Natural evolution mechanisms of the IDH mutant glioma revealed by multi-omics sequencing of a long-term survivor

Purpose Low-grade gliomas(LGG) almost invariably progress into secondary glioblastoma with limited therapeutic options. The currently proposed progression mechanisms were mainly based on treatment intervened patients, seldom is known about the mechanisms under natural evolution. Methods A high-grade glioblastoma (G4) occurred at the posterior of the primary LGG (G2) of a 67-year-old woman carried the glioma for eight years without any serious treatment. Tumor samples and peripheral blood of the patients were collected and subjected to integrated genomic analyses, including whole-exome sequencing, gene expression and DNA methylation proling. Results The center and edge of the tumor were diagnosed as LGG and GBM, respectively. They shared the same trunk mutations including IDH, TP53, and ATRX. They both have mixing cell origins and they have a highly correlated methylation level at the probe level. CIC, BRCA2, and RPA4 mutation which occurred only in G4 with mutant allele frequency(MAF) higher than 15% may contribute to evolution. NAF1 of which the MAF increased by 70% and the mutant RNA reads nearly doubled in G4 may also involve in the evolution. In the pathway level, the MSP-RON pathway was strongly up-regulated in G4. Concomitant with the tumor evolution, we discovered enhanced inammatory signals represented by the up-regulation of the NF-κB pathway and the recruitment of mast cell, of which the absolute cell proportion increased from 3.9% to 6.9%. Contradictorily, the adaptive immune response was suppressed as we found pathways associated with IL17, dendritic cells, and cytotoxic T cells were down-regulated and the inltration level of CD4+ and CD8+ T cells were both decreased. As for the TCR prole, only about 2% of blood clonotypes were detected in the tumor microenvironment. Nevertheless, two clonotypes were found signicantly expanded. Clone frequencies of them were 38% and 19% in G2, respectively. And these were 24% and 11% in G4, respectively. Conclusion. and of MSP-RON pathway could promote glioma evolution under natural conditions. Increased inammatory response and decreased adaptive immune response may also contribute to the process. Besides, two highly expanded TCR clonotypes discovered in this case may serve as a potential adoptive cell therapy source in glioma. Glioblastoma subclasses can be dened by activity among signal transduction pathways and associated genomic

Natural evolution mechanisms of the IDH mutant glioma revealed by multi-omics sequencing of a long-term survivor Long

