key: cord-0895669-8yo5wyx6
authors: Reiterer, Moritz; Rajan, Mangala; Gómez-Banoy, Nicolás; Lau, Jennifer D.; Gomez-Escobar, Luis G.; Ma, Lunkun; Gilani, Ankit; Alvarez-Mulett, Sergio; Sholle, Evan T.; Chandar, Vasuretha; Bram, Yaron; Hoffman, Katherine; Bhardwaj, Priya; Piloco, Phoebe; Rubio-Navarro, Alfonso; Uhl, Skyler; Carrau, Lucia; Houhgton, Sean; Redmond, David; Shukla, Alpana P.; Goyal, Parag; Brown, Kristy A.; tenOever, Benjamin R.; Alonso, Laura C.; Schwartz, Robert E.; Schenck, Edward J.; Safford, Monika M.; Lo, James C.
title: Hyperglycemia in Acute COVID-19 is Characterized by Insulin Resistance and Adipose Tissue Infectivity by SARS-CoV-2
date: 2021-09-16
journal: Cell Metab
DOI: 10.1016/j.cmet.2021.09.009
sha: 4ae89e493d69fe11acec38f4a48ff22d4b4c3eec
doc_id: 895669
cord_uid: 8yo5wyx6

Individuals infected with SARS-CoV-2 who also display hyperglycemia suffer from longer hospital stays, higher risk of developing acute respiratory distress syndrome (ARDS), and increased mortality. Nevertheless, the pathophysiological mechanism of hyperglycemia in COVID-19 remains poorly characterized. Here, we show that hyperglycemia is similarly prevalent among patients with ARDS independent of COVID-19 status. Yet, among patients with ARDS and COVID-19, insulin resistance is the prevalent cause of hyperglycemia, independent of glucocorticoid treatment, which is unlike patients with ARDS but without COVID-19, where pancreatic beta cell failure predominates. A screen of glucoregulatory hormones revealed lower levels of adiponectin in patients with COVID-19. Hamsters infected with SARS-CoV-2 demonstrated a strong antiviral gene expression program in the adipose tissue and diminished expression of adiponectin. Moreover, we show that SARS-CoV-2 can infect adipocytes. Together these data suggest that SARS-CoV-2 may trigger adipose tissue dysfunction to drive insulin resistance and adverse outcomes in acute COVID-19.

The deadly coronavirus disease 2019 pandemic is underscored by the high 2 morbidity and mortality rates seen in certain vulnerable populations, including individuals 3 with diabetes mellitus (DM), obesity, cardiovascular disease and advanced age, with the latter 4 associated with many chronic cardiometabolic diseases (Drucker, 2021; Holman et al., 2020; 5 McGurnaghan et al., 2021; Yang et al., 2021; Zhou et al., 2020) . Hyperglycemia with or without 6 a history of DM is a strong predictor of in-hospital adverse outcomes, portending a 7-fold 7 higher mortality compared to patients with well-controlled blood glucose levels (Zhu et al., 8 2020). Hyperglycemia may be seen as a biomarker that predicts poor prognosis. A 9 retrospective study that compared patients with hyperglycemia that were treated with insulin 10 against those who were not showed increased mortality in those receiving insulin (Yu et al., 11 2021). However, it remains unclear whether insulin treatment is a surrogate for severity of 12 hyperglycemia and overall morbidity, or whether it is an actual causative factor for death. 13 There is thus uncertainty regarding specific treatments for hyperglycemia in acute COVID-19 14 (Lim et al., 2021) . 15 Despite our early recognition of the association between hyperglycemia and perilous 16 outcomes, the pathophysiological mechanisms that underlie hyperglycemia in COVID-19 17 remain undefined (Accili, 2021; Lockhart and O'Rahilly, 2020) . Hypotheses have included a 18 broad range of pathologies, such as direct infection of islets leading to beta cell failure (BCF), 19 and systemic inflammation leading to insulin resistance (IR). Dexamethasone substantially 20 reduces mortality in patients with severe COVID-19 infection requiring oxygen or invasive 21 mechanical ventilation (The RECOVERY Collaborative Group, 2021). Glucocorticoids can also 22 provoke hyperglycemia by inducing insulin resistance and beta cell dysfunction. The 23 J o u r n a l P r e -p r o o f widespread usage of dexamethasone in severe SARS-CoV-2 infection is expected to 1 exacerbate both the incidence and severity of hyperglycemia in COVID-19. However, the 2 contribution of glucocorticoids to hyperglycemia in acute COVID-19 has not been defined. 3 Although COVID-19 is primarily a marked by a respiratory tract infection, severe acute 4 respiratory syndrome coronavirus 2 (SARS-CoV-2) is known to infect other cell types and often 5 leads to extrapulmonary consequences (Gupta et al., 2020; Puelles et al., 2020) . For example, 6 ACE2 and other entry receptors for SARS-CoV-2 can be expressed on pancreatic islet cells, and 7 endocrine cells differentiated from human pluripotent stem cells are permissive to infection 8 (Tang et al., 2021; Wu et al., 2021; Yang et al., 2020) . Early reports of unexpected diabetic 9 ketoacidosis (DKA) in COVID-19 patients fueled concerns for a novel form of acute onset beta 10 cell failure. For example, one case described a patient with new onset diabetic ketoacidosis 11 (DKA) who was found to be autoantibody negative for type 1 DM (T1DM) but showed 12 evidence of prior SARS-CoV-2 infection based on serology results, suggesting the possibility 13 of pancreatic beta cell dysfunction or destruction as a result of COVID-19 (Hollstein et al., 14 2020). However, given the high rates of COVID-19 during this pandemic coupled with low 15 background rates of new onset T1DM, the connection between these two events in this case 16 could be "true, true, and unrelated." Recent studies disagree on whether ACE2 is expressed 17 on pancreatic beta cells or whether the SARS-CoV-2 virus is found in pancreatic beta cells of 18 deceased individuals with COVID-19 ( Wu et al., 2021) . Conversely, the well-known connection between 20 obesity and insulin resistance might lead to impaired immunity and more severe SARS-CoV-2 21 infection (Goyal et al., 2020a ). In fact, population level studies have reported a higher risk of 22 complications in patients with obesity and COVID-19 (Barron et al., 2020 ; Docherty et al., 23 worsened hyperglycemia. In sum, despite much attention, the pathophysiology of 1 hyperglycemia observed in acute COVID-19 remains unknown. 2 In this study, we assessed the pathophysiological mechanism of hyperglycemia in acute and 3 severe COVID-19 and analyzed protein hormones regulating glucose homeostasis. We 4 compared patients with COVID-19 to critically ill control patient groups (those with ARDS and 5 hyperglycemia but without and found striking differences in the characteristics 6 associated with hyperglycemia, further highlighting the metabolic dysfunction seen in this 7 disease. 8 

