Tumor copy number instability is a significant predictor for late recurrence after radical surgery of pancreatic ductal adenocarcinoma

Abstract Objective Our study examined the association between molecular features and clinical results of pancreatic ductal adenocarcinoma (PDAC) patients, aiming to explore the genomic determinants of the recurrence and prognosis of PDAC after surgical removal. Methods This retrospective study analyzed 181 PDAC patients who received pancreatectomy and adjuvant chemotherapy, with 67 patients in the training set. An internal validation set of 48 patients and an external validation set of 66 patients were used to validate the result. Comprehensive genomic profiling was performed on formalin‐fixed paraffin‐embedded (FFPE) tumor specimens to determine genomic features using the designed cancer‐related gene panel based on next‐generation sequencing (NGS). Results Significant differences were identified between the late recurrence (LR) group and early recurrence (ER) group in tumor copy number instability (CNI) levels. Next, the utility of low CNI (the middle and lowest tertile) with regard to predicting LR was confirmed in the training, internal, and external validation sets. Further univariate and multivariate analyses revealed that CNI was an independent predictive and prognostic biomarker, and had higher predictive accuracy for LR than CA19‐9 level, pathological stage, tumor size, and age. In addition, CNI combined with lymph node (LN) metastasis status could provide a more accurate model for predicting LR of PDAC. Conclusion We discovered and validated the association between CNI and clinical outcome in 181 patients with resectable PDAC, demonstrating the utility of lower tumor CNI levels as biomarkers of postoperative LR and favorable prognosis. Moreover, the combination of CNI and LN metastasis status elevated the predictive accuracy and illuminated strategies for patient stratification.


| INTRODUCTION
Pancreatic cancer is one of the most aggressive human malignancies; it carries a poor prognosis, with a 5-year survival rate of 8%, and it ranks fourth among all causes of cancer-related death. 1,2 In China, pancreatic cancer is now the ninth leading cause of cancer-related mortality, with a sharply increasing incidence rate. 3 Pancreatic ductal adenocarcinoma (PDAC) is the most common type and accounts for more than 90% of pancreatic cancer. 4 Despite considerable improvements in multiline therapeutic management, surgical resection still remains the only potentially curative treatment for patients with PDAC. 5 Furthermore, adjuvant chemotherapy with gemcitabine or S-1 and other conservative therapies are necessary to palliate symptoms and improve survival in patients with resectable PDAC. [6][7][8] High incidences of local recurrence and distant metastasis are two of the main reasons for the poor outcome of PDAC. 9 To date, several postoperative features have been reported as predictive and prognostic markers in PDAC. Postoperative features, such as lymph node (LN) metastasis, tumor size (>3 cm), tumor grade, resection margin, and differentiation, have been extensively developed as predictors of survival and recurrence. 6,10-13 Serum carbohydrate antigen 19-9 (CA19-9) is a well-known preoperative and postoperative predictor of clinical outcomes in PDAC. [14][15][16][17] However, the evaluation of patient prognosis mainly relies on pathological features and serum tumor markers. In addition to the four major driver genes (KRAS, CDKN2A, TP53, and SMAD4), few established genomic features have been reported as postoperative predictive markers of outcomes in PDAC. 18 Therefore, it is of great interest to further investigate the molecular features to identify more potential biomarkers to predict postoperative recurrence and survival in patients with resectable PDAC.
Next-generation sequencing (NGS)-based gene panel testing is a clinically useful tool for individualized cancer patient care that investigates genomic features related to targetable gene alterations, tumor mutational burden (TMB), and copy number instability (CNI). To explore the genomic determinants of the recurrence of PDAC after surgical removal, we retrospectively investigated the clinical characteristic and molecular features of 181 PDAC patients, with 67 patients in the training set, 48 patients in the internal, and 66 in the external validation set. The predictive and prognostic abilities of these genomic signatures, as well as clinical and laboratory factors, were extensively investigated. Our investigation helps identify PDAC patients who may not achieve a sustained therapeutic benefit after tumor resection, and for whom further treatment options should be considered.

| Study design and patients
Formalin-fixed paraffin-embedded (FFPE) tumor specimens and matched blood samples were collected from 220 patients who were diagnosed with PDAC. All the patients had undergone pancreatectomy and received first-line adjuvant chemotherapy for at least 6 months.

