five fractions. All patients received either FOLFIRINOX or gemcitabine-based chemotherapy per Supplementary Table S1. Following neoadjuvant therapy, patients received surgery followed by further adjuvant chemotherapy. Surgical samples were scored for inflammation and fibrosis by pathologist.

Patient plasma samples were collected as part of a phase I radiation dose escalation clinical trial (NCT02873598) before, during (6-hour post), and post (6 weeks). Prior to SBRT, these patients all received neoadjuvant chemotherapy. SBRT doses ranged from 9 Gy × 3 fractions to 11 Gy × 3 fractions. Primary outcome measure of this study is MTD of SBRT, with local control, progression-free survival, overall survival, small intestine changes, and vascular and cellular changes as secondary endpoints. This study is still active. Samples were collected and analyzed per COMIRB19-0328.

Cells were passaged to subconfluency and then scratched using Incucyte cell scratch apparatus. Media were changed and cells were then cultured in Incucyte live-cell imaging incubator, and images were collected every two hours. Scratch confluency was calculated using Incucyte software (Essen Bioscience).

Cell lines were generated by CRISPR-CAS9 KO using GFP expressing plasmids obtained from University of Colorado Functional Genomics facility. Cells were transfected with plasmid using standard Lipofectamine 2000 transfection, cells were then flow sorted and grown in individual colonies. Cell lysates were then screened by Western blotting.

PK5L1940 cells and FC1242 KPC cells were lysed using RIPA lysis buffer (Millipore), containing protease inhibitor cocktail (Thermo Fisher Scientific Inc) and phosphatase inhibitor (Sigma) on ice for 30 minutes. Homogenates were centrifuged at 4°C at 13,000 rpm for 20 minutes and lysates were collected. For Western blotting, lysates (50 μg) were loaded onto 10% SDS-PAGE gels. Membranes were probed overnight at 4°C in 1% BSA in PBS-T. Anti-ADAM10 (Abcam 1997), anti-ephrinB2 (Abcam 150411), anti-EphB4 (generous gift from the Vasgene Therapeutics, Inc), and anti-vimentin (Cell Signaling Technology; cat. #5741S) antibodies were used. Horseradish peroxidase–conjugated secondary antibodies were obtained from Sigma.

To determine the subcellular localization of ADAM10, cells were treated with 10 Gy and cultured for 7 days post-radiation. After 7 days, cells were scraped in cold PBS and subjected to subcellular fractionation (Thermo Scientific; cat. #78840) according to the manufacturer's protocol.

Tumors were harvested from flank xenograft experiments in which PK5L1940 cells were injected into the flanks of BL6 mice. Tumors were excised at the time of sacrifice and fixed in formalin for 24 hours. Formalin-fixed tumors were sliced and embedded tumor and mounted onto glass slides in paraffin. For IHC analysis, these mounted tissues were deparaffinized and hydrated followed by antigen retrieval as described earlier (28). Full IHC procedure described in Supplementary Methods.

Multiplexed IHC analysis was performed on Vectra platform (Caliper Life Sciences), described in detail in (29). Full multiplexed IHC analysis is described in Supplementary Methods.

Tumors were harvested from flank xenograft experiments in which PK5L1940 cells were injected into the flanks of BL6 mice. These formalin-fixed tumors were sliced, embedded, and mounted onto glass slides in paraffin. Picrosirius staining and trichrome staining were performed using standard conditions by University of Colorado Cancer Center Pathology Core. Picrosirius birefringence was imaged using Nikon microscope polarized filter. Birefringence signal was quantitated using ImageJ software.

ELISA was performed on human plasma samples obtained from patients with locally advanced pancreatic cancer on protocol COMIRB 16-1139. Mouse and human ELISA kits were obtained from LSBio. Standard kit protocol was used for ELISA. These are described in further detail in Supplementary Methods.

Briefly, snap-frozen tumor samples of approximately 5 mg were pulverized in liquid nitrogen. High salt extraction was used to make the cell fraction. Guanidine extraction was used to make the sECM fraction. Hydroxylamine extraction was used to make the iECM fraction. Fractions were digested with trypsin. Protein (600 ng) was analyzed with nano-UHPLC-MS/MS (Easy-nLC1200, Orbitrap Fusion LumosTribrid, Thermo Fisher Scientific). Files were loaded into Proteome Discoverer 2.2 and were searched against Swissprot mouse and human database. In Excel, protein abundances from the three fractions were summed together per protein. Data were visualized in Excel and MetaboAnalyst 4.0 (30). Mass spectrometry analysis is described in greater detail in Supplementary Methods.

