Accurate Prognosis Prediction of Pancreatic Ductal Adenocarcinoma Using Integrated Clinico-Genomic Data of Endoscopic Ultrasound-Guided Fine Needle Biopsy

Simple Summary Around 10–20% of patients with pancreatic adenocarcinoma (PDAC) have a curative surgery, but most of them perform endoscopic ultrasound-guided fine-needle biopsy (EUS-FNB) for the pathologic confirmation. The aim of our retrospective study was to evaluate the feasibility of targeted sequencing using EUS-FNB minimal specimens for the prognosis prediction of PDACs. We found some clinical factors and genetic alterations significantly related to the metastasis and overall survival, and established a clinico-genomic model by using both clinical parameters and genetic alterations to predict the prognosis of patients with PDACs. Abstract The aim of this study was to investigate the clinical utility of minimal specimens acquired from endoscopic ultrasound-guided fine-needle biopsy (EUS-FNB) and perform targeted deep sequencing as a prognosis prediction tool for pancreatic ductal adenocarcinoma (PDAC). A total of 116 specimens with pathologically confirmed PDAC via EUS-FNB were tested using CancerSCAN® panel for a customized targeted deep sequencing. Clinical prognostic factors significantly associated with survival in PDACs were as follows: stage, tumor mass size, tumor location, metastasis, chemotherapy, and initial CA19-9 level. A total of 114 patients (98.3%) had at least a single genetic alteration, and no mutations were detected in two patients, although they were qualified for the targeted deep sequencing. The frequencies of major gene mutations responsible for PDACs were KRAS 90%, CDKN2A 31%, TP53 77%, and SMAD4 29%. A somatic point mutation of NF1, copy number alteration of SMAD4, and loss-of-function of CDKN2A were significantly associated genetic factors for overall survival. Moreover, BRCA2 point mutation was related to liver metastasis. Finally, a clinico-genomic model was developed to estimate the prognosis of patients with PDAC based on clinical parameters and genetic alterations affecting survival in patients; 20 single nucleotide variants and three copy number variations were selected. Targeted deep sequencing on minimal specimens of PDACs was performed, and it was applied to establish a clinico-genomic model for prognosis prediction.


Introduction
Pancreatic ductal adenocarcinoma (PDAC) is the fourth leading cause of cancer-related mortality in the United States of America [1], and the fifth leading cause in Korea [2]. The five year overall survival (OS) rate for PDAC is approximately 7%, which is attributed to the typically advanced stage at diagnosis, combined with the lack of effective systemic chemotherapy available [3]. Predicting the prognosis and identifying patients who could benefit from specific therapies could improve the OS of patients with pancreatic cancer.
The factors responsible for the survival of patients with PDAC are poorly understood. The favorable prognostic factors of PDAC are primarily clinical factors, such as early disease stage, negative resection margins, and negative lymph nodes [4]. Current clinical prognostic factors produce limitations in predicting the prognosis of PDAC and selecting suitable patients for specific systemic chemotherapy regimens.
Most PDACs are unresectable when first diagnosed and endoscopic ultrasoundguided fine-needle biopsy (EUS-FNB) is the first choice for pathologic confirmation, and those minimal specimens can be widely used for research purposes compared to surgical specimens. Therefore, EUS-FNB has become a most important tool to explore the majority of PDAC tumor biology. Understanding the underlying molecular mechanisms of PDAC could promote the development of a new methodology for early diagnosis, and facilitate improvement in current approaches for PDAC treatment. Molecular profiling of tumor specimens has revealed potential targets for personalized therapy. Next-generation sequencing (NGS) allows parallel sequencing of DNA, which can advance the understanding of the underlying molecular pathophysiology of cancer [5]. Recent NGS genomic analyses of PDAC samples have revealed a complex mutational landscape from surgical specimens [6,7]. Minimal specimens of PDAC may be relatively hypocellular, meaning that small numbers of cancer cells can be outnumbered by stroma or inflammatory cells. Therefore, EUS-FNB specimens frequently yield an insufficient amount of DNA unfit for molecular analysis [8]. Some studies have described the feasibility of NGS using EUS-FNB acquired pancreatic tumor specimens, and demonstrated the superiority of NGS over conventional sequencing in detecting specific genomic alterations in EUS-FNB [9]. The aims of this study were to evaluate the feasibility of targeted sequencing from EUS-FNB minimal specimens, and to determine if the genomic signature of primary PDACs could be correlated to clinical parameters associated with the prognosis of PDACs.

