Abstract
Background
Glioblastoma isocitrate dehydrogenase (IDH)-wildtype (GBM) has been reclassified based on molecular and phenotypic features. Intratumoral hemorrhage is a phenotypic subtype with poorly understood molecular and clinical characteristics. We aimed to characterize the molecular profile and outcomes of hemorrhagic glioblastoma (hGBM) compared with non-hGBM.
Methods
A retrospective analysis of GBMs with pre-operative and post-operative MRI and comprehensive next-generation sequencing of 205 genes was performed. Patients were classified as hGBM or non-hGBM using the Visually Accessible Rembrandt Images criteria. Univariable and multivariable survival analyses were performed. Genes were compared using the Fisher-exact test and corrected for multiple comparisons with the Benjamini-Hochberg method.
Results
176 patients were included, of whom 105 had hGBM. Compared to non-hGBMs, hGBMs were more likely Hispanic (20.0% vs. 7.0%, p = 0.018), presented with motor deficits (38.0% vs. 21.1%, p = 0.020), and had larger tumors (147.6 cm3 vs. 122.2 cm3, p = 0.034). In addition, hGBMs harbored less frequently PDGFRA (13.3% vs. 26.8%, p = 0.031), KIT (8.5% vs. 19.7%, p = 0.040), KDR (6.6% vs. 18.3%, p = 0.027), or PIK3R1 (2.8% vs. 11.3%, p = 0.030) alterations, and a higher frequency of SETD2 alterations (8.5% vs. 1.4%, p = 0.050). These differences were not significant after multiple comparison corrections. No differences in progression-free survival (8.2 vs. 8.3 months, p = 0.337) or overall survival (18.4 vs. 19.7 months, p = 0.800) were identified by hemorrhagic status.
Conclusions
hGBMs might be associated with lower frequencies of PDGFRA, KIT, KDR, and PIK3R1 alterations and higher frequencies of SETD2 alterations compared to non-hGBMs. No outcome differences were observed by intratumoral hemorrhage status in this cohort.
Supplementary Information
The online version contains supplementary material available at https://doi.org/10.1007/s11060-026-05749-3.
Keywords: GBM, Hemorrhagic, PDGFRA, KIT, KDR, PIK3R1
Introduction
Glioblastoma isocitrate dehydrogenase (IDH)-wildtype (GBM) is the most common and aggressive primary central nervous system (CNS) tumor, affecting 3–4 per 100,000 people annually [1, 2]. Despite maximal safe resection and chemoradiation with temozolomide (TMZ), GBM has a poor prognosis [3]. Although remarkable advances in genetic and epigenetic research have shaped glioma classification [4, 5], the relationship between genomic alterations, phenotypic tumor characteristics, and outcomes remains poorly typified. This gap is particularly evident in one frequent radiographic phenotype, intratumoral hemorrhagic GBM (hGBM).
hGBM has been reported in 2.3% to 29.2% of patients with GBM [6–8]. GBM’s hallmarks, microvascular proliferation, endothelial hyperplasia, and areas of necrosis, have been proposed to lead to vascular instability [9]. Similarly, blood–brain barrier disruption has been postulated to contribute to extravasation of blood products into the tumor parenchyma [10]. Recently, a series of 167 glioma patients showed that higher frequencies of CDKN2B, KMT5B, and PIK3CA alterations were harbored in hemorrhagic tumors; however, this study was limited by its heterogeneous population [6]. In addition, the prognostic importance of hGBM remains inconsistent across studies [6, 11].
In this study, we evaluated whether hemorrhagic MRI phenotype correlates with genetic alterations and affects survival in a large, homogeneous, newly diagnosed GBM cohort. We hypothesize that hGBM is associated with a distinct genomic signature compared with non-hGBM.
Methods
Study population
A retrospective review of patients with pathology-proven GBM from 2009 to 2020 was performed. The inclusion criteria were: (1) molecular GBM diagnosis, with cases retrospectively reclassified according to the WHO 2021 classification for CNS [5], (2) age > 18 years, (3) next-generation sequencing (NGS) analysis available, and (4) availability of pre- and post-operative MRI. Patients who did not meet these criteria were excluded from the study. Intratumoral hemorrhage was defined according to the Visually Accessible Rembrandt Images (VASARI) criteria as the presence of intrinsic hemorrhage within the tumor matrix on preoperative MRI, identified by intrinsic foci of T2-weighted hypointensity or T1-weighted hyperintensity [12–14]. Diffusion-weighted imaging (DWI) and apparent diffusion coefficient (ADC) maps were reviewed when necessary to further characterize these signal abnormalities and exclude imaging artifacts. Data were collected from our local hospital’s electronic medical records and managed using a REDCap tool hosted at the University of Texas Health Science Center at Houston [15, 16]. The data included sex, age, race, comorbidities, laboratory parameters at admission, including prothrombin time (PT), activated partial thromboplastin time (aPTT), international normalized ratio (INR), and platelet count, along with Karnofsky performance status (KPS), presenting symptoms, tumor location, pre- and post-operative tumor volumes, adjuvant treatment, recurrence, and salvage treatments. Molecular characteristics collected include NGS information and O6-Methylguanine-DNA-methyltransferase (MGMT) promoter methylation status. The study was approved by the institutional review board (HSC-MS-17-1917), and written informed consent was waived because of the retrospective design, following the 1964 Declaration of Helsinki and its later amendments [17]. This study adheres to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [18].
Brain MRIs were independently evaluated by two blinded readers, including a board-certified neuroradiologist with more than 20 years of experience and a neuroradiology fellow. Disagreements were resolved by consensus. The extent of resection (EOR) was determined using the RANO resect classification system [19]. Volumetric analysis of the tumors was performed using TeraRecon Aquarius software (San Mateo, California) on a separate workstation. Tumor components analyzed included: total tumor volume, enhancing volume, necrotic volume, and perilesional Fluid-attenuated inversion recovery (FLAIR) signal abnormality volume [20].
Next-generation sequencing
Genetic alterations were analyzed from tumor samples using the FoundationOne NGS panel that consistently tested 205 genes and introns of 28 genes in rearrangements (FoundationOne, Foundation Medicine, Inc) [21]. TERT promoter (TERTp) status was available for 135 of the 176 patients.
