key: cord-0979502-xk6rp4e7
authors: Livanos, Alexandra E.; Jha, Divya; Cossarini, Francesca; Gonzalez-Reiche, Ana S.; Tokuyama, Minami; Aydillo, Teresa; Parigi, Tommaso L.; Ladinsky, Mark S.; Ramos, Irene; Dunleavy, Katie; Lee, Brian; Dixon, Rebekah; Chen, Steven T.; Martinez-Delgado, Gustavo; Nagula, Satish; Bruce, Emily A.; Ko, Huaibin M.; Glicksberg, Benjamin S.; Nadkarni, Girish; Pujadas, Elisabet; Reidy, Jason; Naymagon, Steven; Grinspan, Ari; Ahmad, Jawad; Tankelevich, Michael; Bram, Yaron; Gordon, Ronald; Sharma, Keshav; Houldsworth, Jane; Britton, Graham J.; Chen-Liaw, Alice; Spindler, Matthew P.; Plitt, Tamar; Wang, Pei; Cerutti, Andrea; Faith, Jeremiah J.; Colombel, Jean-Frederic; Kenigsberg, Ephraim; Argmann, Carmen; Merad, Miriam; Gnjatic, Sacha; Harpaz, Noam; Danese, Silvio; Cordon-Cardo, Carlos; Rahman, Adeeb; Schwartz, Robert E.; Kumta, Nikhil A.; Aghemo, Alessio; Bjorkman, Pamela J.; Petralia, Francesca; van Bakel, Harm; Garcia-Sastre, Adolfo; Mehandru, Saurabh
title: Intestinal host response to SARS-CoV-2 infection and COVID-19 outcomes in patients with gastrointestinal symptoms
date: 2021-03-04
journal: Gastroenterology
DOI: 10.1053/j.gastro.2021.02.056
sha: 702544b0b1e9191959ba949ffdaf4b31ee7b0ace
doc_id: 979502
cord_uid: xk6rp4e7

Background and Aims Given gastrointestinal (GI) symptoms are a prominent extrapulmonary manifestation of COVID-19, we investigated intestinal infection with SARS-CoV-2, its effect on pathogenesis, and clinical significance. Methods Human intestinal biopsy tissues were obtained from COVID-19 patients (n=19) and uninfected controls (n=10) for microscopic examination, CyTOF analyses and RNA sequencing. Additionally, disease severity and mortality were examined in patients with and without GI symptoms in two large, independent cohorts of hospitalized patients in the United States (n=634) and Europe (n=287) using multivariate logistic regressions. Results COVID-19 cases and controls in the biopsy cohort were comparable for age, gender, rates of hospitalization and relevant comorbid conditions. SARS-CoV-2 was detected in small intestinal epithelial cells by immunofluorescence staining or electron microscopy, in 14 of 16 patients studied. High dimensional analyses of GI tissues revealed low levels of inflammation, including downregulation of key inflammatory genes including IFNG, CXCL8, CXCL2 and IL1B and reduced frequencies of proinflammatory dendritic cells compared with controls. Consistent with these findings, we found a significant reduction in disease severity and mortality in patients presenting with GI symptoms that was independent of gender, age, and comorbid illnesses and despite similar nasopharyngeal SARS-CoV-2 viral loads. Furthermore, there was reduced levels of key inflammatory proteins in circulation in patients with GI symptoms. Conclusion These data highlight the absence of a proinflammatory response in the GI tract despite detection of SARS-CoV-2. In parallel, reduced mortality in COVID-19 patients presenting with GI symptoms was observed. A potential role of the GI tract in attenuating SARS-CoV-2 associated inflammation needs to be further examined.

Gastrointestinal (GI) symptoms comprising nausea, vomiting, and / or diarrhea 1 are a common extrapulmonary manifestation in Coronavirus disease 2019 . Additionally, the presence of GI involvement by severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2) has also been suggested by clinical 2 , non-human primate 3 and in vitro 4, 5 data. However, to date, there is limited evidence of SARS-CoV-2 infection of human intestinal epithelial cells 6 and there are no studies on the response of the GI immune system in COVID-19 patients.