Hyperglycemia in patients with COVID-19 is associated with adverse outcomes. 10 Our study consisted of patients hospitalized at New York-Presbyterian Hospital/Weill Cornell 11 Medical Center and affiliated campuses at Queens and Lower Manhattan Hospital. We (Table 1) , rising to 91.1% and 72.8% among intubated and deceased patients, 16 respectively. Hyperglycemia was thus associated with an increased risk for intubation by more 17 than 15-fold (OR 15.6, Fisher's exact test P < 0.0001) and for death by 3.6-fold (OR 3.6, P < 18 0.0001) compared to non-hyperglycemic patients ( Figure 1A) . Similarly, the median length of 19 hospital stay was more than 2-fold longer for hyperglycemic patients at 10 days compared to 20 5 days for non-hyperglycemic patients (Wilcoxon rank-sum test, P < 0.0001) ( Table 1) . 21 J o u r n a l P r e -p r o o f Patients severely infected with SARS-CoV-2 develop pneumonia, ARDS and require 1 mechanical ventilation (Magleby et al., 2020) . Hyperglycemia increased the risk of developing 2 ARDS by 9.3-fold (OR 9.3, P < 0.0001) for COVID-19 patients and was considerably more 3 prevalent than in patients without ARDS (86.0% compared to 39.8%) ( Figure 1A , Table 1 ). In 4 this study we aimed to further analyze the impact of hyperglycemia on critically ill patients 5 with COVID-19; we thus divided our cohort based on the development of ARDS. We 6 compared the characteristics of these 823 critically ill patients who were ARDS-positive and 7 who had COVID-19 (Table 1) (Table 2 ). Strikingly, a very similar 9 pattern was found among COVID-19-negative control ICU patients, 86% of ARDS-positive 10 controls suffered from hyperglycemia, compared to 63% of ARDS-negative controls ( Figure 1B ). Although the three patient groups differed in the prevalence of obesity and DM, 12 both the overall prevalence and the magnitude of differences were smaller ( Figures 1C,D) . 13 Among both ARDS-positive and -negative ICU controls, hyperglycemia was associated with an 14 increased risk of mortality (OR 2.4, P < 0.0001 and OR 2.5, P < 0.0001), but among patients 15 with ARDS and COVID-19 this was not the case (OR 0.7, P = 0.06) ( Figure 1E ). Median hospital 16 stay of patients with hyperglycemia was longer than for patients without hyperglycemia in all 17 three groups ( Figure 1F ). This prolonged hospitalization effect was most pronounced among 18 patients with ARDS and COVID-19, where the median hospital stay of a patient with 19 hyperglycemia was more than three times longer than for individuals without hyperglycemia 20 with the same condition. For ARDS-positive and ARDS-negative control patients, 21 hyperglycemia was associated with a significantly smaller increase in median hospital stay 22 (36% and 57%, respectively, Poisson regression model, P < 0.0001). 23 J o u r n a l P r e -p r o o f Hyperglycemia in COVID-19 is predominantly associated with insulin resistance and reduced 1 adiponectin levels 2 To determine the mechanism underlying hyperglycemia in patients with a severe acute SARS- 3 CoV-2 infection, we performed a multiplex metabolic protein array targeting hormones 4 known to modulate blood glucose homeostasis. We analyzed plasma samples from patients 5 with COVID-19 and ICU control groups from individuals enrolled in our Biobank of the 6 Critically Ill (BOCI) (Finkelsztein et al., 2017; Siempos et al., 2018; Sureshbabu et al., 2018) . 7 The characteristics of this cohort and the full list of targets, including incretins, islet hormones 8 and adipokines, are shown in Table 3 and described below. The incidence of hyperglycemia 9 within the profiled cohort is comparable to what was observed in the parental cohorts (Table   10 1, Figure S1A ). Patients with COVID-19 and ARDS showed a higher rate of hyperglycemia than 11 patients without COVID but with ARDS, although the effect was not statistically significant (P 12 = 0.056, Figure S1A ). Pre-existing diabetes and body mass index (BMI) were not significantly 13 different between the three groups (Figures S1B,C). Since glucocorticoids are part of the 14 treatment for severe COVID-19 (Sterne et al., 2020;  The RECOVERY Collaborative Group, 15 2021) and known to cause or exacerbate hyperglycemia, we assessed the rate of 16 glucocorticoid treatment among the profiled cohort. There was no difference between 17 patients with COVID-19 and ARDS and those with ARDS only (80% vs 68%, P = 0.35, Figure   18 S1D). One patient in the ARDS-negative control group received glucocorticoids. We also 19 charted when glucocorticoids were administered to patients along with daily glucose values 20 and insulin injections (Figures 1G-J and Supplemental Data File 1). Some patients were 21 persistently hyperglycemic with or without glucocorticoids ( Figure 1H ). Others were initially 22 hyperglycemic but their glucose normalized and remained below the hyperglycemia 23 threshold until hospital discharge ( Figure 1I) . Similarly, some patients were not hyperglycemic 24 J o u r n a l P r e -p r o o f for several days following admission but then experienced a rise in blood glucose, which in 1 some cases was associated with glucocorticoids ( Figure 1J ). We quantified the incidence of 2 each of these patterns within the three patient cohorts and found no significant difference in 3 their distribution (Figures S1E,F). 