Deep Profiling of Plasma Proteoforms with Engineered Nanoparticles for Top-Down Proteomics

Abstract

The dynamic range challenge for the detection of proteins and their proteoforms in human plasma has been well documented. Here, we use the nanoparticle protein corona approach to enrich low-abundance proteins selectively and reproducibly from human plasma and use top-down proteomics to quantify differential enrichment for the 2841 detected proteoforms from 114 proteins. Furthermore, nanoparticle enrichment allowed top-down detection of proteoforms between ~1 μg/mL and ~10 pg/mL in absolute abundance, providing up to a 10⁵-fold increase in proteome depth over neat plasma in which only proteoforms from abundant proteins (>1 μg/mL) were detected. The ability to monitor medium and some low-abundant proteoforms through reproducible enrichment significantly extends the applicability of proteoform research by adding depth beyond albumin, immunoglobins, and apolipoproteins to uncover many involved in immunity and cell signaling. As proteoforms carry unique information content relative to peptides, this report opens the door to deeper proteoform sequencing in clinical proteomics of disease or aging cohorts.

Graphical Abstract

INTRODUCTION

Human plasma is an easily accessible, minimally invasive sample type that contains valuable biological information for biomarker discovery and clinical applications.^1-3 However, analysis of the plasma proteome is challenging due to the overwhelming presence of high-abundance proteins that make up more than 90% of the total plasma proteome, including albumin, immunoglobulins, and other abundant proteins.⁴ Due to the wide dynamic range of protein concentration spanning >11 orders of magnitude, molecular weight (MW) fractionations,⁵ depletion of high-abundant proteins,^6-8 and enrichment of less abundant proteins^9,10 are often required prior to proteomics analysis with mass spectrometry.

To address the dynamic range challenge in analyzing plasma proteome, the Proteograph workflow was established to facilitate deep and broad plasma proteomics measurement at scale.^11-13 The workflow includes contacting biofluids with engineered nanoparticles (NPs) to form protein corona – the spontaneous adsorption of proteins on NP surfaces.^14,15 Varying the physicochemical properties of the nanoparticles results in distinct protein enrichment behaviors, which can be analyzed by mass spectrometry using bottom-up proteomics (BUP) with enzyme digestion. This approach allowed the detection of over 8000 protein groups,^11,16 enabling increased depth for plasma proteome profiling and biomarker discovery. For example, it was used to identify distinct BMP1 isoforms in nonsmall cell lung cancer subjects.¹⁷ The automatic workflow also has advantages for large scale studies, including multiple sample types and/or big cohort analysis. Recent work by Van Eyk and coworkers on high-throughput biomarker discovery across 9 sample types reported that Seer-engineered nanoparticles led to reliable quantification of 3359 protein groups using a 24 min liquid chromatography–mass spectrometry (LC-MS) gradient, which included 137 out of 216 FDA-approved circulating biomarkers.¹⁸ Wilcox and coworkers combined Proteograph with timsTOF HT technology for deep plasma protein biomarker discovery and reported over 4000 quantifiable protein groups.¹⁹ Additionally, the detection of low abundant proteins was not adversely affected among plasma samples with varying lipid levels and extent of hemolysis, which are the two most common variables in matrix composition.¹¹ Together, these applications show that protein corona enrichment using Seer NPs and the Proteograph workflow enables robust deep sampling prior to mass spectrometry-based proteomics.

There is a growing interest in proteoform-level information acquired by top-down proteomics (TDP) to address the limitations of protein inference and post-translational modification (PTM) analysis in BUP.^20-26 To link protein composition more directly to function and disease phenotypes, we recently reported that TDP analysis of peripheral blood mononuclear cells (PBMCs) revealed proteoform biomarker candidates for organ transplant rejection that was reproduced in a multicenter trial.²⁷ We also demonstrated that TDP captured new differences in patient ApoA-I proteoforms enriched from plasma due to lipidation of Lys88 that correlated better than total ApoA-I to several metrics of the cardiovascular disease risk.²⁸ Most recently, we showed that proteoforms in plasma can be signatures of liver cirrhosis progression.²⁹ These examples indicate that analyzing proteoforms can improve disease diagnosis, understanding of the underlying mechanisms, and identification of treatment targets.

