key: cord-0755315-c0cwo240
authors: Kaur, N.; Oskotsky, B.; Butte, A. J.; Hu, Z.
title: Mining transcriptomics and clinical data reveals ACE2 expression modulators and identifies cardiomyopathy as a risk factor for mortality in COVID-19 patients
date: 2020-10-23
journal: nan
DOI: 10.1101/2020.10.20.20216150
sha: 9e06c143417b332ac019803c5803104a790a252d
doc_id: 755315
cord_uid: c0cwo240

Angiotensin-converting enzyme 2 (ACE2) is the cell-entry receptor for SARS-CoV-2. It plays critical roles in both the transmission and the pathogenesis of the coronavirus disease 2019 (COVID-19). Comprehensive profiling of ACE2 expression patterns will help researchers to reveal risk factors of severe COVID-19 illness. While the expression of ACE2 in healthy human tissues has been well characterized, it is not known which diseases and drugs might modulate the ACE2 expression. In this study, we developed GENEVA (GENe Expression Variance Analysis), a semi-automated framework for exploring massive amounts of RNA-seq datasets. We applied GENEVA to 28,6650 publicly available RNA-seq samples to identify any previously studied experimental conditions that could directly or indirectly modulate ACE2 expression. We identified multiple drugs, genetic perturbations, and diseases that modulate the expression of ACE2, including cardiomyopathy, HNF1A overexpression, and drug treatments with RAD140 and Itraconazole. Our unbiased meta-analysis of seven datasets confirms ACE2 up-regulation in all cardiomyopathy categories. Using electronic health records data from 3936 COVID19 patients, we demonstrate that patients with pre-existing cardiomyopathy have an increased mortality risk than age-matched patients with other cardiovascular conditions. GENEVA is applicable to any genes of interest and is freely accessible at http://www.genevatool.org .

Coronavirus disease 2019 is an infectious disease caused by severe acute 31 respiratory syndrome coronavirus 2 (SARS-COV-2). The World Health Organization (WHO) 32 declared the COVID 19 outbreak a pandemic on March 11, 2020. As of October 5, 2020, there 33 have been 34.8 million recorded COVID-19 cases and over 1 million deaths 1 . 34 35

Angiotensin-converting enzyme 2 (ACE2) is the cell-entry receptor for SARS-CoV-2 2 . The 36

binding between ACE2 and spike (S) protein of SARS-COV-2 initiates the viral entry into target 37 cells. ACE2 plays key roles in both the transmission and pathogenesis of SARS-CoV-2, as 38 demonstrated by the following lines of evidence: 1) SARS-CoV-2 fails to infect the lung-derived 39 cell line A549 in the absence of ACE2 expression. The infection is restored after overexpressing 40 ACE2 in the cell line 3 . 2) SARS-CoV-2 fails to infect wild type mice but can infect and cause 41 pneumonia in transgenic mice expressing human ACE2 4, 5 . 3) COVID19 related tissue damages 42 are detected in organs with ACE2 expression, including lungs, intestines, colons, and hearts 6-8 . 1 4) ACE2 expression is increased in the lungs of patients with comorbidities associated with 2 severe COVID-19, suggesting that the level of ACE2 expression is associated with disease 3 severity 9 . Taking these lines of evidence together, it is crucial to comprehensively characterize 4 the ACE2 expression in human tissues. 5 6

To comprehensively profile the expression patterns of ACE2, we not only need to characterize 7 its expression in healthy tissues but also identify diseases, drugs and genetic perturbations that 8 modulate ACE2 expression. The expression of ACE2 in healthy human tissues has been well 9

characterized by resources such as the Human Cell Atlas and GTEx, with the highest 10 expression detected in intestine, testis, lung, cornea, heart, kidney, and adipose tissues 10,11 . 11 However, it is still not clear which diseases and drugs modulate the ACE2 expression. Since 12 ACE2 expression is tightly associated with the pathogenicity of SARS-COV-2, identification of 13 ACE2 modulating conditions will help us reveal and explain risk factors of severe illness from 14 RNA-sequencing data profiles the full transcriptome of samples. Currently, more than 200,000 17 human RNA-seq samples are publicly available, providing an unprecedented opportunity for us 18 to examine ACE2 expression in different human cell types under a variety of conditions and 19

treatments. Data harmonization efforts such as ARCHS4 have uniformly preprocessed the 20 RNA-seq data, making them readily available for analysis 12 . However, fully automated analysis 21 of these datasets faces two main obstacles. First, the metadata are non-standardized and are 22 often unstructured, making it difficult to extract experimental conditions from the studies. 23

