Application of Proteomics Profiling for Biomarker Discovery in Hypertrophic Cardiomyopathy

Abstract

High-throughput proteomics profiling has never been applied to discover biomarkers in patients with hypertrophic cardiomyopathy (HCM). The objective was to identify plasma protein biomarkers that can distinguish HCM from controls. We performed a case-control study of patients with HCM (n = 15) and controls (n = 22). We carried out plasma proteomics profiling of 1129 proteins using the SOMAscan assay. We used the sparse partial least squares discriminant analysis to identify 50 most discriminant proteins. We also determined the area under the curve (AUC) of the receiver operating characteristic curve using the Monte Carlo cross validation with balanced subsampling. The average AUC was 0.94 (95% confidence interval, 0.82–1.00) and the discriminative accuracy was 89%. In HCM, 13 out of the 50 proteins correlated with troponin I and 12 with New York Heart Association class. Proteomics profiling can be used to elucidate protein biomarkers that distinguish HCM from controls.

Introduction

Hypertrophic cardiomyopathy (HCM) is one of the most common genetic cardiac diseases, affecting approximately 600,000 individuals (1 in 500) in the USA [1]. HCM is caused by mutations in the genes coding for proteins constructing myocardial structures including the contractile apparatus. It is often challenging, even for experts, to distinguish HCM from other conditions that cause similar morphological changes in the heart. Such conditions include hypertensive heart disease, athlete’s heart, and HCM phenocopies—e.g., cardiac amyloidosis, sarcoidosis, and Fabry disease [1]. Physicians sometimes have to use advanced diagnostic methods in such cases to make or exclude a diagnosis of HCM. These include genetic testing, cardiac magnetic resonance imaging, and if athlete’s heart is suspected, deconditioning. However, each modality has its own limitations. For example, genetic testing is useful in ruling in HCM if positive; however, the positive rate is generally 50–70% and drops to ~ 30% in late-onset sporadic cases [2]. Magnetic resonance imaging is time and resource-intensive, and sometimes contraindicated (e.g., patients with pacemaker, implantable cardioverter-defibrillator, or claustrophobia). Deconditioning forces the athlete to miss practice sessions and games for a long period of time, which may have a negative impact on his or her athletic career. Development of a set of novel HCM biomarkers that is quick, low-cost, and easily available (e.g., blood) is warranted.

Proteomics profiling with the SOMAscan assay—an advanced unbiased approach that measures concentrations of thousands of proteins using a small amount of sample (~ 50 μl)—has been successfully applied to the discovery of biomarkers in non-cardiac conditions such as Duchenne muscular dystrophy [3], Alzheimer’s disease [4], influenza infection [5], lung and prostate cancer [6, 7], tuberculosis [8], and arthritis [9]. These studies identified biomarkers of disease status and provided important insights into the pathogenesis of disease. In the field of cardiovascular disease, proteomics profiling has been applied to identify novel biomarkers of cardiometabolic risk in the Framingham Heart Study [10]. Furthermore, there is considerable evidence in HCM that increased expression of profibrotic molecules in the heart caused by sarcomeric gene mutation can be detected in plasma [11, 12]. However, to date, no studies have applied proteomics profiling to identify proteins that have high discriminative ability and correlate with disease severity in HCM. Therefore, the objective of the present study was, using plasma samples of patients with and without HCM, to discover proteins that (1) distinguish HCM from controls and (2) correlate with known indicators of advanced disease (e.g., troponin, New York Heart Association [NYHA] class) in the HCM population.

Methods

Study Design and Patient Population

In this case-control study, the cases consisted of patients with HCM followed in the HCM Program of Massachusetts General Hospital (Boston, MA). The control subjects were patients without HCM who were followed in the general cardiology clinic of the same hospital. The diagnosis of HCM was established on the basis of echocardiographic evidence of left ventricular (LV) hypertrophy (max LV wall thickness of ≥15 mm) that is out of proportion to the degree of systemic loading conditions, a nondilated LV, and no other disease capable of producing similar findings (i.e., HCM phenocopies such as Fabry disease and cardiac amyloidosis) [13, 14]. If HCM phenocopy was suspected, appropriate tests—e.g., cardiac magnetic resonance imaging and genetic testing—were performed [13, 14]. The Partners Institutional Review Board approved the study protocol, and all participants provided written informed consent.