Baseline Characteristics
A total of 166 EUS-FNB samples were collected, and 116 samples passed the quality control test of DNA analytes (Table S1) to obtain their genomic profiles through targeted deep sequencing. The schematic flow of study design is shown in Figure 1. Survival data was obtained from the hospital database, and information on mortality was collected via telephone calls in cases of loss to follow-up. The clinical and pathologic characteristics of the 116 minimal specimens from the study patients are shown in Table 1. The median age was 64 years old (range 26-83 years) and male to female ratio was 67:49 (1.4:1). According to the 8th American Joint Committee on Cancer (AJCC) staging system, one (0.9%) patient was stage IB, six (5.2%) were stage IIA, eight (6.9%) were stage IIB, 26 (22.4%) were stage III, and 75 (64.6%) were stage IV. The most common location of the tumor was the head (42.2%), followed by the body (35.4%) and tail (22.4%) of the pancreas. The most common metastatic site was the liver (60%). Among the 116 study patients, 13 (11.2%) patients underwent surgical resection with curative intention, and 11 (84.6%) of them followed by adjuvant therapy. A total of 92 (79.3%) patients were treated with palliative chemotherapy. In addition, 24 (20.7%) patients were treated with best supportive care only.

Clinical Prognostic Factors Affecting OS and Clinical Model
The clinical factors associated with the OS of patients with PDAC were evaluated. Median OS was 8.4 months (range 0.77-103.0 months). Age, gender, BMI, CA 19-9 level, chemotherapy, tumor mass size, tumor location (head, body, tail), clinical stage, and metastasis were used for analysis (Figure 2A Figure S1). For the variables included in the optimized clinical model, chemotherapy, mass size, and CA19-9 were selected using a stepwise method, and age and stage were adjusted ( Figure 3A). This model could separate the high-risk group (median OS, 8.33 months) from the low-risk group (median OS, 9.93 months) for PDAC (HR = 2.11 (95% CI, 1.44-3.10); p = 0.0001) ( Figure 3B).

Development of Clinico-Genomic Model
To develop a prognostic model for PDAC patients, additional useful genetic alterations were included based on the clinical model. Since genetic testing provides additional information to the clinical model, the clinico-genomic model used the forward variable selection method as a way to find genes that additionally contribute to the clinical model ( Figure 6A). Each gene was used as a binary variable according to the presence or absence of alteration, such as SNVs and small InDels in each gene region. When each gene has CNV, it was treated as a separate variable. It successfully estimated the high-risk group (median OS, 5.95 months) and the low-risk group (median OS, 15.27 months) based on clinical factors and genetic alterations (HR = 6.06 (95% CI, 3.84-9.55); p < 0.001, Figure 6B).

