Genetic risk scores to predict the prognosis of chronic heart failure patients in Chinese Han

Abstract Chronic heart failure (CHF) has poor prognosis and polygenic heritability, and the genetic risk score (GRS) to predict CHF outcome has not yet been researched comprehensively. In this study, we sought to establish GRS to predict the outcomes of CHF. We re‐analysed the proteomics data of failing human heart and combined them to filter the data of high‐throughput sequencing in 1000 Chinese CHF cohort. Cox hazards models were used based on single nucleotide polymorphisms (SNPs) to estimate the association of GRS with the prognosis of CHF, and to analyse the difference between individual SNPs and tertiles of genetic risk. In the cohort study, GRS encompassing eight SNPs harboured in seven genes were significantly associated with the prognosis of CHF (P = 2.19 × 10−10 after adjustment). GRS was used in stratifying individuals into significantly different CHF risk, with those in the top tertiles of GRS distribution having HR of 3.68 (95% CI: 2.40‐5.65 P = 2.47 × 10−10) compared with those in the bottom. We developed GRS and demonstrated its association with first event of heart failure endpoint. GRS might be used to stratify individuals for CHF prognostic risk and to predict the outcomes of genomic screening as a complement to conventional risk and NT‐proBNP.

cardiovascular disease have shown the use of genomic information in risk prediction.
Genetic variation in subjects with CHF may determine outcomes, 10,11 but previous studies on the prognosis of HF and related GRS remain unclear. First, CHF has broad spectrum of aetiology and heterogeneity of symptoms 12 ; however, the common characteristics of genetics and/or environment have verified a final common pathway in CHF. 13 Second, the outcomes of HF have cumulative effects for multiple risk factors interaction. In determining the prognosis of CHF, the role of monogenic variant is rare, and this will miss the superposition of minor genetic variations.
Hence, there is a considerable room for improvement to genetic risk assessment for CHF.
Here, we reported a whole exome sequencing wide GRS for CHF to provide prognosis risk evaluation. We re-analysed the mass spectrometry (MS) data for 34 non-failing and failing human left ventricular myocardium. 14 Moreover, in more than three phenotypes of failing heart, such genes will be included only if it has the same variability trend as normal contrast expression (P < 0.05), and thus were detected from whole exome sequencing data of 1000 Han Chinese CHF patients (787 idiopathic dilated cardiomyopathy and 213 ischaemic dilated cardiomyopathy). GRS for HF was constructed utilizing cox regression in HF cohort, to evaluate the stratifying prognosis and risk performance of GRS in 1000 CHF study cohort.

| Data of mass spectrometry of human myocardial tissue
Thirty four hearts samples were all diagnosed and collected by Chen from the Hospital of the University of Pennsylvania. 14 The samples were divided into normal, compensated hypertrophy (cHyp), hypertrophic cardiomyopathy preserved ejection fraction (HCMpEF), hypertrophic cardiomyopathy reduced ejection fraction (HCMrEF), dilated cardiomyopathy (DCM) and ischaemic cardiomyopathy (ICM).

| Study subjects for whole exome sequencing
The Institutional Ethics Committee of Tongji Hospital approved this study, which followed the principles of the declaration of Helsinki. Meanwhile, β-blocker taking as an adjusted factor was collected in different time points. Inclusion and exclusion criteria and the details of data processing and quality control are provided in Appendix S1.

| Whole exome sequencing and bioinformatics workflow
Genomic DNA was extracted from peripheral blood leukocytes

| Data processing and quality control
The WES data were stored with Variant Call Format (VCF). The VCFtools (https ://github.com/vcfto ols/vcftools) was used to perform data analysis, and invalid data were eliminated before establishing available data pools. Considering the repeatability of data processing, we employed appropriate quality control procedures to suit the whole exome sequencing summary statistics adapted for minor allele frequency. PLINK 16 was used to control imputation quality and Hardy-Weinberg equilibrium. Genetic principal components (PCs) 17 was used for cohort structure quality control.

| Construction of GRS
A detailed description is offered in Appendix S1. Briefly, (a) we built a GRS based on mass spectrometry of human myocardial tissue and  19 were used for assessment of GRS.

| Statistical analysis
To test the association of HF susceptibility variants with the prognosis of CHF risk factors, we used a combination of linear and logistic regression adjusting for age at first visit, gender, SNPs and conventional risk factors. These analyses were based on additive models. We considered the significant of SNP risk factor association threshold of P < 0.05.
We analysed individual SNPs and tertiles of genetic risk score, which were adjusted using conditional logistic regression of GRS for age, gender and conventional risk factors, and the significance of the difference in the receiver operating characteristic (ROC) curves was tested with C-index approach. Linkage disequilibrium (LD) was calculated by Haploview version 4.1. Data are expressed as mean ± SEM of experiments. Volcano plots (differential expression of genes for construction of GRS) were plotted by the R package 'ggplot2.' Data analyses were performed using SPSS 24.0 (SPSS, Inc) for Windows (Microsoft Corp).

| Differential gene expression analysis
In the online file (PXD008934), 3764 genes per group were available for analysis, and we filtered meaningful genes of P < .05 per group. The above conditions were for genes that were in more than three groups for HCMpEF, HCMrEF, DCM and ICM. Using Venn diagram to analyse the overlapping gene ( Figure 1A), ultimately, 319 genes were chosen (details seen in Table S1). Baseline data for these samples are listed in https ://www.nature.com/artic les/s41591-018-0046-2#Sec33 .

| Whole exome sequencing
Seventy seven thousand, two hundred and eighty seven variants of Minor Allele Frequency (MAF) > 0.05 were identified in the whole exome sequencing from 1000 CHF objects, with 45 125 LD-prune variants, which were used for the stratification of population ( Figure 1B).

