High Copy-Number Variation Burdens in Cranial Meningiomas From Patients With Diverse Clinical Phenotypes Characterized by Hot Genomic Structure Changes

Meningiomas, as the most common primary tumor of the central nervous system, are known to harbor genomic aberrations that associate with clinical phenotypes. Here we performed genome-wide genotyping for cranial meningiomas in 383 Chinese patients and identified 9,821 copy-number variations (CNVs). Particularly, patients with diverse clinical features had distinct tumor CNV profiles. CNV burdens were greater in high-grade (WHO grade II and III) samples, recurrent lesions, large tumors (diameter >4.3 cm), and those collected from male patients. Nevertheless, the level of CNV burden did not relate to tumor locations, peritumoral brain edema, bone invasion, or multiple lesions. Overall, the most common tumor CNVs were the copy-number gain (CNG) at 22q11.1 and the copy-number losses (CNLs) at 22q13.2, 14q11.2, 1p34.3, and 1p31.3. Recurrent lesions were featured by the CNLs at 1p31.3, 6q22.31, 9p21.3, and 11p12, and high-grade samples had more CNVs at 4q13.3 and 6q22.31. Meanwhile, large tumors were more likely to have the CNVs at 1p31.3 and 1p34.3. Additionally, recurrence prediction indicated the CNLs at 4p16.3 (p = 0.009, hazard ratio = 5.69) and 10p11.22 (p = 0.037, hazard ratio = 4.53) were candidate independent risk factors.


INTRODUCTION
Meningiomas represent the most common primary intracranial tumor type, accounting for 37% of central nervous system neoplasms (1). They are believed to arise from progenitor cells of both the arachnoid cap cells of the arachnoid layer and fibroblasts that reside in the inner dura mater (2). Despite the identification of NF2 mutations or loss of function, recent sequencing studies also revealed mutations involving TRAF7, KLF4, AKT1, SMO, POLR2A, and the ARID1A and TERT promoters of in meningiomas (3)(4)(5)(6). Cytogenetic changes, such as losses of chromosomes 22q, 14q, 1p, 1q, 10q, and 9q, are also commonly reported, some of which are related to tumor progression (7)(8)(9)(10)(11)(12). Copy-number variations (CNVs) of cytobands located at 22q, 1p, and 14q were most common (12)(13)(14). In tumor development, losses of 6q and 4q have been reported to be significantly associated with high-grade lesions (13). Furthermore, meningiomas in specific locations may have featured CNVs; for instance, those at anterior skull base are likely to have intact chromosome 22q, which loses tumor suppressor gene NF2 (15). However, as a relative benign tumor, meningioma had few data from a relatively large cohort to characterize genome-wide CNV changes, which limits efforts on applying them in tumor progression evaluation, prognosis, and the development of new treatments.
Although maximal but safe resection can cure the majority of meningiomas (16), tumor recurrence still occurs even after gross total resection (GTR) (17). The recurrence status cannot be completely predicted by histopathologic grade alone, as it is mainly based on histopathological characterizations of mitotic rate, cellular features of atypia, and local invasion (18). Meningiomas are well-known for their female-biased predominance (19), but tumors in male patients demonstrate not only a higher annual growth rate (20) but also a higher probability of recurrence (21)(22)(23). Previous studies have proposed molecular markers for prognostic scoring systems in recent years (14,21,(24)(25)(26), and a better WHO classification of meningiomas integrated with independent molecular markers may help to predict the recurrence risk and adjust treatment plans for patients with meningiomas. Although genomic structure changes in neurologic tumors are common, extensive efforts are still required to evaluate roles of diverse recurrent CNVs in the models for tumor classification, prognosis scoring, and recurrence prediction.
To our knowledge, we here collected cranial meningiomas by far at the largest sample size in the Chinese population. We performed genome-wide genotyping for all these samples and identified diverse common CNVs. Along with detailed clinical information, we investigated their relations with gender difference, tumor location, grade classification, and recurrence, and we further proposed candidate predictors for tumor recurrence.