K E Y W O R D S
copy number instability, late recurrence, pancreatic ductal adenocarcinoma, predictive marker, prognosis levels or recurrence status, receiving neoadjuvant chemotherapy, or dying from other causes. All patients signed informed consent forms giving their consent to participate in the study prior to their enrollment. All experiments were carried out in accordance with relevant guidelines and regulations.

| Follow-up and definition of recurrence
For the first 3 years after pancreatectomy, the follow-up interval was 3 months and comprised a physical examination, laboratory tests, measurement of tumor markers, and computed tomography (CT). Confirmed recurrence was defined as recurrence suspected on abdominal CT, which was confirmed as recurrence on subsequent follow-up. Progression-free survival (PFS) was calculated from the date of pancreatectomy to the date of recurrence or last follow-up if recurrence did not occur. Patients who relapsed within 6 months after pancreatectomy were defined as experiencing early recurrence (ER), and patients who relapsed after 6 months or without recurrence were defined as experiencing late recurrence (LR). 14,19

| Identification of somatic mutations
FFPE tumors and matched blood samples were submitted for NGS. The NimbleDesign assay was used to identify mutations. We used VarScan2 with the following filters: (a) located in intergenic regions or intronic regions; (b) synonymous single-nucleotide variants (SNVs); (c) allele frequency ≥0.005 in the Exome Aggregation Consortium (ExAC) database; (d) allele frequency <0.01 in the tumor sample; and (e) support reads <5. To identify somatic SNVs and indel mutations, matched blood samples from patients were used as controls for the FFPE tumor samples with mutations.

| Calculation of TMB
To calculate the TMB value of the FFPE tumor samples, the number of somatic nonsynonymous SNVs detected by NGS was quantified by excluding (a) alterations with allele frequency ≥0.002 in the ExAC database; (b) those recorded in the COSMIC database; and (c) those with variant allele frequencies (VAFs) between 0.01 and 1. The TMB was measured in mutations per Mb.

| CNI evaluation
We used the "-normal" parameter to construct the copy number baseline as a negative control, and used a copy number variant (CNV) kit to call copy number variations from the FFPE tumor samples for each patient. For the determination of the FFPE tumor sample CNI, the copy numbers were called after mapping using the Burrows-Wheeler Alignment (BWA) tool. After correction for GC content and mappability using proprietary algorithms for tumor DNA sequencing, the read counts were transformed into log2 ratios and converted into Z-values based on Gaussian transformations vs a normal control group. The target regions that satisfied the Z-score greater than the 95th percentile plus twice the absolute standard deviation of the normal control group were retained, and these Z-scores were summed as the CNI.

| Statistical analysis
The Fisher exact test was applied to explore the relationship between CNI and clinical and molecular characteristics, as well as the distribution of gene mutations in the ER and LR groups. The Mann-Whitney test was used to assess the differences in CNI and TMB levels, which were regarded as continuous variables, between the ER and LR groups. Recurrence curves were obtained by Kaplan-Meier analysis with the log-rank test. Logistic regression analysis was performed to assess the factors predictive of LR, and those with P < .10 were selected for inclusion in the subsequent multivariate analysis. Receiver operating characteristic (ROC) curves were constructed to assess the ability of CNI to predict LR with the area under the curve (AUC) representing the predictive performance. Statistical analyses were carried out using R (https://www.r-proje ct.org/, version 3.6.2), and P < .05 was considered statistically significant.