Discussion
In the present study, we analyzed various clinical factors and genetic alterations in relation to the survival of PDAC patients. Moreover, we developed a clinico-genomic model and successfully predicted the clinical outcome for PDACs. Approximately 10 to 20% of patients with PDAC are operable [3], thus, personalized genetic information is not available for most patients. EUS-FNB is a primary method to acquire tissue samples of pancreatic cancer for pathologic confirmation. We conducted targeted sequencing with EUS-acquired tissue specimens, which are more accessible than surgical specimens. Success rates for NGS analysis of solid tumor tissue specimens were dependent on the types of samples being sequenced. The rate of NGS success was the highest in resection specimens (97%), reduced within biopsy specimens (80%), and was lowest for cytologic specimens (50%) [13]. Several studies have reported that the adequacy of EUS tissue acquisition samples for deep sequencing ranged from 60 to 100% in pancreatic solid mass [14]. In our study, we successfully performed targeted sequencing with EUS-FNB samples in 116 patients (70%) among 166 patients with PDAC. Interestingly, a report showed that the observed EUS-FNA cytology mutational spectrum detected with a commercial 160 cancer gene panel was broader than that of the corresponding surgical pathology specimens [15]. Advanced NGS technologies can be applied to the EUS sample, which allowed the unveiling of potential genomic biomarkers and therapeutic targets [14]. Several studies have reported the adequacy of EUS samples for NGS. Elhanafi et al. [16] analyzed 167 EUS samples of PDAC patients by targeted sequencing with a 47 gene panel, and frequency of genomic alterations were KRAS (88%), TP53 (68%), and SMAD4 (16%). Young et al. [17] performed NGS using a customized gene panel (287 genes) with formalin-fixed paraffin-embedded EUS samples, and the most common mutations were observed in KRAS (78%), TP53 (74%), CDKN2A/B (35%), SMAD4 (17%), and PTEN (13%). Among 116 samples, somatic mutations were identified in 114 samples (98.3%). The missing two samples without any detected mutations might have few neoplastic nuclei due to the minimal tissues of EUS-FNB. For main genetic drivers (KRAS, TP53, CDKN2A, and SMAD4) in PDAC carcinogenesis [3,6,7], KRAS was found to be the most commonly mutated gene (90%) in our PDAC specimens, and TP53 (77%), SMAD4 (29%), CDKN2A (31%), and BRCA2 (4%) were also commonly mutated at comparable rates with other studies using EUS samples [14] and surgical samples [7] (Figure S3). Numerous other genes harbor mutations at lower frequencies. This indicates that we successfully performed targeted panel sequencing with minimal EUS-FNB specimens.
Known prognostic factors related to survival in patients with PDAC are clinicopathologic factors, including CA19-9 levels, disease stage, chemotherapy treatment, performance status, surgical margin status, perineural invasion, and blood vessel invasion [18]. We also identified tumor location and tumor mass size as independent prognostic factors ( Figure 2).
Among the genes in the CancerSCAN ® panel, SNVs of NF1 and deleted copy number loss of SMAD4 were associated with survival in univariate analysis ( Figure 5A,B). A report showed that NF1 deficiency correlated with estrogen receptor phosphorylation and poor survival in breast cancer [19]. Although there is a case report showing an association of NF1 with OS in pancreatic endocrine tumors [20], NF1 mutation in PDAC has not been reported in terms of clinical outcome. SMAD4 deletion has been commonly reported in 50% of PDAC, resulting in aberrant signaling by the transforming growth factor beta (TGF-β) cell surface receptor [3]. Patients with PDAC with SMAD4 protein expression had significantly longer survival [21]. Patients having SMAD4 gene inactivation, by either intragenic mutation or homozygous deletion, had significantly worse survival than patients with intact SMAD4 [22]. Recently, SMAD4 and NF1 mutations were reported as potential biomarkers for poor prognosis to cetuximab-based therapy in metastatic colorectal cancer patients [23]. Moreover, patients with LOF mutation of CDKN2A showed the poor outcome in Figure 5C, and it was also validated in the cohort data from TCGA ( Figure 5D). It was reported that somatic mutations of CDKN2A, a tumor suppressor factor, are present in up to 95% of pancreatic tumors [24]. The inactivation of CDKN2A induced by homozygous deletion or promoter hyper-methylation and point mutation contributes to cancer progression in various cancers, like PDAC [25][26][27].
Integration of genomic and transcriptional features in pancreatic cancer revealed that cell cycle progression was increased in PDAC metastases with driver gene (KRAS, TP53, CDKN2A, SMAD4) loss [28]. Moreover, mutations in known driver genes in the metastases of individual patients with PDAC were remarkably uniform [29], and no significant differences were identified at the gene or pathway level when comparing genomic alterations between primary and metastatic pancreatic tumors [30]. Many studies have reported some genes which may be involved in tumor metastasis, such as KAI1 and nm23-H1 in pancreatic cancer [31]. KAI1, a putative metastasis suppressor gene, appears to be crucial for the development of distant organ metastasis, since its expression was downregulated in lymph node and distant metastasis [32]. Stratford et al. analyzed the gene expression profiles of primary tumors from patients with localized compared to metastatic disease, and identified a six-gene signature (FOSB, KLF6, NFKBIZ, ATP4A, GSG1, SIGLEC11) associated with metastatic disease [33]. These genes were prognostic for survival in PDAC. We also analyzed genetic alterations associated with metastasis, in which BRCA2 mutation was found to be related to liver metastasis rather than other sites, such as lung and peritoneum ( Table 2). BRCA2 germline mutation is also one of the most common causes of familial pancreatic cancer. BRCA2-associated PDAC has a more favorable outcome than non-BRCAassociated PDAC [34]. Moreover, BRCA2 mutations made the disease biologically more chemosensitive, which, consequently, prolong survival despite the prognostically unfavorable disease [35]. Patients carrying BRCA2 pathogenic mutations are more likely to progress to metastasis in prostate cancer [36]. This is the first report that missense mutations in BRCA2 were relevant in PDAC patients with liver metastasis instead of other sites. However, further functional studies are needed to elucidate these mechanisms.
By combining clinicopathologic factors with genetic alterations, we established an integrated clinico-genomic model to predict the survival of patients with PDAC ( Figure 6). Five clinical factors, including age, stage, chemotherapy, tumor mass size, and CA19-9 level, and 23 genetic alterations were selected based on the targeted panel sequencing profile using a stepwise variable selection method. The clinico-genomic model was developed and successfully estimated the prognosis of PDAC. Clinical parameters such as age, stage, chemotherapy, tumor mass size, and plasma CA19-9 level could be collected soon after diagnosis, and EUS-FNB is performed for initial PDAC diagnosis. Therefore, it seems that the clinico-genomic model is useful for prognosis prediction of PDAC patients at the early period of diagnosis with genetic analysis using EUS-FNB samples. However, these require further investigation with more patients with PDAC.
The present study had several limitations. The study design is retrospective, and patient populations were heterogeneous. They received different treatment regimens, and were diagnosed at various stages (resectable and unresectable PDAC). Moreover, we excluded some patients who proceeded directly to surgical resection without EUS-FNB, and they might have had a better prognosis. We performed targeted sequencing using CancerSCAN ® panel instead of whole genome sequencing, which made it impossible to evaluate the correlation between genes not included in the panel and the survival of patients with PDAC. Moreover, we could not verify the clinico-genomic model with the open database, like TCGA, because it was not eligible for some clinical factors, such as chemotherapy, tumor mass size, and CA19-9 level used for this model. Only the validation of genetic alteration related to survival in pancreatic cancer was performed with the TCGA database ( Figure 5). Finally, most novel genetic alterations correlated with survival, metastasis, metastatic sites, and chemotherapy responses and did not elucidate the mechanism of gene function in terms of related factors.
We demonstrated the applicability of targeted NGS and detection of novel genetic mutations associated with prognosis and metastasis in PDAC. Genetic information, in addition to currently utilized clinical information, could help to predict the prognosis of patients, which was proved by the establishment of a clinico-genomic model for prognosis at diagnosis. This provides a more comprehensive view of the genomic landscape of PDAC, with implications for improved clinical decision making in the context of the implementation of tailored therapies. Further functional investigations and validation studies are needed to fully elucidate the mechanism of novel genetic mutations.

