Whole-exome Sequencing Analysis for Mutations Characteristics in Recurrent or Metastatic Gastrointestinal Stromal Tumors

Background: Surgical resection for the metastasis and recurrence of GIST was controversial. It is increasingly important to identify clinical factors related with survival and explore the driver genes and mutations in GIST. Methods: GIST patients who received two surgery for primary and recurrent and/or metastatic tumors between January 2003 and December 2018 were reviewed. Primary outcome was overall survival after reoperation. Kaplan-Meier , Cox proportional hazard regressions, and mean survival time were used to evaluate outcomes. Paired PT(primary tumor), RMT(recurrent and/or metastatic tumor) and normal DNA was whole- exome sequenced to generate comparable data for those specic 8 GIST cases. Results: We identied 39 eligible patients with a median overall survival time of 56.7 months(IQR:9.6-190.3months). Regular TKI(Imatinib) after primary tumor resection (HR:0.568; 95% condence interval(CI):0.211-0.874; P=0.043) was associated with better OS, while presence of liver metastasis were prognostic for worse OS(HR:1.45; 95% CI:1.13-2.02;P=0.032) for those GIST patients who received re-surgery due to recurrent and/or metastatic tumor. Compared with normal tissue, we detected mutation on MUC family both in 8 PT and 7 RMT among the 8 patients. Only in irregular(TKI) group, the KIT mutations between PT and RMT contain correlations and differences, while its inuence exist less on other cases. We also found that 31 genes which were direct correlation with coding regions may associated with RMT. We attained that the Spatial heterogeneity and temporal heterogeneity of the tumor reected on mutation signature and subclone. Conclusions: The MUC mutations were supposed to be a potential predict to recurrent and/or metastatic GIST. The treatment of TKI could inuence the KIT mutations on RMT of GIST. As the heterogeneity exist in PT and RMT, the direction of tumor evolution and progression were not stable and regular. analyzed. Functional annotation and deleterious alterations of germline variants were performed using ANNOVAR 17 . Additional annotation of somatic SNVs to 30 COSMIC (Catalogue Of Somatic Mutations In Cancer) mutational signatures 18 was performed using Somatic Signatures in R package 19 . To the heterogeneity of tumor, we took three samples in different site primary and tumor in case 6, then analyzed the mutations and CNV of each sample. For Somatic SNVs, analyses were conducted on mutation Based on Bayesian clustering method, we chose could take the complexity of or into clone rate cell for Allelic imbalance caused by CNV and normal cell contamination ultimately.


Background
Gastrointestinal stromal tumors GIST are the most common sarcoma subtype and originate from the intestinal cells of cajal (ICCS), most commonly in the stomach and small bowel which are characterized by activating mutations of the KIT or PDGFRA receptor tyrosine kinases in nearly 85-90% of cases [1][2][3] . At present, the mainstay of treatment is surgery for patients with clearly resectable primary GIST 4 . When the FDA approved imatinib(IM) ,a receptor tyrosine kinase inhibitor of KIT, for the management of patients with advanced GIST in 2002, the treatment of GIST entered the era of imatinib and the prognosis of the GIST patients were signi cantly increasing, as GISTs generally resistant to conventional chemoradiotherapy [5][6][7] . And current recommended treatment for GIST patients with a high risk for recurrence is adjuvant imatinib for 3 years at least 8 . Patients with KIT exon 11 deletion mutations were veri ed to bene t most from imatinib 6,9 . But in con rmed cases of KIT exon 9 mutation, raising the dose of imatinib could signi cantly improve the PFS 10 .
However, a number of GIST cases still suffered from recurrence and/or metastasis, leading to patients' death despite comprehensive treatment. It has been reported that surgical resection of residual lesions after disease control with imatinib was bene cial to patients with recurrent or metastatic cases 11,12 . Previous studies have showed that KIT exon 11 mutations in GISTs were signi cantly higher response to imatinib than other mutations in KIT, while PDGFR exon 18 D842V mutantion in GIST was resistant to IM 13 . Some studies revealed that the larger tumor volumes were correlated with adverse outcomes, the chance of molecular evolution and secondary clonal resistance [14][15][16] . As the heterogeneity and IM resistance of GISTs, it is di cult to predict the prognosis of the recurrence and metastasis cases. Therefore, the aims of this study were to evaluate the prognosis factors correlated with surgery for the metastasis and recurrence of GIST patients and explore the gene mutations and tumor evolution between primary tumors and recurrence and/or metastasis tumors by Whole genome sequence (WGS).