Statistical analysis
Fisher’s exact test was used to compare categorical variables, and given that the study cohort was not normally distributed, the Mann–Whitney U test was used for nonparametric continuous variables. Only genes with a count of 5 or more mutations within the cohort were evaluated. Survival outcomes were compared using a univariable log-rank test, with Kaplan–Meier curves plotted. Only patients who received chemoradiotherapy with temozolomide were included in the survival analysis. The primary outcome was overall survival (OS). Secondary outcomes included progression-free survival (PFS) and post-recurrence survival (PRS). OS was defined as the time from the day of radiographic diagnosis to death or last follow-up. PFS was defined from the day of radiographic diagnosis to the date of radiographic progression. PRS was defined as the time from radiographic progression to death or last follow-up. Hazard ratio (HR) estimates were calculated using univariable and multivariable Cox proportional hazard regression models with a 95% confidence interval (CI). Multivariable Cox proportional hazard regression model analysis for OS was adjusted for variables with a p-value of 0.05 or less in univariable analysis, along with variables with a well-known effect on survival, such as age, KPS, EOR, chemoradiotherapy with TMZ, and contrast-enhancing tumor volume [22]. Multiple comparisons adjustment was performed with a false discovery rate (FDR) using the Benjamini–Hochberg formula. Genetic visualization and survival data were performed utilizing the OncoPrinter and statistical tools in cBioPortal [23]. All other statistical analyses were performed using EZR v1.40 (Saitama Medical Center, Jichi Medical University) [24].
Results
Cohort characteristics
A total of 176 patients with GBM met the inclusion criteria (Online resource 1). The median age was 61 years (IQR 53–68). Most patients were males (n = 104, 59.1%) and white (n = 120, 68.2%). The median pre-operative KPS and post-operative KPS were 70 (IQR 70–80) and 80 (IQR 70–90), respectively. The most common tumor location was the temporal lobe (n = 71, 40.3%), followed by the frontal lobe (n = 55, 31.3%). There was a slight predominance in right-sided tumors (n = 95, 54.0%). The most common presenting symptom was headache (n = 77, 43.7%), followed by confusion (n = 74, 42.0%). The median total tumor volume was 137.5 cm3 (IQR 75.0–193.7), with a median contrast-enhancement volume of 19.2 cm3 (IQR 9.4–32.9), a median FLAIR volume of 105 cm3 (IQR 58.1–148.0), and a median necrotic volume of 6.2 cm3 (IQR 2.0-12.8). The presence of hGBM was noted in 105 (59.6%) patients (Fig. 1). Treatment with TMZ was administered to 159 (90.3%) patients, with 156 (88.6%) receiving concomitant radiotherapy. Two patients met the criteria for RANO Class 1; therefore, due to the limited sample size, these cases were combined with RANO Class 2 for subsequent analyses. After the merge, 96 (54.6%) patients underwent Class 1 or 2 resection, 76 (43.1%) underwent Class 3 resection, and 4 (2.3%) underwent Class 4 (biopsy) resection (Table 1).
Fig. 1.
Comparison of MRI sequences between GBM IDH–wildtype with intratumoral hemorrhage (A–D, red arrows) and without intratumoral hemorrhage (E–H). Sequences include T1-weighted (A and E), T2-weighted (B and F), DWI (C and G), and ADC (D and H)
Table 1.
Baseline demographic, clinical, and radiological features of patients with glioblastoma isocitrate dehydrogenase-wildtype with and without intratumoral hemorrhage
| Variable, n (%) | All patients (n = 176) | hGBM (n = 105) | Non-hGBM (n = 71) | p-value |
|---|---|---|---|---|
| Age, median (IQR) | 61 (53–68) | 59 (52–65) | 63 (54–70) | 0.036 |
| Male | 104 (59.1) | 60 (57.1) | 44 (62) | 0.530 |
| Race | ||||
| White | 120 (68.2) | 67 (63.8) | 53 (74.6) | 0.141 |
| African American | 19 (10.8) | 11 (10.4) | 8 (11.2) | 1.000 |
| Hispanic | 26 (14.7) | 21 (20) | 5 (7) | 0.018 |
| Asian | 9 (5.1) | 5 (4.76) | 4 (5.6) | 1.000 |
| Others | 2 (1.1) | 1 (0.9) | 1 (1.4) | 1.000 |
| Social | ||||
| Smoking history | 60 (34.1) | 34 (32.4) | 26 (36.6) | 0.629 |
| Alcohol use | 74 (42) | 43 (41) | 31 (43.7) | 0.876 |
| Substance use | 9 (5.1) | 7 (6.7) | 2 (2.8) | 0.314 |
| Medical history | ||||
| Hypertension | 101 (57.4) | 58 (55.2) | 43 (60.6) | 0.640 |
| Diabetes mellitus | 39 (22.1) | 24 (22.8) | 15 (21.1) | 0.854 |
