Proteometabolomics of initial and recurrent glioblastoma highlights an increased immune cell signature with altered lipid metabolism

Abstract

Background

There is an urgent need to better understand the mechanisms associated with the development, progression, and onset of recurrence after initial surgery in glioblastoma (GBM). The use of integrative phenotype-focused -omics technologies such as proteomics and lipidomics provides an unbiased approach to explore the molecular evolution of the tumor and its associated environment.

Methods

We assembled a cohort of patient-matched initial (iGBM) and recurrent (rGBM) specimens of resected GBM. Proteome and metabolome composition were determined by mass spectrometry-based techniques. We performed neutrophil-GBM cell coculture experiments to evaluate the behavior of rGBM-enriched proteins in the tumor microenvironment. ELISA-based quantitation of candidate proteins was performed to test the association of their plasma concentrations in iGBM with the onset of recurrence.

Results

Proteomic profiles reflect increased immune cell infiltration and extracellular matrix reorganization in rGBM. ASAH1, SYMN, and GPNMB were highly enriched proteins in rGBM. Lipidomics indicates the downregulation of ceramides in rGBM. Cell analyses suggest a role for ASAH1 in neutrophils and its localization in extracellular traps. Plasma concentrations of ASAH1 and SYNM show an association with time to recurrence.

Conclusions

We describe the potential importance of ASAH1 in tumor progression and development of rGBM via metabolic rearrangement and showcase the feedback from the tumor microenvironment to plasma proteome profiles. We report the potential of ASAH1 and SYNM as plasma markers of rGBM progression. The published datasets can be considered as a resource for further functional and biomarker studies involving additional -omics technologies.

Key Points.

1. Proteomics/lipidomics profiling of recurrent glioblastoma (rGBM) reveals a neutrophil fingerprint.

2. ASAH1 is upregulated in rGBM, and functional analysis shows a role in the tumor microenvironment.

3. ASAH1 and SYNM show association with time to recurrence.

Importance of the Study.

Glioblastoma multiforme is considered the most aggressive of brain tumors. With a nearly 100% recurrence rate, current therapy options are widely unsuccessful, making it a high priority to explore novel prognostic markers and therapeutic targets. Proteomics analyses allow for the large-scale evaluation of protein expression and their post-translational modifications, while lipidomics analyses offer an unbiased approach to explore the metabolic status in variable disease conditions. In this study, we performed proteomics and lipidomics analyses of patient-matched initial versus recurrent glioblastomas. Our results revealed upregulation of ASAH1, SYNM, and GPNBM, together with reduced concentrations of ceramides in recurrent tumors. Cellular analyses suggest a role for ASAH1 in tumor-associated immune cells and in the formation of neutrophil extracellular traps. Plasma levels of ASAH1 and SYNM measured via ELISA at initial tumor surgery show an association with time to recurrence.

Glioblastoma multiforme WHO IV (GBM) is the most common, malignant, and aggressive brain tumor. Despite a therapy consisting of surgery, radiotherapy, and temozolomide, the median survival rate of 15 months remains poor, with a recurrence rate of nearly 100%.¹ There is no standard treatment for recurrent GBM, and a second (“repeat”) surgery is deemed feasible in only 20%–30% of cases due to poor accessibility.² Furthermore, the complex dynamics of the tumor microenvironment (TME) have important prognostic implications. Specifically, the infiltration rate of neutrophils into the GBM TME is discussed to be used as a prognostic marker,³ and the formation of neutrophil extracellular traps (NETs) has been described to confer tumor protection and promote its development.⁴ In this context, there is still an urgent need to better understand the molecular mechanisms associated with tumor progression and the onset of recurrence to uncover novel prognostic markers and therapeutic targets.

Sequencing-based molecular approaches have been important for the broader understanding of glioblastoma biology,^5,6 but they are limited in their capacity to reveal the differential regulation of post-translational modifications.⁷ In this context, proteomics rises as the strategy of choice to explore the phenotypic status in different diseases and stages, offering a new valuable layer of information for biomarker research and alternative therapies, including the large-scale probing of proteolytic activity based on the analysis of protein termini (termed “terminomics”).⁸ Pioneer studies have performed large-scale proteogenomic and metabolomic analyses of GBM tumors, revealing proteome-specific subgroups.⁹ Recent efforts have focused on the proteomics characterization of the recurrent tumors, suggesting different drivers of tumor progression^10,11 and treatment resistance.¹² This showcases the increased acknowledgment of the importance of phenotype-focused approaches (i.e. proteomics and lipidomics) to improve our understanding of recurrent GBM (rGBM). Still, an integrative analysis of the proteometabolic rearrangement in rGBM and its implications in the tumor microenvironment remains to be explored.

In the current study, we performed integrative quantitative proteomics, terminomics, and lipidomics analyses on patient-matched initial (iGBM) and and rGBM frozen-fresh tumor samples. Proteomics analyses revealed changes associated with immune cell infiltration and extracellular matrix (ECM) organization. Among the top protein hits, we observed ASAH1 (Acid Ceraminidase 1), SYNM (Synemin), and GPNMB (Glycoprotein Nonmetastatic Melanoma Protein B) as enriched in rGBM. Lipidomics analyses showed a depletion of ceramides in rGBM. We performed functional analyses to evaluate the association of ASAH1 with immune infiltration and report its implication in the reorganization of the TME and NETs formation. Finally, we explore the use of ASAH1, SYNM, GPNMB, and MMP-9 concentrations in patient plasma to differentiate between rGBM from iGBM status as potential markers for disease progression.

