Identification of Novel Biomarkers in Pancreatic Tumor Tissue to Predict Response to Neoadjuvant Chemotherapy

Background: Neoadjuvant chemotherapy (NAC) has been of recent interest as an alternative to upfront surgery followed by adjuvant chemotherapy in patients with pancreatic ductal adenocarcinoma (PDAC). However, a subset of patients does not respond to NAC and may have been better managed by upfront surgery. Hence, there is an unmet need for accurate biomarkers for predicting NAC response in PDAC. We aimed to identify upregulated proteins in tumor tissue from poor- and good-NAC responders. Methods: Tumor and adjacent pancreas tissue samples were obtained following surgical resection from NAC-treated PDAC patients. SWATH-MS proteomic analysis was performed to identify and quantify proteins in tissue samples. Statistical analysis was performed to identify biomarkers for NAC response. Pathway analysis was performed to characterize affected canonical pathways in good- and poor-NAC responders. Results: A total of 3,156 proteins were identified, with 19 being were significantly upregulated in poor-responders compared to good-responders (log2 ratio > 2, p < 0.05). Those with the greatest ability to predict poor-NAC response were GRP78, CADM1, PGES2, and RUXF. Notably, canonical pathways that were significantly upregulated in good-responders included acute phase signaling and macrophage activation, indicating a heightened immune response in these patients. Conclusion: A novel biomarker signature for poor-NAC response in PDAC was identified.


INTRODUCTION
Pancreatic ductal adenocarcinoma (PDAC) has the lowest survival rate of all major cancers (∼6% at 5 years post-diagnosis) and is projected to become the second most common cause of cancer related death by 2030 (1,2). Intrinsic chemotherapyresistance is one of the major clinical problems associated with PDAC, resulting in the failure of currently available therapeutic options (3). Adjuvant chemotherapy in patients with resected PDAC has been shown to extend survival over surgery alone, and more recently, more intensive regimens such as FOLFIRONOX have been shown to be even more effective (4). However, not all patients are capable of commencing let alone completing chemotherapy after surgery for PDAC. As such, there has been an increasing trend toward neoadjuvant chemotherapy (NAC), i.e., pre-operative chemotherapy, in order to effectively deliver systemic chemotherapy since improvements in nodal status and resection margin status have been observed (5,6). However, a subset of patients can be classified as "poor responders" to NAC, failing to demonstrate tumor response with subsequent early disease recurrence and shortened overall survival time (7). While genetic classification of PDAC may help identify a high risk squamous or basal subtype (8), the high costs of these methodologies have prohibited the general clinical use of genetic analysis of individual PDAC patients to help guide therapy. Therefore, there is a need for discovering more readily applicable tissue and/or blood-based secreted biomarkers that can predict a NAC response, which may be detected by more cost-effective tests.
Recently, there has been a surge in interest in the "-omics" approach to biomarker discovery in cancer research. Such approaches allow identification of a myriad of genes, transcripts, proteins and metabolites unique to cancer. They are therefore an invaluable first-step in the process of biomarker identification and validation. SWATH-MS (Sequential Window Acquisition of all Theoretical fragment ion spectra-Mass Spectroscopy) is a high throughput quantitative mass spectrometry method for proteome analysis (9). This technology allows permanent recording of all peptide fragment ions in biological samples, which imparts the advantages of a high throughput shotgun approach, with the consistency and data reproducibility of selective reaction monitoring (SRM) proteomics (10). Here, we report on the use of SWATH-MS as a discovery proteomics approach to identify differences in proteomic profile of good-and poor-NAC responders in PDAC.

Participants and Tissue Collection
Patients who presented with histologically confirmed PDAC at a tertiary centre [Royal North Shore Hospital (RNSH) and North Shore Private Hospital (NSP), Sydney, Australia] were included in the study between 04/03/2016 and 18/07/2017. All patients selected were treated with NAC before surgical resection, following individual discussion by our multidisciplinary team. The NAC regimen was at the discretion of the oncologist. Tumor tissue and adjacent normal pancreas were obtained from patients during the surgical intervention. Pathologically confirmed tumor and adjacent normal pancreas tissue were cut into 2 mm 3 portions and stored in cryotubes in a −80 • C freezer for later analysis.
The NAC response was determined based on the residual tumor viability, as described previously (7). Briefly, at the time of initial surgical pathology reporting, the residual tumor viability was assessed by the reporting pathologist. All histological slides were reviewed to estimate the viable residual tumor as a percentage of the estimated original tumor volume. A case with no response to NAC was recorded as 100% viable and a case with complete regression after treatment was recorded 0% viable.

