Impact of Variations in Commercial Standard Operating Procedure on Plasma Proteome Recovery

Abstract

Advancements in mass spectrometry and complementary technologies now enable comprehensive, high-resolution plasma proteomics. Plasma is a key biofluid for clinical research, harboring potential disease-informative biomarkers. Commercial sources of nondiseased, healthy donor plasma samples are often used for proteomic workflow development and as controls for clinical studies. The overarching assumption is that standard operating procedures for plasma preparation are comparable across different sources. In this study, we investigated the effectiveness of a particle-based protein enrichment strategy against a conventional proteomics workflow on plasma samples from five commonly used commercial sources. We aimed to characterize the extent of variability in plasma proteomes when factors such as freeze–thaw cycles, choice of anticoagulant, and operator-to-operator performance were accounted for in the study setup. Plasma samples were analyzed in data-independent acquisition mode by using two distinct instruments (Exploris 480, timsTOF HT) and search algorithms (CHIMERYS, DIA-NN). Plasma proteome enrichment yields ranged from 2.8- to 6.2-fold more compared to the conventional workflow, achieving yields exceeding 5000 proteins (timsTOF HT). Notably, the observed variability in proteome composition was largely attributable to differences in whole blood-to-plasma operating procedures across commercial sources. While these distinct proteomes remained undetectable with conventional workflows, particle-dependent proteome profiling successfully revealed the procedural differences.

Introduction

In the past 5 years, breakthroughs in biochemical workflows and mass spectrometry innovations have transformed plasma proteomics, enabling high-throughput sequencing that extends far beyond the top 300 abundant proteins. Biochemical strategies that increase the dynamic range of identified proteins include precipitation/depletion of abundant proteins, such as apolipoproteins, − enrichment of plasma extracellular vesicle fractions, and particle-dependent protein corona-forming technologies that enrich proteins based on their chemical properties. , In parallel, mass spectrometers are operating at increasingly faster scan speeds and with improved sensitivity, permitting shorter run times and increased throughput. ,,, These technologies provide various workflow options for plasma and other biofluid research.

Plasma is the primary biofluid for clinical research. Over 54,000 trials at clinicaltrials.gov include the term “plasma” in the study design, representing >10% of all registered trials. Serum (9%), urine (4%), sputum (<1%), and spinal fluid (<1%) represent other biofluids that together provide mass spectrometrists with many opportunities to implement their technologies. Our research interests focus on heart and vascular diseases, entailing studies that rely primarily on explanted tissues from patients undergoing surgical repair of a vessel or valve. − In most of our studies, healthy/nondiseased control tissues are severely limited (if at all obtainable), and autopsy samples are not necessarily suitable alternatives since the tissue and molecular integrity are not guaranteed. In addition, access to control subjects’ plasma, often “healthy donor” samples, is also limited. However, biological specimen banking initiatives and commercial sources facilitate acquisition of plasma from healthy/nondiseased donors. In fact, healthy donor plasma samples are readily accessible from a variety of commercial sources, providing convenient supplies of control specimens not only for clinical studies but also for workflow optimization. Standard operating procedures (SOPs) for specimen collection and storage as part of a clinical study or trial design are pivotal to minimize or account for potential alterations to the specimens’ molecular composition. While we can control SOPs internally, it is imperative that we confirm those of potential external sources.

This report provides a paramagnetic bead-based low-abundant protein enrichment strategy through seeded protein precipitation to evaluate protein recovery from plasma samples derived from five commercial sources within the United States, with the primary aim to evaluate sources of technical variability, including intraoperator variation and potential differences in plasma processing SOPs.

Materials and Methods

Study Design and Statistical Rationale

Human plasma samples were purchased from five independent commercial sources (Sources 15). Each source included samples from two male and two female donors, with ages evenly distributed between 20 and 65 years. All samples were ordered to be shipped frozen, but samples from one source (Source 4) arrived at room temperature. We noted this discrepancy and continued with the sample preparations, as planned. While Sources 1 to 4 used K2-ethylenediaminetetraacetic acid (K2EDTA) as the anticoagulant, ACD-A (acid-citrate-dextrose solution A) was the anticoagulant for Source 5. We further requested a minimum number of freeze–thaw cycles and no filtering of plasma if it was an option on the order form (Source 1 only).