Second, experimental designs are highly variable. While some studies adopt the simple control-24

versus-treatment design, other studies are more complicated, involving multiple time points, 25 combination treatments, or stratified cohorts. The heterogeneous design makes it difficult to 26 analyze the datasets using a single statistical model. 27 28 Multiple tools have been made to analyze transcriptomics data, including CREEDS 13 , 29 scanGEO 14 , GEM-TREND 15 , StarGEO 16 , SIGNATURE 17 , SPIED 18 , Cell Montage 19 , 30

ProfileChaser 20 , ExpressionBlast 21 and SEEK 22 . However, the existing tools have several 31 limitations, preventing them from fully exploring the publicly available RNA-sequencing 32 resources. First, some of the tools annotate the metadata manually and are unable to cover the 33 large number of datasets currently available. Second, the tools focus on differential expression 34 analysis between two groups (e.g., control versus treatment), preventing them from analyzing 35 studies with more complex study designs. 36 37

In this study, we developed GENEVA (GENe Expression Variance Analysis), a semi-automated 38 framework for exploring public RNA-seq datasets. For a given gene, GENEVA identifies the 39 most relevant datasets by analyzing the variance of the gene expression. GENEVA visualizes 40 the relevant datasets for detailed manual analysis. GENEVA is scalable and is agnostic to study 41 designs. Using GENEVA, we identified multiple drugs, genetic perturbations, and diseases that 42 modulate the expression of ACE2, including cardiomyopathy, HNF1A over-expression,and drug 43 treatments with RAD140 and Itraconazole. Our in-depth meta-analysis of seven datasets 44 3 reveals increased ACE2 expression in all cardiomyopathy categories. By analyzing the clinical 1 data of 3936 COVID19 patients at UCSF hospital, we demonstrate that patients with pre-2 existing cardiomyopathy have an increased mortality risk than other patients, including 3

propensity score-matched patients with other cardiovascular conditions. 4 5 6

Analysis of 286650 RNA-seq samples reveals complex transcriptional networks of ACE2. 9

Our study leverages human RNA-sequencing data from the ARCHS4 project, containing 10 286,650 uniformly pre-processed data from 9,124 Gene Expression Omnibus (GEO) series 12 . 11

The large number of RNA-sequencing samples provide an unprecedented resource for studying 12 the expression of ACE2 in different human cell types under a variety of conditions and 13 treatments. 14 15

We first characterized the transcriptional networks of ACE2 using all 286,650 samples. We 16 calculated the Pearson correlation between ACE2 and all other human genes. Because of the 17 large sample size, most of the correlations are statistically significant, even after multiple testing 18

adjustments. Therefore, we focused on the correlation coefficients themselves as a measure of 19 effect size, rather than the p values or significance. While most of the genes have correlation 20 coefficients near 0, a small set of genes are highly correlated with ACE2 (Supplemental Fig.  21 1A). Among 35,238 human genes, 85 (0.24%) genes have correlations higher than 0.5. The top 22 correlated genes include FABP2, MEP1B, and transcription factors such as HNF4G 23 (Supplemental Fig. 1B and C). We examined their expression in different human tissues using 24

GTEx data. We found that these genes are highly expressed in colons and small intestines, an 25 expression pattern similar to ACE2. Similarly, we see high correlations between ACE2 and 26 multiple pathways related to the digestive process (Supplemental Fig. 1D ). These results 27

indicate that the correlations are dominated by the overall transcriptomic difference between the 28 digestive system and other human organs. 29 30

To avoid our analysis being dominated by known cellular or organ differences, we evaluated the 31 correlations between ACE2 and other human genes within each RNA-seq dataset. We then 32 calculated the mean and variances of the correlations across all datasets, weighted by the 33 sample sizes. Our analysis reveals a positive relationship between the mean and variance of 34 the correlation coefficients (Supplemental Fig. 1E ). While some genes have a high average 35 correlation with ACE2, their correlation with ACE2 is highly variable in individual datasets. The 36 results suggest against a common transcriptional network around the ACE2 gene. Rather, 37

ACE2 is co-regulated with different sets of genes under different conditions. 38 39

