Next-generation sequencing of endoscopically obtained tissues from patients with all stages of pancreatic cancer

2.1 Tissue samples and ethics

We retrospectively evaluated 185 endoscopically obtained biopsy samples collected from 172 patients with pancreatic cancer who underwent EUS-FNA or endoscopic biopsy of an invading tumor at the Yamanashi University Hospital between July 2014 and December 2019. The EUS-FNA procedure was performed using 22- or 25-gauge needles of the normal, side hole, or Franseen type (EZ Shot 2 and EZ Shot 3 Plus, Olympus Medical Systems; Expect and Acquire, Boston Scientific). We punctured each sample 1-6 times (20 strokes/puncture) with negative pressure using a 20-mL syringe. Although rapid on-site cytologic evaluation cannot be conducted at our institution, we requested rapid cytologic evaluation at the Department of Pathology after one or two punctures to reduce the number of punctures. This retrospective study used residual samples of EUS-FNA specimens obtained for clinical diagnosis, and no additional puncture was performed for genetic analysis.

After eliminating duplicate samples from the patients and samples with failed NGS analysis or technical errors, 141 samples were finally analyzed by NGS (Figure 1A). A flow chart of the study is shown in Figure 1B. Tissue samples were obtained as 8-μm-thick sections derived from one or two FFPE blocks, and tumor components were separated from these sections using a Laser Capture Microdissection System (LCM, ArcturusXT, Life Technologies). A definite pathological diagnosis requires a certain amount of tumor cells or tumor area because small amounts of tumor-like cells appear similar to cells with degeneration caused by inflammation or mechanical damage. Thus, in samples with negative pathology, we excised the area where the tumor was presumed to be present because it could contain tumor cells.

DNA was extracted from LCM specimens as previously described.^{5, 6} DNA from the biopsied specimens was extracted using GeneRead DNA FFPE Kits (QIAGEN) according to the manufacturer’s specifications. The quantity and quality of the extracted DNA was assessed using a NanoDrop instrument (Thermo Fisher) and Qubit platform (Thermo Fisher). The study was conducted in compliance with the principles of the Declaration of Helsinki and was approved by the Human Ethics Review Committee of Yamanashi University Hospital (Receipt number: 1326 and 1847). Written informed consent was obtained from all study participants. The study protocol was registered at UMIN Clinical Trials Registry (UMIN000044562).

2.2 Genetic mutational analysis of tissue samples using next-generation sequencing

Genetic analysis of the tumor specimens was performed by amplification of the extracted DNA (10 ng) using an Ion AmpliSeq Custom panel (Thermo Fisher) targeting 78 genes (Table S1) and containing 451 primer pairs. The custom panel differed from the ready-made panel used in our previous study.⁸ It was designed using the AmpliSeq Designer (Thermo Fisher) to cover pancreatic cancer driver genes and actionable targets listed in OncoKB.⁹ OncoKB is a knowledge base for precision medicine used to infer whether an existing molecular-targeted drug will be effective. Sequencing was performed with an Ion Chef System and an Ion Proton Sequencer (Life Technologies) using an Ion PI Hi-Q Chef Kit (Life Technologies). Data obtained in the present study were shared with the Japanese Genotype-Phenotype Archive (JGAS000315/JGAD000426).¹⁰

2.3 Identification of gene alterations and suspected microsatellite instability

Gene mutations and CNV were identified using Ion Reporter software version 5.10 (Thermo Fisher). Only mutations and CNV with a mutant allele frequency >4% (with a sequence read depth >100) and copy number >6 were considered truly present in the tissues to avoid false-positive variants due to sequencing errors. In the present study, the number of gene alterations was defined as the number of all mutations and CNV among the 78 cancer-related genes that were sequenced. Thus, all multiple mutations of the same gene and all additional mutations other than hotspot mutations in the COSMIC database were included. Detected gene alterations were then matched and classified as four levels according to OncoKB,⁹ which classifies genetic alterations based on actionability as follows: levels 1-3A, standard therapeutic intervention or compelling clinical evidence for the disease; level 3B, the presence of clinical evidence of another disease; and level 4, the presence of compelling biological evidence.