Materials and Methods

Ethics and Sample Collection

Ethical approval was obtained before this study from the local authorities. At all sites, patients enrolled gave written consent prior to this study. At Philipps University Marburg, the study was approved by the ethics committee (Medical Faculty, file number 185/11). At HULAFE, by the local Ethics committee (La Fe Research Institution—Hospital University and Polytechnic La Fe, file number 0318) and at Bucharest from the Ethics committee of Bagdasar-Arseni (file number 8763 from April 8, 2019). Tissue samples from GBM patients were obtained during surgical resections. The samples were either embedded in paraffin or frozen in liquid nitrogen and stored at −80°C. Blood samples were collected presurgery and 3 to 5 days after tumor resection. After centrifugation at 2000g for 10 minutes, the plasma was stored at −80°C. All patients suffered from isocitrate dehydrogenase (IDH) wild-type GBM WHO grade 4 (Table 1; Supplementary Table S1).

Table 1.

Summary of the patient cohort

Variable	ELISA, N = 301	IHC, N = 101	Lipidomics, N = 101	Mass Spectrometry, N = 111	RT-qPCR, N = 201
Sex
Men	14 (47%)	4 (40%)	5 (50%)	6 (55%)	11 (55%)
Women	16 (53%)	6 (60%)	5 (50%)	5 (45%)	9 (45%)
Age at surgery (years)	62 (54, 68)	51 (50, 61)	62 (54, 63)	51 (46, 58)	61 (52, 64)
TTR2	224 (166, 288)	284 (160, 456)	218 (190, 273)	237 (182, 398)	288 (192, 418)
Unknown	7	0	0	0	0
Latency (days)	437 (336, 603)	576 (410, 775)	546 (431, 766)	681 (516, 1073)	602 (491, 1054)
Unknown	11	3	6	4	10
Type of resection
Subtotal	12 (40%)	4 (40%)	2 (20%)	5 (45%)	9 (45%)
Total	18 (60%)	5 (50%)	8 (80%)	6 (55%)	11 (55%)
Unknown	0 (0%)	1 (10%)	0 (0%)	0 (0%)	0 (0%)

¹ n (%)

² Time to recurrence (days); Median (IQR).

Isolation and Culture of Primary Neutrophils

Neutrophils from healthy donors were isolated from buffy coats obtained from the Blood Bank of the University Hospital Giessen and Marburg. The isolation was performed as described previously.¹³ In brief, dextran sedimentation and low-density Histopaque (1.077 g/mL, Sigma–Aldrich, Germany) gradient centrifugation were applied. Isolated neutrophils were cultured in DMEM (11965-095, Gibco, US) supplemented with 20% FCS (S0615, Sigma–Aldrich, Germany), 25 mM Hepes (H0887, Sigma–Aldrich, Germany), and 1% penicillin/streptomycin (2321115, Gibco, US). Neutrophils were stimulated with either 50 ng/mL Interferon-gamma (IFNɣ; 300-2, Peprotech, Germany) or 20 ng/mL Transforming Growth Factor beta (TGFβ; 7754-BH, R&D Systems, UK). They were induced into either an immune-supportive (IFNɣ) or a tumor-supportive (TGFβ) phenotype. The cell density was 1.6 × 10⁶ cells/mL. Cells were incubated at 37°C and 5% CO₂ for 6 hours.

Coculturing Primary Neutrophils and THP1 Macrophages with Primary GBM Cells

The coculture was performed similarly as previously described.¹⁴ Briefly, the primary GBM tumor cells were seeded on the upper part of the ThinCert Cell Culture Insert (0.4 µm diameter, Greiner-Bio-One GmbH, Germany) at a density of 5 × 10⁵ cells/mL and on a 6-well-plate format. After 24 hours, primary neutrophils were seeded at the lower compartment of the construction at a density of 1.6 × 10⁶ cells/mL for 6 hours before protein isolation. THP1 cells were differentiated into macrophage-like cells as previously described¹⁴ and cocultured with GBM cells as described above (Supplementary methods).

Tissue Cell Sorting and Subsequent Gene Expression Analysis

Tumor cell isolation.

— Liquid-nitrogen shock frozen GBM tissue samples were separated into different cell types, including immune cell types, using antibody-labeled magnetic beads. Subsequently, resulting cell populations derived from tumor bulk tissues (n = 44) were analyzed for ASAH1 mRNA expression via RT-qPCR. Each fraction was isolated using a magnetic stand and washed three times with cell-fraction-specific buffers (Supplementary methods).

Gene expression analysis.

— Cell fractions were lysed in TRIzol Lysis Reagent, and RNA was isolated according to the manufacturer’s instructions. Gene expression was analyzed by Real-Time PCR using TaqMan Gene Expression Assays (Thermo Fisher Scientific, USA) and specific probes for ASAH1, GPNMB, and SYNEMIN. Total RNA isolation was performed as described previously.¹⁴ The relative mRNA expression was calculated with the ΔCt method. (Supplementary methods).

Protein Isolation and Western Blot Analyses

Cells were scraped and homogenized with RIPA buffer, as described previously.¹⁵ The concentration was determined via the BCA Protein Assay Kit (Thermo Fisher Scientific). The protein samples were added to 5× Laemmli buffer and 10× NuPAGE sample reducing reagent (Thermo Fisher Scientific). Blocking was perforned with 5% milk powder (MP) in TBS-T. Secondary antibodies (anti-ASAH1, anti-MMP-9, anti-MPO, and anti-β-Tubulin) were diluted at different concentrations. The chemiluminescence was detected by adding WesternBright Sirius HRP substrate (K-12043-D20, Advansta, USA) and using the ChemiDoc MP Imaging System (Bio-rad Laboratories GmbH, USA) (Supplementary methods).

Immunohistochemistry

Immunohistochemistry (IHC) staining of MPO (monoclonal antibody; MAB3174, R&D Systems, USA) and ASAH1 (polyclonal antibody; HPA005468, Atlas Antibodies, Sweden) was performed on paraffin-embedded tissue. For this, the Vector Laboratories Vectastain Elite ABC Kit mouse IgG (PK-6102, Vector Laboratories, US) was used according to the manufacturer’s protocol. Peroxide blocking was performed for 30 minutes, and the antibody was incubated for 1 hour using an autostainer (AutostainerPlus, DAKO, Hamburg, Germany). The IHC double staining (ASAH1 + MPO; ASAH1 + CD68) was performed as described previously¹⁶ (Supplementary methods).

