Anatomically resolved transcriptome and proteome landscapes reveal disease‐relevant molecular signatures and systematic changes in heart function of end‐stage dilated cardiomyopathy

Dilated cardiomyopathy (DCM), as characterized by the left ventricular dilatation and contractile dysfunction, is one of the molecular mechanisms of which are largely unexplored. Here, we profiled the region‐resolved transcriptome and proteome of healthy and DCM human myocardial tissue and obtained the deep‐coverage dataset consisting 7,605 proteins and 19,880 transcripts in four chambers of the human heart. On the basis of the core proteome and transcriptome characters of the healthy hearts, chamber‐specific proteome alterations were further revealed in end‐stage DCM, among which extracellular matrix (ECM), mitochondrial function, and muscle contraction were the most dysregulated biological processes. Protein–protein interaction network demonstrated divergent functional networks of DCM atrium and ventricle. Additionally, a 4‐biomarker panel (CTSB, vWF, C9, and MFGE8) was established with promising diagnostic potential for the DCM. Collectively, our data provide a global proteomic basis of the chamber‐specific cardiac tissue, and establish a protein catalog that holds promise for better definition and diagnosis of DCM.


INTRODUCTION
Heart is the hardest working muscle organ which located in human chest, slightly left of center, and pumps blood filled with oxygen and nutrients through the blood vessels to the body tissues. Cardiovascular diseases are the leading cause of death globally. Dilated cardiomyopathy (DCM) is one of the highly lethal myocardium diseases defined by the presence of left ventricular dilatation and contractile dysfunction with the absence of abnormal loading conditions and severe coronary artery disease. 1 As a final common manifestation of heterogenous etiologies, DCM often results from myocarditis (mostly viral infection), exposure to drugs and toxins, and systemic endocrine disturbances, while genetic mutations account for approximately 35% cases. 2,3 The estimated prevalence of DCM is approximately 1 in 250 individuals, whereas the prevalence is higher in underdeveloped countries due to unreported or undiagnosed cases. 4,5 Molecular mechanisms and pathophysiological processes of DCM are largely unexplored. Echocardiography and electrocardiography are the first-line imaging tests in the assessment of patients with DCM. The current treatment of DCM is largely limited to the alleviation of clinical symptoms and there is no cure for this disease progression. At the end-stage of DCM, patients eventually develop heart failure and heart transplantation is the final option. 6 Patients with DCM compared with those with ischemic heart disease are typically younger, and five-year mortality rates remain high at approximately 20%. 7 Adverse outcomes highlight the need for molecular mechanism exploration and pathological process characterization.
Recent advances in whole-transcriptome shotgun sequencing (RNA-seq) and mass spectrometry-based proteomics technology allow the in-depth quantitative profiling to assess human health and disease. 8,9 van Heesch et al. applied mRNA-seq and Ribo-seq to human left ventricular cardiac tissue of 65 end-stage DCM patients and 15 non-DCM controls, identifying hundreds of previously undetected microproteins. 10 Wang et al. employed single-cell RNA sequencing to map the transcriptomic landscape of 21,422 cells-including cardiomyocytes (CMs) and non-CMs (NCMs)-from normal, failed and partially recovered adult human hearts, and clarified that CM contractility and metabolism were the most prominent aspects which were correlated with changes in the heart function. 11 Chaffin et al. identified extensive molecular alterations in failing hearts at single-cell resolution by performing single-nucleus RNA sequencing of nearly 600,000 nuclei in left ventricle samples from 11 hearts with DCM and 15 hearts with hypertrophic cardiomyopathy as well as 16 non-failing hearts, and comprehensively characterized both a final common transcriptional pathway and reduced numbers of proliferating resident cardiac macrophages in patients with cardiomyopathy. 12 Doll et al. generated a spatial and cell-type-resolved proteomic map of the healthy human heart, and established proteomic differences between heart regions. 13 Herrington et al. depicted the proteomic architecture of human coronary and aortic atherosclerosis, and further illustrated the potential clinical utility of using tissue proteomics to identify plasma biomarker panel which was highly predictive of angiographically defined coronary disease. 14 Although transcriptomics and proteomics have now enabled qualitative and quantitative analyses of a broad range of genes with an unparalleled level of accuracy and sensitivity, they are still, for the most part, assessed individually with distinct strategies to elucidate potential causal roles of the disordered genes in cardiovascular disease progression.
DCM is characterized by substantial locus, allelic, and clinical heterogeneity which necessitates investigation of individual genes and the pathogenic variant spectrum across clinically overlapping diseases. Genetic studies have identified several molecular and cellular alterations from patients with DCM, mostly contributed to cardiac muscle contractility and relaxation abnormalities. [15][16][17] Due to genetically heterogeneous, mutations in genes encoding cytoskeletal, sarcomeric, mitochondrial, desmosomal, nuclear membrane, and RNA-binding proteins have all been linked to the DCM. Indeed, familial DCM can be caused by mutations of multiple genes, including TTN, 18,19 LMNA, 20 Hierarchical clustering revealed good separation of donor samples from ICM or DCM, and pathological pathways including fibrotic remodeling, oxidative stress, and local cardiac thyroid metabolism were perturbed in both heart failure myocardium. 38 Machine-learning approaches have further applied to define and validate novel prognostically relevant DCM subphenotypes via multiparametric data. Three mechanistically distinct DCM subtypes were identified using a breadth of clinical, imaging, proteomic, and genetic data, which could improve patient stratification and facilitate targeted interventions. 39 By and large, transcript and proteomic analyses in region-resolved cardiac tissue from patients with DCM remain scant, especially for the quantitative proteomics. In this study, we attempted to identify altered molecules and aberrant protein signatures by performing anatomically resolved transcriptional and proteomic profiles, generated diseaserelated networks and elucidated the potential causal roles of those disordered genes in DCM progression. Our findings unveiled a non-invasive diagnostic panel of four proteins significantly altered in serum samples from DCM patients.