Blood Sample Processing and Proteomics Profiling

Venous blood specimens were drawn at the time of outpatient clinic visit. Samples were collected in K₂EDTA-treated tubes and centrifuged for 10 min at 3100 rpm. The supernatant plasma was aliquot and immediately frozen at − 80 °C.

Proteomics profiling was performed using the SOMAscan assay (SOMALogic, Inc., Boulder, CO) [15–17]. This is a recently developed tool for proteomics profiling that is highly multiplexed, sensitive, quantitative, and reproducible [15–17]. It can measure concentrations of 1129 proteins in 50 μl of plasma [15–17]. The assay can quantify proteins spanning over 8 logs in abundance—i.e., from femtomolar to micromolar—with high reproducibility (4.6 median % coefficient of variation) [15–17]. The performance of the assay is generally comparable to that of sandwich enzyme-linked immunosorbent assay [15–17]. To achieve a broad dynamic range, 50 μl of plasma sample was serially diluted to 40%, 1%, and 0.005% of the original concentration. Diluted sample was introduced to the bead-immobilized SOMAmers and equilibrated for 3.5 h at 28 °C, 800 rpm to allow for maximal SOMAmer-protein binding. SOMAmer-bound proteins were then tagged with biotin. The SOMAmer-protein complexes were released from the beads by a photocleavage process (12 min under a 360-nm ultraviolet light). Non-specific SOMAmer-protein pairs quickly dissociate in this process [15–17]. All three dilutions per sample were then recombined. Next, the biotinylated SOMAmer-protein complexes were captured on streptavidin-coated beads. Following further washing steps, SOMAmers were released from the protein targets and collected. The number of each SOMAmer reflects the initial sample protein concentration [15–17]. Samples were loaded onto microarray slides and hybridized for 19 h at 55 °C, 20 rpm. The nucleotides were quantitated using an oligo-array plate reader (Agilent Technologies, Santa Clara, CA) [15–17]. Details of the SOMAscan assay have been published elsewhere [15–17].

Univariable Analysis

Continuous values were represented as mean ± standard deviation if normally distributed, and as median (interquartile range) if not normally distributed. To compare patient characteristics of cases and controls, the unpaired Student’s t test assuming unequal variances was used for continuous variables, and the chi-square test or the Fisher’s exact test for categorical variables. The Mann Whitney-Wilcoxon test was performed to determine the statistical significance of association between concentration of each protein and disease status. Fold changes were calculated by dividing the median in the case group by the median in the control group.

Identification of Candidate Proteins to Discriminate HCM Cases from Controls

The following steps were undertaken to identify a set of 50 candidate proteins with the highest potential to discriminate cases from controls, using the sparse partial least squares discriminant analysis (sPLS-DA). The sPLS-DA is a method of supervised machine learning to examine the discriminative capacity of multi-dimensional data and the relative importance of each feature (i.e., protein in the present study) [18, 19]. First, sample-wise normalization was performed using the median of all protein concentrations in each sample [19]. Second, concentration of each protein was transformed with generalized log-transformation and subsequently normalized with auto-scaling (i.e., mean-centered and divided by the standard deviation of concentration of each protein) [19]. Third, the sPLS-DA model with 2 components and 100 proteins per component was fit to determine the discriminative ability of the data set. The sparse model was chosen instead of the full model using all proteins (i.e., PLS-DA) because proteomics data are known to be sparse (i.e., a small fraction of analyzed proteins contributes heavily to the discrimination of 2 groups) and collinear (i.e., multiple proteins correlate with each other) [18]. Fourth, importance of each protein was determined by the absolute value of loading (i.e., contribution) for the first component. A set of 50 proteins with the highest importance in the sPLS-DA model were used in the subsequent pathway and network analyses.

As a sensitivity analysis, the PLS-DA model (i.e., the full model) was applied to identify the 50 most discriminant proteins to discriminate the 2 groups. The PLS-DA model was repeated using different number of components to reveal the best number of components that achieves the highest Q² value—i.e., goodness of prediction [19]. In this sensitivity analysis using the full model, importance of each protein was determined by the average of the variable importance in projection (VIP) score across all components. VIP score is a weighted sum of squares of the PLS loadings taking into account the amount of explained Y-variation in each dimension and is calculated for each component [19].