Female C57BL6 (6 weeks old) were purchased from The Jackson Laboratories. All the mice were cared for in accordance with the ethical guidelines and conditions set and overseen by the University of Colorado, Anschutz Medical Campus Animal Care and Use Committee. ADAM10-flox Pdgfra-Cre mice were obtained from David Lagares (Massachusetts General Hospital, Boston MA). The protocols used for animal studies were reviewed and approved by the Institutional Animal Care and Use committee at the University of Colorado, Anschutz Medical Campus.

For orthotopic experiments, the protocol is described in detail in (31) and Supplementary Methods.

Metastatic implantation is described in detail in (32) and Supplementary Methods.

Image-guided radiotherapy was performed using the X-Rad SmART small animal irradiator (Precision X-Ray) at 225 kVp, 20 mA with 0.3 mm Cu filter. For animal experiments, the mice were positioned in the prone orientation and a CT scan was acquired. Radiation was delivered at a dose rate of 5.6 Gy/minute. A single 16-Gy dose of X-ray radiation was delivered to mouse pancreata using 10 mm square beam with field edges at mouse midline and below left ribs. These are described in further detail in Supplementary Methods.

We crossed mice with ADAM10 flanked by loxP sites (adam10loxP/loxP mice; ref. 33) to mice that express a tamoxifen-inducible Cre recombinase driven by the mouse promoter of the platelet-derived growth factor receptor, alpha polypeptide (PDGFRα-Cre-ERT transgenic mice; ref. 34). Our results indicate that tamoxifen treatment of offspring that were homozygous for the “floxed” adam10 allele and hemizygous for the PDGFαR-Cre-ERT transgene (adam10loxP/loxP;PDGFRα-Cre-ERT) led to the deletion of the adam10 gene in pancreatic fibroblasts and the generation of adam10 conditional KO mice, herein referred to as adam10-CKO mice. These are described in further detail in Supplementary Methods.

Tumor tissue was fresh frozen, embedded in optimal cutting temperature (OCT) compound. Atomic force microscopy (AFM) was performed on 20-μm LV sections. The nanomechanical and topographical properties of frozen, non-fixed tumor sections placed on glass slides were characterized using a NanoWizard 4a (JPK Instruments). Briefly, the samples were immersed in PBS for 15 minutes to remove OCT and allowed to thaw. The sections were then covered with protease inhibitor at 1× (HALT, Thermo Fisher Scientific), and tissue stiffness was determined using a qp-BioAC-1 (NanoandMore) cantilever with a force constant in the range of 0.15 to 0.55 N/m. Calibration of the cantilever was made using the thermal oscillation method before each experiment. Tissue mapping was performed by quantitative imaging (QITM) mode. Tumor sections were scanned in a 20 × 20 mm area using a set point of 3nN, a Z-length of 3.5 mm, and a pixel time of 30 ms with a resolution of 256×256 pixels. At least four scans were performed on each tumor section. The Hertz model was used to determine the mechanical properties of the tissues using the JPK software (35).

All PDAC tumor specimens were obtained from University of Colorado Hospital patients with informed consent after approval by the ethical committee (COMIRB 08-0439). PDAC tumor organoids were established either from fresh surgical PDAC specimens or from patient-derived xenograft models using established protocols with some modifications (36, 37). Once PDAC tumor organoids were established, short tandem repeat and Mycoplasma analysis were performed, and stocks were cryopreserved. See Supplementary Methods for further details.

Organoid viability was quantified using the CellTiter-Glo 3D Viability Assay (cat. #G9681) according to the manufacturer's protocol. Briefly, cells were lysed, and ATP levels, a marker of metabolic activity, were quantified to determine the number of viable cells.

Organoids were plated at approximately 200 cells per well and cultured in an Incuctye Live-Cell Analysis incubator for seven days with images collected every four hours. Organoid growth was quantified using Incucyte imaging software.

In vitro experiments were conducted for a minimum of n = 2 times in duplicates or triplicates. Quantitative analyses were performed using a Student t test, Mann–Whitney test, one-way ANOVA, or the Mantel–Cox test for survival using GraphPad Prism. P values of <0.05 were considered statistically significant.

We reasoned that there would be global gene-expression changes consistent with fibrosis after SBRT in patients with PDAC and performed RNA-seq analysis of tumor cellular regions of 29 BRPC tumors treated with neoadjuvant combination chemotherapy and SBRT in the University of Colorado pancreas multidisciplinary clinic, comparing global gene expression to 26 pre-neoadjuvant therapy biopsy samples (patient, tumor, and chemotherapy characteristics in Supplementary Table 1, representative images of surgical slides before in Supplementary Fig. S1A, S1C, and S1E, and after region collection in Supplementary Fig. S1B, S1D, and S1F). We performed GSEA of Naba matrisome genes, a set of ECM-associated genes compiled by domain-based organization and associated with activated ECM and fibrosis (38) and found global upregulation of matrisomal genes (Fig. 1A and B) as well as collagens (Supplementary Fig. S1G and S1H) after treatment with SBRT. This change was accompanied by upregulation of Hallmark EMT genes (Fig. 1C and D), suggestive of a more motile and invasive phenotype. To examine induction of fibrosis by a different method, we examined mature collagen formation by picrosirius red staining in surgical samples from five patients without neoadjuvant RT and five with neoadjuvant RT (Supplementary Table 2) and found increased stromal collagen deposition in samples treated with RT (Fig. 1E and F). Both sets had similar collagen anisotropy, consistent with the desmoplastic stroma of these tumors (Fig. 1G).

