Genetic alteration and clonal evolution of primary glioblastoma into secondary gliosarcoma

Abstract Aims Secondary gliosarcoma (SGS) rarely arises post treatment of primary glioblastoma multiforme (GBM), and contains gliomatous and sarcomatous components. The origin and clonal evolution of SGS sarcomatous components remain uncharacterized. Therapeutic radiation is mutagenic and can induce sarcomas in patients with other tumor phenotypes, but possible causal relationships between radiotherapy and induction of SGS sarcomatous components remain unexplored. Herein, we investigated the clonal origin of SGS in a patient with primary GBM progressing into SGS post‐radiochemotherapy. Methods Somatic mutation profile in GBM and SGS was examined using whole‐genome sequencing and deep‐whole‐exome sequencing. Mutation signatures were characterized to investigate relationships between radiochemotherapy and SGS pathogenesis. Results A mutation cluster containing two founding mutations in tumor‐suppressor genes NF1 (variant allele frequency [VAF]: 50.0% in GBM and 51.1% in SGS) and TP53 (VAF: 26.7% in GBM and 50.8% in SGS) was shared in GBM and SGS. SGS exhibited an overpresented C>A (G>T) transversion (oxidative DNA damage signature) but no signature 11 mutations (alkylating‐agents – exposure signature). Since radiation induces DNA lesions by generating reactive oxygen species, the mutations observed in this case of SGS were likely the result of radiotherapy rather than chemotherapy. Conclusions Secondary gliosarcoma components likely have a monoclonal origin, and the clone possessing mutations in NF1 and TP53 was likely the founding clone in this case of SGS.


| INTRODUC TI ON
Gliosarcoma, a rare variant of glioblastoma multiforme (GBM) containing both gliomatous and sarcomatous components, is considered a Grade IV neoplasm according to the WHO classification. 1 Gliosarcoma is clinically similar to GBM and affects mainly 50-70-year-old adults, with a higher proportion found in men. 2 Patients with gliosarcoma and GBM show similarly poor survival outcomes and are typically treated using the same regimen including maximal safe resection, and concomitant radiotherapy and chemotherapy. [3][4][5][6] The pathogenesis of gliosarcoma remains unclear. One hypothesis states that the gliomatous and sarcomatous components originate from different progenitor cells, with the sarcomatous element arising from vascular smooth muscle cells, pluripotent mesenchymal cells of the perivascular adventitia, fibroblasts, or even histolytic cells. [7][8][9][10][11] This concept of biclonal origin of gliosarcoma is based mainly on distinct histological and morphological characteristics of two components. Another prevalent hypothesis, which states that both components share a monoclonal origin from a common progenitor cell, is supported by several studies showing identical genetic alterations (including mutations in TP53 or PTEN, and copy number variation (CNV) of P16 or CDK4) in the two components. [12][13][14][15] Several other genetic variations, which are common in GBM, have also been found in both gliomatous and sarcomatous components of SGS, including CNVs on chromosomes 7, 9p, 9q, 13q, 20q, and X. 13,16 Most gliosarcomas are de novo (primary), while those arising after chemoradiation of primary GBM are termed secondary gliosarcomas (SGS). Radiation therapy is conventionally used for the treatment of GBM. Various radiation-induced intracranial tumors include meningiomas, gliomas, fibrosarcomas, and gliosarcomas. 17 Radiation-induced tumors arise within or adjacent to the previously irradiated field and exhibit a histologically distinct type from the original tumor. 18 SGS tumors occurring after treatment of GBM are distinguishable from radiation-induced gliosarcomas, which develop after intracranial radiation without prior history of GBM. A clinical study by Han et al. 19 showed that patients with SGS and radiationinduced gliosarcomas show distinct survival outcomes and latency periods between radiation and diagnosis of gliosarcoma. In that study, all the patients with SGS had undergone radiotherapy, suggesting that radiation may act as a critical agent for the induction of gliosarcoma. Deb et al. found that TP53 mutations were present in both gliomatous and sarcomatous components of SGS, 20 supporting the theory for a monoclonal SGS origin. However, the association between anti-GBM radiotherapy and histogenesis of SGS sarcomatous elements remains unclear.
Although the clinicopathological, molecular, and genetic characteristics of GBM and SGS have been described previously, 21,22 few studies have examined the genomic alteration and clonal evolution involved in the progression of original GBM into SGS. In this study, we describe a patient with primary GBM that progressed to SGS 10 months after radiochemotherapy. To elucidate the clonal origin of SGS, we performed whole-genome sequencing (WGS) and ultra-deep whole-exome sequencing (WES) using the paired primary GBM and SGS specimens obtained from this patient. Comparison of primary GBM profile against that of SGS using genomic analysis indicated that this case of SGS possessed a monoclonal origin. Mutation signature analysis indicated that therapeutic radiation significantly contributed to the genesis of SGS examined in our present study. To the best of our knowledge, this study is the first to describe a comprehensive genomic profile and clonal architecture of tumor evolution from GMB to SGS.