To reduce confounding results due to potential donor outliers, we pooled each source sample and aliquoted the pool into 4 reactions for each plasma preparation (4 ENRICHplus vs 4 iST-BCT, see below). To evaluate the extent to which the workflow operator impacted proteome yield or consistency, 12 aliquots of the Source 1 pool were made for the ENRICHplus workflow. Researchers T.K., L.W., and J.W. performed four replicates each, whereas only J.W. performed the ENRICHplus workflow for the remaining source samples. These preparations were performed on the same day. Researcher T.K. performed the iST-BCT kit preparations using pooled aliquots generated at the time of the initial sample preparation.

Plasma Proteolysis Steps

Plasma proteome enrichment and proteolysis were performed by using the ENRICHplus kit (PreOmics, GmbH, Germany). For each source, the 4 plasma samples were pooled and 4 aliquots of 50 μL for Sources 2 to 5, and 12 aliquots for Source 1 were created. The 50 μL aliquot was used as the input for the ENplus-BEADS. The remaining steps of the protocol were performed according to the manufacturer’s instructions. The eluted peptides were dried in a speed vacuum (Eppendorf Vacufuge) at 45 °C and resuspended in 22 μL of LC-LOAD provided by the kit. For neat plasma proteomics, 2 μL of plasma was prepared using the iST-BCT kit according to the manufacturer’s protocol (PreOmics, Germany). The dried, eluted peptides were resuspended in 42 μL of LC-LOAD. For each neat and ENRICHplus sample, the peptide concentration was determined using a NanoDrop One Spectrophotometer (Thermo Fisher Scientific). Each peptide sample was diluted to 100 ng/μL using LC-LOAD before injection into the liquid chromatography-tandem mass spectrometry (LC-MS/MS).

LC-MS/MS Analysis

Tryptic peptides were analyzed using one of two instruments, either the quadrupole Orbitrap Exploris 480 coupled to a Vanquish Neo UHPLC system (Thermo Fisher Scientific, ENRICHplus, and iST-BCT preparations) or the timsTOF HT coupled to a proteoElute system (Bruker, ENRICHplus preparations).

For the Orbitrap Exploris 480 acquisitions, peptides (400 ng on column) were fractionated using a dual column setup: a PepMap Neo C18 Trap Cartridge, 5 μm, 300 μm × 5 mm (Thermo Fisher Scientific); and an Easy-Spray PepMap Neo C18 Column, 2 μm, 75 μm × 150 mm (Thermo Fisher Scientific). The column was heated at a constant temperature of 45 °C. The gradient flow rate was 300 nL/min from 5 to 21% solvent B (95% acetonitrile/0.1% formic acid in mass spectrometry-grade water) for 20 min, 21–30% solvent B for 10 min (total acquisition, 30 min), and another 15 min of a 95–5% solvent B sawtooth wash. Solvent A was 0.1% formic acid in water. The mass spectrometry analysis was performed by using data-independent acquisition (DIA). MS1 scans were acquired in profile mode, and MS2 scans were acquired in centroid mode. Spray voltage was set to 2000 V, funnel radio frequency (RF) level at 50, and heated capillary temperature at 275 °C. MS1 was with 120 K resolution, and mass scan range was set to m/z 400–900. The AGC target was set to standard with a maximum injection time of 25 ms. MS2 spectra were acquired with a precursor isolation range of m/z 400–900 divided into 12 Th windows with an overlap of 1 Da. MS2 resolution was set to 15 K and the AGC target value for fragment spectra was set at 1000% with a maximum injection time of auto. Higher-energy collisional dissociation (HCD) collision energy was set to 26%.

For the timsTOF HT acquisitions, peptides (200 ng on column) were separated using an IonOpticks 25 cm × 75 μm, 1.7 μm particle size C18 column over a 30 min gradient. The gradient flow rate was 250 nL/min from 5 to 23% solvent B (acetonitrile/0.1% formic acid) for 18 min, 23–35% solvent B for 4 min, and 35–90% B for 4 min. The column remained at 90% B for an additional 4 min. Solvent A was 0.1% formic acid in mass spectrometry-grade water. The column was held at a constant temperature of 50 °C. The mass spectrometry analysis was performed using parallel accumulation–serial fragmentation coupled to DIA (dia-PASEF). The MS1 scan range was set to m/z = 100–1700. Trapped ion mobility mass spectrometry (TIMS) settings were as follows: 1/K ₀ start and end were set to 0.75 and 1.3 (V·s)/cm², respectively, with the ramp and ion accumulation times set to 50 ms (100% duty cycle). The dia-PASEF MS/MS scheme consisted of 60 variable width and height windows spanning an m/z range of 350.7–1250.6 m/z and a mobility range of 0.6–1.45 1/K ₀, resulting in a cycle time of 1.18 s with 20 MS/MS ramps per MS scan.

