2.1 Study population

Patients undergoing surgery with the diagnosis of isocitrate-dehydrogenase (IDH)-wildtype glioblastoma were identified from a prospectively collected and maintained database of the Department of Neurosurgery at the University Medical Center Hamburg-Eppendorf, Germany. Patients were included when receiving adjuvant combined radiochemotherapy after Stupp [3] or CeTeG [28] regimen and with a follow-up time of at least 3 months after surgery. Data collection, including patient follow-up, was completed on December 31, 2023. Informed written consent was obtained from all patients. Diagnosis was based on the current World Health Organisation (WHO) classification for central nervous tumors (WHO 2021) [29]. The extent of resection was stratified into gross total resection (GTR), near GTR, partial resection (PR), and stereotactic biopsy. A GTR was defined as a complete removal of contrast-enhancing parts, a near GTR as a removal of more than 90% of the contrast-enhancing parts, whereas a resection of lower than 90% was defined as PR. The extent of resection of contrast-enhancing parts was evaluated in the magnetic resonance imaging (MRI) up to 48 h after surgery. Overall survival (OS) was measured from the date of surgery to the date of death from any cause, while progression-free survival (PFS) was measured from the date of surgery to the occurrence of the first progression. Additionally, progression-to-overall survival (POS) was calculated from the date of recurrence surgery to the date of death from any cause. A progress was defined according to Response Assessment in Neuro-Oncology (RANO) criteria [30]. All patients included from the initial surgery were screened for potential re-resection at the time of first local recurrence. Eligibility for recurrence surgery was determined by clinical decision-making of the local tumor board. Patients who underwent re-resection had blood samples taken prior to the re-resection, which were then analyzed and matched to their blood profiles from the initial surgery. Dexamethasone usage was recorded for each patient to assess their daily dosage intake. Patients with chronic dexamethasone treatment prior to their glioblastoma diagnosis were excluded from this study.

2.2 Peripheral blood-derived immune counts and composite scores

Consecutive patients were included if preoperative neutrophil, lymphocyte, platelet, and monocyte counts were available. Preoperative neutrophil, lymphocyte, platelet, and monocyte counts were obtained from a blood withdrawal within 3 days prior to the first surgery. All samples were collected before the administration of surgery-related medication. The composite scores neutrophil–lymphocyte ratio (NLR), platelet–lymphocyte ratio (PLR), and lymphocyte–monocyte ratio (LMR) were all calculated by directly dividing the former by the latter.

2.3 DNA methylation profiling

DNA was extracted from tumor tissue and analyzed for genome-wide DNA methylation patterns using the Illumina Infinium MethylationEPIC array (EPIC)  v2 (850 k). For bioinformatic reasons, converted data to the 450 K array format was used when conducting deconvolution analysis to determine cell type compositions, as previously described [31, 32]. Processing of DNA methylation data was performed with custom approaches [25]. Methylation profiling results from the first surgery were submitted to the molecular neuropathology (MNP) methylation classifier v12.8 hosted by the German Cancer Research Center (DKFZ) [18]. Patients were included if the calibrated score for the specific methylation class was >0.84 at the time of diagnosis in accordance with recommendations by Capper et al. [25] For IDH-wildtype glioblastoma, patients with a score below 0.84 but above 0.7 with a combined gain of chromosome 7 and loss of chromosome 10 or amplification of epidermal growth factor receptor (EGFR) were included in accordance with consortium to inform molecular and practical approaches to CNS tumor taxonomy (cIMPACT-NOW) criteria [33]. Furthermore, a class member score of ≥0.5 for one of the glioblastoma subclasses was required. Evaluation of the MGMT promoter methylation status was made from the classifier output v12.8 using the MGMT-STP27 method [34].

2.4 Processing of methylation arrays

All idats corresponding to methylation array data were processed similarly using the minfi package in R (version 1.40.0) [35]. The data was processed using the preprocessIllumina function. Only probes with detection p-values <0.01 were kept for further analysis. Also, probes with <3 beads in at least 5% of samples, as well as all non-5'-Cytosine-phosphate-Guanine-3'(CpG) probes, single nucleotide polymorphism (SNP)-related probes, and probes located on X and Y chromosomes were discarded. The CpG intensities were converted into beta values representing total methylation levels (between 0 and 1).