Multivariable Receiver Operating Characteristic Curve Analysis

Receiver operating characteristic (ROC) curves were generated using the Monte Carlo cross validation (MCCV) with balanced subsampling. In each MCCV, randomly selected 2/3 of the samples was used to examine the importance of each protein (i.e., the training set), and the remaining 1/3 was used to validate the model created in the first step (i.e., the test set). The 5 most discriminant proteins were used to build the biomarker discrimination models. These steps were repeated multiple times to calculate the average of the area under the curve (AUC) and its 95% confidence interval, and to determine which proteins were most frequently selected in the 5-protein discrimination model. Discriminative accuracy was calculated by dividing the number of accurately classified samples by the total number of samples. In addition to the 5-protein discrimination model, 10-, 25-, 50-, and 100-protein models were developed to compare the discriminative accuracy. As a sensitivity analysis, a discrimination model was constructed using data of male patients (n = 27) and the discriminative accuracy was validated in female patients (n = 10).

Pathway and Network Analyses

Pathway analysis was performed to identify the canonical pathways that are differentially regulated between the HCM and control groups. Associations between the 50 most discriminant proteins identified in the previous step and canonical pathways in the Kyoto Encyclopedia of Genes and Genomes database were determined using the ratio of the number of proteins that map to a canonical pathway divided by the total number of proteins that map to the pathway [20]. Pathways with false discovery rate (FDR) < 0.02 and at least 3 associated proteins were declared as positive (i.e., differentially regulated) pathways. The FDR represented by the pathway analysis further reduces the risk of false-positive findings from the original Mann-Whitney-Wilcoxon tests as pathway components are inter-related rather than independent findings [21]. Network analysis was performed to reveal interconnections between the 50 most discriminant proteins [20].

Correlation with Known Cardiac Biomarkers and Clinical Parameters of Disease Severity

In the 50 most discriminant proteins, the concentrations were correlated with those of known cardiac biomarkers (i.e., troponin I, brain natriuretic peptide [BNP]), interventricular septal thickness, NYHA functional class, and left atrial diameter using the Spearman rank-order correlation test. Pathway and network analyses were performed using proteins that significantly correlated with either troponin I or NYHA functional class.

Metaboanalyst 4.0 (the University of Alberta, Canada) was used to perform the sPLS-DA and PLS-DA [22]. STRING Version 10.5 (String Consortium, Europe) was used for the pathway and network analyses [20]. Stata Statistical Software: Release 12 (StataCorp LP, College Station, TX) was used for the remaining analyses.

Results

Overall, 15 patients with clinically overt HCM and 22 controls without HCM were enrolled. Table 1 summarizes characteristics of study participants. Mean age was 62 years and 27% were female. HCM cases had a significantly higher NYHA class. Patients with HCM were more likely to be on β-blocker while control subjects were more likely to be taking angiotensin-converting enzyme inhibitor and/or angiotensin II receptor blocker. HCM cases had thicker interventricular septum, smaller LV end-diastolic diameter, higher degree of mitral regurgitation, and higher prevalence of systolic anterior motion of mitral valve leaflet. Details of cardiovascular conditions in the control group are displayed in Supplemental Table 1.

Table 1.

Baseline clinical characteristics of the study participants

Characteristicsa	HCM	Control	P value
	n = 15	n = 22
Demographics
Age (year)	63 ± 11	61 ± 11	0.65
Female	5 (33)	5 (23)	0.71
NYHA class ≥ 3	13 (87)	0	< 0.001
Race/ethnicity
Caucasian	10 (67)	19 (86)	0.14
African-American	0	1 (5)
Undefined	5 (33)	2 (10)
Medical history
Prior AF	3 (20)	3 (14)	0.67
Prior VT/VF	1 (7)	0	0.41
Prior non-sustained VT	2 (13)	3 (14)	>0.99
Prior syncope	3 (20)	1 (5)	0.28
Family history of sudden cardiac death	5 (33)	1 (5)	0.03
Family history of HCM	3 (20)	2 (10)	0.38
Medications
β-blocker use	14 (93)	5 (23)	< 0.001
Calcium channel blocker use	8 (53)	8 (36)	0.34
ACE inhibitor or ARB use	2 (13)	11 (50)	0.04
Diuretic use	4 (27)	4 (18)	0.69
Loop diuretic	4 (27)	1 (5)	0.14
Thiazide	0	3 (14)	0.26
Potassium sparing diuretic	2 (13)	0	0.16
Disopyramide	3 (20)	0	0.06
Amiodarone	1 (7)	0	0.41
Physiological measurements
Left atrial size (mm)	43 ± 7	39 ± 8	0.16
Systolic blood pressure (mmHg)	129 ± 15	131 ± 12	0.69
Diastolic blood pressure (mmHg)	72 ± 14	76 ± 8	0.38
Interventricular septum thickness (mm)	18 ± 3	12 ± 2	< 0.001
Posterior wall thickness (mm)	12 ± 1	11 ± 2	0.10
Left ventricular outflow tract gradient (mmHg) at rest	68 [22–90]	–	–
Left ventricular outflow tract gradient (mmHg) with Valsalva maneuver	82 [27–103]	–	–
Left ventricular ejection fraction (%)	67 ± 9	66 ± 8	0.35
Left ventricular end-diastolic diameter (mm)	42 ± 5	47 ± 4	0.01
Left ventricular end-systolic diameter (mm)	27 ± 6	30± 4	0.17
Systolic anterior motion of mitral valve leaflet	13 (87)	1 (5)	< 0.001
Degree of mitral regurgitationb	2.5 [2–3]	1 [1–1.5]	< 0.001