Mass Spectral Processing

The Exploris 480 spectra were analyzed using Proteome Discoverer (PD, version 3.2 with CHIMERYS, version 4.0), Thermo Fisher Scientific or DIA-NN version 2.0 Enterprise. The timsTOF HT spectra were analyzed using a DIA-NN 2.0 Enterprise.

PD3.2+CHIMERYS

The spectra were queried against the UniProt human database (downloaded May 2025; 42,421 entries). Trypsin/P was set as the digestion enzyme, allowing up to 2 missed cleavages and a minimum peptide length of 7–30 amino acids (1–6 peptide charge). Oxidation (+15.995 Da) of methionine was set as a variable modification, and carbamidomethylation (+57.021 Da) of cysteine was set as a static modification. The Quantification Type was set to the default setting MS2 Apex (Quan in all files).

DIA-NN

The spectra were queried against the same Uniprot human database as above, in library-free mode. Trypsin/P was set as the digestion enzyme, allowing up to 1 missed cleavage and a minimum peptide length of 7–30 amino acids (1–4 peptide charge). N-terminal methionine excision and carbamidomethylation (+57.021 Da) of cysteine were considered. For the algorithm settings, match between runs and protein inference were selected. The quantification strategy was set to Quant UMS (high precision), and the cross-run normalization was retention time-dependent.

Our study reports proteins that are supported by a single peptide (1% false discovery rate (FDR)) that also had a quantification value in at least 1 of 4 replicates. Figure S1A comprises a summary of the consensus proteomes and the figures in which they were used.

Platelet Enzyme-linked Immunosorbent Assays

Quantification of plasma P-selectin (CD62P, DY137, R&D Systems) and platelet factor 4 (PF4, DY795, R&D Systems) levels was performed according to the manufacturer’s instructions. Individual donor plasma samples were assayed, but measurements for Source 5 samples exceeded the upper limit of detection of the assay and thus were excluded from the analysis.

Statistical and Gene Set Enrichment Analyses (GSEA)

Data processing and statistical analyses were conducted using R (ver. 4.4.3) (R Core Team, 2025) in a specific computational environment with various packages. The Tidyverse suite of packages (Wickham et al., 2019) was employed, which includes dplyr (ver. 1.1.4) for data manipulation, tidyr (ver. 1.3.1) for data reshaping, and readr (ver. 2.1.5) for data import. To clean and standardize the data, the janitor package (ver. 2.2.1) (https://github.com/sfirke/janitor) was utilized, and summary statistics were calculated using the rstatix (ver. 0.7.2) package (https://rpkgs.datanovia.com/ggpubr/).

For data visualization, ggplot2 (ver. 4.0.0) was used, along with additional extensions such as ggpubr (ver. 0.6.1) (https://rpkgs.datanovia.com/ggpubr/) to create publication-ready visuals and ggthemes (ver. 5.1.0) (https://jrnold.github.io/ggthemes/) for styling. Protein overlap among different sources was illustrated through Venn diagrams created with ggVennDiagram (ver. 1.5.4). Multipanel figures were assembled using the patchwork package (ver. 1.3.2) (https://patchwork.data-imaginist.com), which allowed for the combination of multiple ggplot objects with customized layouts.

Principal component analysis (PCA) was performed on log₁₀-transformed protein abundance matrices using the prcomp­() function in base R. Only proteins with complete abundance data across all source–method combinations were included in the analysis. Pearson correlation coefficients between protein abundance profiles from different sources and analytical methods were calculated with the cor­() function in R. The significance of these correlations was assessed using the stat_cor­() function from ggpubr, obtaining p-values with the cor.test­() function. Heatmaps of correlation matrices were generated using pheatmap (ver. 1.0.13) (https://CRAN.R-project.org/package=pheatmap), applying hierarchical clustering with complete linkage and Euclidean distance. Hierarchical clustering based on correlation distances (1correlation coefficient) was performed using the hclust­() function, and the optimal number of clusters was determined by cutting the dendrogram to create k = 2 clusters with the cutree­() function.