| Dyslipidemia | 69 (39.2) | 38 (36.1) | 31 (43.7) | 0.431 |
| Ischemic stroke/TIA | 15 (8.5) | 9 (8.6) | 6 (8.4) | 1.000 |
| Hemorrhagic stroke | 6 (3.4) | 6 (5.7) | 0 (0) | 0.082 |
| Anemia | 13 (7.4) | 7 (6.7) | 6 (8.4) | 0.772 |
| Heart failure | 8 (4.5) | 5 (4.8) | 3 (4.2) | 1.000 |
| Arrhythmias | 7 (3.9) | 5 (4.8) | 2 (2.8) | 0.702 |
| Coronary artery disease | 15 (8.5) | 8 (7.6) | 7 (9.8) | 0.784 |
| Myocardial infarction | 5 (2.8) | 3 (2.8) | 2 (2.8) | 1.000 |
| Peripheral vascular disease | 8 (4.5) | 3 (2.8) | 5 (7) | 0.274 |
| Medication use history | ||||
| Antihypertensive | 84 (47.7) | 50 (47.6) | 34 (47.8) | 1.000 |
| Statins | 58 (32.9) | 33 (31.4) | 25 (35.2) | 0.744 |
| Antiplatelets | 36 (20.4) | 19 (18.1) | 17 (23.9) | 0.447 |
| Anticoagulation | 5 (2.8) | 1 (0.9) | 4 (5.6) | 0.160 |
| DVT prophylaxis (pharmacological) | 30 (17) | 18 (17.1) | 12 (16.9) | 1.000 |
| Lab values at admission | ||||
| PT (IQR) | 13.40 (12.90–13.90) | 13.40 (12.90–13.90) | 13.50 (12.90–13.90) | 0.644 |
| aPTT (IQR) | 28.00 (25.20–30.60) | 28.00 (26.20–30.20) | 28.20 (24.87–30.80) | 0.923 |
| INR (IQR) | 1.01 (0.96–1.05) | 1.01 (0.96–1.05) | 1.01 (0.97–1.05) | 0.650 |
| Platelet count (IQR) | 229 (190–267) | 235(188–272) | 217 (192–259) | 0.615 |
| Pre-Op KPS, median (IQR) | 70 (70–80) | 80 (70–80) | 70 (70–80) | 0.552 |
| Post-Op KPS, median (IQR) | 80 (70–90) | 80 (70–90) | 80 (70–90) | 0.534 |
| Presenting symptoms | ||||
| Confusion | 74 (42) | 50 (47.6) | 24 (33.8) | 0.086 |
| Dizziness | 14 (8) | 6 (5.7) | 8 (11.2) | 0.256 |
| Gait problems | 29 (16.4) | 20 (19) | 9 (12.6) | 0.305 |
| Headache | 77 (43.7) | 50 (47.6) | 27 (38) | 0.220 |
| Language deficit | 69 (39.2) | 34 (32.3) | 35 (49.2) | 0.028 |
| Memory loss | 22 (12.5) | 11 (10.4) | 11 (15.5) | 0.358 |
| Motor deficit | 55 (31.25) | 40 (38) | 15 (21.1) | 0.020 |
| Seizures | 46 (26.1) | 24 (22.8) | 22 (31) | 0.294 |
| Sensory deficit | 41 (23.2) | 28 (26.6) | 13 (18.3) | 0.210 |
| Visual deficit | 37 (21) | 23 (21.9) | 14 (19.7) | 0.851 |
| Vomiting | 13 (7.4) | 8 (7.6) | 5 (7) | 1.000 |
| Tumor location | ||||
| Frontal | 55 (31.3) | 32 (30.5) | 23 (32.4) | 0.869 |
| Temporal | 71 (40.3) | 41 (39) | 30 (42.2) | 0.754 |
| Parietal | 33 (18.8) | 22 (21) | 11 (15.5) | 0.433 |
| Occipital | 5 (2.8) | 4 (3.8) | 1 (1.4) | 0.649 |
| Insular | 6 (3.4) | 3 (2.8) | 3 (4.2) | 0.686 |
| Tumor side | ||||
| Left | 75 (42.6) | 41 (39) | 34 (47.9) | 0.278 |
| Right | 95 (54) | 61 (58.1) | 34 (47.9) | 0.281 |
| Midline/Bilateral | 6 (3.4) | 3 (2.8) | 3 (4.2) | 0.686 |
| CE volume, median (IQR) | 19.2 (9.4–32.9) | 20.0 (11.9–35.2) | 13.7 (6.8–31.5) | 0.007 |
| FLAIR volume, median (IQR) | 105.0 (58.1–148.0) | 110.3 (70.4–150.0) | 92.2 (45.3-142.5) | 0.067 |
| Necrotic volume, median (IQR) | 6.2 (2.0-12.8) | 7.2 (2.9–14.2) | 4.2 (0.8–10.1) | 0.012 |
| Tumor volume, median (IQR) | 137.5 (75.0-193.7) | 147.6 (97.0-195.1) | 122.2 (62.1-180.9) | 0.034 |
| MGMT status | ||||
| Methylated | 28 (16) | 17 (16.2) | 11 (15.5) | 0.832 |
| Unmethylated | 48 (27.2) | 34 (32.4) | 14 (19.7) | 0.084 |
| Not available | 100 (56.8) | 54 (51.4) | 46 (64.8) | 0.089 |
| EOR | ||||
| Class 1 and 2 | 96 (54.6) | 57 (54.2) | 39 (54.9) | 1.000 |
| Class 3 | 76 (43.1) | 47 (44.7) | 29 (40.8) | 0.643 |
| Class 4 | 4 (2.3) | 1 (0.9) | 3 (4.2) | 0.304 |
| Adjuvant therapies | ||||
| Chemoradiotherapy with TMZ | 156 (88.6) | 93 (88.6) | 63 (88.7) | 1.000 |
| Radiotherapy only | 4 (2.3) | 2 (1.9) | 2 (2.8) | 1.000 |
| TTF | 19 (10.8) | 13 (12.3) | 6 (8.4) | 0.467 |
| Bevacizumab | 14 (7.9) | 7 (6.6) | 7 (9.8) | 0.572 |
| Virus injection | 2 (1.1) | 2 (1.9) | 0 (0) | 0.516 |
| Immunotherapy | 0 (0) | 0 (0) | 0 (0) | 1.000 |
| Others† | 15 (8.5) | 9 (8.6) | 6 (8.4) | 1.000 |
| Recurrence | 135 (76.7) | 83 (79) | 52 (73.2) | 0.467 |
| Treatment at recurrence | ||||
| Reoperation | 49 (36.3) | 30 (36.1) | 19 (36.5) | 1.000 |
| Salvage TMZ | 64 (47.4) | 40 (48.2) | 24 (46.1) | 0.861 |
| Salvage irinotecan | 39 (28.9) | 25 (30.1) | 14 (26.9) | 0.846 |
| Salvage bevacizumab | 84 (62.2) | 51 (61.4) | 33 (63.4) | 0.857 |
| Salvage TTF | 29 (21.5) | 20 (24.1) | 9 (17.3) | 0.396 |
| Radiotherapy/SRS | 63 (46.6) | 38 (45.7) | 25 (48.1) | 0.860 |
| Best supportive care | 14 (10.4) | 9 (10.8) | 5 (9.6) | 1.000 |
| Others§ | 11 (8.1) | 7 (8.4) | 4 (7.7) | 1.000 |