However, TDP of plasma proteoforms also suffers from the wide dynamic range of the plasma proteome and presents the additional challenge that a single gene can give rise to hundreds of proteoforms via combinations of genetic variation, alternatively spliced RNA transcripts, endogenous proteolysis, and PTMs. Lee and coworkers used extensive MW fractionation (10–15 < 30 kDa fractions) to identify 442 unique proteoforms from 71 proteins.³⁰ Fornelli and coworkers reported a record number of 1481 unique proteoforms from 130 proteins via high-field asymmetric waveform ion mobility spectrometry (FAIMS) to further simplify serum proteoform mixtures from only two MW fractions of <30 kDa plasma proteins.³¹ Most recently, Sun and coworkers demonstrated an example of TDP proteoform identification from protein corona.³² In their work, polystyrene nanoparticles (PSNPs) were incubated with human plasma to form protein corona, which was then eluted into SDS buffer, cleaned, and analyzed with capillary zone electrophoresis (CZE) for identification of 263 unique proteoforms from 50 proteins in the 3–70 kDa size range. While most of the proteoforms came from abundant proteins such as apolipoproteins (ApoA-I-II 100 proteoforms, ApoC-I-III 77 proteoforms), albumin, and complement C3, this work hinted at the promising future of TDP characterization of protein corona, particularly of those tailored to enrich lower abundant proteoforms.

Here, we present the results of TDP analysis of engineered NPs that differentially interrogate human plasma using a modified Proteograph workflow. We observed a record number of unique plasma proteoforms (2841) from 114 proteins in LC-MS with clear evidence of reproducible NP enrichment of plasma proteins. By mapping the identified proteins to the Human Plasma Proteome Project (HPPP) database,³³ our data showed identification of low abundant proteins at concentrations less than 10 pg/mL by HPPP estimates and associated with pathways in immune responses and cell signaling. This work opens the door for deep quantitative analysis of proteoforms in plasma and could reveal biological differences that were previously hindered by abundant proteins present in plasma.

EXPERIMENTAL SECTION

Plasma Sample Preparation

Human plasma samples were purchased from BioIVT, where samples were collected from three healthy donors (two Hispanic females, aged 35 and 47, and one African American male, aged 32). Plasma was used either neat or processed with Seer-engineered NPs using a modified Proteograph workflow (XT well A and XT well B). Each well was incubated with 100 μL of plasma and washed as previously described.¹¹ For each type of XT well, 40 wells were combined, eluted into 250 μL of SDS lysis buffer in the Proteograph assay kit, and dried in a vacuum centrifuge.

Molecular Weight Fractionation

Proteins were fractionated using a polyacrylamide-gel-based prefractionation for the analysis of intact proteoforms and protein complexes by mass spectrometry (PEPPI-MS)³⁴ workflow (<50 kDa) prior to LC-MS analysis. Neat plasma (7 μL, ~500 μg total protein) and NP eluates (~250 μg total protein) were resuspended in 40 μL of the nonreducing NuPAGE sample buffer (Thermo Fisher Scientific) and boiled at 95 °C for 10 min. Proteins were resolved in a 1.5 mm NuPAGE 4–12% gel (Thermo Fisher Scientific) at 70 V for 10 min and 150 V for 10 min. Bands under 50 kDa were excised, crushed, homogenized, and shaken (1400 rpm, rt, 20 min) in 100 mM ammonium bicarbonate, pH 9, with 0.1% SDS. The slurry was transferred to a 0.45 μm centrifugal filter (Corning) and spun at 14 000 rpm for 10 min. The flow-through was stored at −80 °C. Samples were precipitated with methanol/water/chloroform and resuspended in 100 μL (neat plasma) or 50 μL (NPs) sample buffer (94.8% water, 5% acetonitrile, and 0.2% formic acid) for LC-MS. A detailed PEPPI-MS protocol can be found in our previous work.²⁶