Given the immune dysregulation seen in COVID- 19 7, 8 , we aimed to document infection of the GI tract in patients with COVID-19, to define the cellular and transcriptomic changes within the GI tract, and to determine the impact of GI symptoms on COVID-19 outcomes. Here, we present findings from well-characterized cohorts of COVID-19 patients hospitalized in tertiary care centers, from both New York City, USA and Milan, Italy, where we conducted high dimensional analyses of mucosal and systemic immune parameters and investigated disease outcomes associated with GI involvement in COVID-19 patients.

Endoscopic biopsies were obtained from 20 COVID-19 and 10 control patients undergoing clinically indicated endoscopic procedures after informed consent with Mount Sinai Hospital (MSH) IRB approved protocol (IRB . The demographic characteristics of these patients J o u r n a l P r e -p r o o f and controls are provided in Supplementary Table 1 and 2. COVID-19 severity is defined in   Supplementary Table 3 and Supplementary Methods.

634 subjects with COVID-19, admitted to MSH between April 1, 2020 and April 15, 2020, who met study inclusion criteria were enrolled in a Discovery Cohort under an IRB approved protocol (IRB-20-03297A) (Supplementary Methods).

We analyzed a cohort of 287 patients admitted to a tertiary care center in Milan, Italy between 

A distinct 'Internal Validation Cohort' of patients who were hospitalized at MSH between April 16, 2020 and April 30, 2020 (Supplementary Methods) was analyzed using a predictive model.

Formalin fixed, paraffin embedded tissue was analyzed (Supplementary Methods). Primary and secondary antibodies are summarized in Supplementary Table 16 .

Biopsy specimens and infected Vero E6 cells (positive control) were examined by electron RNA-sequencing (RNA-seq) was performed on RNA isolated from the EC and LP samples obtained from COVID-19 cases and controls (Supplementary Methods).

For univariable statistical analyses, Graph Pad Prism (version 8) was used to calculate unpaired two tailed t-test for continuous variables and either Fisher's exact test or the Chi-square test for categorical variables.

A multivariate logistic regression was utilized to model each outcome as function of GI symptoms and clinical variables including age, gender, body mass index (BMI) and comorbidities. Significant associations were determined based on 95% confidence interval (CI) based on 1000 bootstrap iterations (Supplementary Methods).

Only age and BMI were adjusted for, since they were the only variables significantly associated with both outcomes across different GI symptoms models in the Discovery Cohort (Supplementary Table 9 ). Then, the estimated model was utilized to predict the outcome of patients in the Internal Validation Cohort.

Likelihood Estimation) package available in R Cran 10 .

SARS-CoV-2 viral loads were determined as previously reported 11 (Supplementary Methods).

The ELLA cytokine platform measured TNF-α, IL-6, IL-8, and IL-1β 8 . Unpaired two-tailed ttests were used to compare individual cytokines quantified by the ELLA panel between GI symptomatic and asymptomatic groups. P-values were adjusted via Benjamini-Hochberg 12 .

Multiplexed proteomic inflammation panel (Olink, 92 inflammation-related proteins) was used to quantify circulating cytokines using an antibody-mediated proximity extension-based assay.

The Benjamini-Hochberg procedure was used to adjust P values for multiple testing.

Consensus clustering was performed on the abundance of the 92 cytokines across all 238 samples using the R package ConsensusClusterPlus 13 . Associations between GI symptoms and Olink proteomic data were derived using unpaired t-test comparing the symptomatic and asymptomatic groups. P-values were adjusted via Benjamini-Hochberg (10% FDR threshold of significance). 

Robust expression of ACE2 was noted on the small intestinal brush border in both controls and There were a few cells positive for the viral nucleocapsid protein but negative for MUC2 which tended to be located at the base of the crypts (Figure 2 , P to Q). The more diffuse viral antigen staining in the ileum as compared to the duodenum is not explained by apparent differences in ACE-2 protein expression ( Figure 2 , A to D), however, may be explained by increased goblet cells in the ileum 15 and this data appears to be consistent with organoid cultures 4 . As negative controls, 5 duodenal and 6 ileal biopsies from 10 patients collected prior to the pandemic (Supplementary Table 5 

Next, we performed TEM in 16 patients. Eight of these patients showed presence of 70-110 nm viral particles in the intestinal epithelial cells of the duodenum and/or ileum by TEM (Supplementary Table 4 

J o u r n a l P r e -p r o o f

We inoculated Vero E6 cells with the supernatants of homogenized intestinal tissues, but did not observe any apparent cytopathic effects or plaque formation after 7-days culture. In addition, cell culture supernatants did not reveal the presence of viral RNA by RT-qPCR.