%: Percentage; aPTT: activated partial thromboplastin time; CE: Contrast enhancing; DVT: Deep vein thrombosis; EOR: Extent of resection; FLAIR: Fluid-attenuated inversion recovery; hGBM: hemorrhagic glioblastoma; INR: International normalized ratio; IQR: Interquartile range; KPS: Karnofsky Performance Scale; MGMT: O-6-Methylguanine-DNA methyltransferase; n: number; PT: Prothrombin time; SRS: Stereotactic radiosurgery; TIA: Transient ischemic attack; TMZ: Temozolomide; TTF: Tumor-treating fields
†Others: Depatuxizumab mafodotin; ICT-107; Lapatinib; Plerixafor; Ribociclib; Trametinib; Veliparib; Verubulin. Others§: Adegramotide/Nelatimotide; Carmustine; Depatuxizumab mafodotin; Everolimus; ICT 107; Immunotherapy; Intrathecal chemotherapy; Lomustine; Nivolumab; Plerixafor; SVN53-67/M57-KLH; TPI 287; Virus injections. P values in bold indicate statistical significance (p ≤ 0.05)
Twenty-four genetic alterations with an incidence greater than or equal to 5 in this cohort were examined. The 10 most common genetic alterations across the entire cohort were TERTp (n = 111/135, 82.2%), CDKN2A/B (n = 117, 66.4%), PTEN (n = 78, 44.3%), EGFR (n = 77, 43.7%), TP53 (n = 66, 37.5%), PDGFRA (n = 33, 18.7%), NF1 (n = 30, 17%), PIK3CA (n = 25, 14.2%), RB1 (n = 25, 14.2%), and KIT (n = 23, 13.1%), Table 2.
Table 2.
Molecular differences among Glioblastoma Isocitrate Dehydrogenase-Wildtype with and without Intratumoral Hemorrhage
| Gene, n (%) | All patients (n = 176) | hGBM (n = 105) | Non-hGBM (n = 71) | p-value |
|---|---|---|---|---|
| CDKN2A/B | 117 (66.4) | 70 (66.6) | 47 (66.2) | 1.000 |
| TERT (promoter only) | 111 (82.2) | 69/84 (82.1) | 42/51 (82.3) | 1.000 |
| PTEN | 78 (44.3) | 48 (45.7) | 30 (42.2) | 0.757 |
| EGFR | 77 (43.7) | 45 (42.8) | 32 (45.1) | 0.877 |
| TP53 | 66 (37.5) | 40 (38.1) | 26 (36.6) | 0.875 |
| PDGFRA | 33 (18.7) | 14 (13.3) | 19 (26.8) | 0.031 |
| NF1 | 30 (17) | 17 (16.2) | 13 (18.3) | 0.838 |
| PIK3CA | 25 (14.2) | 12 (11.4) | 13 (18.3) | 0.271 |
| RB1 | 25 (14.2) | 15 (14.3) | 10 (14.1) | 1.000 |
| KIT | 23 (13.1) | 9 (8.5) | 14 (19.7) | 0.040 |
| CDK4/CDK6 | 22 (12.5) | 15 (14.2) | 7 (9.8) | 0.488 |
| KDR | 20 (11.3) | 7 (6.6) | 13 (18.3) | 0.027 |
| MDM4 | 16 (9.1) | 9 (8.6) | 7 (9.8) | 0.794 |
| PIK3C2B | 11 (6.2) | 6 (5.71) | 5 (7) | 0.758 |
| PIK3R1 | 11 (6.2) | 3 (2.8) | 8 (11.3) | 0.030 |
| MDM2 | 10 (5.7) | 7 (6.6) | 3 (4.2) | 0.742 |
| SETD2 | 10 (5.7) | 9 (8.5) | 1 (1.4) | 0.050 |
| BRAF | 9 (5.1) | 5 (4.8) | 4 (5.6) | 1.000 |
| PTPN11 | 7 (4) | 6 (5.71) | 1 (1.4) | 0.244 |
| STAG2 | 6 (3.4) | 3 (2.8) | 3 (4.2) | 0.686 |
| TET2 | 6 (3.4) | 3 (2.8) | 3 (4.2) | 0.686 |
| ATRX | 5 (2.8) | 5 (4.8) | 0 (0) | 0.082 |
| FGFR3 | 5 (2.8) | 4 (3.8) | 1 (1.4) | 0.649 |
| LRP1B | 5 (2.8) | 3 (2.8) | 2 (2.81) | 1.000 |
%: percentage; hGBM: hemorrhagic glioblastoma; n: number. TERT (promoter only) percentages calculated with the 135 patients tested
P values in bold indicate statistical significance (p ≤ 0.05)
Clinical differences among glioblastoma patients with and without intratumoral hemorrhage
Patients with hGBM were younger, with a median age of 59 (IQR, 52–65), compared with 63 (IQR, 54–70) in non-hGBM (p = 0.036), while sex was comparable across the groups. Hispanic ethnicity was higher in hGBM (n = 21, 20.0%) compared with non-hGBM (n = 5, 7.0%), p = 0.018. Comorbidities, prior medications, including antiplatelet (n = 19 [18.1%] vs. n = 17 [23.9%], p = 0.447), anticoagulant (n = 1 [0.9%] vs. n = 4 [5.6%], p = 0.160), and pharmacological DVT prophylaxis (n = 18 [17.1%] vs. n = 12 [16.9%], p = 1.000), as well as admission hematologic laboratory values, including PT (13.4 vs. 13.5, p = 0.644), aPTT (28 vs. 28.2, p = 0.923), INR (1.01 vs. 1.01, p = 0.650), and platelet count (235 vs. 217, p = 0.615), were comparable between patients with hGBM and non-hGBM. hGBMs were associated with larger tumor burden, including greater median total tumor volume (147.6 cm³, IQR, 97.0–195.1 vs. 122.2 cm³, IQR, 62.1–180.9; p = 0.034), higher contrast-enhancing volume (20 cm³, IQR, 11.9–35.2 vs. 13.7 cm³, IQR, 6.8–31.5; p = 0.007), and greater necrotic volume (7.2 cm³, IQR, 2.9–14.2 vs. 4.2 cm³, IQR, 0.8–10.1; p = 0.012), while FLAIR volume was comparable across groups. Other clinical characteristics, including pre-operative and post-operative KPS, tumor location, tumor side, EOR, and receipt of adjuvant therapies, were similar between cohorts (Table 1).