Sample Collection
This study was approved by the Institutional Review Board of Beijing Tiantan Hospital affiliated with Capital Medical University. Three hundred and eighty-three frozen meningioma samples were collected at Beijing Tiantan Hospital, Capital Medical University, between August 2008 and August 2017. Signed informed consent forms were acquired from all patients or their guardians before surgery. Tumor specimens from meningioma samples were stored in liquid nitrogen immediately following collection. Genomic DNA was purified from tumor samples using a Biomek 3000 automated workstation with an E.Z.N.A Mag-Bind Tissue DNA Kit (Omega Bio-Tek, Norcross, GA, USA). DNA quality and quantity were determined using a NanoDrop 1000 instrument (Thermo Scientific, Wilmington, DE, USA).

Clinical Data Collection and Follow-Up
Clinical information for 383 patients, including gender, age, primary or recurrent, degree of resection, tumor location, tumor diameter, bone invasion, peritumoral brain edema, pathological subtype, WHO grade, and follow-up results (recurrence and survival), was collected and summarized in Table 1. Pathological diagnosis was reviewed according to the 2016 WHO classification for meningiomas. Tumor recurrence was defined as tumor reemergence after GTR (gross total resection), or tumor regrowth with a minimum change of 25% increase of any tumor diameter after non-GTR based on contrast-enhanced MRIs (27). The degree of resection was decided according to the criteria of Simpson grading and classified as GTR (Simpson grade I to III) or STR (subtotal resection), verified by postoperative magnetic resonance images (MRIs) (28). Recurrence-free survival was defined as the period from the time of present surgery in our hospital to tumor recurrence (or last follow-up visit).

Whole-Genome Single-Nucleotide Polymorphism (SNP) Genotyping and Statistical Analysis
Whole-genome SNP array analysis for 383 meningioma samples was performed on Illumina Human Infinium CoreExome BeadChips (Illumina, San Diego, CA, USA). Raw intensity values were processed to obtain a normalized B allele frequency (BAF) and a log R ratio (LRR) for each probe using the GenomeStudio Software v2.0.4 (Illumina, San Diego, USA). LRR values were segmented with Genome Alteration Detection Analysis (GADA, Juan R. González, Barcelona, Spain) using parameters of T > 10 and segment lengths containing ≥50 continuous probes. For loss of heterozygosity (LOH) analysis, the sliding window approach was used with a window size of 100 informative SNPs. A window was considered to represent LOH if more than 80% of the SNPs had a minor allele frequency ≤0.9. A segment was defined either as normal or as having one of 3 types of alteration status based on the following criteria: (1) normal, |LRR| < 0.075 and retaining heterozygosity; (2) gain, LRR ≥ 0.075; (3) loss, LRR ≤ −0.075; and (4) copy-number neutral loss of heterozygosity (CNNLOH), |LRR| < 0.075.
To assess genome instability, the genomic fractions of CNVs, and CNNLOH were estimated by dividing the number of SNPs undergoing a specific alteration by the total number of SNPs present in the respective chromosome or in the respective sample. To identify minimal common regions (MCRs) of copy-number gains and losses, the Genomic Identification of Significant Targets in Cancer (GISTIC, Broad Institute, Boston, USA) algorithm was utilized. Thresholds of LRR were set at 0.1 and −0.1 to allow GISTIC to identify amplifications and deletions, respectively. Q-values of minimal common regions <0.01 were defined as significant, and 0.99 was used as the confidence level to determine regions that contained potential driver genes. For genes within candidate CNV markers, the differential gene expression analysis was performed using NCBI GEO2R for the dataset GSE74385 in Gene Expression Omnibus (GEO) (29); the survival analysis for diverse tumors using the gene expression data and clinical information in The Cancer Genome Atlas (TCGA) project via the portal UALCAN (30).

Statistics
Chi-squared tests and Wilcoxon rank-sum tests were performed using R scripts. The two-sided significance level was set at p ≤ 0.01, two-sided. Gene set enrichment analysis (GSEA) was performed using online tools (31), and the two-sided significance level was set at q ≤ 0.05. Prognostic analyses were conducted using Kaplan-Meier analysis and a Cox proportional regression model. Before conducting the prognostic analysis, patients with history of surgery, or with STR, or with postoperative radiotherapy were excluded. The two-sided significance level was set at p ≤ 0.05.