We have previously shown that inhibition of ephrinB2 ligand binding with EPHB4 decreases RT-induced fibrosis and sensitized PDAC to RT. It was recently reported that ADAM10 induction led to ephrinB2 cleavage in a bleomycin model of pulmonary fibrosis, driving the fibrotic phenotype, leading us to hypothesize that ADAM10 may be induced following RT and facilitate ephrinB2 function in PDAC tumors. We evaluated pancreatic cancer patients' survival following neoadjuvant SBRT stratified by ADAM10 and ephrinB2 expression in our 29 post-surgical human PDAC patient samples. High expression of ADAM10 and ephrinB2 post-SBRT resulted in significantly worse prognosis (Fig. 1H). We also observed enriched matrisome protein expression in these patients (Supplementary Fig. S1I). Examination of ADAM10 and ephrinB2 expression in RT-naïve PDAC patients in TCGA revealed that patients with high ADAM10 or high ephrinB2 expression had worse prognosis compared with patients with low expression of both ADAM10 and EFNB2, with worse median OS (median: 24.2 months, vs. median 17.25 months, HR 1.403, 95% CI of HR 0.8235–2.392, P = 0.04; Fig. 1I) and DFS (median: 20.3 months vs. median 12.3 months, HR 1.645, 95% CI of HR 1.045–2.591, P = 0.02; Fig. 1J). Interestingly, individual expression of ADAM10 or ephrinB2 alone was not independently prognostic (Supplementary Fig. S1J and S1K). Additionally, correlations were observed between ADAM10 (Supplementary Fig. S2A; R² = 0.21, P = 1.73×10^–11) and EPHB4 expression (Supplementary Fig. S2B; R² = 0.15, P = 3.13×10^–8) with EFNB2, suggesting commonality to aggressive tumors. To ensure that expression of these proteins was independently predictive of survival and not only associated with other tumor cellularity, we examined tumor purity between tumors in the TCGA with low and high ADAM10 and ephrinB2 expression. We found no correlation between ADAM10 expression and tumor purity (Supplementary Fig. S2C), although EFNB2 expression was associated with higher tumor purity (Supplementary Fig. S2D and S2E). However, no correlation was observed between tumor purity and patient prognosis (Supplementary Fig. S2F). We also tested whether ADAM10 and EFNB2 expression was associated with known tumor subtypes. We found that high EFNB2 was associated with a difference in tumor subtype, as characterized in Bailey and colleagues, containing more squamous and progenitor types, versus low ephrinB2 expression, which was associated with more immunogenic and ADEX types (3). ADAM10 was not associated with differences in the Bailey and colleagues subtypes (Supplementary Fig. S2G; ref. 3), and neither ADAM10 nor EFNB2 was associated with enrichment for the basal and classic subtypes characterized in Moffitt and colleagues (Supplementary Fig. S2J and S2K; ref. 4). High EFNB2 expression was associated with the progenitor Bailey and colleagues subtype (Supplementary Fig. S2H). Interestingly, neither Bailey nor Moffitt subtypes were predictive of survival in the TCGA cohort (Supplementary Fig. S2I and S2L). These together suggest that ADAM10 and ephrinB2 coexpression is independently predictive of survival and not a consequence of other known tumor characteristics such as cellularity or published subtype.

