Introduction

Diffuse gliomas with histologically lower-grade features (hLGG) are heterogeneous tumors with different clinical outcomes. Some patients may experience long stable disease, while others rapidly deteriorate with fatal outcomes. Classifications based on histological findings have been used to discriminate subgroups with distinct prognosis outcomes. Molecular studies have reported hLGG sub-classification into different molecular subgroups [13, 14, 17, 55]. The first group is IDH mutant-type (IDHm) and chromosome 1p/19q co-deleted tumors which correspond to oligodendrogliomas. These tumors generally have good prognoses. The second group is IDHm with intact 1p/19q and alterations in ATRX and TP53, corresponding to astrocytomas. In contrast, IDH wild type (IDHw) hLGG are molecularly and clinically distinct [2, 13, 14, 55].

In adults, IDHw gliomas may contain several pediatric-type tumors [31]; these include biologically more favorable glial and glioneuronal tumors with BRAF alterations, those gliomas with altered MYB-/MYBL-1, and high-grade gliomas with H3F3A mutation or RTK2. The other IDHw hLGG, which mainly arise from the adult supratentorial brain, frequently have molecular features of glioblastomas; TERT promoter mutation (pTERTm), epidermal growth factor receptor amplification (EGFRamp), or chromosome 7 gain and 10 loss (CH7/10). New WHO guidelines have categorized such groups as glioblastoma, IDH wild type [11]. IDHw hLGGs with molecular features of glioblastomas are typically known as molecular glioblastoma (mGBM).

Recent studies have revealed that mGBM has specific clinical features; tumors often have diffuse infiltration patterns (gliomatosis), multicentric loci or gyriform spread [39, 57]. Consequently, patients are likely to undergo biopsy rather than resection. Furthermore, they were less likely to receive adjuvant radiochemotherapy. Thus, while most studies have suggested the prognostic correspondence between histological glioblastoma (hGBM) and mGBM, the equivalence has not been sufficiently demonstrated. Reports have indicated that mGBM may reflect under-sampling phenomena or early-stage hGBM [49, 65]. For the former, treatment outcomes are worse if adjuvant radiochemotherapy is withheld, while for the latter, the outcomes depend on lead-time to hGBM development.

Molecular marker frequency in IDHw hLGGs differ extensively between studies. pTERTm rates vary from 15 to > 70% in IDHw hLGGs [33, 57], while EGFRamp rates vary from 10 to > 50% [19, 34]. It is unclear if such differences are caused by methodological or other factors, however, differences may affect tumor classification and prognostication.

As IDHw hLGGs are relatively rare, large-scale studies are scarce. We systematically reviewed IDHw hLGG studies and performed a meta-analysis to explore the variability underlying molecular marker frequency and their prognostic implications for mGBM.

Methods

We conducted this study according to Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines. It was not registered on The International Prospective Register of Systematic Reviews.

Literature search and data extraction

A PRISMA flow diagram is outlined (Fig. 1). From 2009 onward, we searched for relevant English studies using the keywords “glioma,” “astrocytoma,” AND “IDH” OR “isocitrate dehydrogenase” AND “TERT promoter” OR “EGFR” OR “chromosome 7” OR “chromosome 10” in PubMed, Ovid MEDLINR, Cochrane library and Scopus. Literature searches and data extraction were independently conducted by two study authors. EndoNote® (version 20) was used for data processing. Adult hemispheric IDHw LGG studies including ≥ 20 cases were incorporated if molecular testing in at least one of the followings was included; EGFRamp, pTERTm or CH7/10. Exclusion criteria: (1) molecular data could not be extracted, (2) studies with IDH1 data alone, (3) public database studies (e.g. The Cancer Genome Atlas, etc.), or (4) familial or pediatric cancers. For studies from the same institute, newer or relevant studies were selected. Finally, studies were selected by authors through discussion and consensus.