Immunocytochemistry

Neutrophils were seeded on coverslips (12 mm diameter) coated with Poly-D-Lysine (P1149, Sigma–Aldrich, Germany) and fixed with 4 % (w/v) PFA in PBS and subsequently permeabilized with 0.3 % (v/v) Triton X in PBS. Unspecific binding sites were blocked with 5 % (w/v) BSA in PBS. Primary antibodies were diluted in 5 % (w/v) BSA in PBS (ASAH1, APREST70303, 1:100, Sigma–Aldrich, Germany; MMP-9, AF911, 1:40, R&D Systems, USA). Neutrophils were incubated with secondary antibodies and subsequently stained with Hoechst (H33258, Abcam, UK) for visualization of DNA and NETs formation. Coverslips were mounted on slides using Fluorescence mounting medium (Supplementary Methods).

Enzyme-Linked Immunosorbent Assay

GBM patients’ plasma samples and plasma samples from healthy age-matched donors were analyzed for their protein amount of soluble ASAH1 (HUDL00256, AssayGenie, IRL), SYNM (ABIN6224490, Antibodies-Online), GPNMB (HUFI04984, AssayGenie, IRL), and MMP-9 (DY911, R&D Systems, USA) according to the manufacturer’s protocol.

Protein Extraction, Sample Preparation, and LC-MS/MS Measurement

GBM samples were processed to be analyzed via liquid chromatography coupled to mass spectrometry as previously described,¹⁷ using a TMT labeling approach. Samples were distributed into three TMT mixtures, each of one containing iGBM/rGBM patient sample pairs. (Supplementary methods).

Lipid Extraction, Sample Preparation, and LC–MS/MS Measurement

iGBM/rGBM tissue samples were used from each patient. Lipids and water-soluble molecules were separated via phase separation (Supplementary methods). Non-targeted measurement of lipids by LC–MS was carried out as described previously¹⁸ using an Agilent 1290 Infinity II UHPLC in line with a Bruker Impact II QTOF-MS operating in negative or positive ion mode (Supplementary methods).

Proteomics LC–MS/MS Data Processing

Three bioinformatic approaches were applied to extract different layers of information from the spectral data: (1) general large-scale proteomics identification and quantitation, (2) analysis of proteolytic processing, and (3) proteogenomics. The FragPipe pipeline^19–21 (v17) was used for peptide/protein identification and quantitation. In brief, spectral data were searched against an in-silico-digested protein sequence database (EBI Human canonical proteome, version 2021_03) assuming experimental tryptic digestion. Quantitation was based on TMT reporter ion intensities from MS2 spectra.²² Analysis of proteolytic processing was based on a semi-tryptic search. Proteogenomics analyses were performed after generating a GBM-specific database from publicly available RNA-seq sequencing data, including single amino acid variants (SAAVs), using a Galaxy-based²³ bioinformatics pipeline (Supplementary methods).

Statistical Analysis of Proteomics and Lipidomics Results

Differential abundance analysis was based on a linear model including the patient random effect, to account for the patient-match nature of the cohort. Differentially abundant proteins were subjected to functional enrichment analyses based on Reactome. The analysis of proteolytic processing was performed using the semi-specific search results. We focused our proteogenomics analyses on detecting differentially abundant SAAVs. R scripts, intermediate results, and reproducible reports are available via Zenodo²⁴ (Supplementary methods).

The lipidomics features were preprocessed for visualization, exploratory analyses, and quality control. Quantitative features were standardized and filtered using our exclusion criteria (Supplementary methods). Patient-matched linear models were used for the differential abundance analysis. Resulting up- or downregulated features were evaluated using functional enrichment tools. The integration of proteomics and lipidomics data was based on sparse partial least squares regression (sPLS) (Supplementary methods).

Analysis of Plasma-ELISA Results

Log₂ transformed distributions of protein concentrations were compared pairwise for control, iGBM, and rGBM. sPLS-DA was performed to visualize the separation between samples based on plasma levels of the targeted proteins. The association between protein concentrations in plasma and time to recurrence (TTR) was tested using Kaplan–Meier analyses and a Cox-proportional hazards model (Supplementary methods).

Data and Code Availability

Spectral files used for peptide and protein identification and quantitation can be accessed via the European Phenome Archive (EGA) (https://ega-archive.org/studies/EGAS00001006395), under restricted access (due to human data sharing restrictions). Requests for data access will be referred directly to our Data Access Committee associated with EGA https://ega-archive.org/dacs/EGAC00001002750. The Reproducible reports and R scripts used for the processing and analysis of proteomics, lipidomics, and plasma data are available via Zenodo.²⁴ The proteogenomics pipeline can be accessed via Galaxy through the link: https://usegalaxy.eu/published/workflow?id=438dc46201bb8996

Results and Discussion

Cohort Description

This study is based on a cohort of 53 patients (22 with non-methylated, 31 with methylated MGMT promoter), whenever patient-matched tumor tissue samples from iGBM and rGBM (Table 1) were available. For LC-MS/MS-based proteomics analyses, samples from surgically resected glioblastoma specimens were obtained from each of the 11 patients, resulting in a patient-matched cohort of 22 samples of initial and recurrent tumors. Overall, the cohort size is comparable with previous proteomics studies.¹⁰ ELISA tests were performed on plasma samples from 30 patients, and IHC assays were performed on tissue slides from 10 GBM patients. LC–MS/MS lipidomics analyses were performed from paired rGBM and iGBM tissues from a set of 10 patients. These were aimed to integratively explore the association of variable protein expression in rGBM and metabolic profiles. In agreement with present clinical guidelines, initial GBM patients (53 in total) were treated by surgical resection and adjuvant chemoradiotherapy. In total, 37 patients received chemoradiotherapy, 12 radiotherapy only, 2 chemotherapy only, 1 only surgery but no treatment (for details, see Supplementary Table S1).