Introduction
Gliomas are a heterogeneous group of brain tumors with distinct biological and clinical properties 1,2 .
Subtypes and grades of gliomas were de ned by histological features traditionally. The 2016 update to the World Health Organization (WHO) classi cation of tumors integrated molecular and phenotypic features into the classi cation of glioma. IDH1/2 mutation, as well as codeletion of 1p/19q, indicates a better prognosis 3,4 . Meanwhile, transcriptome analyses of glioma have classi ed them into four subtypes denoted proneural, mesenchymal, classical, and neural [5][6][7][8] , which were correlated with different neural lineages and drug sensitivity. Representative molecular biomarkers of the four subtypes were EGFR, NF1, PDGFRA, and IDH1, respectively. However, in some cases boundaries of the four subtypes were blurred in that a sample might show patterns of more than one subtype [5][6][7][8][9] . Furthermore, several epigenetic markers have also shown prognostic and/or predictive values 10,11 . O6-methylguanine-DNA methyltransferase (MGMT) and cytosine-phosphate-guanine (CpG) island methylator phenotype (CIMP) were the two biomarkers that were wildly studied. CIMP was closely correlated with IDH for nearly all IDH-mutant gliomas were CIMP 10,12,13 . Recently, two subsets of IDH-mutant/CIMP gliomas were discovered, CIMP-low and CIMP-high. The former presented a relatively low degree of DNA methylation and poorer outcome compared to the latter 10,12−14 .
Low-grade gliomas (LGG) almost invariably progress into secondary glioblastoma (sGBM) within 5-10 years after diagnosis [15][16][17] . sGBM accounts for about 20% of GBM and is generally identi ed by harboring IDH mutation 15,18 . Although the evolutionary landscape 19,20 of GBM has been widely studied, the mechanisms behind remain elusive. Bai and colleagues 21 compared the progressed GBM samples to their lower-grade counterparts in 41 IDH mutant glioma patients. They found nonlinear clonal expansion of the original tumors and identi ed some key oncogenic pathways that drive progression, including activation of the MYC and RTK-RAS-PI3K pathways. Besides alterations in oncogenic pathways, local immunosuppression also plays an important role in glioma progression [22][23][24][25] . As well-known, glioma presents a more immunosuppressive tumor microenvironment compared with other tumors. Fewer leukocytes in ltration caused by blood-brain-barrier is the main reason. T cell exhaustion and apoptosis via PD-1 ligands (PD-L1/2) expressed by tumor cells also play a role 26 . Effector T cells could be silenced by a combination of soluble factors, such as IL10 and TGFb 22,27,28 , and be inhibited by myeloid-derived suppressor cells (MDSCs, CD33 + HLA-DR), tumor-associated macrophages, and regulatory T cells (Tregs) 29 . Furthermore, Mohme, et al. 22 identi ed a distinct exhaustion signature in tumor-in ltrating lymphocytes (TILs) compared to peripheral blood lymphocytes (PBLs) and discovered a contracted repertoire of TCR in recurrent GBM compared to newly diagnosed GBM.
Although a variety of mechanisms have been proposed to clarify glioma progression, all of them are based on the comparison of surgically resected specimens at diagnosis and at relapse, and a majority of them included patients who received chemo/radio therapy, which re ected the arti cial selection. How gliomas evolve under natural conditions remains unknown. Here, we try to answer this question by comparing the multi-omics sequencing data of LGG and GMB specimens resected simultaneously from a long-term survivor who carried a tumor for more than eight years without any formal medical interference.

Materials And Methods
Biospecimen acquisition and histological diagnosing The two adjacent tumors were resected surgically and were applied to histological examination separately. Tumor purity was evaluated histologically. Antibodies used in IHC staining included Anti-ATRX antibody (ID: Sc-55584, Santacruz), Anti-Ki67 antibody (ID: ab15580, Abcam), and Anti-Olig2 antibody (ID: ab109186, Abcam). DNA and RNA of the tumor were extracted using AllPrep DNA/RNA Mini Kit (ID: 80204, QIAGEN). Peripheral blood of the patient was collected using tubes coated with EDTA before surgery, then DNA was puri ed from the white blood cells (WBC) centrifuged.

Whole exome sequencing and analyzing
Exonic regions of DNA were captured in a solution using the Agilent SureSelect approach according to the manufacturer's instructions (Agilent, Santa Clara, CA). Paired-end sequencing was performed using the HiSeq X Ten next-generation sequencing instrument (Illumina, San Diego, CA). The average data output was about 200×coverage. Somatic variants were identi ed using GATK Best Practices Pipeline 30 .
Somatic single-nucleotide variants (SNVs) were detected using MuTect 31 , while somatic indels were identi ed using Strelka 32 . Cancer cell fraction was inferred using EXPANDAS 33 , clone phylogeny between different samples was estimated by CITUP 34 , and was visualized by TimeScape 35 .
mRNA sequencing and analyzing.
RNA was isolated and sequenced. The reads were aligned to the Ensembl GRCh37 human genome assembly with default parameters. And were normalized by FPKM. Tumor purity was evaluated by ESTIMATE 36 using FPKM. ssGSEA was performed using the R package GSVA 37 . Transcriptomic pro les of the two tumors were compared using ssGSEA based on three collections of annotated gene sets from the Molecular Signature Database v7.1 (C6 oncogenic genesets, C7 immunologic genesets, and BIOCARTA subset of Canonical pathways of C2) 38  T cell receptor (TCR) sequencing and analyzing DNA puri ed from the two tumors and peripheral blood was analyzed by high-throughput sequencing of the full length of TCRβ chain by the ImmuHub TM TCR pro ling system (ImmuQuad Biotech, Hangzhou, China). Brie y, sequencing was performed on an Illumina MiSeq® system with PE300 mode (Illumina, San Diego, California, the USA). MiTCR (milaboratory) algorithm was applied to raw sequencing data for PCR and sequencing error correction and V, D, J, C gene segments mapping with IMGT ® . The resulting nucleotide and amino acid sequences of CDR3 of TCRβ chain were determined and those with out-offrame and stop codon sequences were removed from the identi ed TCRβ repertoires. The Clonality Index specialized in characterizing the state of clonal proliferation of T cells and the Shannon Index utilized to evaluate clonotype diversity was calculated as previously reported 45 .

