Proteomics Research in Cardiovascular Medicine and Biomarker Discovery

Abstract

Proteomics is a systems physiology discipline to address the large-scale characterization of protein species within a biological system, be it a cell, a tissue, a body biofluid, an organism, or a cohort population. Building on advances from chemical analytical platforms (e.g., mass spectrometry and other technologies), proteomics approaches have contributed powerful applications in cardiovascular biomedicine, most notably in: 1) the discovery of circulating protein biomarkers of heart diseases from plasma samples; and 2) the identification of disease mechanisms and potential therapeutic targets in cardiovascular tissues, in both preclinical models and translational studies. Contemporary proteomics investigations offer powerful means to simultaneously examine tens of thousands of proteins in various samples, and understand their molecular phenotypes in health and disease. This concise review introduces study design considerations, example applications and use cases, as well as interpretation and analysis of proteomics data in cardiovascular biomedicine.

Overview

Over the past 20 years, there has been an accelerating growth in the application of proteomics in cardiovascular biomedicine, building upon pioneering studies with translational significance (1,2) and mechanistic insights (3–8). The majority of human genes functions to create proteins, which are the molecular workhorses that carry out virtually all metabolic, signaling, and physiological functions in life. Many human diseases may be characterized by the elicited changes in proteome configurations. In complex late-onset diseases in particular, considerable environmental components exist, such that individual genetic variants are often poorly predictive of disease states. Telltale molecular signatures of acquired cardiovascular diseases may instead manifest through intermediate phenotypes, including transcript and protein abundance. Beyond what may be learned from gene and transcript information alone, molecular changes at the protein layer, in particular, can provide independent insights into disease mechanisms in the following 3 areas:

First, it has gradually emerged from large-scale studies that transcript and protein changes in a system may, at times, correspond poorly—with transcript levels explaining as little as 10% to 30% of variations in protein abundance. Although the exact contribution of transcripts to protein-level abundance is debated, it is evident that in the heart and other organs, a large number of post-transcriptional modulators can alter disease-driver protein expression and function during pathogenesis, without changes in transcript abundance (9,10). These modulators include components of the nonsense-mediated decay pathway, long noncoding RNAs, and microRNAs (11–14).

Secondly, plasma proteins provide an accessible readout of the status of potentially all tissues, and are the sources of many current biomarkers in use (Table 1). Clinically useful biomarkers can include circulating proteins that have been secreted or leaked directly into the plasma from resident cells (e.g., myocytes) following diseases or injuries (15), and are thus spatially uncoupled from the transcript change at the tissue of origin. The tissues of origin of circulating proteins may be unknown, inaccessible, or invasive to procure, making it impractical to hunt for these potential biomarkers via transcript measurements, instead of protein measurements.

Table 1.

Selected Protein Biomarkers for Cardiovascular Diseases

Protein Biomarker	Abbreviation	Disease Relevance	Required Assay Sensitivity (Estimate)	*Discovery Period	Reference
Apolipoprotein A-I	APOA	Cardiovascular event risk	~1 mg/ml	Mid-1980s	(96)
Apolipoprotein B	APOB	Cardiovascular event risk	~1 mg/ml	Mid-1980s	(97)
B-type natriuretic peptide	BNP	Heart failure, acute coronary syndrome	~100 pg/ml	Early 2000s	(98)
C-reactive protein	CRP	Cardiovascular event risk	~10 µg/ml	Late 1990s	(99)
Creatine kinase- myocardial band	CK-MB	Acute myocardial infarct, myocardial necrosis	~1 ng/ml	1960s–1970s	(100)
Cystatin-C	CST3	Cardiovascular event risk	~1 µg/ml	2000s	(101)
Fibrinogen	FBN	Cardiovascular event risk	~1 mg/ml	1980s	(99)
Lipoprotein- associated phospholipase A2	Lp-PLA2	Coronary heart disease risk	~100 ng/ml	2000s	(102)
Myeloperoxidase	MPO	Ischemic heart disease; acute coronary syndrome	~10 ng/ml	1980s	(103)
Myoglobin	MYO	Myocardial infarction, necrosis	~ 10 ng/ml	Late 1970s	(104)
Serum amyloid A	SAA	Coronary artery disease	~10 µg/ml	1990s	(105)
Troponin I	cTnI	Myocardial injury, myocardial infarction	~10 pg/ml	1970s–1990s	(106)
Troponin T	cTnT	Myocardial injury, myocardial infarction	~10 pg/ml	1970s–1990s	(107)

^*Periods of initial discovery and/or assay development (108,109).