Molecular features of glioblastoma patients with and without intratumoral hemorrhage
In the comparison of the incidence of genetic mutations between cohorts, hGBM tumors showed lower frequencies of PDGFRA (13.3% vs. 26.8%, p = 0.031), KIT (8.5% vs. 19.7%, p = 0.040), KDR (6.6% vs. 18.3%, p = 0.027), and PIK3R1 (2.8% vs. 11.3%, p = 0.030), and a higher frequency of SETD2 (8.5% vs. 1.4%, p = 0.050) compared with non-hemorrhagic tumors (Table 2; Fig. 2). After applying the Benjamini–Hochberg FDR correction for multiple comparisons, these differences in genetic incidences were not sustained. MGMT status did not differ significantly across the groups.
Fig. 2.
Visual summary of cancer-related gene mutations in 105 GBM IDH–wildtype cases with intratumoral hemorrhage and 71 without intratumoral hemorrhage
Effects of intratumoral hemorrhage on glioblastoma survival
Univariable log-rank analysis revealed no differences in PFS (8.2 vs. 8.3 months, p = 0.337; Fig. 3A), PRS (13.3 vs. 13.5 months, p = 0.766; Online resource 2), or OS (18.4 vs. 19.7 months, p = 0.800; Fig. 3B) between patients with hGBM and those without hGBM. In a subanalysis of the 105 patients with hGBM at presentation, after again excluding those who did not receive chemoradiotherapy with TMZ and those who received best supportive care at first recurrence, 70 patients were included. Of these, patients who received salvage bevacizumab at first recurrence (n = 48) had similar PRS (13.4 vs. 16.1 months, p = 0.119; Fig. 3C) and OS (21.6 vs. 24.8 months, p = 0.185; Fig. 3D) compared with those who did not receive bevacizumab (n = 22). Similarly, patients with non-hGBM (n = 43) showed no differences in PRS (14.9 vs. 19.2 months, p = 0.227; Fig. 3E) or OS (22.7 vs. 21.9 months, p = 0.851; Fig. 3F). Cox multivariable regression showed that age (HR 1.02, CI 95% 1.01–1.04, p = 0.001) correlated with a significant higher risk of death, whereas, parietal tumor location (HR 0.59, CI 95% 0.34–0.99, p = 0.046) and methylated MGMT status (HR 0.50, CI 95% 0.29–0.86, p = 0.013) were associated with lower risk of death (Online resource 3).
Fig. 3.
A GBM IDH–wildtype stratified by their intratumoral hemorrhage status, in which patients harboring intratumoral hemorrhage had a non-significant difference in PFS. B GBM IDH–wildtype stratified by their intratumoral hemorrhage status, in which patients harboring intratumoral hemorrhage had a non-significant difference in OS. C GBM IDH–wildtype with intratumoral hemorrhage only, stratified by salvage bevacizumab status, in which patients who received salvage bevacizumab had a non-significant difference in PRS. D GBM IDH–wildtype with intratumoral hemorrhage only, stratified by salvage bevacizumab status, in which patients who received salvage bevacizumab had a non-significant difference in OS. E GBM IDH–wildtype with non-intratumoral hemorrhage only, stratified by salvage bevacizumab status, in which patients who received salvage bevacizumab had a non-significant difference in PRS. F GBM IDH–wildtype with non-intratumoral hemorrhage only, stratified by salvage bevacizumab status, in which patients who received salvage bevacizumab had a non-significant difference in OS
Discussion
The present study demonstrated that hGBM was identified in 59.6% of patients. Clinically, hGBMs presented more frequently with motor deficits. Although exploratory molecular analyses suggested a higher frequency of SETD2 mutations and lower frequencies of PDGFRA, KIT, KDR, and PIK3R1 mutations among hGBMs, these associations did not remain statistically significant after correction for multiple comparisons. Therefore, these molecular findings should be interpreted as hypothesis-generating rather than definitive. Importantly, hGBM was not associated with prognosis.
The reported incidence of hemorrhage in brain tumors varies widely across the literature [7]. Prior studies have reported hemorrhage rates ranging from 2.3% to 29.2% in primary brain tumors, including GBMs, whereas metastatic brain tumors exhibit higher rates, ranging from 36.0% to 50.0% [6–8, 25]. However, interpretation of these estimates is limited by heterogeneity of prior cohorts, as most prior cohorts grouped different brain tumor types or glioma subtypes and relied on earlier WHO CNS classifications. In contrast, our study evaluated a uniformly defined cohort of patients with GBM and found that 59.6% showed hGBM, a higher rate than in reports of primary brain tumors. Differences in radiographic criteria used to identify hemorrhage may also contribute to variability across studies, and further studies incorporating susceptibility-weighted imaging (SWI) may help assess these reported incidence rates. Similarly, differences in MRI acquisition parameters, such as slice thickness, and scanner characteristics (1.5-T vs. 3-T), may have influenced the detection of hemorrhage both within our cohort and across studies. Although direct comparisons are limited for the previously mentioned study design differences, prior reports have suggested that higher-grade gliomas are associated with hemorrhage [8]. This observation is consistent with the findings of the present study and supports the concept that hGBM may reflect, at least in part, the aggressive vascular biology characteristic of GBM.