We adjusted for the variances by calculating the standardized correlation coefficient, defined as 40 the mean correlation divided by the standard deviation of the correlation. The standardized 41 correlation coefficient allows us to prioritize genes with relatively conserved correlations with 42

ACE2 across all studies. The top genes include MYO7B, PDZK1 and transcription factors such 43 as HNF1A (Supplemental Fig. 1 F and G) . Pathway analysis identified many metabolic 44 4 processes to be correlated with ACE2 expression. The findings are consistent with previous 1 observations that ACE2 is involved in glucose metabolism and energy stress responses 23,24 . 2 3

Our analysis identified three transcription factors in the hepatocyte nuclear factor family, 4

including HNF4G, HNF1A, and HNF4A (Supplemental Fig. 1 C and G) . HNF4G has the highest 5 overall correlation with ACE2 while HNF1A has the highest standardized correlation coefficient 6

within studies. We then tested the causal relationship between the transcription factors and 7 ACE2 expression. We identified two RNA-seq datasets that compared human cells with or 8 without genetic perturbation of HNF4G and HNF1A. While HNF1G is positively correlated with 9 ACE2 expression (Supplemental Fig. 1 C) , overexpression of HNF1G does not lead to 10 significantly increased ACE2 expression (Supplemental Fig. 1I ). Rather, there is a trend of 11 reduction in ACE2 expression. HNF1A overexpression leads to increased ACE2 expression and 12

HNF1A knockdown reduced ACE2 expression (Supplemental Fig. 1J Next, we hope to identify conditions that modulate the expression of ACE2. We developed a 24 computational framework named GENEVA (Gene Expression Variance Analysis) to identify the 25 most relevant datasets for visualization and detailed manual analysis ( Fig. 1A and Methods).

GENEVA prioritizes the datasets that have a large variance of ACE2 expression. The rationale 27

is that datasets with large ACE2 variance are likely to contain conditions that modulate the 28 ACE2 expression. At the same time, GENEVA controls for the overall heterogeneity of the 29 samples to prioritize datasets in which ACE2 is specifically modulated by experimental 30 conditions rather than due to tissue type differences. In addition, GENEVA embeds the meta-31 data into numerical space and prioritizes datasets with high correlations between ACE2 32 expression and the metadata (Fig. 1B) . This allows GENEVA to identify datasets in which ACE2 33 is regulated by experimental conditions rather than randomness or unexplained factors. While 34 our study focuses on ACE2 and its role in COVID-19 disease, GENEVA is applicable to all 35 genes. We created a web application that allows researchers to apply GENEVA to their gene of 36 interest [http://genevatool.org]. 37 38

We tested the significance of the GENEVA scores using a permutation procedure. We randomly 39

shuffle the samples across studies to generate a null distribution. We compared each GENEVA 40 score to the null distribution to calculate the p-value. We adjusted for multiple-testing using the 41 false discovery rate (FDR) method 26 . We identified 27 significant datasets with FDRs less than 42 0.05 (Table 1) . Interestingly, GENEVA identified HNF1A as an ACE2 modulator, which was 43 also identified in our correlation analysis (Supplemental Fig. 1J ). GENEVA additionally identified 44 5 multiple drugs and diseases that modulate the ACE2 expression, revealing potential risk factors  1  for severe illness from COVID-19.  2  3 Here, we highlight three ACE2 modulating conditions, manually picked based on their effect on 4 ACE2 expression and their potential impact on public health. Data from GSE89714 show up-5 regulated expression of ACE2 in hypertrophic cardiomyopathy ( Fig. 2 A, B) . Hypertrophic 6 cardiomyopathy is the most common inherited heart disease, affecting an estimated 15,188,000 7 individuals (0.2%) worldwide 27 . Our finding is consistent with an increased death rate in COVID-8

19 patients with heart conditions 28-30 and suggests that higher ACE2 expression can contribute 9

to the increased risk. Data from GSE104177 showed that RAD140, a selective androgen 10 receptor modulator, induces ACE2 expression in human breast cancer xenografts (Fig. 2 C, D) . 11

Data from GSE114013 show that Itraconazole, an antifungal drug, up-regulates ACE2 12 expression in two colorectal cancer cell lines, HT55 and SW948 (Fig. 2 E regulates ACE2 while other studies show that the drug has no effect on ACE2. The effect of the 25 drug will be overestimated if a researcher only includes the studies with positive results. 26