Thirdly, disease processes may be mediated not by the altered abundance of gene products, but by other functional parameters of the proteome, including protein post-translational modifications (PTMs), protein-protein interactions, and protein degradation, which take place after the protein molecules are synthesized and cannot be predicted from genetic information a priori.

Proteomics technologies allow researchers to measure protein function on a large scale, and hence interrogate the molecular layer that closely abuts physiological phenotypes (16). However, the complexity of the human proteome also presents a daunting analytical challenge for large-scale characterization. For example, circulating proteins in human plasma, found at a concentration of ~70 mg/ml, are estimated to comprise at least 10,000 distinct protein species (17) over a concentration range of >10 orders of magnitude 10-billion-fold differences) (18). Approximately 3,000 proteins in plasma may be viewed as classical resident plasma proteins, which tend to occupy the higher end of the abundance spectrum. These resident proteins include extremely abundant species, such as albumin, immunoglobulins, transferrins, and fibrinogens, which are found in the mg/ml concentration range and account for >90% of plasma protein content. Other plasma proteins come from secretion or leakage from various organs before being diluted into the 5 liters of peripheral blood, and are found in the µg/ml to ng/ml range. Very low-abundance proteins, including interleukins 6 and 12, tumor necrosis factor (TNF)-alpha, and other cytokines, circulate in the pg/ml range, and are typically masked by higher-abundance proteins in untargeted proteomics analyses (Figure 1). Adding to this complexity, plasma proteins may be proteolytically processed, creating numerous degradation product species, and repetitive sequences in proteins can give a spurious appearance of higher abundance. These challenges are compounded by the absence of a protein equivalent of the polymerase chain reaction, such that amplification of minute samples is not possible.

Figure 1. Sensitivity and Coverage of Various Proteomics Study Designs.

Properties of 3 common proteomics study designs are listed. Discovery (shotgun), targeted, and targeted discovery proteomics approaches take different strategies to address proteome coverage, number of samples, and sensitivity limits. Discovery proteomics experiments have high coverage (up to 10,000 proteins in a single sample), but have limitations in throughput; fewer samples/subjects can be analyzed. Targeted discovery approaches focus on analyzing a panel of high-potential targets in sufficient numbers of samples. Targeted proteomics is able to achieve the highest sensitivity, which allows the detection of low-abundance plasma markers, such as TNF and IL-6, but at the expense of scope and throughput. Graph (center) shows the typical sensitivity range of discovery, targeted discovery, and targeted approaches, juxtaposed with the concentrations of selected plasma proteins and disease markers. Example technological platforms and additional notes are shown on the right. CK-MB = creatine kinase-myocardial band; CRP = C-reactive protein; ELISA = enzyme-linked immunosorbent assay; IL-6 = interleukin 6; LDL = low-density lipoprotein; MS = mass spectrometry; NT-proBNP = N-terminal B-type natriuretic peptide; TNF = tumor necrosis factor.

In the following 3 segments of this review, we will discuss 3 aspects of proteomics applications in cardiovascular biomedical research: 1) study design for biomarker research and discovery, with discussions on approaches to circumvent analytical challenges; 2) characterization of multidimensional protein parameters in disease mechanism research; and 3) interpretation and validation of proteomics data/methodologies.

Study Designs for Biomarker Research and Discovery

With precision medicine initiatives demanding a broader pool of measurable proxies of individual characteristics, the call for investigations to discover new candidate biomarkers will likely continue to increase. Successful screens of protein biomarker candidates require obtaining data: 1) with good coverage of all the proteins present in a sample; and 2) in a sufficient number of subjects to ensure adequate power for discovery. Unlike microarrays or RNA sequencing (RNA-seq), a single proteomics experiment often cannot simultaneously obtain high coverage and throughput, necessitating tradeoffs in experimental designs. The advantages and disadvantages of technological platforms should therefore be evaluated comprehensively on the basis of the scope and depth of analysis required, and determined prior to the commencement of chemical analysis. Three common designs, all of which can provide both qualitative and quantitative data on proteins, are: 1) global discovery; 2) targeted; and 3) targeted discovery. Each has distinct strengths and may be better suited to particular experimental goals. Each may be performed in suitable technological platforms that offer different throughputs, analytical depth, and sensitivity. We provide here a brief summary on the strengths and considerations for each type of approach. The Central Illustration presents an overview of data acquisition and analysis; see Table 2 for a glossary of selected technological terms commonly in use.