Fig. 1
figure 1

PRISMA flow diagram and search strategy

Full size image

We collected patient age (mean or median) and sex information, molecular data, treatment-related factors, including tumor removal degree and adjuvant therapy, and overall survival (OS). Molecular diagnosis methods and tissue types (formalin-fixed paraffin-embedded (FFPE) or fresh-frozen (FF)) were also recorded. Studies were discriminated into Asian or non-Asian regions. When only median values and ranges were available, the mean and standard deviation were calculated using the method of Hozo et al. [26]. Individual patient data (IPD) were recorded if available. We re-calculated relevant statistical data from IPD, whereas tumors with H3F3A or BRAF mutations were excluded where possible. We used hazard ratio (HR) to determine treatment effects. When HRs were unavailable, we calculated values according to the method by Tierney et al. [59] or from Kaplan–Meier curve data. We used WebPlotDigitizer (https://automeris.io/WebPlotDigitizer/) to extract survival data from Kaplan–Meier curves [43, 44].

Statistical analysis

We used the R software program (v4.03) (https://www.r-project.org/) to perform statistical analyses (Meta, Metafor and EZR packages). An inverse-variance approach, with a random-effects method, was applied. Studies were assessed for heterogeneity (I2 statistic). A single-arm meta-analysis was performed to assess specific mutation rates using the metaprop function with a generalized linear mixed model. Meta-regression analyses were performed to identify heterogeneity related factors. Analyses were performed using the rma.uni function in Meta software package. In Kaplan–Meier curves from reconstructed data, log-rank and Cox-proportional hazard tests were used for survival analyses. Covariates of age, sex, type of surgery, adjuvant-treatment type, Karnofsky performance status score (KPS), and promoter methylation status of O6-methylguanine-DNA methyltransferase (MGMT), in addition to molecular status, were included in multivariate analyses when applicable. Two-sided P < 0.05 values were statistically significant.

Risk of bias

We used the Joanna Briggs Institute’s (JBI) critical appraisal checklist for case series (Supplementary Table 1) (https://jbi.global/critical-appraisal-tools) to evaluate risk of bias. Publication bias was evaluated using a funnel plot including > 10 studies. A linear regression analysis was also performed.

Ethics approval and informed consent

This review did not involve direct human investigations; therefore, no informed consent was required.

Results

We performed our literature search on September 4th, 2022 and retrieved 212 studies out of 688 records for full-text assessment (Fig. 1). We excluded 163 studies (lack of relevant data = 64, < 20 case = 29, public database only = 27, review articles = 9, hGBM only = 8, overlapping data = 19, and other reasons = 7). Finally, 49 studies were selected: 6 studies for comparing mGBM with hGBM [3, 6, 15, 37, 50, 58], 33 studies for calculating molecular marker frequency [2, 4, 7,8,9,10, 12, 16,17,18,19,20, 23, 24, 29, 30, 33,34,35,36, 38,39,40, 45, 48, 51,52,53,54, 56, 60, 61, 63, 66], and 8 studies for both [25, 28, 32, 42, 46, 47, 57, 64]. Two studies were utilized for analyses of prognostic factor in mGBM [21, 62]. IPD were collected from 12 studies [2, 4, 6, 15, 19, 21, 23, 25, 51, 61, 62, 64].

Risk of bias

Almost all studies were retrospective case series’ with low evidence levels. The JBI checklist for case series showed that most of them had one or two deficits (Supplementary Table 1).