Microsatellite instability was detected with the MSIsensor2 tool,¹¹ which uses machine learning methods to calculate the length distributions of microsatellites per site from sequence reads. The Ion AmpliSeq Custom panel used in the present study contained 20 homopolymer sites and three microsatellite sites consisting of a 2-6 base repetition in its target region. The microsatellite instability (MSI) score was calculated as the percentage of abnormal length distribution in the repeating sequences among total homopolymer or microsatellite sites. MSI was confirmed when the MSI score was ≥88, which was the cutoff value determined by a comparison of immunohistochemistry (IHC) of mismatch repair (MMR) genes (Figure S1).

2.4 Immunohistochemistry of mismatch repair genes

Anti–MLH1 (1:250 dilution; ab92312; AbCam plc), anti–MSH2 (1:8000 dilution; ab227941; AbCam plc), anti–MSH6 (1:500 dilution; ab92471; AbCam plc), and anti–PMS2 (1:100 dilution; ab110638; AbCam plc) antibodies were used as primary antibodies. IHC staining was performed according to the manufacturer’s instructions. Briefly, 3-μm-thick deparaffinized sections of FFPE were stained using primary antibodies specific for the above MMR genes. Antigens were retrieved by boiling tissue sections in Target Retrieval Solution (Dako). Envision+Dual Link HRP (Dako) was used as a secondary antibody and diaminobenzidine was used as the chromogen. IHC staining was blindly examined by two independent investigators.

2.5 Statistical analysis

Data that did not display a normal distribution were presented as the median (range) or median (interquartile range [IQR]). Factors associated with overall survival were identified using Cox multivariate regression analysis, in which P < .05 was considered significant. All statistical analyses of recorded data and graphic creations were performed using the Lifelines program (https://zenodo.org/badge/latestdoi/12420595) with the Python platform.

3.1 Patient characteristics and qualitative assessments of extracted DNA and next-generation sequencing

Among 147 samples that underwent NGS testing, 141 (96%) were successfully analyzed (Figure 1A); therefore, gene alterations in samples from 141 patients were analyzed with their clinical information. Table 1 shows the clinical characteristics of the 141 patients included in the study. Tumor stages were defined according to the TNM Classification of Malignant Tumors by the Union for International Cancer Control,¹² and 4, 62, 16, and 59 patients in this study were classified as stages I, II, III, and IV, respectively. A total of 59 (42%) patients had stage IV pancreatic cancer and 82 (58%) received chemotherapy. Tissue samples were obtained by EUS-FNA or endoscopic biopsy in 129 (91%) and 12 (9%) cases, respectively (Table S2). A pathological diagnosis of malignancy was made using histology in 100 (71%) patients and cytology in 34 (24%) patients. However, a pathological diagnosis could not be made prior to therapy in seven (5%) patients, among whom three (2%) were diagnosed with malignancy using pathology of resected tissues, and four (3%) were diagnosed based on clinical course, respectively.

The median (range) amounts of extracted DNA from the FFPE samples obtained by EUS-FNA or endoscopic biopsy were 20.6 (1-339) ng, except for 10 samples that had DNA concentrations below the limit of detection (LOD). In the NGS analyses of endoscopically obtained tissues, the target regions of the 78 cancer-related genes included 44 140 bases, with median (range) sequenced read depths of 4568 (448-29 865). Data for EUS-FNA and duodenal biopsies are shown in Table S2. The yield of extracted DNA (median [IQR]) was lower in EUS-FNA samples compared with that from duodenal biopsy (19.2 [10.2-34.2] ng vs 132.9 [79.7-202.7] ng, respectively; P < .001), whereas the number of samples with mutations in KRAS and any gene was not different between these two sets of samples (110/129 [85%], EUS-FNA vs 10/12 [83%], duodenal biopsy; P = 1.000) and (122/129 [95%], EUS-FNA vs 11/12 [92%], duodenal biopsy; P = .518) (Table S2).