Top-Down Liquid Chromatography–Mass Spectrometry

Proteins (10 μL each run) were separated using a Vanquish Neo UHPLC chromatographic system (Thermo Fisher Scientific). Reversed-phase LC was performed on a MAbPac EASY-Spray column (150 mm length by 150 μm inner diameter, Thermo Fisher Scientific) with an in-house packed PLRP-S trap (25 mm length by 150 μm i.d., Agilent). The total run time was 120 min using a gradient of mobile phase A (99.9% water and 0.1% formic acid) and mobile phase B (19.9% water, 80% acetonitrile, and 0.1% formic acid). The flow rate was set at 1 μL/min, and the gradient used to resolve proteins was 5% B at 0 min, 20% B at 5 min, 70% B at 110 min, 99% B from 111 to 114 min, and 5% B from 115 to 120 min. The column outlet was coupled inline to an EASY-Spray source and an Orbitrap Eclipse mass spectrometer (Thermo Fisher Scientific) operating in intact protein mode with 2 mTorr of N₂ pressure in the ion routing multiple (IRM). The transfer capillary temperature was set at 320 °C, the ion funnel RF was set at 60%, and 15 V of source CID was applied. MS¹ spectra were acquired at 120 000 resolving power (at m/z 200), a normalized AGC target of 1000%, 100 ms maximum injection time, and 1 μscan. The data-dependent top-N-2 s MS² method used 32 NCE for HCD to generate fragmentation spectra acquired at 60 000 resolving power (at m/z 200), with a normalized AGC target of 2000%, 1200 ms maximum injection time, and 1 μscan. Precursors were isolated with a quadrupole using a 3 m/z isolation window, dynamic exclusion of 60 s duration, and threshold of 1 × 10⁴ intensity. All mass spectrometry .raw data and .tdReport files were uploaded to MassIVE (repository number MSV000095086). A list of proteoforms (PFR #) and proteins (accession #) identified and the proteoform intensity sheet can be found in Table S4.

Top-Down Data Search and Analysis

The raw data files were searched with a publicly available TDPortal v4.1.0 (https://portal.nrtdp.northwestern.edu/) workflow based on the Galaxy Project³⁵ that generated results reported with a 1% context-dependent false discovery rate (FDR) assigned at the protein, isoform, and proteoform levels.³⁶ A proteoform database was created from the SWISS-PROT human database (Taxon 9606 – June 2020) that contained 2.4 million proteoform entries. A – 14-ppm m/z spectral shift was applied to account for instrument calibration. Proteins and proteoforms were filtered for 1% FDR for identification (upset plots). For quantitative proteoform analysis, a CSV intensity sheet was generated by utilizing an isotopic fitting algorithm across the chromatogram to obtain the intensities of all proteoforms identified in the whole study with 10% FDR confidence in each sample. This includes those previously considered as “un-identified” and not meeting the 1% FDR cutoff for identification.³⁷ Box–Cox transformed intensity values were subjected to a hierarchical linear model-based ANOVA, with a Benjamini and Hochberg FDR correction (α = 0.05), to find proteoforms that were differentially enriched between sample groups. Volcano plots were generated, where each proteoform was represented as a function of the estimated effect size (in log₂ fold-change) and the statistical confidence that differences between the two samples were significantly different (−log₁₀ of instantaneous q values). Q values above 0.05 were considered to be significant. Proteoform heatmaps were generated using the “Complex-Heatmap” R package. Identified proteins were matched with the HPPP database³³ using accession numbers for the waterfall plots. Gene ontology (GO) analyses were performed on the metascape website (https://metascape.org/) v3.5.20240101.³⁸

RESULTS AND DISCUSSION

Adapting the Proteograph Workflow for Top-Down Proteomics

The Proteograph workflow is well-established for BUP applications. The automatic sample processing steps include protein corona formation on mixtures of NPs engineered to maximize plasma proteome coverage, washes, enzyme digestion, and peptide cleanup.^11,17-19 Two mixtures of NPs are currently available in the latest generation Proteograph XT workflow: XT well A and XT well B. We modified the workflow with SDS elution of the protein corona for the TDP study of plasma from three healthy donors (biorep01–03). As illustrated in the scheme in Figure 1, plasma was used neat (7 μL, ~500 μg total protein) or enriched with XT well A and XT well B (40 wells each) NPs. Then, the proteins were fractionated for <50 kDa (PEPPI-MS). Each sample (including neat and NP-enriched) was prepared in three technical replicates (techrep01–03) and finally analyzed with LC-MS over a 120 min gradient in three injections (injrep01–03). The study involved 81 randomized LC-MS runs.

Figure 1.