AF atrial fibrillation, ACE angiotensin-converting enzyme, ARB angiotensin II receptor blocker, HCM hypertrophic cardiomyopathy, NYHA New York Heart Association, VT/VF ventricular tachycardia or ventricular fibrillation

^aData are expressed as number (percentage), mean ± standard deviation, or median [interquartile range]

^bDegree of mitral regurgitation was converted to numerical values according to the following rule: trace = 1, trace to mild = 1.5, mild = 2, mild to moderate = 2.5, moderate = 3, moderate to severe = 3.5, severe = 4

Discrimination of the disease status using the sPLS-DA model showed that the overall proteomic profile was distinctly different between the cases and controls (Fig. 1). Figure 2 displays the 50 most discriminant proteins that distinguish HCM cases from controls. The sensitivity analysis using the PLS-DA model revealed a substantial overlap between the 2 lists of 50 most discriminant proteins—94% (47 out of 50) of proteins were overlapping and selected as discriminant with both sPLS-DA and PLS-DA methods (Supplemental Table 2). The PLS-DA classification model using different number of components confirmed that the two-component model used in the sPLS-DA achieves the highest Q² value and that addition of the third component and thereafter did not significantly increase Q² (Supplemental Figure 1).

Fig. 1.

Two-dimensional score plot using the sparse partial least squares discriminant analysis model. Green circles represent the proteomic profile of HCM cases, and red circles are that of controls. HCM hypertrophic cardiomyopathy, sPLS-DA sparse partial least squares discriminant analysis

Fig. 2.

The 50 most discriminant proteins to distinguish hypertrophic cardiomyopathy cases from controls, identified with the sparse partial least squares discriminant analysis. Red box on the right indicates that the protein concentration was increased in HCM, and green box means that was decreased in HCM. P values were computed with the Mann Whitney-Wilcoxon test. Fold change was calculated by dividing the median in case by the median in control. The blue bars indicate importance of each protein to discriminate HCM cases from controls, which was determined by the contribution to the discriminative model. HCM hypertrophic cardiomyopathy, MAPK mitogen-activated protein kinase, NEGF neurite growth-promoting factor

The multivariable ROC curve analysis revealed that the average AUC was 0.94 (95% confidence interval, 0.82–1.00, Fig. 3). The discriminative accuracy of the 5-protein model was 89%, which was higher than that of the other models (Fig. 4). Proteins that were most frequently selected in the 5-protein model are displayed in Supplemental Figure 2.

Fig. 3.

Area under the receiver operating characteristic curve using 5-protein discrimination model. AUC area under the curve, CI confidence interval

Fig. 4.

Discriminative accuracy using different number of proteins in the discrimination model. Numbers on the x-axis represents the number of proteins used in each model

The sensitivity analysis using the data of male patients for derivation and those of female patients for validation showed that the discriminative accuracy was 90%—i.e., 9 out of 10 female patients were classified correctly (Supplemental Figure 3).

Pathway analysis using the 50 most discriminant proteins identified through the sPLS-DA displayed that Ras pathway (FDR = 1 × 10⁻⁵) and multiple upstream signaling pathways related to cell proliferation and angiogenesis were significantly upregulated in HCM cases compared to controls (Supplemental Table 3). Network analysis demonstrated that there were significantly larger number of edges (98 edges) found in the network of the 50 most discriminant proteins than expected (48 edges) with a protein-protein interaction enrichment P value of 1.2 × 10⁻¹⁰. Proteins that map to Ras-MAPK pathway were at the hub of the interaction network (Supplemental Figure 4A).

