Main

Initially, the pandemic COVID-19, caused by SARS-CoV-2, was considered to be an exclusive lung disease, eventually leading to serious respiratory symptoms1. In the meantime, accumulating experimental and clinical studies have suggested that SARS-CoV-2 may also cause lesions in the kidneys, heart, brain, and gastrointestinal and endocrine organs2,3,4,5,6,7. SARS-CoV-2 tropism towards distinct tissues is governed by cellular factors expressed on target cells such as the viral entry receptor angiotensin-converting enzyme 2 (ACE2)8 and the transmembrane serine protease 2 (TMPRSS2)8. ACE2 messenger RNA9,10,11,12,13 and protein12,13,14 expression within the islets of Langerhans has been reported, but not yet been shown, to allow SARS-CoV-2 entry9,12,15. Diabetes mellitus presents Janus like in COVID-19 (refs. 3,16): first, pre-existing diabetes is a highly prevalent comorbidity observed in 11–22% of patients and as such increases the risk of a severe disease, requiring more intense interventions and increasing mortality17,18,19,20,21,22. Second, SARS-CoV-2 infection seems to affect the exocrine pancreas, manifesting as pancreatitis in 32.5% of critically ill patients23, and pancreatic enlargement and abnormal amylase or lipase levels in 7.5–17% of patients9,22. Third, metabolic dysregulation has been observed in patients with COVID-19 as: (1) increased hyperglycaemia in patients with type 2 diabetes24; (2) ketoacidosis in 2–6.4% of diabetic and non-diabetic patients18,25; and (3), in case studies reporting ketoacidosis on SARS-CoV-2 infection, accompanied by (4) new-onset type 1 diabetes mellitus (T1DM) in the absence of autoantibodies26,27,28. In a cohort study of patients with diabetes, hyperglycaemia was reported in more than 50% of all cases, and almost a third experienced diabetic ketoacidosis29. Finally, a multicentre study found an 80% increase of new-onset T1DM in children during the COVID-19 pandemic30. In accordance, a recent meta-analysis summarizes that severe COVID-19 is associated with increased blood glucose levels31. However, the formal proof of SARS-CoV-2 as a β-cell tropic virus, potentially leading to diabetes, is still missing, and the only correlative evidence stands in the light of conflicting experimental and clinical findings13,22,32,33,34. Prospectively collected acute and long-term outcomes on new-onset diabetes cases, together with thoughtful interpretations of emerging data up to final clarification of this debate, are warranted33.

Accordingly, it is unclear whether and how SARS-CoV-2 might trigger β-cell injury, but it may occur via either immune-mediated β-cell ablation or direct perturbation of β-cell function, both eventually leading to so-called infection-related diabetes according to the current World Health Organization (WHO) classification35. Recent evidence suggests that SARS-CoV-2 can infect human endocrine cells in vitro12,36. However, this finding was obtained with stem-cell-derived, immature human β-cells that express ACE2 and TMPRSS2, and viral replication or impact on β-cell function has not been analysed in detail36. ACE2 or TMPRSS2 expression was also detected in exocrine and endocrine cells in human pancreatic tissue; however, varying expression patterns across distinct pancreatic cell types were reported9,10,12,13,14,15,34,36,37,38. Thus, it is imperative to clarify whether human pancreatic endocrine cells organized within an islet of Langerhans are permissive for and affected by SARS-CoV-2 infection, and to elucidate the mechanisms underlying a potential endocrine dysfunction associated with COVID-19 (refs. 16,21,22,36).

In the present study, we (1) defined ACE2 and TMPRSS2 expression patterns in human pancreatic endocrine and exocrine cell types, (2) employed human pancreatic islet cultures to demonstrate susceptibility to SARS-CoV-2 infection and viral replication in β-cells and (3) showed that SARS-CoV-2 infection affects glucose-stimulated insulin secretion (GSIS). In addition, we (4) visualized viral particles replicating in endocrine pancreatic cells and defined their subcellular localization patterns and finally (5) presented examples of multiorgan infection patterns including the pancreata of patients who died from COVID-19.