4 Hyperglycemia can be caused by two distinct mechanisms: insulin resistance (IR) or BCF due 5 to diminished beta cell mass and/or insufficient insulin secretion from failing beta cells 6 (DeFronzo et al., 2015). IR is characterized by hyperinsulinemia and hyperglycemia, as the 7 pancreatic beta cells are still functional and attempt to overcome hyperglycemia by increasing 8 insulin secretion. In contrast, individuals experiencing BCF are no longer able to secrete 9 appropriate amounts of insulin, resulting in insulinopenia and severe hyperglycemia. 10 To divide the patients with hyperglycemia into these two subgroups, we assessed their serum 11 C-peptide levels. As a by-product of insulin secretion by the pancreas, C-peptide is a direct 12 indicator of endogenous insulin production and has a longer half-life compared to insulin. 13 Insulin levels were also assessed, but our assay could not distinguish between endogenous 14 and therapeutically administered exogenous insulin. Many of the patients involved in this 15 study received insulin treatment (Figures 1H-J and Supplemental Data File 1), confounding 16 the insulin measurement results. We observed that the patients in the ICU with COVID-19 17 displayed significantly elevated levels of plasma C-peptide compared to either ARDS-positive 18 and ARDS-negative controls, who did not differ from each other (Figure 2A and Figure S2A ). 19 This was independent of the higher rate of hyperglycemia in patients with COVID-19 since it 20 persisted even if only hyperglycemic patients from each cohort were compared ( Figure 2B   21 and Figure S2B ). Similarly, it was also independent of glucocorticoid treatment ( Figure S2C ). To dissect the mechanism of hyperglycemia in patients with COVID-19, study participants ] 10 with hyperglycemia were divided into subtypes reflecting a predominantly IR or BCF 11 phenotype ( Figures S2G-L) . A study participant with type 1 diabetes was adjudicated to have 12 BCF, validating our classification scheme. The incidence of IR was more than three times 13 greater among patients with COVID-19 than among ARDS-positive controls, and more than 14 six times greater than among ARDS-controls ( Figures 2E,F) . Glucocorticoids are known to 15 cause hyperglycemia, by inducing either IR or beta cell damage (Tamez-Pérez, 2015). We 16 found no difference in the incidence of glucocorticoid treatments among the IR and BCF 17 subgroups within the COVID-19 or ARDS-positive control patients ( Figure S3A ). One of the 18 patients assayed in the ARDS-control group received glucocorticoids. There was also no 19 difference in the prevalence of obesity between BCF and IR subtypes in any of the three 20 patient groups. A pre-existing diagnosis of type 2 diabetes on the other hand was significantly 21 associated with BCF over IR in both the COVID-19 and ARDS-positive control groups (Fisher's 22 exact test, P = 0.02 and 0.04 respectively). This suggests that BCF occurs mostly in people with 23 the higher incidence of glucocorticoid treatments in the COVID-19 ARDS-positive group and 1 DM among ARDS-positive control patients ( Figures S1B,D) , we constructed a logistic 2 regression model that adjusted for these differences. We still found a significantly higher rate 3 of IR in patients with COVID-19 compared to both ARDS-positive (OR 8.3, P = 0.009) and ARDS- 4 negative controls (OR 15.1, P < 0.0001) ( Figure S3B ). No difference was observed in the 5 proportion of IR between the two controls, or in the distribution of BCF between any of the 6 groups. 7 Beta cell function and insulin sensitivity are conditioned by endocrine stimuli between tissues. 8 Over the last two decades, it has become increasingly clear that the adipose tissue (AT) acts 9 as a major source of such stimuli. Proteins and cytokines secreted from the AT into the 10 bloodstream, collectively called adipokines, have been shown to regulate beta cell function 11 and mass and to modulate insulin sensitivity in peripheral tissues (Gómez-Banoy and Lo, 12 2019). Dysfunctional adipokine secretion has been associated with metabolic disorders such 13 as obesity, metabolic syndrome, and type 2 diabetes. 