Therefore, after the GENEVA analysis, unbiased meta-analyses are required to confirm the 27 findings. 28 29

We performed a comprehensive search for datasets related to the three highlighted conditions, 30

including cardiomyopathy, itraconazole treatment, and RAD140 treatment. We did not find 31 additional datasets related to itraconazole and RAD140 treatment. For cardiomyopathy, we 32 identified a total of 7 datasets. We performed a meta-analysis using a mixed-effect model, 33

taking data from all 7 datasets into account. The result confirmed that ACE2 expression is 34 significantly elevated in heart tissue samples from cardiomyopathy patients (p-value < 0.001 Although previous studies have demonstrated 37 , how 42 ACE2 is regulated in other types is unknown. Within the 7 datasets, we were able to identify all 43 the common cardiomyopathy types. Our analysis revealed significantly increased ACE2 44 6 expression in most of the cardiomyopathy types (Fig. 3) , including DCM, HCM, RCM, and 1 LVNC. Although the result of ARVC is not statistically significant, the data show a clear trend of 2 ACE2 upregulation (Fig. 3D ). 3 4 COVID-19 patients with pre-existing cardiomyopathy show an increased mortality rate 5 While COVID19 patients with cardiovascular conditions show a higher mortality rate, it is not 6 clear how cardiomyopathy, in particular, affects the survival of the patients. Because the ACE2 7 expression is significantly elevated in the heart of cardiomyopathy patients, we hypothesize that 8 pre-existing cardiomyopathy leads to increased mortality in patients with COVID19. 9

We identified 3936 COVID19 patients from the electronic health records (EHR) of the University 11 of California San Francisco (UCSF) hospital. We divided the patients into three groups, 12

including patients with pre-existing cardiomyopathy (N = 43), patients with other pre-existing 13 cardiovascular diseases (N = 624), and patients without cardiovascular diseases ( N = 3269) 14 (Table 2 ). The most common non-cardiomyopathy cardiovascular diseases include 15 hypertension (N = 424), atherosclerotic heart diseases (N = 120) and Cardiac arrhythmia (N = 16 105). 17 18 We first compared the cardiomyopathy patients to patients without cardiovascular diseases. 19

Patients with cardiomyopathy have a larger proportion of males and older ages. They also have 20 a higher percentage of patients with preexisting conditions such as cancer, diabetes, and 21 hyperlipidemia. A higher percentage of cardiomyopathy patients have severe COVID-19 22 disease presentations, including ventilator use, respiratory failure, chest pain, and death (Table  23 2). We then performed survival analysis to test the effect of cardiomyopathy while controlling for 24 differences in age, gender, and pre-existing conditions using a multivariable Cox proportional-25 hazards model. We confirmed that cardiomyopathy is significantly associated with the risk of 26 death (p = 0.004) ( Figure 4A ). 27 28 We next compared the cardiomyopathy patients to patients with other cardiovascular diseases. 29

Cardiomyopathy patients have a higher proportion of males compared to patients with other 30 cardiovascular diseases. Age, race, and pre-existing conditions are comparable between the 31 two groups. Again, we observe that a higher percentage of cardiomyopathy patients have 32 severe COVID-19 presentations, including ventilator use, chest pain, and death ( Table 2) . 33

Multivariable Cox proportional-hazards regression confirms that cardiomyopathy is significantly 34 associated with the risk of death (p = 0.038, 438 % increase in observed death rate) ( Figure  35 4A). We further confirmed the increased mortality by comparing cardiomyopathy patients with a 36 propensity score-matched cohort of patients with other cardiovascular diseases ( Figure 4B and 37 Table S1 ). 38 7 1 We then examined the survival of cardiomyopathy patients who are COVID-19 negative. We 2 compared the survival of COVID-19 negative cardiomyopathy patients (N = 2250) with a 3 propensity score-matched cohort of patients with other cardiovascular diseases (N = 18000). 4

The two cohorts are comparable in demographics and non-cardiovascular diseases (Table S2) . 5

The 5-year mortality rate is only slightly higher in the cardiomyopathy patients (p = 0.034, 22% 6 increase in observed death rate) (Fig. 4C ). When we consider the patient's survival at 160 days, 7 a time frame comparable to the COVID-19 positive dataset, there is no significant difference 8 between the survival of the two groups (Fig. 4C) . 9