Differential protein expression between source groups (group 1: Sources 1–2 versus group 2: Sources 3–5) was evaluated using two-sample t tests. log₂ fold changes were computed as the difference in the mean log₁₀ abundances between the groups. p-Values were adjusted for multiple testing using the Benjamini–Hochberg false discovery rate (FDR) method via the p.adjust­() function. Proteins with an adjusted p-value <0.05 and an absolute log₂ fold change >1 were considered significantly differentially expressed. Volcano plots were created to visualize the relationship between fold change and statistical significance.

For functional enrichment analysis, the fgsea package (ver. 1.32.2) was utilized, implementing the fast Gene Set Enrichment Analysis algorithm. Reactome pathway gene sets were obtained from the Molecular Signatures Database (MsigDB) via the msigdbr package (ver. 25.1.1) (https://igordot.github.io/msigdbr/), specifically selecting C2 curated gene sets from the CP:REACTOME subcollection. Proteins were ranked by log₂ fold change, and enrichment was assessed using fgseaMultilevel­() with default parameters (minSize = 15, maxSize = 500). Pathway significance was determined with FDR-adjusted p-values <0.05.

Data Availability

The mass spectrometry and resulting search data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the data set identifiers PXD069212 and PXD069325.

Results

Study Goals

We sought to determine whether plasma procured from distinct sources could potentially impact protein yields when using a bead-dependent plasma proteome enrichment strategy, in this case, enlisting the ENRICHplus method. Plasma samples were purchased from five sources within the United States. Specifications for the plasma orders are detailed in the Study Design and Statistical Rationale section.

For each source, a plasma pool was created and then divided into 4 × 50 or 4 × 2 μL aliquots for the bead-based enrichment (ENRICHplus kit) or conventional neat (iST-BCT kit) plasma proteolysis preparations, respectively (Figure A). One researcher performed all ENRICHplus preparations, but to evaluate potential operator-to-operator bias in sample preparations, the plasma pool from Source 1 was divided into 12 aliquots permitting the 3 lead researchers (T.K., L.W., J.W.) to independently perform the bead-based enrichment protocol. Peptides were analyzed using data-independent acquisition (DIA) mass spectrometry, performed on the benchtop quadrupole Orbitrap (Exploris 480) or the timsTOF HT (diaPASEF). We confirmed that 400 ng (200 ng for the timsTOF HT) generated a strong and consistent base peak signal on the Exploris 480 (Figure B).

1.

Study overview. (A) Plasma sample collection and processing strategy. Four individual donor plasma samples were procured from five sources. Plasma samples were pooled and then divided into four aliquots of 50 μL for the ENRICHplus (bead-dependent protein enrichment) or 2 μL for the iST-BCT (neat plasma) technical replicates. *Source-1 was aliquoted for 12 replicates for the ENRICHplus workflow. Tryptic peptides were analyzed using data-independent acquisition (DIA) strategies performed on the quadrupole Orbitrap (Exploris 480) or the timsTOF HT system. Data were analyzed using CHIMERYS via Proteome Discoverer or DIA-NN. PASEFparallel accumulation–serial fragmentation. (B) Base peak traces for representative ENRICHplus replicates analyzed on the Exploris 480 (NL, normalized level).

Plasma Protein Numbers Vary with Commercial Source

To streamline interinstrument comparisons, we used DIA-NN for spectral annotations, but CHIMERYS was also used for the Exploris 480 spectra (Figure S1A). We generated single (source-specific, Figure S1B) and combined (five sources) consensuses to evaluate the impact of complementary spectral assignments on a given data set (Figure A). The most striking result of the ENRICHplus output is the number of proteins distinguishing Sources 1 and 2 from Sources 3 to 5. For the Exploris 480 data, we recovered 720/1062 (CHIMERYS/DIA-NN) and 787/1112 proteins for Sources 1 and 2 proteins, respectively, whereas those from Sources 3 to 5 exceeded 1500/2300 proteins (Figure A, Tables S1 and S2). The largest disparity in protein numbers is between Sources 1 (720/1062) and 3 (1682/2558). While we anticipated some differences in proteome yields, those exceeding 2-fold were not expected. Due to the higher number of protein identifications for Sources 1 and 2 achieved with the timsTOF HT compared with the Exploris 480, the disparity between sources is less prominent but nonetheless notable (e.g., 3480 vs 5434 for Source 1 versus Source 3, respectively; Table S3). Despite the lower number of Exploris 480 protein identifications, most had in common with the timsTOF HT data. For example, when considering the DIA-NN data set, 2647 of 2709 (98%) Exploris 480 proteins were identified in the timsTOF HT data set (Figure B).