Central Illustration. Overview of Proteomics Analyses for Cardiovascular Diseases.

This figure lists major components in common proteomics workflows, as well as their associated experimental considerations in biomarker discovery and disease studies. From top to bottom, protein samples are collected from the plasma of human cohorts or cardiac tissues in animal models according to study goals (either biomarker discovery or mechanistic studies). Three major study approaches (discovery, targeted, and targeted discovery) take different strategies between proteome coverage and analytical throughput, and utilize different technological platforms, including mass spectrometry and protein arrays. Following the acquisition of large-scale datasets, data are processed to identify the protein species present and to deduce their relative quantities across samples. Subsequently, a number of statistical and bioinformatics workflows are used in data interpretation to extract insights from datasets. Network analysis casts proteins in the context of interaction networks and/or altered cellular pathways. Statistical learning and modeling methods connect the identified molecular features to orthogonal phenotypes, identify signatures, and offer information on subject classification or predictive analysis. The identified protein signatures will require validation, which can be achieved by complementary translational studies, including in vitro biochemical analysis and large cohorts. PTM = post-translational modifications.

Table 2.

Glossary of Selected Technologies Used in Cardiovascular Proteomics

Terms	Description
Amino acid sequence	A segment of the primary structure of the protein where individual amino acids are linked by peptide bonds.
Aptamer array	A non-MS proteomics technology where single-stranded DNA oligonucleotides with affinity for specific protein targets are used as a means to detect and quantify up to ~1,000 different protein targets.
Liquid Chromatography	A common front-end separation technology in proteomics experiments to fractionate complex sample mixture. It is used to resolve and enrich peptides by their chemical properties such as size, hydrophobicity, or charge, and the like.
Mass spectrometry	An analytical technology to measure the mass of an analyte (e.g., peptide, protein, amino acid, and so on) in the format of mass-to- charge ratio.
Mass spectrum	The data output of MS. A mass spectrum is a plot displaying the relative intensity or abundance of the analyte ions (y-axis) and their mass-to-charge ratio (x-axis).
Mass/charge (m/z)	The principal measurement of a mass spectrometer: the mass of the ionized analytes divided by the charge they carry. Also commonly referred to as mass-to-charge ratio.
Multiple-reaction monitoring	A targeted proteomics approach in which a particular type of mass spectrometer is programmed to only measure a few peptides and their fragment ions of a target protein of interest, while filtering out all other peptides and proteins in the sample, thus enhancing specificity and sensitivity of the detection/quantification of the targets. Also known as selected reaction monitoring (SRM).
Search engine	A software tool used in sequence database search to identify proteins from MS data (e.g., SEQUEST (110)).
Sequence database search	A method to identify proteins from MS data by matching the collected experimental mass spectra to theoretical mass spectra generated from a protein sequence database, such as UniProt (19) or RefSeq (111).
Spectral library search	A method to identify peptides by matching MS data to previously known experimental spectra in a library, such as COPaKB (21) or PeptideAtlas (20).
Shotgun proteomics	A method to identify and quantify proteins on a large scale, in which proteins are first digested into short segments (peptides) that are amenable to sequence determination via tandem MS analysis. The identity of each present protein is then inferred from the sequenced peptides. Also known as bottom-up proteomics.
Tandem mass spectrometry	An analytical method involving the fragmentation of peptides (or other analytes) via multiple stages of MS analyses to obtain sequence/structural information of the analytes. Also known as MS/MS or MS2.
Tandem mass tags (TMT)/iTRAQ	Commercially available chemical labels containing stable isotopes, which can be used to label protein samples for multiplexed analysis and comparison.

