Orbitrap Mass Spectrometry and High-Field Asymmetric Waveform Ion Mobility Spectrometry (FAIMS) Enable the in-Depth Analysis of Human Serum Proteoforms

Abstract

Blood serum and plasma are arguably the most commonly analyzed clinical samples, with dozens of proteins serving as validated biomarkers for various human diseases. Top-down proteomics may provide additional insights into disease etiopathogenesis since this approach focuses on protein forms, or proteoforms, originally circulating in blood, potentially providing access to information about relevant post-translational modifications, truncations, single amino acid substitutions, and many other sources of protein variation. However, the vast majority of proteomic studies on serum and plasma are carried out using peptide-centric, bottom-up approaches that cannot recapitulate the original proteoform content of samples. Clinical laboratories have been slow to adopt top-down analysis, also due to higher sample handling requirements. In this study, we describe a straightforward protocol for intact proteoform sample preparation based on the depletion of albumin and immunoglobulins, followed by simplified protein fractionation via polyacrylamide gel electrophoresis. After molecular weight-based fractionation, we supplemented the traditional liquid chromatography–tandem mass spectrometry (LC-MS²) data acquisition with high-field asymmetric waveform ion mobility spectrometry (FAIMS) to further simplify serum proteoform mixtures. This LC-FAIMS-MS² method led to the identification of over 1000 serum proteoforms < 30 kDa, outperforming traditional LC-MS² data acquisition and more than doubling the number of proteoforms identified in previous studies.

Introduction

Top-down proteomics (TDP) is defined as the high-throughput investigation of protein forms, or proteoforms,¹ typically performed by mass spectrometry (MS).² Despite recent technological advances in separation techniques³ and high-resolution mass spectrometry,^4,5 TDP still offers relatively limited throughput compared to its peptide-centric counterpart, bottom-up proteomics (BUP).⁶ However, while BUP analyzes short peptides derived from the enzymatic digestions of proteins, and thus suffers from the protein inference problem,⁷ TDP offers the possibility to interrogate the actual protein molecules present in the cell or tissue of interest. This feature makes TDP a powerful tool for the investigation of clinical samples and, more specifically, for the discovery and validation of biomarkers.

In light of this, it is not surprising that the application of TDP has seen a rapid expansion in translational studies over the past decade.⁸ Intact protein analysis by MS has been used on a variety of clinical samples and tissues, including (but not limited to) cerebrospinal fluid,⁹ cardiac tissue,^10,11 gingival crevicular fluid,¹² brain tissue,^13,14 and saliva.¹⁵⁻¹⁸ Kelleher and co-workers analyzed the various proteoforms of peripheral blood mononuclear cells (PBMCs) in an attempt to identify biomarkers for transplant rejection.¹⁹ Such work has recently led to the creation of the largest proteoform repository to date, the Blood Proteoform Atlas, which includes >30 000 unique proteoforms.²⁰

Blood serum and plasma arguably represent the most widely used and easily accessible samples for clinical studies, and also the largest sources of known disease biomarkers.^21,22 Historically, BUP has been extensively applied to the characterization of these tissues,²³ whose complexity arises from the extreme dynamic range of concentration (spanning 11 orders of magnitude) of their protein components. More than 90% of the total protein content in human plasma is attributed to only 22 highly abundant proteins (HAPs).²⁴ Therefore, alongside strategies based on the direct analysis of the whole protein content of serum and plasma,²⁴ approaches based on the depletion of HAPs, and particularly human serum albumin and immunoglobulins, have been proposed and validated and are now easily applicable in any clinical proteomics laboratory thanks to a large variety of commercially available protein depletion kits.^25,26 Surprisingly, the application of MS-based top-down proteomics for the large-scale exploration of proteoforms present in human serum and plasma has been limited. A notable exception is given by the study from Cheon et al.,²⁷ who applied extensive molecular weight (MW)-based off-line fractionation of human plasma proteoforms using gel-eluted liquid fraction entrapment electrophoresis (GELFrEE).²⁸ By combining multiple HAP depletion methods and through the analysis of multiple GELFrEE fractions, the authors identified a total of 442 low MW proteoforms (<25 kDa) from 71 proteins.²⁷ To our knowledge, this study represents the largest high-throughput TDP study on human plasma to date, since other studies either focused on the peptidome (i.e., mostly proteins <10 kDa)²⁹ or selectively targeted the proteoforms of plasma or serum proteins of interest, such as apolipoproteins (e.g., ApoC-III and ApoA-I)³⁰⁻³³ or transthyretin.³⁴