3.2 Overview of genetic abnormalities of endoscopically obtained specimens in pancreatic cancer

Alterations to pancreatic cancer genes in the endoscopically obtained tissue samples were identified in 75 of the 78 analyzed cancer-related genes, and alterations of any type were detected in 134 cases (91%, Figure 2). Gene alterations were detected in eight of the 10 samples with extracted DNA concentrations below the LOD. The four most frequently altered genes in the tissue samples were KRAS (84%), TP53 (57%), SMAD4 (19%), and RET (18%), followed by SMO (16%), FGFR3 (14%), BRCA2 (10%), NF1 (10%), and CDKN2A (9%). When summarized according to signaling pathway, the Ras (KRAS, FGFR3, EGFR, ERBB2, FGFR1, ABL1, FGFR2, EGFR-AS1, FLT3, KIT, MET, PDGFRA, BRAF, RAF1, MEK, and ERK), mTOR (KRAS, RET, FGFR3, PIK3CA, MTOR, PTEN, ERBB2, FGFR1, ABL1, FGFR2, ALK, KIT, MET, PDGFRA, NRAS, AKT1, MTOR-AS1, ERGR, and PI3K), cell cycle (TP53, CDKN2A, and RB1), JAK/STAT (ERBB2, ABL1, ALK, FLT3, JAK2, JAK2, and ERGR), and Wnt (APC, CTNNB1, and CDH1) signaling pathways were activated by their constituent gene alterations in 89%, 88%, 62%, 11%, and 9% of samples, respectively.

3.3 Actionable genes in pancreatic cancer

We next matched the detected gene alterations with OncoKB, which classifies genetic alterations into four levels according to an actionability scale. Oncogenic mutations in KRAS were most frequently detected in the cohort, and they were associated with resistance to molecular targeted drugs, such as cetuximab and panitumumab (Table S3). The next most frequently detected actionable genes were BRCA2 and PIK3CA, which were associated with the efficacy of talazoparib, olaparib, rucaparib, and fulvestrant plus alpelisib. In total, 19.9% of patients with gene alterations corresponded to levels 1-2 of OncoKB (Table S4). Furthermore, germline mutations in homologous recombination repair (HRR)-associated genes, which are efficacy markers for olaparib,¹³ were detected in 3% of patients. None of the patients had a G12C mutation in KRAS, a specific target for sotorasib,¹⁴ and MSI, which is an efficacy marker for immune checkpoint inhibitors,¹⁵ was detected in 5% of the patients (Figure 2).

3.4 Genetic diagnosis of malignancy in pathologically negative cases

Pathological diagnoses of endoscopically obtained samples were classified into three groups: histology-positive, with histological malignancy findings using EUS-FNB or duodenal biopsy (n = 113, 80.1%); FNB-negative, with negative findings using EUS-FNB but positive findings using EUS-FNA cytology (n = 19, 13.5%); and fine-needle aspiration biopsy (FNAB)-negative, with negative findings using EUS-FNB and EUS-FNA cytology (n = 9, 6.4%). Two, three, and four of the nine FNAB-negative patients were diagnosed with malignancy using cytological testing of bile juice or ascites, pathology of surgical specimens, and imaging and bad clinical course, respectively. Patients who were pathologically negative using either EUS-FNB or EUS-FNA cytology had gene alterations in KRAS or GNAS and at least one of 78 genes in four (44%) and six (67%) of the nine cases, respectively, suggesting the diagnostic ability of genetic analysis even in FNAB-negative cases (Figure 3).

3.5 Genetic and clinical factors associated with overall survival

Survival analysis was performed because the relationship between the genetic and clinical factors, including overall survival, remains poorly understood. We first analyzed the survival rates of patients treated using FOLFIRINOX (n = 20) according to HRR gene status because HRR status is related to the therapeutic effect of the platinum-containing regimen. We revealed a tendency toward better survival in cases with HRR mutations (Figure 4A). We next performed Cox regression analysis for overall survival with genetic and clinical factors (Table 2) and revealed that the presence of metastasis (P < .005; hazard ratio [HR], 3.30; Figure 4B) and serum CEA levels >5 ng/mL (P = .01; HR, 1.89) were associated with worse prognosis. Because metastasis was found to have a strong influence on overall survival, we then performed Cox regression analysis by stratifying metastasis and revealed that the number of mutations <3 (P = .03; HR, 2.48, Figure 4C) and serum CEA levels >5 ng/mL (P < .005; HR, 3.94; Figure 4D) were associated with worse prognosis in cases without and with metastasis, respectively (Table 3). We also stratified patients into those who underwent surgical resection from the viewpoint of clinical common sense and then performed Cox regression analysis (Table S5). No factors influenced the prognosis in surgically treated cases, whereas metastasis (P < .005; HR, 4.16) and CEA levels >5 ng/mL (P < .005; HR, 2.28) were associated with poor prognosis in patients who did not undergo surgical resection.