iTRAQ = isobaric tags for relative and absolute quantitation; MS = mass spectrometry

Global Discovery Proteomics

Global discovery approaches offer an unbiased view of the protein species that may be found in a sample. Shotgun proteomics is one of the most widely used and standardized global discovery strategies. In shotgun analysis, proteins are first enzymatically digested (e.g., using trypsin) into peptides in vitro, and then subjected to tandem mass spectrometry (MS) analysis. The resulting mass spectra are then searched against a protein sequence database (19) or spectral library (20,21) for protein identification. Various downstream data analyses, such as spectral counting and network analysis may be performed to extract other qualitative and quantitative information from the data. A major strength of shotgun proteomics is that it casts the widest net towards potential biomarkers (covering thousands of proteins) (22) and does not require the development of individual assays for each target (23,24). Two weaknesses are its lower overall sensitivity, which may bias it against biologically significant proteins at low abundance (see later discussion on Improving Detection of Low-Abundance Proteins and Throughput), and less accurate quantification compared with targeted proteomics.

Targeted Proteomics

Targeted proteomics offers the highest sensitivity and specificity among MS-based proteomics approaches. MS-based targeted approaches require the programming of a specific instrument (e.g., a triple quadrupole mass spectrometer), and development of methods (e.g., multiple reaction monitoring [MRM]) to monitor only a limited number of analytes (via predetermined peptide mass/charge and fragment information) to achieve its high sensitivity. It was demonstrated that targeted MS, in conjunction with stable isotope dilution, was able to verify a panel of cardiovascular biomarker candidates at a limit of quantification (LOQ) ranging from 2 to 15 ng/ml (25). Targeted MS is also applicable to post-translationally modified proteins with PTM site specificity (26), with the absolute quantity of 14 phosphorylation sites on cardiac Troponin I as a notable example (27). Ongoing effort is being devoted to the development of targeted protein assays aimed at cardiac biomarker analysis (28–30). A disadvantage of targeted proteomics is that the practice of only utilizing 1 or a few peptides in the experiments may not represent the behavior of the full-length protein (e.g., due to proteolysis). Specific assays must be developed for each protein being targeted, which requires substantial technical expertise from specialized laboratories. However, once developed, other investigators and facilities in clinical and research settings may readily adapt the assays. Due to the labor and costs required to validate assays, the number of proteins that can be realistically analyzed in a targeted biomarker search may be limited (~100 or fewer). Accordingly, targeted proteomics is most suitable for characterizing subsets of likely candidate biomarkers with prior disease implications. A limited number of non-MS-based antibody assays are also available for targeted quantification of low-abundance proteins in the plasma (e.g., enzyme-linked immunosorbent assays [ELISA]).

Targeted Discovery Approaches

A middle ground between a targeted and an untargeted approach is sometimes sought, which may be called targeted discovery. The goal is to attain the best of both worlds: focusing on a finite panel of targets, while maintaining sufficient throughput to analyze large number of samples. Oftentimes, this comes at the expense of scope (number of proteins analyzed) and flexibility, but, on the upside, panels of “likely bets” can be preselected by manufacturers or by the investigators to increase the probability of discovery (31). A number of MS-based (e.g., sequential window acquisition of all theoretical mass spectra [SWATH] (32)) and non-MS-based affinity proteomics (e.g., antibody- and aptamer-based arrays (31,33,34)), are being used to discover candidate biomarkers. Antibody-based arrays have been utilized to discover markers for ischemic stroke, using a proteomics chip of 92 high-potential cardiovascular proteins (35), including B-type natriuretic peptide (BNP), growth differentiation factor (GDF)-15, matrix metalloproteinase (MMP)-12, and others. Aptamer arrays that use nucleotides in place of antibodies for target binding have been utilized to discover protein myocardial infarct markers and to evaluate a panel of blood proteins as risk factors for prediction of coronary heart diseases (33,34,36,37).