Here, we propose a simplified method for the in-depth analysis of the low MW intact proteome (<30 kDa) of human serum. In the attempt to make the experimental protocol accessible to the largest possible number of laboratories, we replaced GELFrEE (whose commercial version has been discontinued) with a method named polyacrylamide-gel-based prefractionation for analysis of intact proteoforms and protein complexes by mass spectrometry (PEPPI-MS),³⁵ which is based on a simple SDS-PAGE apparatus typically available to any protein biochemistry research group. We opted for the use of HAP depletion using two different commercial kits targeting albumin and immunoglobulins to limit the quantity of antibody light chain (∼25 kDa) contaminating the 0–30 kDa portion of the serum proteome under examination. The performances of the two kits were compared to evaluate whether their use may result in the unsought enrichment of specific protein groups, as previously reported for other kits benchmarked using BUP.^36,37

To improve the depth of intact proteome coverage, in our Orbitrap MS-based TDP experiments we supplemented in-solution PEPPI-MS proteoform fractionation with gas-phase fractionation (GPF) of proteoform cations using high-field asymmetric waveform ion mobility spectrometry (FAIMS).³⁸ Recent top-down studies,³⁹ also involving complex human samples such as brain tissue from Alzheimer’s patients,¹³ demonstrated that FAIMS can substantially increase the number of identified proteoforms, particularly low-abundance ones. In FAIMS, protein cations are separated based on their differential mobility once subjected to an asymmetric waveform (i.e., including low and high electric field components). A compensatory DC voltage (CV) is applied to prevent collision with the electrodes of ions with specific mobility characteristics. Therefore, incremental changes in the applied CV values allow for the transmission of different fractions of the ions originally generated via electrospray ionization. While in the first FAIMS-TDP studies the CV was maintained constant throughout the whole duration of a liquid chromatography-tandem mass spectrometry (LC-MS²) experiment and progressively increased across consecutive runs (a practice known as “external stepping”),^13,39 more recently Tholey and co-workers have demonstrated that an “internal stepping” strategy, whereby multiple CV values are applied in cycle within the same LC-MS² experiment, can be efficiently implemented, thus reducing the overall number of runs required for thorough sample characterization.⁴⁰⁻⁴²

Our TDP analysis of two PEPPI fractions (0–15 and 15–30 kDa) with and without FAIMS (with internal CV stepping) confirmed the utility of GPF via differential mobility spectrometry, which in conjunction with MW-based fractionation can lead to the identification of more than 1000 low MW serum proteoforms from a limited number of LC-FAIMS-MS² runs.

Experimental Section

Serum Protein Extraction and Fractionation

Human pooled serum was purchased from Millipore Sigma (catalog number H4522). Two commercial kits were used to deplete 60 μL of serum of albumin and immunoglobulins following the manufacturers’ protocols: High Select HSA/Immunoglobulin Depletion Midi Spin Columns (catalog number A36367; Thermo Fisher Scientific, Rockford, IL) and ProteoExtract (catalog number 122642; Millipore Sigma, MA, USA). Two duplicate samples were prepared using each kit, with each duplicate equally split in two prior to protein fractionation (see below) so that the same input samples would be later analyzed in the non-FAIMS and FAIMS experiments.