14 We therefore tested if severe COVID-19 is associated with aberrations in glucoregulatory 15 hormones such as those arising from the islets, incretins, and adipokines (Figures 3A-F and 16 Figures S4A-J). Furthermore, we assessed whether the abundance of any glucoregulatory 17 hormone differed between the hyperglycemic subgroups ( Figures S5A-Q) . Indeed, the groups contained slightly higher percentages of patients with obesity than the COVID-19 1 group ( Figure S1C ). Furthermore, the ARDS-positive control group had a higher percentage of 2 patients with diagnosed diabetes ( Figure S1B ), suggesting that the decrease in adiponectin 3 associated with COVID-19 may even be underestimated. We found no difference in the 4 abundance of adiponectin among the IR and BCF hyperglycemic subgroups within the patients 5 with COVID-19 ( Figure S5G ). Adipsin, an adipokine known to promote beta-cell survival and 6 insulin secretion (Gómez-Banoy et al., 2019), was also decreased among the patients with 7 COVID-19 compared to ICU controls, although this difference was not statistically significant 8 (overall P = 0.06, Figure S4H ). Leptin, an adipokine regulator of appetite and energy 9 expenditure, was increased with respect to ARDS-positive but not ARDS-negative controls 10 ( Figure 3B ). The adiponectin-leptin ratio, a biomarker of metabolic health and adipose 11 function, was severely depressed in patients with COVID-19 by 10-fold compared to ARDS-12 positive and 6-fold compared to ARDS-negative controls ( Figure 3C ) (Frühbeck et al., 2019). 13 We also found an increase in amylin among patients with COVID-19 ( Figure 3D ). Like C- 14 peptide, this protein is co-secreted with insulin from pancreatic beta cells, further confirming 15 the beta cell hyper-secretory phenotype present in a majority of patients with COVID-19 and 16 ARDS. PAI-1, an adipokine that acts as an inhibitor of fibrinolysis, was also found to be 17 significantly increased with respect to ARDS-negative but not ARDS-positive controls ( Figure   18 3E), as was reported previously for patients with severe COVID-19 (Zuo et al., 2021) . However, 19 increased PAI-1 levels did not correlate with hyperglycemia ( Figure S5O ). Resistin, a Figures 4J,K) . Adipoq was also halved in expression in 4 mouse but not human adipocytes after SARS-CoV-2 infection ( Figure 4L and Figure S6E ). 5 Collectively these results indicate that SARS-CoV-2 incites a robust inflammatory response in 6 adipose tissues in part by direct infection of adipocytes to cause tissue dysfunction. 7 Discussion beta cell failure in a minority of patients, many of them had pre-existing advanced diabetes 1 marked by the use of 2 or more anti-hyperglycemic medications or insulin, or an elevated 2 percent hemoglobin A1c (%HbA1c), signifying prior poor blood glucose control. We also note 3 that hyperglycemia is associated with an increased risk of mortality in both control patient 4 groups without COVID. In contrast, the presence of hyperglycemia in patients with COVID-19 5 and ARDS did not portend a higher risk of death, further supporting a fundamental difference 6 in the mechanism of hyperglycemia in COVID-19. Perhaps classifying the mechanism of 7 hyperglycemia as IR or BCF in an individual infected with SARS-CoV-2 by insulin, C-peptide, 8 and glucose measurements may guide glucose lowering therapy. Insulin is the accepted 9 treatment for hyperglycemia in hospitalized patients. Our study raises the question for patients with COVID-19 and hamsters infected with SARS-CoV-2, circulating leptin levels are 1 higher in patients with COVID-19 whereas Lep is decreased in the hamsters. The hamster 2 model may capture some but not all features of the human disease. Of note, the hamster 3 model used in this study is comparatively mild as the hamsters do not need to undergo 4 mechanical ventilation unlike the patients with COVID-19 and ARDS. In the future, the 5 hamster is a promising model that can be used to dissect the pathophysiological mechanisms 6 of acute and long COVID pertaining to glucose homeostasis and diabetes. 7 Strikingly, both human and mouse adipocytes have relatively higher expression of TFRC, 8 NRP1, and FURIN than ACE2 and TMPRSS2, the latter two being the better studied viral entry 9 factors. Future studies will be needed to determine which receptor(s) SARS-CoV-2 uses to 10 infect adipocytes. We also find that Adipoq is decreased in mouse adipocytes following in vitro 11 infection with SARS-CoV-2 but ADIPOQ in human breast adipocytes is not. This may be due to 12 the fat depot origin of the cells, specific donors or participation of other factors in vivo. 13 Collectively our results implicate direct viral infection of adipose tissues as one potential 14 mechanism for adipose tissue dysfunction and insulin resistance. While we show that 15 adipocytes are capable of being directly infected by SARS-CoV-2, it is possible that other cell 16 types within the adipose, such as endothelial cells and preadipocytes, may be susceptible.