2.

ENRICHplus plasma proteomes. (A) Number of proteins quantified for each mass spectrometer platform and spectral annotation software. (B) Commonly and uniquely identified proteins from panel (A). (C) Commonly and uniquely identified proteins considering only the Exploris 480 data (CHIMERYS). (D) corresponding principal component analysis for panel (C) (n = 4 replicates per source).

For the main analyses, we focused on the Exploris 480 (CHIMERYS) data. In the ENRICHplus workflow, 29% (511 of 1792) of the proteome was quantified in all five data sets, with data sets 3 to 5 sharing 35% (621) of the proteome that was not identified in data sets 1 and 2 (Figure C). Principal component analysis (PCA) shows that the workflow replicates clusters by commercial source (Figure C). We examined more closely potential operator-to-operator bias by evaluating the distribution of the average protein abundance in the Source 1 plasma proteome that was prepared by three researchers (three researchers prepared four replicates each, Figure S2A–C). Protein recovery was comparable across Source 1 replicates, with 85% of proteins (487 of 572 proteins) common to all 3 researcher data sets, and a Pearson correlation of ≥0.945 when the protein abundances were compared. Although we identified interoperator variability between the average abundance of the 487 commonly identified proteins, the calculated effect size was small (Kendall’s W = 0.0385, Figure S2). These results suggest that while there is a significant difference between researchers, the interoperator effect falls within the negligible range as compared to the variation we observed (Figure D).

Neat Plasma Proteomes Support Commercial Source as a Major Source of Variability

The ENRICHplus workflow improved proteome yields from 2.8-fold (Source 1) to 6.2-fold (Source 3) greater than the conventional neat plasma method (Exploris 480+CHIMERYS analyses; Table S4). Specifically, the neat plasma proteome yields were between 237 and 336 proteins (Figure A). Source 5’s neat proteome uniquely contributes 19% of the total proteome (65 of 346 proteins), and 65% (226 proteins) of the proteome was quantified in all data sets (Figure B). Apart from Source 5, the remaining data sets form a cluster that is distinct from the ENRICHplus clusters (Figure C), underscoring that the magnitude of the aforementioned interplasma proteome discrepancies would not have been appreciated in conventional neat plasma proteome profiling. We also noted that the distributions of protein intensities for the neat proteomes are slightly wider than those of their ENRICHplus counterparts (Figure D). A comparison of the commonly identified proteins in the neat and ENRICHplus data sets shows a positive correlation in their abundances, with Sources 1, 2, and 5 exhibiting correlations between r = 0.51 (Source 2) and r = 0.64 (Source 5), whereas Sources 3 and 4 are r = 0.29 and 0.34, respectively (Figure E). Having the highest number of neat plasma proteins, and their relatively high correlation with their ENRICHplus counterparts, indicates that Source 5 plasma underwent a process that modified the proteome’s dynamic range (more below).

3.

Summary statistics of neat plasma proteomes. (A) Bar plot showing the number of proteins identified in each commercial source data set, for ENRICHplus and the neat (iST-BCT) method. (B) Overlap of neat plasma proteins identified across data sets. (C) PCA displaying the clustering of samples based on protein abundance profiles, colored by source data set and shaped by method. (D) Density plots showing the distribution of protein abundances for each data set, with separate traces for ENRICHplus (solid) and neat (dashed) plasma proteomes. Vertical lines indicate the peak abundance for each method. (E) Scatter plots comparing protein abundances between ENRICHplus and neat plasma data set, with Pearson correlation coefficients and significance levels indicated on each plot. Dashed line is a reference for r = 1; the solid line indicates ordinary least-squares regression.