Differential Proteome Features of Recurrent GBM

Proteomic analysis identified and quantified 5954 proteins after database search against canonical human protein sequences. Four thousand four hundred and sixty four were present in at least seven initial/recurrent GBM (iGBM/rGBM) pairs (or >60% of samples) (Figure 1A). These numbers showcase a broader proteome coverage at similar sample proportions based on our TMT-based approach, compared to recently published data.^10,11 Log2-transformed and normalized protein abundances were highly comparable across the cohort (Figure 1B). sPLS-DA analysis resulted in a model that was able to segregate iGBM from rGBM samples (Figure 1C). These observations indicate that our approach allows for robust quantitation of proteomics features with differential behavior between iGBM and rGBM.

Figure 1.

Shotgun proteomics reveals molecular rearrangement with a changing immune fingerprint in recurrent tumors. (A) The number of proteins identified and quantified per TMT mixture (Mix). Quant. in >4 represents the number of proteins quantified in at least four paired samples (iGBM and rGBM). (B) Normalized median protein abundance values per sample in the study. (C) Sparse least square discriminant analysis (sPLS-DA) results. (D) Differential abundance analysis results between iGBM and rGBM. Black dots represent differentially abundant proteins, gray dots are not significantly different. Red dots represent ASAH1, SYNM, GPNMB, and CD14. (E) Normalized abundances of ASAH1, GPNMB, and SYNM. Lines join samples belonging to the same patient. (F) Protein coverage of ASAH1 and SYNM based on tryptic peptide identifications via mass spectrometry. (G) Significantly enriched pathways from either up- or downregulated proteins at rGBM, based on the Reactome database. GeneRatio: the proportion of up-/downregulated proteins observed in the annotated pathway. P-adjust: adjusted P-value of the over-representation test. (H) Network representation of the significantly enriched pathways and their associated up-/downregulated proteins. (I) ASAH1 mRNA expression from different cell types isolated from tumor, based on RT-qPCR. Glast: glutamate-aspartate transporter; CD31: marker of endothelial cells; CD14: marker for immune cells; CD133: marker for cancer stem-like cells. A one-tailed paired student’s t-test was applied to determine differences between groups (*P-value < .05). (J) MPO-positive neutrophils visualized via IHC staining utilizing the monoclonal antibody MAB3174. (R&D Systems, USA). Representative pictures were taken comparing one patient with iGBM and rGBM. Scale bar = 200 µm.

We used a linear model to explore the association of protein abundance with iGBM or rGBM, including the patient random effect as a covariate to control for the inter-patient heterogeneity. One hundred and six proteins were differentially abundant between rGBM and iGBM, with 116 upregulated in rGBM and 30 being downregulated (Figure 1D, Supplementary Table S2).

The protein acid ceramidase (ASAH1, involved in sphingosine metabolism) showed the smallest adjusted p-value, as it is consistently upregulated in rGBM (Figure 1E). The ASAH1 peptides identified account for ~50% of protein coverage (Figure 1F). Furthermore, we observed glycoprotein nonmetastatic melanoma protein B (GPNMB, transmembrane protein with melanogenic activity) as upregulated in rGBM, showing the highest positive fold-change of all dysregulated proteins, and increased abundance in most patients (Figure 1E). Finally, Synemin (SYNM, intermediate filament with cytoskeletal role) was detected as consistently upregulated in all patients in this study (Figure 1E) with >60% of protein coverage (Figure 1F).

Functional enrichment analyses encountered five major biological themes as enriched by upregulated proteins. Three of these are associated with immune processes such as neutrophil degranulation and interleukin signaling. Further enriched pathways were associated with cell surface interactions and extracellular matrix organization (Figure 1G–H).

Expression levels of these proteins were evaluated by qPCR in an independent set of patients. SYNM and GPNMB mRNA levels were increased in rGBM (p-value = .03 and p-value = .042), which aligns with the higher enrichment in rGBM observed via quantitative proteomics. In contrast, there is no difference in ASAH1 mRNA expression comparing iGBM and rGBM (p-value = .64) (Supplementary Figure S1) pointing to differential mRNA and protein biology of ASAH1, in line with earlier reports about ASAH1 expression in brain.²⁵

Upregulation of ASAH1 Shows an Association with Neutrophil and Macrophage Infiltration in Recurrent GBM

We observed that three of the five enriched biological themes are associated with immune processes, together with an increased expression of the granulocyte-related protein CD14 in rGBM (Figure 1H). We then questioned whether the increased abundance of ASAH1 in rGBM could be related to immune infiltration. We isolated cell populations from GBM tissue samples and quantified the mRNA expression of ASAH1 on each. Interestingly, ASAH1 is enriched specifically in CD14+ cells (Figure 1I). This observation suggests that the high expression of ASAH1 in rGBM tissue is mainly driven by monocytes and granulocytes. The reanalysis performed on publicly available single-cell expression data derived from initial and recurrent human glioblastoma myeloid cells²⁶ reveals that ASAH1 shows a higher expression in cells annotated as tumor-associated macrophages (TAM), mast cells, and monocytes. ASAH1 also shows overlapping expression with ELANE (neutrophil marker), MPO (macrophage marker) (Supplementary Figure S2A, IBA1 (microglia), and CD68 (bone marrow-derived macrophages) (Supplementary Figure S2B), suggesting that ASAH1 seems to be ubiquitous among these cell populations. Single-cell data reanalysis shows minimal differential abundance of ASAH1 between iGBM and rGBM in cells annotated as monocytes and tumor-associated macrophages (Supplementary Figure S2C). We observed a high infiltration of neutrophil granulocytes via IHC, especially in rGBM (Figure 1J). These results underline the critical role of tumor-infiltrating immune cells not only in iGBM but even higher in rGBM and the possible association of rGBM infiltrating immune cells with ASAH1.