Using multispectral imaging analysis, we found ADAM10, EphB4, and ephrinB2 to be expressed in human pancreatic tumor samples (Fig. 2A). ADAM10 displayed patchy expression on some tumor islands, as determined by morphology (Fig. 2A) and cytokeratin staining (Supplementary Fig. S3A), with little stromal expression, as determined by vimentin and αSMA (Fig. 2A; Supplementary Fig. S3B and S3C). EphB4 displayed consistent expression on tumor islands, whereas ephrinB2 displayed punctate patchy tumor cell as well as stromal expression (Fig. 2A; Supplementary Fig. S3A–S3C). Using the Human Protein Atlas, we found ADAM10 and EphB4 to be expressed on tumor cells, whereas ephrinB2 occurred on stroma and some tumor cells by morphology (Supplementary Fig. S3D). When we examined syngeneic mouse flank tumors (using the KPC cell line PK5L1940) by IHC, we observed ADAM10 and EphB4 to have a cytoplasmic/membranous expression pattern on morphologic tumor islands, whereas ephrinB2 displayed more spindle-shaped expression between clusters of tumor cells, consistent with predominantly stromal expression (Supplementary Fig. S3E). Consistent with these findings, using multispectral imaging of orthotopically implanted KPC tumors, we found EphB4 to be expressed on cell clusters not expressing vimentin (Supplementary Fig. S3F). Additionally, by sequential immunofluorescence analysis, ADAM10 staining was observed in regions of EpCAM staining in these tumors (Supplementary Fig. S3G). Examination of ADAM10 (Supplementary Fig. S4A), EFNB2 (Supplementary Fig. S4B), and EPHB4 (Supplementary Fig. S4C) gene expression in tumors in the UCSC Cell Browser Treehouse Cancer Compendium data set (39) revealed that all had higher than average expression in PDAC tumors compared with other tumor types. Also, in the UCSC Cell Browser Adult Pancreas data set, ADAM10 and EPHB4 had higher ductal expression in normal pancreas, consistent with the cell type of origin of PDAC (Supplementary Fig. S4D and S4E), whereas EFNB2 had both ductal and mesenchymal expression (Supplementary Fig. S4F). We also examined expression of all three proteins in published scRNA-seq data sets of human PDAC tumors (Supplementary Fig. S5A–S5C; ref. 40) and mouse PDAC tumors (Supplementary Fig. S5D, S5E and S5F; ref. 41) and found all three to be expressed in subsets of ductal, endothelial, and fibroblast cells.

To determine if ADAM10 was induced by RT, we examined ADAM10 expression in PK5L1940 PDAC cells in vitro one week following irradiation and found that expression of precursor and active isoforms of ADAM10 increased with increasing RT doses (Fig. 2B and C). We then generated a CRISPR-CAS9 KO of ADAM10 in PK5L1940 KPC cells to test the impact of ADAM10 on fibrosis and survival following RT (Supplementary Fig. S5G). We performed orthotopic tumor implantation to properly recapitulate the local tumor microenvironment with the host stromal contribution, followed by in vivo irradiation with an image-guided small animal irradiator (Fig. 2D). Wild-type (WT) and ADAM10 KO graft tumors were implanted into BL/6 mice and protein analysis revealed ADAM10 upregulation in WT tumors with increasing doses of RT. A cleaved band of ephrinB2 was also observed in WT tumors at the 16-Gy dose of RT, but was absent in the ADAM10 KO grafts (Fig. 2E and F). Interestingly, this cleavage did not occur at a dose of 8 Gy, suggestive of a threshold effect. There was still some detectable ADAM10 in these tumors consistent with some stromal contribution by the host mouse.

To examine whether ADAM10 was driving fibrosis in response to radiation, we implanted PK5L1940 PDAC WT or ADAM10 KO cells orthotopically into syngeneic immunocompetent mouse pancreata, followed by in vivo treatment with RT (Fig 2D). Analysis of mature collagen formation by Masson's Trichrome and Picrosirius red staining (Fig. 2G) demonstrated RT-induced mature collagen formation (Fig. 2H) and increased collagen organization (by anisotropy) in WT tumors, but not in ADAM10 KO tumors (Fig. 2I). These results were validated in flank PK5L1940 tumors treated with RT and the ADAM10-specific inhibitor GI254023x. Treatment with the ADAM10 inhibitor blocked RT-induced fibrosis (Supplementary Fig. S5H–S5J) and, when combined with RT, resulted in a reduction in tumor growth compared with untreated tumors (Supplementary Fig. S5K). This reduction in tumor growth was also observed with the combination of RT and ADAM10 inhibition using another KPC cell line, FC1242 (Supplementary Fig. S5L).

Examining the effect of ADAM10 KO on orthotopic pancreatic tumor size two weeks following in vivo irradiation, we found that tumor mass and volume were reduced in ADAM10 KO tumors, whereas the combination of ADAM10 KO and RT further reduced tumor mass beyond ADADM10 KO or RT alone (Fig. 2J and K). We also observed a significant reduction in the growth rate of flank tumors with the combination of ADAM10 KO and RT (Supplementary Fig. S5M). Similarly, we found that treatment with the combination of ADAM10 inhibitor GI254023x and RT resulted in a reduction in orthotopic tumor volume and mass beyond inhibitor alone, whereas treatment with RT alone resulted in a significant reduction in tumor mass only (Fig. 2L and M). Examining median survival, we found that ADAM10 KO (35 days) or treatment with RT alone (33 days) did not significantly improve survival above untreated WT tumors (28 days; Fig. 2N and O). However, the combination of ADAM10 KO and RT resulted in a significant improvement in survival (48 days), suggesting a synergistic effect of the two treatments and demonstrating the importance of intact ADAM10 expression for resistance to RT (Fig. 2N and O).