Among the 50 most discriminant proteins to distinguish the disease status, 13 had significant correlation with troponin I, none with BNP, and 2 with interventricular septal thickness (Table 2). With regard to the directionality of the correlation, 12 out of the 13 proteins with significant correlation with troponin I were concordant (i.e., proteins that were increased in HCM had positive correlation and those decreased in HCM had negative correlation). Concentrations of 12 out of the 50 proteins significantly correlated with NYHA class, and all of these 12 proteins had concordant directionality (Table 3). The analysis of correlation between the 50 proteins and left atrial diameter revealed significant correlation with 9 proteins in HCM, in which 3 were concordant (Table 4). In the pathway analysis of the 23 proteins that had significant correlation with either troponin I or NYHA class, Ras pathway was again noted to be significantly upregulated (FDR = 5 × 10⁻⁵) along with its upstream and downstream pathways (Supplemental Table 4). Network analysis with these proteins showed significantly larger number of edges (51 edges) than expected (16 edges). Protein-protein interaction enrichment P value was1.9 × 10⁻¹². These proteins were centered around members of Ras-MAPK pathway in the network (Supplemental Figure 4B).

Table 2.

Proteins with significant correlation with troponin I among the 50 most discriminant proteins

Protein namea	P value	Spearman’s ρ	Increased/decreased in HCM
cGMP-specific 3′,5′-cyclic phosphodiesterase	0.006	+ 0.44	Increased
MAPK-3	0.02	+ 0.39	Increased
Caspase-3	0.047	+ 0.33	Increased
Protein kinase Cα	0.047	+ 0.33	Increased
Complement 4b	0.0002	− 0.57	Decreased
SHC-transforming protein 1	0.02	+ 0.38	Increased
Hepatocyte growth factor	0.001	+ 0.50	Increased
Lymphocyte antigen 86	0.0003	− 0.56	Decreased
Sphingosine kinase 1	0.04	+ 0.33	Increased
Calpastatin	0.03	− 0.35	Increased
Apolipoprotein A-I	0.04	− 0.35	Decreased
Disintegrin and metalloproteinase domain-containing protein 9	0.01	+ 0.41	Increased
Cathepsin H	0.01	+ 0.42	Increased

MAPK mitogen-activated protein kinase

^aThe order of proteins is the same as Fig. 2

Table 3.

Proteins that significantly correlate with the New York Heart Association among the 50 most discriminant proteins

Protein namea	P value	Spearman’s ρ	Increased/decreased in HCM
Glycogen synthase kinase-3 α/β	0.03	+ 0.55	Increased
MAPK-3	0.01	+ 0.61	Increased
Caspase-3	0.02	+ 0.59	Increased
Proto-oncogene tyrosine-protein kinase Src	0.04	+ 0.52	Increased
Protein kinase Cβ	0.05	+ 0.51	Increased
Pyruvate kinase PKM	0.02	+ 0.59	Increased
Tyrosine-protein kinase LYN	0.01	+ 0.65	Increased
Sorting nexin-4	0.04	+ 0.54	Increased
Tyrosine-protein kinase BTK	0.02	+ 0.59	Increased
Ribosome maturation protein SBDS	0.02	+ 0.61	Increased
β-Adrenergic receptor kinase 1	0.01	+ 0.65	Increased
Tyrosine-protein kinase CSK	0.02	+ 0.59	Increased

MAPK mitogen-activated protein kinase

^aThe order of proteins is the same as Fig. 2

Table 4.

Proteins that significantly correlate with left atrial diameter among the 50 most discriminant proteins

Protein namea	P value	Spearman’s ρ	Increased/decreased in HCM
Midkine/NEGF2	0.007	− 0.66	Increased
Secreted frizzled-related protein 1	0.02	− 0.60	Increased
Hepatocyte growth factor	0.002	− 0.72	Increased
Lymphocyte antigen 86	0.05	+ 0.52	Decreased
Carbonic anhydrase 13	0.04	+ 0.54	Increased
Fibroblast growth factor 18	0.04	− 0.53	Increased
Complement C3d fragment	0.03	− 0.56	Decreased
MAPK-14	0.03	+ 0.56	Increased
40S ribosomal protein SA	0.04	+ 0.53	Decreased

MAPK mitogen-activated protein kinase, NEGF neurite growth-promoting factor

^aThe order of proteins is the same as Fig. 2

Discussion

In this case-control study that comprehensively profiled the plasma proteome, we discovered proteins with high discriminative accuracy to distinguish HCM cases from controls. Furthermore, multiple proteins significantly correlated with the conventional cardiac biomarker and clinical parameters of disease severity. Our study is the first investigation that applies proteomics profiling to reveal plasma protein biomarkers of HCM and to examine their correlations with known markers of advanced disease. These findings provide an important step towards developing a panel of plasma biomarkers to help physicians accurately diagnose HCM.

