ABSTRACT: Introduction: Cerebrospinal fluid (CSF) is the proximal body fluid to the brain and contains biomarkers for pathological brain processes. Discovery of protein biomarkers by mass spectrometry (MS) is challenging due to the high abundance range of CSF proteins, issues with sampling and processing, and the complex interplay of pathology and human variability. We previously quantified more than a thousand proteins in about 200 samples; establishing a promising biomarker panel for Alzheimer’s disease. However, due to technological constraints, studies including ours typically had to compromise on the number of quantified proteins or the sample size. Here, we developed and optimized a high-throughput label-free workflow which we validate by analyzing thousands of CSF samples to focus on study size scalability. Methods: Our high-throughput CSF proteomics workflow uses a streamlined, data independent acquisition (DIA) strategy. It is built around the Evosep chromatography system run with 21-minute gradients (60 samples per day) and the Bruker timsTOF Pro 2. MS-Fragger and DIA-NN were chosen for library and DIA data analysis, respectively. We optimized workflow aspects including sample preparation, MS acquisition, spectral library generation, raw data processing and normalization algorithms. We demonstrate scalability by measuring more than 6000 CSF samples from patients with Multiple Sclerosis, and diverse neurological conditions. Additionally, we investigated factors causing analytical variability and used differential expression and machine learning approaches to characterize how pathology, age, sex, disruption of blood brain barrier (BBB), and sampling-related biases impact the CSF proteome. Preliminary data: We found that the above protocol – combined with in-solution digestion, off-line peptide clean-up and single-run analysis, yielded optimal CSF proteome depth. It permitted sample preparation streamlining for eight 96-well plates at a time. Furthermore, we enhanced protein identification by adapting the DIA acquisition scheme to the CSF precursor distribution in the m/z dimension and focusing on enhanced ion mobility separation instead of covering the complete ion mobility range. We generated a CSF-specific DIA spectral library of initially 3000 proteins using MS-Fragger, by depletion of abundant proteins, 96 high-pH fractions, data-dependent measurement, complemented by deep-leaning peptide intensity predictions of the entire proteome. These libraries were pruned to about 2,000 proteins experimentally observed in cohort samples, which enabled processing in only three days. We achieved stable identification of about 1200 proteins per sample across the cohort, totaling 2000 proteins in the entire dataset, of which 1000 were quantified in more than 75% of all samples. Quantitative precision was high (CV 13% intraplate and 23% overall). Quality control samples revealed a drastic effect of blood contamination on CSF proteomes; we excluded affected patient samples. The impact of circadian sampling time was minor. Across the cohort, disease and the covariates sex, age, and disease-unspecific BBB impairment accounted for 11, 6, 11, and 38% of variability in the dataset, respectively. The BBB impairment had an indirect effect on the relative abundance of the other CSF proteins (those not expected to leak from blood to CSF). In the CSF from Multiple Sclerosis patients, we found an immunoglobulin signature, which was expected and provides positive control. Another set consisted of known and novel proteins, that were dominantly MS associated but partially also perturbed in other conditions including autoimmunity or infection. Novel aspect: We present a powerful CSF protein biomarker discovery workflow amenable to unprecedented study sizes at good proteome coverage and quantitation.
Mario Oroshi 