Rare germline variants in pancreatic cancer and multiple primary cancers: an autopsy study

Background There is a lack of information on rare germline variants of pancreatic cancer-predisposing genes. Risk genes for multiple primary cancers may overlap with those for pancreatic cancer. Methods A retrospective study of autopsy cases with a negative family history in the Japanese single nucleotide polymorphism for geriatric research database examined rare germline variants in the protein-coding regions of 61 genes. Targeted sequencing of these genes was performed and classified for pathogenicity using the American College of Medical Genetics and Genomics guidelines. Polyphen-2, SIFT and LoFtool algorithms were used to predict damage to protein function. Results Of the 189 subjects used (90 cancer and 99 non-cancer controls), 72 patients had pancreatic cancer (23 had multiple primary cancers) and 18 had no pancreatic cancer in multiple primary cancers. APC, BRCA2, BUB1B, ENG and MSH6 were associated with cancer predisposition, and pathogenic/likely pathogenic (P/LP) variants occurred in 6% [pancreatic cancer (4/72); all-cancer (5/90)] and 54% (49/90) carried only variants of uncertain significance (VUS) among cancer patients. Of these VUS, in pancreatic cancer patients, four DNA mismatch repair (MMR) genes (MLH1, MSH2, MSH6 and PMS2), and POLQ in men were significantly associated (odds ratio = 3.83; P = 0.025; P = 0.027, respectively). The most abundant predictor of functionally damaging variants was POLQ. Conclusions The frequency of P/LP variants in patients with sporadic pancreatic cancer suggests the need for genetic evaluation of individuals with no family history. VUS of MMR genes (MLH1, MSH2, MSH6 and PMS2) and POLQ may be useful in predicting genetic trends in the potential risk of pancreatic cancer, especially in individuals lacking P/LP.


Introduction
Most patients with pancreatic cancer are sporadic (Vietri et al., 2022) and the incidence of pancreatic cancer is estimated to increase further. The risk factors for pancreatic cancer include sex, smoking, alcohol consumption, diabetes and family history. Pancreatic cancer is refractory and has a low 5-year survival rate. The etiology of pancreatic cancer is not fully understood. The close agreement between the incidence and mortality rates indicates that only a small fraction of patients are detected in the early stages of efficient intervention. These features highlight the need for effective detection methods for high-risk populations (Amundadottir et al., 2004;Kleeff et al., 2016). However, little is known about germline variants that predispose to pancreatic cancer. primary cancers and non-cancers, using 61 gene panel sequencing. Most cancers occur in the elderly, imposing great public health and socioeconomic burden (Fane and Weeraratna, 2020). This case-control study is based on a database of consecutive autopsies registered for a study of geriatric disease, a design that characterizes the risk of pancreatic cancer in elderly individuals with a negative family history. Furthermore, this study is unique in that it includes centenarians of the same age and sex in the case and control groups.

Patient selection and ethics committees
Subjects for the case-control study were collected from consecutive autopsy samples of elderly individuals in Japan, referred to as JG-SNP: Japanese single nucleotide polymorphism for geriatric research (Sawabe et al., 2004) for the study of geriatric diseases. Autopsy samples were obtained from patients with pancreatic cancer and multiple primary cancers and those without cancer who were enrolled at Tokyo Metropolitan Geriatric Hospital between 1995 and 2012 and had no relevant family history. The control group consisted of non-cancer patients whose sex, age, smoking and alcohol consumption matched the patients in the case group as closely as possible. Patient medical records were referenced to obtain clinical data. Written informed consent was obtained from a family member of each patient involved in this study for the collection and research. This study was approved by the ethics committees of Tokyo Medical and Dental University (approval No. 02016-011), Tokyo Metropolitan Geriatric Hospital (approval No. R16-55) and Juntendo University (approval No. M18-0240).

DNA extraction, quantification and quality control
DNA was extracted from the kidneys of autopsy samples using a standard phenol/chloroform extraction method. The purity and concentration of the DNA extracted from all samples were measured using a NanoDrop 1000 spectrophotometer (Thermo Scientific, Wilmington, Delaware, USA) and a Qubit fluorometer (Life Technologies, Carlsbad, California, USA). All the samples fulfilled the minimum quality and purity requirements of 1 mg in less than a 130-ml volume, with an optical density OD260/280 of 1.7-1.9 and an OD260/230 of greater than 2. DNA fragmentation was assessed by agarose gel (2%) electrophoresis.
Polyphen-2 used the HumDiv model, an evaluation model for rare genes identified by genome-wide association studies, and the annotation by the respective algorithms of SIFT and LoFtool is mainly based on the Ensembl Variant Effect Predictor (https://asia.ensembl.org/Homo_sapiens/ Tools/VEP). The predictive annotations for these three algorithms were classified as possibly damaging (PD), benign (B) for Polyphen-2, deleterious (D), deleterious low confidence (DLC), tolerated low confidence (TLC), tolerated (T) for SIFT, and PD, B for LoFtool.