Patients and tumor samples
This study was divided into two parts. In the rst part, we retrospectively analyzed 39 patients who had recurrence or metastasis after the rst surgery and underwent one or multiple surgery at Ruijin Hospital, Shanghai Jiao Tong University school of Medicine, from January 2003 to December 2018. Diagnosis was con rmed using standard histology as well as immunohistochemistry for CD117 (KIT) and sometimes DOG-1.
In the second part, we selected 8 clinical cases of the 39 patients on the basis of different groups. The reasons for the 8 patients as candidates: 1) based on prognostic factors attained through rst part; 2) the time of two surgeries were all after 2010 on account of the degradation of DNA and RNA. The samples contain primary tumor, recurrent or metastasis tumor and normal appearing tissue. All tissues were manually microdissected from unstained, 10-μm-thick formalin-xed, para n-embedded (FFPE) sections. The integral process was showed in Figure.1.A. The study was reviewed and approved by Shanghai Jiaotong University School of Medicine Ruijin Hospital Ethics Committee.

Data Collection and Follow-up
Demographic and clinicopathological data were reviewed retrospectively, including as following age, sex, characteristics of the primary tumor (PT) and recurrent and/or metastasis tumor (RMT)(size, site or recurrence/metastasis site , mitotic rate ), tumor growth pattern(Invasive, Dilative), histologic variant (spindle, epithelioid, or mixed), margins from surgery resection(R0,R1/2), and signi cant dates(date of diagnosis, surgery time, recurrence interval time, death, last follow-up). Fifty high-powered elds (HPF) were counted by the experienced pathologist to determine mitotic rate. Recurrence interval time (RDT) represents the period from the rst surgery to recurrence. Overall survival was de ned as the time between recurrent surgery and death of any cause.

Statistical Analysis
For statistical analysis, qualitative data are presented as number (%). Continuous variables are expressed as mean with standard deviation (SD). Actuarial recurrence free survival was calculated using the Kaplan-Meier method. Factors associated with recurrence were tested by univariate log-rank analysis.
Variables that were signi cant in univariate analysis were entered into multivariate analysis. Multivariate analysis was performed with the Cox proportional hazard regression model. Hazard ratio (HR) for comparison of the 2 groups was summarized with its 95% con dence interval (CI) and P-value using logistic regression. Statistical analysis and calculations were done by using IBM SPSS statistics 24 performing independent samples t-test and Kaplan-Meier survival analysis. A p-value less than 0.05 was considered statistically signi cant.

Genomic analysis, Bioinformatic Analysis and annotation
We adopted the Agilent_60M exon targeting sequence enrichment system to capture human exon sequences in our study. Firstly, we tested the purity and concentration of captured hybridization DNA library, and illumina HiSeq Platform sequencing (GeneX Health Co. Ltd, Beijing, China) was performed. Taking the accuracy of sequencing data into account, we sequenced each sample two times, then cleared the repetition by merging two sequencing data, and constructed the sequencing results of the case at that time. The average sequencing depth of the target regions was >100×.
Take the Raw data into bioinformatics analysis process 1 Sequencing quality assessment : collect the sequencing error rate, data volume, comparison rate; evaluate whether the library sequencing meets the standard, then perform the following analysis ; 2) Information mining and analysis: detect Somatic SNV(single number variation) CNV(copy number variation); annotate and analyze each database; explore and compare gene mutation spectrum and mutation characteristics, CNV distribution, Tumor evolution, Tumor heterogeneity between PT and RMT.
In our study, small variants include single nucleotide variants (SNVs) and insertions or deletions. We took Control-FREEC, CNVkit, Contra to detect Somatic