Patients and clinical specimens
Human cardiac specimens and serum samples were obtained from Zhongshan Hospital, Fudan University, with the approval of the Research Ethics Committee from this hospital (Identifier No. B2014-084). Written informed consent was obtained from all study subjects. For heart donors, informed consent was obtained from the next of kin. Ten end-stage DCM hearts who suffered heart transplantation from November 2015 to January 2016 were reviewed, and the cause of ten enrolled DCM hearts is idiopathic. Ten healthy hearts were collected from trauma individuals, which did not present any heart injury or signs of cardiac malfunction. As for serum samples, a cohort of DCM patients (n = 53) and healthy volunteers (n = 35) were enrolled. Detailed descriptions of clinical specimens can be found in the Supplemental Experimental Procedures.

Cardiac tissue processing
Based on clinically diagnosed disease state, four heart cavities were collected from ten end-stage DCM patients. The four heart cavities included left atrium (LA), left ventricle (LV), right atrium (RA), and right ventricle (RV). Accordingly, these four healthy heart cavities from ten adult individuals were collected after trauma rescue failure during an autopsy. Frozen human cardiac tissues, 20 mg/sample, were washed with cold Phosphate buffered saline (PBS) and grinded using a pre-cooled mortar and pestle under the continuous addition of liquid nitrogen. Then, powdered tissues were collected for total RNA isolation and RNA-seq.
Minced cardiac tissues, ideally 100 mg/sample, were washed three times with cold PBS and powdered in liquid nitrogen. Label-free quantification (LFQ) experiments or iTRAQ-labeled quantitative proteomic analyses were performed.

Transcriptome sequencing
RNA sequencing experiments were performed on four heart cavities of cardiomyopathy patients with end-stage DCM (n = 3) and non-DCM controls (n = 3). The RNAseq procedure on tissue samples was carried out as follows: total RNA was isolated using TRIzol reagent (Invitrogen, USA) and quantitated via Qubit 2.0 (Life technologies, USA). The quality and integrity of RNA samples were controlled by a BioAnalyzer 2100 system (Agilent, USA). For library preparation, 1 μg total RNA were captured by Dynabeads Oligo(dT) 25 (Invitrogen), sheared to fragments of ∼200 bp, and reverse transcription reactions were performed by SuperScript III cDNA Synthesis Kit (Invitrogen). cDNA was end-repaired, polyA-tailed, ligated to adapters, and amplified by standard PCR. RNA-seq library preparation was performed by TruSeq RNA Sample Prep Kit (Illumina, USA). The sequencing libraries were qualified by the Bioanalyzer 2100 (Agilent), then sequenced on Illumina HiSeq X System via 2×150 bp paired-end sequencing, which were controlled by HiSeq Control Software (HCS). Human heart mRNA-seq libraries were sequenced to an average depth of 63M (min. 36M, max. 82M) raw reads. RNA-seq reads were mapped to hg19 with TopHat (version 2.0.14). Fragments per kilobase of exon model per million mapped reads (FPKM) values were calculated per gene with Cufflinks (version 2.2.1), using the GTF file downloaded from UCSC Genome Browser. Differential expression analysis between DCM and healthy control samples using Bioconductor software with limma package (version 3.46.0). 40 False discovery rate (FDR) ≤ 0.05 and a fold change (FC) ≤ 0.5 or ≥ 2 were used as the genome-wide significance thresholds.