Study design and workflow. Plasma from three human subjects (bioreps) was enriched with two reaction wells from Proteograph XT (XT well A and XT well B). Neat plasma and nanoparticle eluates were extracted for proteins <50 kDa (PEPPI), and an established top-down LC/MS workflow in discovery mode and TDportal search were used to identify and quantify proteoforms. Each nanoparticle enrichment of the sample was performed in triplicate (techreps). All LCMS injections were performed in triplicate (injreps). Proteoforms were filtered for 1% FDR, and quantitative analysis was performed in RStudio.

Nanoparticles Improve the Coverage of the Intact Plasma Proteome

The data were searched against a curated human proteoform database, and hits passing a 1% FDR at the protein and proteoform levels were considered identified (ID). With neat plasma and NPs combined, we identified 2841 unique proteoforms from 114 proteins, a 4-fold and 6-fold increase, respectively, over neat plasma only conditions. 52% of the proteoforms were identified at level 1 where PTMs were unambiguously assigned.³⁹ Figure 2A,B shows upset plots of proteoform and protein IDs, respectively, at different intersections of treatments (neat plasma, XT well A, and XT well B). Each of the three treatment groups had large numbers of unique proteoforms (673, 676, and 1491, respectively), while the overlaps between neat plasma and NPs were very small. There were 27 proteoforms shared between neat plasma and NPs. Among them, 13 proteoforms were identified in all three conditions; two proteoforms were only shared between neat plasma and XT well A, and 12 proteoforms were only shared between neat plasma and XT well B. Similarly, at the protein level, there were only 12 proteins shared between neat plasma and NPs. This observation demonstrates that the NP protein corona enrichment enables greater depth in the interrogation of the plasma proteome, which is typically hindered by high-abundance proteins in standard TDP and BUP analyses. We also note that there were some overlaps in proteoform (223) and protein (21) IDs between XT well A and XT well B. These two groups of NPs were engineered to interrogate orthogonal groups of plasma proteome while not mutually exclusive as demonstrated by BUP.^11-13

Figure 2.

Proteoforms and proteins were identified in this study. We report 2841 unique proteoforms (A) and 114 proteins (B) identified in neat plasma and NPs combined. Compared to neat plasma only, the NPs increase the proteoform identifications by over 4-fold. We note that the intersections between neat plasma and XT well A and B are extremely small, supporting that NPs interrogate parts of the plasma proteome that were inaccessible without enrichment. (C) Mass distribution of the identified proteoforms, all of which were <50 kDa using the top-down PEPPI-LCMS workflow.

Figure 2C shows a histogram of proteoform IDs by their monoisotopic mass. The distribution is as expected with predominantly small proteoforms since proteins were specifically sought at <50 kDa. However, we note that TDP LC-MS has predominantly focused on <30 kDa proteoforms in the past due to difficulties in separation, signal dilution in multiple charge states, ion decay, and acquisition of isotopically resolved spectra for identification/quantification.^40,41 With the recent improvement in separation technology and MS instrumentation, we can now tackle the proteoforms in the 30–50 kDa range with quantifiable data acquired in data-dependent acquisition (DDA) mode. Looking closer into the proteoforms we identified, many of them have canonical sequences above 50 kDa in MW. For example, prothrombin (accession no. P00734) has a canonical sequence of 622 amino acids (a.a.) and a MW of 70 kDa. We identified two proteoforms of its C-terminal fragments: PFR 8996840 (amino acids 528–622, 10.9 kDa) and PFR 8996850 (amino acids 515–622, 12.2 kDa). Another example is Rho GTPase-activating protein 45 (accession #Q92619-1), a 1136-a.a. protein with a 125 kDa MW. Its N-terminal fragment (PFR 9014676, a.a. 1–69, 7.2 kDa) was found with N-terminal acetylation. These two examples illustrate the importance of using TDP to acquire proteoform level information to accurately assign protein fragments and PTM sites and associate with their biological functions, which may be drastically different from their canonical form and will be neglected and indistinguishable by simply assigning them into a protein group in the BUP approach.

