H3K27M-mutant diffuse midline gliomas should be further molecularly stratified: an integrated analysis of 669 patients

H3K27M-mutated diffuse midline gliomas (H3-DMGs) are aggressive tumors with a fatal outcome. This study integrating individual patient data (IPD) from published studies aimed to investigate the prognostic impact of different genetic alterations on survival of these patients. We accessed PubMed and Web of Science to search for relevant articles. Studies were included if they have available data of follow-up and additional molecular investigation of H3-DMGs. For survival analysis, Kaplan–Meier analysis and Cox regression models were utilized, and corresponding hazard ratios (HR) and 95% confidence intervals (CI) were computed to analyze the impact of genetic events on overall survival (OS). We included 30 studies with 669 H3-DMGs. TP53 mutations were the most common second alteration among these neoplasms. In univariate Cox regression model, TP53 mutation was an indicator of shortened survival (HR 1.446; 95% CI 1.143–1.829) whereas ACVR1 (HR 0.712; 95% CI 0.518–0.976) and FGFR1 mutations (HR 0.408; 95% CI 0.208–0.799) conferred prolonged survival. In addition, ATRX loss was also associated with a better OS (HR 0.620; 95% CI 0.386–0.996). Adjusted for age, gender, and tumor location, the presence of TP53 mutations, the absence of ACVR1 or FGFR1 mutations remained significantly poor prognostic factors. We outlined the prognostic importance of additional genetic alterations in H3-DMGs and recommended that these neoplasms should be further molecularly segregated. This may aid neuro-oncologists in appropriate risk stratification.


Introduction
Diffuse midline gliomas are predominantly seen in children and likely have a lethal prognosis. These tumors typically arise in the pons, thalamus, or spinal cord and have a dismal survival of 12-18 months [1]. About 70-80% of them carry missense mutations in the Histone H3 genes, including the HIST1H3B/C (H3.1), HIST2H3A/C (H3.2), or H3F3A (H3.3) [2,3]. The H3.3 protein share some similarities and differences in functions with its H3.1 and H3.2 counterparts. These proteins maintain the core function of the nucleosome. H3.3 is accumulated into the transcriptionally active regions to replace nucleosomes [4] whereas H3.1 and H3.2 are coupled to DNA synthesis [5,6]. The most common H3 mutations are missense mutations of G34R/V and K27M [2]. These neoplasms are considered World Health Organization (WHO) grade IV, implying their advanced aggressiveness.
Published studies have established an adverse prognosis of midline gliomas harboring H3K27M mutations as compared to wild-type (wt) tumors [2,3,7,8]. In addition, the presence of H3K27M mutations is also an indicator of shortened survival among IDH-wt gliomas and GBMs [9][10][11]. It is still debated as to whether all H3K27M-mutant diffuse midline gliomas (H3-DMG) have a universally poor prognosis, and reports have suggested that these tumors are clinically heterogeneous [8]. It is of clinical importance to further stratify these neoplasms molecularly to better predict patient outcomes and make appropriate therapeutic decisions. In this study, we integrated individual patient data (IPD) from published studies and examined the prognostic impact of various genetic biomarkers in H3-DMGs.

Search term and literature search
We accessed PubMed and Web of Science to search for relevant articles from inception to June 2021 using the following search term: Glioma AND (H3K27M OR H3-K27M OR H3 K27M OR H3F3A OR HIST1H3B OR HIST1H3C). We followed the recommendations of the Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) statement [12].

Selection criteria and abstract screening
Duplicates from the two electronic databases were removed after results are merged and imported into EndNote (Clarivate, PA, USA). Two reviewers independently screened the title and abstract of the articles using the designated selection criteria. Studies were included if they fulfill all following criteria: (i) studies providing IPD of H3-DMG with follow-up data available and (ii) studies with data of at least one genetic event of H3-DMGs available. We excluded studies if: (i) not relevant to research aims; (ii) studies without data of K27M mutations; (iii) case reports; (iv) reviews, theses, or books; (v) conference or proceeding papers; or (vi) studies with duplicated populations.

Full-text screening and data extraction
Two reviewers independently reviewed the full-text of potential studies and extracted data into a standardized excel form. The following IPD were extracted from the articles: authors, institution, country, year of publication, study period, patient identification number, H3 genotypes, detection methods, demographic information, tumor location, WHO grades, treatments, overall survival (OS) time, OS status, and genetic alterations of H3-DMGs. We subsequently removed cases in non-midline locations or cases without follow-up data among the extracted datasets.

