Intact-Mass Analysis Facilitating the Identification of Large Human Heart Proteoforms

Abstract

Proteoforms, the primary effectors of biological processes, are the different forms of proteins that arise from molecular processing events such as alternative splicing and post-translational modifications. Heart diseases exhibit changes in proteoform levels, motivating the development of a deeper understanding of the heart proteoform landscape. Our recently developed two-dimensional top-down proteomics platform coupling serial size exclusion chromatography (sSEC) to reversed-phase chromatography (RPC) expanded coverage of the human heart proteome and allowed observation of high-molecular weight proteoforms. However, most of these observed proteoforms were not identified due to the difficulty in obtaining quality tandem mass spectrometry (MS2) fragmentation data for large proteoforms from complex biological mixtures on a chromatographic timescale. Herein, we sought to identify human heart proteoforms in this dataset using an enhanced version of Proteoform Suite, which identifies proteoforms by intact mass alone. Specifically, we added a new feature to Proteoform Suite to determine candidate identifications for isotopically-unresolved proteoforms larger than 50 kDa, enabling subsequent MS2 identification of important high-molecular weight human heart proteoforms such as lamin A (72 kDa) and trifunctional enzyme subunit α (79 kDa). With this new workflow for large proteoform identification, endogenous human cardiac myosin binding protein C (140 kDa) was identified for the first time. This study demonstrates the integration of the sSEC-RPC- MS proteomics platform with intact-mass analysis through Proteoform Suite to create a catalog of human heart proteoforms and facilitate the identification of large proteoforms in complex systems.

Graphical Abstract

Proteoforms are the different forms of proteins, which result from sources such as genetic variation, RNA editing and splicing, and post-translational modifications (PTMs).¹ A proteoform family consists of the different proteoforms from a single gene.² The identification and quantification of proteoforms are important to understanding biological systems because different proteoforms exhibit distinct biological functions.^3–7 Altered proteoforms have been observed in heart disease, the leading cause of death worldwide, suggesting the importance of proteoform analysis to understanding cardiac dysfunction. ^{4, 8–9} Mass spectrometry-based top-down proteomics is the most powerful available tool for identification and quantification of proteoforms. In a typical top-down proteomic analysis, precursor mass spectra (MS1) of intact proteoforms are acquired, the most abundant peaks are selected for fragmentation, and tandem mass spectra (MS2) of the fragment ions are acquired.¹⁰ Top-down proteomic analysis of the human proteome is technically challenging due to its wide dynamic range and high complexity. Analysis of high-molecular weight (MW) proteoforms is particularly difficult because there is an exponential MS signal-to-noise ratio (S/N) decrease with increasing MW.¹¹ Signal suppression due to the co-elution of highly abundant low-MW proteoforms imparts even more challenges for the analysis of high-MW proteoforms from complex mixtures. Thus, size-based separations are critical to observing and identifying high-MW proteoforms.