This study is not powered to detect rare events, nor does it rule out potential SARS-CoV-2 1 infection of pancreatic islet cells but suggests that it is not a major etiology of hyperglycemia survivors and those with "Long COVID" are needed to monitor for future IR and BCF. Early 10 data suggests that there may be persistent insulin resistance post COVID-19 (Montefusco et   11 al., 2021). It remains to be determined if patients who recovered from hyperglycemia prior to 12 discharge will have an increased future risk of developing diabetes. 13 Limitations of Study 14 Our study has a number of limitations. First, the study was designed retrospectively and 15 plasma samples were derived from consenting patients. Consequently, the COVID-19 and 16 control groups are not matched for demographic variables, comorbidities, or in-hospital collected early, generally during the first 2 days of ICU admission. IR and BCF are not mutually 1 exclusive and hyperglycemia can be dynamic, especially in acutely ill patients who may be 2 experiencing septic shock, evolving through different phases of acute viral infection, and 3 being treated with glucocorticoids. Our data in critically ill patients precludes extensive 4 physiological assays of insulin secretion and insulin sensitivity. Future studies with 5 longitudinal assessment of IR and BCF over the course of acute SARS-CoV-2 infection may help 6 to better understand the dynamic nature of hyperglycemia. Postprandial C-peptide/glucose 7 ratios are an established marker of beta cell function. However, they are usually assessed at 8 a set time following a standardized meal. This is not possible in the context of the critically ill, 9 intubated patients assessed in this study. 10 Third, the mechanism behind the decreased adiponectin expression in patients with COVID- 11 19 remains to be determined. Although we show the presence of replicating SARS-CoV-2 12 genetic material within hamster adipose tissue and human and murine adipocytes in vitro 13 following SARS-CoV-2 infection, it is still unclear to what extent a similar mechanism indeed 14 exists in people infected with COVID-19. Baseline adiponectin levels prior to infection were 15 not available in the patients to unequivocally demonstrate diminished adiponectin following 16 infection. Furthermore, individuals with acute COVID-19 often experience decreased food 17 intake, if and how this contributes to decreased adiponectin remains to be determined. Lastly, 18 fat biopsies were not possible in these critically ill patients to ascertain active adipose viral 19 infection. Further information and requests for resources and reagents should be directed to and will 4 be fulfilled by the lead contact, James Lo (jlo@med.cornell.edu). 5 Materials availability 6 This study did not generate new unique reagents. 7 Data and code availability 8  RNA-seq data have been deposited at GEO and are publicly available as of the date of 9 publication. Accession numbers are listed in the Key Resources Table. 10  Original western blot images have been deposited at Mendeley and are publicly 11 available as of the date of publication at 12 https://data.mendeley.com/datasets/hvp58h6jc7/1 13  Additional Supplemental Items are available from Mendeley Data at 14 https://data.mendeley.com/datasets/sckg5ygndm/2 15  This paper does not report original code. 16  Any additional information required to reanalyze the data reported in this paper is 17 available from the lead contact upon request. All requests will be reviewed to verify if 18 the request is subject to any intellectual property or confidentiality obligations. Any 19 data and materials that can be shared will be released via a Material Transfer 20 Agreement. 21 Study design and human cohort description 1 The COVID-19-positive component of the study cohort was sourced from multiple 2 institutional resources designed to facilitate COVID research. These resources are aggregated 3 and integrated into a database called the COVID Institutional Data Repository (COVID-IDR). 4 The COVID-IDR contains data extracted automatically from electronic health record (EHR) 5 systems, as well as data abstracted manually, using REDCap (Harris et al., 2009 ), an 6 established research data collection tool, for patients testing positive for SARS-CoV-2 via RT-7 PCR who were admitted or seen in the emergency department at three hospitals within the 8 NewYork-Presbyterian (NYP) healthcare system between March 3, 2020 and May 15, 2020. 9 Some variables (including comorbidities and outcomes) were derived from the REDCap 10 project via manual abstraction, while others (e.g. in-hospital medication usage) were derived 11 from automatically extracted EHR data (Goyal et al., 2020b) . 12 The control population was drawn from an existing data base of patients with an ICU vasopressor requirements were also available. These data are listed in Table 1 (patients with   1 COVID-19) and Table 2 (ICU controls) . 2 For patients with COVID-19, data pertaining to BMI, diabetes status, and %HbA1c were 3 manually extracted by trained and quality controlled medical student reviewers (Goyal et al., 4 2020b). BMI was determined from height and weight recorded during hospital admission or 5 from an outpatient encounter within the past 3 months. Diabetes status was defined by 6 physician documentation of type 1 or type 2 DM in the medical history/diagnosis codes or by 7 %HbA1c ≥ 6.5. %HbA1c levels recorded within 10 days of hospitalization or within 90 days 8 prior to admission were included. Diabetes medications were automatically extracted from 9 the HER and manually confirmed for plasma-sampled individuals. 10 For both patients with COVID-19 and ICU controls, hyperglycemia was defined as having a 11 peak glucose value of > 170 mg/dL. Peak glucose was defined as the highest glucose value 12 during the entire hospitalization. Glucose measurements were extracted from the electronic 13 health record. Patients without glucose values were excluded from the analysis. For study 14 participants with plasma profiled samples, two separate glucose values > 170 mg/dL were 15 required to be classified as hyperglycemic. In addition, dextrose infusions were assessed to 16 ensure that iatrogenic hyperglycemia was not being identified. 17 The study was approved by the institutional review board of WCMC (1405015116, 20-18 05022072, 20-03021681). 19 Human Plasma Samples during the COVID-19 pandemic, an electronic informed consent was obtained from all 2 individuals who were SARS-CoV-2-positive or from their surrogates for inclusion. 3 Demographics, comorbidities, diagnoses, and laboratory values for the plasma-sampled 4 subset of patients are shown in Table 3 . Mycoplasma Detection Kit (lonza), They were not authenticated by an external service but 3 were derived directly from ATCC. 3T3-L1 adipocytes (Mus musculus, male) were cultured in 4 DMEM (Corning) with 10% FBS at 37 °C, 5% CO2. L1 cells were not tested for mycoplasma and 5 validated by differentiation. 6 Human adipocyte donors 7 Breast adipose tissue was obtained from patients undergoing mastectomy or breast reduction 8 surgery under approved Institutional Review Board protocols (20-01021391 and 1510016712) 9 at Weill Cornell Medicine. Two independent experiments with adipocytes from different 10 donors were carried out. The patient characteristics were as follows: 13 Biobank sample collection and ELISA 14 For each participant, whole blood (6-10 mL) was obtained. Briefly, whole blood samples were 15 drawn into EDTA-coated blood collection tubes (BD Pharmingen, San Jose, CA). Samples were 16 stored at 4°C and centrifuged within 4 hours of collection. Plasma was separated and divided 17 into aliquots and kept at -80°C. Plasma insulin and C-peptide were measured by ELISA and Resistin, Lipocalin 2, PAI-1. 9 Glycemic categorization of patients 10 Patients classed as hyperglycemic were subdivided into IR or BCF. The C-peptide reading from 11 the Multiplex Metabolic Protein Array (pg/mL) was divided by the glucose measurement 12 closest to the time of plasma collection (mg/dL) to calculate C-peptide/glucose ratios. Patients 13 who had a ratio of less than 20.5 and were either hyperglycemic or receiving insulin treatment 14 at sampling time were classed as BCF. Approximately 20 mL of tissue was dissected, avoiding blood vessels and fibrous areas. The 1 tissue was then minced into 2-3 mm pieces and digested in 50 mL total volume of Ham's F12 2 media (Corning) supplemented with 10% FBS, 1% penicillin/streptomycin, 10 mg/mL 3 collagenase type I (Sigma Aldrich) and 10 ug/mL hyaluronidase (Sigma Aldrich). Tissue was 4 digested for 1 hr at 37 °C on a rotator. The digested tissue was then centrifuged at 500 g for 5 10 min to separate the adipocyte fraction (floating layer) and stromal vascular fraction 6 (pellet). Oil from lysed adipocytes was removed by aspiration and a wide orifice pipet tip was 7 used to collect the adipocytes below the oil layer. Adipocytes were washed twice cultured in 8 Ham's F12 media (Corning) supplemented with 10% FBS, 1% penicillin/streptomycin (37 °C, 9 5% CO2). Viral infections were performed within 24 h following adipocyte isolation.