Patients
Overall, 166 patients with pathologically confirmed PDAC who underwent EUS-FNB between 30 November 2011 and 31 December 2014 at the Samsung Medical Center were enrolled. Among them, 116 who passed a DNA quality control test were analyzed for their genomic profiles by targeted CancerSCAN ® panel sequencing. Written informed consent was obtained from each patient. This study was approved by the Institutional Review Board (IRB) of Samsung Medical Center (IRB No. 2014-04-048). Clinicopathologic data was collected, including age, gender, body mass index (BMI), smoking status, performance status (Eastern Cooperative Oncology Group, ECOG), operations, chemotherapy responses, tumor size, tumor location, overall/TNM stage of the disease, serum levels of carcinoembryonic antigen (CEA), carbohydrate antigen 19-9 (CA19-9), and overall survival. Electrical medical records were reviewed and clinical AJCC 8th system were used for the staging.

EUS-FNB Procedure
All EUS-FNB procedures were performed to obtain the primary pancreatic cancer tissue specimen under conscious sedation by experienced endosonographers with GF-UE160-AL linear EUS apparatus (Olympus, Tokyo, Japan) equipped with an Aloka ProSound SSD-5000 (Wallingford, CT, USA). EchoTip ® Ultra Endoscopic Ultrasound Needle or En-doTip ProCore ® HD Ultrasound Biopsy Needle (19, 22, or 25 gauges; Wilson-Cook Inc., Winston-Salem, NC, USA) were used, allowing several passes under direct endosonographic visualization to obtain sufficient cellular aspirates for cytologic analysis. A transgastric approach was performed on lesions in the body or tail of the pancreas, and a transduodenal approach was used for lesions in the head or uncinate process. For tissue retrieval, the stylet was introduced into the needle, or the needle was flushed with a 5 to 10 mL air-filled syringe. The extruded material was first used for primary gross inspection, and remaining tissues were snap-frozen in a screw vial and stored at −80 • C for DNA extraction. Informed consent for purpose and complication of EUS-FNB was obtained from all patients prior to procedure.