Protein extraction and digestion
Powered cardiac tissue were lysed in lysis buffer (2% SDS, 20 mM HEPES, pH = 8.0) containing protease inhibitors (Thermo Fisher Scientific, USA) and 2 μl Benzonase (Sigma-Aldrich, USA), then sonicated for 3 min (5 s on and 5 s off, amplitude 25%). The supernatant of the lysate was collected as whole-tissue extract. Each aliquot of 100 μg cardiac proteins was reduced with 10 mM TCEP for 30 min at 56 • C, followed by alkylation with 20 mM iodoacetamide for 30 min in the dark. Proteins were precipitated overnight at -20 • C using analytical grade acetone (Sigma-Aldrich). The recovered proteins were washed and resuspended with 100 μl 100 mM TEAB solution. Sequencing-grade rLys-C (Promega, USA) was added to a final protease:protein ratio of 1:100 (w/w) and incubated for 3 h at 37 • C. Samples were then digested by sequencing-grade modified trypsin (Promega) at ratio of 1:50 (w/w) for 16 h at 37 • C. Peptides were collected by centrifugation at 16,000×g for 20 min.

iTRAQ labeling
Peptides were labeled with iTRAQ8-plex reagent respectively according to manufacturer's protocol. The labeled peptides were combined and vacuum-dried. The dried peptides were desalted using a Sep-Pak C18 cartridge (Waters, USA) and the eluted peptides were also dried using a vacuum concentrator. The mixtures were fractionated by high pH reversed phase liquid chromatography (Waters, USA).

LC-MS/MS analysis
Label-free peptides or iTRAQ-labeled peptides were dissolved in 0.1% formic acid. Separation was performed on a reverse C 18 column (C 18 3 μm, 100Å 75 μm×25 cm) from Thermo, with elution gradient from 8% to 38% buffer B (0.1% FA, ACN) with a flow rate of 300 nl/min for 2 h by Eksigent 1D plus. Peptides were injected into Triple Time of Flight (TOF) 5600 (AB Sciex, USA) operated in positive mode with an ion spray voltage at 2.3 kV. Survey scans were acquired from 350 to 1500 m/z, while MS/MS scans were from 100 to 1250 m/z in the high-sensitivity mode. The 20 most intensive precursors were respectively selected for fragmentation per cycle. The total LC-MS running time for each IDA injection was 120 min. Additionally, a subset of randomly iTRAQ-8plex samples was analyzed on Orbitrap mass spectrometer (Thermo, Q Exactive and Fusion) using the same LC setup and gradient as described earlier of the Q TOF-based analysis.

Proteomic data analysis
For LFQ experiments, protein identification and quantification were performed with ProteinPilot TM software (version 4.5) against a Uniprot_Human database. For iTRAQ-labeled experiments, relative peptide quantification and protein identification were performed using Proteome Discoverer TM software (version 2.1) with Mascot search engine. Search parameters were set as follows: (i) species, Homo sapiens; (ii) protein database, UniProtKB/Swiss-Prot (including 20264 sequences) ; (iii) quantification method, iTRAQ 8-plex (peptide labeled); (iv) digestion, Trypsin, allowing up to two missed cleavage; (v) static modifications, Carbamidomethyl, iTRAQ8plex; (vi) dynamic modifications, methionine oxidation; (vii) precursor mass tolerance = 10 ppm, fragment mass tolerance = 0.05 Da. Quantification is based on the relative intensities of reporter ions which appear in the low mass range of MS/MS spectra. The mass spectrometry proteomics data have been deposited to the iProx Consortium (http://www.iprox.org). Proteins with FDR < 0.01 on both protein and peptide level and matching ≥ 1 unique peptide were considered as positively identified proteins. Proteins with quantification p-value < 0.05 (DCM cases vs. healthy controls) and with FC ≥ 1.3 or ≤ 0.769 (the average FC of repeat experiments) were considered as differentially expressed proteins (DEPs).
The gene ontologies (biological processes and molecular functions) of all IDs were searched against the Gene Ontology database using Blast2GO-Functional Annotation and Genomics (http://www.blast2go.com/) as well as KEGG Mapping-GenomeNet (http://www.genome.jp/kegg/). Pathway analyses for protein-protein interactions and upstream regulations of differentially expressed candidates were performed using Ingenuity Pathway Analysis (IPA) software (QIAGEN, USA). Cluster analysis was performed using STRING networks and visualized via Cytoscape software (version 3.8.0). 41,42 The specific statistical tests used are indicated in the figure legends or appropriate methods section and were performed within the R statistical environment.

2.9
Serum preparation and enzyme-linked immunosorbent assay DCM patients (n = 53) and healthy volunteers (n = 35) were enrolled in this experiment. Serum samples were obtained from Zhongshan Hospital and stored at −80 • C immediately after collection, with the approval of the Research Ethics Committee from this hospital of Fudan University. Written informed consent was obtained from all study subjects. Protein biomarkers in serum samples were validated using Enzyme-Linked Immunosorbent Assay (ELISA) kits, including CTSB (Abcam, UK), vWF (Abcam, UK), C9 (Abcam, UK), DKK3 (R&D Systems, USA), MFGE8 (R&D Systems, USA), APOA2 (Abcam, UK), FABP4 (R&D Systems, USA), DES (RayBiotech, USA). Experiments were performed according to manufacturer's instructions.