Next, we performed mass cytometry (CyTOF) based immunophenotypic analysis on GI tissue and peripheral blood from a subset of COVID-19 cases (GI tissue, n = 13; blood, n = 10) and controls (GI tissue, n = 9; blood, n = 9) (Supplementary Table 1 In the LP, CD206 + CD1c + cDC2 (conventional DCs-0.4-fold decrease, p=0.01) and

plasmacytoid DCs (pDCs) were reduced in COVID-19 cases (0.5 fold decrease, p=0.07) ( Figure   3 , D and E), analogous to changes described in the blood 16 . Effector (PD-1 + CD38 + ) CD4 + and CD8 + T cells ( Figure 3F ) as well as CD8 + CD103 + T cells (tissue resident memory) (Supplementary Figure 7A) Among PBMCs, effector (PD-1 + CD38 + ) CD4+ and CD8+ T cells were significantly increased in patients ( Figure 3I ). Alterations in monocytes, T REG and IgG + plasma cells are shown in Supplementary Figure 9 . Finally, a significant increase in activated (CD29 + CD38 + ) Figure 10A) and a non-significant increase of these activated T cells in the LP of patients compared to controls (Supplementary Figure 10B) .

Details of all immune population changes are provided in Data file S2.

Altogether, intestinal tissues of COVID-19 patients showed altered distribution of immune cell subsets, most notable for reduced frequencies of CD206 + CD1c + cDC2 and pDCs and an increased frequency of effector T cells.

Next, we performed RNA-Seq on the EC and LP in 13 COVID-19 patients and 8 controls. The Figure 4B ).

We considered the possibility that the observed expression changes could imply alterations in relative cell type proportions (in addition to transcriptional alterations within cells).

Therefore, we interrogated data derived from single-cell RNA-seq 18 Key inflammatory genes including IFNG, IL1B, CXCR4, TNFSF14, CXCL2,   CSF-1, CXCL8, IL18R1, NRP1 and IL18BP were downregulated in LP of patients compared to controls ( Figure 4F ).

Together, these data reveal a dynamic remodeling of GI tissues by SARS-CoV-2, notably with a significant downregulation of pathways associated with inflammation and antigen presentation in the LP with a concomitant activation of antiviral response signaling genes in the EC.

Given the observed downregulation of key inflammatory genes, we hypothesized that intestinal involvement in COVID-19 is associated with a milder disease course. We tested this hypothesis Table 6 ).

J o u r n a l P r e -p r o o f Table 7 ). These findings were further emphasized by Kaplan-Meier estimates of survival over short-term follow-up of 25 days (p<0.001 log-rank test) ( Figure 5A and Supplementary Figure   14 , A and B). Consistent with prior reports 8 older age and higher disease severity were associated with higher mortality (Supplementary Table 8 ).

Next, we created a multivariate model, adjusting for age, BMI, gender, race/ethnicity, diabetes, HTN, chronic lung disease and heart disease to determine the impact of GI symptoms on COVID-19 outcomes (Table 1) . Consistent with published literature 20 age and BMI were positively associated with COVID-19 severity and mortality (Supplementary Table 9 ). The presence of any GI symptoms, as well as diarrhea, nausea, and vomiting individually, were inversely associated with COVID-19 severity and mortality ( Figure 5B , Supplementary Table 9 ).

Patients who presented with GI symptoms had 50% reduced odds of having severe disease (odds ratio (OR) of 0.56) and death from COVID-19 (OR of 0.54), compared to the patients who presented without GI symptoms ( Figure 5B , Supplementary Table 9 ).

Next, we confirmed our findings in an External Validation Cohort in which GI symptoms on admission were characterized as presence or absence of diarrhea (Supplementary Figure 5C ). In 270 patients in which treatment data was available, no specific treatment was associated with GI symptoms (p-values > 0.05) (Supplementary Table 11 ). In addition, diarrhea was significantly associated with mortality after adjusting for all treatments (Supplementary Table 11 ). Thus, our observations from this External Validation Cohort were in alignment with those from the Discovery Cohort. 

