2.1 Patient clinical and genetic characteristics

A total of 30 patients with high-grade pNENs were included in our study after histologic review (25 specimens were taken from the pancreas and 11 samples from liver metastasis). According to the WHO 2017 classification for pNENs, the well-differentiated G3 pNENs were classified as pNETs G3 (n = 13), while the poorly differentiated were pNECs (n = 13). Four cases were diagnosed as “ambiguous” due to the difficulty of distinguishing between pNETs G3 and pNECs. The representative histopathological images along with corresponding Ki-67 immunohistochemistry results for these cases are illustrated in Figure S1. To better distinguish between high-grade (G3) pNETs and pNECs, 15 patients with pNETs G1 and 10 patients with pNETs G2 were also enrolled. As a cohort of 55 patients, the median age was 53 years (range 21−75 years). The proportion of male patients (n = 29, 53%) was similar to females. Approximately 73% of patients had symptoms, and 31% already had liver metastasis (ENETS Stage IV) at diagnosis (Tables 1 and S1).

The sequencing of 55 patients revealed 245 mutations in 94 different genes. The mutations per case ranged from 1 to 12, with a mean of 4.6. At least one DDR gene alteration was found in 29 patients. TP53 and RB1 were affected in 11 and eight patients, respectively. MEN1 and ATRX/DAXX were affected in 21 patients and 13 patients, respectively. All patients’ recurrent mutations (3% or higher) are shown in Figure 1A. Other mutations that only occur in single cases were included in Figure S2.

2.2 Chromothripsis in patients with pNENs

Chromothripsis has been reported to potentially lead to the rapid malignant progression of some tumors. In our study, we analyzed the occurrence of chromothripsis in pNENs. In total, 14 out of 55 patients had chromothripsis involving different chromosomes. We detected chromothripsis on chromosome 9 in eight cases, chromosome 8 in seven cases, chromosome 1 in six cases, chromosome 4/5/19 in five cases, chromosome 2/6/11 in four cases, chromosome 7/10 in three cases, chromosome 3/13/18/20/21 in two cases, and chromosome 12/14/15/16/17 in one case (Figure 1B). Chromothripsis-positive patients presented a higher TP53 alteration (p < 0.001) and most of them (10 out of 14) were in the pNECs group (p = 0.001). However, there was no significant difference in DDR alteration (p = 0.702) and liver metastasis (ENETS Stage IV) at pathological diagnosis (p = 0.098) (Table 2).

2.3 Chromothripsis defined a group of patients with poor overall survival

By June 2020, 33 patients had survived, 20 had died, and two had lost follow-up. In the entire cohort, patients with chromothripsis had a significantly shorter overall survival (OS) of 658 days (95% CI, 389−927) compared with 2056 days (95% CI, 1809−2303) for those without chromothripsis (log-rank p < 0.001). In the high-grade (G3) pNENs group, patients with chromothripsis also had a shorter OS of 567 days (95% CI, 340−794) compared with 1534 days (95% CI, 1089−1979) for those without chromothripsis (log-rank p = 0.001). The association between different classifications and OS is shown in Figure 2A,B.

In the entire cohort, sex, TP53 alteration, grade, ENETS Stage IV, and chromothripsis were significantly correlated with OS. The multivariable analysis showed ENETS Stage IV and chromothripsis are independent prognostic indicators for OS. In the high-grade pNENs group, only grade, ENETS Stage IV, and chromothripsis were significantly correlated with OS. The multivariable analysis showed that only chromothripsis remains an independent prognostic indicator for OS (Table 3).

Interestingly, two chromothripsis-negative patients, whose pathological diagnosis was pNECs, had a survival time of 1617 and 2389 days (from the day of the pathological diagnosis to June 2020), respectively. Conversely, one chromothripsis-positive patient whose pathological diagnosis was pNETs G3 died 280 days after diagnosis. Furthermore, in the ambiguous group (n = 4, pNETs G3 or pNECs cannot be clarified by pathologists), one chromothripsis-positive patient also had shorter OS than three chromothripsis-negative patients.

2.4 Chromothripsis-positive patients had a higher risk of distant metastasis

By June 2020, 10 patients had distant metastasis after follow-up. In the entire cohort, patients with chromothripsis had a significantly higher risk of distant metastasis (92%) compared with those without chromothripsis (46%) (p = 0.005). In the G3 pNENs group, patients with chromothripsis tended to have a higher risk of distant metastasis (100%) compared with those without chromothripsis (69%) although there was no significant difference (p = 0.098) due to the low number of patients in this subset. The association between different classifications and distant metastasis is shown in Figure 2C,D.

2.5 TP53 mutations in chromothripsis-positive patients preceded chromothripsis

TP53 mutation has been reported to potentially lead to chromothripsis in cells. Conversely, the occurrence of chromothripsis in cells can impact various key genes, thereby promoting the rapid malignant progression of tumors. To explore such a relationship in pNENs, an algorithm using sequencing data was developed to infer the timing of driver mutation, like TP53 mutation, relative to chromothripsis events. We calculated the tumor fraction in the same specimen by CNV and driver mutation, respectively. If the tumor fraction calculated by CNV is higher than by driver mutation, it means that the chromothripsis occurred earlier; otherwise, the driver mutation occurred earlier. The results showed that all TP53 mutations in chromothripsis-positive patients preceded chromothripsis. CTNNB1 mutation in patient 33 and SMAD4 mutation in patient 41 also occurred before chromothripsis (Figure 3).