Protein content of depleted serum was quantified using a NanoDrop OneC microvolume UV–vis spectrophotometer (Thermo Fisher Scientific, Rockford, IL) measuring absorbance at 280 nm, and a volume corresponding to 300 μg of protein was acetone precipitated overnight at −20 °C. Each protein pellet was resuspended in 1% SDS (w/v) and subsequently fractionated based on molecular weight (MW) into two fractions, 0–15 and 15–30 kDa, using PEPPI-MS on a 12% T polyacrylamide gel. Additional details for the PEPPI-MS protocol are provided in the Supporting Information. A 10 μL aliquot of recovered protein solution from each fraction was used for visualizing the result of PEPPI using SDS-PAGE followed by silver staining. Immediately prior to LC-MS² analysis, samples were desalted and SDS removed using methanol/chloroform/water precipitation and reconstituted in mobile phase A (vide infra). Alternatively, the other two SDS cleanup methods were applied for a specific comparison. Anion-exchange disk-assisted sequential sample preparation (AnExSP) was performed based on a previously published method.⁴¹ Briefly, the sample volume was reduced using centrifugal 3 kDa MWCO ultrafiltration columns (Amicon Ultra, Millipore Sigma, St Louis, MO). Next, SDS was removed from samples by buffer exchanging twice with 8 M urea using the centrifugal column. Following SDS removal, the sample was buffer exchanged twice into 100 mM ammonium bicarbonate (ABC) at pH 8.5 using spin filtration and concentrated to ∼15 μL prior to recovery based on the manufacturer’s suggested protocol. Next, a StageSpinTip was prepared using a 6 mm diameter disk of Empore anion SPE (3M, Saint Paul, MN) in a 200 μL pipet tip. The StageSpinTip was conditioned twice with 80 μL of methanol (7000g for 3 min for remaining centrifugation steps) and then equilibrated twice with 80 μL of 100 mM ABC. The sample was then loaded onto the tip and the volume increased to 200 μL with 100 mM ABC prior to centrifugation. The sample was washed three times with 200 μL of 100 mM ABC. Finally, the sample was eluted in two steps using 40 μL of 0.5% formic acid (v/v) and 50% methanol (v/v). The sample volume was vacuum concentrated to ∼5 μL, then the volume was increased to 20 μL with mobile phase A (vide infra). Sample cleanup based on HiPPR Detergent Removal Columns (Thermo Scientific, catalog no. 87779) followed the manufacturer’s suggested protocol using LC-MS grade water. Cleaned samples were vacuum concentrated to a volume of ∼5 μL, which was brought up to 20 μL with mobile phase A (vide infra) prior to column injection.

Intact Protein Mass Spectrometry

Proteoform separation was obtained using an Ultimate 3000 UHPLC system (Thermo Scientific, Sunnyvale, CA), with trap and analytical columns (150 μm i.d. by 30 mm length and 100 μm i.d. by 250 mm length, respectively) packed in-house with macroporous polystyrene-divinylbenzene PLRP-S material (5 μm particle size, 1000 Å pore size; Agilent, Santa Clara, CA). Columns were heated at 55 °C, and the analytical flow rate was set at 1 μL/min. Mobile phase A consisted of 5% acetonitrile and 0.2% formic acid (v/v) in water, while mobile phase B was 5% H₂O and 0.2% formic acid in acetonitrile (v/v). An 85 min chromatographic method (total run time), including two final washes preceding column re-equilibration, was used (details in the Supporting Information). The column outlet was connected online to a custom-made nanoelectrospray ionization source (built according to the instructions provided by the University of Washington Proteomics Resources and fitted with narrow mounting posts to be positioned in front of the FAIMS unit) coupled to an Orbitrap Eclipse Tribrid mass spectrometer (Thermo Scientific, San Jose, CA), which was operated in intact protein mode. Data-dependent acquisition was based on broadband mass spectra (MS¹) recorded at 120 000 (at 200 m/z) resolving power (r.p.) over a 400–2000 m/z window, and HCD MS² spectra were recorded at 60 000 r.p. (at 200 m/z) after the fragmentation of the most abundant ions with charge states between +5 and +50. HCD normalized collision energy (NCE) of 35% was applied, and dynamic exclusion was enabled for 45 s. The target values for the automatic gain control (AGC) were set at 4.8e5 and 2.5e5 charges for MS¹ and MS², respectively. Both MS¹ and MS² acquisitions were based on a single time-domain transient (i.e., “microscan”) per mass spectrum. In experiments not using FAIMS, a 3 s data acquisition cycle time was used, and three technical replicates of each PEPPI fraction were run (total number of collected RAW files: 24). In experiments using FAIMS Pro (Thermo Scientific, San Jose, CA), the compensation voltage (CV) was stepped during the runs, utilizing the values reported in Table S1, applying 3 CV values per LC-FAIMS-MS² run. In this case, each 0–15 kDa PEPPI fraction was run three times, while larger 15–30 kDa fractions were subjected to two LC-MS² runs (total number of collected RAW files: 20). FAIMS experiments also used shorter MS¹–MS² cycle times of 1.5–2 s to accommodate for the use of multiple CV values in the same run. All mass spectrometry data files are uploaded on MassIVE (repository number MSV000092574).