Clinicopathologic features and outcomes
For the primary surgery , the median age of the patients at the time of diagnosis was 56.2 years (range:35.1-75.3). As the tumor location, the small intestine presented the most common primary site n=28,71.8% followed by the stomach n=4,10.3% and colorectum (n=4,10.3%). The median tumor size was 9.18 cm (range:3.2-17.9). Mitotic rate of 12 patients with primary tumor was less than 5/50, those of 5 patients more than 25/50.The median RDT was 41.3 months (IQR:33.5-55.7). 37 patients underwent radical surgery, and the other 2 patients only underwent R1 surgery.
After the primary surgery, all patients had recurrence, metastasis or disease progression in 41.3 months (range :12.3-117.5) On account of recurrent site, peritoneum was involved in 28 cases (17 patients with local recurrence, the others with new lesions ) liver in 9 cases (4 patients only with liver metastasis, the other 5 patients with two or more sites), rare sites including lymph nodes 1 patient , cutaneous of chest wall and inguinal(1 patient). All the cases received surgery within 6-12 months 34 patients accepted the radical surgery, 2 patients' pathology revealed the margins was positive, the other 3 patients only received palliative resection (Table 1).
In our study, 23 cases died from the tumor and 6 cases lived with neoplasm until the last follow-up(2020-12-31) excepted for 2 cases who were lost to followup. The median overall survival of the entire GIST cohort was 56.7 months (range:9.6-190.3). On survival analysis, liver metastasis and without IM treatment predicted poor prognosis (Figure1-B,C). On univariate analysis, recurrent site and IM treatment were associated with overall survival. Taking the two into multivariate analysis, they were proved to be independent factors associated with prognosis. (Table 2) 4. Whole-genome Analysis

KIT mutations
Of the 8 clinical cases, 5 harbored mutations in KIT, which containing two with both primary and recurrent tumor mutations, the other three only had primary tumor mutation; 1 harbored mutations in PDGFRA among the RMT, the two remaining patients were wild type. We discovered that the KIT mutations between primary tumor and recurrent tumor not only had correlation that all had same mutations in exon 11 clustered in the proximal part between codons 568 and 576 and consisted of small in-frame deletions and point mutations ; but also had difference that as Table 3 shows. In the remaining three patients, two detected mutations in exon11 and one harboured in exon 17, these mutations were only detected in PT. In the mutated exons of the KIT gene, the types mainly included point mutations, base deletions and amino acid substitution.

CNV variations
The 8 patients were grouped according to IM treatment after primary surgery (Table 4).
All samples were obtained deletions in chromosome1p, 14, 15; while deletions in the chromosome13q were frequent especially in metastatic and/or recurrent tumors Fig. 2A,B . In PT samples Fig. 2A , the deletions in chromosome10 and 22 were detected in 7 cases, the left one was in No drug group. While in RMT samples (Fig. 2B), the ampli cation in chromosome5, especially in 5p, was generally detected (6/8) and the two in No drug group were also not detected variation mentioned above.
We could not obtain any deletion in chromosome10 of correspondent cases in RMT compared with PT. However, new CNVs were detected in chromosome 22 in 2 cases, and the deletions of chromosome22 attained in PT did not exist in RMT upon the irregular group. Only in the liver metastasis group, we acquired deletions in chromosome18 both in PT and RMT, and its feature was consistent in two tumors. which only occurred in patients receiving IM treatment.

Somatic SNV results
Firstly, High frequency mutations (top 20) in different code genes were ltrated among all the samples, while one sample was excluded because it did not contain those genes Fig. 2C1 . Surprisingly, the mutations on the MUC family were obtained both in each sample of PT and RMT, such as MUC4, MUC6, MUC16, MUC17 et al, except one sample of RMT . We also found the frameshift mutation on FRG1 and ZNF717 had common tendency. The remained mutation tendency between PT and RMT existed correlations and distinctions (Fig. 2C2, C3).
Two recurrent tumors of the 8 patients were detected mutation in AJAP1 resulting in the loss of two amino acids of its encoding product. In two samples of RMT(25%), we observed the mutation in PRL19 encoding an essential structural constituent of ribosome, Additionally, mutations of coding region of the LOC101927550 (ncRNA) were harbored in three cases. which in uences cell proliferation and growth velocity by affecting the synthesis of the protein.