mGBM marker frequency in IDHw hLGGs

In our analyses, 33 studies reported pTERTm rates in IDHw hLGG (Supplementary Table 2) [2,3,4, 8,9,10, 12, 18, 19, 24, 25, 28,29,30, 32, 33, 36, 38,39,40, 42, 45,46,47,48, 51, 52, 56, 57, 61, 63, 64, 66]. The pooled pTERTm proportion was 47.5% ([95% confidential interval (CI) 39.8–55.3]) (N = 2044, I2 = 86.2%) (Table 1) (Fig. 2A). A funnel plot showed no asymmetry (P = 0.34) (Fig. 2B). In subgroup analyses, the pooled proportion was significantly lower in studies from Asia when compared with non-Asian regions (36.2% [CI: 29.8–43.1] vs. 56.2% [CI: 45.2–66.5], P = 0.002), and significantly higher in studies using next generation sequencing (NGS) when compared with other methods (Sanger sequencing, Pyrosequencing or others) (68% [CI: 48.5–82.8] vs. 42.3% [CI: 35.5–49.3], P = 0.015), and significantly higher in studies using FFPE when compared with other methods (FF or either FFPE or FF) (30 studies; N = 1904: 56.2% [CI: 45.1–66.7] vs. 38.7% [CI: 28.5–50.0], P = 0.03). Meta-regression analyses also showed that the pooled pTERTm proportion increased by mean patient age (P = 0.02) and was possibly affected by grade III tumor rates (G3rate) (P = 0.068) (Table 1). Multiple meta-regression analyses showed that only regions (Asia or non-Asia) reached significance (P = 0.03) but FFPE exhibited borderline significance (P = 0.08). However, multiple meta-regression analyses still showed high residual heterogeneity (I2 = 84.8%). We hypothesized that undetermined factors may have affected our results. Such factors may have affected pTERTm rate in 1p/19q co-deleted gliomas (Codel). Indeed, we identified a significant correlation between pTERTm rates in IDHw LGG and Codel tumors (R = 0.49, P = 0.012) (Supplementary Fig.S1A). Moreover, a quantile–quantile (Q-Q) plot showed that distribution of pTERTm rates in Codel was normal when we excluded cases with rate < 90% (Supplementary Fig.S1B). For this reason, we selected 19 articles from institutes which reported ≥ 90% pTERTm rates in Codel for multiple meta-regressions. Then, the geographical region (P < 0.001) and method (NGS vs others, P = 0.002) were significant for the pTERTm proportion, and heterogeneity disappeared (I2 = 0%). The pooled pTERTm rate in IDHw hLGG was 41.1% [CI: 33.4–49.4] (7 studies, N = 335) in Asia but 64.1% [CI: 54.1–73.0] (12 studies, N = 588) in non-Asia regions (P < 0.001).

Table 1 Frequency of molecular markers in IDH wild-type histologically lower-grade gliomas
Full size table
Fig. 2
figure 2figure 2

A Forest plot of proportion for TERT promoter mutation (pTERTm) in IDH wild-type lower-grade gliomas. Subgroup analyses showing a higher frequency in Asian regions (P = 0.0024). B Funnel plot showing no asymmetry. C Forest plot showing molecular glioblastomas (mGBM) proportions for pTERTm rates. Subgroup analysis presenting a higher frequency in Asian regions (P = 0.002). D Kaplan–Meier curves from individual patient data showing overall survival in patients with mGBM and IDH wild-type lower-grade gliomas “not elsewhere classified” (NEC), in Asian and non-Asian regions

Full size image

EGFRamp rates in IDHw hLGGs (N = 2225) (Table 1) were reported by 27 studies [2,3,4, 7, 9, 16, 17, 19, 20, 25, 28, 32, 34, 36, 38, 39, 42, 45, 47, 51,52,53,54, 56, 57, 60, 61]. The pooled EGFRamp proportion was 23.6% [CI: 18.6–29.4] (I2 = 83.1%). We did not identify publication bias (P = 0.28). It was low in Asia studies when compared with other regions (15.9% [CI: 10.6–23.2] vs. 28.8% [CI: 22.9–35.5], P = 0.008). Methodologies (NGS vs others (FISH, MLPA, array CGH etc.)) did not affect result (P = 0.43), but tissue type did and was significant (FFPE 29.6% [CI: 22.2–38.3] vs FF and FF or FFPE 17.1% [CI: 10.3–27.1], P = 0.047). Meta-regression analyses showed that G3rates were significant factors (P = 0.035), but mean age was not (P = 0.23). Multiple meta-regression analyses also revealed that geographical region (P = 0.01), tissue type (P = 0.028) and G3rates (P = 0.037) were significant factors impacting EGFRamp rates (I2 = 67.6%).