Proteomic Sample Preparation and SWATH-MS Analysis
Protein Digestion and LC-MS/MS Analysis All tissue samples were lysed in 100 mM triethylammonium bicarbonate (TEAB) and 1% sodium deoxycholate buffer using a probe sonicator. Protein concentrations were estimated using the bicinchoninic acid protein assay (Thermo Scientific, Waltham, MA). The cysteine residues were reduced in the presence of 10 mM dithiothreitol (DTT, Bio-Rad, Hercules, CA) at 60 • C and alkylated with 10 mM iodoacetamide (IAA, Bio-Rad) at room temperature in the dark. Trypsin (sequencing grade; Promega, Madison, WI) was added in a 1:50 ratio and proteins were enzymatically degraded overnight at 37 • C. By adding 1 µL formic acid (FA; Thermo Scientific) the digestion was quenched and the sodium deoxycholate (SDC) precipitated and removed by centrifugation (14,000 rpm) for 5 min. Samples were lyophilized and reconstituted in 2% acetonitrile (ACN; Sigma Aldrich, St. Louise, MO) and 0.1% FA.
Digested samples were pooled, by combining a small fraction of each tissue sample from the tumor and adjacent normal pancreas, and subjected them to basic reverse phase chromatography high performance liquid chromatography (HPLC), using an extended C18 column 2.1 mm × 150 mm, 3.5 µm (Agilent, Santa Clara, CA), on an Agilent 1200 series HPLC. One hundred microgram of peptides per pool were precleaned with Sep-Pak C18 and then injected at a flow rate of 0.3 mL/min at room temperature onto the column. The peptides were separated over a 1 h gradient from using Buffer A of 5 mM ammonia at approximately pH 10.4 and Buffer B of 90% ACN/5 mM ammonia, and eluting peptides were collected in fractions of 1 min. Concatenated pooling of the fractions was performed.
For data dependent MS/MS acquisition to build a spectral library of the basic reverse phase fractionated samples, the 20 most intense m/z values exceeding a threshold >250 cps on the TripleTOF 6600 with charge stages between 2+ and 4+ were selected for analysis from a full MS survey scan and excluded from analysis for 20 s to minimize redundant precursor sampling.
In data independent acquisition, a 100 variable window method was used over a range of 400-1,250 m/z with window sizes based on precursor densities in the LC-MS/MS acquisition. Collision energies were calculated for 2+ precursors with m/z values of lowest m/z in window + 10% of the window width. The data were acquired over an 80 min ACN gradient.

Protein Identification and Quantification
Spectral libraries for SWATH-MS quantitation were generated with ProteinPilot TM software 5.0 using the Paragon TM algorithm (SCIEX) in the thorough ID mode including biological modifications and chemical modifications. MS/MS data were searched against the human UniProt database (release February 2016, 20,198 entries) with carbamidomethyl as a fixed modification for cysteine residues. An Unused Score cut-off was set to 0.05 and the false discovery rate (FDR) analysis was enabled.
Generated Paragon group files were imported into PeakView TM software 2.1 using the SWATH MicroApp 2.0 (release 25/08/2014) to generate a sample specific spectral library which was matched against SWATH-MS data. After retention time calibration with endogenous peptides, data were processed using following processing settings; 100 maximal peptides per protein, maximal 6 transitions per peptide, peptide confidence threshold of 99%, transition false discovery rate < 1%, 5 min extraction window and fragment extraction tolerance of 75 ppm.