To determine whether activation of the extracellular matrix was occurring with RT in an ADAM10-dependent manner, we performed mass spectrometry on WT and ADAM10 KO tumors treated with RT. Hierarchical clustering resulted in the discovery of global proteomic changes occurring in WT but not ADAM10 KO tumors (Fig. 3A and B), indicating that ADAM10 is required for changes in the proteomic expression of tumor stroma following high-dose RT. Using the Core Matrisome gene set (42), we discovered that numerous matrisome proteins had increased expression with higher dose of RT (Fig. 3C and D). These proteins did not change in the ADAM10 KO tumors, suggesting that RT-induced ECM activation and fibrosis is dependent on ADAM10 activity. Notably, on partial least-square analysis, the protein with the greatest significance was collagen I (Fig. 3E), which is known to regulate invasion, metastasis, and apoptotic pathways in PDAC (8). Re-interrogating our human post-SBRT surgical samples, we found an enrichment of matrisome proteins upregulated in our human samples treated with RT versus the pre-SBRT biopsy samples (Supplementary Fig. S5N and S5O), indicating conservation of a common set of upregulated ECM genes following RT. We also examined the effect of ADAM10 inhibition in the presence of RT on proteomic expression in tumors by treating WT grafts with GI254023x and found there were ADAM10 activity-dependent changes in matrisome proteins after RT (Supplementary Fig. S6A–S6D). Reactome pathway analysis of our WT-ADAM10 KO comparison found the five most overrepresented pathways in WT tumors treated with SBRT included ECM organization and collagen formation (Supplementary Fig. S6E), consistent with ADAM10 driving RT-induced ECM activation and fibrosis. The most overrepresented pathways in ADAM10 KO tumors included Notch signaling, the intrinsic apoptosis pathway, and interferon signaling. Additionally, we found a large number of negatively prognostic proteins downregulated by ADAM10 inhibition, as well as genes involved in promoting cell motility, invasion, and metastasis (Supplementary Table S3).

To confirm our proteomic analysis, we examined markers of EMT by Western blotting following RT and found induction of vimentin and snail to occur in WT tumors treated with 16 Gy RT. When we examined ADAM10 KO tumors (Fig. 3F and G), we see a reduction in expression of these markers as compared with control RT-treated tumors, confirming this induction of EMT is mediated by ADAM10. We also see a reduction in these markers after treatment with GI254023X (Supplementary Fig. S6F).

We examined whether these ADAM10-dependent matrisomal changes had an effect on the tissue tension within PDAC tumors, as that has been shown to drive global gene-expression changes and subsequent aggressive tumor behavior (7). Using AFM, we found the nanomechanical stiffness of WT PDAC tumors increased after RT (Fig. 4A–C). This increase with RT did not occur in the ADAM10 KO tumors (Fig. 4B and D), and compared with WT tumors, stiffness was lower in ADAM10 KO tumors with or without RT (Fig. 4B and D).

We next examined the effect of ADAM10 on the migration of the KPC cell line PK5L1940. First, we found no difference in cell proliferation by MTT assay between WT and ADAM10 KO cells to explain differences in migratory capacity (Fig. 4E). In wound-healing assays of KPC cells, cell migration was greater in WT cells compared with ADAM10 KO cells, consistent with ADAM10 activity enhancing migration (Fig. 4F). To determine if this migration could be further enhanced by fibroblast-expressed ephrinB2, we incubated KPC cells in coculture with NIH-3T3 fibroblasts for one week prior to scratch assay. Fibroblasts enhanced the motility of WT KPC tumor cells (Fig. 4G), an effect that was not observed in ADAM10 KO cells (Fig. 4H). To determine if migration was dependent on the stimulatory effect of ephrinB2–EphB4 interaction, we treated PK5L1940 cells with a pegylated peptide inhibitor (TNYL-RAW, labeled B4i) that blocks the ephrinB2–EphB4 interaction (43). In the presence of TNYL-RAW, there was no increase in motility with the addition of fibroblasts (Fig. 4I). These suggest that ADAM10 enhances motility, and the addition of fibroblasts enhances tumor cell motility in ADAM10 and ephrinB2–EphB4-dependent manners. When we treated WT cells with RT, we found an increase in migration (Fig. 4J), which did not occur in ADAM10KO cells (Fig. 4K). We also observed an increase in migratory capacity of WT cells treated with RT with the addition of fibroblasts (Fig. 4L). Furthermore, when we treated fibroblast cocultured WT cells with RT, we detected a significant increase in their migratory capacity (Fig. 4M), whereas with cocultured ADAM10KO cells, this difference was no longer observed (Fig. 4N).