Results

ACE2 and TMPRSS2 expression in endocrine cells and a ductal subpopulation

As pancreatic ACE2 and TMPRSS2 expression is currently under debate12,13,34,37, we initiated our validation analysis with two reference antibodies (ab15348 and ab92323, Abcam; Extended Data Fig. 1), which have been previously extensively characterized in immunofluorescence and immunohistochemistry studies (ACE2 (refs. 12,36,39,40,41,42) and TMPRSS2 (refs. 43,44,45,46,47)). First, quantitative PCR (qPCR) analysis of ACE2 and TMPRSS2 in human lung Calu-3 cells and EndoC-βH1 cells, a model of human β-cells48, was performed and revealed detectable RNA levels of both viral entry proteins (Extended Data Fig. 2a). Immunoblotting confirmed detection of the correct molecular mass of ~110/120 kDa and recently reported12,49 short 50-kDa isoforms of endogenous ACE2 in both cell types (Extended Data Fig. 2b). We found all isoforms of ACE2 to be expressed in fresh-frozen human pancreatic tissue comprising exocrine and endocrine cell types (Extended Data Fig. 2b). Similarly, the TMPRSS2 antibody detected proteins of 54 kDa and 26 kDa, consistent with glycosylated forms of full‐length TMPRSS2 and the cleaved serine protease domain, as previously reported50 (Extended Data Fig. 2c). Notably, ACE2 and TMPRSS2 expression of the different isoforms varied across participants (Extended Data Fig. 2b,c), in line with previous findings12. Immunofluorescence imaging revealed similar staining patterns for ACE2 or TMPRSS2 expression in differentiated air–liquid interface cultures of primary human airway epithelial cells (HAECs)51 as well as in EndoC-βH1 cells (Extended Data Fig. 2d,e). Preincubation of the employed anti-ACE2 antibody with an epitope-matching blocking peptide abrogated ACE2 detection in EndoC-βH1 cells (Extended Data Fig. 2e).

After validation of antibody specificity, we imaged ACE2 and TMPRSS2 expression in tissue sections derived from five histologically healthy human pancreata. Fluorescence staining of both SARS-CoV-2 entry factors was observed in the islets of Langerhans in all samples (Fig. 1a,c, and Extended Data Figs. 3 and 4). Strong ACE2 expression was detected in endothelial cells (Extended Data Fig. 3a–c, white arrowheads) and in a subpopulation of cytokeratin 19 (CK19)-positive ductal cells (Extended Data Fig. 3d). Moderate ACE2 signals were observed in endocrine cells (Fig. 1a and Extended Data Fig. 3a–c) and detection was prevented by the ACE2-blocking peptide during immunostaining (Extended Data Fig. 3e). ACE2 was only faintly expressed in GATA-binding protein 4-positive acinar cells (Extended Data Fig. 3f). Similarly, TMPRSS2 was detected in the endocrine compartment (Fig. 1c and Extended Data Fig. 4a–c) and in some ducts (Extended Data Fig. 4d). TMPRSS2 expression in acinar cells was barely detectable (Extended Data Fig. 4e). Co-staining for endocrine cell types and viral entry proteins revealed a heterogeneous staining pattern across the five donors with varying coefficients (Fig. 1b,d). The highest coefficients were found for C-peptide (C-pep)-positive β-cells co-stained for ACE2 (mean: 0.40 (minimum–maximum: 0.22–0.78)) and TMPRSS2 (mean: 0.73 (minimum–maximum: 0.59–0.86)) (Fig. 1b,d). The α- and δ-cells expressing either glucagon (GCG) or somatostatin (SST), respectively, revealed a smaller ACE2 (mean: 0.15 (minimum–maximum: 0.07–0.23); mean: 0.18 (minimum–maximum: 0.04–0.35)) or TMPRSS2 (mean: 0.12 (minimum–maximum: 0.01–0.28); mean: 0.17 (minimum–maximum: 0.03–0.42)) double-positive fraction with less variance across the five patients (Fig. 1b,d). Thus, pancreatic exocrine and endocrine cells express SARS-CoV-2 entry factors.

Fig. 1: Pancreatic β-cells express SARS-CoV-2 entry factors ACE2 and TMPRSS2.
figure 1

a,c, Adult pancreatic tissue sections from five healthy participants, stained with antibodies against ACE2 (a; red) or TMPRSS2 (c; red), and C-pep (green), GCG (green) or SST (green). Cell nuclei were visualized using DAPI (blue). Representative confocal sections of participant 1 are shown; for participants 2–5, see Extended Data Figs. 3 and 4. Scale bars, 10 µm. b,d, Colocalization of cell-specific markers C-pep, GCG or SST with ACE2 (b) or TMPRSS2 (d) was analysed using Pearson’s correlation coefficients and Fiji. Numbers of investigated healthy human pancreata were as follows: ACE2/C-pep: n = 5; TMPRSS2/C-pep: n = 4; ACE2/SST or GCG: n = 4; TMPRSS2/SST or GCG: n = 4. Four different islets of Langerhans were analysed per patient. Data are presented as mean ± s.e.m. quantified for each participant and staining combination. Ordinary one-way analysis of variance (ANOVA) with Tukey’s post-test was used.