Statistical analyses
For studies with potentially overlapping populations, we selected the study with the largest sample size for analysis. Kaplan-Meier analysis and Cox proportional hazards model were conducted to assess the impact of various clinical and molecular parameters on survival of H3-DMGs. The deviance residuals and the dfbeta values were used to examine influential observations. Hazard ratios (HR) are presented as mean and 95% confidence interval (CI). A two-sided p-value of < 0.05 was considered statistically significant. The statistical analyses were performed using the R software, version 3.6.1 (The R Foundation, Vienna, Austria).

Results
We identified 714 articles for the title and abstract screening and included 65 studies for full-text reading. Subsequently, 35 studies were further excluded due to missing survival data, without IPD, without genetic event investigation, or duplicated populations. Finally, we included 30 studies with 669 H3-DMGs for integrated analyses [3,7,8,11, (Fig. S1). The characteristics of 30 included studies are presented in Table 1.

Clinical characteristics of H3-DMGs
The median patient age was 11 years (range 1-82) and there was no gender predilection. Genetically, H3.3 K27M mutation was predominant, followed by H3.1; only 2 H3.2 1 3 K27M mutations were found. Stratified by tumor locations, 91.3% of cases were intracranial midline gliomas with pons and thalamus being the most common sites, while a subset of cases (8.7%) located in the spinal cord. More than 50% of patients died within the first year of follow-up and the median OS was 11.6 months. The clinical features of H3-DMGs are shown in Table S1. Table S2 lists the prevalence of concurrent genetic alterations in H3-DMGs. Mutations in the TP53 genes were the most common genetic event, accounting for more than 50% of cases. Other common genomic events included ATRX loss/mutations, PDGFRA amplification, ACVR1, PIK3CA, FGFR1, and PPM1D mutations, and MGMT methylation. Other rare somatic mutations with a prevalence of less than 10% were BRAF, NF1, PDGFRA, PIK3R1, PTEN, and TERT mutations. Genetic variations in the EGFR were also uncommon. Table S3 shows the associations of H3-DMG genetic alterations with H3 genotypes (H3.1 and H3.3). ACVR1 and PIK3CA mutations more commonly arose with H3.1 mutation whereas ATRX and TP53 mutations, and PDG-FRA amplification were more frequently associated with H3.3 mutation. Table S4 presents the concurrent occurrence rate of TP53 mutations with other genetic alterations. ACVR1, FGFR1, and PPM1D mutations were most likely mutually exclusive with TP53 mutations whereas PDGFRA alterations were likely to concomitantly occur with mutations of the TP53 gene.

Impact of genetic events on the prognosis of H3-DMGs
There was no statistical difference in OS of patients with H3. 3 (Table S5). Adding TP53, ACVR1, FGFR1 mutations, and ATRX loss alternately into a multivariate Cox regression model adjusted for age, gender, tumor locations and the extent of resection, the presence of TP53, ACVR1, or FGFR1 mutations remained independent prognostic markers for patient survival (Table 2). However, ATRX loss had no effect on patient outcome in the multivariate analysis (HR 0.624; 95% CI 0.344-1.133; p = 0.121).