Mutations among clinical group
We detected mutations occurred on UTR region of the MTMR11 and GOLGA6L4 only in regular treatment group Fig. 3A1, B1 . Interestingly, we acquired frameshift mutations in the EMX2,C2CD4D gene besides above-mentioned in liver metastasis group with regular treatment Fig. 3A2 (Fig. 4A). In the type of GIST, the mutation of T>G/A>C accounted for the high proportion of the 6 types, and the proportion of each mutation was signi cantly different in three spatial locations, while its tendency and feature in primary and recurrent tumor was similar.
To infer the mutation pattern of recurrent and/ or metastasis tumor, bases at upper and downstream positions of 1bp of point mutation were taken into consideration and we classi ed it into 96 varieties according to Mutational Signatures (Figure4-B)., Mutational characteristics were obtained through those varieties. As it (Figure4-C) shows, the difference of pattern between the primary and recurrent tumor was mainly re ected on 8 signatures Signature1 Signature2 Signature3 Signature5 Signature11 Signature12 Signature20 Signature30 .

Somatic CNV
The loss of copy number at Chr1, Chr2, Chr13, Chr15 was detected both primary and recurrent tumors. Besides, we also obtained increase at Chr5 and decrease at Chr11, Chr14 in RMT, while decrease was detected in Chr4, Chr10, Chr12, Chr22, ChrX only in PT. As it (Fig. 4D) shows, the CNVs were not only existed different expects, but also had correlations at different points. 4.5.3 The subclones, evolution and heterogeneity of tumor cells Compared with primary tumor, the number of subclone and the cluster of low prevalence were decreased signi cantly in recurrent tumor. But tumor in different sites in the same sample, the number and the cluster were almost conformable both in two tumors (Fig. 4E-1). Take the all samples of PT and RMT into subclone analysis (Figure4-E3), 3 subclones cluster were divided. The prevalence of three clusters in recurrent tumor was quietly higher than it in primary tumor, while the difference in the same tumor was not signi cant both in the two tumors respectively (Figure4-E2). To know the evolution of different tumor, we attained that the evolutionary relationship in PT was P3>P2>P>1 and in RMT was R3>R2>R1 in Clonal Phylogeny (Fig. 4F1). Compared with PT, the samples of RMT were in the end of evolutionary tree (Fig. 4F2,). And the correlation between P2 and P1 was low in PT, while it existed close correlation between R2 and R1 in RMT. The relatedness was quietly low among P3, R3 and other correspondent samples. There existed the same and different mutation sites in different locations of the same tumor through comparing the functional mutations of each sample, and it was con rmed that the same results were acquired both PT and RMT (Fig. 4G). It was attained that the mutation frequency of some genes was presented to be signi cantly different between the each sample of the two tumors, especially in MUC19, KIT, PABPC3, ZNF208, TOP3A, RGS19. Those genes were taken into functional enrichment, and we found that the genes involving in biology process and its biological functions were closely connected with tumor cell growth regulation extensively, such as in uencing cell differentiation, regulating the expression of nuclear genes, and regulating translation level.