Molecular features
A previous study of 167 glioma patients found higher rates of CDKN2B, KMT5B, and PIK3CA alterations in tumors with hemorrhage compared to non-hemorrhagic tumors, with the authors suggesting that loss-of-function alterations in KMT5B may suppress VEGFR2 expression in endothelial cells [6]. In our series, we did not observe an association between hGBM and the previously reported alterations in CDKN2B and PIK3CA, while KMT5B was not assessed. Instead, our exploratory molecular analysis identified a different pattern: PDGFRA, KIT, and KDR amplifications, as well as PIK3R1 mutations, were more frequent in non-hGBMs, whereas SETD2 mutations were more frequent in hGBMs. These differences may reflect methodological and cohort-level distinctions, including the previous study’s broader population of diffuse gliomas and our molecularly defined GBM cohort. Furthermore, we used the standardized VASARI imaging criteria to define hGBM, providing a reproducible framework for future validation of our findings.
PDGFRA, KIT, and KDR are located within the receptor tyrosine kinase (RTK) cluster on chromosome 4q12 and are frequently co-amplified in glioblastomas [26, 27]. This co-amplification in glioblastomas has been associated with worse survival outcomes [27–29]. Interestingly, despite the higher frequency of these amplifications in non-hGBMs in our cohort, we observed no significant differences in survival outcomes based on hemorrhagic status. PDGFRA is activated by PDGF ligands and contributes to glial and mesenchymal cell proliferation and migration [30–33]. PDGFRA amplification, observed in 18.7% of our cohort, has been associated with GBM growth [32, 33]. KIT activation induces MAPK and PI3K–AKT signaling [31, 34, 35], while KDR, a VEGF receptor, mediates angiogenesis and vascular sprouting in GBM [31, 35]. Amplification of the 4q12 locus activates RTK-driven PI3K signaling, and the RTK–PI3K axis may also promote vascular stabilization and pericyte recruitment, although these effects are thought to occur primarily through PDGFRB signaling [36, 37]. The reduced and heterogeneous amplification of 4q12 RTK genes in patients with hGBMs in our cohort may reflect decreased pathway activation, whereas non-hGBMs may develop more stabilized and dense vascular networks. Also, non-hGBMs may instead be characterized by microvascular thrombosis, leading to hypoxia and pseudopalisading necrosis rather than distinct vascular or endothelial properties [38]. In preclinical models, hypoxia-driven angiogenesis in GBM involves stabilization of HIF-1 A, upregulation of VEGF expression, and formation of immature, highly permeable vessels with low pericyte coverage, potentially predisposing tumor vasculature to leakage and hemorrhage [39, 40]. Although both RTK-driven and hypoxia-driven pathways converge on VEGF signaling, their upstream activation may differentially influence vascular maturation, permeability, and hemorrhagic potential.
Among the molecular findings in our cohort, the higher frequency of SETD2 mutations in hGBMs is particularly intriguing. SETD2 encodes the histone methyltransferase responsible for H3K36 trimethylation (H3K36me3), a chromatin mark associated with active transcription. Through regulation of transcriptional elongation, RNA splicing, and chromatin organization, SETD2 helps maintain gene expression programs, and its disruption may promote epigenetic dysregulation and tumorigenesis [41]. Preclinical studies have also implicated SETD2 in angiogenic gene expression and vascular development, as SETD2 knockout mice exhibit forebrain vascular abnormalities and hemorrhage, potentially due to impaired neurovascular formation in neuroepithelial and neural crest–derived cells [41]. Although preliminary, these findings raise the possibility that SETD2 alterations may contribute to vascular instability in a subset of hGBMs.
Clinical presentation and survival
The clinical presentation of hGBMs likely depends on tumor size, location, hemorrhage volume, and involvement of eloquent cortex or subcortical pathways. Patients may present with focal neurological deficits, including hemiparesis, sensory loss, or visual deficits, and in some cases mimic a hemorrhagic stroke. Prior studies have suggested that ~ 3.0% of presumed stroke cases are later found to harbor an underlying tumor [42, 43]. In our series, hGBMs showed a higher frequency of motor deficits at presentation compared to non-hGBMs. This may reflect tumor location and larger tumor size, hemorrhage-related disruption of motor pathways, or greater acute mass effect. A prior cohort of 457 gliomas reported greater neurological impairment and lower preoperative KPS in hemorrhagic tumors compared to non-hemorrhagic tumors [6]. In contrast, our study found similar preoperative KPS between groups, suggesting that although hemorrhage may influence presenting symptoms and tumor size, it did not translate into a measurable difference in baseline functional status in our cohort.
The prognostic significance of hGBMs remains uncertain. Prior studies have reported conflicting data, with a cohort of 162 GBMs demonstrating worse survival among patients with hemorrhage in univariable analysis [6, 11]. In contrast, our cohort found similar PFS and OS between hGBMs and non-hGBMs. Discrepancies across studies may reflect differences in cohort composition, imaging definition of hemorrhage (20.0% of hemorrhage in the prior largest cohort compared to ~ 59.6% in our cohort), molecular classification, and adjustment for prognostic factors. Thus, while hGBM may be associated with distinct clinical and molecular features, its independent effect on survival remains unclear.
Limitations
This study has several limitations inherent to its retrospective design. Selection bias, as not all GBM patients at our institution underwent NGS and/or postoperative MRI within 72 h for volumetric assessment. As this study used the standardized VASARI definition for intratumoral hemorrhage, non-contrast CT and susceptibility-sensitive MRI sequences (SWI or T2*-weighted imaging) were not used. Therefore, these imaging sequences are needed to validate our findings and further future larger, multi-institutional studies incorporating characteristic hemorrhagic patterns, including distinguishing microscopic hemorrhage and chronic hemosiderin deposition from clinically significant hemorrhage, and their associated molecular features. In addition, future studies incorporating susceptibility-sensitive imaging may further characterize hemorrhagic patterns and determine whether they are associated with distinct molecular features and underlying biological mechanisms. Furthermore, future studies incorporating intraoperative and pathological imaging may provide direct confirmation of macroscopic hemorrhage and facilitate radiographic–intraoperative–pathologic correlation. In addition, MGMT promoter status was unavailable for many patients due to its recent adoption at our institution. The molecular analyses were also limited by sample size and the need for correction across multiple comparisons, which reduced the ability to identify statistically significant associations. Therefore, our genetic findings between hGBMs and non-hGBMs should be interpreted as exploratory. Despite these limitations, this study provides a homogeneous analysis of genetic alterations in hGBM vs. non-hGBM. External validation in larger datasets or external cohorts is needed to confirm our findings.