| Mutation detection
Low-quality reads and sequencing adaptors were removed from raw sequencing data to obtain clean reads. Human reference genome GRCH37, downloaded from the Ensembl database, was used as reference to align reads using Burrows-Wheeler Aligners (BWA). 23 The Picard tools were used to sort the alignment and remove duplicate reads from the aligned results. GATK-Mutect2 was employed to detect somatic single nucleotide variants (SNVs) and to call small insertions and deletions (indels). 24 SNVs and indels were annotated by variant effect predictor (VEP) and ANNOVAR. 25,26

| Detection of structural variants
Structural variants (SVs) were predicted by DELLY2 using BAM files containing alignments of paired tumor and peripheral-blood control samples. 27 Tumor-specific somatic SVs were identified by comparing SVs found in samples of GBM/SGS tissues against those in control blood samples. All somatic SVs were annotated using ANNOVAR to determine their type and functional classification.

| Detection of copy number variants
Somatic copy number variants (CNVs) were detected by Control-FREEC using aligned reads of matched tumor and peripheral-blood control samples as input. Control-FREEC can analyze overdiploid tumor samples and samples contaminated by normal cells. 28 Control-FREEC was also used to identify loss-of-heterozygosity events. All detected CNVs were consequently annotated according to the Decipher, DGV, and ISCA databases.

| Mutational signature analysis
Mutational signature analysis was performed using MuSiCa to visualize the somatic mutational profile and extract the contribution of reported mutational signatures. 26 All SNVs identified using WES deep-sequencing data were utilized as input for analysis using MuSica.

| Clonal evolution analysis
Deep-sequenced mutations in WES data were used for the analysis of clonal evolution. Somatic non-silent SNVs (nonsynonymous mutations, nonsense mutations, and mutations affecting splicing sites) with VAF > 0.1% in the samples of GBM or SGS tissues were selected for the inference of clonal population structures. Pyclone was used to identify and quantify clonal populations in our samples of GBM and SGS tissues. 26

| Analysis of mutation incidence in GBM patients from databases
To determine whether the mutations identified in this study commonly occur in primary GBM, we screened the data of mutations in several available databases including the Chinese Glioma Genome F I G U R E 1 Magnetic resonance imaging and histopathological staining of braintumor tissues. (A) Sagittal T1-weighted contrast-enhanced MRI imaging of initial and recurrent tumors. (B) Representative images of primary glioblastoma (GBM) and secondary gliosarcoma (SGS) tissues stained using hematoxylin and eosin (HE), and immunolabeled using antibodies to glial fibrillary acidic protein (GFAP) and reticulin Atlas (CGGA, http://www.cgga.org.cn), The Cancer Genome Atlas (TCGA, http://cance rgeno me.nih.gov), and GSE16011 from Gene Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo/). These databases provide the molecular, genomic, and clinical data of the patients with different types of tumors, and are commonly used to screen the biomarkers with implications of prognosis and/or therapy resistance. [29][30][31] 3 | RE SULTS

| Molecular characteristics of glioblastoma multiforme and secondary gliosarcoma
The patient described in our present study was diagnosed with primary GBM using standard and molecular pathological analyses.

PTEN (T232fs) and P53 (V203M) mutations and EGFR amplification
were identified in the primary tumor tissue, but mutations in IDH1/2, BRAF, and TERT, and 1p/19q co-deletion were not detected. MGMT promoter methylation status was negative as indicated by pyrosequencing. The patient recovered, showing a Karnofsky Performance Scale score of 90 after the first gross total resection, but was diagnosed with SGS 10 months after the first GBM resection, and died from tumor progression 5 months after she was diagnosed with gliosarcoma. The recurrent tumor contained GFAP-expressing gliomatous tissue and reticulin-rich sarcomatous elements ( Figure 1B). The molecular and genetic characteristics of SGS tissue were identical to those of primary GBM. Many literatures report that the genetic profile of gliosarcoma is similar to that of GBM, except for absent or minimal EGFR amplification and rare cases of IDH mutations. 14,32 In this patient, EGFR amplification was identified in both primary GBM and SGS using fluorescence in situ hybridization (FISH).