Logistic regression model 12 Models to determine if group membership predicted the outcome of euglycemia, insulin 13 resistance, or beta cell failure were determined from the assays. The outcome was assumed 14 to be ordinal; ranks of 0, 1, and 2 were assigned to euglycemic, insulin resistant, and beta cell 15 failure patients respectively. As such, ordinal logistic regression models with a cumulative 16 logit link function was employed. Models were tested to meet the assumption of proportional 17 odds. When the assumption was violated, we used the unequal slopes option where the 18 model assumes separate slopes for each level of the outcome. 19 Data analysis 20 The COVID and control populations were profiled and tabulated. The subsample of patients 21 with the multiplex arrays were profiled and tabulated and statistical tests were conducted to 22 determine group differences (C+A+, C-A+, C-A-). Hamsters were randomly assigned to virus-23 infected or mock-control groups. Experiments were conducted in batches of 3 hamsters per 1 group, they were not blinded and no power calculations were done. No animals or samples 2 were excluded. Fisher's exact test and chi-squared test were used to test association of 3 categorical variables, odds ratios were used to determine the strength of the association. 4 Continuous data was tested for normality (Shapiro-Wilk). Normally distributed data were 5 compared using Welch's t-test, otherwise non parametric tests were used (Kruskal-Wallis test 