To test whether RT affected the invasive capacity of our KPC tumor cells in a manner dependent on ADAM10 cleavage of ephrinB2, we cocultured WT or ADAM10 KO cells with fibroblasts for one week following in vitro irradiation and examined invasion by Xcelligence invasion assay 24 hours following removal of fibroblasts and initiation of serum starvation. We found a trend toward an increase in invasion in WT cells with the addition of RT (Fig. 4O). However, in our ADAM10 KO cells, a significant reduction in invasion after treatment with RT occurred, which was reversed with the addition of EPHB4 stimulating ephrinB2 mimetic ephrinB2 FC protein (Fig. 4P). We did not observe a significant difference in cell viability in recovered cells following RT (Fig. 4Q). This suggests that preservation of invasion capability following RT is dependent on ADAM10 and ephrinB2. We did not observe a reduction in ADAM10 KO cell viability with RT, or an increase in viability with the addition of ephrinB2 FC (Fig. 4R), suggesting this invasion effect is not a function of cell survival.

We utilized the human tumor organoid line Panc193, cocultured with human CAFs (ephrinB2 expression shown in Supplementary Fig. S6G), to determine whether ADAM10 inhibition could enhance the growth delay seen with RT in a human tumor model. A significant decrease in tumor growth constant (k) was observed with RT in both the vehicle-treated and GI254023x-treated tumor organoids (Fig. 5A), with a trend toward more reduction of tumor growth in the presence of inhibitor treated. Bioluminescent organoid viability was not reduced by RT alone, but the combination of ADAM10 inhibition with RT was sufficient to significantly reduce viability (Fig. 5B). Using the EphB4–ephrinB2 inhibitor, TNYL-RAW (labeled B4i), a similar reduction in growth in organoid viability was observed (Fig. 5C). These data indicate that both ADAM10 inhibition and EphB4 inhibition enhanced the tumor cell killing effect of RT in this system.

To test the effect of ADAM10 KO on metastasis of KPC cells in vivo, we utilized a hemispleen model of liver metastases (Fig. 5D; ref. 32). WT and ADAM10 KO cells were irradiated in vitro (10 or 0 Gy) and cocultured with NIH-3T3 fibroblasts prior to mouse injection to mimic the stimulatory effect seen in our motility assays (Fig. 5D). Two weeks post-implantation, the number of metastatic lesions and area of metastases were reduced in the livers of mice implanted with ADAM10 KO cells treated with RT (Fig. 5E–G). Neither ADAM10 KO nor treatment of WT cells with RT improved median survival compared with untreated WT cells (Fig. 5H). Combination RT with ADAM10 KO, however, led to a significant improvement in overall survival (Fig. 5H). These data suggest that inhibiting ADAM10 induction following RT can improve survival by reducing the metastatic potential of tumor cells, though the modest delay in mortality in this experiment suggests that these tumor cells retain their proliferation potential once metastases are seeded.

We next generated a CRISPR-CAS9 KO of ephrinB2 on NIH-3T3 fibroblasts to determine whether the contribution of ephrinB2 from fibroblasts was important for the metastatic potential of KPC cells. We cocultured PK5L1940 KPC tumor cells with ephrinB2 KO fibroblasts (Supplementary Fig. S6H) and performed hemispleen implantations to analyze the rate of liver metastasis formation. We found that 14 days following implantation, both liver weight and surface area of PDAC metastases were reduced significantly by ephrinB2 KO on the cocultured fibroblasts (Fig. 5I and J), confirming the role of stromal contributed ephrinB2 in metastasis formation.

In order to further demonstrate that tumor cell and not host stromal ADAM10 was responsible for the resistant EMT phenotype we observed, we then utilized a mouse KO model of ADAM10 for tumor implantation experiments. As mice that are globally ADAM10 deficient die at day 9.5 of embryogenesis with multiple defects of the developing central nervous system, somites, and cardiovascular system (44), we utilized mice in which we could conditionally delete ADAM10 in pancreatic fibroblasts. Mice with ADAM10 flanked by loxP sites (adam10loxP/loxP mice; ref. 33) were crossed with mice that express a tamoxifen-inducible Cre recombinase driven by the mouse promoter of the platelet-derived growth factor receptor, alpha polypeptide (PDGFRα-Cre-ERT transgenic mice; ref. 34). PDGFRα-Cre mice have been used to label and/or target genes in glial progenitor cells in the CNS (34) as well as dermal (45), lung (46), muscle (47), and liver (48) fibroblasts during the development of fibrosis. PDGFRα has also been used to investigate CAF functions in vivo during the development of cancer due to its ability to target a distinct subset of tumor resident CAFs (49, 50) compared with Acta2-Cre transgenic mouse (which targets both myofibroblasts and smooth muscle cells) or FSP-Cre transgenic mouse (which targets both fibroblasts and a subset of myeloid cells). Our results indicate that tamoxifen treatment of offspring that were homozygous for the “floxed” adam10 allele and hemizygous for the PDGFαR-Cre-ERT transgene (adam10loxP/loxP;PDGFRα-Cre-ERT), led to the deletion of the adam10 gene in pancreatic fibroblasts and the generation of adam10 conditional KO mice, herein referred to as ADAM10-CKO mice. Littermates treated with corn oil vehicle alone were used as controls and are referred to herein as adam10-C mice.