Targeted Deep Sequencing
Genomic DNA was extracted using a QIAamp DNA mini kit (Qiagen Inc., Valencia, CA, USA) according to the manufacturer's instruction. DNA concentration and purity were checked using a Nanodrop 8000 UV-Vis spectrometer (Thermo Scientific, Waltham, MA, USA) and Qubit 2.0 Fluorometer (Life Technologies, Grand Island, NY, USA). The degree of DNA degradation was measured using a 200 TapeStation Instrument (Agilent Technologies, Santa Clara, CA, USA). Qualified DNA was sheared using a focused-ultrasonicator (Covaris, Woburn, MA, USA) and used for library construction using CancerSCAN ® (Cancer Somatic mutation call for Clinical reports with Annotation) version 1 based on a customized SureSelect XT reagent kit (Agilent, Santa Clara, CA, USA). CancerSCAN ® panel, a targeted sequencing platform designed at Samsung Medical Center, offered flexibility to include target genes curated from the literature as requested by the researchers and clinicians. These selected genes covered variants associated with the targeted cancer therapies (i) approved by the Korean MFDS and US FDA, (ii) in the clinical trials at the Precision Oncology Clinic at Samsung Medical Center, or (iii) reported as having association with response of therapy in the public databases and the literature. It was designed to enrich the exons of 83 genes (Table S3), covering 366.2 kb and 2.59 Mb of the human genome [37]. After the enriched exome libraries were multiplexed, the libraries were sequenced on the HiSeq 2500 sequencing platform (Illumina, San Diego, CA, USA). Sequence reads were mapped to the human genome (hg19) using the Burrows-Wheeler Aligner. Duplicate read removal was conducted using Picard and SAMtools. Local alignment was optimized by The Genome Analysis Toolkit. Variant calling was only conducted in targeted regions of CancerSCAN ® . To detect SNVs, results of three types of variant callers (MuTect, Vardict, and LoFreq) were integrated to increase sensitivity. Pindel was used to detect InDels. CNVs were calculated for targeted regions by dividing read depth per exon by estimating normal reads per exon using an in-house reference. In analysis pipeline, we first removed the very common germline variant based on the population frequency after mutation calling, and flagged the benign mutation based on the known public database, pathogenicity database, and in-house database (data from about 6000 people). TCGA (PMID: 29625048) data was downloaded from cBioPortal, and we compared binary gene mutation status, such as nonsynonymous, truncation, and copy number alteration with survival. R version 4.0.2 was used for statistical analyses, and survival v3.1.12 and survminer v0.4.8 packages used for survival analysis.

Clinical and Clinico-Genomic Model
We used overall survival time as the endpoint of survival analysis. The clinical model was constructed using only clinical variables, and both clinical and genetic variables were included to develop a clinico-genomic model. In order to explain the necessity of genetic testing in clinic, initially, an optimal clinical model was developed using only clinical parameters. To prove clinical usefulness of genetic testing, a clinico-genomic model should show better performance than a clinical model. Therefore, a clinico-genomic model was developed by adding genetic parameters that help to improve performance based on the clinical model (similar to the forward variable selection method). All statistical models were developed with multiple Cox's proportional hazard regression method, which uses clinical parameters and mutation information for each gene as independent variables, and overall survival time and censoring data as response variables. The score calculated in this regression model was divided based on the median of calculated Cox score to separate the high-risk group and the low-risk group. Overall survival performance of each group was shown using a Kaplan-Meier graph, and log-rank method was used for significance testing.

Statistical Analyses
For all of the tests, significance level was set at 5% and reported as two-tailed pvalues. Statistical analyses were carried out using R software for Linux version 3.6.2 (The R Foundation, St. Miami, FL, USA).

Conclusions
It was successful to perform targeted sequencing with minimal specimen from EUS-FNB, and analyze genetic alterations associated with the prognosis of patients with PDAC. Combined with clinical information, genetic alterations were useful to predict the survival of PDACs.
Supplementary Materials: The following are available online at https://www.mdpi.com/article/10 .3390/cancers13112791/s1, Figure S1: Kaplan-Meier plots for clinical factors not significantly associated with survival, Figure S2: Additional genomic alterations identified by targeted deep sequencing in EUS-FNB specimens of patients with PDAC, Figure S3: Comparison of mutation frequency of main genetic drivers in PDAC,   Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.

Data Availability Statement:
The data presented in this study are available on request from corresponding authors.

Conflicts of Interest:
There is no conflict to disclose.