Source data

SARS-CoV-2 replicates in human pancreatic islets

To determine the susceptibility to ex vivo infection, human pancreatic islets isolated from four human donors were exposed to SARS-CoV-2, and expression of viral spike (S) and nucleocapsid (N) protein, as well as endocrine cell markers, was analysed. S and N proteins were not detected at day 1 (not shown), but became readily detectable at days 3 (Fig. 2a) and 5 (Fig. 2b) post-infection (Extended Data Fig. 5). Pancreatic islets treated with 5 µM remdesivir, a polymerase inhibitor with potent in vitro anti-SARS-CoV-2 activity52, did not stain positive for S or N proteins, indicating suppression of SARS-CoV-2 replication. Quantification of viral N-protein expression in infected islets confirmed robust infection ranging between 20% N-positive cells at day 3 and 34% at day 5 per infected islet (Fig. 2c). Only a few cells stained positive for cleaved caspase 3 (CASP3) across all conditions, suggesting no increased apoptosis at this stage (Fig. 2a,b). Although some cells exhibited double positivity for the pancreatic hormones C-pep/chromogranin A and the viral N/S proteins, which was not observed for GCG and SST, most of the SARS-CoV-2-infected cells appeared to lack hormone expression (Fig. 2a,b,d and Extended Data Fig. 5). Staining with non-endocrine markers such as the endothelial marker platelet endothelial cell adhesion molecule (PECAM-1, CD31), the ductal marker CK19 or the acinar marker chymotrypsin revealed only scattered positive cells in pancreatic islet preparations, thus making the bias of the results due to preparation impurities unlikely (Extended Data Fig. 6). To probe lineage identity of infected cells, we stained for pancreatic and duodenal homoeobox 1 (PDX1) and NKX6.1, both markers mostly labelling endocrine cells in the adult pancreas53,54,55, and found that infected hormone-negative cells were still positive for PDX1 or NKX6.1, suggesting that they are endocrine cells that lose hormones upon infection (Fig. 2d). Quantification of N/NKX6.1 double-positive cells per infected islet of donor 3 at day 5 post-infection revealed that approximately 21% of putative β-cells were infected (Extended Data Fig. 6d,e). A more definite quantification was impeded by the amount of analysable material and the fact that N/S-positive cells of the endocrine lineage appear to be insulin negative on infection. However, increasing intra- and extracellular viral RNA levels of the islets in the absence of remdesivir indicate progressive viral replication (Fig. 2e,f). Productive viral replication in islets of all donors was confirmed by increasing infectious viral titres in the respective supernatants (Fig. 2g). On remdesivir treatment, almost no infectious virus was detected in the supernatants of islets (Fig. 2g), indicating efficient inhibition. This is in line with low viral RNA levels (Fig. 2e,f) and the absence of N or S protein in confocal microscopy analyses (Fig. 2a,b and Extended Data Fig. 5) in the presence of remdesivir. Thus, pancreatic islets are susceptible to SARS-CoV-2 infection, which can be inhibited by remdesivir.

Fig. 2: SARS-CoV-2 productively infects human pancreatic islets.
figure 2

Human pancreatic islets were mock treated with medium or infected with SARS-CoV-2 and cultivated in the presence or absence of remdesivir (5 µM). Displayed images represent three donors. a,b, Islets of donor 2 fixed 3 d (a) or 5 d (b) post-infection were stained for SARS-CoV-2 N protein (SARS-CoV-2 N, red), C-pep (green), cleaved CASP3 (white) and nuclei (DAPI, blue). Representative images are shown. Scale bars, 10 µm. c, Quantification of fraction of SARS-CoV-2 N-protein-positive cells in infected islets of donors 2 and 3 (see Extended Data Fig. 5 for donor 3 stainings). Data are presented as mean ± s.e.m. from seven (day 3) and six (day 5) individual infected islets and from four uninfected islets (unpaired, two-sided, Student’s t-test). d, Islets from donor 3 were infected and SARS-CoV-2-infected cells (SARS-CoV-2 N, red), β-cells (C-pep, green) and endocrine lineage cells (PDX1, blue) were visualized. Nuclei stained with DAPI are pseudo-coloured in white. Scale bar, 5 µm. SARS-CoV-2-infected cells frequently became hormone (C-pep) negative but remained lineage positive (PDX1). Uninfected cells remained double positive for C-pep and PDX1. e, Supernatants of islets (donors 1–4) were harvested over 5 d and SARS-CoV-2 ORF1b-nsp14 was quantified by qPCR. Data are presented as the mean values of two replicates. For each donor (Do), changes relative to day 0 (viral input RNA) are visualized. f, SARS-CoV-2 N was quantified in cellular RNA isolates of donors 2 and 3 at days 3 and 5 post-infection and normalized to GAPDH RNA. Data are presented as mean values of two replicates. g, Supernatants of all four donors from e were assessed for infectivity by TCID50 endpoint titration. Data are presented as the mean values of two replicates and the dotted line indicates the lower limit of quantification.

