Patient demographics

We identified 87 patients with a pathological diagnosis of gliosarcoma in our institutional cohort. Of these, 51 had known treatment information while the remainder were lost to follow-up after surgery. 62.1% were male. Median age at diagnosis was 60.4 years (IQR 16.2 years; Table 1). The median overall survival (OS) was 14.0 months, and 64.4% of patients were deceased at the time of data collection. Of those patients with known baseline Karnofsky Performance Score (KPS; 46% of all patients), 92.5% had a KPS of 80 or above. Of the 40.2% of patients for whom extent of resection was known, 42.9% underwent subtotal resection while 51.4% received a gross total resection. Most (52.9%) patients were not tested for MGMT methylation status. Primary and secondary gliosarcoma did not differ significantly in sex, KPS, MGMT status, or extent of resection, though age at diagnosis was younger in patients with secondary gliosarcoma (Supplemental Table 1).

Table 1 Demographic characteristics of patients

We also identified 492 IDH-wildtype glioblastoma patients. There were no significant differences between the gliosarcoma and glioblastoma patients in regards to age, extent of resection, or MGMT status, although the proportion of patients with known extent of resection and MGMT status was greater for glioblastoma (Table 1). However, patients with glioblastoma had a lower baseline KPS overall (p < 0.0001), with 25.6% of patients having a KPS under 80.

Treatment characteristics

Of the 87 gliosarcoma patients, 51 patients had known treatment status (58.6%). All 51 patients received at least one course of radiation therapy (RT), and all but one patient received temozolomide (TMZ) at least once (Table 1). The median number of adjuvant cycles of TMZ received was five. Gliosarcoma and glioblastoma did not significantly differ in the proportion of patients who received first-line RT or TMZ, although glioblastoma patients were less likely to receive TMZ at least once. Of the 405 (82.3%) glioblastoma patients with available treatment history, 93.8% received first line RT and 86.4% received first-line TMZ. When treated, the median number of cycles of TMZ were the same for patients with glioblastoma as that of gliosarcoma (5). There was insufficient data to compare subsequent lines of therapy including targeted therapies.

Genomic landscape of gliosarcoma

We examined the distribution of commonly altered genes in gliosarcoma. Among the 38 gliosarcoma samples that were sequenced with the NGS Solid Tumor Panel, 18 genes had an alteration frequency of greater than 10% (Fig. 1a). Over half of the patients in our cohort had alterations in PTEN (63%), TERT (55%), or TP53 (55%). CDKN2A and CDKN2B were co-deleted in 6 patients (16%). Of these 6 patients, 5 also possessed a MTAP co-deletion. No tumors harbored pathogenic IDH1 or IDH2 mutations.

While the majority of sequenced gliosarcoma cases were identified as primary gliosarcomas (n = 29), several were identified at tumor recurrence as secondary gliosarcomas that had displayed other histology at initial diagnosis (n = 7). All but one of the secondary gliosarcomas had been diagnosed with glioblastoma at the time of initial encounter. The remaining patient was initially diagnosed with a high-grade CNS embryonal tumor, NOS. There were no significant differences in the frequency of common alterations between primary and secondary gliosarcomas (Supplemental Fig. 6).

Fig. 1

Genomic features of gliosarcomas. (A) Co-mutation plot for all gliosarcoma samples with genomic profiling data (n=38) grouped by gliosarcoma type (primary vs. secondary). Genes altered in greater than 10% of samples and IDH1/2 canonical alterations are shown in descending order of frequency. (B) Stacked barplots comparing gene alteration frequencies between glioblastoma (GBM) and gliosarcoma (GSM) for genes altered in greater than 10% of gliosarcoma samples. Benjamini-Hochberg adjusted p-values < 0.05 are displayed

We next evaluated whether the pattern of common genomic alterations in gliosarcoma was different from glioblastoma (Fig. 1b). We observed several differences. Notably, EGFR was more commonly altered in glioblastoma (p-adj = 0.00055) whereas NF1, PTEN, and TP53 were more commonly altered in gliosarcoma (p-adj = 0.00055, 0.047, and 0.0035, respectively). In contrast to other published gliosarcoma cohorts, we identified no BRAF alterations in our cohort [8, 9, 15,16,17, 20, 21].

Survival characteristics in patients with gliosarcoma

We evaluated the prognostic impact of a histopathologic diagnosis of gliosarcoma. There were no statistically significant differences in OS between gliosarcoma and glioblastoma patients (p = 0.95), although primary gliosarcoma exhibited inferior progression-free survival compared to glioblastoma (p = 0.0099, Fig. 2, Supplemental Fig. 2). Additionally, there were no differences in OS between primary and secondary gliosarcoma patients when evaluating survival from time of first cancer diagnosis (p = 0.38; Supplemental Fig. 3). We observed that male sex and older age were associated with a worse outcome (p = 0.065, HR = 2.4; p = 0.0053, HR = 1.1, respectively; Fig. 3a, Supplemental Table 2). In our cohort, there was no survival impact of MGMT status, KPS, extent of resection, or primary versus secondary gliosarcoma, likely due to the limited sample size (Supplemental Table 2).