Thus far, two studies have applied proteomics technique to HCM in humans [23, 24]. The first was a cross-sectional study involving 20 male patients with HCM using two-dimensional gel-based approach. This study showed a negative correlation between haptoglobin concentration and LV outflow tract gradient [23]. However, significance of this finding is uncertain because it could be due to subclinical hemolysis at the site of LV outflow tract obstruction [25], as a reduction in haptoglo bin is a known marker of hemolysis. Lack of inclusion of female patients also limits the generalizability of this study. The second study involved 47 patients with HCM and 46 healthy controls, and reported 8 of 128 proteins tested were differentially expressed [24], including laminin subunit α-2 which is known to be involved in fibrosis. However, these findings could be driven largely by type I error because of lack of correction for multiple testing. By contrast, our study used a conservative cutoff FDR value of 0.02, which would yield 1 false-positive declaration out of 50. To date, no study has applied comprehensive proteomics profiling to identify plasma biomarkers in HCM. In this context, the present study adds to the body of knowledge by revealing proteins that can distinguish HCM from controls and correlate with known markers of advanced HCM.

In our study, the discrimination model based on the proteomics profile performed well with a high AUC. The use of MCCV, where 2/3 of the samples was used for discrimination model derivation and 1/3 for validation, addresses the issue of model overfitting. The use of the 5-protein model further mitigates potential overfitting, and the discriminative accuracy of this model was at least as good as the other models including up to 100 proteins. Additionally, we have performed a sensitivity analysis using the data from male patients for derivation and those from female patients for validation, again revealing high discriminative accuracy. The findings in the present study elucidate the potential utility of proteomics profiling to discover plasma HCM biomarkers and suggest that a plasma biomarker panel consisting of a small number of proteins offers high discriminative accuracy. Development of such a novel diagnostic panel that is quick to perform, low-priced, and easy to obtain would aid in making a diagnosis of HCM when combined with conventional clinical features.

If proteins with different concentrations in HCM can also function as markers of advanced HCM, the concentrations of these proteins would correlate with known markers of disease severity within the HCM population. Indeed, in our study, multiple discriminant proteins significantly correlated with the objective and subjective markers of disease severity—i.e., troponin I and NYHA class. Our findings suggest that these proteins are not only helpful in discriminating HCM cases from controls but also serve as markers of severe form of HCM.

Interestingly, a number of upstream and downstream pathways associated with upregulation of Ras-MAPK pathway were found to be upregulated in HCM in our analysis using the 50 most discriminative proteins. The results were similar in the pathway analysis using the 23 proteins that correlated with either of the 2 markers of disease severity, and this list had a substantial overlap with the list of proteins that were most frequently selected in the 5-protein discrimination model. These lists had several sets of proteins related to Ras MAPK signaling pathway in common. The first set consists of members of Ras-MAPK signaling cascade that are known to cause cardiac hypertrophy and positive inotropic effects in response to β₁ adrenergic stimulation (e.g., protein kinase Cα and protein kinase Cβ) and downstream proteins activated by protein kinase C (e.g., glycogen synthase kinase-3 α/β) [26, 27]. The second set of proteins, pleiotrophin/NEGF1 and midkine/NEGF2, belongs to the same protein family and exerts pleiotropic properties (i.e., capable of inducing cell proliferation) through MAPK signaling [28, 29]. The third set includes SHC-transforming protein 1 and proto-oncogene tyrosine-protein kinase Src, both of which are known to activate Ras-MAPK signaling cascade [30, 31]. Additionally, caspase 3 is a member of MAPK pathway [32], and the expression of pyruvate kinase PKM is increased through MAPK-dependent pathway [33]. Our findings suggest that Ras-MAPK and associated pathways not only contribute to the pathogenesis of HCM but also play an important role in the disease progression to severe form of HCM.