Detection of Low-Abundant Proteins and Their Proteoforms

Next, we investigated the depth of proteome coverage afforded via NP-based plasma enrichment. We mapped the proteins identified in each treatment condition (neat plasma, NPs) to the HPPP database for their rankings and estimated concentrations in plasma to understand whether NPs enrich low-abundant proteins.³³ The results are illustrated in waterfall plots with an x-axis of concentration ranking and a y-axis of estimated concentration (Figure 3A-C, left). Each circle represents a protein identified in this study with TDP, and the five least abundant proteins in each condition are labeled with their UniProt accession numbers and gene names. Figure 3A shows that in neat plasma, the proteoform identification was limited to proteins that had an estimated concentration of above 1000 ng/mL from the top 10% population in plasma proteome, while in Figure 3B,C, the identification was made significantly deeper with NP enrichment, enabling detection of low-abundant proteins. Remarkably, the least abundant protein that we identified was the V-type proton ATPase 16 kDa proteolipid subunit (accession #P27449), which had an estimated concentration of 9 pg/mL and ranked 4027 by HPPP (Table S1).³³ The proteoform of this ATPase subunit (PFR 1499) was found to have N-terminal methionine removal and acetylation.

Figure 3.

Waterfall plots of identified proteins and their abundances reported in HPPP (left) and gene ontology analysis (right). The individual waterfall plots of protein identified in neat plasma (A, 20 proteins) and XT NPs (B, 80 proteins; C, 57 proteins) show that the NPs enable the detection of more proteoforms of proteins that have low estimated abundances in the HPPP. GO analysis revealed that more proteins associated with immune responses and cell signaling were identified under NP-enriched conditions.

Furthermore, the HPPP concentration estimation is based on protein level information and does not factor in signal dilution by complicated proteoform landscapes of each protein. For example, we previously found that ApoA-I, despite being the 11th most abundant protein in plasma, had a wide dynamic range within its own proteoform landscape. These comprised less than 5% (glyco-proteoforms) and <2% (acylated proteoforms) of total ApoA-I and were far more correlated with indices associated with the risk of cardiovascular disease.²⁸ Therefore, proteoform-specific enrichment is also of particular interest in TDP analysis that distinguishes it from BUP. An example in this work is the identification of death-associated protein 1 (DAP1, accession no. P51397) and its phospho-proteoforms (Figure 4A). DAP1 is a low-abundant plasma protein that has an estimated concentration of 0.09 ng/mL and ranks 3047 in the HPPP database (Table S1).³³ It was not identified in neat plasma but was found in both XT wells A and B after NP enrichment (see waterfall plots in Figure 3B,C). Interestingly, in XT well B, only the canonical proteoform (PFR 2628, a.a. 2–102, with N-methionine removal and acetylation, Figure 4B) was identified. However, in XT well A, there were four additional proteoforms (PFR 8616, 13512, 15907, and 9039157) identified with various combinations of PTMs at different sites including phosphorylations at S48 and S50, acetylation at K28, and N-terminal truncation from cleavage between G21 and G22 (Figure 4C, and full proteoform sequence can be found in Table S2). We note that PFR 8616 and PFR 13512 were identified at proteoform level 4 (Table S2); additional fragmentation experiments are needed in the future to unambiguously localize the phosphorylation. The identifications described here are based on a 1% FDR cutoff. To further validate them, we used label-free quantification (LFQ) to compare proteoform abundances under each treatment condition (see box plots in Figure S1A). We found that the intensity of the canonical DAP1 proteoform (PFR 2628) was significantly higher in XT well A and B than in neat plasma, supporting that it was identified in both NP wells, and the DAP1 phospho-proteoforms (PFR 8616, 13512, and 9039157) were all significantly more abundant in NP well A compared to well B, indicating differential enrichment. Considering the diversity of the proteoform landscape of DAP1, each proteoform could be well under the estimated concentration of 0.09 ng/mL. The differential enrichment of phosphorylated DAP1 proteoforms between XT well A and B can be utilized for understanding proteoform specific changes in signaling.

Figure 4.

DAP1 proteoforms. (A) DAP1 protein undergoes several post-translation modifications including proteolytic cleavage, acetylation, and phosphorylation that result in several proteoforms. (B) Representative sequence coverage map of one DAP1 proteoform (PFR 2628). (C) Five DAP1 proteoforms were identified in NP-enriched conditions.