Discussion
Prior to the era of imatinib, there was no effective treatment for recurrent or metastatic GIST and surgery was often attempted in the absence of any alternative or as an emergency for bleeding or gastrointestinal perforation or obstruction. It was previously reported that the rate of complete gross resection was low and the median survival was only 15-19 months 21,22 . And many research 23,24 revealed that the prognosis of patients diagnosed with advanced or metastatic GIST was signi cantly raised by the optimal use of imatinib. Kang 12,25 retrospectively analyzed the correlation between IM and surgery in patients with metastatic and /or recurrent GISTs and concluded that surgical resection of residual lesions after disease control with imatinib is likely to be bene cial to patients. In our series, the OS of the entire group was 56.7months after the second surgery for recurrent t and/or metastatic tumors. But the survival outcome in patients who took IM regularly after resection for primary tumor was signi cantly better than those who did not take IM or take it irregularly(p 0.043), and was also better than previous research in terms of OS. The importance of IM and radical surgery resection should be highlighted at the maximal clinical response of GIST that may be associated with survival bene ts 16,26,27 . Surgical resection, however, improved the prognosis of the patients who were resistant or unresponsive to IM that may contributed to secondary gene mutation to it 28 .
We presented the result of the surgery resection for liver metastasis of GISTs that the OS was obviously lower than those with other sites recurrence or metastasis(p=0.021). As liver was reported to be the most common site of the GIST metastasis, 50-60 % of patients were found to have a liver involvement during the disease process 25,29,30 . Hou et al. 31 revealed that there was no statistical difference in the survival of patients who underwent the radical surgery with different metastasis or recurrence only through IM therapy. A multicenter prospective study 32 got a rough view that liver metastasis of gastrointestinal stromal tumor may not be controllable by surgery alone and require concomitant imatinib therapy. Thus, the therapeutic model and surgical opportunity of those patients should be seriously considered.
Based on the clinical results, we used deep Whole-genome sequence (WGS) to pro le genomic variation in both primary tumors and recurrent and/or metastatic tumors. As we all know, the majority of GIST(75-80 ) harbor gain of function KIT mutation in exons 9,11,13,14,17, and 5-10 of GIST have mutations in platelet-derived growth factor receptor a (PDGFRA) gene in exons 12, 14, 18 33 . As the amount of our samples are small, we observe an interested result that the KIT mutations existed correlations and difference between PT and RMT in Irregular Medication group , and in Regular-medication group, the KIT mutations only were found in PT. With the use of IM, the mutations in KIT could not been attained in the RMT. While the mutations were actually harbored both PT and RMT in irregular group, but these mutations were not completely equal. Miselli FC 34 raised that KIT ampli cation may be a mechanism of drug resistance in GIST. As to reactivation of KIT signaling by tumor subclones with heterogeneous secondary KIT mutations, it resulted in oncogenically-activated KIT to be the key driver of GIST proliferation and survival [35][36][37] . Serrano et al. 38 reported on the activity of nine TKIs which have either been approved or are under clinical investigation as KIT inhibitors for GISTs, against imatinib-resistant GIST cell lines with different secondary KIT mutations, also showed that rapid alternation of sunitinib and regorafenib is more effective than monotherapy using either drug in vitro. Mutations in PDGFRA were only detected in the RMT of the regular group. The difference of KIT /PDGFRA mutations between PT and RMT was likely to be caused by secondary mutation or tumor evolution.
MUC4 was veri ed to play an important role in cell proliferation and differentiation of epithelial cells by inducing speci c phosphorylation of ERBB2 and affecting the formation of a MUC4-ERBB2-ERBB3-NRG1 complex leading to down-regulation of CDKN1B, resulting in repression of apoptosis and stimulation of proliferation 39,40 . The mutations in ASH1L, MUC4 and KMT2D seemed to have coordinated variation between PT and RMT. ASH1L is a histone methyltransferase of lysine speci cally trimethylating 'Lys-36' of histone H3 forming H3K36me3 41 ; MUC4 was supposed to affect the intracellular regulation of ASH1L and KMT2D through intracellular cascade. The decrease of FRG1 expression may in uence tumor progression by regulating cell migration and invasion 42 , and ZNF717, as a transcription factor, involves in nucleic acid binding and DNA-binding 43 . We had reasons to believe that ZNF717 promoted the progression of tumor through the combination with FRG1. The tumor cell growth mainly depended on the MUC family in the two tumors (mentioned above), while the effect of KIT/PDGFRA on RMT decreased signi cantly. And the mutation of ASH1L and KMT2D tended to occur on RMT.