2.10
Machine learning approach for diagnostic panel discovery Potential serum biomarkers were assessed by receiver operating characteristic (ROC) analysis. The area under the ROC curve (AUC) of the proposed protein panel was used as a metric to evaluate the sensitivity and specificity of the biomarker performance. A random forest (RF) model is one of the most-used supervised machine learning algorithm which is applied to predict disease without hyper-parameter tuning. [43][44][45][46] The accuracy of the RF model at predicting DCM was tested using least absolute shrinkage and selection operator (LASSO) regression. There are several other variable selection methods that have traditionally been used to build models, however, the LASSO is based on minimizing mean squared error, which is based on balancing the opposing factors of bias and variance to build the most predictive model and minimize prediction error. 47,48 LASSO can be used as a variable selection method via cross-validation in three steps. First, separating the data into a training set and a test set, then building the model in the training set, lastly, estimating the outcome in the test set using the model from the training set. Specifically, the data were split into five equal parts with the model built in 80% of the data and then tested in the remaining 20% of the data. Each part served as the test set and the mean squared error was averaged across all the folds. Samples of the training dataset were selected randomly with replacement 200 times and the proteins with repeat occurrence frequencies were selected as the diagnostic panel of DCM. The parameter lambda was determined based on a 10-fold cross-validation. LASSO analysis was conducted in R version 3.5.3 with the glmnet package. Sensitivity and specificity were used to evaluate the performance of the RF model. Sensitivity represents the correctly classified positive samples to the total number of positive samples, while specificity represents the correctly classified negative samples to the total number of negative samples.

Region-resolved proteomic and transcriptomic profiling of cardiac tissues
Heart biospecimens of 10 DCM patients underwent orthotopic heart transplantation and 10 healthy donors were reviewed (Table S1). The workflow of this study was presented in Figure (1A). We first selected three healthy hearts and applied proteome and transcriptome analyses to the four chambers-LA, LV, RA, and RV, then RNAseq analysis was applied to compare different transcripts between three DCM hearts with three healthy ones, lastly the iTRAQ-labeled proteomic approach was performed to depict disease-relevant molecular alterations through ten DCM hearts versus ten controls ( Figure 1A). Venn diagrams were used to exhibit the overlapping genes/proteins between transcriptomic and proteomic datasets. In total, 19,880 transcripts and 7,605 proteins were identified in human heart tissues ( Figure 1B), among which 14,840 transcripts and 6,335 proteins were identified from four chambers of the human healthy heart ( Figure 1C,D).
Proteome and transcriptome dataset from three healthy hearts were divided into high-, medium-, and lowabundant fractions based on intensity-based absolute quantification (iBAQ) value and fragments per kilobase per million mapped fragments (FPKM) value, respectively. Chamber-enriched biological processes of each fraction were annotated, and four lines represented four chambers. High-abundant proteins/transcripts were highly similar within each chamber, for example, energy production, nucleotide biosynthesis, and muscle contraction. Mediumabundant proteins were relevant to translation, mitochondrion organization, and mRNA processing; while mediumabundant transcripts were associated with RNA splicing, mitotic cell cycle, and lipid biosynthesis. However, lowabundant proteins displayed divergent characteristics, proteins were related with immune response, GTPase activity, and actin filament process; transcripts were implicated Three protein fractions were presented as below: Log 2 (iBAQ) between 0-10 were defined as low, between 10-15 were medium, while larger than 15 were high. Three transcript fractions were classified (low: Log 2 (FPKM) <4, medium: Log 2 (FPKM) between 4-9, and high: Log 2 (FPKM)>9).
in chromatin modification and DNA damage checkpoint ( Figure 1E,F).