Identification of Independently Significant Prognostic CNVs in Predicting Tumor Recurrence
Based on common CNVs, we tried to predict the tumor recurrence. After excluding patients with recurrent lesions, with subtotal resection, or having postoperative radiotherapy, 267 patients were included for further prognostic analysis, and the detailed clinicopathological features of this subcohort are shown in Table 1. In the follow-up (mean period, 60 months), 12 patients suffered from tumor recurrence. All common CNV regions and clinical features were included in univariate Cox analysis of tumor recurrence. As shown in Table 6  for tumor recurrence. In particular, most significant independent risk factors for recurrence were loss of 4p16.3 (p = 0.009, HR = 5.69, multivariate Cox analysis) and 10p11.22 (p = 0.037, HR = 4.53). As shown in Figure 4, patients with losses of both 4p16.3 and 10p11.22 were more likely to suffer from tumor recurrence than patients with loss of either one, or patients with neither of these CNV changes. Calculated by Cox analysis, the hazard ratio (HR) increased by 5.10 (95% CI: 2.35-11.08, p = 3.7 × 10 −5 ) for each additional prognostic CNV. Eight genes located within these two CNLs: ZNF141, ABCA11P, ZNF595, ZNF721, ZNF718, ZNF876P, ZNF732 in 4p16.3, and CCDC7 in 10p11.22 (Table S1). In differential gene expression analysis between non-recurrent (grade I, 13 and grade II, 6) and recurrent (grade I, 7 and grade II, 8) lesions in a public gene expression dataset of meningioma (Method), ZNF141 and ZNF595 showed the tendency to have lower expression levels in recurrent samples (unadjusted p < 0.05, Table S1). We further examined the effects of expression levels of these eight genes on survival time for 31 tumor types in TCGA (Method) and identified 28 associations with significance (p < 0.05, Figure S1). All of these genes had lower expression levels in certain tumor types from patients with shorter survival, which indicated their decreased functions related to malignant phenotypes. Particularly, most of these genes (five out of eight) had the same effects on patient survival in head and neck squamous cell carcinoma (HNSC), and the low expression of ZNF718 (p = 0.0027), CCDC7 (p = 0.01), ZNF141 (p = 0.012), ZNF721 (p = 0.029), and ZNF732 (p = 0.045) all demonstrated significant associations with the shorter survival of patients.