Differential Enrichment of Apolipoproteins and Platelet Proteins

Guided by relative protein numbers and the PCA (Figure D), we first grouped Sources 1 and 2 (group 1) and compared them to grouped Sources 3 to 5 (group 2). Of the shared proteome (848), 17 proteins increased in group 1 and 257 increased in group 2 (log FC > 1, adj. p-value <0.05) (Figure S3A). Gene set enrichment analysis (GSEA) of these group 1 proteins indicated Reactome pathways associated with, for example, complement functions and lipoprotein biology, whereas group 2 proteins associated with platelet activation, membrane trafficking, and vesicle-mediated transport (Figure S3B, Table S5). Intriguingly, we noted that many of the increased proteins in group 2 are known to be released from activated platelets, while proteins increased in group 1 comprise several apolipoproteins (Figure S3C). To corroborate the increase in platelet proteins in group 2, neat plasma levels of soluble platelet factor 4 (PF4) and membrane-associated P-selectin (SELP/CD62P) from the individual donors were quantified by enzyme-linked immunosorbent assay (ELISA). Readouts from Source 5 plasma samples exceeded the upper limit of detection and were thus excluded from further analysis. Nonetheless, both proteins were depleted in group 1 by 2.1-fold for CD62P (p = 0.0011) and 1.3-fold for PF4 (p = 0.28) (Figure S3D), confirming what is indicated in the enriched plasma proteome.

To evaluate the impact of different plasma sources objectively, we characterized proteins based on their abundance profiles, representing each protein as a vector of abundances. We conducted a PCA where the entities were the proteins and the variables were the sources (Figure A). The loadings, indicated by arrows on the PCA plot, show that the abundances from Source data sets 1 and 2 are opposed to those of data sets 3, 4, and 5 across the second principal component (PC2), which accounts for 16.24% of the variance. The variance explained by the first principal component (PC1) is 78.16% and is likely related to the overall abundance of the proteins, as depicted in the plot. To further investigate this finding, we calculated the Pearson correlation coefficients between proteins based on their abundance profiles across the data sets (Figure B). Hierarchical clustering of the correlation matrix revealed two main clusters of proteins. Cluster 1 contained 402 proteins that included 68 apolipoproteins and 87 platelet proteins. These 87 platelet proteins were enriched in Source data sets 3 to 5 (Figure C). In contrast, Cluster 2 comprised 109 proteins (21%), including 54 known apolipoproteins but no platelet proteins (Figure C). These 54 apolipoproteins were more enriched Source data sets 1 and 2 (Figure C). Leveraging the available timsTOF HT data that identified a larger number of proteins (Figure A), we conducted a similar analysis. The PCA shows a similar trend in which Source data sets 1 and 2 are opposed to those of data sets 3 to 5 across PC2 (13.07%) and the variance in PC1 (82.53%) is the overall abundances (Figure S4A). Subsequent clustering of these data also generated two main clusters (Figure S4B,C). In this case, the apolipoproteins dominated a single cluster and were enriched in Source data sets 1 and 2, whereas the platelet proteins were equally distributed between the two clusters, exhibiting an enrichment in data sets 3 to 5 (Figure S4C). The overall abundance distributions of proteins identified using two independent platforms indicate that the apolipoproteins were depleted and platelet activation increased in the SOPs of plasma derived from Sources 3, 4, and 5.

4.

PCA and cluster analysis of proteins based on their abundance profiles across data sets. (A) PCA plot of proteins colored by type (apolipoproteins, platelet proteins, and other proteins). Areas represent the density of each of types. Arrows represent the loadings of each source on the PCA. (B) Heatmap showing the Pearson correlation coefficients between proteins based on their abundance profile across data set. Rows and columns are clustered using hierarchical clustering with complete linkage. (C) Using the two main clusters from the hierarchical clustering, a bar plot shows the number of proteins in each cluster categorized by their type. (D) Box plots illustrating the distribution of protein abundances for each cluster, labeled by protein type and colored by source. Cluster 2 has no platelet proteins. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001, ****p ≤ 0.0001.