Source data

SARS-CoV-2-infected endocrine cells show subcellular and functional changes

Infection of endocrine cells by SARS-CoV-2 was further analysed using transmission electron microscopy (TEM). Pancreatic islets from human organ donors 2 and 3 were infected with SARS-CoV-2 and analysed after 3 and 5 d ex vivo culture (Fig. 3a and Extended Data Fig. 7). Infection of islet cells with SARS-CoV-2 resulted in dilatation and vacuolization of the endoplasmic reticulum (ER)–Golgi apparatus complex, a finding suggestive of ER stress and Golgi body swelling56,57,58. These vacuoles contained viral particles showing coronavirus morphology56,57,58, indicating productively infected endocrine cells. The virion-containing vesicles were formed in the perinuclear region and processed to the cell surface. Furthermore, infection occurred in cells containing secretory vesicles that seemed to be enlarged and maintained in the perinuclear region (Fig. 3a). Granule numbers per cell in infected islets were decreased by 2.2-fold on day 3 (Fig. 3b) and 2.4-fold on day 5 (Fig. 3c) post-infection. In contrast, we did not detect intracellular viral particles and observed fewer morphological alterations in remdesivir-treated islet cells (Fig. 3a and Extended Data Fig. 7a). Thus, SARS-CoV-2 behaved in human endocrine cells similar to previously reported TEM phenotypes of infected lung- and gut-derived cells56,57,58. To analyse whether SARS-CoV-2 infection and associated subcellular changes of the islets affect function, we assessed the islet response towards a high glucose pulse. We found that GSIS was induced in all conditions, but the magnitude of induction from low to high glucose was reduced in infected islets (Fig. 3d,e). Of note, overall glucose responsiveness was lower in two-islet preparations, and varied between preparations, most probably due to the limitations of prolonged ex vivo culture. These data corroborate that SARS-CoV-2 replicates in endocrine cells and suggest that infection may affect glucose-dependent insulin secretion in pancreatic islets. However, to obtain more definite conclusions more islet preparations will be necessary.

Fig. 3: SARS-CoV-2 infects and replicates in pancreatic islets, thus resulting in impairment of β-cell function.
figure 3

Human pancreatic islets of donor 3 were infected with SARS-CoV-2 and cultivated with or without 5 µM remdesivir, or left uninfected. a, At day 5, islets were fixed and sectioned for TEM analysis. Electron micrograph (i) and magnified inlet (ii) of the infected preparation show cells with endocrine secretory vesicles (orange arrowheads) and dilated Golgi vacuoles (red asterisks) containing virus particles (blue arrows). Vacuoles and viral particles were absent in the uninfected (iii) and remdesivir-treated (5 µM) (iv) samples (see Extended Data Fig. 7a for further micrographs at equal magnifications as in i and ii). b,c, Endocrine secretory vesicles from donor 3 were manually identified by two independent individuals at day 3 (b) or 5 (c) post-infection and quantified blinded using Fiji. Data are presented as mean ± s.e.m. of 8 TEM images (day 3) and 8, 5 and 7 TEM images for uninfected, infected and remdesivir-treated islets (day 5), respectively, containing on average 19 nuclei. Statistical significance was calculated by ordinary one-way ANOVA with Tukey’s post-test. d,e, At day 3 post-infection, islet functionality of donors 1, 2 and 4 was analysed by static GSIS. Islets were exposed first to 2 mM and then to 20 mM glucose for 1 h each, and insulin secretion into the buffer was determined by ultrasensitive insulin ELISA. Calculated insulin secretion indicated in the corresponding bars (d) and averaged as fold inductions (e) is shown. Data are presented as mean ± s.e.m. of two technical replicates (donor 1) and six replicates (donors 2 and 4). Statistical significance of fold insulin induction was calculated using an unpaired, two-sided Student’s t-test.