DISCUSSION
In the present study, we clarified the CNV characteristics of cranial meningiomas in 383 Chinese patients. Particularly, we compared the CNV burdens of meningiomas in diverse phenotypes. We found more CNVs in the samples of highgrades, recurrent lesions, tumor diameter over 4.3 cm, and samples from male patients. Meanwhile, CNV burden may not relate to tumor locations, peritumoral brain edema, bone invasion, and multiple lesions. Moreover, we also identified featured CNVs in each clinical group. Besides, we found two candidates as independent prognostic CNVs in predicting tumor recurrence.
The CNVs frequently identified in patients with distinct clinical features hold clues for further functional studies. For instance, two CNLs of 1p31.3 and 1p34.3, commonly seen in meningiomas of high-grade and recurrent or large lesions, contain a lot of genes with functional importance. At 1p34.3, the SFPQ gene participates in transcriptional regulation, DNA double-strand break repairs, and suppression of RNA:DNA-hybrid-related telomere instability (33,34). At 1p31.3, the USP1 gene, involved in multiple DNA repair pathways, can function as a key senescence regulator controlling genomic integrity (35); autophagy protein ATG4C participates in controlling the unregulated cell growth (36). Reduced levels of autophagy have been described as being linked to malignant tumors (37). Functional changes related to these genes may also contribute to the progression of meningiomas, which needs further studies for validation.
The CNG at 10q23.31 was the only CNV more commonly seen in multiple meningiomas rather than in single lesions. It covers only one gene, KIF20B, an oncogene involved in cytokinesis. A recent study suggested to target the KIF20B gene in the treatment for hepatocellular carcinoma (38). Inhibition of KIF20B can block mitosis at both metaphase and telophase, which enhance the cytotoxicity of two chemotherapeutic drugs, hydroxycamptothecin, and mitomycin C (39). The role of KIF20B in tumorigenesis of meningiomas, especially multiple lesions, suggests that its suppression might be a novel strategy in the treatment for multiple meningiomas in the future. Moreover, the CNG at 6p21.33, more frequently found in lesions from male patients, is where HLA-B and HLA-C are located, indicating the existence of immune factors underlying gender difference of meningioma occurrence; the CNG at 20q13.33, more frequently identified in patients with peritumoral brain edema, covers the SYCP2 gene, which is related to the depth of cervical invasion in squamous cell carcinoma (40).
Tumor recurrence is an important issue for patients with meningiomas, and patients with meningiomas prone to recurrence need adjuvant radiation therapy or close follow-up. Meanwhile, patients with low risk of tumor recurrence could be spared from the toxicity of radiation therapy. Nevertheless, these patients are not accurately identified by WHO grading (41). Here, we demonstrated the potential of CNV profiling in recurrence prediction. Loss of 1, 4, 9, and 10p and gain of 1q or other chromosomal regions have been revealed to be risk factors for tumor recurrence in previous studies (7-11, 14, 21, 24). In our observation, the CNLs of 4p16.3 and 10p11.22 were independent risk factors for cranial meningioma recurrence. The CNL of 4p16.3 covers MiR-571, ABCA11P, ZNF141, ZNF595, ZNF718, ZNF721, ZNF732, and ZNF876P. A recent study identified miR-571 as the first miRNA that prevents aberrant DNA replication, and the Cdk2-c-Myc-miR-571 axis was identified as a new pathway for regulating DNA replication, cell cycle, and genomic stability in cancer cells (42). As a result, loss of miR-571 may lead to genomic instability. Although some studies have reported differential expression or mutation occurrence of ZNF595 (43), ZNF721 (44), ZNF718 (45), and ZNF141 (46), their functions remain unclear. Besides, potential roles of ABCA11P, ZNF732, and ZNF876P are novel in meningioma recurrence. In the 10p11.22, CCDC7, also known as Biot2, highly expressed in CD133-positive stem cells, functions as a risk factor for poor prognosis in colorectal cancer (47,48). In our study, the CNL at 10p11.22 (CCDC7) was an independent risk factor of tumor recurrence, and the underlying mechanisms need further investigation.
The cross-sectional analysis in the entire cohort compared primary and recurrent lesions from different groups of patients, and some primary tumors may also harbor CNVs contributing to tumor recurrence. It may undermine the ability to identify CNVs related to recurrence, which may explain the missing of the CNLs of 4p16.3 and 10p11.22 in the comparison. Meanwhile, the comparison results may be cofounded by differential CNVs present in the early stage of tumor development between two groups of patients. Therefore, the follow-up study provides us an opportunity to identify those CNVs related to recurrence. The recurrence rates of patients with these two CNVs were over 20% (loss of 4p16.3, 21%, 4/19; loss of 10p11.22, 27%, 3/11), significantly higher than the recurrent rate (about 4%) in patients without them. Nevertheless, only 12 patients (4.5%, 12/267) had tumor recurrence during a mean follow-up period of 5 years in our subcohort for prognostic analysis. Although it is similar to previous observation, which is 3% for WHO grade I meningiomas and 30% for WHO grade II meningiomas in GTR patients (28), the prediction power of these two candidate markers requires further evaluation in a larger group of patients with tumor recurrence in the future follow-up. Besides, the recurrence factors may have heterogeneity, and 4p16.3 and 10p11.22 together accounted for the 13% (loss of 4p16.3, 4/45; loss of 10p11.22, 4/45, losses of both, 2/45) recurrent lesions in the cross-sectional analysis. It needs further efforts to dissect other CNVs related to tumor recurrence.

CONCLUSIONS
Based on a large number of patients with cranial meningiomas, we identified that the CNVs of 22q, 1p, and 14q were the most prevalent. Meningiomas of high WHO grades, recurrent tumors, large size, and male gender were likely to have more CNVs, especially of large size (>500 kB). Additionally, the CNLs at 4p16.3 and 10p11.22 were promising candidates as independent risk factors for tumor recurrence prediction.

DATA AVAILABILITY STATEMENT
The datasets in this study have been publicly deposited. They can be accessed at: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi? acc=GSE147673.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by Institutional Review Board of Beijing Tiantan Hospital Affiliated with Capital Medical University. Signed informed consent forms were acquired from all patients or their guardians.