Conclusions
In this cohort of patients with GBM, intratumoral hemorrhage may be associated with distinct exploratory molecular patterns, including lower frequencies of KDR, KIT, PDGFRA, and PIK3R1 alterations and a higher frequency of SETD2 mutations compared with non-hGBMs. No significant differences in PFS or OS were observed between hGBMs and non-hGBMs.
Supplementary Information
Below is the link to the electronic supplementary material.
Acknowledgements
None.
Author contributions
Study design: AB, AD, and YE. Data recollection: AB, AD, OA, CAC, and LO. Data analysis: AB. Manuscript writing: AB, AD, OA, and YE. Manuscript revision and editing: AB, AD, SH, MA, JJZ, AIB, RFR, LYB, NT, and YE. Study supervision: YE. Approved final manuscript: all authors.
Funding
None.
Data availability
The datasets generated and/or analyzed during the current study are available from the corresponding authors upon reasonable request.
Code availability
Not applicable.
Declarations
Ethical approval
This study was approved by the institutional review board of The University of Texas Health Science Center at Houston and Memorial Hermann Hospital, Houston, TX (HSC-MS-17-1917), and it was in accordance with the 1964 Helsinki Declaration and its later amendments. This study also adheres to the STROBE guidelines.
Consent to participate and publication
Not applicable.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
- 1.Ostrom QT, Price M, Neff C et al (2022) CBTRUS statistical report: Primary brain and other central nervous system tumors diagnosed in the United States in 2015–2019. Neuro Oncol 24:v1–v95. 10.1093/neuonc/noac202 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Fortunato JT, Scharf A, Shuman AG, Diamond EL (2026) Staging prognostic discussions about glioblastoma. J Neurooncol 176:162. 10.1007/s11060-025-05356-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Stupp R, Mason WP, van den Bent MJ et al (2005) Radiotherapy plus concomitant and adjuvant temozolomide for glioblastoma. N Engl J Med 352:987–996. 10.1056/NEJMoa043330 [DOI] [PubMed] [Google Scholar]
- 4.Torres J, Bundrant BE, Dono A et al (2025) Cystic glioblastoma IDH-wildtype: genetic alterations and outcomes. J Neurooncol 175:515–523. 10.1007/s11060-025-05143-5 [DOI] [PubMed] [Google Scholar]
- 5.Louis DN, Perry A, Wesseling P et al (2021) The 2021 WHO classification of tumors of the Central Nervous System: A summary. Neuro Oncol 23:1231–1251. 10.1093/neuonc/noab106 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Shi Y, Kang X, Ge Y et al (2024) The molecular signature and prognosis of glioma with preoperative intratumoral hemorrhage: a retrospective cohort analysis. BMC Neurol 24:202. 10.1186/s12883-024-03703-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Schrader B, Barth H, Lang EW et al (2000) Spontaneous intracranial haematomas caused by neoplasms. Acta Neurochir (Wien) 142:979–985. 10.1007/s007010070052 [DOI] [PubMed] [Google Scholar]
- 8.Kondziolka D, Bernstein M, Resch L et al (1987) Significance of hemorrhage into brain tumors: clinicopathological study. J Neurosurg 67:852–857. 10.3171/jns.1987.67.6.0852 [DOI] [PubMed] [Google Scholar]
- 9.Wu Y, Wu Y, Xia S et al (2025) JMJD6-driven epigenetic activation of COL4A2 reprograms glioblastoma vascularization via integrin α1β1-dependent PI3K/MAPK signaling. Acta Neuropathol Commun 13:194. 10.1186/s40478-025-02114-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Sarkaria JN, Hu LS, Parney IF et al (2018) Is the blood-brain barrier really disrupted in all glioblastomas? A critical assessment of existing clinical data. Neuro Oncol 20:184–191. 10.1093/neuonc/nox175 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.McGahan BG, Neilsen BK, Kelly DL et al (2017) Assessment of vascularity in glioblastoma and its implications on patient outcomes. J Neurooncol 132:35–44. 10.1007/s11060-016-2350-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Gutman DA, Cooper LAD, Hwang SN et al (2013) MR imaging predictors of molecular profile and survival: multi-institutional study of the TCGA glioblastoma data set. Radiology 267:560–569. 10.1148/radiol.13120118 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Setyawan NH, Choridah L, Nugroho HA et al (2024) Beyond invasive biopsies: using VASARI MRI features to predict grade and molecular parameters in gliomas. Cancer Imaging 24:3. 10.1186/s40644-023-00638-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.VASARI research project https://wiki.cancerimagingarchive.net/display/Public/VASARI+Research+Project. Accessed 26 Feb 2026
- 15.Cheng AC, Duda SN, Taylor R et al (2021) REDCap on FHIR: Clinical Data Interoperability Services. J Biomed Inf 121:103871. 10.1016/j.jbi.2021.103871 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Bueno A, Dono A, Amezquita-Contreras C et al (2026) Molecular features of glioblastoma isocitrate dehydrogenase-wildtype in extreme survival groups after supramaximal and maximal resection. Neurosurgery. 10.1227/neu.0000000000003929. [DOI] [PubMed]
- 17.Ballantyne A, Eriksson S (2019) Research ethics revised: The new CIOMS guidelines and the World Medical Association Declaration of Helsinki in context. Bioethics 33:310–311. 10.1111/bioe.12581 [DOI] [PubMed] [Google Scholar]
- 18.von Elm E, Altman DG, Egger M et al (2007) The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Lancet 370:1453–1457. 10.1016/S0140-6736(07)61602-X [DOI] [PubMed] [Google Scholar]
- 19.Karschnia P, Young JS, Dono A et al (2023) Prognostic validation of a new classification system for extent of resection in glioblastoma: A report of the RANO resect group. Neuro Oncol 25:940–954. 10.1093/neuonc/noac193 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Dono A, Wang E, Lopez-Rivera V et al (2020) Molecular characteristics and clinical features of multifocal glioblastoma. J Neurooncol 148:389–397. 10.1007/s11060-020-03539-z [DOI] [PubMed] [Google Scholar]