Improving Detection Of Low-Abundance Proteins And Throughput

To improve detection of low-abundance proteins in plasma for MS-based analysis, commercial immunodepletion columns, including the Human IgY14 and SuperMix columns (Sigma-Aldrich, St. Louis, Missouri), are used to remove top- to medium-abundance proteins. Simplifying the sample mixture helps unmask low-abundance proteins for detection, but many proteins or fragments of interest may be inadvertently codepleted by the process. Typically, a sample containing 100 µl of plasma may only yield ~140 µg of proteins after 98% of high-abundance species are depleted, an amount sufficient for a limited number of replicate analyses. To target challenging proteins in plasma and other samples, additional preprocessing steps, including size-exclusion chromatography, multidimensional liquid chromatography, and immunocapture may be used, but necessitate greater starting amounts of samples (e.g., 0.01 to 1 ml of plasma) and additional processing time.

To achieve multiplexing, minimize experimental variations, and increase throughput, multiple protein samples can be barcoded with stable isotopes and analyzed in a single experiment at the expense of costs. Commercially available isobaric tags for relative and absolute quantitation (iTRAQ) and tandem mass tag (TMT) labels, small amine-reactive molecules covalently linked to differential mass reporters and mass balancers, can be used to tag peptides such that up to 10 sets of barcoded samples may be coanalyzed in 1 MS experiment. During tandem MS, the mass reporters separate from the balancers and act as surrogates for comparison of protein quantity of each sample. This multiplexing strategy has been profitably used to identify potential infarct markers in a study that analyzed the plasma of 16 patients at multiple time points following septal ablation treatment. With the aid of 4-plex iTRAQ labels, the proteomics result quantified >3,000 human plasma proteins in 16 patients and discovered 333 regulated proteins (38), including known markers of spontaneous myocardial infarction, such as creatine kinase-myocardial band (CK-MB) and troponin. Using 6-plex TMT labels, the proteomic analysis of the plasma samples of ~1,000 human subjects has been recently demonstrated (39). The popularization of new 10-plex TMT labels will further increase throughput.

Multidimensional Protein Parameters in Disease Mechanism Research

Contemporary proteomics applications in basic disease mechanism research often incorporate additional experimental techniques that target increasing numbers of “dimensions” of protein properties. Whereas previous proteomics studies focused largely on gene product abundance (up- and down-regulation in quantity as primary drivers of changes in function), there has been a growing appreciation of other, equally significant proteome parameters that modulate cardiovascular biology. These parameters include post-translational regulation, protein-protein interactions, and protein turnover, each of which may be altered in disease and present potential therapeutic targets (9). These parameters are aiding in efforts to fully comprehend the disease proteome as a dynamic, multidimensional entity. Ongoing developments in these areas may expand the parameter space from which future biomarkers may be found.

Protein Post-Translational Regulation

Post-translational regulation through covalent modification of proteins is a major mechanism of cardiac signaling. Over 200 protein PTMs that can alter the chemistry and interaction surface of a protein, and thereby its structure, activity, binding partners, or subcellular localization, are known (40). Although over 500,000 PTM sites are known to exist (41), available site-specific antibodies target relatively few of them. MS-based proteomics techniques do not require separate reagents (i.e., antibodies) for each site, and hence can be effectively used to examine new sites or sites for which antibodies are not available.

For a number of reasons, PTM studies often require additional sample preparation and enrichment considerations (e.g., with titanium dioxide [TiO₂] (42)) before data acquisition,. First, modified proteins frequently exist in lower abundance than the total protein population (although in some instances, such as in the case of secreted glycoproteins, almost every protein molecule is modified). For example, if only 10% of a protein species is modified, then the modified proteins are effectively 10 times more difficult to detect than the total (modified plus unmodified) population. Secondly, whereas identifying a protein only requires detection of any of its multiple proteolytic peptides along its primary structure, PTM analysis requires the precise detection of the peptide containing the PTM site, which sometimes can lie in a region of the protein that is not amenable to MS detection (e.g., sequences without trypsin digestion sites). Thirdly, many PTMs are chemically labile and biologically transient, making their capture and enrichment before signal measurement essential.

Post-translational modifications under intense investigation for their roles in cardiovascular diseases include phosphorylation (43,44), acetylation (45,46), SUMOylation (47), glycosylation (48,49), S-nitrosylation (50,51), ubiquitination (52), and others (53,54). The remainder of this section highlights proteomics approaches to understand protein phosphorylation, acetylation, and oxidative modifications.