Can COVID-19 cause diabetes?

Gene expression of leptin, resistin, 4 and adiponectin in the white adipose tissue of obese patients with non-alcoholic fatty liver 5 disease and insulin resistance

Associations of type 1 and type 2 diabetes with COVID-8 19-related mortality in England: a whole-population study

Admission hyperglycaemia as a predictor of mortality 13 in patients hospitalized with COVID-19 regardless of diabetes status: data from the Spanish 14 SEMI-COVID-19 Registry

Simulation of the Clinical and Pathological 17 Manifestations of Coronavirus Disease 2019 (COVID-19) in a Golden Syrian Hamster Model: 18 Implications for Disease Pathogenesis and Transmissibility

SARS-CoV-2 Cell Entry Factors ACE2 and TMPRSS2 Are 21

Expressed in the Microvasculature and Ducts of Human Pancreas but Are Not Enriched in β

Adiponectin-leptin Ratio is a Functional Biomarker of Adipose 2 Tissue Inflammation

Adipokines as key players in β cell function and failure

Adipsin preserves beta cells in 7 diabetic mice and associates with protection from type 2 diabetes in humans

Obesity and COVID-19 in New York City: A 11 Retrospective Cohort Study

Clinical Characteristics of Covid-19 in

A randomized multicenter Phase II study of 17 docosahexaenoic acid in patients with a history of breast cancer, premalignant lesions, or 18 benign breast disease

Extrapulmonary manifestations of 21 COVID-19

Research 1 electronic data capture (REDCap)-A metadata-driven methodology and workflow process for 2 providing translational research informatics support

Autoantibody-negative insulin-5 dependent diabetes mellitus after SARS-CoV-2 infection: a case report