Analysis of Top-down MS Data

RAW files were searched against a human UniProt database (DB) derived from an .xml file using ProSight PD 4.2 (Proteinaceous, Inc., Evanston, IL), part of Proteome Discoverer 3.0 (Thermo Scientific). Standard processing and consensus workflows for “high–high” data analysis were used. For FAIMS experiments, the FAIMS CV was kept unspecified to account for varying CV values within a single RAW file (so that all MS¹ spectra were considered in the analysis, regardless of the CV value used for their acquisition). MS¹ spectra were deconvoluted within the high–high cRAWler node by using the Xtract algorithm. The DB search was two-tiered, including an “annotated proteoform” (mass tolerances: 2.2 Da and 10 ppm for precursor and fragment ions, respectively) and a “subsequence search” (mass tolerances set at 10 ppm). Up to four PTMs were considered per proteoform, based on the UniProt annotation. Unless otherwise stated, three proteoform-spectrum matches (PrSMs) were allowed per precursor. PrSMs were filtered using a three-level, context-dependent FDR calculation with a cutoff set at 1%.⁴³ The FDR was recalculated based on the grouping of the file, i.e., one FDR calculation was performed for the FAIMS versus non-FAIMS comparison and another one for the High Select versus ProteoExtract comparison. Plots were generated by using GraphPad Prism 9 (GraphPad Software). GO analysis was performed using DAVID.⁴⁴

Results and Discussion

FAIMS Improves the Coverage of the Intact Serum Proteome

TDP inherently suffers from signal dilution problems that are proportional to the mass of the analyzed proteins. The combination of broad isotopic clusters and wide charge state envelopes that characterize protein ions generated via electrospray ionization typically restricts TDP to the analysis of proteins <30 kDa and limits the spectral dynamic range.^45,46 Therefore, multidimensional prefractionation of proteoforms is widely applied in TDP to reduce sample complexity. However, these strategies increase sample preparation efforts and produce a large number of protein fractions, ultimately requiring longer data acquisition and analysis times. Here, GPF via FAIMS was evaluated as a means to counterbalance limited sample prefractionation. To benchmark the utility of FAIMS, depleted protein samples were fractionated by PEPPI-MS into only two fractions, according to the workflow graphically described in Figure 1.

Figure 1.

Workflow followed for the TDP analysis of commercial serum. Note that two identical sets of samples were subjected to PEPPI-MS to minimize variation in the proteoforms to be identified via Orbitrap MS with or without the use of FAIMS.

As shown in Figure 2A, PEPPI is characterized by a high degree of reproducibility which is also important for the implementation of label-free quantitative TDP studies.⁴⁷ Each protein fraction was later subjected to LC-MS² analysis, with or without FAIMS. Notably, MW-based fractionation is advantageous because CV values in LC-FAIMS-MS² experiments can be adjusted to the expected MW of the proteoforms under analysis, following the findings by Gerbasi et al.³⁹

Figure 2.

Summary of the results in the presence or absence of FAIMS Pro. (A) Results of MW-based fractionation of serum proteins with PEPPI-MS, visualized on a silver stained PAGE gel. (B) Mass distribution of proteoforms (at 1% FDR) without (blue bars) and with (red bars) FAIMS. (C) Venn diagram of unique accession numbers identified in the two groups of experiments (without FAIMS, blue; FAIMS, red). (D) Venn diagram of unique number of proteoforms with and without FAIMS. The results presented here include the analysis of serum samples depleted using both kits under evaluation.