Region-specific proteome and transcriptome of healthy heart
Global proteome and transcriptome were defined in three healthy hearts to explore physiological characteristics in four chambers. Partial least-squares discriminant analysis (PLS-DA) model created with the label-free identified proteins from four chambers of healthy hearts (n = 3) classified the proteomes into two groups of ventricles and atria ( Figure S1A). Loading scatter plot was shown in Figure (S1B). The dynamic range of proteins detected by label-free coupled mass spectrometry spanned six orders of magnitude and that of transcripts detected by RNAseq spanned about four orders of magnitude ( Figure S1C). The relationship between mRNA and protein expression were analyzed, slope of 0.64693 for LA and between 0.65 and 0.67 for other three chambers ( Figure S1D), the number of protein molecules produced per molecule of mRNA appeared to be a little larger for low-than for high-abundance transcripts.
Besides contractile function of myocardium, the omics data of each chamber revealed chamber-specific functions. LV proteins mainly enriched in mitochondrial transport, actin-mediated contraction, and ATP biosynthesis. LA proteins enriched in mitochondrion organization and muscle contraction, as well as RNA splicing and oxygen transport. Proteins in RV focused on Golgi vesicle transport and microtubule polymerization, while RA focused on cytoskeleton organization and protein-lipid complex assembly (Figure 2A). At the transcriptomic level, LV prominently involved in the muscle contraction function. RV mainly enriched in cell adhesion and chemotaxis. Biological functions among two atriums could be categorized as homeostasis process, immune response, and substance metabolism ( Figure 2B).
The vast differences in the abundances of mRNA and protein expressed within each heart chamber were visualized by plotting the ranked order of relative intensities of transcripts and proteins ( Figure 2C). For example, ACTC1 (muscle protein actin) was the most abundant protein in four chambers owing to the contractile function of heart organ, and myosin-related genes (MYL and MYH families) presented to be rich in both protein abundance and transcriptome abundance from each chamber of the heart. As for LV chamber, three of the top five abundant transcripts were troponins (TNNI3, TNNC1, and TNNT2), while three of the top five proteins were myosin isoforms (MYH7, MYL3, and MYL2). Troponins are cardiac regulatory proteins that control the calcium mediated interaction between actin and myosin. 49 Myosin isoforms are actinbased motor molecules with ATPase activity which are essential for muscle. The abundance distributions of transcripts and proteins exhibited different characteristics in each heart chamber.
To reveal specific protein markers for each heart chamber, protein expression values were calculated from the individual median averaged protein LFQ intensities and the corresponding median LFQ intensities of the other three chambers ( Figure 2D). With the criteria log 2 fold expression > 2.585 (intensity ratio > 6) when compared iBAQ intensity in the indicated heart chamber with other three chambers, 47, 33, 42, and 18 chamber-specific markers in LA, LV, RA, and RV chamber, respectively, were delineated. The well-known cardiac markers such as MYL7, MYBPHL and SNTG2, were clearly recovered as LA chamber-specific proteins. LV-specific markers included ANKRD2 which could regulate gene expression during the muscle development and in response to muscle stress. Although muscle contractile in heart is the predominant role, we proved again that heart is a highly differentiated organ and each chamber contains unique functions.

Proteome changed in dilated cardiomyopathy cases versus healthy controls
Up to 50% of the DCM cases its exact cause remains initially unknown; this condition is called idiopathic DCM. 50,51 Here, comprehensive proteome profiling was generated from ten idiopathic DCM hearts and ten healthy controls via iTRAQ-based quantitative proteomics. We checked the ten iTRAQ-labeled proteomic datasets and deleted three datasets with the criteria deviation of significant regulated proteins < 20% (Table S2). Boxplot figure exhibited the global distribution of the remaining seven proteomic datasets ( Figure S2). Correlation of protein expression profiling in each cardiac chamber showed strong correlation between protein expression in healthy heart chambers, while low correlation in DCM heart chambers, indicating cardiac chambers went through divergent biological events during DCM progression ( Figure S3).
Proteome measurements of all seven hearts from end-stage DCM patients and seven from healthy controls resulted in a total of 7,177 protein groups at a 1% FDR at both protein and peptide levels. Among which 557 proteins were significantly changed, in detail, 192 proteins were remarkably altered in LA chamber (71 down-regulated and 121 up-regulated), 174 proteins were notably changed in LV chamber, whereas 202 and 238 proteins were markedly changed in RA and RV chambers, respectively  Table S3). p-Values were calculated from the data of seven DCM hearts versus seven healthy controls. As shown in the volcano plots, mitochondrial electron transport related proteins (such as ME3, COX5A, and NDUFA7) as well as mitochondrial oxidoreductase HIBADH were decreased, while actin-binding proteins (such as NRAP and MYOZ1) and microtubule-associated protein 4 (MAP4) were increased in LA chamber. Cell adhesion-related proteins (such as POSTN and VCAN) and von Willebrand factor (VWF) were increased in all four chambers; meanwhile, myosin-related proteins MYH16 and MYL9 as well as actin-binding protein ACTN1 were increased in LV chamber. Venn diagram illustrated the numbers of common and unique differential expressed proteins (DEPs) in four chambers. For the increased DEPs, 16 common proteins in four chambers were delineated, including MFGE8, LTBP2, SERPINH1, and so on ( Figure 3E). For the decreased DEPs, 4 common proteins in four chambers were exhibited ( Figure 3F). Chord diagram for Gene ontology (GO) analysis of 557 DEPs was presented ( Figure S4). The link between cardiac chambers and pathways was described grounded on GO biological processes. The thickness of each line represented the number of DEPs. Different chambers and pathways were color-coded according to the catalog. Iron ion transport, ATP synthesis, actin cytoskeleton organization, protein folding, etc. were remarkably enriched in each chamber. Collectively, DEPs between four chambers presented a large variation in protein expression pattern, indicating that each chamber undergoes its specific pathological processes.