Discussion
Diffuse midline gliomas are lethal neoplasms because of their infiltrative nature and difficulty to achieve gross tumor removal as well as resistance to conventional therapy. Over the past few decades, patient survival remains unchanged despite recent advances in understanding their biology and genetic landscape. Based on their aggressiveness, diffuse midline gliomas can range from WHO grade II to IV, but they have always been considered grade IV clinically because of their universally poor prognosis [39,40]. More recently, landmark studies demonstrated that histone mutations are the most common genetic marker of these tumors and are associated with a poor outcome regardless of their primary location [2,41,42]; there is also increasing evidence that H3K27M-mutant gliomas are clinically and prognostically heterogeneous [8,19,43]. Therefore, further segregation of H3K27M tumors into subgroups would facilitate a refined assessment of natural history and risk profiling. Our study demonstrated several genetic markers that are of clinical importance.
TP53 mutations, encoding the TP53 protein, have long been known to be the driver mutation of many cancers, Number at risk including brain tumors [44]. Notably, TP53 and H3K27M mutations are prevalent among diffuse midline gliomas and GBMs and more than half of H3-DMGs concurrently harbor TP53 mutations [2,36]. TP53 mutations were found to be associated with H3.3 rather than H3.1 K27M mutations [8,36,45]. Experimental studies of diffuse intrinsic pontine gliomas (DIPG) demonstrated that H3K27M mutations are early recurrent mutations and subsequently accompanied by an obligate partner mutation of either the growth factor signaling (e.g., RAS and PI3K), RB1, or TP53 pathways [46]. Furthermore, studies reported that mutations in the TP53 tumor suppressor gene could play a role in tumor resistance against radiotherapy in DIPG cell models [47]. Clinically, TP53 mutations were found to worsen survival of DIPG patients and reduce response to radiation treatment compared to TP53-wt DIPGs [47]. However, the impact of the parallel occurrence of TP53 and H3K27M mutation in midline gliomas warrants further investigation. Several studies have reported a potential link of TP53 mutations with adverse prognosis in H3K27M-mutated tumors [8,36,43] whereas other studies failed to establish this association [2,3,19,32,33]. Our study confirmed the association of TP53 mutations with poor survival of H3-DMGs.
In contrast to TP53 aberration, mutations in ACVR1 are more frequently associated with H3.1 K27M mutations and lower-grade histology [25,30,33,43]. ACVR1 and FGFR1 mutations have been proved to be independent predictors of prolonged survival in midline gliomas [16,30]. These two mutations are almost mutually exclusive with TP53 mutations but could co-occur with H3 mutations in a subset of midline gliomas [33,43]. The concomitant occurrence of H3 and FGFR1/ACVR1 may complicate the diagnostic decisions towards an aggressive or favorable tumor, particularly in small biopsies in which the tumor tissue is too small for low-and high-grade discrimination. Therefore, further investigation into the effects of ACVR1/FGFR1 alterations in H3-DMGs is essential. This study integrating 264 and 153 analyzed cases for ACVR1 and FGFR1 alterations demonstrated that these two mutations were correlated with a superior outcome. Because TP53, ACVR1, and FGFR1 mutations mostly do not co-occur with each other, we did not analyze them in a multivariate model. Multivariate analyses further confirm the independent prognostic effects of these mutations regardless of age, gender, and tumor locations. These findings would be helpful for further risk stratification of midline gliomas and may open up potential possibilities of targeted therapies. Our results support additional risk stratification of H3-DMGs to accurately refine the prognosis of this entity. The rapid development of next-generation sequencing and other advanced molecular techniques could help improve patient care and finalize the standard treatments for these malignant gliomas. ATRX deficiency is a common event in IDH-mutated but rarely occurs in IDH-wt cortical gliomas [48]. Astrocytic tumors with ATRX loss have a significantly better prognosis than patients who express ATRX [48]. In IDH-wt diffuse midline gliomas, however, ATRX loss is more frequently observed in about 25-40% of cases [8,34]. Our results showed that the presence of ATRX loss alone was also a positive marker of midline glioma survival. However, after adjusting for confounding factors, we could not reproduce the significant association of this biomarker and patient survival. A few studies reported PDGFRA amplification as a marker for unfavorable outcome of diffuse midline glioma [19,25]. Our integrated analysis revealed a similar trend but our result did not reach statistical significance. Other mutations of the MAPK pathway including BRAF mutation have been correlated with favorable survival in glioma patients [49,50]. BRAF mutation is very rare among H3-DMGs and we did not uncover any meaningful association of this mutation with patient survival. Unlike in IDH-wt cortical GBMs [51][52][53], EGFR alterations and TERT promoter mutations were uncommon in midline gliomas and had no impact on patient survival. Genetic alterations of the PI3K/AKT pathway such as PIK3CA, PIK3R1, or PTEN mutations showed no effects on patient outcome. Castel et al. reported a prolonged survival of H3.1 K27M-mutant tumors in comparison to H3.3-mutant but this finding was not confirmed in our study [8].
This integrated analysis demonstrated the prognostic importance of various genetic biomarkers in a large population of H3-DMGs. In the 2021 WHO classification of central nervous system tumors, the following key biomarkers have been recommended to better stratify the survival of H3-DMGs including TP53, ACVR1, PDG-FRA, EGFR, and EZHIP [54]. This recommendation is primarily based on published institutional results and our pooled data from 30 studies demonstrated that the prognostic values of PDGFRA and EGFR alterations are not significant. On the other side, we highlighted that FGFR1 mutation and ATRX loss could be used to further stratify the H3-DMGs. However, our study has limitations. Firstly, most of the included studies relied on retrospective patient selection so we could not avoid the risk of selection bias stemming from these studies. Next, important clinical covariates such as radiation and chemotherapy administration, and chemotherapy regimen detail were missing in over 50% of the included studies so we could not include Number at risk them in the multivariate models. Finally, the prevalence and identification of genetic markers could be affected by the application of different detection methods among the included studies. Prospective multicenter studies with a large sample size are needed to confirm our study findings and refine H3-DMGs into different prognostic subgroups.
In summary, our study demonstrated several important molecular markers that can influence the survival outcome of H3-DMGs including TP53, ACVR1, FGFR1 mutations, and ATRX loss. Our results suggest that these aggressive gliomas do not have a universally dismal survival, and that routine molecular profiling may help inform our understanding of tumor behavior.