Further IHC staining was performed to investigate the implication of ASAH1 in the tumor microenvironment. MPO was used as a marker for neutrophil granulocytes, while CD68 was used for macrophages (Figure 2A). The semi-quantitation shows a higher occurrence of ASAH1 in rGBM (Figure 2B, p-value = .002). In the case of MPO, a trend of higher neutrophil abundance in rGBM was traced (Figure 2B, p-value = .08). In a second set of patients, this trend was confirmed with a higher infiltration rate in rGBM (p-value = .01, Supplementary Figure S3). A higher macrophage infiltration in rGBM was observed based on CD68-positive cells (Figure 2B, p-value = .02). Correlation analyses showed a small but significant association between the high abundance of ASAH1+ and MPO+ or CD68+ cells, respectively (Figure 2C and 2D). Costaining between ASAH1 and MPO/CD68 was performed to evaluate colocalization (Figure 2E). Notably, more ASAH1, more neutrophils, and especially more macrophage infiltration were visible in rGBM. Interestingly, the upper right picture shows ASAH1 being secreted by a neutrophil granulocyte (Figure 2E), suggesting an intracellular and extracellular role as part of cell-to-cell communication in the tumor microenvironment (TME). With this, we provide evidence that ASAH1 is expressed not only by tumor cells but also by neutrophils and macrophages in the TME of glioblastoma.

Figure 2.

The high occurrence of ASAH1 in rGBM is associated with elevated neutrophil and macrophage infiltration in rGBM. (A) Ten iGBM and rGBM-paired tissue slides (n = 20) were stained via 3,3ʹ-Diaminobenzidine (DAB) immunohistochemistry. (A&B) Representative pictures of ASAH1 (upper panel), MPO (middle panel), and CD68 (lower panel) are shown. Scale bar = 50 µm. Arrows indicate MPO-positive cells. (B) Semi-quantitation of positive cells (QuPath-0.2.3). (*P-value < .05; **P-value < .01, based on Wilcoxon signed rank test). (C and D) A Pearson correlation analysis of stained cells analyzing the association of ASAH1+ and MPO+ or CD68+ cells. Red dots indicate iGBM samples, whereas blue dots depict rGBM tissue slides. (E) The slides of patient 53 with an iGBM (left) and rGBM (right) were costained for ASAH1 (brown) and MPO (pink), as shown in the upper panel, as well as for ASAH1 (brown) and CD68 (pink) in the lower panel. Arrowheads indicate a coexpression of indicated proteins. Scale bar in all images = 20 µm.

Primary Neutrophils Display ASAH1 in NETs when Cocultured with Primary GBM Cells

We performed coculture experiments to further analyze the implications of neutrophils expressing ASAH1 in the context of GBM TME. Primary neutrophils were obtained from healthy donors and stimulated with either IFNɣ or TGFβ and were positively tested for viability (Figure 3A and 3B) (Supplementary methods). We aimed to mimic the immune or tumorigenic phenotype of neutrophils being exposed to either IFNɣ or TGFβ (Figure 3C), which are naturally both present in the GBM TME.³

Figure 3.

Functional expression analysis of ASAH1 in primary neutrophils and the tumor microenvironment. (A) Schematic description of the workflow to separate primary neutrophil granulocytes from buffy coats via density gradient centrifugation. Theron neutrophils were exposed to either IFNγ (50 ng/mL) or TGFβ (20 ng/mL) for 6 hours. The schema was created with BioRender.com. (B) Representative microscopic image of neutrophils in culture. Scale bar = 200 µm. After culturing the primary neutrophils, RNA was isolated and an RT-qPCR analysis for ASAH1 mRNA (C), MMP9 mRNA (D), and CD95 mRNA (E) was conducted. (F) A schematic description of co-culture experiments. Created with BioRender.com. (G) Immunocytochemistry images were taken after the coculture with GBM42 or GBM100. Coverslips were coated with Poly-D-Lysine before seeding primary neutrophils. Hoechst (H33258, Abcam, UK) stained chromatin inside the nuclei or released via NETs formation. For ASAH1 visualization, the polyclonal antibody purchased from Atlas Antibodies (HPA005468) was utilized, for MMP9, the antibody was purchased from R&D Systems (AF911). Scale bar = 20 µm. (H–K) Representative Western Blot results for ASAH1 and MMP-9 expression are shown after coculture experiments. The quantitation of biological replicates relative to the internal MPO (neutrophils) or β-Tubulin (THP1-macrophages, GBM42) signals was conducted using the software ImageJ. All results are shown with mean values ±SD of at least three repetitions and neutrophil donors marked in individual dot-shapes (n = 3) (+ = coculture; − = control). (*P-value < .05;**P-value <.01. Based on unpaired two-tailed t-test for [C–E] and one-tailed Student’s t-test for [H–K]).

Exposure to TGFβ not only induced MMP-9 expression marking the tumorigenic phenotype²⁷(p-value < .001) (Figure 3D), but also ASAH1 mRNA levels were slightly increased (Figure 3C, p-value = .057). Contrary, increased expression of CD95 (Fas receptor) in neutrophils exposed to IFNɣ indicates an immune supportive phenotype (Figure 3E, p-value < .0001). Thus, we suggest an enhanced ASAH1 expression in neutrophils exposed to TGFβ is more likely to be associated with a tumor-supportive phenotype of neutrophils.