Source data

Transcriptional changes in human islets after SARS-CoV-2 infection

To assess potential transcriptional changes induced by SARS-CoV-2, we performed bulk RNA-sequencing (Smart-Seq2) of uninfected and infected (with or without remdesivir), cultured human islets from two donors in an explorative analysis. First, respective transcriptomes obtained 5 d post-infection were clustered. Most of the sample variance was determined by the two islet preparations differing also in the donor sex (Fig. 4a). However, transcriptomes from SARS-CoV-2-infected cells clearly separated from uninfected counterparts, whereas remdesivir treatment resulted in intermediate clustering (Fig. 4a). Among the top upregulated genes in SARS-CoV-2-infected islets were several interferon (IFN)-stimulated genes (ISGs) such as IFITMs59, OAS2, IFI27 and ISG15, whereas genes linked to β-cell physiology or diabetes59,60,61,62,63,64,65, such as SYT4, PASK, PEX6 and PLCXD3, were significantly downregulated (Fig. 4b). Of note, ISGs were upregulated not only after SARS-CoV-2 infection, as compared with uninfected cases, but also in remdesivir-treated islets (Fig. 4c). Gene ontology (GO) term analysis confirmed an initiation of a transcriptional cellular defence reaction in response to SARS-CoV-2 infection. Terms such as ‘defence response to virus’ and ‘regulation of viral genome replication’ were strongly upregulated after SARS-CoV-2 infection (Fig. 4d). Comparing SARS-CoV-2 infection with remdesivir-treated infected cultures revealed a similar but less pronounced enrichment of INF-related terms such as ‘IFN α/β signaling’ and ‘type I IFN signaling pathway’, indicating a partial reversion of transcriptional changes (Fig. 4f). A similar pattern was observed when focusing on COVID-19-related disease terms (Fig. 4e,g). Gene set enrichment analysis (GSEA) further confirmed the enrichment of IFN signaling in SARS-CoV-2-infected islets against uninfected and remdesivir-treated infected islets (Fig. 4h–k). In addition, a trend indicating loss of β-cell identity, as revealed by several gene sets66, as well as defects in protein secretion in virally infected islets, could be detected (Fig. 4h–k). Vice versa, these defects were attenuated on remdesivir treatment, indicating that the observed changes are caused by SARS-CoV-2 infection, which is in accordance with our functional ex vivo experiments (Fig. 3). Thus, on a transcriptional level, infected islets show innate defence reactions and transcriptional changes indicative of loss of β-cell identity. Of note, viral infection and IFN type I response have been shown to trigger development of T1DM in individuals with genetic predisposition67,68,69. However, more studies with higher sample sizes will be necessary to confirm these conclusions.

Fig. 4: Transcriptional changes in human islets after SARS-CoV-2 infection.
figure 4

Human pancreatic islets of donors 2 and 3 were infected with SARS-CoV-2 and cultivated with or without 5 µM remdesivir, or left uninfected, and prepared for RNA-seq. a, Smart-seq2 expression heatmap illustrating proximity between different treatment conditions and experiments (n = 2 islet donors). b,c, Volcano plots with depicted genes of interest for comparison of islets infected with SARS-CoV-2 and uninfected (b) or remdesivir-treated (c) islets. Significant genes were highlighted in blue (adjusted P < 0.1) and significant genes with a log2(fold-change) > |1| in red. Statistical significance was tested using DESeq2 (ref. 104). d,f, Selection of significantly enriched gene sets comparing differentially expressed genes in islets infected with SARS-CoV-2 versus uninfected (d) or versus remdesivir-treated (f) islets in overrepresentation analyses against common databases; the enrichment test for significance was performed using g:Profiler105. e,g, Selection of significantly enriched gene sets in overrepresentation analyses against COVID-19-related disease terms comparing differentially expressed genes in islets infected with SARS-CoV-2 versus uninfected (e) or versus remdesivir-treated (g) islets; the enrichment test for significance was performed using EnrichR106 (***adjusted P < 0.001, **adjusted P < 0.01, *adjusted P < 0.05). h,j, Normalized enrichment scores for a selection of gene sets in GSEA of islets infected with SARS-CoV-2 compared to uninfected (h) or to remdesivir-treated (j) islets (FDR; ***P < 0.001, **P < 0.01, *P < 0.05). ROS, reactive oxygen species. i,k, GSEA plots for selected gene sets of islets infected with SARS-CoV-2 compared to uninfected (i) or to remdesivir-treated (k) islets. The false discovery rate (FDR) was determined using the GSEA (Broad Institute) desktop tool107 as detailed in Methods.