The depth of the plasma proteome is connected with myriad molecular functions. For example, proteins secreted from tissues that regulate cell adhesion, signaling, and developmental functions as well as cytokines associated with immunity are ranked >1000 in concentration.^42-44 We performed gene ontology (GO) analysis using all the protein accession numbers identified in each condition. The results are displayed on the right-hand side of each waterfall plot in Figure 3. The proteins identified in neat plasma are limited to endocytosis, blood coagulation, and acute responses, consistent with the function of the most abundant proteins in plasma. In XT wells A and B, the proteome depth identified increases drastically even beyond the top 1000 proteins, revealing pathways associated with immune response and cell signaling. For example, we identified several low abundant chemokines including C─C motif chemokines 5, 14, and 18 (accession #P13501, #Q16627, and #P55774, respectively). The majority of the C─C motif chemokine 5 proteoform landscape was comprised of PFR 40350, a fragment (a.a. 26–91, 7.6 kDa) of the canonical sequence found in both XT well A and B (Table S3). Additionally, in XT well A, we found two less abundant proteoforms: PFR 18966 (a.a. 24–91, 7.8 kDa) with two additional residues at the N-terminus and PFR 5015352 (a.a. 26–91, 7.6 kDa) having the same length as PFR 40350 but with a disulfide bond between C34 and C57. Similarly, for the C─C motif chemokine 18 (Table S3), we identified two proteoforms, PFR 53607 (a.a. 21–89, 7.9 kDa) and PFR 155657 (a.a. 21–88. 7.8 kDa), in both XT wells A and B containing single amino acid variants (SAAVs) as well as an intramolecular disulfide bond between C30 and C54 in PFR 5019691 (a.a. 21–89, 7.9 kDa). The LFQ box plots of C─C motif chemokine 5 and 18 proteoforms can be found in Figures S1B and S1D. The ability to identify and quantify these chemokines and their disulfides at the proteoform level is important for understanding their function and associated immunoresponses. For example, the disulfide between C30 and C54 in the C─C motif chemokine 18 is known as essential for its activity.⁴⁵ We found that in healthy donors, the ratio of reduced vs disulfide proteoforms (PFR 53607 vs 5019691) is roughly 1.25:1 (Table S3). It will be relevant to correlate the activation of the chemokine with diseased patients to understand their immune responses. We also note that the presence of disulfides would be lost in the routine BUP workflow in the reduction–alkylation step.

Lastly, we identified two C─C motif chemokine 14 proteoforms: PFR 98045 (a.a. 20–93, 8.7 kDa) and PFR 98050 (a.a. 20–93, 8.7 kDa, K61E). Instead of a PTM, we found a natural variant (K61E) that was reported in the dbSNP database (ref: rs16971802).⁴⁶ Both C─C motif chemokine 14 proteoforms were identified only in XT well B with the 1% FDR threshold (Table S3). Curiously, the LFQ data showed contradicting results where there were no significant differences (PFR 98045) or little significance due to large variances in XT well B (PFR 98 050) between neat plasma and NP enrichment conditions (Figure S1C). We speculated that this variance resulted from individuals expressing the different variants and sought to compare wild type (WT, PFR 98045) and K61E (PFR 98050) proteoform expression in each plasma donor. Box plots in Figure S2 show that WT was differentially expressed in all three donors, and K61E was detected at the level of noise in biorep01. The fold-changes of WT/K61E were 54, 0.24, and 0.1 for biorep01, biorep02, and biorep03, respectively, with all p-values of <2 × 10⁻⁵. The potential difference in ionization efficiency due to the single site mutation was not considered in this analysis. However, the drastic fold-changes in the LFQ data still pointed to the conclusion where the donor for biorep01 only expressed WT, whereas the donor for biorep03 was homozygous for the K61E variant and therefore expressed no WT proteoform. Biorep02 had a slightly significant difference (q = 0.03) in WT expression compared to biorep03, so we could not rule out that the donor for biorep02 coexpressed both proteoforms. However, the ratio of WT/K61E in the biorep02 was 0.24, different from the 1:1 ratio that one would expect from a heterozygous gene variation (i.e., one copy of each in the genome). Therefore, further genome sequencing data of this individual will be complementary to TDP to address whether the single-site mutation contributes to the proteoform stability, resistance to degradation, or upregulation in gene expression. It is also unknown how widespread the variant is and whether it is functionally different. This example demonstrates that TDP quantitatively captures nuanced differences in low-abundant chemokine proteoforms that may assist the development of precision medicine in immunology. In addition to genomics, we now have a new tool to study gene variants at the proteoform level, which tightly connects their composition to their molecular functions.