Although no prior study has shown the involvement of Ras-MAPK pathway in the disease development of HCM in humans, it has been known that mice with a gain-of-function mutation in the Raf1 gene (a key molecule in Ras-MAPK signaling pathway) display HCM [34, 35]. Among humans with such a mutation in Raf1, 95% were found to have HCM-like phenotype [34–36]. Further, clinical syndromes caused by mutation in the genes encoding proteins of Ras MAPK pathway—i.e., the “RASopathies” [37]—can develop LV hypertrophy resembling HCM. For example, the prevalence of HCM-like cardiac phenotype was significantly higher in multiple RASopathies such as LEOPARD (73%), cardio facio-cutaneous (33%), and Noonan (20%) syndromes than the general population (~ 0.2%) [37]. Together with these prior animal and human investigations, our observations would generate a hypothesis that Ras-MAPK may be a common molecular pathway shared between HCM and the RASopathies that exhibit HCM-like LV hypertrophy.

Nevertheless, the inferences from the present study need to be interpreted with caution as these data were derived from plasma proteomics while the aforementioned pathways involve transmembrane proteins as well as intracellular cascades. Investigators have previously reported that patients with HCM can have higher plasma concentrations of proteins associated with cardiac fibrosis [11, 12], indicating that it is possible to gain information on differences in the heart at the molecular level using proteomics analysis in plasma. One potential explanation is that intracellular components and transmembrane structures are included in exosomes, at least a part of which originate from cardiomyocytes, and therefore can be detected in plasma. Another possible explanation is continuous, low-grade destruction and turnover of myocardium with release of intracellular components into the blood—this is known to be the reason why troponins are detected in plasma in patients with heart failure even without myocardial infarction [38]. Indeed, in our study, 13 out of the 50 discriminative proteins had significant correlation with troponin I, which is a larger number than what would have been observed by chance (i.e., 2–3 proteins at α = 0.05). By contrast, none of the 50 proteins correlated with BNP—a hormone released to the extracellular space. Moreover, in our study aldosterone-regulated sodium reabsorption, adrenergic signaling, and Wnt signaling pathways were upregulated in HCM, all of which are known to be activated in this population [39–41]. For instance, in a mouse model of HCM induced by cardiac-specific H-Ras-G12 V, Wnt signaling pathway was upregulated in conjunction with phosphorylation of glycogen synthase kinase-3 α/ β in the course of pathogenic myocardial hypertrophic transformation [41]. Thus, these pathways serve as “positive controls” in our study. These data provide indirect evidence to support the notion that at least a part of these proteins have originated from the myocardium and that plasma proteomics profiling reflects differential regulation of molecular pathways between the HCM and non-HCM populations.

Our study has several potential limitations. It is a single center study with patients at a tertiary care facility. Therefore, our findings may not be generalizable to patient populations with less severe HCM. The present study did not assess temporality or causality between the differentially regulated proteins and HCM pathogenesis or disease progression. The sample size was small in this study, and because of this, multivariable analysis could not be performed in the analysis of correlation with known markers of disease severity. All proteins correlated with NYHA class showed concordant directionality while some proteins that correlated with left atrial diameter were discordant. This could be because these two parameters may reflect different biological processes, as left atrial diameter can be affected by a number of factors (e.g., mitral regurgitation, left ventricular and atrial remodeling, and duration of elevated filling pressures) in addition to the degree of heart failure at the time of evaluation. Not all patients had data on genetic testing and year of diagnosis. Comparison between HCM cases and healthy individuals was not performed. Myocardium specimens were not available.

Conclusions

This study is the first investigation that demonstrates the role of proteomics profiling in HCM to discover plasma protein biomarkers that are associated with disease status and severity. Our data displayed high discriminative accuracy and exhibited correlations of these proteins with known markers of advanced HCM. Our study serves as a step towards development of HCM biomarkers that can be measured in blood, which is readily available in the real-world clinical settings. Our findings should also facilitate further research into the mechanisms linking the identified molecular pathways to the HCM pathobiology.

Abbreviations

AUC

Area under the curve

BNP

Brain natriuretic peptide

FDR

False discovery rate

HCM

Hypertrophic cardiomyopathy

LV

Left ventricular

MCCV

Monte Carlo cross validation

NYHA

New York Heart Association

sPLS-DA

Sparse partial least squares discriminant analysis

VIP

Variable importance in projection