10 Taken together, the results show that cardiomyopathy itself does not pose large additional risk 11 of mortality among patients with cardiovascular diseases. Rather, the interaction between 12 SARS-CoV-2 infection and pre-existing cardiomyopathy leads to increased mortality in patients. 13

Our transcriptomics analysis suggests that the up-regulated ACE2 expression may contribute to 14 disease severity of COVID-19 in patients with pre-existing cardiomyopathy. However The disease severity of COVID-19 patients varies from asymptomatic to life-threatening. While 21 we do not fully understand the reason behind such variation, it is clear that the disease severity 22

is determined by multiple factors, including age, gender, the status of the immune system, and 23 the pre-existing conditions 7,38-40 . ACE2 expression is a key determinant of the disease severity, 24

as shown by multiple studies in humans and in animal models 3,5,9 . Therefore, it is critical to 25 identify conditions that modulate ACE2 expression, as the information will help us reveal and 26 explain factors associated with increased risk of severe illness from COVID-19. 27 28

We leveraged the massive amount of publicly available RNA-seq data to identify the ACE2 29 modulating conditions. While many tools exist for analyzing bulk GEO data, they are not 30 optimized for this purpose. First, some tools require researchers to search for datasets using 31 keywords, such as the name of a drug or a disease. These tools do not address our needs, as 32

we are looking for any conditions that modulate ACE2 expression. Second, some tools require 33 manual annotation of experimental groups in the studies, which are not scalable and often only 34 cover a small subset of currently available datasets. Finally, the existing tools focus on 35 differential expression analysis of two groups and are unable to address more complex 36 experimental designs. 37 38 We address the problems using a variance analysis approach. Instead of comparing two 39 experimental groups, we quantify the variance of the gene expression across all samples in a 40 8 study. The rationale is that datasets with large ACE2 variances are likely to contain conditions 1 that modulate the ACE2 expression. We improved our approach by using two modifications. 2

First, we numerically embedded the metadata and calculated the regression coefficient R 2 3 between ACE2 and the embedding. This allows us to prioritize datasets in which the ACE2 4 variation is associated with metadata. Second, we controlled for the overall heterogeneity of 5 samples in the study, this allows us to prioritize datasets in which ACE2 are specifically 6 modulated rather than as a result of cell-type differences. 7 8

Our study identifies multiple diseases, conditions, and genetic perturbations that modulate 9 ACE2 expression. When interpreting these findings, readers should take into account several 10 limitations of our study. First, many of the ACE2 modulating conditions are discovered based on 11 data from one study with small sample sizes. These results should be viewed as data-driven 12 hypotheses rather than definitive proofs. Additional data is required to confirm the findings. 

Data preparation 2

We downloaded the uniformly processed RNA-seq data from ARCHS4 website 3 (https://amp.pharm.mssm.edu/archs4/download.html) on August 03, 2020. The downloaded 4 data include gene-level count data of 286650 samples from 9124 datasets and sample-level 5 metadata. We transformed the gene count data into percentile rank data, which reduces the 6 influences of library size, batch effects, and extreme values 41,42 . We downloaded study-level 7 metadata using the entrez_search and entrez_summary function from the rentrez library 43 . 8 9

Correlation analysis 10

We first calculated the Pearson correlation between ACE2 and other genes using data from all 11 286,650 samples. To calculate intra-study correlation, we selected a subset of studies in which 12 the variance of ACE2 is greater than 10. We then calculated the Pearson correlation within each 13 study. Transcription factors are identified by selecting genes with the Gene Ontology term 14 "DNA-binding transcription factor activity (GO:0003700)". 15 16

To identify pathways associated with ACE2, we used the pairwise correlations with ACE2 as a 17 signature. We used the signature to query the Gene Ontology Biological Process database 44 . 18

The fgsea function from the fgsea library was used to calculate the enrichment score 45 . 19 20

Metadata embedding 21

We first concatenated the metadata of each sample into a single string, including the title, tissue 22 type, and other characteristics (e.g. demographics, time points, treatment, genetic information, 23 and disease status). We then calculated the pairwise Levenshtein distance between the strings 24 that belong to the same study (GEO series). We applied multidimensional scaling to the 25 pairwise Levenshtein distance and embedded the strings into 2-dimensional space for 26 visualization and downstream analysis. 27 28 GENEVA analysis 29