We modeled a coculture system to better emulate the physiological conditions of neutrophil activation by GBM cells (Figure 3F). Initially, we looked for morphological changes in neutrophils after coculture (Figure 3G). The neutrophil extracellular traps (NETs) formation was significantly higher in neutrophils cocultured with GBM42 and GBM100 (Figure 3G; lower panel) as revealed by DNA staining suggesting that soluble factors released by GBM cells can induce NETs formation. Both ASAH1 and MMP-9 were found to be localized in NETs and increased in expression after coculture, as shown for four different patient-derived GBM cell lines (Supplementary Figure S4).

Western blot analyses show that neutrophils cocultured with GBM42 exhibit a slightly enhanced protein expression of ASAH1 (Figure 3H; p-value = .042) and a significant upregulation of MMP-9 (Figure 3I; p-value = .03). ASAH1 was only moderately expressed, and its expression did not change after coculturing in GBM42 (Figure 3J; p-value = .056). MMP-9 was strongly detected in GBM42, with a significantly higher expression after coculture (Figure 3K; p-value = .005).

Whereas the release of MMP-9 with NETs has been demonstrated,^3,28 to our knowledge, we hereby showcase the first description of the presence of ASAH1 in NETs. With these insights, we justify more experimental approaches regarding ASAH1 being expressed by immune cells in the TME of GBM and potentially contributing to GBM progression and resistance mechanisms. In coculture with macrophages (THP-1), expression levels of ASAH1 and MMP-9 are rather downregulated (Figure 3J) in macrophages, while only MMP-9 expression in GBM cells is highly induced (Figure 3K), as seen for all four GBM cell lines (Supplementary Figure S4D). To clarify if ASAH1 is released from NETs, we performed further biochemical experiments monitoring the activity of ELANE after either activation of NET formation with PMA (1 and 3 hours) or inhibition with selevestat (Supplementary Figure S5). NET formation was revealed by significantly higher activities of ELANE after PMA stimulation, whereas selevestat reduced ELANE activity. Notably, ASAH1 was not detected by ELISA under all experimental conditions, suggesting that despite NET formation and relocation of ASAH1 into NETs, the protein is not released into cell supernatants in vitro and thus not detected as a soluble protein. This contrasts with our in vivo observation with serum samples and suggests that in vivo NET formation requires neutrophil degradation during longer time periods or the binding of ASAH1 to NETs is weaker in the serum compartment. Furthermore, immunodetection of ASAH1 in NETs was confirmed by ELANE activity and DNA staining (Supplementary Figure S6).

Analysis of Semi-Specific Peptides Reveals Increased Proteolytic Processing of GPNMB and GFAP in rGBM

We performed a semi-specific type of database search together with in-house developed scripts for data processing to extract quantitative information on peptides produced from intrinsic proteolytic activity.²⁴ Consistently, 13 211 peptides were identified and quantified among all three TMT mixtures, among which 1121 can be considered as products of proteolysis (Figure 4A). The general percentage of proteolytic products does not differ between iGBM and rGBM (Figure 4B). Differential abundance analysis shows 21 of these with increased abundance in rGBM, while four were found to decrease (Supplementary Figure S7A, Supplementary Table S3).

Figure 4.

Analysis of proteolytic products and lipidomics reveals differential proteolytic activity and rearrangement of ceramide profiling in rGBM. (A) Number of identified and quantified peptides based on their specificity toward experimental cleavage by trypsin. Semi-specific peptides are non-tryptic either at the N-term or the C-term and are considered products of endogenous cellular proteolysis. (B) Comparison of percentage of semi-specific peptides between iGBM and rGBM. (C) Correlation of fold-changes (rGBM/iGBM) between protein abundances and their correspondingly quantified proteolytic products. (D) Protein coverage identification of GPNMB and GFAP based on semi-specific peptides. Colored bars indicate the fold-change of peptides consistently quantified in all samples. (E) Differential abundance analysis results between rGBM and iGBM based on lipidomics data. The red dashed line represents an adjusted P-value threshold of .05. (F) Over-representation test results of up-/downregulated lipid features based on Lipid Ontology. (G) Lipid Ontology over-representation test results of lipid clusters identified as highly correlated with protein abundances in rGBM, based on multiomics integrative analyses (sPLS). (H) Circle plot representing the proteins and lipids identified as highly correlated in terms of their patient-matched fold-changes between iGBM and rGBM. Dots clustered together represent features (proteins and/or lipids) with a significant positive correlation. Opposite clusters represent features with significant negative correlations. Clusters with a 90° positioning to each other indicate no correlation. (I) Gene Ontology overrepresentation test results of protein clusters identified as highly correlated with lipid abundances in rGBM, based on multiomics integrative analyses (sPLS).

The correlation analysis between log2 fold-changes of protein abundances versus the log2 fold-changes of the abundances of their associated semi-specific peptides shows a mild but significant positive correlation (Pearson score 0.5; p-value << .001). Interestingly, a considerable fraction of the differentially abundant proteolytic products is either not part of differentially expressed proteins or shows a different abundance pattern compared to the latter (Figure 4C). The low-quantitative correlation of semi-specific peptides with their associated protein abundances suggests the independence of these proteolytic products as characteristic features that deserve particular interpretative focus.

Several proteolytic products of GPNMB were identified, with one of them found in all TMT mixtures and significantly upregulated in rGBM compared to iGBM (DVYVVTDQIPVFVTM; C-terminally cleavaged; Supplementary Figure S7B). Three of these GPNMB semi-tryptic peptides arise from the same protein region, presenting a repetitive proline structure in the C-term (Supplementary Figure S7B, Supplementary Table S4). It has been reported that the cleavage of the extracellular domain (ECD) of GPNMB promotes cell migration in breast cancer cells,⁹ which suggests its potential implication in ECM rearrangement and tumor progression.