The comparison of non-FAIMS results (based on the analysis of each PEPPI fraction in triplicate) versus the FAIMS results (based on three runs for each 0–15 kDa fraction and two runs for each 15–30 kDa fraction, always using different internally stepped CV values) demonstrate a +35% increase (i.e., 894 unique proteoforms identified with FAIMS versus 661 without) in the identification of proteoforms < 15 kDa when FAIMS is used (Figure 2B). As displayed in Figure 2C and D (and summarized in Table S2), FAIMS globally outperforms standard experiments in both the number of identified proteoforms (1165 versus 947) and UniProt accession numbers (105 versus 81). These results are obtained through efficient GPF, as exemplified in Figure 3 and Figure S1, which show that the application of different CV values can unveil the presence of different coeluting proteoforms with various masses and charge state distributions. Importantly, a comparison of technical replicate runs performed with or without FAIMS demonstrates that when FAIMS is not applied, most of the identified proteoforms in each run remain the same; conversely, the application of different CV values across runs changes substantially the identifications obtained across LC-FAIMS-MS² experiments on the same PEPPI fraction (Figure S2).

Figure 3.

Effect of different compensation voltage values (CV) on the detection of human serum proteoforms. This example refers to the 0–15 kDa fraction. (A) Total ion chromatograms at CV values of −50 (top), −40 (middle), and −30 V (bottom). Up to three different CV values were used in a single LC-MS² run. (B) Corresponding MS¹ spectra obtained at the same retention time by applying different CV values. The three spectra show different species and prevalent charge states of the same proteoforms.

Additional analysis of the spectral intensity of precursor ions linked to identified proteoforms demonstrates that GPF does not uniquely facilitate the identification of particularly low-abundance proteoforms. Figure 4 shows the distribution of intensities of precursor ions corresponding to three groups of identified proteoforms: the comparison of the precursors of proteoforms identified in both non-FAIMS and FAIMS data sets (in blue and pink, respectively) shows a statistically significant increase in the precursor intensity when GPF is applied.

Figure 4.

Comparison of the intensities of proteoforms shared between FAIMS and non-FAIMS to the intensities of proteoforms exclusive to FAIMS. Shared proteoform intensities from non-FAIMS acquisition are in blue; those for FAIMS acquisition are in pink, while intensities for FAIMS-specific proteoforms are in red. Dotted lines represent the 75% (top) and 25% (bottom) quartiles. Dashed lines represent the median. Significance was determined by the Kolmogorov–Smirnov test; ***P value = 0.0007; ****P value < 0.0001.

Even though the precursor ions of proteoforms uniquely identified in FAIMS (displayed in red) are characterized by reduced intensity compared with the rest of the precursors that led to an identification in the LC-FAIMS-MS² runs, their intensity is not low in absolute terms (but rather similar to the majority of precursors included in the non-FAIMS data set). This reflects the capability of FAIMS to reduce spectral congestion so that additional precursors can be selected for fragmentation in data-dependent acquisition experiments. Overall, these results indicate that the complexity of the intact proteome of serum is so high that only a minor fraction of proteoforms (631, corresponding to 43% of the total) and proteins (59, corresponding to 45% of the total) is shared between the two data sets: when the identifications of FAIMS and non-FAIMS experiments are combined, the number of identifications reaches 1481 proteoforms from 130 protein accessions. The LC-FAIMS-MS² strategy is clearly advantageous over regular LC-MS² (+23% and +30% in the number of identified proteoforms and UniProt accession numbers, respectively), and this is demonstrated also by the results of a DB search where the third technical replicate for the FAIMS 0–15 kDa fraction was not considered (Figure S3), yet the number of identifications is substantially unaltered. Collectively, these results suggest that the use of three FAIMS-based runs per sample with more closely spaced CV values could help further expand the proteome coverage in future studies, and they also indicate that the third LC-FAIMS-MS² run for the 0–15 kDa fraction, which used particularly negative CV values (usually applied to small polypeptides, like in BUP experiments), did not account for a significant number of unique proteoform identifications—possibly also due to limitations in the sample preparation (vide infra).