Integrative analysis of transcriptome and proteome between dilated cardiomyopathy patients and healthy controls
We randomly chose three DCM patients and three healthy controls from collected specimens, then performed transcriptome study for four cardiac chambers via RNA-Seq approach. A total of 14,880 transcripts were detected, and 987 transcripts were changed among four chambers compared with healthy controls (Table S4). The top 28 most significant GO terms (p < 0.05) in down-regulated and up-regulated transcripts were enriched ( Figure 4A). In the down-regulated transcripts, the hydrogen peroxide metabolic process was the most significant term from four chambers; moreover, mitochondrial translation was the specific term enriched in the LV chamber. As for up-regulated transcripts, ECM organization was the most significant term in LA, LV, and RV chambers.
We further integrated global transcriptomic and proteomic changes to explore molecular features of DCM ( Figure 4B-E, Table S5). A total of 6,405 transcript-protein pairs were quantified, and the majority of pairs were remained unchanged when compared DCM hearts with healthy controls (gray dots). Notably, a total of 44 nonredundant mRNA-protein consistent pairs were observed in four chambers (red dots), mainly focused on ECM organization and small molecule metabolism. According to gene ontology cellular component (GOCC) classification, 34 genes (including COL8A1, FBN1, VCAN, SOD3, etc.) were associated with ECM and 7 genes (including DECR1, ABAT, COQ10A, PRDX1, etc.) were related to mitochondrion. Most genes were only changed at either transcript level (yellow dots) or protein level (blue dots). Transcripts altered while proteins unchanged pairs were enriched in signal transduction, lipid metabolism, and mitochondrion organization. Proteins altered while transcripts unchanged pairs were enriched in protein folding, anatomical structure development, cell junction organization, and so on. In brief, the discrepancies between transcriptome and proteome changes reflected the significance of proteomic research, correspondingly, the post-transcriptional regulation could play an important role in DCM development.

Landscape of systematic changed proteins during dilated cardiomyopathy progression
The total 557 DEPs in DCM hearts versus healthy controls were classified into six groups according to their variation trends in four chambers, and the relevant pathways were analyzed ( Figure 5A,B). Routinely, the altered proteins in LV and RV chambers from patients with DCM were mainly associated with actin filament organization, muscle contraction, and mitochondrial translation. However, other chambers uncovered intriguing changes which were different from ventricular chambers, such as ubiquitination, immune regulation, mRNA regulation, and so on. Six major functions categories were further focused and subclassed using 99 DEPs, namely mitochondrial electron transport, mitochondrial translation, muscle contraction, extracellular matrix (ECM) organization, actin filament organization, and adaptive immune response ( Figure 5C). Mitochondrial electron transport-related DEPs were overexpressed in LV chamber, whereas ECM organizationrelated DEPs were upregulated mainly in LA chamber and partially in LV chamber. The mitochondrial electron transport process and mitochondria translation process tended to be downregulated in the RV chamber. As for the muscle contraction process, desmin (DES), myosin regulatory subunits (MYL12B and MYL9), and tropomyosin alpha-3 chain (TPM3) were up-regulated when compared each chamber from DCM hearts with healthy controls, while   (C) Six major functions categories of DEPs, namely mitochondrial electron transport, mitochondrial translation, muscle contraction, extracellular matrix organization, actin filament organization, and adaptive immune response myosin heavy chain isoforms (MYH4, MYH8, MYH6, MYH13, and MYH1), ATPase subunits (ATP1B1, ATP1A1, and ATP2A2), as well as serine/threonine-protein kinase mTOR (MTOR) were down-regulated in four chambers. Our data emphasized the commonality of mitochondrial dysfunction and ECM remodeling in the end-stage DCM.
A STRING network was further retrieved for the total 557 DEPs deemed significantly changed in patients suffering from DCM. Clustering via Cytoscape software was relative to the confidence score of the interaction, and enrichment analysis was performed on the resulting clusters. 41 Eight clusters were annotated in Figure (S5), including extracellular structure organization, ATP metabolic process, mitochondrial translation, sarcomere, and so on. These processes represented the major molecular changes during DCM progression.