Multiple major kinase pathways operate in the heart, including protein kinase A (PKA), protein kinase B (PKB/Akt), mitogen-activated protein kinase (MAPK)/extracellular signal-regulated kinase (ERK), protein kinase C (PKC), glycogen synthase kinase 3b (GSK3b), calcium/calmodulin-dependent protein kinase II (CamKII), and others; hence, it is no surprise that phosphoproteomics is commonly applied in the study of heart diseases. A well-known example of a phosphorylated protein essential for cardiac energetics is pyruvate dehydrogenase (PDH) (55), which converts pyruvate to acetyl coenzyme A for respiration, and whose activity is dictated by its phosphorylation by PDH kinases (PDKs). PDH phosphorylation thus serves as a major regulatory checkpoint of cardiac energy metabolism. Classically, phosphorylation sites were mapped by phosphorus 32 (³²P)-labeling and Edman degradation. Up to ~50,000 phosphorylation sites may now be identified in a single study using shotgun proteomics (56) combined with enrichment strategies. Mapping in the heart has unearthed a large number of phosphorylation sites, including those common to multiple tissues and many that are unique to the cardiovascular system (57). Targeted mapping efforts may also be directed towards selected cellular compartments (e.g., mitochondria (58)), pathways (e.g., those targeted by phosphodiesterase 9A [PDE9A]/phosphodiesterase 5 [PDE5] (59) and CamKII (60)), specific proteins (e.g., the myosin-binding protein C [MyBP-C] phosphoproteome (61)), or multiple specific sites on the same proteins (26).

Acetylation of histones plays a key role in gene expression by modulating the accessibility of transcriptional elements to DNA wound around the histones. A family of enzymes known as histone acetyltransferases (HATs) catalyzes the addition of the acetyl group to proteins, whereas histone deacetylases (HDACs) revert the modification. Although protein acetylation was classically described on histones, other proteins are also modified. Protein acetylation participates in nonchromatin processes, including metabolism and protein localization, as well as protein stability (45,46), and can be mapped with proteomics techniques following enrichment by anti-acetyl-lysine antibodies (62,63). Over one-third of mitochondrial proteins are subjected to acetylation, including a large number of metabolic enzymes. Acetylation in mitochondria may function in cardioprotective contexts (46), whereas hyperacetylation of succinate dehydrogenase in mitochondria may drive energy imbalance in the development of heart failure (63).

Reactive oxygen and nitrogen species can confer covalent modifications (e.g., S-nitrosylation, S-sulfhydration, and S-glutathionylation) on cysteine residues in redox-sensitive proteins. These modifications can change the activity and binding surfaces of target proteins, and thus directly participate in cardiovascular biology in a number of disease contexts. For example, it is known that nitric oxide (NO) serves as a critical signaling messenger in ischemic preconditioning. NO signals partly by causatively increasing protein S-nitrosylation, which has been proposed to mediate cardioprotection by reducing other, pernicious forms of cysteine oxidation on the target proteins during reperfusion injury (64,65). To detect oxidative modification en masse, biotin switch proteomics are used that first chemically block free thiols, then uses a specific reducing agent to reduce different reductive states of oxidative PTMs (e.g., ascorbic acids may be used to reduce only S-nitrosylated thiols, but not other modifications), and finally labels the freed thiols with biotin or isotope-labeled tags (cys-TMT or iodo-TMT) (51,66).

Protein Temporal Dynamics

Protein homeostasis and proteolysis are broadly implicated in heart diseases and injuries. Classical studies of protein turnover required pulse-chase of radioisotopes and isolation of target proteins. With recent advances, proteomics studies can measure turnover dynamics of thousands of proteins in multiple tissues and organisms (67,68). Combining stable isotope labeling, mathematical modeling, and shotgun proteomics, recent investigations into the turnover of ~5,000 cardiac proteins in mouse cardiac hypertrophy (69,70) revealed novel correlations between proteome profiles and phenotypic functional limitations. For example, although the hypertrophic heart is known to switch in fuel preference from fatty acid to glucose, the expression of glycolytic proteins does not change accordingly (71). Instead, the turnover of most glycolytic proteins was found to accelerate drastically in hypertrophy without concomitant expression changes. Thus static expression profiling experiments alone would not have discovered this aspect of cardiac pathophysiology had there been no a priori knowledge of increased glycolysis, whereas a large-scale experiment of protein turnover might have led to the generation of relevant hypotheses. A possible connection between turnover and function is that old proteins may have reduced activity compared with young, functional proteins. For example, the glycolysis enzyme triose-phosphate isomerase (TPI) can become chemically deamidated upon catalytic cycling. In another example, the stem cell factor/KIT ligand (KITLG) has been found to lose biological potency by 50-fold after chemical deamidation (72).