Our group has deep interests in studying proteoforms in blood to identify rejection biomarkers for organ transplantation.^23-27 These studies were mostly performed using PBMCs, which are less accessible and less uniform compared to plasma samples. The NP enrichment allows us to quantify proteoforms involved in immune responses as well as several proteoforms we previously identified as “immunoproteoforms” in a panel that could be potential biomarkers for the liver-transplant outcome such as PFR 18628, 18631 (platelet factor 4, accession #P02776), and PFR 1464 (thymosin beta-4, accession #P62328).²⁷ This workflow will enable the study of immunoproteoform responses in plasma with greater depth via TDP to expand the impact of translational proteomics in organ transplant and autoimmune disease research.

Nanoparticles Reproducibly Fractionate Proteoforms in Plasma

To further demonstrate that NPs differentially fractionate proteoforms from neat plasma and from each other, we used a quantitative TDP approach. The volcano plots comparing neat plasma versus XT well A (Figure 5A) and neat plasma versus XT well B (Figure 5B) clearly reflect the differences of proteoforms identified in neat plasma and NP-enriched conditions. The proteoforms on the right side of the plots in red indicate enrichment in neat plasma, and those on the left side in blue indicate enrichments in NPs. We observed differential proteoform fractionation of over 1024 (2¹⁰)-fold differences in some cases, indicating an extreme differential and reproducible affinity for NPs, which were not observed in neat plasma. Figure 5C compares the two NP-enrichment conditions we used—XT well A versus XT well B. The high confidence in large fold changes (>2⁵-fold) again supports that the two wells were engineered to interrogate largely orthogonal proteins and their proteoforms with little overlap. The use of both wells improves the proteome coverage and serves as another level of plasma protein fractionation—not by MW but their affinities to the functionality of engineered NPs to form unique protein coronas.¹¹ We also found that proteoforms from the same family largely clustered together on volcano plots (see the plots highlighting DAP1 and chemokines in Figure S3). This is consistent with the observation that proteoforms from the same gene are often enriched together in the protein corona.

Figure 5.

Nanoparticles differentially fractionate proteoforms (10% FDR) in human plasma. (A-C) Volcano plots comparing neat plasma/NP and different NPs. Differential enrichments were characterized by large fold changes in proteoform abundances between neat plasma and nanoparticles (A, 4256 proteoforms; B, 4173 proteoforms) as well as between different nanoparticles (C, 4172 proteoforms). (D) Unsupervised clustering reveals unique clusters of proteoforms in neat plasma enriched by different NPs. (E) Source of variance plot shows that the major contributors of differences observed are from different NP treatments.

To assess the effect of NP enrichment in another way, we next performed unsupervised clustering of proteoform abundances under three treatment conditions (neat plasma, XT well A, and XT well B). In the heatmap output (Figure 5D), each condition is represented with a unique cluster of enriched proteoforms, resulting in three distinct proteoform profiles. Sample clustering suggests that both NP-enriched samples are more similar to each other than neat plasma, consistent with the larger proteoform and protein intersections observed in Figure 2A,B. Furthermore, the enriched clusters represent proteoforms uniquely identified in each treatment condition, correlating with additional GO functions revealed by the NPs in Figure 3. In summary, these quantitative analyses support that NPs differentially interrogate proteoforms in plasma.

A box plot of percent variation explained by all sources (Figure 5E), including biological replicates (different plasma donors), technical replicates (complete sample preparation procedure including Proteograph NP-enrichment and PEPPI steps), injection replicates (multiple LC-MS injections of the same samples), treatments (neat plasma, XT well A and XT well B), and residual (all other potential sources), shows that the majority of variance outside of the residual came from the different NP treatments. This is consistent with the observation that NPs differentially interrogate the proteoforms described above. The next largest variation source came from biological reps, representing differences in proteoforms from individuals. In Figure S4, we compare proteoforms from different donors in different treatment conditions (neat plasma, NPs) using volcano plots. We found that both neat plasma and two NP-enrichment wells allowed quantitative comparison of proteoform abundances between individuals. We note that the fold changes (<2⁵) and confidence level (−log₁₀FDR < 15) were much smaller than the comparison between treatments in Figure 5A-C, consistent with the smaller variance explained in Figure 5E. We also observed that biorep01 may be more different from biorep02 and biorep03 from more significantly changing proteoforms in volcano plots in all three treatment conditions (Figure S4), including the C─C motif chemokine 14 gene variation discussed above and highlighted in Figure S2. While the sample size and additional information about the donors are limited, the observation points to the future application of the assay in analyzing clinical samples from patients for precision medicine. Lastly, both technical and injection replicates demonstrated low levels of variation, suggesting that NP enrichment using the Proteograph workflow combined with PEPPI-MS for TDP is robust and reproducible.