Data visualization
Figures were visualized using matplotlib 3.2.1 except for the evolutionary tree which was drawn using the R package TimeScape.

Clinical presentation and histopathological considerations
A 67-year-old woman, of whom an intracranial lesion was detected and a closer follow-up observation was prescribed in June 2010. In October 2013, a low-density area with a suborbicular shadow was observed at the right frontal lobe. Attributed to a variety of reasons including no deteriorating symptoms, she didn't receive any formal medical care. In May 2018, she suffered aggravated symptoms, imaging examination showed the tumor at the right frontal lobe increased, while a new enhanced shadow appeared at the posterior of the primary tumor, which indicated the occurrence of a secondary highergrade glioma there (Fig.1A). The primary and secondary tumors were resected and diagnosed as grade II diffuse astrocytoma (G2 in abbreviation) and grade IV glioblastoma (G4 in abbreviation), respectively.
The immunohistochemical result showed that Ki67 expression was higher in G4 than in G2, Olig-2 (oligodendrocyte transcription factor 2) expression was positive in both samples and was higher in G4 ( Fig.1B).

The molecular origin of G4
Clinical trajectory suggested that the G4 originated from G2. We explored it at the molecular level. Tumor purity of G2 and G4 were both 100% evaluated histologically, and those were 97% and 96% separately estimated by ESTIMATE (36), which guaranteed the comparability of them. A total of 117 and 96 somatic mutations were detected in G2 and G4, respectively. Among them, 53 mutations were shared between the two samples ( Fig.2A). Of note, Minor allele frequencies (MAF) of glioma driver genes IDH(p.R132S), TP53(p.R234C), and ATRX(p.T1582NfsTer19) were approximately equal between the two samples (Fig.  2B). TERT promoter mutations, EGFR VIII fusion, and 1p/19q co-deletion were detected in neither samples. Therefore, the mutation subtypes of these two specimens were both IDH mutant and 1p/19q non-codel, which were consistent with the characteristics of diffuse astrocytoma.
Transcriptionally, IDH mutations suggested these two tumors were of proneural subtype. ssGSEA analysis using the signature de ned by Verhaak et al. 5 veri ed the result, proneural got the highest score among the four transcription subtypes in both samples, which was 32.4% in G2 and 32.8% in G4. However, ssGSEA scores of the rest three transcriptional subtypes were also comparable, which implied the mixing subtypes (Fig.2C). And cell of origin estimated by ssGSEA using signature de ned by Cahoy et al. 39 also favored that both samples were of mixing cell origination. Oligodendrocytes accounted for 28.7% and 29.4% in G2 and G4 separately, while astrocytes accounted for 26.6% and 28.3% in G2 and G4 separately ( Fig.2C).
At the methylation level, beta values at all CpG loci of G2 and G4 were highly correlated, with R 2 equals 0.991. And only 4,035 out of 761,759 CpGs showed a difference in beta value greater than 0.2 between these two samples (Fig.2D). MGMG promoter status of G2, as well as that of G4, was methylated, with the probability of methylated status equals 0.99 estimated by MGMT-STP27 mode for both samples. Epigenetic subtype analysis revealed that both samples were CIMP+ as the probability density of betavalues of probes used to de ne CIMP+ reached a peak when the beta-value was about 0.76, which was in accordant with that both samples were IDH mutant (Fig.2E). Furthermore, both samples were CIMP-high as the majority of beta-values of probes utilized to de ne CIMP-high distributed around 0.75 (Fig.2F).