Protein-Protein Interactions

Physical liaison between proteins is a primary means whereby information is transduced along signaling pathways. The target specificity and activity of PKA are markedly regulated by which of the ~40 A-kinase anchoring proteins (AKAPs) it primarily interacts with (73), to mediate kinase signaling, and which can change during heart disease (74). Protein-protein interaction analysis is pursued to characterize signaling networks, discover the components of protein complexes, and identify the contexts in which a protein may function in health and disease (75). A powerful approach to analyze protein-protein interaction is affinity purification-mass spectrometry (AP-MS) (76,77), which precipitates bait proteins, along with their interacting partners, using antibodies or another purification method. Using unbiased AP-MS, a recent study discovered the association of the cardiac transcription factor TBX5 with the nucleosome remodeling deacetylase (NuRD) repressor complex, revealing a novel mechanism for TBX5 in cardiac development and congenital heart diseases (78). Although it is the current mainstay of interactome research, AP-MS specificity can vary. Even with careful washing, nonspecific binding occurs and contaminants can remain. Thus, identified interactors should be verified by complementary methods. New methods are currently being developed to ameliorate this challenge, including in silico filtering of interactome data (79,80) and cross-linking mass spectromety (XL-MS) (81). The latter uses reactive chemicals to physically link together proteins in apposition for confident MS identification (reviewed in (82)).

Interpretation and Validation of Proteomics Data

Several considerations are provided here regarding how to interpret proteomic results, as well as how to utilize new workflows and technologies.

Scope and sensitivity

With incomplete coverage, the old adage that the absence of evidence is not evidence of absence is especially true for proteomics experiments. The number of proteins and peptides in a shotgun experiment is an important indicator of the depth of coverage in data acquisition. Hence, if a known biomarker is not reported or reproduced in a discovery-based experiment in the cohort, a pertinent question is whether the technological platform used has sufficient sensitivity and precision to support reproducible measurements at endogenous concentration.

The predictive value of a putative biomarker depends not only on the magnitude of its difference in normal versus diseased/at-risk individuals in the population, but also on the effect size, defined as the number of standard deviations between 2 measured means. Hence, for the purpose of protein biomarker discovery, the coefficient of variation (CV) of measurement is essential for determining whether there is sufficient power to distinguish significant differences between healthy and disease samples. Beyond the precision characteristics of the technological platforms, sample processing, as well as biological variations in human cohorts (both intrasubject and intersubject), may also impart significant variance in protein measurements. Thus, individual putative protein markers often require lengthy verification and validation pipelines that involve large cohorts (83). Proteomics assays have reported typical CVs in the range of <5% for intralaboratory measurements (34,84) to <20% for interlaboratory measurements (85). Other measurements of analytical performance that can have a drastic impact on the likelihood of payoff in downstream verification and validation processes include the limit of detection (LOD), limit of quantification (LOQ), and linearity of measurement, which are covered at length elsewhere (83).

False Discovery Rate

As in other large-scale approaches, both the number of identified proteins and the significance of quantitative comparisons in a quantitative proteomics experiment may become inflated if controls for false discovery rates (FDRs) are not adequate. In protein identification, the FDR of a database search provides an estimate for the proportion of incorrect protein identification in the result list. An FDR of 1% is commonly deemed acceptable, meaning that in a profiling experiment claiming to identify 5,000 proteins, up to 50 may be misidentifications. FDR is sometimes calculated with the aid of decoy databases, which contain scrambled or reversed protein sequences that are compared to experimental data, along with real protein sequences. The number of identified decoy sequences provides an estimate of the number of false discoveries among regular database sequence matches. For quantitative proteomics comparisons between samples, as many hypotheses as the number of proteins compared are being tested per sample pair. Multiple-testing corrections (e.g., Bonferroni corrections) are used to control for type I errors (incorrectly rejecting the null hypothesis) that arise from making multiple comparisons (1 per protein per sample pair). However, it must be noted that even infinitesimal p values are no indication that results are biomedically important (86), and do not obviate the need to validate with follow-up studies, including in vitro experimentation or in cohorts.