The Ge laboratory recently developed serial size exclusion chromatography (sSEC), which utilizes MS-compatible solvents for high resolution size-based fractionation of complex protein mixtures.^12–13 sSEC fractionation was combined with online reversed-phase chromatography (RPC) in a two-dimensional (2D) separation platform and top-down proteomic analysis of human heart tissue lysate was performed on a quadrupole-time-of-flight (Q-TOF) mass spectrometer.¹² Our 2D sSEC-RPC analysis resulted in the detection of 5360 proteoforms, with 47 proteoforms larger than 60 kDa. In this previous study, mostly MS1 data was acquired in order to profile the intact human heart proteome and determine high-MW candidates for subsequent targeted MS2 analysis, and targeted broadband MS2 data was acquired for a select set of high-MW proteoforms. Despite MS2 availability for 18 unique masses over 30 kDa, only two proteoforms were identified, creatine kinase (43 kDa) and trifunctional enzyme subunit beta (47 kDa), using the top-down search algorithm, MS-Align+. It is worth noting that these two proteoforms were isotopically resolved by the Q-TOF mass spectrometer, so the monoisotopic mass could be used for database searching. No proteoforms larger than 50 kDa were identified in the previous study, despite availability of acquired targeted MS2 data¹². It is a recognized challenge in online LC-MS/MS top-down proteomics to obtain sufficient fragmentation ions for identification of proteins larger than 50 kDa.^{10, 14} Indeed, a number of issues challenged successful top-down proteomic analysis of larger proteoforms. Though sSEC-RPC separation allowed for the MS1 detection of larger heart proteoforms, proteoforms larger than 50 kDa were not isotopically resolved by the Q-TOF mass analyzer, which prevented the use of monoisotopic mass in top-down search algorithms. Furthermore, MS2 of larger proteoforms on an LC-MS timescale often provides sparse fragmentation and low S/N for fragment ions. Co-isolation of more than one charge state of a parent ion during the targeted MS2 acquisition was used to improve the fragmentation efficiency and fragment ion S/N.¹⁵ However, this strategy also increased the MS2 spectral complexity because these spectra included fragment ions from multiple, co-eluting proteoforms. Confident identification of large proteoforms typically also requires manual validation of the fragment ions from the MS2 spectra due to the complexity of the data. However, this process requires a candidate sequence to query. These challenges combined prevented identification of the proteoforms >50 kDa in the original study.

The Smith laboratory has recently developed the freely available and open-source software program Proteoform Suite (https://smith-chem-wisc.github.io/ProteoformSuite/), which uses MS1 intact-mass measurements to identify proteoforms by comparing the observed experimental proteoform masses to theoretical masses derived from a database and to co-eluting experimental proteoform masses.^16–17 Proteoform Suite has been used to analyze proteomic datasets derived from S. cerevisiae^16–17, E.coli¹⁸, and murine mitochondria¹⁹ on an Orbitrap mass spectrometer. Although MS1 measurements do not provide PTM localization and the false discovery rate is typically higher than for MS2 analyses, Proteoform Suite enables MS1 intact-mass identification of proteoforms that were either not selected for or unable to be identified by MS2 analysis.^{17, 19} For deconvolution results, Proteoform Suite can input any .tsv file with mass, intensity, and retention time columns, making it versatile for any instrument or deconvolution method of choice. We used Proteoform Suite here to identify human heart proteoforms in the top-down proteomic dataset described above. This study, which integrated MS-compatible size-based fractionation and intact-mass analysis, identified 409 proteoforms <50 kDa, with 80 proteoform masses between 30 and 50 kDa. Notably, we devised a workflow that combines two of our freely-available software tools, Proteoform Suite and MASH Suite Pro, to identify 13 proteoforms larger than 50 kDa (Supporting Figure S-1). We implemented a new feature in Proteoform Suite that determines theoretical candidates for isotopically-unresolved proteoforms larger than 50 kDa by intact-mass analysis. We then used MASH Suite Pro to deconvolute MS2 spectra and query fragment ions against the candidate sequences determined in Proteoform Suite. This workflow enabled identification of important large human heart proteoforms that were previously unidentified.

The experimental methods are reported in Supporting Information. Briefly, proteins were extracted from human heart tissue and fractionated by size using sSEC. Fractions were analyzed by RPC-Q-TOF-MS, and MS1 data were deconvoluted to determine observed masses. Separate analyses in Proteoform Suite version 0.3.5 (https://github.com/smith-chem-wisc/ProteoformSuite/releases/tag/0.3.5) were performed for proteoforms <50 kDa (isotopically-resolved and monoisotopic mass reported) and proteoforms >50 kDa (isotopically-unresolved and apex of the charge-state deconvoluted peak reported). Proteoforms >50 kDa were not isotopically-resolved on the Q-TOF mass spectrometer; therefore, deconvolution could not provide the monoisotopic mass. Instead, the mass at the apex of the charge-state deconvoluted spectrum was reported, which is close to the average mass of the proteoform.²⁰ The average mass for each proteoform in the theoretical database was determined from the chemical formula. We enabled Proteoform Suite to identify candidates for high-MW proteoforms by implementing a notch search²¹ against the theoretical database using the average mass with a 2 Da mass tolerance. The 2 Da tolerance was chosen to be wide enough to capture the apex of the observed unresolved isotopic envelopes, while rejecting most theoretical proteoform matches.