CH7/10 status was evaluated in 13 studies (N = 1194) (Table 1) [4, 9, 19, 23, 32, 35, 36, 38, 51, 54, 56, 57, 61], although definition were different (e.g. whole CH7/10, 7p( +)/10q(−), 7( +)/10q(−), or EGFR gain/PTEN loss) (Supplementary Table 1). The pooled proportion in IDHw hLGG was 33.2% [CI: 21.5–47.4] (I2 = 89.8%). A Funnel plot showed no publication bias (P = 0.20). The rate was also lower in Asian regions (15.0% [CI: 6.7–30.4]) when compared with non-Asian regions (50.9% [CI 42.9–58.9]) (P = 0.0003). Methodologies (NGS vs. others) did not affect rates (P = 0.07), but tissue type did (FFPE 53.9% [CI: 44.6–63.0] vs. FF and FFPE or FF 24.7% [CI: 12.4–43.4] (P = 0.007). Meta-regression analyses showed that mean age (P = 0.03) and G3rate (P = 0.046) were significant factors. Multiple meta-regression analyses also revealed that geographical region (P < 0.0001), tissue type (P < 0.0001) and G3rate (P = 0.0028) significantly affected the CH7/10 rates. When evaluation was limited to studies using whole CH7/10 [9, 23, 32, 35, 36, 54, 56, 57], pooled proportion in IDHw hLGG was 34.8% [CI: 20.0–53.2], which is comparable to the abovementioned value. In this case, subgroup and meta-regression analyses did not reach significance except in terms of mean age (P < 0.001); however, they showed a tendency similar to the described analyses.

The prognostic implications of difference in mGBM marker rates

We calculated the pooled mGBM rate in IDHw hLGG studies examining pTERTm. The pooled mGBM rate in Asian regions (38.9% [CI: 33.4–44.8], 14 studies, N = 868) was lower than that in non-Asia regions (61.9% [CI: 49.0–73.2], 19 studies, N = 1215) (P = 0.0015) (Fig. 2C). Interestingly, the rate increased in studies examining multiple molecular markers (three markers 79.0% [CI: 68.9–86.5], two markers = 78.0% [CI: 50.3–92.5] vs. one marker (pTERTm) = 45.1% [CI: 34.2–56.4]) in non-Asian regions (P < 0.0001), while rates did not change in Asian regions (three markers = 39.6% [CI: 28.2–52.2], two markers = 32.0% [CI: 25.4–39.4], and one marker (pTERTm) = 41.1% [CI: 34.3–48.3]) (P = 0.38). This was true when we analyzed studies showing > 90% pTERTm rates in Codel tumors (Asian regions: P = 0.94 and non-Asian regions: P < 0.0001). Therefore, pTERT wild-type hLGGs in Asian regions rarely expressed other GBM molecular markers (4.1% when analyzed in IPD), whereas those in non-Asian did.

As molecular marker analyses showed, studies using FFPE showed a significantly higher mGBM rates (62.5% [CI: 49.5–74.0]) when compared with those using other tissue (FF or either FF or FFPE) (41.4% [CI: 31.4–52.1]) (P = 0.014).

We compared OS rates of IDHw hLGG tumors between Asian and non-Asian studies using IPD. The median OS rate in Asian mGBM tumors (21.1 months [CI: 16.1–24.7], N = 185) was similar to in non-Asian tumors (20.8 months [CI: 17.9–22.7], N = 275) (P = 0.45) (Fig. 2D). The median OS rate of IDHw hLGG tumors, “not elsewhere classified” (NEC) in Asian regions (38.1 months [30.4–46.8], N = 304), was also similar to non-Asian regions (35.2 months [CI: 26.4–53.3], N = 123) (P = 0.6).

Differences between mGBM and hGBM

mGBM and hGBM survival data were provided by 14 studies (Table 2) [3, 6, 15, 25, 28, 32, 37, 42, 46, 47, 50, 57, 58, 64]. Five showed that biopsied tumors were more frequent for mGBM when compared with hGBM [6, 37, 50, 57, 58] (P < 0.001 in each study, Fisher’s exact tests), while two studies reported comparative total removal rates between the two [15, 32]. Eight studies described adjuvant radiochemotherapy frequency rates [6, 15, 32, 37, 42, 50, 57, 58]. Four studies reported that the patients with mGBM underwent radiotherapy less often when compared with patients with hGBM. Three studies reported that the patients with mGBM underwent chemotherapy less often when compared with patients with hGBM. Treatments frequency differences in other studies failed to reach significance.