Tumor evolution trajectory from G2 to G4 43 mutations were identi ed only in G4 (private mutations, Fig.3A). Among them, CIC mutation (p.Thr761LeufsTer163) had a MAF of 23.3%, and at the transcription level, mutant CIC reads accounted for 10% of overall CIC reads detected in RNA sequencing. Two genes RPA4 (p.S2IfsTer50) and BRCA2 (p.M3217V), which were related to homologous recombinations (HR) pathway, had MAFs of 24% and 15.4%, respectively.
Seven clones were identi ed (Fig.3B, Table 1). Driver genes with high MAF including IHD, TP53, and NAF1 were all assigned to clone 1, which was the earliest event. Of note, the N219K mutation of NAF1 has never been reported before. MAF of it increased from 44.4% in G2 to 75.0% in G4, and the proportion of the mutant RNA reads increased from 37.5% to 70.0%, correspondingly. Clone 2, which was also an early event, composed the second-largest cell population in G4. NOTCH1(p.A1967V) mutation, of which MAF increased from 20.5% in G2 to 37.5% in G4, was assigned to this clone. Clone 4 was the biggest in G4, and private mutations of G4 including CIC, BRCA2, and MMRN1 were all assigned to this clone, indicating their role in tumor progression. Clone 5 was a new event in G4, MAF of all genes assigned to this clone was less than 10%, except FFAR1 (p.A17V). Contradictorily, clone 6 and clone 7, which occurred in G2, were almost eliminated in G4.
Oncogenic pathways of the two samples were further compared by ssGSEA. Pathways of which the absolute difference of normalized enrichment score (NES) calculated between G4 and G2 was greater than 20% of the NES of G2 were analyzed. (Fig.3C). The highest up-regulated pathway in G4 was the MSP-RON pathway, activation of which can stimulate the Ras-PI3K signaling pathway(46) which was reported to be altered in 68% of recurrent glioma patients (21). The second up-regulated pathway in G4 was the nuclear factor-κB (NF-κB) pathway, activation of which could enhance cancer cell proliferation and induce immunosuppression. Contrarily, the down-regulated pathways included the dendritic cell pathway, which can stimulate Th1 and Th2 cell differentiation, IL17 pathway, and cytotoxic T cell pathway, indicating the immunosuppression in G4.

Immune microenvironment alteration concomitant with tumor evolution
Immune cell fraction in the two tumors was deconstructed using CIBERSORT(47) (Fig.4A). The in ltration levels of activated mast cells, memory B cell, and follicular helper T cell were higher in G4 (Fig.4A), suggesting their role in defending glioma progression. However, The in ltrations of resting CD4+ memory and CD8+ T cells were decreased in G4 noteworthily, which was consistent with the down-regulation of Th1, Th2, Th17, and cytotoxic T cell pathway aforementioned and implied the evasion of the adaptive immune response in G4.
TCR sequencing of the two tissue samples and peripheral blood was performed to pro le epitope recognition patterns during glioma evolution. Of the 1926 clonotypes discovered in peripheral blood, only 42 (2.13%) were detected in G2 or G4 (Fig.4B). Diversity estimated by Shannon-Index was higher in G4 than in G2, correspondingly, clonality was lower in G4 than in G2, which suggested the occurrence of new epitopes during glioma evolution. 15 clonotypes were identi ed in common between G2 and G4 (Fig.4C), which accounted for 13.2% and 11.4% of all clonotypes of G2 and G4 separately. clone frequencies of all these shared clonotypes were decreased from G2 to G4. Among them, two clonotypes (CASGDRVSWGYTF and CASSLSRGQLNGYTF) had the highest clone frequency in all of G2, G4, and peripheral blood (Fig.4C-D).