Using causal inference methodology, we quantified the ATE of GI symptoms on COVID-19 outcomes while accounting for potential confounders. We performed this analysis on the MSH Given recent reports suggesting that NP SARS-CoV-2 viral loads are correlated with disease outcomes 11 , we compared NP viral loads in a subset of Discovery and Internal Validation Cohort (n=329, where data available). Patients with and without GI symptoms had comparable SARS-CoV-2 NP viral loads (mean log 10 copies/mL 5.1 (SD 2.3) and 5.6 (SD 2.4), respectively) (p=0.07); furthermore, no significant differences were observed for each individual GI symptom ( Figure 5F ).

To correlate the observed mortality difference with GI symptoms with known biomarkers for severe COVID-19, we examined IL-6, IL-8, TNF-α, and IL-1β levels measured on admission.

IL-6 and IL-8, which are known to be directly associated with poor survival 8 , were found to be significantly reduced in circulation of patients with GI symptoms (FDR 10%) (Supplementary Cluster 4 (Fisher's exact test 10% FDR). These pathways were downregulated in patients with diarrhea (p<0.05 from t-test) ( Figure 6B ), suggesting a reduced inflammatory response in patients with GI symptoms. Additionally, Clusters 1, 2, 3, 5 and 6 were significantly downregulated in patients with GI symptoms compared to those without (FDR 15%) ( Figure   6C ). This seemed to be driven mostly by diarrhea since the same clusters were significantly downregulated in patients with diarrhea (FDR 10%). We observed a similar, albeit a reduced signal for nausea and vomiting likely due to the smaller sample size (n=29 for vomiting, n=54

for nausea). Table 15 ).

Overall, GI symptoms are associated with significantly reduced levels of key inflammatory cytokines including IL-6, IL-8, IL-17 and CCL28 that are known to be associated with poor COVID-19 outcomes.

Given the robust expression of ACE2 on the small intestinal epithelium, we hypothesized that the intestines would be susceptible to SARS-CoV-2 infection. Here, we detailed for the first In two distinct and large cohorts of COVID-19 patients, we observed a significant reduction in mortality in patients presenting with GI symptoms compared to those without GI symptoms, even after adjusting for multiple confounders including age and comorbidities, which is consistent with findings in two smaller cohorts 30, 31 . Notably this finding is different from early reports suggesting increased severity with GI symptoms 32 , likely attributable to the inclusion of abnormal liver function tests which are associated with poor outcomes.

We duly acknowledge some limitations of our study. GI biopsies were performed on a distinct set of patients undergoing clinically indicated procedures and therefore, they were not all in the acute phase of illness. Furthermore, given only 3 patients in the biopsy cohort had GI In summary, our data detail the previously unappreciated GI tissue response to SARS-CoV-2 and provide the rationale for future mechanistic studies to understand a possible attenuation of SARS-CoV-2 pathogenicity by the intestinal environment. Table 3 ).

Patients A total of 634 subjects were included in the Discovery Cohort (Supplementary Figure 11) . In addition to demographic information (including race and ethnicity, age and gender), clinical characteristics, laboratory data and outcomes data was extracted from the medical charts. Co-variates that were studied included: BMI (obesity defined as BMI >30) and comorbid conditions including, hypertension, diabetes, chronic lung disease (including asthma and COPD), heart disease (including coronary artery disease, atrial fibrillation and heart failure), chronic kidney disease, cancer, HIV, and inflammatory bowel disease (IBD).

GI symptoms were defined as more than one episode of either diarrhea, nausea, and/or vomiting at the time of admission. If only one episode of either diarrhea, nausea, and/or vomiting was specifically documented, patients were not considered to have GI symptoms. Additionally, we did not consider GI symptoms that developed during the course of hospitalization, as they could reflect nosocomial or treatment-related effects and only considered the GI symptoms that were present at the time of hospital admission so as to avoid including iatrogenic confounders (treatments or hospital acquired illnesses that can result in diarrhea, nausea and vomiting).

Disease severity (as described above) and mortality were considered as outcomes variables.