Table 2 Clinical and molecular features of IDH wild-type histological and molecular glioblastomas and “not elsewhere classified” lower-grade gliomas
Full size table

While most studies showed that mGBM patients had comparable OS rates when compared with hGBM patients, the pooled HR value of mGBM against hGBM was 0.82 in univariate analyses ([CI: 0.69–0.98], I2 = 46%) (P = 0.028) (Fig. 3A). No publication bias was detected (P = 0.92) (Fig. 3B). Moderate heterogeneity improved in the pooled HR in multivariate analyses (including covariates of age (five studies), sex (four studies), surgery (four studies), radiotherapy and/or chemotherapy (three studies), KPS (three studies), MGMT (two studies) and CDKN2A/B (one study)), which showed a significantly better OS rate in mGBM (HR 0.61 [CI: 0.50–0.74], I2 = 0%) (P < 0.0001) (Fig. 3C). We identified no differences in pooled HR values between Asian and non-Asian regions (P = 0.80). Kaplan–Meier curves from IPD showed an identical OS rate between hGBM and mGBM with grade III histology (P = 0.39), but better OS for mGBM with grade II histology (P = 0.014) (Fig. 3D).

Fig. 3
figure 3figure 3

A Forest plot showing hazard ratio (HR) values from overall survival analyses (univariate) between patients with molecular glioblastoma (mGBM) and histological glioblastoma (hGBM). B Funnel plot showing no asymmetry. C Forest plot (from multivariate analyses) showing significantly lower HR values in mGBM against hGBM (P < 0.0001), with low heterogeneity. *data from individual patient data (IPD):**extracted data from Kaplan–Meier curves: †data from https://www.surgery.cuhk.edu.hk/BTC/HSBC/molecular_grading_paper_data.pdf. D Kaplan- Meier curves of IPD showing better OS rates for mGBM with grade II histology when compared with mGBM with grade III histology or hGBM

Full size image

Prognostic factors in mGBM

We meta-analyzed prognostic factors related to OS rates in mGBM (16 studies) (Table 3) [3, 4, 9, 15, 19, 21, 25, 32, 38, 46, 47, 50, 51, 61, 62, 64]. A younger age (HR1.028 [CI: 1.011–1.045], P = 0.001) and extent of surgery (HR 0.643 [CI: 0.446–0.926], P = 0.018) were identified as significantly better prognostic factors. Although no molecular markers reached significance, patients with histologically grade III gliomas had worse OS rates when compared with grade II tumors in pooled results from both univariate (HR 1.58 [CI: 1.02–2.429], P = 0.003) and multivariate analyses (HR 1.314 [CI: 1.041–1.660], P = 0.022) (Supplementary Fig. S2A, B).

Table 3 Pooled results of prognostic factors in overall survival of molecular glioblastomas
Full size table

Discussion

mGBM has significantly better OS rates when compared with hGBM. Additionally, mGBM with grade II histological features had better OS rates when compared with that with grade III histological features. Although mGBM rates in IDHw hLGG varied extensively among studies, differences were partly explained by geographical regions. Rates in Asia were significantly lower when compared with non-Asian regions. Interestingly, pTERT wild-type tumors in Asia rarely expressed other molecular markers, however, despite these differences, mGBM in Asian and non-Asian regions showed identical OS rates using IPD.

Diagnosis and frequency of mGBM

We identified one reason which possibly explained mGBM rate differences across studies, i.e. the geographical region. While low pTERTm frequency rates in glioblastomas in Asia were previously reported [5], this was also true for IDHw LGGs. Not only pTERTm, but also EGFRamp and CH7/10 also occurred less frequently in Asian regions. As lower-grade glioma incidences in Asia were lower when compared with eastern countries [41], mGBM incidence was assumed to be lower. We observed differences in molecular markers combination patterns between Asian and non-Asian regions. Fujimoto et al. [19] reported that pTERT mutation status was necessary and sufficient to diagnose IDHw hLGG with molecular features of glioblastoma. This appeared to be true for Asian but not in non-Asian regions. However, if only the test for pTERTm is used, approximately 4.1% of mGBM may be missed from the analyses even in Asia.