As further proof of the inherent complexity of the serum proteome, we compared these TDP results against a recently published BUP data set collected using the same exact commercial pooled serum sample and utilizing the same commercial HAP depletion kits (Figure S4).⁴⁸ Surprisingly, only 12% of UniProt accession numbers are shared between the TDP and BUP data sets (when considering only the proteins < 30 kDa identified in the BUP study, the fraction decreases to 9%). By extending the comparison to include the results of a larger BUP study based on the analysis of clinical serum samples,⁴⁹ the differences in protein identification across data sets are even more apparent.

Main Sources of Variability among Human Serum Proteoforms

Regarding chemical modifications discovered by TDP, 535 and 520 proteoforms of the non-FAIMS and the FAIMS data sets, respectively, include at least one PTM (corresponding to 56% and 45% of the identifications, respectively), indicating that serum proteoforms are frequently modified (Table S2 and Figure S5). Acetylation and phosphorylation are the most frequently observed PTMs. Protein truncation also plays a substantial role in expanding the proteoform variability in serum. Considering the five UniProt accessions represented by the largest number of proteoforms in this TDP data set, N-terminal sequence truncation (beyond the removal of initiator Met or of signal peptide) is present in 12%–66% of the cases. Double truncation (i.e., contemporary truncation at the N- and C-terminus) is observed in up to 41% of the identified proteoforms derived from a single gene (Figure S6).

We posit that truncation is one of the reasons that explain why the majority of identified proteoforms have a mass < 15 kDa (Figure 2B). The almost complete lack of proteoforms in the 20–25 kDa mass bin is justified when considering that the BUP analysis of the same pooled serum used for the present TDP study has led to substantially more identifications in the 0–15 kDa range than in the 15–30 kDa bin (61 vs 33, respectively, as reported in Figure S4). Additionally, while the gel displayed in Figure 2A shows an apparent protein band at MW ∼ 25 kDa, that more than likely corresponds to just one protein, a residual immunoglobulin light chain. The DB search identified both lambda and kappa light chains of human immunoglobulin; however only a few distinct sequences are included in the DB, and thus the number of identified proteoforms is not as high as for other abundant proteins (e.g., apolipoprotein A-I). The large number of small proteoforms present in the sample explains also why the use of FAIMS had a major impact almost exclusively on analytes < 15 kDa. In separate studies based on the analysis of different samples such as whole cell lysates, FAIMS increased the number of proteoform identifications throughout the whole 0–30 kDa range.⁴⁷ Therefore, we conclude that the results reported here reflect the specific compositional characteristics of the analyzed pooled serum sample.

A Closer Look to Serum Depletion and TDP

Differences were observed when RAW files (with and without FAIMS) collected after the depletion of albumin and immunoglobulins were compared with the High Select and ProteoExtract kits (Figure 5). Figure 5A and B show that, although the majority of proteoforms are shared between the two data sets (772 unique proteoforms, corresponding to 51% of the 1510 total), only 64 UniProt accession numbers (corresponding to 42% of the 152 total) were identified in both data sets.

Figure 5.

Summary of comparative analysis of High Select and ProteoExtract serum depletion kits. High Select results are in brown, while ProteoExtract results are in purple. (A) Venn diagram of unique accession numbers (at 1% FDR). (B) Venn diagram of unique proteoforms (at 1% FDR). (C) Mass distribution of proteoforms identified using the two kits with almost identical profiles. (D) Comparison of most prominent functional categories of some of the protein accessions uniquely identified by High Select or ProteoExtract.