Mortality was calculated as patient status (dead or alive) at 25 days post admission. If no information was available after discharge, patients were censored at the time of hospital discharge.

This cohort consisted of 287 patients admitted to a tertiary care center in Milan, Italy between 

The 

The 

Sections (5µm) of formalin fixed, paraffin embedded tissue were dewaxed in xylene and rehydrated in graded alcohol and then washed in phosphate-buffered saline (PBS). Heat-induced epitope retrieval was performed by incubating slides in a pressure cooker for 15 minutes on high in target retrieval solution (Dako, S1699). Once slides cooled to room temperature, they were 

Three-10x non-overlapping IF images were taken for each biopsy. Twelve biopsies (10 duodenum, 2 ileum) from 11 COVID-19 patients in the biopsy cohort were analyzed along with 9 uninfected controls (5 duodenum, 4 ileum). CD3+ intraepithelial lymphocytes (IELs) and CD3+ CD8+ (IELs) were quantified for each image. The length of epithelium in each image was measured in ImageJ 2 . Biopsies from COVID-19 patients and controls were compared via unpaired t test.

Following post-fixation in 1% osmium tetroxide, tissues were serially dehydrated and embedded in epoxy resin in standard fashion. One-micron toluidine-stained scout sections were prepared for light microscopic orientation; 80 nm ultrathin sections for EM were stained with uranyl acetate and lead citrate and examined in a Hitachi 7650 transmission electron microscope at 80kV. #15140-122). Cells were grown in a humidified incubator at 37ºC with 5% CO2. Vero E6 cells seeded in six well dishes and infected with SARS-CoV-2 at a multiplicity of infection of 0.01 for 48 hours before fixing and preparing for electron microscopy. Cells were pre-fixed with 3% glutaraldehyde, 1% paraformaldehyde, 5% sucrose in 0.1M sodium cacodylate trihydrate, removed from the plates and further prepared by high-pressure freezing and freeze-substitution as described below.

Tissue samples were fixed with 3% glutaraldehyde to meet biosafety requirements. Images were recorded with a 2k x 2k CCD camera (XP1000; Gatan, Pleasonton, CA).

Tomographic tilt series and large-area montages were acquired automatically using the SerialEM software package 3 . For dual-axis tomography, images were collected at 1° intervals as samples were tilted +/-62°. The grid was then rotated 90° and a second tilt-series was acquired about the orthogonal axis. Tomograms were calculated, analyzed and modeled using the IMOD software We distinguished virions inside a cytoplasmic exit compartment from the inner vesicles of an MVB based on differences in size (MVB inner virions are generally smaller in diameter than coronaviruses) and the presence of surface spikes and internal puncta (MVB inner vesicles do not present surface spikes or internal puncta).

African green monkey kidney epithelial cells (Vero E6) were originally purchased from 

Biopsies were transferred to 10 Cryovials were immediately transferred to -80C until the sample was acquired for mass cytometry as detailed below.

Briefly, 15ml of Lymphosep -Lymphocyte Separation Medium (MP Bio.) was added to each 50 ml centrifugation tube. Blood was diluted with PBS to bring the volume up to 30ml and diluted blood was layered gently over Lymphosep. Tubes were then centrifuged at 2000 rpm for 20 mins with the brakes and acceleration off. After centrifugation, the buffy coat containing

PBMCs was transferred to another tube and was centrifuged at 1800 rpm to pellet the cells.

Pellets were resuspended in PBS and tubes were centrifuged at 1800 rpm. Finally, the pellets were resuspended in the freezing medium (10% DMSO + 44% FBS in RPMI) and cryopreserved at -80 °C.

Cells were processed as previously described by Geanon et al 13 Devneg. It is already been described that SmartTube-based fixation protocols take into account previously described mass cytometry artifacts such as cell-cell multiplets, isotopic spillover or oxidation, or mass cytometer instrument configuration 13 .

Pre-gated viable CD45+ cells were first clustered and annotated using the Astrolabe Cytometry 

Base-calling and quality scoring of sequencing data were done through Illumina's Real-Time Analysis (RTA) software. RNA-seq data processing and reference mapping were done with custom analysis scripts combining publicly available tools as previously described 15 with modifications as follows, reads were mapped to a custom reference that combined the human hg38 reference genome (Release 34, GRCh38.p13) and the SARS-CoV-2 genome (RefSeq NC_045512) for simultaneous quantification of host and virus transcripts.