Statistical analysis
We statistically compared the associations between patients with and without cancer. In addition, the patients were examined separately according to their sex. The association between rare germline variants and cancer risk was estimated using the odds ratio (OR), confidence interval (CI), and P value for Fisher's exact test and calculated using the MedCalc (https://www.medcalc.org) software. Disease prevalence and gene counts were compared between groups using SPSS statistics software (version 20 IBM, USA) and confirmed using the parametric Student's t-test. The significance threshold for all tests was set at P < 0.05.

Results
Patient characteristics are summarized in Table 1. The mean age of the patients was 82 years and ± 8.09 standard deviation (cancer patients: 82 ± 8.003; controls: 82 ± 8.21). Of the 90 patients with cancer, 72 (80%) had pancreatic cancer, including 49 with pancreatic cancer only (PaC) and 23 with pancreatic cancer and multiple primary cancers (PaMC). Eighteen (20%) patients had no pancreatic cancer in multiple primary cancers (NPaMC). The control group comprised 99 patients.  Table 2 summarizes the rare germline P and LP variants. APC (P), BRCA2 (P), BUB1B (LP), ENG (LP) and MSH6 (P) were identified in patients with cancer.
were identified in men. In non-cancer patients, 43 variants were identified in women and 33 variants were identified in men. In the cancer group, 54% (49/90) of patients carried only VUS. The most frequent genes in patients with cancer were DNA polymerase theta (POLQ; n = 6) and MSH6 (n = 4) in women and POLQ (n = 4) in men (Fig. 1a,b).
We examined the association between VUS and the presence of cancer by using two statistical approaches in a case-control study. Table 3 summarizes the association between VUS and the presence or absence of cancer, using Fisher's exact test.
MSH6, four MMR genes (MLH1, MSH2, MSH6 and PMS2), and POLQ were significantly associated with both pancreatic cancer and all-cancer groups. MSH6 was significantly associated with pancreatic cancer (P = 0.029) and all-cancer groups (P = 0.023; women: P = 0.044). The ORs for MMR genes (MLH1, MSH2, MSH6 and PMS2) were significantly higher in both the pancreatic cancer (OR = 3.83; P = 0.025) and all-cancer groups (OR = 4; P = 0.019). POLQ was significantly associated with pancreatic cancer in men (P = 0.027) and all-cancer groups (OR = 4; P = 0.041). The mean number of VUS for MMR genes (MLH1, MSH2, MSH6 and PMS2) was significantly higher in the pancreatic cancer group (cancer patient, 0.13 vs. control, 0.04; P = 0.033) than that in the control group. In the all-cancer group, the mean number of VUS for MMR genes (MLH1, MSH2, MSH6 and PMS2) was significantly higher (0.14 vs. 0.04; P = 0.015) than that in the control group. The mean number of VUS was significantly higher in the all-cancer group (0.18 vs. 0.05; P = 0.044) compared to the control group in women.

Predicting functional effects of missense variants
In silico analysis using Polyphen-2, SIFT and LoFtool was annotated in cancer patients and controls for predicted effects of missense variants on protein function (Supplementary Table 1   Association of rare germline variants of uncertain significance of four DNA mismatch repair genes (MLH1, MSH2, MSH6 and PMS2) and POLQ in cancer patients and controls Cancer type The significance threshold was P < 0.05 and bold values indicate a significant difference. Pancreatic: includes patients with multiple primary cancers.
During data collection, some information was unclear, such as the rs ID numbers for variants, but was scrutinized as closely as possible. Further updating of information on the data sets used in this study is needed.