For a given gene in a given dataset, we first calculated the variance of the gene (VARg). We 30 measure the overall heterogeneity of the samples by calculating the average variance of all 31 genes (VARm). We run a regression using the expression of the gene as the dependent 32

variable and the embedded metadata as independent variables (expression ~ first embed 33 dimension + second embed dimension). The regression coefficient (R 2 ) represents the 34 association between the expression of the gene and the embedded metadata. The product 35

between VARg and R 2 represents the variance of the gene explained by the embedded 36 metadata. The GENEVA score is defined as VARg × R 2 / VARm. 37 38

To test the significance of the GENEVA scores, we shuffled the samples across all datasets. 39 We then calculated the GENEVA scores of all shuffled datasets to create a null distribution. 40

Given a GENEVA score G, its p-value is defined as the probability that the null distribution is 41 greater than G: p-value = Prob(null > G). We adjust the p values for multiple testing using the 42 false discovery rate method. 43 44 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.20.20216150 doi: medRxiv preprint

We searched the gene expression omnibus using the keyword "cardiomyopathy". We then filter 2 the results to only include studies that 1) profiled the transcriptome of heart tissues from 3 humans and 2) compared cardiomyopathy samples with healthy samples. We identified 7 4 studies. We used a mixed effect model to test the effect of cardiomyopathy on ACE2 5 expression: ACE2 expression ~ study (random effect) + cardiomyopathy status (fixed effect). 6 7

To examine the ACE2 expression in different types of cardiomyopathy, we separated the 8 cardiomyopathy samples based on their subtype. We matched the cardiomyopathy samples 9

with healthy controls within the same study. We used unpaired T-tests to test the effect of each 10 cardiomyopathy type on ACE2 expression. Since data from multiple studies are available for 11 dilated cardiomyopathy (DCM), we used a mixed effect model to test the effect of DCM on 12 ACE2 expression: ACE2 expression ~ study (random effect) + cardiomyopathy status (fixed 13 effect). 14 15

Analysis of electronic health records 16

The We thank all researchers who have contributed RNA-seq datasets to the Gene Expression 1

Omnibus. We thank Dr. Alexander Lachmann, Dr. Avi Ma'ayan, and other co-authors for 2 creating ARCHS4, which enabled our study. We thank Dr. Douglas Arneson and Sanchita 3

Bhattacharya for helpful discussion. We thank the UCSF COVID-19 Data Mart for providing the 4 electronic health record data of COVID-19 patients. This work was partially supported by the 5 National Institute of Allergy and Infectious Diseases (Bioinformatics Support Contract 6 HHSN272201200028C). The content is solely the responsibility of the authors and does not 7 necessarily represent the official views of the National Institutes of Health. The analysis of the 8 electronic health record data is conducted with approval from the UCSF institutional review 9

board (IRB #: 20-31107). 10 11

12

Atul Butte is a co-founder and consultant to Personalis and NuMedii; consultant to Samsung, 13

Mango Tree Corporation, and in the recent past, 10x Genomics, Helix, Pathway Genomics, and 14

Verinata ( . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 23, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 23, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.20.20216150 doi: medRxiv preprint + ++ + + + + + + + + ++ ++ + + + + + + + + + + ++ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + ++ + ++ + + ++ + + + + + + + + + ++ + ++ ++ + + + + + + + + + + +++ + + ++ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + ++ + + + ++ + ++ ++ + + + + + + + + + ++ ++ + + ++ + + + + + + + + ++ ++ + + + + + + + + + + ++ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + ++ + + + + ++ + + +++ + ++ + + ++ + ++++ ++ + Pre-existing cardiomyopathy

Other cardiovascular diseases P = 0.043 + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 23, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 23, 2020. Stars indicate the significant differences when compared the group to the cardiomyopathy group. * P< 0.05, ** P<0.01,*** P<0.001 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted October 23, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted October 23, 2020. Table S1 : Demographic and clinical information of COVID-19 patients with cardiomyopathy and a propensity score matched cohort of COVID-19 patients with other cardiovascular diseases . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.20.20216150 doi: medRxiv preprint Table S2 : Demographic and clinical information of Non-COVID-19 patients with cardiomyopathy and a propensity score matched cohort of patients with other cardiovascular diseases . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10. 1101 /2020 

Other cardiovascular diseases (N=18000) Overall

6%) 15055 (83.6%) 16937 (83.6%) diabetes Yes 370 (16.4%) 2874 (16.0%)