Comparison with Existing Techniques

In addition to technical considerations, criteria to nominate a differentially expressed protein as a candidate biomarker can vary. Currently, no gold standard exists for what constitutes a good candidate (87). Decision justifications in published reports range from formal statistical inferences of individual proteins (88), panels of proteins after feature selection (89), to additional criteria, such as cutoffs in the raw magnitude of fold-change (e.g., >5-fold changes) in at least n number of patients (90), to more custom modeling that adjust for covariates in risk assessments (31,35). In determining whether the presented candidates are worthy of prioritized verification studies, it stands to reason that the results of discovery experiments should be carefully scrutinized for both nominal significance (e.g., p values), and for the magnitude and variance of biological changes (effect size and confidence interval).

Ultimately, a candidate biomarker or protein with potential biomedical significance needs to be validated by orthogonal approaches in independent cohorts or models. In silico methods may be used to triage a list of differentially expressed proteins and prioritize candidates for validation, for example, by determining their tissue expression specificity and half-lives. Promising biomarkers may be further credentialed with targeted discovery approaches that quantify the status of the candidates in larger numbers of subjects and conditions. For instance, protein profiles may be compared in paired controls before and after unrelated procedures, or in peripheral blood versus proximal tissue fluids, such as coronary sinus effluents, to filter out spurious associations. An in-depth discourse on biomarker verification and validation can be found elsewhere (83).

Outlook

With continued technological progress, new interactions of proteomics with genetics and genomics are now becoming possible. Proteomics studies applied to large cohorts can be integrated with genetic polymorphisms to identify protein quantitative trait loci (pQTL) that control protein abundance, and perhaps thereby control phenotypic traits. Proteogenomics studies, combining RNA-seq and proteomics data from common datasets, have been undertaken in cancer research to identify the protein coding consequences of cancer variants and mutations (91), whereas a recent multiomics study in 80 mouse genetic backgrounds demonstrated that multiple types of omics data can complement each other to characterize mitochondrial functional regulation (92). Further integration of proteomics and genomics methods will allow examination of nonstandard gene products, including short open reading frames, unannotated transcripts, and alternative splicing isoform proteins. By greatly expanding the depth of the proteome that may be experimentally accessed, the development of isoform proteomics, in particular, will have important implications for cardiovascular research. These isoforms may present with differential expression levels or patterns, localizations, interactions, and PTMs in different cell types and during disease progression, including in ischemic cardiomyopathy (93) and hypertrophy (94). Although RNA-seq has discovered many alternative isoforms at the transcript level, the majority of isoform transcripts have not been validated at the protein level. Many alternatively spliced transcripts may lead to frame-shifts and be targeted by nonsense-mediated decay. The functional significance of many transcript isoforms may soon be addressed by proteomics using custom sequence databases translated from RNA-seq transcripts.

Broader applications will also be facilitated by continued improvements in throughput and coverage. For example, newly-developed neutron encoding (NeuCode) techniques will allow over 30 samples to be combined for MS analysis (95), leading to a 5-fold increase in throughput for investigating large cohorts. With new affinity proteomics platforms, such as protein and aptamer arrays, we have already seen further development of preselected cardiovascular-specific protein panels (33), which could allow the disease status of the likeliest candidates to be screened across large sample pools. In summary, with core proteomics technologies having now matured to provide high protein and sample coverage, new applications are increasingly available to interrogate the dynamic parameters that connect gene expression to physiological functions. We envision that proteomics-based discovery-driven platforms will continue to support molecular phenotyping of individuals in the era of precision medicine, a9nd will provide ample opportunities for innovative breakthroughs in cardiovascular medicine for years to come.

ABBREVIATIONS AND ACRONYMS

FDR

false discovery rate

MS

mass spectrometry

PTM

post-translational modification