To determine the contribution of ADAM10 on stromal fibroblasts to tumor growth, we implanted adam10-C and adam10-CKO mice with WT or ADAM10 KO flank tumors and compared their growth rate to tumors implanted in WT mice. Although there was a reduction in tumor growth 27 days following implantation of ADAM10 KO tumor cells compared with WT tumor cells in both WT and fibroblast-specific ADAM10-deficient mice, there was no tumor growth delay in the ADAM10-CKO mice (Fig. 5K; Supplementary Fig. S6I). Similarly, we found no difference in size of WT tumors implanted in WT or ADAM10-CKO mice 20 days after treatment with 16 Gy RT (Fig. 5L). These together suggest ADAM10 expression on tumor cells, rather than fibroblasts, is required for the observed protumorgenic, prosurvival effect.

To determine whether serum-soluble ephrinB2 could be detected following SBRT, we tested plasma from locally advanced pancreatic cancer patients on clinical trial NCT02873598 with samples collected before, during, and after treatment with three fractions of increasing doses of SBRT for ephrinB2 expression by ELISA (Fig. 6A). We could detect ephrinB2 in pancreatic cancer patients' plasma and found that this detection increased with time after SBRT and with increasing doses of SBRT (Fig. 6A). We also tested serum collected from tumor-bearing mice two weeks following RT and found increasing quantities of soluble plasma ephrinB2 with increasing dose (Fig. 6B). We compared serum ephrinB2 in mice with WT or ADAM10 KO PDAC tumor grafts and found that ADAM10 KO significantly reduced the levels of serum ephrinB2 (Fig. 6C). Our findings that RT induced ADAM10 cleaves ephrinB2 with increasing doses suggest that ephrinB2 cleavage may be a useful serum marker for RT-induced fibrosis and its disappearance a marker of efficacy of ADAM10-directed therapy.

Cytotoxic therapies, including RT, are core tools in the oncology armamentarium. RT is a powerful locoregional modality that can both directly ablate tumor cells and stimulate local and systemic immune antitumor responses (51), which are both crucial for local and distant control of pancreatic cancer. However, any cytotoxic therapy can apply selective pressure to a tumor population, and may be made more effective by targeting the mechanisms of adaptation tumors use to become resistant to these therapies. This is especially relevant in the context of a cancer whose tumorigenesis parallels the late effects of radiation, and whereas the role of dose escalation is still being investigated, a recent major trial of SBRT failed in improving outcomes and may in fact worsen them (13). It should be noted that preclinical models of RT have shown activation of adaptive resistance mechanisms in other tumor histologies. Fractionated RT has been shown to lead to EMT in breast cancer through Notch-mediated activation of JAK/STAT3 signaling (52), which is also a key regulator of the desmoplastic stroma in pancreatic cancer (7). As EMT is fundamental to normal wound healing (53), it follows that cytotoxic therapies would lead to EMT given the ensuing local inflammatory response (54). This is more significant in pancreatic cancer, where it has been shown that chronic inflammation is central to the disease's natural history (55), with ECM activation further enhancing its aggressiveness (3, 7). Here, we found ECM and EMT pathways were upregulated following RT in patient and mouse pancreatic cancer tumors, but were able to abrogate these changes by targeting ADAM10. ADAM10 inhibition was able to block ephrinB2 cleavage, mature collagen deposition, and tumor cell motility, enhancing tumor killing by radiation and decreasing metastasis formation. These findings suggest that targeting the cellular and microenvironmental response to radiation may enhance the efficacy of standard treatments, and may raise the possibility of cure with cutting-edge radiation delivery.

There has been interest in the targeting of metalloproteinases (MMP), as these proteins have been found to play roles in multiple pathways implicated in tumor aggression (56). Trials of MMP inhibitors, to which ADAM10 enzymatic activity is related, did not lead to large-scale adoption in part due to their nonspecificity and ubiquitous nature, resulting in a broad range of effects on many tissues. This led to dose-limiting toxicities, notably musculoskeletal pain, which, while reversible, made it an unattractive option in a long-term maintenance setting (57). Importantly, as in the case with ADAM10, inhibition was not inherently cytotoxic and without combination with a cytotoxic agent would not result in tumor killing. Although solid tumors are prone to dysregulation of major cellular pathways to increase their invasive potential such as the TGFB, Wnt, Notch, and Hedgehog pathways (58), targeting of these pathways does not typically result in durable cytotoxicity as in the case of oncogene-addicted hematologic malignancies (59). The lack of cytotoxicity does not discount the fact that therapies targeting mediators of fibrosis and EMT may have utility in enhancing the effect of traditional cytotoxic therapies by blocking the phenotypic changes that allow for tumor adaptation and survival.