Discussion
This case had drawn our attention because a secondary glioblastoma appeared beside the primary LGG without any formal treatment interference in eight years, which enabled us to explore the evolution mechanisms of glioma under natural conditions. Both of the two tumors harbored driver mutations including IDH, TP53, and ATRX, mixing cell origins, and of CIMP-High, which guaranteed that the G4 originated from G2. Two processes were generally considered to contribute to the progression, tumor intrinsic proliferation, and host immune suppression.
Tumor evolution trajectory revealed that clone 4 composed of private mutations in G4 mainly expanded most. Among them, CIC mutation indicates the proliferation of oligodendrocytes, RPA4 and BRCA2 mutations suggest the inactivation of DNA damage repair pathway. Clone 1, which was the earliest event, included tumor cells harboring IDH, TP53, and NAF1 mutations. The p.N219K of NAF1 has not been recorded in COSMIC and was rstly reported in this case to our knowledge. N219K locates at the Cbf5binding domain of NAF1. The function of this mutation is unknown, however, as MAF of NAF1 mutation increased nearly 70% and the proportion of mutant RNA reads of NAF1 increased by about 90%, this mutation probability supplied growth advantage. And overexpression of NAF1 was reported to promote tumorigenesis and progression of glioma through modulating ribosome assembly and protein synthesis 48 . At the pathway level, we found the activation of the MSP pathway played an important role in the natural progression of glioma. MSP is the ligand for the receptor tyrosine kinase RON, MSP-RON can activate two signaling pathways: RAS-ERK and PI3K-AKT, which were altered in 68% of recurrent glioma patients 21 . However, Bai, et al. 21 also reported there was a weak association of RTK-RAS-PI3K pathway activation with treatment (q = 0.05). Our case favored the perspective that RTK-RAS-PI3K pathway activation was a genomic event that underlines glioma progression rather than a selection result of treatment.
In the prespective of immune defending, immune microenvironment analysis revealed that the in ltration levels of CD4 + memory T cells and CD8 + T cells were lower in G4, and signal pathways including IL17 and cytotoxic T cell were down-regulated in G4, indicating the immune suppression of T cell response.
this result was a little different from the report of Mohme, et al 22 . They found CD8 + T cell fraction was signi cantly higher in recurrent GBM than primary LGG, while CD4 + T cell fraction was signi cantly lower. The discordance of CD8 + T cell in ltration reveals the unique immune microenvironmental feature during evolution without arti cial interference. We also detected the up-regulation of the NF-κB pathway in G4, activation of which could enhance cancer cell proliferation and induce immunosuppression 49 .
Furthermore, Mohme, et al. 22 have found that the single most expanded clonotype occupied a signi cantly greater space in the recurrent GBM TIL, especially a few top-ranked clonotypes of which clone frequency can be higher than 15%. In our case, clone frequencies of the top two clonotypes were separately 38% and 19% in G2 and were separately 24% and 11% in G4, which were higher than those reported previously. The reason may be that the patient in our case was a long-term tumor carrier.
Inderberg and colleagues 50 proposed long-term surviving cancer patients as a source of therapeutic TCR. The top two TCRs discovered in our case may have a great value in adoptive cell therapy of glioma.
In conclusion, we discovered biological processes including mutation of CIC, BRCA2, and NAF1, activation of MSP-RON and NF-κB pathways, and immune evasion contribute to the natural evolution of glioma. We identi ed two highly expanded TCR clonotypes that may recognize and eliminate glioma cells and may serve as potential adoptive cell therapy sources.