4.1 Patients and pathologic re-evaluation

It was a retrospective study and approved by the Ethics Committee of West China Hospital, Sichuan University. High-grade pNENs were searched from patients who underwent biopsy or surgical resections of their tumors (from 2008 to 2018) and had a definite pathological diagnosis of pNENs with mitotic rate > 20/10 HPF or Ki-67 index > 20%. Cases were excluded if they contained a non-neuroendocrine component or the pancreas could not be diagnosed as the primary site. Some samples with a definite pathological diagnosis of low-grade pNETs were also enrolled as a reference to better distinguish high-grade (G3) pNETs from pNECs. For certainty, all samples included were histologically examined again by our pathologists, and each case was also reassessed for mitotic indices and Ki-67 labelling index. There were no changes in them relative to the original diagnoses. Then, we graded all pNENs we collected (from 2008 to 2018) according to the newest WHO 2017 classification. The diagnostic review was blinded to molecular genetic findings in our study. Formalin-fixed and paraffin-embedded tumor tissues and paired normal tissues were retrieved from the files of the Department of Pathology, West China Hospital, Sichuan University. Clinical information was obtained from related Electronic Case Report Forms.

4.2 DNA extraction and targeted next-generation sequencing

Genomic DNA was extracted from slides of tumor samples and paired normal tissues with MagPure Tissue DNA DF Kit (Magen Inc.; MD5112-TL-06) according to the manufacturer's protocol. Extracted DNA was quality-controlled by Qubit dsDNA HS assay (Thermo Fisher Scientific) and Agilent 2100 Fragment Analyzer. We processed the DNA according to a modified single-stranded DNA sequencing library preparation protocol. Briefly, genomic DNA was sonicated with 3′-poly-A tailed with terminal transferase, ligated with a poly-dT-tailed P5 sequencing adaptor, and subjected to several cycles of linear amplification using a P5 sequencing primer. A P7 adaptor with a random nucleotide 3′ overhang was then ligated to the 3′ end of the linear amplification product. From there, the sequencing library was amplified using P7+P5 primers to a sufficient amount. Hybrid capture was performed using standard DNA-probe-capture practice. The post-capture libraries were sequenced on Illumina Novaseq.

For panel sequencing, a customized target capture panel targeting 341 genes (Euler Technology), which covers common oncogenes, tumor suppressor genes, as well as genes of interest for clinical oncology, was used. The panel also included ∼1076 common SNP (common in East Asian population, with MAF in [0.25, 0.8]), which evenly span the genome for copy number variation analysis. The capture assay was validated using external and internal standards to meet a limit-of-detection of 0.5% for short nucleotide variation (SNV) at a given standard sequencing depth (∼2000× deduplicated). Genes were identified as DDR-related in our study based on Pubmed searches^{45, 46} and authoritative databases (such as GenCards Database). Specifically, they included MLH1, MSH2, MSH6, MRE11A, PMS1/2, BRCA1/2, PARP1/2, RAD50, RAD54L, BRIP1, BARD1, FANCA, FANCC, FANCL, ERCC3, ATR, ATM, CHEK1/2, and POLE in this study.

4.3 Chromothripsis detection

In our study, chromothripsis was analyzed by pattern recognition aided by machine learning, with a bioinformatic pipeline adapted from ShatterSeek (https://github.com/parklab/ShatterSeek) that modified for capture panel sequencing data. The chromothripsis-positive threshold was set as “three or more consecutive genomic regions with typical copy number sequence (monotonic and linear change in log copy number) as generated by bridge-fusion-bridge-class, or four or more consecutive genomic regions with typical copy number sequence (repetitive oscillating change in log copy number) as generated by chromosome scattering.” The absolute integer copy number and the exact physical position of each genomic region were computed by linear regression of GC-adjusted, read-depth-normalized, circular segmented output from CNVkit.

4.4 Timing of TP53 mutation and chromothripsis

To compute the tumor fraction with chromothripsis, we first select choose genomic regions containing the copy number variation related to chromothripsis. Then, we calculate exon-wise copy number log-ratio of these genomic regions using CNVkit. For each “integer” of $$\vec{x}$$, the observed copy number log-ratios resulted from a probability function, which could be modeled as a gaussian process. Hence, a gaussian mixture model was applied to all exon-wise copy number log-ratios to decompose them into several possible gaussian distributions (3), each with a mean log-ratio change corresponding to an “integer” in $$\vec{x}$$. Linear regressions were performed with all possible integer combinations of $$\vec{x}$$ (for example, for a 5-mode distribution of copy number log-ratio, the vector of $$\vec{x}$$ would choose from five integer elements, such as [−1,0,1,2,3], [−2,−1,1,2,4], [1,2,3,4,5], etc.) to $$\vec{y}$$ according to the abovementioned Equation (2). To make the computation possible, we match an integer to all CNV region belonging to a single mode of gaussian distribution in (3).

A best fit solution was taken at the adjusted p < 0.01 and R² > 0.95 for the linear regression solution. By extracting the slope coefficient from this best-fit linear regression, we obtain the tumor fraction of chromothripsis-containing tumor cells.

The abovementioned Equations (1) and (2) were validated with pure cancer cell line sequencing data as well as artificial mixtures of cancer DNA into normal PBMC genomic DNA to show that in these cases, both SNV-inferred and CNV-inferred tumor fraction were accurate.

4.5 Statistical analysis

The OS was assessed as the period from the pathological diagnosis to death or last follow-up. It was estimated using the Kaplan–Meier method and compared with the log-rank test. Uni- and multivariable Cox regression analyses were used to evaluate the impact of selected clinicopathologic factors. The HRs and 95% CIs were calculated. Differences in characteristics by groups were tested by Chi-squared or Fisher exact test. Statistical analysis was conducted using SPSS 25.0 software (IBM Corporation, Armonk, New York). A two-tailed p value < 0.05 was considered statistically significant.