Data Analysis
Survival data was compared using Kaplan-Meier curve analysis. The statistical differences in the survival curve were analyzed by the Log-rank test. Proteomic data was initially analyzed by the principal component analysis (PCA) to observe inherent groupings within the data set. Further, proteins which were markedly up-or down-regulated (log 2 ≥ 2 or ≤ −2) were compared using multiple t-test analysis (p < 0.05; q < 0.1; false discovery rate was determined with Q = 1%). The predictive model for selected proteins was validated by the Area Under the Receiver Operating Characteristic (AUROC) curve. All analysis was performed using either GraphPad Prism (GraphPad Software, San Diego, California) or JMP (SAS Institute, Cary, North Carolina) statistical software. Pathway analysis was performed using Ingenuity Pathway Analysis (IPA; Qiagen Bioinformatics, Redwood City, CA) (11). The proteins which were markedly (log 2 ≥ 2 or ≤ −2) and significantly (p < 0.05; q < 0.1) differentially expressed were inputted into IPA.

Population Demographics and Survival Data
A total of 18 PDAC patients (7 males, 11 females) were recruited for this study. All PDAC patients underwent neoadjuvant chemotherapy (NAC) before surgical resection. Patient characteristics (age, sex, tumor stage, NAC received, residual tumor viability) are described in Figure 1A.
The patients were divided on the basis of their response to NAC, which was determined by the residual tumor viability in the specimen. Based on the previously described classification methods (13), the tumors with ≤30% viable tumor cells (i.e., HTRG grade 0, CAP grade 0; HTRG grade 1, CAP grade 1; and HTRG grade 2, CAP grade 2: complete to moderate response) were graded "good-responders, " while tumors with >30% viable tumor cells (HTRG grade 2, CAP grade 3; poor response) were graded as "poor-responders." The good-responders had significantly (p < 0.05) longer overall survival compared to poor-responders ( Figure 1B).

Principal Component Analysis: Distinct Tissue Samples
Using SWATH-MS analysis, a total of 3,156 proteins were identified in both tumor tissue and adjacent normal pancreas. Principal component analysis (PCA) was performed on the proteomic data obtained by SWATH-MS analysis of tumor tissue and adjacent normal pancreas. PCA is an unsupervised class recognition approach, to observe inherent groupings (14). Tissues were observed to be clustered according to their class grouping (i.e., tumor tissue or adjacent normal pancreas) for all patients together (Figure 2A), good-responders (Figure 2B),  or poor-responders ( Figure 2C). These results indicate that a clearly distinct tumor and adjacent normal tissue specimens were obtained from the patients.

Differentially Regulated Proteins
There were 236 differentially expressed (log 2 > 2; p < 0.05) proteins in the tumor tissue in good-responders compared to their adjacent normal pancreas (Supplementary Table 1). Of these, 134 proteins were over-expressed and 102 proteins were under-expressed in the tumor tissue. In poor-responders, only 67 proteins were differentially expressed (23 over-expressed and 44 under-expressed; Supplementary Table 2). The top 10 over-and under-expressed proteins for both good-and poor-responders based on fold-change are reported in Tables 1, 2. The over-expressed proteins in good-and poorresponders showed distinct functional activity. In contrast, the majority of proteins which were under-expressed in both goodand poor-responders, shared similar functional (proteases or peptidase) activity with 7 out of top 10 proteins being the same.

Comparative Pathway Analysis
Next, based on the identified differentially regulated proteins in both good-and poor-NAC responders, pathway analysis was performed using Ingenuity Pathway Analysis. A number of canonical pathways were observed to be differentially regulated in good-and poor-NAC responders (Figure 3A and Supplementary Table 3). Notably, immune response pathways, such as acute phase signaling and macrophage mediated nitric oxide and reactive oxygen species production, were upregulated in good responders but remained unaffected in poor-responders. Similarly, analysis of predicted disease and functions based on differential protein expression using IPA, supported an immunogenic phenotype in good-responders, while poor-responders showed only mild inflammatory response and phagocyte migration ( Figure 3B).

Biomarker Analysis
There were 19 proteins which were markedly (log 2 > 2) and significantly (p < 0.05) over-expressed in tumor from the poor-responders compared to good-responders ( Table 3). The ability to these proteins to predict chemo-resistance to NAC was determined by area under the receiver operator characteristic (AUROC) curve. Four biomarkers, namely GRP78, CADM1, PGES2, and RUXF, demonstrated very high predictive performance with AUROC ≥ 0.92.