2.5 Cell state composition analysis

To infer the abundance of cell type and cell state in the samples, we subjected each sample to a bulk DNA methylation assay using EPIC arrays and applied the Silverbush et al. deconvolution method [36]. The deconvolution method is a reference-free method that uses a hierarchical matrix factorization approach inferring both cell types and the cell states therein. The method was trained on the DKFZ glioblastoma cohort and tested on the The Cancer Genome Atlas (TCGA) glioblastoma cohort and was able to infer the abundance of cell types in the microenvironment (immune, glia, and neuron) and malignant cell states (malignant stem-like cells component and two differentiated cells components). We applied the method as described in Silverbush et al. using the cell type and cell state encoding provided in the manuscript and via the engine provided in EpiDISH [37] package, with the robust partial correlation (RPC) method and a maximum iterations of 2000.

2.6 Cell type deconvolution

Non-negative least square (NNLS) linear regression was used in deconvolving the beta values of methylation arrays into cell type components [38-40]. As a reference, a publicly available signature was obtained from Moss et al. consisting of gene expressions for 25 cell type components (Monocytes_EPIC, B-cells_EPIC, CD4T-cells_EPIC, NK-cells_EPIC, CD8T-cells_EPIC, Neutrophils_EPIC, Erythrocyte_progenitors, Adipocytes, Cortical_neurons, Hepatocytes, Lung_cells, Pancreatic_beta_cells, Pancreatic_acinar_cells, Pancreatic_duct_cells, Vascular_endothelial_cells, Colon_epithelial_cells, Left_atrium, Bladder, Breast, Head_and_neck_larynx, Kidney, Prostate, Thyroid, Upper_GI, and Uterus_cervix) and 6105 unique CpGs [38].

2.7 3D volumetric segmentation

We analyzed T1-weighted as well as T2-weighted fluid attenuated inversion recovery (FLAIR) MRI axial images of all glioblastoma patients before surgery. The program BRAINLAB was used for all analyses. To measure tumor volume, the tumor region of interest was delineated with the tool “Smart Brush” in every slice by hand, enabling a multiplanar 3D reconstruction. With this methodology, the volume of contrast enhancement and of FLAIR hyperintensity was assessed in cubic centimeter.

2.8 Statistical analysis

The data collection, including the patient follow-up, was completed on December 31, 2023. After that, the differences in continuous variables were analyzed with the Mann–Whitney U test, and differences in proportions were analyzed with the chi-square test or Fisher exact test. Overall and disease-free survival was evaluated with the Kaplan–Meier method. Univariate and multivariate Cox proportional hazards models were used to assess the effects of variables on seizure outcome and to compute odds ratio (OR). A two-sided p-value less than 0.05 was considered as statistically significant. All analyses were performed using SPSS Inc. (Chicago, IL, USA).

3.1 A higher peripheral neutrophil count is associated with unfavorable overall survival in newly diagnosed glioblastoma

To address the question of whether peripheral immune cell counts (neutrophils, lymphocytes, platelets, and monocytes) or composite scores (NLR, PLR, and LMR) can serve as predictive markers for the prognosis of glioblastoma patients, we conducted a comprehensive survival analysis involving all 176 patients mentioned in this study. All patients underwent combined radiochemotherapy following surgery and had a minimum follow-up period of 3 months. As of the final data collection, 137 (77.8%) patients were dead (Figure S1).

Given the existing discord in the literature concerning the impact of dexamethasone on immune cell functionality and counts [41-44], we initially divided the study population into two cohorts: one comprising patients who received dexamethasone before the first surgery and blood collection and the other consisting of those who did not (Figure S2). In this initial analysis, no significant differences were observed with respect to immune cell counts and composite scores between these two subgroups (Figure S2). Consequently, we decided to include all patients in our survival analysis, regardless of their preoperative dexamethasone treatment, as no survival difference was observed in the multivariate analysis (Table SS2). As the further correlation between patient survival and continuous immune cell counts did not yield significant results (Table SS3), patients were subsequently categorized based on the median value of each immune cell count, as detailed in Table SS4. Further correlation with additional clinical features revealed lower platelet levels in older patients (Figure S3) but no other associations with additional clinical variables (Figures S4 and S5).

The results of our survival analysis unveiled a notably improved OS among patients with lower peripheral neutrophil counts (p = 0.01, median OS 17.0 vs. 10.0 months; Figure 1A). However, none of the other immune cell counts (Figure 1B–D) or composite scores (Figure 1E–G) demonstrated prognostic significance in the context of newly diagnosed glioblastoma when dichotomized by the median (Table SS4).