Differential gene expression (DGE) analysis was performed with the Bioconductor edgeR package 16 using as input a combined matrix of mapped paired-end read raw counts, with genes in rows and samples in columns. Prior to DGE analysis, gene counts were converted to fragments per kb per million reads (FPKM) with the RSEM package with default settings in strand-specific mode 17 .

Genes with less than 1 FPKM in at least 50% of the samples were removed. The remaining gene counts were then normalized across samples using the weighted trimmed mean of M-values (TMM) method 18 . The dispersion was estimated by fitting a generalized linear model (GLM) as implemented in edgeR, sex was fitted as a covariate on a per-patient paired design. Pairwise comparisons were performed between sample groups (i.e., between tissue sections, and between cases and controls). Significant expression differences were selected based on eBayes adjusted p values corrected for multiple testing using the Benjamini-Hochberg method (q ≤ 0.05).

KEGG pathway and gene ontology (GO) biological process (BP), molecular function (MF), and/ or cellular component (CC) enrichment analyses were performed using the gProfileR R v0.6.8 package 19 . The background gene set was restricted genes with detected expression (defined as genes with expression levels above 1 FPKM in at least 50% of samples). Genes with differential expression were ranked by log 2 fold change and used as an ordered query. P values were corrected using the g:SCS algorithm to account for multiple comparisons. , were tested for significant (p ≤ 0.05) enrichment using Fisher's exact tests and using Bonferroni correction for multiple comparisons.

Additionally, GSEA 23 was carried out on a rank ordered list of the infected EC versus control molecular analysis. The ranking metric used was logFC * -logP value, however, the results were similar when logFC metric was also used (data not shown). For the COVID-19 associated datasets, we curated two signatures from infected organoids 24 : hSIOs-COVID-19:

human small intestinal organoids (hSIOs) grown in either i) Wnt high expansion (EXP) medium (at adjP<0.05) or ii) differentiation (DIF) medium (at adjP<0.1). The standard GSEA settings were used, namely 'meandiv' for normalization mode, 'weighted' enrichment statistic, and '1000' permutations. GSEA using the Hallmark database (v7.1 25 ) was also performed with the same settings.

For this analysis, we considered 570 patients with clinical descriptors including as age, gender, race/ethnicity, BMI, comorbidities (including hypertension, diabetes, chronic lung disease (including asthma and COPD), heart disease (including coronary artery disease, atrial fibrillation and heart failure), and GI symptoms. A multivariate logistic regression was utilized to model severity and mortality as function of each of the GI symptoms and clinical variables including race, age, gender, BMI, heart and lung diseases and hypertension.

In particular, race was stratified as White (Caucasian), Black (African-American),

Hispanic and others; lung disease was set equal to 1 if the patient was either affected by COPD or asthma and zero otherwise; heart disease was set equal to 1 if the patient was either affected by coronary artery disease, atrial fibrillation or heart failure and 0 otherwise. The severity indicator was set equal to 1 for severe and severe with EOD patients and 0 for mild and moderate COVID-19 patients; mortality was set equal to 1 for deceased patients and 0 otherwise.

Significant association were determined based on 95% confidence interval (CI) based on 1000 bootstrap iterations. At each bootstrap iteration, patients were sampled with replacements and logistic regressions were estimated considering as outcome severity and mortality. Then, 95% CI of coefficients and odds ratio were estimated across bootstrap iterations.

For this analysis, we considered 228 patients with clinical data such as age, gender and GI symptoms as described in Aghemo et al 1 . A multivariate logistic regression was utilized to model mortality, ICU admission and the composite outcome of ICU admission or mortality as function of presence or absence of diarrhea and clinical variables including age, gender, BMI, heart disease, COPD, diabetes and hypertension. Heart disease was set equal to 1 if the patient was either affected by coronary artery disease or atrial fibrillation and 0 otherwise. CI of odds ratio

were computed based on 1000 bootstrap iterations as above.