In the analysis of proteoforms <50 kDa, there were 2253 unique experimental proteoforms, 409 of which were identified by Proteoform Suite at 7.5% FDR (Supporting Table S-1 and Supporting Figure S-2). An experimental proteoform corresponds to a mass observed in the deconvoluted MS1 spectrum, after applying the intensity cutoff and aggregation described in the Supporting Information. As previously discussed^{16–17, 19}, Proteoform Suite can currently only calculate a global FDR for intact-mass identifications because individual identifications are unable to be scored. There were also 45 ambiguous identifications, which matched to more than one proteoform identification (Supporting Table S-2). The remaining 1799 experimental proteoforms were unidentified, meaning they either did not form an accepted experimental-theoretical relation or did not match identification criteria¹⁶. The 409 identified experimental proteoforms were derived from 268 unique proteins (the number of unique accessions, see Supporting Table S-1). Proteoform families were constructed in Proteoform Suite from accepted mass differences between theoretical and experimental proteoforms (Supporting Figure S-3) as described previously.^{2, 16–17} A summary of the experimental proteoforms and proteoform families from the <50 kDa analysis is shown in Supporting Figure S-2. Examples of proteoform families and the charge-state deconvoluted MS1 spectra for heart proteoforms derived from the genes LASP1, CRIP2, HMGN1, and HSPB1 are shown in Supporting Figure S-3.

Previously published Proteoform Suite analyses were performed on samples separated by size using gel-eluted liquid fraction entrapment electrophoresis (GELFrEE) prior to RPC-MS analysis with an Orbitrap mass spectrometer, and the largest proteoform identified by intact-mass analysis was 31 kDa.¹⁷ Here, we used Proteoform Suite to analyze sSEC-fractionated data acquired using a Q-TOF mass spectrometer, and identified 80 proteoforms between 30 and 50 kDa. A histogram of identified experimental proteoform masses is shown in Supporting Figure S-4. The <50 kDa intact-mass analysis identified many heart proteoform families >30 kDa that are critical to heart function, including creatine kinases and actin isoforms (see Supporting Information and Supporting Figure S-5).

In the >50 kDa analysis, Proteoform Suite found at least one theoretical proteoform candidate for 36 experimental proteoforms (Supporting Table S-3), 30 of which had more than one theoretical match. The FDR obtained for the >50 kDa analysis was quite large, 82.7%, due to the wide 2 Da search tolerance, which results in a high chance of matching with a decoy proteoform mass. Given this high FDR, we do not consider these experimental proteoforms to have been identified, but rather to be candidates for subsequent analysis of the available targeted MS2 data.

The proteoform candidates determined in this manner were used to guide MS2 data analysis in order to identify proteoforms. For example, a co-eluting group of 65 kDa and 72 kDa proteoforms (Figure 1A) had previously been targeted by broadband co-isolation and collisionally activated dissociation but were not able to be identified with a typical database search in MS-Align+, likely due to the use of average precursor mass in place of monoisotopic mass and the MS2 spectral complexity due to daughter ions originating from multiple parent ions. By intact-mass analysis, Proteoform Suite determined the theoretical candidate for the group of 72 kDa proteoforms to be acetylated lamin A (gene LMNA), including both mono- and bis- phosphorylated proteoforms. Proteoforms in the LMNA gene family are intermediate filament proteins, which make up the nuclear lamina.²² Since changes to LMNA phosphorylation have been linked to cardiomyopathies, it is important that we can detect and quantify these specific proteoforms in the heart.^22–23 We used MASH Suite Pro to query the MS2 fragments against the LMNA sequence determined by Proteoform Suite, and we confirmed the identity of the acetylated lamin A proteoform with both N-terminal and C-terminal sequence tags that were manually validated in the raw data, as previously described²⁴ (Figure 1B). The lamin A identifications led us to hypothesize that the three co-eluting 65 kDa proteoforms were three proteoforms from the lamin C isoform from gene LMNA: acetylated, acetylated and mono-phosphorylated, and acetylated and bis-phosphorylated. Likewise, we used the same MS2 spectrum to confirm the identification of the lamin C isoform with a C-terminal sequence tag (Figure 1C, Supporting Figure S-6 ). Lamin C had not been identified as a candidate for the 65 kDa proteoforms in Proteoform Suite because it was absent in the theoretical database downloaded from UniProt, which contained only canonical protein sequences. This study demonstrates how intact-mass analysis in Proteoform Suite can guide MS2 analysis to enable the identification of isotopically-unresolved proteoforms >50 kDa. A full list of validated MS2 fragment ions for lamin A and lamin C can be found in Supporting Tables S-4.1 – S-4.4.