| Differential gene loci and prognosis of patients with heart failure
To confirm the association between genetic alterations of common genes in the MS of heart failure tissues and the prognosis of CHF patients, 441 SNPs harboured in differentially expressed genes (Table   S2) Table 2, Table S7). The following genes AGT, SLC25A13, HRG, APOB, SOD3, SYNM and TLN2 were included in the GRS study. In addition, volcano plot was used to display the variance of candidate genes among the different types of failing hearts ( Figure 1C-F).

| Predictive effect of GRS on the prognosis of HF
For each individual, we calculated CHF-specific genetic scores using the weighted sum of the risk allele (zero, one, or two for risk alleles at each locus; Figure 2A). These scores were weighted according to the size effect reported in the endpoints studies. Genetic risk score was strongly associated with the prognosis of CHF by univariable analysis  Figure 2B, Table S4).
When the genetic risk score was divided into quartiles as previous tertiles, results were consistent and associated with the prognosis of CHF; details are provided in  Figure 2C, Table S4).

TA B L E 3
Results of univariable and multivariable cox proportional hazard analyses for cardiac events characteristic (ROC) curves. The N-terminal B-type natriuretic peptide (NT-proBNP) was built as a contrast scale, which has been widely recognized as an important prognostic indicator of heart failure. [21][22][23] NT-proBNP in GRS of tertiles and quartiles showed no differentiation with ANVOA (P = 0.895; P = 0.704; Table S8). In Cox regression of prognosis of CHF, models based on age had higher C-index (C = 0.626; 95% CI: 0.595-0.656) than any of the individual conventional risk factors, with the second-best model being GRS on assessment (C = 0.620; 95% CI: 0.589-0.650; Figure 3A Figure 3B and C, Tables S5 and S6).

| D ISCUSS I ON
Genetic counselling is recommended for CHF patients and their family members. 24 It has always been of clinical focus to foresee the risk of heart failure via genetic background characteristics. GRS was used to identify patients with Mendelian and complex disease 25 patterns known as loci for HF risk factors, such as CAD, Cardiometabolic Disease, 8,26 Blood Pressure 9 and atrial fibrillation. 27 Heart failure has similar or worse prognosis when compared to most cancers. 28 Traditional risk factors 20 have been used in risk prediction. However, obtaining GRS to predict the prognosis of HF remains challenging.
Here, we made an effort to incorporate genetic risk scores into clinical practice for determining the outcome of CHF.
In analysing the data from case-control mass spectrometry of human left ventricle tissue in this cohort study, 441 genes were is an antioxidant enzyme that catalyses the conversion of superoxide radicals into hydrogen peroxide and oxygen, which may protect the heart from oxidative stress. 33 SYNM is an intermediate filament (IF) family member, which primarily functions as mechanical stress, maintains structural, and related to smooth muscle cell cytoskeleton of the heart, with its absence causing ventricular dysfunction in mice. 34 TLN2 belonging to the talin protein family is a cytoskeletal protein, which is highly expressed in cardiac muscle when loss of talin-1 and talin-2 leads to dilated cardiomyopathy and cardiac dysfunction in mice. 35 Clinicians have a growing need for tools to assess the prognosis of heart failure accurately in the individual patient, especially to obtain credible information regarding the prognosis in the early stage and not just collecting clinical characteristics of symptoms and the results of invasive test in the acute exacerbation stage. SNPs as factors of the innate genetic background are involved in the pathophysiological processes of heart failure. This is a new attempt to build a genetics risk score consisting of SNPs to predict the outcomes of heart failure. We found that genetic risk score including eight SNPs (rs4273214, rs33958047, rs2301629, rs1042464, rs679899, rs2536512, rs3134587 and rs1320191) harboured seven genes were associated with heart failure (even after we have ac-

| Study limitations
First, the SNPs used for GRS panel construction should be enriched.
For example, more potential differential genes can be obtained after augmenting the sample size of MS control-case. Meanwhile, extensive whole exome sequencing cohort potential candidate variants will unfold with GRS. Second, there are racial differences between mass spectrometry (MS) samples and whole exome sequencing population. Another population or other ancestries are needed to verify the results. However, the potential clinical use of GRS can be predicted. Third, the expansion of the sample size of CHF population is needed, and future studies on large multiethnic cohorts will validate GRS. Fourth, other widely accepted heart failure prognostic risk scores, such as Seattle heart failure model should be included as control in subsequent studies. Fifth, we did not include the index of family history in traditional risk. The information about disease and its direct relations is often vague, which may be related to the level of medical care in the past.

| CON CLUS ION
We developed and assessed a genetic score based on eight SNPs in the current study, and we demonstrated that it was associated with the first event of heart failure endpoint. We attempted the concept of engaging genomic information to stratify individuals for CHF prognostic risk and to improve risk reclassification for participants, and we demonstrated a hypothesis to predict the outcomes of genomic screening to complement conventional risk and NT-proBNP.

ACK N OWLED G EM ENT
We Editage (www.edita ge.com) for English language editing.

CO N FLI C T O F I NTE R E S T
The authors declare no competing financial interests.

AUTH O R CO NTR I B UTI O N S
Shiyang Li and Senlin Hu developed the study concept, design and interpreted the data and drafted the manuscript. Dong Hu and Sun Yang performed the data analysis. Lei Xiao, Yanghui Chen, Huihui Li performed the research, Guanglin Cui, Dao Wen Wang supervised the design of the study and revised the manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
The authors confirm that the data supporting the findings of this study are available within the article and its supplementary materials.