We also observed six upregulated cleaved peptides from the Glial fibrillary acidic protein (GFAP) in rGBM (6 out of 25 total hits; Supplementary Table S3). It was identified with a coverage of >80% based on semi-specific peptides (Figure 4D), which suggests high proteolytic processing. Although the general cleavage pattern for all eight GFAP semi-specific peptides combined does not appear clearly defined (Supplementary Figure S7C and 7D), we report increased proteolytic processing of GFAP in rGBM, a noteworthy observation that could point out a progression marker.²⁹ We hypothesize that the increased proteolysis of GFAP could be associated with the observed rearrangement of glial tissue and TME during rGBM and is worth further exploration.

Lipidomic Profiling of Recurrent GBM

The increased expression of ASAH1 in rGBM evoked the question of a potential association between these proteomic changes and metabolic rearrangement. The explorative lipidomics analysis of 10 patient-matched iGBM/rGBM specimens identified more than 1700 lipids, including ceramides, sphingolipids, and glycerolipids. Comparative quantitative analyses showed 206 differentially abundant lipidomic features between initial and recurrent tumors (Figure 4E; Supplementary Table S5). Sixty-one differentially abundant features were successfully mapped to a lipidomic identification (30 increased and 31 decreased in rGBM). Ceramides showed a trend toward decreased abundance, with most ceramides presenting negative fold-changes, and five of them were observed as differentially abundant (Figure 4E). Lipid Ontology analyses³⁰ revealed that downregulated lipids are mostly representatives of ceramides, sphingolipids, glycerolipids, and triacylglycerols, while upregulated features are over-represented by glycerophosphoglycerols (Figure 4F).

The proteolipidomics integrative bioinformatic analyses revealed two main lipid clusters by their strong (negative or positive) correlation with protein abundance (Figure 4G). The biggest lipid cluster (Figure 4G, yellow dots left) is overrepresented by glycerophosphocholines and glycerophospholipids (Figure 4F, C1). This group shows a positive correlation in rGBM with proteins of the HLA presentation system and transmembrane transporter activities while showing a negative correlation with proteins associated with phosphotransferase and acetyltransferase activity (Figure 4H). The second lipid cluster (Figure 4G, yellow dots, bottom) is represented by glycerolipids and triacylglycerols (Figure 4F, C2). These lipids show a strong negative correlation with transcriptional and ion channel activity proteins (Figure 4H).

The ceramide metabolism has been explored in GBM and other cancer types. Chemotherapy and radiation treatment may promote sphingomyelinases to generate ceramides from sphingomyelins, inducing apoptosis.³¹ The processing of ceramides mediated by the upregulation of ASAH1 has been found to confer resistance to apoptosis in GBM, leading to the exploration of the sphingolipid system as a therapeutic target.³² In our study, we evidence the importance of ceramide regulation in the resistance to GBM treatment, which can be emphasized as a hallmark for the development of rGBM.

Proteogenomics Reveals Expressed Single Amino Acid Variants

We generated a custom GBM-specific database from publicly available transcriptomic data,⁹ to include potentially identifiable SAAVs. The presence of 232 SAAVs was identified after a database search (Supplementary Table S6) with 214 accurately quantified in at least three samples. Only 30 of these were consistently identified and quantified in all samples (Figure 5A). Only one identified SAAV was observed with increased abundance in rGBM (FAM63A_T433K; Ubiquitin carboxyl-terminal hydrolase MINDY-1; Figure 5B–5E). The identified variant has been already described as part of a reference-quality genomic annotation of a Puerto Rican individual.³³ A Cox-proportional hazards model showed no association of FAM63A_T433K abundance in iGBM tissue with time to recurrence (Supplementary Figure S8). This suggests that the observed SAAV represents the differential abundance of a MINDY protein isoform. Overall, our workflow demonstrates the feasibility of proteogenomics for exploratory research for GBM recurrence and suggests similar levels of SAAVs in iGBM and rGBM.

Figure 5.

Proteogenomics reveals a differentially abundant single amino acid variant (SAAV). (A) Heatmap of the median normalized abundances of peptides containing one of the 30 SAAVs consistently quantified in all GBM samples. (B) Differential abundance analysis of SAAVs. The red dot represents the differentially abundant SAAV, green dots are not differentially abundant. (C) Boxplots of abundances of the FAM63A_T433K SAAV in iGBM and rGBM tumors. Lines represent the association of observations in the same patient. (D) Diagram of the identified peptide sequence containing the FAM63A_T433K SAAV. Numbers in brackets at the N-term represent the experimental biochemical modification by TMT. The fact that most amino acids were identified by both y(orange)- and b(green)-ions indicates a good quality peptide identification based on a tandem mass spectrum. (E) Tandem mass spectrum used for the identification of the peptide containing the FAM63A_T433K SAAV.

Plasma Levels of ASAH1 at iGBM Surgery Show Association with Time to Recurrence

The observation of differential proteomics abundance of ASAH1, SYNM, and GPNMB led us to question whether these proteins could have the potential to act as molecular signatures of tumor progression in plasma. The plasma concentrations of selected proteins were determined in samples from ten healthy donors and compared to fourteen GBM patients before the first (iGBM) and second (rGBM) surgery, using ELISA. ASAH1, SYNM, and GPNMN were selected based on their proteomics behavior (Figure 1). As for MMP-9, coculture experiments indicated an intriguing connection of ASAH1 to MMP-9 expressing tumorigenic neutrophils (Figure 3), and recent studies propose a predictive value of plasma MMP-9 in GBM patients.²⁷ASAH1 tends to be more abundant in plasma of rGBM patients, which is consistent with the proteomics behavior. On the other hand, MMP-9 concentration tends to attenuate in rGBM cases. GPNMB features comparable plasma levels in control, iGBM, and rGBM samples. SYNM shows a significant increase in concentration in iGBM compared to controls (Figure 6A). We then explored the association between the plasma concentrations of these proteins at initial surgery (iGBM) with TTR in GBM patients. Interestingly, the Kaplan–Meier analyses show that higher concentrations of ASAH1 and SYNM in iGBM plasma show a positive association with TTR (Figure 6B). Following this trend, the application of a multivariate Cox-proportional hazards model shows that decreased concentrations of ASAH1 are associated with decreased TTR, although for SYNM the model does not show its significant association with TTR (Figure 6C). It is worth emphasizing that this experiment is limited to a sample size of 14 GBM patients and only two-time points (pre-surgery iGBM and pre-surgery rGBM). Albeit preliminary and exploratory, the significant effect of ASAH1 and SYNM concentrations on TTR showcases the potential use of these protein markers in post-surgery follow-up schemes.