In 270 patients from this cohort treatment data was available. Treatments included hydroxychloroquine, antiviral treatments including lopinavir-ritonavir and darunavir-cobicistat, tocilizumab, steroids, antibiotics including ceftriaxone, azithromycin, piperacillin-tazobactam, statins, angiotensin-converting-enzyme (ACE) inhibitors and angiotensin II receptor blockers (ARBs). Using this data, we performed fisher's exact test to determine whether any treatments were associated with diarrhea. Additionally, we computed 95% CI of odds ratio based on 1000 bootstrap iterations.

For this analysis, we considered 233 patients with clinical data including age, BMI, and GI symptoms. In order to evaluate the predictive performance of each model, bootstrapping was performed. Specifically, at each bootstrap iteration, we randomly sampled patients in the Discovery Cohort with replacement and estimated a logistic regression to model each outcome as function of a particular GI symptom, age and BMI. In this analysis, only age and BMI were adjusted for since they were the only variables significantly associated with both outcomes across different GI symptoms models in the Discovery Cohort ( Figure 5B , Supplementary Table   9 ). Then, the estimated model was utilized to predict the outcome of patients in the Internal Validation Cohort. This procedure was repeated for 1000 bootstrap iterations. For each iteration,

Receiving Operating Characteristic (ROC) curve and area under the curve (AUC) were computed. For comparison purposes, the distribution of AUC across 1000 bootstrap iterations from the predictive model based on age and BMI only was considered. Figure 5D shows for 1000 bootstrap iterations. Following the strategy above, at each bootstrap iteration, patients were sampled with replacement. Figure 5E shows the 95% confidence intervals of difference in AUC 27 . For patients with multiple NP swabs available, the first swab was used for analysis.

The ELLA platform is a method for rapid cytokine measurement using microfluidics ELISA assays. The assay measured TNF-α, IL-6, IL-8, and IL-1β, previously validated by the Mount Sinai Human Immune Monitoring Center (HIMC) using plasma from multiple myeloma patients and recently reported for large cohort of COVID-19 patients admitted to MSH.

For analysis of circulating cytokines, we used a multiplexed proteomic inflammation panel (Olink), which consists of 92 inflammation-related proteins quantified by an antibody-mediated proximity extension-based assay. Samples with normalized protein expression values below the limit-of-detection in >75% of samples were excluded from further analysis. For the remainder of analytes, any sample under the limit of detection was assigned a value of the limit-of-detection divided by the square root of 2. The log2 fold-change over the median healthy control protein expression was then calculated, and the Benjamini-Hochberg procedure was used to adjust P values for multiple testing.

For this analysis, we considered 238 samples with GI symptoms annotation. Consensus clustering was performed based on the abundance of 92 cytokines across all 238 samples.

Consensus clustering was performed using the R packages ConsensusClusterPlus based on zscore normalized data. Specifically, markers were partitioned into six clusters using the K-means algorithm, which was repeated 1000 times. Then, markers in each cluster were considered in order to derive cluster z-score signatures via package GSVA. Based on these signatures, the association between different clusters and GI symptoms were derived via logistic regression with outcome corresponding to each GI symptom. Figure 6C shows the signed FDR (-log10 scale). Pvalues were adjusted via Benjamini-Hochberg. The pathway analysis for the clusters described above was carried out considering the entire KEGG and HALLMARK databases. 

Extrapulmonary manifestations of COVID-19

AGA Institute Rapid Review of the GI and Liver Manifestations of COVID-19, Meta-Analysis of International Data, and Recommendations for the Consultative Management of Patients with COVID-19

Respiratory disease in rhesus macaques inoculated with SARS-CoV-2

TMPRSS2 and TMPRSS4 promote SARS-CoV-2 infection of human small intestinal enterocytes

SARS-CoV-2 productively infects human gut enterocytes

Evidence for Gastrointestinal Infection of SARS-CoV-2

Imbalanced Host Response to SARS-CoV-2 Drives Development of COVID-19

An inflammatory cytokine signature helps predict COVID-19 severity and death

A Streamlined CyTOF Workflow To Facilitate Standardized Multi-Site Immune Profiling of COVID-19 Patients

tmle: An R Package for Targeted Maximum Likelihood Estimation

SARS-CoV-2 viral load predicts COVID-19 mortality

Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing

ConsensusClusterPlus: a class discovery tool with confidence assessments and item tracking