Protein network analysis and discover potential serum biomarker panel
To explore the etiology of DCM, interactive networks were retrieved to identify functional modules for all proteins deemed significant by ANOVA analysis (Table S6). Systematically changed proteins derived from the DCM atrium were involved in essential functions as developmental disorder, protein synthesis, cellular assembly and organization, lipid metabolism, and so on ( Figure 6A). Correspondingly, systematically changed proteins in the DCM ventricle were largely related to cell cycle and cell morphology, skeletal and muscular disorders, cardiac dilation and cardiac enlargement, energy production, and so on ( Figure 6B).
To discover potential DCM biomarkers, we utilized dysregulated proteins identified in cardiac tissue and potentially secreted into the blood as biomarker candidates. Notably, 14 dysregulated secreted proteins were quantified in serum samples from a cohort of 53 DCM patients and 35 healthy donors. Using ELISA analysis, 8 proteins differed significantly between the DCM group and the control group, including cathepsin B (CTSB), vWF, complement component C9, dickkopf-related protein 3 (DKK3), etc. (*p < .05, **p <.001, t-test; Figure 6C). ROC curves were shown in Figure (S6). With the criterion of ROC area >0.75, 4 proteins were screened as potential biomarkers for DCM diagnosis, including CTSB, vWF, C9, and DKK3. Of which CTSB has been reported that increased myocardial expression was found in patients with DCM; 52 whereas C9 has been demonstrated to be effective in detecting acute myocardial damage. 53 Moreover, DKK3 could prevent familial DCM development in mice through activation of the canonical and inhibition of the noncanonical Wnt signaling pathway. 54 Supervised machine learning RF model was further applied to classify DCM patients and healthy controls with the relative quantitative data of 8 proteins ( Figure 6D). Based on the attribute importance obtained from average impurity decrease algorithm, combination of CTSB, vWF, C9, and milk fat globulin protein E8 (MFGE8) were recommended for DCM prediction. Using this optimal combination of predictors, only 4 of the 35 healthy individuals and 3 of the 53 DCM patients were wrongly predicted, hence the discriminative accuracy was 0.92. Additionally, the AUC had improved from 0.855 for CTSB to 0.989 in the combined model ( Figure 6E).
Among which, CTSB, vWF and C9 were screened by ELISA experiment earlier with the criterion of ROC area >0.75. MFGE8 was supplemented to the biomarker panel by machine learning algorithm (AUC = 0.663). MFGE8 is a pleiotropic-secreted glycoprotein which plays a crucial role in the maintenance of tissue homeostasis and the prevention of inflammation. 55 Notably, MFGE8 was the first time reported to be associated with cardiovascular disease. The discovery of this newly developed panel would open opportunities for better definition and diagnosis of the end-stage DCM.