| Genomic analysis
To understand the genomic-alteration and clonal-evolution processes involved in the progression of primary GBM into SGS, we used WGS to analyze the paired samples of primary GBM and recurrent SGS, and samples of peripheral blood mononuclear cells used as matching controls. WGS has a sequencing depth of approximately 30×. To increase the quantification accuracy of variant-allele frequency and detection sensitivity for low-abundance variants, we used ultra-deep WES to generate more than 50 Gbp of sequencing data for both GBM and SGS. Alignment of the WES data obtained in this study covered 95% of the exome region with >100× depth; average sequencing depth for GBM and SGS was 553× and 652×, respectively.
A total of 3729 somatic mutations were identified in the GBM WES data, of which 353 were non-silent SNVs (missense, nonsense, and splicing-site changing) and 80 were indels in the coding regions (Data S1). In the SGS WES data, we identified 1050 mutations, including 186 non-silent SNVs and 24 coding indels (Data S1). Among the non-silent mutations, 42 were shared between the two samples, while 391 and 168 were specific to GBM or SGS, respectively. Tumor mutation burden, calculated as the number of mutations per megabase, was 7.2 for GBM and 3.5 for SGS.
Whole-genome sequencing data were used to identify wholegenome SNVs, small indels, and larger variants including CNVs and SVs. We identified a total of 20,191 somatic mutations (18,   10% of GBM tumors harbor an NF1 somatic mutation. 34 Considering that the VAF of NF1 was nearly 50%, that the mutation was heterozygous, and that no CNVs were detected in the NF1 gene region, we conclude that this NF1 mutation must have been present in nearly all the tumor cells of primary GBM and SGS, indicating that the two components of SGS may originate from the same ancestral cancer cell.

| Clonal evolution
Another important cancer-driving gene in cluster 7 was TP53.
Only one mutation site in TP53 (V203 M/c.607 G>A) was observed in our study. The VAF of this TP53 mutation increased from 26.7% in GBM to 50.8% in SGS ( Figure 3D

| Mutational signatures
The overall frequency of SNVs identified in our WES data was 127.  Figure 4D). Compared with that of GBM, a higher contribution of COSMIC signature 3,9,12,15, and 21 was observed in SGS ( Figure 4D). Although the relationship between these signatures and radiation exposure is still unclear, most of these signatures are associated with defects in DNA repair. 36 Interestingly, signature 11 was not found in the SGS examined in our study ( Figure 4D). Because signature 11 is characterized predominantly by C>T (G>A) mutations and is associated with exposure to alkylating

| Mutation incidences of NF1 and TP53 in GBM patients from TCGA database
To determine whether NF1 and TP53 mutations commonly occur in primary GBM, we searched the data of these mutations from TCGA, CGGA, and GEO databases. Finally, the relevant data were screened out only from TCGA database, including 390 primary GBM samples (without chemoradiotherapy before surgical resection) and 10 recurrent GBM (occur after radiochemotherapy). The overall incidences of TP53 and NF1 alterations in primary GBM patients are 31% (121/390) and 12% (47/390), respectively ( Figure 5).
We also attempted to investigate whether radiotherapy increased the incidence of NF1 and TP53 mutations by comparing the mutations rates between the paired primary (without radiotherapy) and recurrent (occur after radiotherapy) GBM cases from TCGA. However, there were only nine paired cases in the database, and most of them received both radiotherapy and chemotherapy (data not shown). Therefore, we did not perform any further analysis of the mutation incidence after radiotherapy. Feigin and Gross were the first to detail gliosarcoma and proposed that the sarcomatous elements arose from neoplastic transformation of blood vessels induced by the malignant glial cells. 38 However, this hypothesis is hindered by inconsistent expression of vascular markers in the sarcomatous component. Some studies have proposed that the sarcomatous component is derived from pluripotent mesenchymal cells of perivascular adventitia, fibroblasts, or even histocytes. [8][9][10][11] However, genetic studies have identified a similar genetic profile in both components, 12-16 strongly indicating that gliosarcoma has a monoclonal origin. In the case described herein, a cluster of mutations was shared in both primary GBM and SGS, including a founding mutation in the two tumor suppressor genes NF1 and TP53. Therefore, our results also indicate that gliosarcomas have a monoclonal origin.