Source data

Cross-organ and pancreatic infection during severe COVID-19

To examine whether the ex vivo observed infection of islets also naturally occurs in pancreatic tissue, we performed postmortem studies on four individuals who died from COVID-19 (Table 1). One patient was SARS-CoV-2 negative on admittance to the hospital but developed respiratory symptoms 4 d later, and consecutive SARS-CoV-2 tests revealed increasing viral RNA levels in analysed nasopharyngeal swabs (Fig. 5a). A past medical history included metformin-controlled type 2 diabetes, arterial hypertension and coronary artery disease (Table 1). Under conservative therapy, the patient rapidly deteriorated, accompanied by a progressive oxygen demand, and died due to respiratory failure 3 d after the onset of clinical symptoms (Fig. 5a). The other three patients had longer disease courses, ranging from 9 to 28 d of hospitalization and 6 to 24 d of ventilation before death (Table 1). One patient had diabetes and was taking oral medication and one had exocrine pancreatic insufficiency. These three patients developed acute kidney injury at various stages. In total, two patients died from respiratory failure, one from gastrointestinal bleeding and one from multiorgan failure (Table 1).

Table 1 Patient data of COVID-19 deceased patients
Fig. 5: Multiorgan infection in COVID-19 deceased patients.
figure 5

a, Timeline of disease course of patient 1, hospitalized due to a planned transcatheter aortic valve implantation (TAVI) procedure, including blood glucose levels throughout hospitalization and thoracic X-ray at day 5 of hospitalization. b, Multiple organs of patient 1 were stained for viral N protein (red). Sections of the lung showed strong staining of shed alveolar macrophages as well as of alveolar lining cells. In the kidney, epithelial cells of the tubulus (arrows) are N-protein positive, as well as some cells within the glomerulus (arrowhead). The spleen shows no positive cells. In the liver, N protein is detected in some cells of the bile duct as well-isolated hepatocytes, whereas lymph nodes and heart muscle appear negative. Representative images from four biopsy sections of the lung and one biopsy section of other organs were selected. c, Lungs, kidneys and spleens of patients 2–4 were stained for viral N protein. For patient 2, the lung tissue appeared negative, whereas oedema (hash) and thrombus in a small vessel (asterisk) were seen. In the kidney, some tubulus cells were N-protein positive whereas in the spleen only very few cells showed weak staining. Patient 3: the lung tissue was negative for SARS-CoV-2 N, showing some initial oedema and thrombus in a small pulmonary artery (asterisk). In the kidney, cells of the proximal tubulus were positive (arrow), as were some cells in Bowman’s capsule of the glomerulus (arrowhead). No N protein was seen in the spleen. Patient 4: lung tissue appeared negative except for one positive cell in the interstitial space. A thrombus is marked by an asterisk; the hash marks the alveolar space. In the kidney some tubular cells were N-protein positive, whereas the glomerulus appeared negative. In the spleen one single positive cell cluster was detected. Inset shows a high magnification to illustrate this observation. A representative image from one biopsy section was selected. Scale bar, 100 µm.

At the postmortem examination, we systematically stained different organs for viral SARS-CoV-2 N protein using two independent anti-N antibodies, validated for immunohistochemistry in a SARS-CoV-2 gut organoid model70 (Extended Data Fig. 8a). In patient 1, we observed massive N-protein staining in pneumocytes lining the alveolar space, as well as alveolar macrophages resembling acute pneumonia with diffuse alveolar damage (Fig. 5b and Extended Data Fig. 8b), explaining the rapid death of this patient due to acute respiratory failure66. In addition, some epithelial cells of the kidney tubules and a few hepatocytes and cholangiocytes stained N-protein positive, whereas heart muscle, lymph nodes and the spleen showed no signs of SARS-CoV-2 infection (Fig. 5b and Extended Data Fig. 8b). No clear viral N-protein signal was detected in the lung tissues of the other three patients (Fig. 5c and Extended Data Fig. 8b) with a longer disease course, but rather we observed macrothrombi (the asterisk in Fig. 5c), which is in line with previous observations during later stages of COVID-19 (refs. 4,71). Similar to patient 1 (Fig. 5b and Extended Data Fig. 8b), we found viral N protein in all analysed kidneys, supporting previous findings of a renal tropism of SARS-CoV-2 (refs. 3,4,6,71,72,73,74,75,76,77) (Fig. 5c and Extended Data Fig. 8b). Human kidney obtained from a non-COVID-19 postmortem examination served as a negative control to ensure valid staining (Extended Data Fig. 8c). In the spleens, just patients 2 and 4 showed some scattered infected cells without gross abnormalities (Fig. 5c and Extended Data Fig. 8b).