Discussion
We explored rare germline variants that encode proteins that predispose patients to sporadic pancreatic cancer. Panel sequencing of the 61 genes was performed using a case-control design in patients with a negative family history. Sequencing results revealed that 6% (4/72) of the patients, including one 104-year-old centenarian woman, with pancreatic cancer carried pathogenic cancer predisposition P/LP variants ( Table 2). VUS of LS-related germline MMR genes and POLQ were significantly associated with pancreatic cancer. Interestingly, these results suggest that trends in cancer risk may differ in women and men, respectively (Tables 3 and 4).
Our report on the pathogenic frequency of rare germline variants provides evidence that reaffirms the importance of predicting pancreatic cancer risk using genetic screening in individuals without a family history. Previous studies have reported that the frequency of germline pathogenic variants in patients with familial pancreatic cancer ranges from approximately 5 to >10% (Bartsch et al., 2004;Petersen et al., 2006;Chaffee et al., 2018;Rainone et al., 2020), supporting the frequency of 6% observed in the present study. Pancreatic cancer-predisposing genes were also present in multiple cancer phenotypes, and similar results were observed for the following two cancer subsets: all-cancer (5/90) and NPaMC (1/18) groups. Despite current cost concerns, we speculate that comprehensive genetic evaluations, such as multigene panels, may be beneficial for early detection and therapeutic measures even for individuals with a negative family history.
We identified five genes associated with pancreatic cancer predisposition in patients with cancer. P/LP variants of APC, BRCA2, BUB1B and ENG were identified in patients with pancreatic cancer, and MSH6 was identified in an NPaMC patient with both colorectal and esophageal cancers. No pathogenic P or LP variants were identified in any of these five genes in the control group (Table 2). In jMorp, a large Japanese database, BUB1B c.2441G>A (p.Arg814His) has been reported as a rare variant (MAF = 0.0039%); however, no records are available for APC c.4666dupA (p.Thr1556AsnfsX3), BRCA2 c.3853_3854insG (p.Glu1285fs), ENG c.728_729insCG (p.Pro244fs) and MSH6 c.1444C>T (p.Arg482Ter). The ENG c.728_729insCG variant is not listed in the ClinVar database (accessed 30 December 2022) and appears to be a novel variant. Meanwhile, the P/LP variants of AXIN2, BMPR1A, MSH2 and SMARCA4 were identified in 5% (5/99) of the subjects in the non-cancer control group, including one 104-year-old woman (Table 2), which did not significantly differ from the rate in the cancer group, and therefore the significance of risk is uncertain. These results also provide evidence that cancer risk prediction should be interpreted with caution and emphasize the complexity of understanding cancer mechanisms.
Germline variants can influence somatic variant patterns and genomic instability; ancestral populations with the same germline variants may exhibit different cancer phenotypes (Oak et al., 2020). One of our interesting findings is that de novo APC c.4666dupA (p.Thr1556AsnfsX3), which has been reported to cause an aggressive Gardner syndrome in a 2-year-old European boy (Kiessling et al., 2019), reproduced as a pancreatic, duodenal and colorectal cancer phenotype in a 104-year-old Japanese woman ( Table 2). The effect of predisposition, such as family history, other genes, or environment, on the phenotypic differences between the two patients, i.e. the boy and the centenarian, remains unclear. Whether the pathogenic variant APC c.4666dupA varies significantly by sex, age and race in terms of disease phenotype, age of onset and prevalence require clarification.
Overall, the observed landscapes of rare germline variants were diverse and heterogeneous (Fig. 1a,b) We then examined whether VUS carriers of specific cancer predisposition genes are more likely to develop cancer. Our study evaluated the potential effects of VUS on the risk of developing cancer in a case-control study and confirmed statistical associations among the following genes: MSH6, four MMR genes (MLH1, MSH2, MSH6 and PMS2) and POLQ. VUS for these genes was more abundant in patients with pancreatic cancer than in the non-cancer control group, and the same trend was observed in the all-cancer group (Tables 3 and 4). MSH6 tends to delay cancer development compared to MLH1 and MSH2 (Kolodner et al., 1999;Berends et al., 2002;Bonadona et al., 2011). Patients with multiple primary cancers (PaMC and NPaMC) with variants in MLH1, MSH2, MSH6 and PMS2 showed a trend consistent with various LS-related cancers (Geary et al., 2008;Kastrinos et al., 2009;Weissman et al., 2012;Win et al., 2012;Møller et al., 2018;Gambichler et al., 2021), although whether they had LS was unknown (Supplementary Table 3 Predicted effects of variants of uncertain significance (VUS) in patients with pancreatic cancer. The missense variants were annotated as potentially damaging protein functions in Polyphen-2, SIFT, and LoFtool. VUS in pancreatic cancer patients with a total rating of possibly damaging/deleterious (PD/D) were listed. POLQ was the most frequently identified variant in this study. CDH1 c.2638G>A (p.Glu880Lys) was also identified in one 67-year-old woman with hepatocellular carcinoma and colon cancer; not shown in this figure.
