Leveraging the CSF proteome toward minimally-invasive diagnostics surveillance of brain malignancies

Abstract

Background Diagnosis and prognostication of intra-axial brain tumors hinges on invasive brain sampling, which carries risk of morbidity. Minimally invasive sampling of proximal fluids, also known as liquid biopsy, can mitigate this risk. Our objective was to identify diagnostic and prognostic cerebrospinal fluid (CSF) proteomic signatures in glioblastoma (GBM), brain metastases (BM), and primary central nervous system lymphoma (CNSL). Methods CSF samples were retrospectively retrieved from the Penn State Neuroscience Biorepository and profiled using shotgun proteomics. Proteomic signatures were identified using machine learning classifiers and survival analyses. Results Using 30 µL CSF volumes, we recovered 755 unique proteins across 73 samples. Proteomic-based classifiers identified malignancy with area under the receiver operating characteristic (AUROC) of 0.94 and distinguished between tumor entities with AUROC ≥0.95. More clinically relevant triplex classifiers, comprised of just 3 proteins, distinguished between tumor entities with AUROC of 0.75-0.89. Novel biomarkers were identified, including GAP43, TFF3 and CACNA2D2, and characterized using single-cell RNA sequencing. Survival analyses validated previously implicated prognostic signatures, including blood brain barrier disruption. Conclusions Reliable classification of intra-axial malignancies using low CSF volumes is feasible, allowing for longitudinal tumor surveillance.

Systematic Review of Cerebrospinal Fluid Biomarker Discovery in Neuro-Oncology: A Roadmap to Standardization and Clinical Application

“…63 There are also more unique proteins in ventricular than lumbar CSF, and proteins detected at both sites have higher concentrations in ventricular CSF (Fig 1A). 64,65 These differences affect diagnostic performance, as seen in patients with GBM where diagnostic miRNA biomarkers were more specific in lumbar CSF, but more sensitive in cisternal CSF. 66 Similarly, diagnostic classifiers trained using ventricular CSF had lower performance when applied to lumbar CSF.…”

“…55,70 BBB disruption has a characteristic plasma protein signature in CSF that is prevalent across CNS malignancies, including GBM, PCNSL, and BrM. 64,65 The extent of disruption indicates disease severity and is negatively associated with survival. 64,65 There are several strategies to identify BBB disruptionindependent markers.…”

“…64,65 The extent of disruption indicates disease severity and is negatively associated with survival. 64,65 There are several strategies to identify BBB disruptionindependent markers. The protein-albumin quotient Q normalizes query CSF proteins by CSF albumin levels, thereby controlling for BBB-related leakage:…”

“…101 Single-cell transcriptomic sequencing can also establish cell origin of RNA and proteomic biomarkers. 64 However, caution should be exercised when interpreting transcriptomic data because of potential discordance between RNA and protein expression. Biomarkers can also be identified in conditioned media from in vitro tumor cultures; however, this does not guarantee tumor specificity.…”

“…These represent a BBB disruption signature that can monitor the severity and prognosis of disease. 64,65 miR-21 miR-21 is an ubiquitously expressed miRNA that has been claimed as a biomarker for at least 29 unique diagnoses. 104 In CNS malignancies, it is detected in CSF supernatant and EVs/ exosomes, and is elevated in CSF from patients with PCNSL, glioma, and BrM (Data Supplement, Table S6).…”