We then analysed pancreatic involvement of the four patients. During COVID-19 treatment we observed hyperglycaemia in patients 1, 2 and 3, accompanied by a progressive insulin demand in patients 2 and 3. Pancreatic histopathology revealed the presence of SARS-CoV-2 N protein, with varying numbers of positive cells in all four patients, indicating a persistent infection during severe COVID-19 independent of early (patient 1) or late (patient 2-4) COVID-19 disease stage (Fig. 6a and Extended Data Fig. 8b). Specifically, N protein was detected in some small ducts (CK19) and in single or grouped acinar cells, but was negative in an uninfected control specimen (Fig. 6a and Extended Data Fig. 9), in agreement with the viral entry protein expression pattern (Fig. 1, and Extended Data Figs. 3 and 4). Of note, one patient had elevated lipase levels, indicating a certain degree of exocrine damage due to SARS-CoV-2 infection (Table 1). N-protein-positive cells were not randomly scattered across the human pancreas, but instead occurred in clusters of infected cells, suggesting localized viral spread (Fig. 6a,b). To probe infection of human β-cells, we performed immunohistochemical double staining for the viral N protein and insulin, but observed only a few double-positive cells (Fig. 6a, close-ups marked with a hash). Nevertheless, N-protein-positive cell clusters were located close to the islets of Langerhans, indicating a certain degree of association between SARS-CoV-2 infection and the endocrine compartment. This was quantified by a vicinity score based on the distance between N-protein- and insulin-positive cell clusters, and classified in cells with a distance <100 µm or ≥100 µm against a randomly calculated reference distance. On average, 51% of SARS-CoV-2-infected cell clusters were located close to human islets, with a significant maximum of 60% and 83% in patients 1 and 2, as well as similar trends with 40% and 31% in patients 3 and 4, respectively (Fig. 6b,c). Again, some insulin-positive cells revealed a faint red N-protein signal pointing towards spreading infection (Fig. 6d, arrowheads). The morphology of some of the clearly infected cells did not resemble ductal, acinar or endocrine morphology, indicating a certain degree of plasticity occurring after infection. This is in line with the observed loss of hormones and endocrine granules as suggested by immunostaining and TEM images of infected islet explants (Figs. 2 and 3). Indeed, high N-protein signals appeared to go along with low insulin-staining intensity (Fig. 6d, close-up). To further address this, we co-stained for N protein and NKX6.1, which is exclusively expressed by β-cells within the adult pancreas55. Indeed, we detected N-/NKX6.1-double-positive cells in four out of four patients, in close proximity to the islets of Langerhans and SARS-CoV-2-infected cell clusters (Fig. 6e,f and Extended Data Fig. 10). Under the assumption that clusters of NKX6.1-positive cells represent or are derived from the endocrine compartment, we estimated the percentage of infected endocrine cells on average at 46% (range 23–65%) (Extended Data Fig. 10). This indicates that infection of β-cells might result in hormone loss, an observation matched by our ex vivo analyses (Fig. 2 and Extended Data Fig. 5). To conclude, pancreatic SARS-CoV-2 infection can occur in severe cases of COVID-19, including the exocrine and endocrine compartment. Nevertheless, the still low sample size, suboptimal tissue quality due to autolytic necrosis and limited clinical data precluded a correlation with the individual endocrine function and outcome in patients.

Fig. 6: Pancreatic infection pattern in COVID-19 deceased patients.
figure 6

a, Pancreatic tissue sections from four different COVID-19 deceased patients stained for SARS-CoV-2 N protein (red) and insulin (brown). Rectangles mark areas of higher magnification in the next row. Insets show further high magnification to illustrate specific patterns. Asterisks mark the magnification of stained areas outside the illustrated regions but corresponding to the patients from the respective column. Infection occurred as N-positive clusters in all four patients with positivity of some ductal cells and a few acinar cells. Insulin and N-protein double-positive cells were observed in three out of four patients (marked with a hash). A representative image from one biopsy section of each patient is displayed. b, N-positive cells are often located in close vicinity to the islets of Langerhans or even mixed in islet-like structures together with insulin-positive cells. The morphology of N-positive cells frequently resembled non-acinar/non-ductal morphology. A representative image is derived from one biopsy section of patient 1. c, Vicinity ratio of N-positive regions (>5 positive cells) located near insulin-positive endocrine cells (<100-µm distance to endocrine cell) divided by all N-positive regions (>5 positive cells) reveals that SARS-CoV-2 is not randomly distributed across the pancreas, but rather located close to endocrine structures. On average, 51% of N-positive regions are located close to endocrine cells or islets. For statistical testing, two-sided Fisher’s exact test was applied to the absolute numbers (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001). d, Viral N- and insulin-double-positive cells are rare (arrowheads) but interlaced into the islets of Langerhans. The highest N-protein signal in a cell cluster or even in individual cells correlates with the lowest insulin signal (close-up). e,f, Consecutive sections were stained for either insulin (e) or NKX6.1 (f), together with N protein. Rectangles in (i) mark areas of higher magnification (ii). Dashed areas connected with double-headed arrows mark corresponding regions that are highly N positive, insulin negative and NKX6.1 positive (left) or N negative, insulin positive and NKX6.1 positive (right). Representative images are derived from one biopsy section of patient 2. Scale bars, 100 µm; insets, 10 µm.