Limitation and Future Directions

While we report a new record number of proteoforms identified and depth in plasma, this work is still limited by common challenges in TDP, particularly in the analysis of high MW proteoforms and the amount of material needed. As described here, we began with an SDS-compliant workflow to assess the potential and effect size of the NP approach, but the analysis of large proteoforms (>50 kDa) by LC-MS remains challenging due to several restrictions: 1) instrument resolving power and ion decay—limiting acquisition of isotopically resolved spectra of large/highly charged proteoform ions⁴⁰; 2) separation technology—proteoforms are poorly separated due to peak broadening and overlapping, resulting in extremely complex spectra that are hard to deconvolute; and 3) inefficient sample preparation and cleanup—due to the challenge in separation, fractionation of samples is often needed. In addition, these methods require precipitation of intact proteins to remove detergents, which typically requires >100 μg of total proteins.^26,34 Development of a streamlined protocol to deliver the protein corona to an efficient TDP process will greatly enable this approach.

New acquisition methods continue to improve the detection of large proteoforms. For examples, Ge and coworkers recently showed that using LC-MS combined with charge state deconvolution (low resolution, not isotopically resolved) can detect proteoforms up to 220 kDa for highly abundant proteins in specific tissues.⁴⁷ Sun and coworkers combined CZE separation and low resolution MS¹ acquisition to achieve plasma proteoform characterization in the 3–70 kDa “mid” mass range.³² However, low injection volumes (nL scale) substantially limit the sensitivity of CZE-MS and its ability to detect low abundant proteoforms.⁴⁸ Although limited to <30 kDa proteoforms, Fornelli and coworkers’ unique FAIMS approach directly addresses the challenge by reducing the complexity of the spectra and minimizing the need for multiple fractionation steps in the TDP analysis of plasma and serum.³¹ Approaches such as proton transfer charge reduction (PTCR)⁴⁹ and individual ion mass spectrometry (I²MS)^50,51 have recently been used for better MS¹ and MS/MS analyses of large proteoforms and their fragments that can be applied to the study of plasma proteome.

Compared with BUP analyses of the plasma proteome, this TDP workflow is still limited in proteome depth and the amount of material needed. Typically, in BUP, one well of Proteograph XT enrichment is sufficient to characterize 1500–2000 protein groups (Figure S5) using an Orbitrap Exploris 480 mass spectrometer, while the current TDP workflow is limited by MW, abundance, and sample preparation efficiency that only allows the characterization of proteoforms from 114 proteins. Given that proteoforms are the new currency in proteomics,²² we are encouraged by the ability to detect proteoforms in the low pg/mL concentration range and continue to improve our sample preparation workflow. For example, we are experimenting with combining XT wells A and B to improve protein recovery from gel fractionation as well as several detergent-free elution methods for direct infusion to LC-MS. As the proteomics field starts to recognize that proteoforms correlate more tightly to phenotypes in patient cohorts than tryptic peptides,^20-22,27 continual improvement of the sample preparation workflow and reducing the amount of sample needed will be crucial for the future application of plasma TDP in translational studies where patient samples are limited in quantity.

CONCLUSIONS

Nanoparticle-based protein corona enrichment of low-abundance proteins is a fast-growing approach in deep proteomics profiling of plasma proteome. Here, we adapt the BUP-based Proteograph workflow for TDP analysis and report a record number of 2841 proteoforms identified from 114 plasma proteins, revealing pathways previously hindered by abundant proteins, such as albumin and apolipoproteins. This approach will allow proteoform-level analysis and the discovery of biomarkers in diseases, which are more closely related to molecular function than protein group-level information. The workflow can also be coupled with emerging TDP acquisition methods such as low-resolution MS¹, CZE-MS, PTCR, and I²MS to further reduce spectral complexity, improve separation, and extend the mass range beyond 50 kDa.