Fig. 2

Overall survival in gliosarcoma (GSM) versus glioblastoma (GBM). Kaplan-Meier and log-rank test of patients who received temozolomide or radiotherapy as part of first-line treatment for each group are included. Median overall survival is represented as dashed lines for each group

We then examined the association between specific genomic alterations and outcomes in our institutional cohort. We evaluated genes that were altered in more than 10% of patients. Among the 18 genes interrogated, only CDKN2A/CDKN2B/MTAP alterations showed a significant association with prognosis on univariate analysis (p = 0.023, HR = 5.4). TSC2 alterations showed a trend towards being protective, but did not meet statistical significance (p = 0.12, HR = 0.31). These effects for CDKN2A/CDKN2A/MTAP (p = 0.043, HR = 4.8) and TSC2 (p = 0.14, HR = 0.20) remained when controlling for MGMT methylation status, a well-known prognostic marker in glioblastoma and likely gliosarcoma, in a subgroup of 30 patients with known MGMT status [22, 23]. RB1 alterations (p = 0.036, HR = 0.21) were also significantly associated with improved survival when controlling for MGMT status (Supplemental Fig. 4).

Impact of molecular alterations on survival in Project GENIE gliosarcomas

Recognizing that limited sample size impacted our analysis, we sought to validate our findings using a larger, independent cohort of gliosarcoma patients from AACR Project GENIE. We identified 93 gliosarcoma patients with both survival and sequencing data. This cohort did not differ significantly from our institutional cohort with regards to age, sex, or survival status (Supplemental Table 3). Extent of resection, KPS, MGMT status, and treatment information were not available. There was no significant difference in OS between our institutional and GENIE cohort using estimated OS for the GENIE cohort as described above (Supplemental Fig. 5).

Genomic alterations were similar between the institutional and GENIE cohort. As in our institutional cohort, TERT, PTEN, TP53, and NF1 were the topmost altered genes in the GENIE cohort (Supplemental Fig. 6a). RB1, CDKN2A, and CDKN2B were also altered at high frequencies in both cohorts. There were no significant differences between the two cohorts in the alteration frequencies of the most commonly altered (> 10% of samples tested) genes in either cohort (Supplemental Fig. 6b).

Univariate Cox proportional hazards modeling was performed on the GENIE data set for age, sex, and the 18 genes identified in our institutional cohort. There was no effect from age or sex, unlike the institutional cohort. CDKN2A/B (p = 0.096, HR = 1.9) and LRP1B (p = 0.043, HR = 2.7) were associated with worse OS, while RB1 (p = 0.015, HR = 0.44) were associated with improved prognosis (Fig. 3b, Supplemental Table 4). Of note, the direction of effect was preserved between the two cohorts for multiple genes.

Fig. 3

Genetic factors impacting survival in gliosarcoma. (A-C): Univariate Cox proportional hazards results for select variables for the (A) institutional, (B) Project GENIE, and (C) pooled cohorts. Hazards ratios (HR) and 95% CI displayed for all comparisons. (C-D): Kaplan-Meier curves and log-rank test for pooled gliosarcoma patients from both cohorts stratified by CDKN2A/B, LRP1B, RB1, or TSC2 alteration status. Median overall survival is represented as dashed lines

Impact of molecular alterations on survival in gliosarcoma in pooled data

We next performed pooled analysis using both cohorts to determine the effect on univariate cox analysis (Fig. 3c, Supplemental Table 5). As expected, CDKN2A/B (p = 0.039, HR = 1.8), RB1 (p = 0.016, HR = 0.51), and LRP1B (p = 0.050, HR = 2.0), which were significant in at least one of the cohorts and had consistent directionality, were significantly associated with survival. CDKN2A/B loss and LRP1B alterations were associated with inferior survival, while RB1 alterations were protective. Furthermore, TSC2, which was not significant in either cohort individually, was significantly associated with better prognosis in the combined data (p = 0.048, HR = 0.31). Kaplan-Meier curves of the pooled patients stratified by CDKN2A/B, LRP1B, RB1, and TSC2 alteration status are shown in Fig. 3d-g.

We also fit multivariate Cox proportional models to the pooled data to determine whether associations between CDKN2A/B, LRP1B, RB1, and TSC2 and survival remained after controlling for demographic variables. The effects of each gene were preserved when individually fit with age, sex, and cohort of origin (i.e. either institutional or GENIE); CDKN2A/B (p = 0.021, HR = 2.2) and LRP1B (p = 0.024, HR = 2.2) remained significantly associated with shorter survival, while RB1 (p = 0.011, HR = 0.47) and TSC2 (p = 0.030, HR = 0.27) remained associated with improved survival (Supplemental Fig. 7).