Figure 1.

A) Charge-state deconvoluted MS1 spectra for lamin A and lamin C isoforms of LMNA (fraction 5, 28–29.3 min). Proteoform Suite identified three candidate proteoforms from the lamin A isoform: with acetylation alone, with both acetylation and mono-phosphorylation, and with both acetylation and bis-phosphorylation. Three additional co-eluting masses were also observed in the same retention window. Data from two MS2 experiments (15 eV and 20 eV CAD energies) were combined to generate sequence tables shown in panels B) and C). B) Lamin A isoform sequence with matching MS2 fragments and highlighted C-terminal sequence tag (pink). Zoom-in of a representative MS2 spectrum (15 eV CAD) from 910–970 m/z shows y-ions corresponding to the lamin A C-terminal sequence tag. C) Lamin C isoform sequence with matching MS2 fragments and highlighted C-terminal sequence tag (blue). Zoom-in of a representative MS2 spectrum (15 eV CAD) from 870–910 m/z shows y-ions corresponding to the lamin C C-terminal sequence tag. Additional zoom-in of the MS2 spectrum for the highlighted fragment ions are found in Supporting Figure S-6.

Proteoform Suite determined that two observed 79 kDa proteoforms closely matched the unmodified and succinylated trifunctional enzyme subunit α (gene HADHA). Similarly, we used a previously acquired targeted MS2 spectrum to confirm this candidate identification (Supporting Figure S-7B and S-7D). Proteoform Suite also identified trifunctional enzyme subunit β in the <50 kDa analysis (47 kDa, gene HADHB), which was validated with targeted MS2 analysis in our previous study (Supporting Figure S-7A and S-7C, Supporting Tables S-4.5 and S-4.6).¹² Upon inspection of the charge-state deconvoluted MS1 spectrum for trifunctional enzyme subunit β, we observed a peak shifted by +100 Da from the unmodified proteoform, which likely corresponds to its succinylated proteoform.

Proteoform Suite identified candidate proteoforms for cardiac myosin binding protein C (cMyBP-C, 140 kDa) encoded by MYBPC3 gene, and myosin heavy chain (MHC, 223 kDa) encoded by MYH7 gene, two major constituents of the cardiac sarcomere. Examination of previously acquired MS2 data against the full-length cMyBP-C sequence (aa 1–1274) determined by Proteoform Suite revealed matching y-ions from the C-terminus. Proteoform Suite also identified a proteoform candidate as bis-phosphorylated cMyBP-C with methionine cleaved (aa 2–1274). Matching the MS2 data against the methionine-cleaved cMyBP-C sequence revealed matching b-ions and an N-terminal sequence tag (Figure 2, Supporting Table S4.7). Other co-eluting intact masses indicate mono-phosphorylated full-length, mono-phosphorylated methionine-cleaved, and tris-phosphorylated methionine-cleaved cMyBP-C proteoforms. The recombinant cMyBP-C had previously been studied by Ge et al²⁵, however we report for the first time the identification of the endogenous proteoform from human heart tissue. This proteoform identification is highly relevant to the study of heart function and dysfunction because phosphorylation of cMyBP-C plays a key role in regulating heart contraction and relaxation.²⁶ While MS2 data was not available for the verification of the candidate 223 kDa proteoform, MHC is a known major constituent of the cardiac sarcomere and has been previously identified in mouse heart tissue lysate by bottom-up MS/MS data.²⁷

Figure 2.