In regular group, we detected the mutations in MTMR11, GOLGA6. MTMR11 is one of Myotubularin Related Protein, involving in process of glycophosphatidylinositol dephosphorylation. The high frequency mutations of GOLGA6L4 and the expression of MTMR11 were found to be obvious abnormal in many cancers 44 . The mutations in EMX2, FAM101A were only attained in liver metastasis group with regular treatment. EMX2 encodes a homeobox-containing transcription factor which could in uence the growth and development of cells and animals 45 ; and it was proved to interact with FLNA to regulate lamin framework around nucleus and the shape of the nucleus. Consequently, we inferred that the mutations had correlation with these patients.
In irregular group, the mutations on SMG7 , RAD54L2 and RBPJ were detetcted . We acquired that SMG7 and SMG5 were involved in cell nonsense-mediated mRNA decay 46 . RAD54L2 is one of DNA helicases, and RBPJ is a transcriptional activator for the Notch signal pathway 47 . The frameshift mutations of the two genes may result in escaping from the effect of IM on tumors. In No-Drug group, mutations in PTPRG, STK35/FANCD2 were detected. As we know, the encoding conduct of PTPRG constitutes Tyrosine protein phosphatase receptor activating a series of signaling cascade to affect protein synthesis ,which was con rmed to play a key role in the development and progression of many cancers 48 . Therefore, its mutation could result in the abnormity of the activation activity, thus in uencing subsequent signaling cascade. STK35 is one of serine/threonine protein kinase and FANCD2, as one of histones, prevents DNA breakage and loss of Chromatin and participants in the stabilization of chromosomes and DNA damage repair 49 .
As an adhesion protein, AJAP1 may be translocated to the nucleus, via its interaction with β-catenin complexes, where it can regulate gene transcription, then possibly have a potent impact on cell cycling and apoptosis and also participates in tumor cell adhesion and intercellular metastasis 50,51 , implying that its mutation may have effect on tumor metastasis or recurrence.
The mutations of coding region of the LOC101927550 existed overlap with SMURF1 encoding a ubiquitin ligase that is speci c for receptor-regulated SMAD proteins in the bone morphogenetic protein (BMP) pathway. So we speculated that LOC101927550 in uenced the targeted combination with miRNA of SMURF1 that regulated the expression of SMURF1, and affected the development of tumors 52 .
For the tumor heterogeneity and evolution, we obtained that the mutations and variants of different sites in the same tumor were signi cantly different. And the differences between PT and RMT were greatly discrepant, mainly re ecting in point mutations and subclones which were related with the grade malignancy and IM resistance in RMT.
Heidi M 53 demonstrated that mutational analysis by use of liquid biopsies can capture the molecular heterogeneity of the whole tumour, and also revealed that multiple resistance mutations were synchronously present. It has been shown that imatinib-resistant disease frequently harbours up to two resistance mutations within a single tumour or metastasis, or up to ve mutations in separate metastases from one patient [54][55][56] .
As B Liegl et al 55 also showed that extensive intra-and inter-lesional heterogeneity of resistance mutations and gene ampli cation in patients with clinically progressing GIST and also underscore the heterogeneity of clinical TKI resistance, and highlight the therapeutic challenges involved in salvaging patients after clinical progression on TKI monotherapies. Thus, we deduce that the spatial and temporal heterogeneity contributes to the occurrence or metastasis of tumor and resistance toTKI.

Conclusions
Although our study is limited by small population size and its restrospective design, it had several strengths, including the rst genome sequencing for GIST within PT and RMT, different modes of TKI treatment, deep whole genome sequencing to analyze the correlations and distinctions within PT and RMT, and the analysis of the tumor evolution and heterogeneity among different times and sites. The different treatment modes of TKI may induce different mutations in code gene, while the irregular group is more likely to contribute to various mutations. And we also highlight the TKI treatment for high risk GISTs after resection for primary tumor. We acquire that mutations in MUC family were very likely to have close connection with the recurrence and metastasis of GIST. Thus, we insist that the genotype and precision medicine for GIST is quietly signi cant and meaningful 57