One substantial molecular diagnostic issue in this field is that TERT promotor areas have high guanine-cytosine content and easily form secondary structures which lead to a poor amplification [27]. Thus, a diagnosis may be misleading when tumor cell density is low. While it is generally accepted that FF tissues are better materials, our meta-regression analysis showed that pTERTm rates were higher in FFPE samples. EGFRamp and CH7/10 detection rates had similar frequencies. This was most likely due to sampling selection spots for molecular diagnostics. IDHw hLGGs sometimes grow invasively and adopt gliomatosis cerebri forms. In such cases, tumor cells may be sparse in tumor areas by preoperative imaging. A more appropriate area for molecular diagnoses may be selected in FFPE rather than FF samples. While different methodologies have different molecular diagnostic sensitivities [1, 22], specimen condition, especially tumor cell density, appears to be more important. To exclude the problems in material and methodological issues, we re-analyzed pTERTm data in studies that showed pTERTm > 90% of Codel tumors. This generated a small change in the pTERTm rates in mGBM but our overall conclusions were unaltered.

Prognostic factors in mGBM

Although mGBM in most studies exhibited similar OS rates to hGBM, pooled HR analyses showed better OS rates in mGBM when compared with hGBM. Furthermore, pooled HR values from multivariate analyses, including a covariate analysis of age, sex and surgery, showed a much better OS rates with lower heterogeneity (HR 0.61 [CI: 0.50–0.74], I2 = 0%). Berzero et al. [9] reported better OS rates in mGBM with grade II histology when compared with grade III, although their study lacked multivariate analyses. As mGBMs with grade III features showed worse OS rates when compared with grade II (HR 1.582 [CI: 1.030–2.429], P = 0.003), differences in OS rates between mGBM and hGBM appeared to be due to the grade II subgroup in mGBM, as indicated by Kaplan–Meier curves (Fig. 3D).

However, even mGBM in grade II histology sometimes shows rapid growth with new ring-like contrast enhancement by magnetic resonance imaging [28, 39], and hGBM transformation at recurrence [49, 52]. Thus far, specific molecular abnormalities showed no significant difference in OS rates [21, 50]. We also showed that no molecular markers significantly affected prognoses other than histological grade. The prognostic significance of the extent of mGBM surgery remains controversial. Ruda et al. [52] identified no survival benefit from gross total resection of mGBM with grade II histology (they did not include stereotactic biopsy cases). Ramos-Fresnedo et al. [50] did not demonstrate OS differences between biopsy and GTR (P = 0.079). Nevertheless, we showed that pooled HR favored more extensive surgery (HR 0.644 [CI: 0.420–0.990], P = 0.045). Zhang et al. [65] reported that the prospective identification of mGBM resulted in more aggressive patient management and improved clinical outcomes when compared with a biologically matched historical control patient cohort receiving standard-of-care therapy based on histomorphologic diagnoses alone. However, the effects of early aggressive treatment for mGBM must be examined in future studies.

Study limitations

Our study had several limitations. Firstly, most studies were retrospective and had low evidence levels. Consequently, heterogeneity was high in pooled data and multiple meta-regression analyses were required to identify reasons for this. Secondly, while we largely excluded tumors with BRAF or H3F3A mutation, some studies presented no information. Most of the studies had no data regarding other pediatric-type gene changes, such as altered MYB/MYB-L1. Therefore, NEC numbers in IDHw hLGG may have been overestimated in some studies. Thirdly, in some cases, we extracted survival data from Kaplan–Meier curves in figures of articles. Although we confirmed overlapping of original and extracted curves, measurement errors in graph line thickness may have inadvertently occurred. Fourthly, some multi-institutional studies may have had overlapping data with other studies. While we carefully selected studies, some duplicates data may have been included. Potential data overlaps are recorded in supplementary Table 1.

Conclusion

Patients with mGBM have better OS rates when compared with patients with hGBM, especially when grade II histology is indicated. We could not identify other molecular markers that differentiated mGBM prognosis, although patient age and surgical extent were prognostic factors. We identified differences between Asian and non-Asian regions in terms of molecular marker frequency pattern for mGBM in IDHw hLGG. However, OS rates in patients with mGBM were concordant between Asian and non-Asian cohorts. Although a considerable number of “NEC” in IDHw hLGGs were identified especially in Asian patients, further studies are warranted for their classification. Controversies continue to persist in the new WHO classification of IDHw hLGG. DNA methylation profiling is an evolving method that will further facilitate classification [31]. Prospective clinical studies in IDHw hLGG with DNA methylation profiling are necessary in the future.