cancer included predicted possibly damaging and deleterious variants in MLH1, MSH2 and MSH6. Interestingly, POLQ were more abundant than those of other genes and some VUS might be loss-of-function variants in DNA double-strand break repair-related pathways. Genetic POLQ dysregulation plays an important role in genomic double-strand break repair and genome maintenance and is associated with the risk of pancreatic, breast, ovarian and other cancers, and overexpression is associated with poor prognosis (Higgins et al., 2010;Wood and Doublié, 2016;Slater et al., 2021). Here, our results also revealed that the VUS of MSH6 and POLQ suggested potentially different trends in association with cancer in women and men, respectively (Table 3). Our results may indicate that VUS in cancer patients may affect the potential accumulation of cancer risk, although there is no evidence of a causal relationship between VUS and cancer risk at this stage.
We could not examine microsatellite instability (MSI) as it relates to error repair in the germline; therefore, the association between MSI and the VUS of MMR genes is unknown. MMR gene-related cancers, including pancreatic cancer, are associated with MSI and MMR gene dysfunction, and the presence of MMR gene dysfunction or MSI-high is a known LS predictor and increases cancer risk (Laghi et al., 2012;Hause et al., 2016;Nojadeh et al., 2018). MSI status is also interesting because it predicts immunotherapy susceptibility such as immune checkpoint inhibitors (Baretti and Le, 2018;Latham et al., 2019;Ghidini et al., 2020). Previous studies that focused on sex differences in MSI in somatic MMR gene variants reported significant sex differences in esophageal and gastric cancers as well as an association among the differences in the incidence, prognosis, and treatment response of cancer (Li et al., 2018;Kohlruss et al., 2021). Further research on germline variants including MSI may facilitate the development of personalized therapies that focus on sex-based differences.
Information retrieved from the ClinVar database indicates no consensus on the involvement of VUS in cancer predisposition. The number of individual variant carriers of this study was limited and the small sample size was insufficient to establish the potential role of VUS in cancer. Nevertheless, the potential pathogenic risk of VUS is important for variant carriers. The clinical management of genetic diagnosis with risk prediction using VUS has several limitations. The current scrutiny of VUS pathogenicity has not kept pace with the number of VUS that have been detected. Indeed, many VUS are subsequently downgraded to benign variants whereas some are classified as pathogenic (Solomon et al., 2017;Macklin et al., 2018). In addition, a study also reported that the coexistence of VUS in MMR genes with predicted nonpathogenicity with other VUS of MMR genes might subtly increase cancer risk (Kantelinen et al., 2012). Hence, identification of VUS, which is very common in patients with cancer, and the collection of information on at-risk VUS by functional and epidemiological analysis is warranted. Findings of previous studies, as well as those of the current study, have suggested the importance of cancer risk prediction in individuals carrying rare germline VUS that lack the P/LP variants.
This study had several limitations. First, the rare germline variants identified in a population with a negative family history are likely de novo variants. However, no absolute evidence was found because no genetic information was available from the parents, families or relatives. Second, the effects of rare germline variants may be influenced by disease combination, smoking and drinking status, ethnicity, geographical region and database classification interpretation. Thus, the extent of variant reflection in the development of cancer remains unclear. Third, the presence of variants that were not included in the cohort and the uncertainty of the statistical power may stem from the small population size. Our results reflect only partial information on the individual patients in the study cohort. Prospective case-control studies including all age groups in various large populations of different ancestries are warranted to validate the current study findings.
In conclusion, our results provide important evidence for P/ LP variants in 6% of elderly patients with pancreatic cancer with a negative family history. As for VUS, MMR genes (MLH1, MSH2, MSH6 and PMS2) and POLQ may help predict potential trends in pancreatic cancer risk. MSH6 and POLQ may have slightly different potential cancer-associated trends in women and men, respectively. Given the current limitations of pancreatic cancer risk prediction in the general population, algorithmic considerations for the VUS distribution of cancer-predisposing genes will be important in estimating trends in individual genetic risks in the future. Further studies are required to confirm these findings.