3.2 Peripheral immune cells are decreased at the time of progression and predict overall survival in recurrent glioblastoma

In our study, a total of 176 patients were included, of which 41 underwent re-resection for local recurrent glioblastoma following the completion of combined radiochemotherapy. For these patients, peripheral immune cell counts were obtained 3 days prior to the re-resection. We conducted a detailed analysis of immune cell counts and composite scores in this matched-pair patient cohort (Figure 2A–G). At the time of re-resection, significant alterations were observed in various blood parameters. Specifically, lymphocytes (p = 0.00004, Figure 2A), monocytes (p = 0.0435, Figure 2B), neutrophils (p = 0.0078, Figure 2C), and platelets (p = 0.00016, Figure 2D) exhibited a marked decrease. Additionally, the LMR showed a significant reduction (p = 0.0096, Figure 2E). However, no significant differences were observed in the NLR (p = 0.925, Figure 2F) and the PLR (p = 0.155, Figure 2G) when compared with their counterparts with newly diagnosed glioblastoma when dichotomized by the median.

All patients were followed up after re-resection with 28 patients (68.3%) were reported as dead. The patient cohort was stratified based on the median peripheral immune cell counts at the time of recurrent surgery (Figure 2H–N). Survival analysis revealed an unfavorable prognosis for patients with elevated monocyte (p = 0.03; Figure 2I), neutrophil (p = 0.04; Figure 2J), and platelet counts (p = 0.01; Figure 2K). However, the composite scores (LMR, NLR, and PLR) were found to be insufficient for predicting patients' survival following re-resection when dichotomized by the median (Figure 2L–N).

Our findings underscore the changes in peripheral immune cell counts at the time of first recurrence after combined radiochemotherapy and highlight the association between higher cell counts and poorer survival outcomes in these patients.

3.3 Integrated DNA methylation analysis alters the predictive potential of peripheral immune cell counts based on subclass

It is evident that there are DNA methylation subclass-specific signaling pathway aberrations [45], and the response to immunotherapy might be selective for certain subclasses [27]. Therefore, we conducted genome-wide DNA methylation analysis on all 176 samples obtained from the initial diagnosis and subclassified patients based on their DNA methylation subclass, namely RTK I, RTK II, or MES [18, 19]. The DNA methylation subclass-stratified comparison of peripheral immune cell counts and composite scores showed significantly higher neutrophil counts in the blood of RTK I patients compared to RTK II patients (p = 0.02; Figure 3A). However, all other immune cell counts and composite scores did not differ between the subclasses (Figure 3B–G).

When we performed survival analysis for each DNA methylation subclass using the aforementioned cutoff values (Table SS4), we found a significant influence of monocytes in the RTK II subclass (p = 0.04, median OS 16.0 vs. 9.0 months; Figure 3H), and of neutrophils (p = 0.007, median OS 22.0 vs. 14.0 months; Figure 3I), as well as lymphocytes (p = 0.04, median OS 19.0 vs. 13.0 months; Figure 3J) in the MES subclass on the OS of newly diagnosed glioblastoma patients. All other immune cell counts were not predictive of the OS when dichotomized by the median, regardless of whether classified as RTK I (Figure SS6), RTK II (Figure SS7), or MES (Figure SS8).

In summary, these findings demonstrate comparable peripheral immune cell counts across DNA methylation subclasses but underscore the importance of subclassification when analyzing survival outcomes.

3.4 Correlation of peripheral immune cell counts with cell states and infiltrating immune cells in matched tumor tissue

Utilizing a deconvolution approach of DNA methylation data has emerged as a powerful tool for defining cell states and the composition of tumor tissue, as demonstrated in previous studies [38, 46]. These analyses have revealed significant enrichment of different cell types between subclasses [47, 48]. We posed the question of whether peripheral immune cell counts are accurately reflected in the cell states and infiltrating cells within the tumor tissue. To investigate this, we performed deconvolution analysis on matched tumor tissue samples from all 176 patients (Figure 4). Using a reference dataset provided by Silverbush et al., we observed a significant association between platelets (p = 0.02, r = −0.17; Figure 4A), monocytes (p = 0.01, r = 0.20; Figure 4A), and the PLR (p = 0.03, r = −0.17; Figure 4A) with a more differentiated, tumor-progressive cell state (“differentiated-2”). Additionally, platelets exhibited significant correlations with immune (p = 0.04, r = 0.15; Figure 4A) and neuron (p = 0.04, r = 0.14; Figure 4A) signatures. Further, when we delved into a more detailed breakdown of the immune compartment using a reference signature by Moss et al. [38], we found an association between peripheral monocytes and tissue-infiltrating B cells (p = 0.004, r = 0.23; Figure 4B). However, it is important to note that all other peripheral immune cell counts did not show a consistent reflection in the tumor tissue deconvolution approach across all tumors (Figure 4B). Upon stratification into DNA methylation subclasses, we observed no significant association between peripheral immune cell counts and tumor tissue-infiltrating cells in the RTK I (Figure S9A) and RTK II (Figure S9B) subclasses. However, in the MES subclass, especially tumor-infiltrating B cells, we observed significant correlations (Figure S9C).