A) Original and charge-state deconvoluted MS1 spectra for the cMyBP-C proteoforms (fraction 4, RT 33–35 min). B) MS2 spectrum acquired by co-isolating all charge states in the 700–800 m/z range over the protein elution window (CAD energy 18 eV). C) Sequence for cMyBP-C with methionine cleaved with matching b- and y- fragment ions and highlighted (green) N-terminal sequence tag. Zoom-in of the MS2 spectrum between 520–650 m/z shows b-ions corresponding to the highlighted N-terminal sequence tag (green) and other ions in the zoom-in spectrum that match to the proteoform sequence (grey).

There are top-down search software programs available that allow precursor mass searches with average mass instead of the monoisotopic mass, such as ProsightPC and ProsightPTM²⁸, which was also able to identify the full-length cMyBP-C proteoform when the precursor and fragment masses were input manually. Another approach to identify high-MW proteoforms uses ultra-high-resolution mass spectrometers, such as the FT-ICR mass spectrometers^{4, 9, 13, 25, 29–32}, to isotopically resolve high-MW proteoforms and thereby facilitate their identification by monoisotopic mass analysis. The intact-mass analysis method we present here makes top-down analysis of larger proteoforms in complex samples accessible to laboratories with access to more widely available instrument platforms. Due to the difficulty in obtaining high-quality MS2 spectra of large proteoforms, particularly on an LC-MS time scale, both the acquisition and analysis of MS2 data performed in MASH Suite Pro are targeted techniques. In this study, we only reported >50 kDa proteoform identifications if a manually-validated sequence tag was obtained. Analysis in Proteoform Suite is more automated and amenable to large-scale analyses, but intact-mass analysis does not provide identifications with as much confidence as MS2 data, particularly in isotopically unresolved data and in complex systems such as human heart tissue. ¹² However, as identified proteoforms are increasingly catalogued in repositories such as the Consortium for Top-Down Proteomics Proteoform Atlas (http://atlas.topdownproteomics.org/), the ability to automatically match intact masses with candidate theoretical proteoforms will facilitate both qualitative and quantitative proteoform analyses. We also note a major benefit of intact-mass analysis and proteoform family construction is its potential use for guided targeted MS2 data acquisition. If an intact mass is matched to a candidate proteoform with biologically interesting PTMs or from a gene of interest, subsequent targeted MS2 analysis can be performed on the sample. Additionally, the experimental-experimental comparison can reveal unidentified proteoform families of interest (e.g., those with multiple phosphorylation events), that can be followed-up with targeted MS2 analysis.

CONCLUSIONS

We used the open-source and freely available software program Proteoform Suite to identify proteoforms by intact mass in a dataset of serial size exclusion chromatography-separated fractions of human heart tissue lysate. Proteoform Suite identified 409 unique proteoforms <50 kDa, with 80 identified proteoforms of 30–50 kDa, whereas the largest proteoform identified by intact mass in previous Proteoform Suite analyses was 31 kDa. While sSEC fractionation had enabled 45 unique experimental proteoforms >50 kDa to be observed in a previous top-down study, the high-MW candidates which were targeted for MS2 in that work were left unidentified by a conventional top-down search algorithm due to either excessive spectral complexity or inefficient fragmentation. Proteoform Suite determined theoretical candidates for many of these proteoforms, some of which were subsequently confirmed with previously acquired targeted MS2 data. The integration of the MS-compatible sSEC proteomics platform with intact-mass analysis through Proteoform Suite enabled identification of many important human heart proteoforms and proteoform families. Proteoform Suite can identify proteoform candidates for subsequent targeted MS2 analysis and proteoform quantification, making it a valuable tool for top-down proteomics.