| Clinical characteristics of studied PDAC patients
Of the 220 patients, 39 were excluded after clinical data quality control. Finally, 181 PDAC patients were included in this study. Of the 115 enrolled patients from Ruijin hospital, 67 were in the training set and 48 were in the internal validation set. The other 66 patients from the First Affiliated Hospital and Fujian provincial hospital were regarded as an external validation set ( Figure S1). As shown in Table 1

| Differential molecular features between ER and LR groups
To investigate the predictive markers for PDAC, molecular features were assessed in the current study. As demonstrated in Figure 1 and Table 2, in the ER and LR groups, of the four major driver genes, KRAS was mutated most frequently, with rates of 85.7% (18/21) and 87.0% (40/46), respectively. Moreover, TP53 had mutation rates of 71.4% (15/21) and 60.9% (28/46), CDKN2A had mutation rates of 9.5% (2/21) and 17.4% (8/46), and SMAD4 had mutation rates of 9.5% (2/21) and 19.6% (9/46) in the ER and LR groups, respectively. Unfortunately, there were no significant differences in the distributions of these gene mutations between the ER and LR groups (Fisher's exact test, P > .05).
In addition to gene alterations, the CNI and TMB were further calculated in our cohort. The results showed that the median CNI in the training set was 5490.03 (IQR from 4587.06 to 7629.59), and the median TMB value was 5.99 mutations/Mb (IQR from 4.00 to 7.97). As shown in Table 2, the CNI was significantly higher in the ER group than in the LR group (Mann-Whitney test, P = .002). However, the TMB value was comparable between the ER and LR groups (Mann-Whitney test, P = .735).

| The predictive performance of CNI in PDAC
Based on the significant difference in CNI between the ER and LR groups in the training set (Mann-Whitney test, P = .002, Table 2 and Figure 2A), we next identified that CNI was also significantly lower in the LR group than in the ER group in both the internal and external validation sets (Mann-Whitney test; internal validation set, P = .006; external validation set, P = .004; Figure 2B,C). Subsequently, the ROC analyses indicated a significant value of CNI for predicting LR in all three cohorts, with an AUC of 0.716 (training set, P = .005), 0.729 (internal validation set, P = .012), and 0.711 (external validation set, P = .009), respectively ( Figure 2D-F).
To further investigate the ability of CNI to predict LR, we defined the CNI-high group as that with a CNI of highest tertile and the CNI-low group as that with a CNI of middle and lowest tertile. The LR rate was compared between the CNIhigh and CNI-low groups. Surprisingly, the CNI-low group had a very high rate of LR (Fisher's exact test, training set, 82.2%, P = .002, Figure 3A; internal validation set, 81.2%, P = .020, Figure 3B; external validation set, 81.8%, P = .016, Figure 3C). The above data indicated that a low CNI could be a predictive signature for LR in PDAC.
Clinical and laboratory features, such as LN metastasis, preoperative CA19-9, pathological stage, tumor size, and age, have been reported to be associated with recurrence in PDAC. 6,17,20,21 Herein, univariate and multivariable logistic regression analyses were performed to assess whether the following characteristics were risk factors for LR: age, gender, pathological stage, T status, LN metastasis status, tumor size, preoperative CA19-9, and preoperative carcinoembryonic antigen (CEA). Univariate analysis demonstrated T status, LN metastasis status, and CNI as considerable risk factors for LR (P < .050, Table 3). Further, multivariable analysis showed that both CNI and LN metastasis status were independent predictive factors for recurrence in PDAC patients (CNI: HR = 7.918, 95% CI, 2.128-29.458, P = .002; LN metastasis status, HR = 4.230, 95% CI, 1.153-15.518, P = .030), and that CNI might be a better predictive marker than LN metastasis status (Table 3).

| The prognostic value of CNI in PDAC
Given the predictive performance of CNI in PDAC, we further investigated the prognostic value of CNI in PDAC. K-M plotter was used to compare PFS in the CNI-high and CNI-low groups (CNI-high: highest tertile, CNI-low: middle and lowest tertile). Of note, the CNI-low group had a significantly longer PFS than the CNI-high group in both the three cohorts (log-rank test, training set, median PFS, 11.0 vs 5.5 months, HR = 0.368, P = .002, Figure 3D; internal validation set, median PFS, 10.5 vs 6.0 months, HR = 0.444; P = .021, Figure 3E; external validation set, median PFS, 12.0 vs 5.5 months, HR = 0.367, P < .001, Figure 3F). These results indicated that CNI was inversely related to prognosis in pancreatic cancers.