We previously discovered that ephrinB2–EphB4 inhibition reduces stromal fibrosis in pancreatic tumors and leads to improved tumor killing by RT. Other Eph-ephrins have been shown to be regulated by ADAM10 cleavage in development, and the action of ephrinB2 has been shown to be both positively (19) and negatively (60) regulated by ADAM10 cleavage. ADAM10 activity has been found to be dependent on EPH function in other developmental contexts (61). Fibroblast activation by ephrinB2 driving myofibroblast differentiation and resulting pulmonary fibrosis was found to be dependent on the activity of ADAM10 in response to bleomycin (19). One novel finding of that publication was the detection of a shorter soluble cleaved isoform of ephrinB2 that could function in a paracrine manner to induce this myofibroblast differentiation. ADAM10 has also been seen to be induced in endothelial cells in response to RT, leading to an increase in vascular permeability (23), itself an important step in initiating tissue fibrosis (62). Our detection of soluble ephrinB2 in this context has significant implications, given the prior finding that ephrinB2 could stimulate fibrosis, and the importance of fibrosis in driving the invasive nature of pancreatic cancer (63). Although preliminary, the clinical and mechanistic significance of systemic ephrinB2 warrants further study, and could represent a therapeutic target to enhance the efficacy of RT and decrease tumor resistance and normal tissue toxicity.

Fibrosis of normal tissues is a normal consequence of RT (15), which has traditionally been thought to be related to damage to tissues residing in the treated field (64). But as outcomes improve and advanced technologies allowing higher doses per treatment such as SBRT come into greater use, it will be important to understand the impact of cytotoxic stimulation of fibrosis has on the tumor microenvironment and systemic response. Both of these potential sequelae highlight the potential benefit of combining agents targeting tumor fibrosis, through pathways such as ADAM10–ephrinB2, in not only improving local outcomes, but preventing adaptive resistance and decreasing normal tissue toxicity.

S.D. Karam receives clinical trial research funding unrelated to this work from AstraZeneca. No disclosures were reported by the other authors.

A.C. Mueller: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft. M. Piper: Conceptualization, validation, investigation, visualization, methodology, writing–review and editing. A. Goodspeed: Formal analysis, writing–review and editing. S. Bhuvane: Investigation. J.S. Williams: Investigation. S. Bhatia: Conceptualization, methodology, writing–review and editing. A.V. Phan: Investigation. B. Van Court: Conceptualization, methodology, writing–review and editing. K.L. Zolman: Investigation. B. Peña: Investigation. A.J. Oweida: Conceptualization, methodology, writing–review and editing. S. Zakem: Investigation. C. Meguid: Data curation. M.W. Knitz: Conceptualization, methodology, writing–review and editing. L. Darragh: Conceptualization and methodology. T.E. Bickett: Conceptualization and methodology. J. Gadwa: Conceptualization and methodology. L. Mestroni: Resources and methodology. M.R.G. Taylor: Conceptualization and methodology. K.R. Jordan: Data curation, investigation, and visualization. P. Dempsey: Conceptualization, resources, and methodology. M.S. Lucia: Investigation. M.D. McCarter: Conceptualization, data curation, and investigation. M. Del Chiaro: Conceptualization, data curation, and investigation. W.A. Messersmith: Conceptualization, resources, and data curation. R.D. Schulick: Conceptualization, investigation, and project administration. K.A. Goodman: Conceptualization, resources, data curation, investigation, and methodology. M.J. Gough: Conceptualization, resources, and methodology. C.S. Greene: Conceptualization, resources, data curation, and formal analysis. J.C. Costello: Conceptualization, resources, data curation, and methodology. A.G. Neto: Formal analysis and investigation. D. Lagares: Conceptualization, resources, and formal analysis. K.C. Hansen: Conceptualization, resources, data curation, and formal analysis. A. Van Bokhoven: Conceptualization, resources, data curation, investigation, and methodology. S.D. Karam: Conceptualization, resources, formal analysis, supervision, funding acquisition, investigation, methodology, writing–review and editing.

This work was supported by Cancer Center Support Grant (P30CA046934), R01-DE028282 (S.D. Karam), R01-DE028529 (S.D. Karam), Paul Sandoval Funds (A.C. Mueller), RSNA Resident Research Grant (A.C. Mueller), Cancer League of Colorado Grant (A.C. Mueller), and by the Wings of Hope Foundation (S.D. Karam and K.A. Goodman). D. Lagares gratefully acknowledges funding support from the NIH (grant R01 HL147059-01).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.