Notably, such a discrepancy is not explained in terms of proteoform mass bias, since the mass distributions of proteoforms identified using the two kits are virtually identical (Figure 5C). Due to the above-illustrated complexity of the serum proteome and the stochastic nature of precursor selection in data-dependent acquisition, minor discrepancies across data sets are to be expected. However, a more detailed analysis of the lists of UniProt accession numbers identified uniquely in one of the two data sets suggests that minor differential protein enrichment may have occurred as a consequence of the use of the two different HAP depletion kits. Specifically, GO analysis on the 43 UniProt accession numbers unique for the samples depleted with High Select reveals that most of these proteins are hydrophilic, with several being glycosylated or generally polar. Conversely, some of the 45 UniProt accession numbers unique to the ProteoExtract kit are enriched in two major protein groups: those including immunoglobulin-like domains and protein kinases (Figure 5D). This may imply that the ProteoExtract kit has potentially a lower affinity and/or binding capacity than the High Select one for antibody chains and other immunoglobulin-like proteins when compared to the High Select option. Grand average of hydropathy (GRAVY) analysis performed on both lists of unique UniProt accessions confirms that the average hydrophobicity of the proteins identified using the High Select is slightly lower than in the case of the ProteoExtract kit (i.e., the hydrophobic proteins being 9.3% and 11.3% in High Select and ProteoExtract, respectively); however, no statistically significant difference is observed when comparing the two sets of GRAVY values (Figure S7). This seemingly confirms that the observed difference in depletion selectivity is likely to be attributed to unintentional, nonspecific enrichment of proteins characterized by particular domains or post-translational modification. However, while some of the proteins potentially enriched by each HAP depletion kit are represented by multiple proteoforms (this is the case of proteins included in the “Ig-like domains” category, for instance), others are represented by single proteoforms, as in the case of proteins linked to the “kinase” category enriched by the use of the ProteoExtract kit. Additionally, several of the functional categories into which the GO analysis grouped the two groups of proteins uniquely identified by each kit were the same. All considered, the actual differences in performance between the two tested depletion kits seem to be limited.

The Relevance of Detergent Cleanup Methods

Since the analysis of the initially collected 44 RAW files demonstrated that the majority of identifiable proteoforms have a mass < 15 kDa, we benchmarked also SDS cleanup methods that are known to outperform the traditional MeOH/CHCl₃/H₂O precipitation method in the recovery of small proteins. A separate set of serum samples was subjected to HAP depletion with the High Select kit and subsequent PEPPI-MS fractionation prior to LC-FAIMS-MS² analysis (details in Table S3). SDS was removed from samples using the MeOH/CHCl₃/H₂O procedure, a commercial HiPRR detergent removal cartridge, and the anion exchange method described by Takemori et al.⁴¹ The summary of results is displayed in Figure 6, which shows that the MeOH/CHCl₃/H₂O method is globally outperformed by the other two, with differences that are particularly marked for the smaller proteoforms: while the standard method led to the identification of 87 and 125 proteoforms in the 0–5 and 5–10 kDa mass bins, respectively, the HiPRR cartridge produced 186 and 223 proteoform identifications (an increase of +114% and +78%, respectively), and the anion exchange method returned 316 (+263%) and 308 (+146%) proteoform IDs.

Figure 6.

Summary of the comparison of three SDS removal methods. (A) Global count of identified UniProt accession numbers and unique proteoforms. (B) Mass distribution of identified proteoforms. (C) Venn diagram of identified UniProt accession numbers. (D) Venn diagram of identified proteoforms. Results of the three cleanup methods are color-coded based on the legend included in the figure.

These observations are in line with what was reported by Takemori et al. and suggest that future TDP studies of human serum or plasma may be based on different SDS removal protocols depending on the targeted proteoform mass range. Finally, it is important to underline that while the anion exchange cleanup protocol produced the highest number of proteoform identifications < 15 kDa, it is by far the most complex and time-consuming of the three tested methods.

Brief Considerations on Serum Biomarkers

Ideally, increasing the number of proteoforms identified from human serum or plasma could lead to new biomedical insights or the ability to perform more advanced clinical analyses. Therefore, to evaluate the applicability of TDP to biomarker identification, we considered a list of validated serum and plasma biomarkers compiled by Anderson in 2010.⁵⁰ The list includes a total of 109 biomarkers approved by the FDA throughout 2008, in addition to the other 96 protein targets used in various clinical assays in the USA (Table S4). Importantly, while only a small fraction of these proteins have masses < 30 kDa (e.g., just 34% of the FDA-approved ones), TDP also detects truncation products of larger proteins, as previously implied. Figure 7 and Figure S8 show that the list of UniProt accession numbers globally identified in this study includes a total of 20 of the 109 FDA-approved biomarkers and 30 of the extended list.