Apheresed Plasma Increased Proteome Yields

Our study design minimized confounding factors such as donor sex, age, freeze–thaw cycles, and storage temperature when ordering plasma, and accounted for operator bias for the workflow itself. The ELISAs of neat plasma confirmed the disparity in platelet protein abundances between groups 1 (Sources 1 and 2) and 2 (Sources 3 to 5); thus, the most probable remaining source of variation in the plasma proteome is the preanalytical processing of whole blood samples. To corroborate our hypothesis, we inquired about plasma preparation SOPs from all commercial sources (Figure S5). Four of five responded to our request for detailed information on their SOPs that were not explicit on their websites. Their responses were consistent with the proteomics. Differences in centrifugation protocol, storage time, and overall procurement varied in a manner that corroborated the distinctions between data set groups 1 and 2. Source 5 provided plasma collected via apheresis, a medical procedure that separates blood into cellular subsets and plasma. During blood donation apheresis procedures, blood is anticoagulated and separated into several components by centrifugation. The plasma is typically passed through a filter or membrane system before entering the collection bag, potentially inadvertently depleting subsets of (apo)­lipoproteins. Nevertheless, these findings are consistent with both the higher number of protein identifications and the distinct position in the PCA plot for the Source 5 neat plasma samples. Depletion of higher-abundant proteins during the apheresis procedure allows for the quantification of lower-abundant proteins even without additional enrichment.

Source 3, on the other hand, provided plasma whose blood was stored at 4 °C for up to 4 h before centrifugation. Prolonged whole blood storage prior to plasma preparation increases platelet activation in EDTA-anticoagulated whole blood, potentially contributing to our observed proteomic changes. Moreover, centrifugation of whole blood at 1500g may also result in incomplete removal of platelet debris that further contributes to their enrichment in group 2. Taken together, SOPs and the resulting proteomic data, protocols that intentionally or inadvertently deplete apolipoproteins may be permissive to improved sequencing depth.

Discussion

Mass spectrometry-enabled plasma proteomics is currently trending, inciting enthusiasm and caution at the same time. Multiple novel technologies are on the market, whose distinct chemistries are likely to enrich distinct proteins. It is therefore important that for a given workflow, the impact of the plasma preparation itself is evaluated. In this study, we employed the ENRICHplus plasma proteome workflow, a particle-based strategy to enrich lower-abundant proteins.

We anticipated that distinct commercial sources would impact the plasma proteome yields. For instance, a previous study using immunodepletion of high-abundant proteins to process samples procured from a multicenter network ascribed the main source of variability to the study center. The propensity to detect actin and other cytoskeleton proteins associated with specific centers, indicating deviations from the presumed SOP. The anticoagulant, centrifugation, sample hold time, and temperature conditions were also tested, of which variations in sample hold time before freezing and the anticoagulant (EDTA vs heparin) entailed the largest sources of variability in protein recovery. For instance, longer hold times for blood, especially at room temperature, increased the prevalence of canonical cytoskeleton proteins, and heparin treatment reduced overall recovery of plasma proteins. These findings are consistent with our initial assumptions that requesting plasma with minimal freeze–thaw cycles and the same anticoagulant would diminish potential variability in SOP that may confound proteome comparisons. In addition, by ordering plasma from presumably healthy donors, we expected to minimize potential disease-associated plasma signatures such as platelet activation that results in the release of their protein constituents. − For instance, variations in SOPs can predispose blood to artificial platelet activation or insufficient platelet removal. Others have reported that prolonged blood storage times also induce leukocyte activation, and enzyme-mediated degradation of proteins. ,

Variables that we did not anticipate were apheresed blood (Source 5), blood holding times of up to 4 h on ice before centrifugation (Source 3), and delivery of plasma at room temperature (Source 4). Since we did not receive additional information from Source 4, we do not know which specific processing step (undisclosed or shipping condition) contributed to the relatively high protein yield. The disparity between group 1 (Sources 1 and 2) and group 2 (Sources 3 to 5) data sets was evident regardless of the instrument platform and spectral annotation workflow.

Donor apheresis primarily uses ACDA-anticoagulated blood, as ACDA provides a slightly acidic milieu, imperative to maintain platelet quiescence during the platelet collection and storage process. EDTA-anticoagulated plasma, however, is predominantly collected for translational/clinical research and hematological testing, pertaining to its role in effectively chelating various metal ions (among others Ca²⁺, Mg²⁺, or Zn²⁺) that act as cofactors for enzymes, thus preventing enzymatic degradation of proteins or analytes of interest. During the conception of this study, we included ACDA plasma, expecting that its proteome would be distinct from the K2EDTA groups. We had not associated ACDA with the apheresis protocol; therefore, the impact of ACDA itself cannot be evaluated. Being aware of and understanding the inherent differences in anticoagulant use for distinct blood donation procedures may therefore allow researchers to infer upstream plasma preparation methods and to make informed decisions with respect to downstream experimental setups. Plasma proteomes from Sources 1 and 2 were prepared using a similar centrifugation protocol that entailed immediate centrifugation of blood after collection, thus resulting in very similar proteomes.