- 21.Dono A, Amsbaugh M, Martir M et al (2021) Genomic alterations predictive of response to radiosurgery in recurrent IDH-WT glioblastoma. J Neurooncol 152:153–162. 10.1007/s11060-020-03689-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Thakkar JP, Dolecek TA, Horbinski C et al (2014) Epidemiologic and molecular prognostic review of glioblastoma. Cancer Epidemiol Biomarkers Prev 23:1985–1996. 10.1158/1055-9965.EPI-14-0275 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.cBioPortal for cancer genomics::oncoprinter. https://www.cbioportal.org/oncoprinter. Accessed 21 Feb 2025
- 24.Kanda Y (2015) [Statistical analysis using freely-available EZR (Easy R) software]. Rinsho Ketsueki 56:2258–2266. 10.11406/rinketsu.56.2258 [DOI] [PubMed] [Google Scholar]
- 25.Kim DG, Park CK, Paek SH et al (2000) Meningioma manifesting intracerebral haemorrhage: a possible mechanism of haemorrhage. Acta Neurochir (Wien) 142:165–168. 10.1007/s007010050019 [DOI] [PubMed] [Google Scholar]
- 26.Dono A, Torres J, Nunez L et al (2024) Imaging predictors of 4q12 amplified and RB1 mutated glioblastoma IDH-wildtype. J Neurooncol 167:99–109. 10.1007/s11060-024-04575-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Carlotto BS, Trevisan P, Provenzi VO et al (2023) PDGFRA, KIT, and KDR gene amplification in glioblastoma: Heterogeneity and clinical significance. Neuromolecular Med 25:441–450. 10.1007/s12017-023-08749-y [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Dono A, Ramesh AV, Wang E et al (2021) The role of RB1 alteration and 4q12 amplification in IDH-WT glioblastoma. Neurooncol Adv 3:vdab050. 10.1093/noajnl/vdab050 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Joensuu H, Puputti M, Sihto H et al (2005) Amplification of genes encoding KIT, PDGFRalpha and VEGFR2 receptor tyrosine kinases is frequent in glioblastoma multiforme. J Pathol 207:224–231. 10.1002/path.1823 [DOI] [PubMed] [Google Scholar]
- 30.McKinnon RD, Waldron S, Kiel ME (2005) PDGF alpha-receptor signal strength controls an RTK rheostat that integrates phosphoinositol 3’-kinase and phospholipase Cgamma pathways during oligodendrocyte maturation. J Neurosci 25:3499–3508. 10.1523/JNEUROSCI.5049-04.2005 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Nobusawa S, Stawski R, Kim Y-H et al (2011) Amplification of the PDGFRA, KIT and KDR genes in glioblastoma: a population-based study: PDGFRA amplification in glioblastoma. Neuropathology 31:583–588. 10.1111/j.1440-1789.2011.01204.x [DOI] [PubMed] [Google Scholar]
- 32.Hermanson M, Funa K, Hartman M et al (1992) Platelet-derived growth factor and its receptors in human glioma tissue: expression of messenger RNA and protein suggests the presence of autocrine and paracrine loops. Cancer Res 52:3213–3219 [PubMed] [Google Scholar]
- 33.Cancer Genome Atlas Research Network (2008) Comprehensive genomic characterization defines human glioblastoma genes and core pathways. Nature 455:1061–1068. 10.1038/nature07385 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Ida JA Jr, Dubois-Dalcq M, McKinnon RD (1993) Expression of the receptor tyrosine kinase c-kit in oligodendrocyte progenitor cells. J Neurosci Res 36:596–606. 10.1002/jnr.490360512 [DOI] [PubMed] [Google Scholar]
- 35.Zwick E, Bange J, Ullrich A (2001) Receptor tyrosine kinase signalling as a target for cancer intervention strategies. Endocr Relat Cancer 8:161–173. 10.1677/erc.0.0080161 [DOI] [PubMed] [Google Scholar]
- 36.Engelman JA (2009) Targeting PI3K signalling in cancer: opportunities, challenges and limitations. Nat Rev Cancer 9:550–562. 10.1038/nrc2664 [DOI] [PubMed] [Google Scholar]
- 37.Guo P, Hu B, Gu W et al (2003) Platelet-derived growth factor-B enhances glioma angiogenesis by stimulating vascular endothelial growth factor expression in tumor endothelia and by promoting pericyte recruitment. Am J Pathol 162:1083–1093. 10.1016/S0002-9440(10)63905-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Brat DJ, Van Meir EG (2004) Vaso-occlusive and prothrombotic mechanisms associated with tumor hypoxia, necrosis, and accelerated growth in glioblastoma. Lab Invest 84:397–405. 10.1038/labinvest.3700070 [DOI] [PubMed] [Google Scholar]
- 39.Pore N, Liu S, Haas-Kogan DA et al (2003) PTEN mutation and epidermal growth factor receptor activation regulate vascular endothelial growth factor (VEGF) mRNA expression in human glioblastoma cells by transactivating the proximal VEGF promoter. Cancer Res 63:236–241 [PubMed] [Google Scholar]
- 40.Mosteiro A, Pedrosa L, Ferrés A et al (2022) The vascular microenvironment in glioblastoma: A comprehensive review. Biomedicines 10:1285. 10.3390/biomedicines10061285 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Chen B, Zhang C, Rui H et al (2025) Histone H3K36 methyltransferases NSD1 and SETD2 are required for brain development. Hum Genet 144:529–543. 10.1007/s00439-025-02740-2 [DOI] [PubMed] [Google Scholar]
- 42.Richard SA, Ye Y, Li H et al (2018) Glioblastoma multiforme subterfuge as acute cerebral hemorrhage: A case report and literature review. Neurol Int 10:7558. 10.4081/ni.2018.7558 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43.Li L, Yin J, Li Y et al (2013) Anaplastic astrocytoma masquerading as hemorrhagic stroke. J Clin Neurosci 20:1612–1614. 10.1016/j.jocn.2012.09.041 [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The datasets generated and/or analyzed during the current study are available from the corresponding authors upon reasonable request.
Not applicable.