Figure 6.

Plasma proteomics signature for classification of recurrent glioblastoma multiforme. (A) The log2 transformed plasma protein concentrations of ASAH1, MMP-9, GPNMB, and SYNEMIN in serum from control, iGBM, and paired rGBM samples determined by ELISA, respectively. (B) Kaplan–Meyer plots show the association between higher ASAH1 and SYNM abundance in plasma with a longer time to recurrence. (C) Cox-proportional hazards ratio model results. When holding other variables constant, lower concentrations of ASAH1 are associated with a shorter time to recurrence.

Conclusion

To our knowledge, we report the first integrative proteo-lipidomics analysis focused on recurrent glioblastoma in a patient-matched human cohort. The proteomics analyses revealed an up-regulation of ASAH1, GPNMB, and SYNM in rGBM; an observation consistent with an immune fingerprint and extracellular matrix rearrangement. The large-scale terminomics analyses revealed the differential abundance of products of endogenous proteolysis (cleavage of GFAP and GPNMB); this substantiates a hypothesis of cellular and extracellular rearrangement in support of tumor resistance and progression. After functional exploration, an association of ASAH1 abundance with neutrophil and macrophage infiltration was observed especially in rGBM. Furthermore, coculture experiments suggest the presence of ASAH1 in neutrophil extracellular traps (NETs) when exposed to GBM cells. Moreover, our metabolomic analyses revealed a significant downregulation of ceramides and sphingolipids in rGBM. In terms of progression markers, there is evidence of association between plasma concentrations of ASAH1 during iGBM with TTR, as observed by our Kaplan–Meyer analyses and Cox proportional hazards models. It is important to acknowledge that this study represents initial observations based on a limited sample size. Nevertheless, we consider these findings to reinforce the idea that the integration of multiple omics data sources can enhance the identification of potential therapeutic targets and aid in the development of personalized treatment strategies for patients with recurrent glioblastoma. The identification of novel biomarkers and pathways may also contribute to the development of more effective diagnostic and prognostic tools, which may improve patient outcomes and ultimately lead to a better understanding of this devastating disease.

Conflict of interest statement

None declared.

Funding

Authors acknowledge funding by the Deutsche Forschungsgemeinschaft (DFG, projects 446058856, 466359513, 444936968, 405351425, 431336276, 43198400 (SFB 1453 “NephGen”), 441891347 (SFB 1479 “OncoEscape”), 423813989 (GRK 2606 “ProtPath”), 322977937 (GRK 2344 “MeInBio”), the ERA PerMed program (BMBF, 01KU1916); the ERA TransCan program (project 01KT2201,“PREDICO”), the German Consortium for Translational Cancer Research (project Impro-Rec), the investBW program (project BW1_1198/03 “KASPAR”), and the BMBF KMUi program (project 13GW0603E, project ESTHER) to OS; JWB was funded by the ERA PerMed program PerProGlio (BMBF, 01KU1916B), by the Deutsche Forschungsgemeinschaft (DFG), project BA1606/3-1, and by a Research Grant from the University Medical Center Giessen and Marburg (UKGM). LC is funded by the von Behring-Röntgen Foundation, grant number 70_00034. M.F. was funded by the Medical-Scientist-Programme, Faculty of Medicine, University of Freiburg. The authors acknowledge the support of the Freiburg Galaxy Team: Björn Grüning, Bioinformatics, University of Freiburg (Germany) funded by the Collaborative Research Centre 992 Medical Epigenetics (DFG grant SFB 992/1 2012) and the German Federal Ministry of Education and Research BMBF grant 031 A538A de.NBI-RBC.

Data Availability

Spectral files used for peptide and protein identification and quantitation can be accessed via the European Phenome Archive (EGA) using the following link https://ega-archive.org/studies/EGAS00001006395, under restricted access (due to human data sharing restrictions). Requests for data access will be referred directly to our Data Access Committee associated with EGA https://ega-archive.org/dacs/EGAC00001002750. The Reproducible reports and R scripts used for the processing and analysis of proteomics, lipidomics, and plasma data are available via Zenodo 24. The proteogenomics pipeline can be accessed via Galaxy through the link: https://usegalaxy.eu/published/workflow?id = 438dc46201bb8996. All other data are available from the corresponding authors (J.W.B. and O.S.) upon reasonable request.

Author contributions

Conceptualization, experimental design and execution: J.W.B., O.S. Data processing integration, interpretation, and manuscript writing: M.C.C., A.S., J.W.B., O.S. Sample collection: A.S., J.W.B, D.B., Z.H., O.K., K.Z., S.S., A.V.G., F.M.B., J.F.L., G.R., L.C.A., L.M.B., C.N. Mass spectrometry for proteomics: sample preparation: C.Y.C., C.N., A.S., S.S., LC-MS: M.L.B., data analysis: M.C.C., N.P., M.N.S. Mass spectrometry for lipidomics: sample preparation: M.F., L.M., measurement: N.C.W., J.B., data analysis: J.B., M.C.C. Experimental and functional analyses: A.S., J.V.H., D.B., O.K., Z.H., K.Z., L.C., S.S.,R.I., T.W.; Statistics and modeling: J.S., M.C.C., H.R., A.A., A.V.S.