Localization of mucin (MUC2 and MUC3) messenger RNA and peptide expression in human normal intestine and colon cancer

The intestinal stem cell

A dynamic COVID-19 immune signature includes associations with poor prognosis

SARS-CoV-2 Cell Entry Depends on ACE2 and TMPRSS2 and Is Blocked by a Clinically Proven Protease Inhibitor

Intra-and Inter-cellular Rewiring of the Human Colon during Ulcerative Colitis

A novel role for constitutively expressed epithelial-derived chemokines as antibacterial peptides in the intestinal mucosa

The importance of overweight in COVID-19: A retrospective analysis in a single center of Wuhan

Following TRAIL's path in the immune system

Harnessing the biology of IL-7 for therapeutic application

Histopathologic Changes and SARS-CoV-2 Immunostaining in the Lung of a Patient With COVID-19

Single-cell landscape of bronchoalveolar immune cells in patients with COVID-19

Longitudinal analyses reveal immunological misfiring in severe COVID-19

Enteric involvement of severe acute respiratory syndrome-associated coronavirus infection

COVID-19 autopsies: Procedure, technical aspects and cause of fatal course. Experiences from a single-center

Pathogenesis and transmission of SARS-CoV-2 in golden hamsters

Susceptibility of ferrets, cats, dogs, and other domesticated animals to SARS-coronavirus 2

Gastrointestinal Symptoms and Coronavirus Disease 2019: A Case-Control Study From the United States

COVID-19 Digestive System Involvement and Clinical Outcomes in a Large Academic Hospital in

Manifestations and prognosis of gastrointestinal and liver involvement in patients with COVID-19: a systematic review and meta-analysis

Immunoglobulin responses at the mucosal interface

The regulation of IgA class switching

Enhanced SARS-CoV-2 Neutralization by Secretory IgA in vitro

Mosaic nanoparticles elicit crossreactive immune responses to zoonotic coronaviruses in mice

COVID-19 Digestive System Involvement and Clinical Outcomes in a Large Academic Hospital in

NIH Image to ImageJ: 25 years of image analysis

Automated electron microscope tomography using robust prediction of specimen movements

Automated tilt series alignment and tomographic reconstruction in IMOD

Correction for non-perpendicularity of beam and tilt axis in tomographic reconstructions with the IMOD package

Molecular Architecture of the SARS-CoV-2 Virus

Structures and distributions of SARS-CoV-2 spike proteins on intact virions

SARS-CoV-2 structure and replication characterized by in situ cryo-electron tomography

In situ structural analysis of SARS-CoV-2 spike reveals flexibility mediated by three hinges

Multivesicular bodies mimicking SARS-CoV-2 in patients without COVID-19

FcRn-mediated antibody transport across epithelial cells revealed by electron tomography

Electron tomography of late stages of FcRn-mediated antibody transcytosis in neonatal rat small intestine

A Streamlined CyTOF Workflow To Facilitate Standardized Multi-Site Immune Profiling of COVID-19 Patients

Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing

Integrated Transcriptome and Network Analysis Reveals Spatiotemporal Dynamics of Calvarial Suturogenesis

edgeR: a Bioconductor package for differential expression analysis of digital gene expression data

RSEM: accurate transcript quantification from RNA-Seq data with or without a reference genome

A scaling normalization method for differential expression analysis of RNA-seq data

Profiler--a web-based toolset for functional profiling of gene lists from large-scale experiments

Intra-and Inter-cellular Rewiring of the Human Colon during Ulcerative Colitis

Single-Cell Analysis of Crohn's Disease Lesions Identifies a Pathogenic Cellular Module Associated with Resistance to Anti-TNF Therapy

Imbalanced Host Response to SARS-CoV-2 Drives Development of COVID-19

Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles

SARS-CoV-2 productively infects human gut enterocytes

The Molecular Signatures Database (MSigDB) hallmark gene set collection

tmle: An R Package for Targeted Maximum Likelihood Estimation

SARS-CoV-2 viral load predicts COVID-19 mortality

We would like to thank the clinical staff, physicians and patients who participated in this study.This research was partly funded by NIH/NIDDK123749 0S1 (S.M.). Additional support was J o u r n a l P r e -p r o o f