Our findings have been particularly informative for our own research, raising awareness that we anticipate will also be of interest to other laboratories. We rely on commercially sourced plasma for a variety of experiments, including developing and testing ELISAs and optimizing lipoprotein or extracellular vesicle isolation protocols. The choice of source can be influenced by several factors, such as sample costs and the ability to request specific patient demographics. Our findings do not suggest that one SOP is superior to another. If, however, platelet-depleted plasma is a prerequisite for studies (e.g., extracellular vesicle isolation), it is imperative that researchers evaluate the likelihood that the SOPs may inadvertently activate platelets. When used as a comparative healthy control cohort, inadvertent platelet activation in commercial plasma may mask platelet activation in disease contexts, such as infection, cardiovascular disease, or cancer, where platelet activation is a well-recognized pathological mechanism. − Likewise, SOPs that deplete apolipoproteins are likely to have perturbed their corresponding lipoprotein fractions, which will in turn impact not only apolipoprotein but also lipid/cholesterol measurements. The neat plasma proteomic data would not have necessarily alerted us to these potential confounders. These findings underscore the value of the continual development of plasma proteomic technologies, but also careful consideration and awareness of preanalytical variables and how they may differentiate these technologies. This latter point is particularly important for plasma profiling of retrospective study cohorts, in which preanalytical phases cannot be regulated. Although distinct methods have been reported to enrich distinct subproteomes, these differences convey the true complexity of the plasma compartment and how its proteome informs us of technical and/or biological sources of variation.

Conclusions

Nanoparticle-based enrichment of low-abundance plasma proteins uncovered how variations in routine laboratory practices can significantly influence research outcomes. Our study also highlights the strength of mass spectrometry to independently corroborate, through its unbiased proteome profiling, the varied blood-to-plasma operating procedures.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jproteome.5c01012.

• ENRICHplus plasma proteome yields (Figure S1); operator-to-operator performance analysis for ENRICHplus workflow (Figure S2); differential expression and gene set enrichment analysis (GSEA) between proteins identified in two groups of commercial sources (Figure S3); timsTOF HT + DI-NN data analysis (Figure S4); reported standard operation procedures (SOPs) for initial plasma orders (Figure S5) (PDF)

• Combined consensus proteome for 5-source ENRICHplus workflow (Exploris, CHIMERYS) (Table S1); combined consensus proteome for 5-source ENRICHplus workflow (Exploris, DIA-NN) (Table S2); combined consensus proteome for 5-source ENRICHplus workflow (timsTOF HT, DIA-NN) (Table S3); combined consensus for neat plasma proteome (Exploris, CHIMERYS) (Table S4); reactome pathways differentially enriched between group 1 and group 2 (Table S5) (XLSX)

∇.

T.K., L.W., and J.W. equally contributed to this study. T.K., J.W., and L.W. prepared and analyzed human plasma enrichment samples. T.K. and Z.J.K. collected mass spectrometric data. T.K., L.W., J.T.M., G.S., and S.A.S. performed data analysis. S.A.S. supervised the study. M.A., E.A., and S.A.S. provided overall supervision, resources, and funding. All authors contributed to the editing of the manuscript.

This work is supported by grants from the National Institutes of Health (R01HL149302 to M.A., and R01HL174066 to M.A., E.A., and S.A.S.); Leducq Foundation PRIMA Network (22RAF02 to E.A.); the American Heart Association (2024A003947 to S.A.S.); The European Union’s Horizon Europe Research and Innovation Program under the Marie Skłodowska-Curie Actions (101203749 to L.W.); and Kowa Company, Ltd. (Nagoya, Japan, to M.A.).

The authors declare the following competing financial interest(s): TK is an employee of Kowa Company, Ltd., Nagoya, Japan, but also a visiting scientist at Brigham and Womens Hospital and Harvard Medical School when the study was conducted. JW is an employee of PreOmics, Inc. ZJK is an employee of Bruker Scientific.