| DISCUSS ION
Most patients with SGS were previously managed with radiotherapy for the original GBM. 19,32,39 Gliosarcoma can also occur after radiation therapy in patients with other tumor phenotypes (e.g., leukemia, meningioma, and low-grade glioma), which is termed radiationinduced gliosarcoma. Some cases of radiation-induced gliosarcoma develop within the radiation field but at a location separate from that of the primary tumor. 19 These observations indicate a causal association between therapeutic radiation and induction of sarcomatous components. Several anticancer drugs, including alkylating agents, are also mutagenic. Because the patient described in our present study was administered both radiotherapy and treatment with temozolomide, we examined the mutational signatures of tumor tissues F I G U R E 5 Somatic alterations in TP53 and NF1 in glioblastoma multiforme (GBM) patients from TCGA database. There are 390 primary and 10 recurrent GBM samples. In primary GBM patients, the somatic alterations of TP53 and NF1 occurred in 121 and 47 samples, respectively. In recurrent GBM patients, the alterations of TP53 and NF1 occurred in 6 and 1 samples, respectively obtained from this patient in order to investigate the possible etiology of tumor evolution from GBM to SGS. Our results indicate that the prevalence of C>A (G>T) transversions was higher in SGS than in GBM. This mutational alteration is not in accord with the pattern induced by temozolomide, which is characterized by a predominance of C>T (G>A) mutations. 37 Consistently, the contribution of signature 11 in SGS was scant; this signature is associated with exposure to alkylating agents. The C>A (G>T) transitions in SGS can be explained by ROS-induced mutagenic DNA lesions following radiotherapy. Our results indicate that therapeutic radiation was a significant contributor to the somatic mutations observed in this SGS tumor.
The clone containing mutations in NF1 and TP53 may have been the founding clone in this case. The TP53 gene, known as the guardian of the genome, is critical in detecting DNA damage and preventing damaged cells from passing the damaged DNA to their daughter cells. Our analysis of the genomic data from TCGA showed that the overall incidences of TP53 mutations were 31% (121/390). Genetic studies have showed that gliosarcoma has a higher frequency of TP53 mutation than primary GBM. 14,32 Consistently, we observed a higher prevalence of TP53 mutations (V203M) in SGS compared with that in primary GBM. A recent study demonstrated that temozolomide can induce two TP53 missense mutations (R110C/c.328 C>T and R175H/c.524 G>A) in glioma spheres derived from primary GBM, and that these mutations may facilitate epithelial-to-mesenchymal transition. 40 These findings indicate that TP53 mutations likely participate in driving the development of GBM to SGS. Two recent studies, examining molecular and genetic profiles in a gliosarcoma patient with multiple recurrences and an extracranial metastasis, showed that several somatic mutations that are key in primary gliosarcoma, such as a TP53 mutation, were shared in recurrent and metastatic tumors. 41,42 Our analysis demonstrated that the overall incidences of NF1 mutations were 12% (47/390) in primary GBM patients from TCGA cohort. The NF1 mutation in gliosarcoma patients has also been reported in two studies, with the frequency of 21% (3/14) 32 and 30% (3/10), 43 respectively. Due to the small number of patients in the two reports, further studies are needed to reveal the overall incidence of NF1 mutations in gliosarcoma. Tumor stem cells theory is an alternative explanation for tumor development and progression.
Although our study indicated that the clone possessing mutations in NF1 and TP53 was the founding clone in this case, whether the clone originates from glioma stem cells or other progenitor cells is unknown. To investigate whether the glioma stem cells harbor NF1 and TP53 mutations, we have searched the mutation profile of glioma stem cells from two databases, including COSMIC Cell Lines Project (https://cancer.sanger.ac.uk/cell_lines/) and GSE23806 (download from GEO website, https://www.ncbi.nlm.nih.gov/geo/). 44 .However, we did not find the relevant information in the two databases.
Our results indicate that mutation rate decreased significantly in SGS (3.5 per Mb) as compared with that in GBM (7.2 per Mb).
Interestingly, our GBM samples exhibited a high burden of somatic mutations and low CNVs, whereas our SGS samples showed a low burden of somatic mutations and high CNVs. The inverse relationship between these two genomic aberrations may indicate the presence of a compensatory mechanism in tumor evolution. 45 When mutation load is high, the decline in CNVs is driven by cellular autonomous mechanisms and immune response to neoantigens. When mutation load decreases following chemoradiotherapy, CNVs increase because of suppression of immune surveillance in the tumor microenvironment. Further studies are necessary to delineate the dynamic correlations between these genomic aberrations.
In conclusion, herein, we described the first genome-wide deep sequencing of paired primary GBM and SGS samples obtained from the same patient. Our results provide genomic evidence for the monoclonal origin of the two components of SGS and for the relationship between therapeutic radiation and SGS pathogenesis.
Therapy-driven tumor evolution is a major impediment in the management of GBM. Improving our understanding of the molecular and genetic mechanisms driving therapy-driven tumor evolution will facilitate the development of more effective therapeutic strategies.
Due to the rarity of SGS, only one patient with SGS was described in this study. Future studies enrolling more patients will help reveal the mechanisms involved in the transformation of GBM into SGS.

ACK N OWLED G M ENTS
We thank the Wuhan Frasergen Bioinformatics Company Limited for performing the next-generation sequencing described in our study.

CO N FLI C T O F I NTE R E S T S
The authors declare that this study was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
The sequencing data were deposited into the Sequence Read Archive database under accession number PRJNA720573. These data will be made available to the public after publication of this manuscript.