Figure 7.

Biomarkers in the TDP data set. This is a comparison of the proteins (i.e., UniProt accession numbers) from this TDP study against the extended list of plasma biomarkers proposed by Anderson (201 total UniProt accessions without including immunoglobulin chains). (A) Venn diagram showing the biomakers included in the FAIMS and non-FAIMS TDP data sets. (B) Venn diagram differentiating the biomarkers identified by the mean of the High Select and ProteoExtract depletion kits.

GPF is advantageous also for biomarker identification: the LC-FAIMS-MS² strategy slightly outperforms the standard LC-MS² acquisition with 29 identified biomarkers versus 25, respectively (Figure 7A). The two depletion kits have overall comparable performance but also show a total of five biomarkers identified by only one of the two kits (Figure 7B). However, it is important to underline that the biomarkers uniquely identified by either of the two depletion kits are represented by 1–4 proteoforms each, and hence we cannot rule out the possibility that some of these differential identifications are to be interpreted not as the product of actual kit-dependent enrichment but rather as an effect of the stochastic nature of data-dependent acquisition (as visible in Figure S2).

Conclusions

Recent advances in mass spectrometry instrumentation, including new ion mobility devices such as FAIMS, and the introduction of sophisticated data acquisition strategies are enabling fast and extensive BUP analysis of the plasma/serum proteome for clinical analysis.⁵¹ However, these impressive achievements necessarily lack the possibility of precisely reconstructing the original landscape of proteoforms circulating in patients’ blood, thus limiting the potential for biomarker discovery. The top-down study presented here demonstrates that modern Orbitrap technology coupled with FAIMS allows researchers to investigate a sufficiently large portion of the serum proteome when supported by streamlined sample preparation and protein MW-based fractionation as a viable clinical analysis technique with a relatively higher potential for biomarker discovery. While commonly applied detergent removal strategies such as methanol/chloroform/water precipitation can lead to overall satisfactory results, the present study also demonstrates that alternative cleanup protocols may be beneficial for specifically targeting clinically relevant low molecular weight serum proteoforms.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jproteome.3c00488.

• Table S1, list of compensation voltage values for main study analysis; Table S3, list of compensation voltage values for SDS cleanup comparison; Figure S1, total ion chromatograms obtained at different compensation voltages in the LC-FAIMS-MS² experiments; Figure S2, analysis of technical reproducibility across runs; Figure S3, comparison of the number of identified UniProt accessions and proteoforms as a function of PrSMs per precursor; Figure S4, comparison of top-down and bottom-up results; Figure S5, analysis of post-translational modifications in each data set; Figure S6, analysis of type and frequency of sources of variability found on the five UniProt accessions represented by the highest number of distinct proteoforms; Figure S7, GRAVY index score of serum proteins; Figure S8, comparison of the TDP data sets (UniProt accession numbers) against the list of FDA-approved plasma biomarkers proposed by Anderson (PDF)

• Table S2, list of identified proteoforms in experiments with and without FAIMS (XLSX)

• Table S4, extended list of FDA-approved plasma biomarkers according based on Anderson 2010 (XLSX)

Author Contributions

J.T.K. performed research and analyzed data. R.T.F., J.B.G., K.R.D., D.M.H., and S.M.G. contributed to data analysis and data analysis software development. M.W.B., C.L.B., R.H., and J.J.D. contributed to mass spectrometry and FAIMS setup. N.A. codesigned the experiments and performed data analysis. L.F. contributed to data analysis and experimental design and wrote the manuscript. All authors read and approved the final manuscript.

The study was supported by the research grant number R35SGM147397 awarded by the National Institute of General Medical Sciences of the National Institutes of Health to L.F. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

The authors declare the following competing financial interest(s): Michael W. Belford, Cornelia L. Boeser, Romain Huguet, Sylvester M. Greer, David M. Horn, and Jean-Jacques Dunyach are employees of Thermo Fisher Scientific. Ryan T. Fellers, Joseph B. Greer, and Kenneth R. Durbin are part of Proteinaceous, Inc.