DISCUSSION
Numerous efforts have been fostered to characterize DCM heterogeneity in a clinicopathological relevant manner, however, timely and precise diagnosis remains a major clinical challenge. An overview of heart with anatomically resolved information is necessary for cardiac biology study and DCM mechanism exploration. In this study, we applied region-resolved proteomic and transcriptomic analyses to identify a total of 7,605 cardiac proteins and 19,880 transcripts from four chambers of end-stage DCM hearts and healthy controls ( Figure 1B). The distribution of proteins and transcripts from healthy hearts was outstandingly enriched in the processes of oxidative phosphorylation and muscle contraction (Figure 2A,B). The abundant proteins were mainly observed as structural and mitochondrial function related ones (e.g., MYLs, MYHs, TNNT2, and ACTC1), which supported the strong contractive 'pump' and the high energy demand of the cardiac tissue ( Figure 2C). Practically, the cardiac myocyte is about one-third mitochondria by volume and is capable of utilizing all classes of energy substrates for a staggering 30 kg of ATP/day to maintain circulation and contraction. 56,57 The low-abundant proteins were divergent between each chamber ( Figure 1E) and chamberspecific markers were also investigated when compared the indicated chamber with other three chambers from healthy hearts ( Figure 2D), which suggested that heart was a highly differentiated organ and the specific functions engaged by each chamber. More important, we uncovered disease-specific expression changing patterns in each chamber which could significantly contribute to the DCM development. LV chamber manifests a robust plasticity response and pathological remodeling which involves re-organization of myocytes, interstitial cells and vessels leading to increased stiffness and/or impaired contractility. 58,59 Research group of Benjamin Prosser revealed a consistent upregulation and stabilization of intermediate filaments and microtubules in failing human hearts via proteomic analysis of LV tissue. Specifically, proliferated, de-tyrosinated microtubules acted as compression resistance elements to impair contraction in the failing heart and increase myocyte stiffness. 60 Cluster analysis demonstrated DCM-specific systematic changes that extracellular structure organization as well as cell adhesion binding process was dysregulated, apart from sarcomere structure and oxidative phosphorylation process ( Figure S5). Six major functional categories were depicted via region-resolved expression patterns when compared DCM hearts with healthy controls, whereas 34 genes were delineated in ECM organization ( Figure 5A,C). ECM constitutes a dynamic molecular network providing structural support to cardiac tissue homeostasis, 61 our data highlighted that ECM remodeling was a crucial pathologically change during DCM progression. In line with previous studies, which illustrated inflammatory cell infiltrated in myocardial biopsy samples of patients with DCM, 62 or hyper-activated NLRP3 inflammasome with pyroptotic cell death of CMs were presented in the myocardial tissues of DCM patients, 63 our dataset revealed that the inflammation process was enriched in LA chamber from DCM patients. This study supported that increased filling pressures in response to LV failure would further impair the functions of the RV and atria.
Omics analyses indicate that post-transcriptional regulation of gene expression plays an overriding role in the normal and diseased heart. Integration of multi-omics in Cardiovascular diseases (CVDs) has begun to present high potentials for translational discoveries. Research group of Elena Aikawa analyzed human stenotic aortic valves obtained from valve replacement surgeries via spatiotemporal transcriptomics and global unlabeled and label-based tandem-mass-tagged proteomics. Calcific aortic valve disease (CAVD) occurs in the context of a complex, multi-layered tissue architecture. A high-resolution spatiotemporal atlas of the combined valvular tissue, layer, and cell proteome, and a comprehensive repository of molecular drivers of CAVD were performed to demonstrate side-specific valve cellular calcification potential and identify a novel layer-specific cell marker. Disease stage-and layer-specific omics profiling was the first to examine CAVD with a tissue resolution, thereby paving the way for clinically meaningful investigations and development of novel antifibrotic and anti-inflammatory therapeutics for the treatment of CAVD. 64 Kalyanasundaram et al. explored the function of adult human sinoatrial node (SAN)specific fibrosis in atrial fibrosis and arrhythmias. Intact SAN pacemaker complex was dissected from cardioplegically arrested explanted nonfailing hearts (non-HF) and human failing hearts. Fibroblasts from the central intramural SAN pacemaker compartment and right atria were isolated and subjected to comprehensive high-throughput next-generation sequencing of whole transcriptome, microRNA, and proteomic analyses. Proteomic signatures and signaling pathways associated with ECM flexibility, stiffness, focal adhesion, and metabolism were uniquely altered in HF SAN fibroblasts compared with non-HF SAN. 65 In this study, we integrated global transcriptomic and proteomic changes to explore molecular features of DCM ( Figure 4B,E), the unbiased omics data provided a glimpse into the molecular framework underpinning altered mitochondrial energetics, oxidative stress, and ECM in the heart undergoing pathological hypertrophy.
Biomarkers are increasingly recognized as an effective means for risk prediction in chronic heart failure and guiding management. 66,67 Brain natriuretic peptide (BNP) appears to be the gold standard in supporting the daily clinical treatment of patients with heart failure, 68 however, that is not DCM specific. Linking our findings with published researches would discover novel DCM biomarkers. ELISA experiments were applied to screen four proteins (including CTSB, vWF, C9, and DKK3 with AUC >0.75) as potential serum biomarkers for early detection of DCM ( Figure 6C). A 4-biomarker panel, containing CTSB, vWF, C9, and MFGE8, was further developed for DCM prediction based on the supervised machine learning model which had an AUC of 0.989 ( Figure 6D,E). These four proteins were functional complementary, among which CTSB may amplify the in-progress inflammatory response and have a direct role in plaque destabilization through degradation of the fibrous cap in patients with atherosclerosis. 69 vWF plays critical role in platelet recruitment to the injured vessel wall, whereas the acute release of vWF in myocardial infarction predicts death and heart failure. 70,71 C9 is the pore-forming subunit of the membrane attack complex that plays a key role in the innate and adaptive immune response. 72 MFGE8 is a pleiotropic-secreted glycoprotein which plays a crucial role in the maintenance of tissue homeostasis and the prevention of inflammation. 55 Notably, this study reported MFGE8 was associated with cardiovascular disease for the first time. Although this newly developed biomarker panel is quite promising, it will require a more thorough analysis including a larger patient cohort for further clinical application.
This study established anatomically resolved transcriptome and proteome landscapes of healthy hearts and end-stage DCMs, which revealed disease-relevant molecular changes and pinpoint potential mechanism of DCM etiology. A promising 4-protein panel was developed for potential clinical DCM diagnosis. In summary, these integrated datasets provide diverse and rich resources to investigate the molecular basis of heart physiology and pathology including cardiomyopathies.

A U T H O R C O N T R I B U T I O N S
Ling Lin and Shanshan Liu designed, performed, and interpreted the MS-based proteomic analysis of clinical specimens and wrote the paper. Zhangwei Chen, Yan Xia, and Juanjuan Xie performed the bioinformatics analysis, generated the figures, and revised the paper. Mingqiang Fu, Danbo Lu, and Yuan Wu provided patient material and clinical data and revised the paper. Huali Shen, Pengyuan Yang, and Juying Qian designed and supervised the project, developed a protein biomarker panel, reviewed, and edited the paper. We gratefully acknowledge Institutes of Biomedical Sciences, Fudan University for providing mass spectrometry by request. We sincerely appreciate Le Kang for assistance in tissue histology examination, Guoquan Yan for technical assistance in the quantitative proteomic experiments, Yang Zhang and Lujie Yang for bioinformatic tools. We thank many past and present members of Chinese Human Proteome Project (CNHPP) Consortium for thoughtful suggestions and supports.

C O N F L I C T O F I N T E R E S T
The authors declare no conflict of interest.