Risk factors for COVID-19-related mortality in people 9 with type 1 and type 2 diabetes in England: a population-based cohort study

TβRIII Expression in Human Breast Cancer Stroma and the Role of Soluble TβRIII in Breast 13 Cancer Associated Fibroblasts. Cancers (Basel)

Propagation, Inactivation, and Safety Testing of 15 SARS-CoV-2

Expression of SARS-CoV-2 Entry Factors in 18 the Pancreas of Normal Organ Donors and Individuals with COVID-19

Postprandial C-peptide to glucose ratio as a predictor of β-cell

WebGestalt 2019: gene set 3 analysis toolkit with revamped UIs and APIs

COVID-19 and diabetes mellitus: from 5 pathophysiology to clinical management

When Two Pandemics Meet: Why Is Obesity 7 Associated with Increased COVID-19 Mortality? Med

Effects of Moderate and Subsequent 10

Progressive Weight Loss on Metabolic Function and Adipose Tissue Biology in Humans with 11 Obesity

Impact of Severe Acute Respiratory Syndrome 14

Coronavirus 2 Viral Load on Risk of Intubation and Mortality Among Hospitalized Patients 15

With Coronavirus Disease

Risks of and risk factors 18 for COVID-19 disease in people with diabetes: a cohort study of the total population of 19

CoV-2 host range to mice

Acute and long-term disruption 3 of glycometabolic control after SARS-CoV-2 infection

SARS-CoV-2 infects and replicates in cells of the human 6 endocrine and exocrine pancreas

Glycogen Synthase Kinase-3α Promotes Fatty Acid Uptake 9 and Lipotoxic Cardiomyopathy

Multiorgan and Renal Tropism 12 of SARS-CoV-2

edgeR: A Bioconductor package for 14 differential expression analysis of digital gene expression data

Postprandial C-Peptide to Glucose Ratio as a Marker of β Cell Function: 16 Implication for the Management of Type 2 Diabetes

Postprandial serum C-peptide to plasma glucose ratio as a predictor of 19 subsequent insulin treatment in patients with type 2 diabetes

CEDAR): An automated method to support intensive care units with electronic health record data

Secondary Use of Patients' Electronic Records (SUPER): An Approach 3 for Meeting Specific Data Needs of Clinical and Translational Researchers

Symp. Proc. 2017

Pathogenesis and transmission of SARS

CoV-2 in golden hamsters

RIPK3 mediates pathogenesis of 10 experimental ventilator-induced lung injury

Impaired glucose metabolism in patients with diabetes, 13 prediabetes, and obesity is associated with severe COVID-19

Association between Administration of 16

Systemic Corticosteroids and Mortality among Critically Ill Patients with COVID-19: A Meta-17 analysis

RIPK3 promotes sepsis-20 induced acute kidney injury via mitochondrial dysfunction

Steroid hyperglycemia: Prevalence, early detection and therapeutic recommendations: A narrative review

Mouse Adipose Tissue Collection and Processing 2 for RNA Analysis

SARS-CoV-2 infection induces beta cell transdifferentiation

Dexamethasone in Hospitalized Patients with 7

Fasting blood glucose at admission is an independent predictor for 28-day mortality 10 in patients with COVID-19 without previous diagnosis of diabetes: a multi-centre 11 retrospective study

Factors associated with COVID-19-related 14 death using OpenSAFELY

The membrane-anchored serine protease, TMPRSS2, activates PAR-2 in prostate cancer 17 cells

Hypoxic and pharmacological 20 activation of HIF inhibits SARS-CoV-2 infection of lung epithelial cells

SARS-CoV-2 infects human pancreatic β cells and 1 elicits β cell impairment

A Human Pluripotent Stem Cell-based Platform to Study SARS

CoV-2 Tropism and Model Virus Infection in Human Cells and Organoids

Estimating the infection-fatality risk of SARS-CoV-2 in

New York City during the spring 2020 pandemic wave: a model-based analysis

Effect of SARS-CoV-2 spike mutations on animal ACE2 usage and in vitro 12 neutralization sensitivity

Insulin Treatment Is Associated with Increased 14 Mortality in Patients with COVID-19 and Type 2 Diabetes

Clinical course and risk factors for mortality of adult inpatients with COVID

China: a retrospective cohort study

Association of Blood Glucose Control and Outcomes in Patients with COVID-19 20 and Pre-existing Type 2 Diabetes

report that hyperglycemia in critically ill patients with COVID-19 is caused mainly by insulin resistance and is associated with decreased circulating adiponectin. SARS-CoV-2 is shown to directly infect human adipocytes, trigger an inflammatory antiviral response in the adipose tissue and cause its dysfunction