Deep Proteome Analysis of Cerebrospinal Fluid from Pediatric Patients with Central Nervous System Cancer

Abstract

The cerebrospinal fluid (CSF) is a key matrix for discovery of biomarkers relevant for prognosis and the development of therapeutic targets in pediatric central nervous system malignancies. However, the wide range of protein concentrations and age-related differences in children makes such discoveries challenging. In addition, pediatric CSF samples are often sparse and first prioritized for clinical purposes. The present work focused on optimizing each step of the proteome analysis workflow to extract the most detailed proteome information possible from the limited CSF resources available for research purposes. The strategy included applying sequential ultracentrifugation to enrich for extracellular vesicles (EV) in addition to analysis of a small volume of raw CSF, which allowed quantification of 1351 proteins (+55% relative to raw CSF) from 400 μL CSF. When including a spectral library, a total of 2103 proteins (+240%) could be quantified. The workflow was optimized for CSF input volume, tryptic digestion method, gradient length, mass spectrometry data acquisition method and database search strategy to quantify as many proteins a possible. The fully optimized workflow included protein aggregation capture (PAC) digestion, paired with data-independent acquisition (DIA, 21 min gradient) and allowed 2989 unique proteins to be quantified from only 400 μL CSF, which is a 340% increase in proteins compared to analysis of a tryptic digest of raw CSF.

Keywords: CSF input volume; biomarker; cerebrospinal fluid; extracellular vesicle; protein aggregation capture.

Figure 1

Experimental design and workflow (created with BioRender.com). (A) Pooling of CSF in subsets of patients with ALL, a brain tumor or nontumor control. These disease-specific subgroup-pools were pooled 1:1:1 prior to each experiment and sequential ultracentrifugation. (B) workflow for tryptic digest methods. (C) Data acquisition modes. (D) workflow for data analysis. Created using BioRender.com.

Figure 2

(A) Heatmap of unsupervised clusters comprising all fractions. The experiment was repeated two times (“Initial” is the first experiment, while “Repeated” was the same experiment 3 months later). (B) Dynamic range and proteome depth stratified for sample input volume and each fraction. The intensity corresponds to median of MS2-intensity across the workflow triplicates. (C) Predicted subcellular location using the DeepLoc algorithm. (D) Cryo-transmission electron microscopy of 20K and 120K EVs. (E) Western blotting probed for common EV markers. In this experiment, equal sample masses were loaded. (F) Comparison of MS2-intensities between fractions with emphasis on EV markers. In this experiment, equal volumes were loaded. (G) Nanoparticle tracking analysis of 20K and 120K EVs stratified for patients with ALL or a brain tumor as well as nontumor controls. (H) Biomarker coverage of previous reported potential biomarkers, summarized in ref (10).

Figure 3

(A) The total number of unique proteins was analyzed using a linear regression model, with adjustments for the mass loaded onto the EvoTip and the CSF input volume. This model evaluated the impact of these variables on the number of unique proteins while accounting for their interrelationships. P-values for statistical significance are provided below each estimate to indicate the significance of the CSF input variable. (B) Reproducibility based in workflow triplicates and measured by the coefficient of variation across different sample input volumes. (C) Total number of proteins quantified with and without application of the gas-phase spectral library. Stratified for input sample volume and volume-based load. (D) Total number of proteins quantified within each fraction as a function of peptide mass loaded (measured on A280). (E) Total sample hydrophobicity (all peptides summed) measured as Gravy score and as a function of peptide mass loaded (A280 based).

Figure 4

(A) Venn diagram of total numbers of proteins identified across all fractions and by each digest method. The identified proteins were quantified from an input sample volume of 400 μL and search via the gas-phase spectral library. (B) The dynamic range and proteome depth in each digest and protein extraction method in within the individual fractions, stratified for usage of the gas-phase spectral library. All proteins identified were included. (C) The same as B, but now including the addition of unique proteins identified within each fraction. I.e., the unique proteins were allocated to the fraction in which it was first observed. (D) A Venn diagram illustrating the overlap between 20K and 120K EVs identified using the PAC-based vs urea-based digest (using the gas-phase spectral library). (E) Unique proteins identified in the 120K and EV-depleted fraction on top of what have been identified in the Raw +20K EV fraction. Here, only the PAC-based and urea-based approach and search using the gas-phase spectral library. (F) Missed cleavages in each fraction while stratified for digest and protein extraction method as well as usage of the gas-phase spectral library. (G) Similarly stratified, here showing unique proteins and unique peptides identified in each fraction. (H) Total sample hydrophobicity based on all peptides summed. The hydrophobicity was measured as the Gravy score. Stratified for each method used for protein extraction and fraction.

Figure 5

(A) Unique proteins identified in each fraction, here stratified for gradient length (100, 60, 40 SPD) while comparing DDA vs DIA, with and without the gas-phase spectral library using the sample pool. (B) The same, but for unique peptides. (C) Number of unique proteins identified in this study vs previous literature. (D) The efficiency of quantifying the number of unique proteins, here measured as . (E) The overlap between unique proteins identified in this study and compared to previous studies on CSF from healthy donors. (G) The same, but the CSF study was performed on EV-enriched samples from pediatric patients with a brain tumor.

Figure 6

Methodology applied to individual samples (in contrast to pooled samples). (A) A heatmap based on unsupervised clusters. Here showing that PAC-based on-bead digest performs instable compared to the urea-based in-solution digest. (B) Number of unique proteins identified across each fraction and within the PAC-based on-bead and urea-based in-solution digest. (C) Volcano plots comparing the log 2-transformed MS2-intensities obtained from analyzing four FCM-positive and four FCM-negative patients, here including Raw, 20K EVs and 120K EVs for both PAC-based on-bead and urea-based in-solution digest.