| Joint utility of CNI with LN metastasis status
Given the fact that CNI and LN metastasis status were independent predictive factors for LR in our study population (Table 3), we further explored the possibility of combining CNI with LN metastasis status for the prediction of LR in PDAC patients. Intriguingly, we observed that CNI-low   Figure 4A) and 87.5% (24/26) in the external cohort (the external validation set, Figure 4C), respectively. Moreover, patients within the CNI-low/LNnegative subgroup had a significantly longer PFS than the other three subgroups (log-rank test for trend, internal cohort, P = .003, Figure 4B; external cohort, P < .001, Figure 4D). In all, the above data suggested that there is value in combining CNI and LN metastasis status when evaluating the prognosis of PDAC patients.

| DISCUSSION
Despite considerable achievements in diagnosis and therapeutic strategies, the overall 5-year survival rate of PDAC has remained dismal in the past few years. [22][23][24] Therefore, more reproducible and utilizable biomarkers are needed to provide appropriate adjuvant treatment protocols for patients with resected PDAC. In the present study, we retrospectively evaluated the baseline genomic and clinical features of a PDAC cohort of patients who underwent pancreatectomy and adjuvant chemotherapy. Here, we show that the CNI level may have superior predictive value compared with other signatures, including CA19-9, T status, LN metastasis status, SNV, and TMB. Previously, four major driver gene alterations were found to be associated with poor prognosis in PDAC patients. 18,[25][26][27][28][29] However, no significant differences in the distribution patterns of these genes were identified in our cohort, which may possibly be the result of the imbalanced proportions of patients between the mutation and wild-type groups. A larger sample size may further support the significance of the driver gene signature. Moreover, previous studies have demonstrated the prognostic value of LN metastasis status. 21 As expected, we herein identified that LN metastasis status was significantly associated with LR and was also independent predictive factor for recurrence in PDAC patients (Table 3). Furthermore, clinical and laboratory features (including CA19-9 level, tumor size, and pathological stage), which are well-known prognostic factors, 6,11,19 were analyzed in our cohort as well. Although we observed higher odds ratios (ORs) with regard to LR for CA19-9 level, pathological stage, and tumor size, there was no statistical significance.
Importantly, for the first time, we present evidence showing that CNI is an independent predictive biomarker for LR in PDAC patients, surpassing the value of LN metastasis status and pathological stage (Table 3). Similarly, plasma CNI has been demonstrated to be a predictor of the response to systemic PDAC therapy and immunotherapy across different types of advanced cancers. 30,31 More specifically, the CNI might be a prognostic biomarker for PDAC. Herein, we found that a low CNI was significantly correlated with a longer PFS ( Figure 3D-F), which was in accordance with previous data showing that a high CNI was associated with worse clinical outcomes in head and neck squamous cell carcinoma (HNSCC), 32 gastric cancer, 33 and PDAC, 34 respectively. Taken together, lower CNI levels might be a strong signature for the identification of PDAC patients who might benefit from adjuvant therapy after pancreatectomy.
Based on the above findings, we next evaluated the joint utility of CNI and LN metastasis status ( Figure 4). Intriguingly, the CNI-low/LN-negative subgroup had a particularly higher LR rate than the other three subgroups, indicating that the combination of CNI and LN metastasis status may help identify patients prone to LR after surgery. Genomic analysis is easy to implement with the rapid development of NGS. We have identified a threshold to define CNI-high and CNI-low subgroups, and the relationship between CNI Abbreviations: CA19-9, preoperative carbohydrate antigen 19-9; CEA, preoperative carcinoembryonic antigen; CI, confidence interval; CNI, copy number instability; LN, lymph node; OR, Odds ratio; TMB, tumor mutation burden.
and clinical outcome was verified in different NGS panels. Therefore, the joint utility of CNI and LN metastasis status is worth for further investigation and validation in larger size cohorts in the near future.

| CONCLUSION
Taken together, we present evidence demonstrating the association between tumor CNI and clinical outcomes in patients with resected PDAC. Our data indicated that tumor CNI is a potential predictive and prognostic biomarker in patients with resectable PDAC. Notably, the combination of CNI and LN metastasis status may enhance predictive accuracy and improve patient stratification. Further prospective studies are warranted to develop a predictive algorithm to define the optimal threshold for CNI in PDAC and even across different cancer types.