DISCUSSION
Neoadjuvant chemotherapy (NAC) is being increasingly given to PDAC patients with borderline/locally advanced disease and is also being evaluated in upfront operable patients. Previous  (15)(16)(17). Moreover, RUXF and PGES2 are known to be involved in chemo-resistance in ovarian and colorectal cancer, respectively (18,19). On the other hand, CADM1 is shown to be a good prognostic marker in other cancers (20,21). Four proteins (i.e., TMED2, AGR2, JTB, and CADM1) among the over-expressed proteins in poor-responders ( Table 3) were predicted to be secreted extracellularly. This is important, as detection of these proteins in plasma/serum from the PDAC patients could be used to develop a simple blood-based test for determining NAC response in PDAC patients. Of note, TMED2, AGR2, and JTB are known to be associated with poor prognosis in other cancers (22)(23)(24) and thus, could be explored as novel biomarkers for predicting chemo-resistance in PDAC. Future studies, assessing the levels of these biomarkers in serum or plasma from the NAC-treated PDAC patients, will be required to confirm the clinical utility of these biomarkers as an indicator of chemo-resistance in PDAC.
This study also compared tumor tissue with adjacent normal pancreas in both good-and poor-NAC responders. A distinct proteomic profile of over-expressed proteins in tumors was observed in good-and poor-NAC responders compared to the adjacent normal pancreas. Rho guanine nucleotide exchange factor 18 (Uniprot: ARHGI_Human) was the most over-expressed protein in good-responders. This latter protein is known to be up-regulated in response to reactive oxygen species (25), which are known to be increased in tumor tissue treated with chemotherapy (26). Periostin (Uniprot: POSTN) was identified as the highest over-expressed protein in poor-NAC responders. Periostin is an extracellular matrix protein, which is known to play an important role in cancer progression (27). Notably, periostin expression has also been shown to be associated with chemoresistance in pancreatic and other cancers (28)(29)(30). Previous studies have also demonstrated periostin as a poor prognostic biomarker in PDAC and other cancers (31)(32)(33)(34). In PDAC, periostin is produced by pancreatic stellate cells and it is shown to establish a microenvironment that is supportive for cancer growth and progression (31,35). Identification of periostin as a highly over-expressed protein in poor-NAC responders in this study further supports its important role in PDAC chemo-resistance.
Assessment of pathway analysis revealed that the SPINK1 pathway was upregulated in both good-and poor-NAC responders, with the latter having more pronounced involvement of this pathway. SPINK1 is a serine protease inhibitor which has an anti-trypsin activity and is known to play an important role in protecting the normal pancreatic tissue from inadvertent activation of trypsin (38). Moreover, SPINK1 is also shown to play a role in cancer cell survival and progression (39)(40)(41). In this study, we observed that levels of SPINK1 (Uniprot: ISK1_Human) were decreased in tumor tissue compared to adjacent normal pancreas. This is consistent with previous studies demonstrating higher levels of SPINK1 in normal pancreatic tissue compared to tumor (42,43).
The pathway analysis further demonstrated that innate immune response was highly activated in tumors from the good-NAC responders, while only moderate immune activity was observed in poor-NAC responders. It can be postulated that initial response to NAC in good-responders could have resulted in a heightened immune infiltration into these tumors, resulting in an overall increased anti-tumor response. Studies have also shown similar immune-stimulatory effect of chemotherapy in other cancers (44,45), but this is the first study to observe this effect in PDAC.
The main limitation of this study is a relatively small cohort size. Future multi-institutional studies with a larger group of patients will be required to independently validate the identified proteins and their predictive value. This study utilized tumor specimens obtained at the time of surgical resection after chemotherapy treatment. Future studies will be required to further validate these findings using pre-NAC endoscopic ultrasound (EUS) core biopsies. Notably, EUS core biopsy provides sufficient amount of protein (∼1 µg) required for SWATH-MS analysis, which highlights the future clinical utility of these biomarkers in selecting patients for NAC prior to surgery.

CONCLUSION
Overall, this exploratory study has demonstrated the successful application of SWATH-MS proteomic analysis to pancreatic tumor and normal pancreas tissue samples, resulting in the identification of novel potential biomarkers which may predict for a chemo-resistant tumor phenotype in PDAC patients treated with NAC. Further research in a larger patient cohort is required to validate these findings.

DATA AVAILABILITY STATEMENT
The datasets generated for this study can be found in the ProteomeXchange via the PRIDE database (46) (Accession: PXD017051).

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by