Source data

Discussion

The disease course in patients with COVID-19 can be perturbed by diabetes mellitus in two ways. On the one hand, diabetes is a risk factor for severe disease3,16,17,18,19,20,21,22 and, on the other, SARS-CoV-2 infection has been associated with altered glucose metabolism3,16,24,29. Specifically, ketosis and ketoacidosis were observed during and after SARS-CoV-2 infection, both being at least indirect clinical indicators of a lack of insulin due to β-cell loss or malfunction18,25,26,27,28,29,30. In the present study, we: (1) dissected pancreatic expression patterns of SARS-CoV-2 viral entry proteins, and (2) demonstrated permissiveness of β-cells to SARS-CoV-2 infection and replication, which affects (3) subcellular morphology and glucose responsiveness. Moreover, we (4) demonstrated the presence of SARS-CoV-2 viral antigen in pancreata and, most importantly, in NKX6.1-positive β-cells from COVID-19 deceased patients, some of whom had prediagnosed diabetes.

We analysed human pancreatic islets for SARS-CoV-2 entry factor expression and consistently observed ACE2 positivity of intra-islet endothelial cells as previously reported12,13,34,37. We also demonstrated that the endocrine cells of human pancreatic islets express TMPRSS2 as well as the long and short isoforms of ACE2, in line with recent findings12,36. However, partially conflicting data have been reported about the expression of the long ACE2 isoform in human β-cells13,34. These different experimental outcomes might be attributed to intra- and interindividual variations in the investigated donor tissue, or utilization of antibodies that have various affinities to the respective isoforms12,13,34,36. In the present study, we used established antibodies in our immunohistochemistry analysis that detect the short and long isoforms and, in addition, performed an extensive in-house characterization of these antibodies and the applied methodology12,36,49. Fignani et al.12 provide a comprehensive analysis on ACE2 expression in β-cells, including mass-spectrometry-based detection. Briefly, this study showed that the short ACE2 isoform is expressed in human pancreatic islets, where it is preferentially expressed in subsets of insulin-producing β-cells12,49. It is interesting also that Kusmartseva et al. reported detectable mRNA levels of ACE2 in a proportion of endocrine cells13. Of note, other viral entry factors might potentiate low ACE2 expression levels in β-cells. Specifically, neuropilin-1, a factor expressed in pancreatic β-cells80, and the high-density-lipoprotein scavenger receptor B type 1, also expressed in human endocrine cells81,82, have been shown to facilitate SARS-CoV-2 infectivity83,84. However, according to the current state of knowledge, ACE2 expression remains the major determinant of SARS-CoV-2 entry and thus organ tropism. Several proteases can prime the coronavirus S protein of which pancreatic TMPRSS2 expression also matched our infection pattern of COVID-19 deceased patients6,8,37,85,86. We found that ACE2 and TMPRSS2 colocalize less with markers of δ- and α-cells. However, this lower frequency does not preclude infection as suggested by previous data36. We acknowledge that further studies are necessary to reveal the exact entry mechanisms of SARS-CoV-2 into β-cells and to assess infection patterns of other endocrine cells in more detail. However, independent of a potential debate on viral entry protein expression, we observed productive SARS-CoV-2 infection of ex vivo cultured islets using state-of-the-art, molecular virology-based assays, providing evidence for the presence of functional entry factors.

Human pluripotent, stem-cell-derived β-cells can be infected by SARS-CoV-2 (ref. 87). In endocrine cells of human ex vivo cultured Islets of Langerhans, we detected viral proteins, viral RNA and increasing infectious viral titres, and we also visualized SARS-CoV-2 particles inside vacuoles in the perinuclear region56 by TEM. The most striking observation was an enlarged and vacuolized ER–Golgi intermediate compartment, similar to observations in SARS-CoV-2-infected intestinal, kidney and airway epithelial cells56,57,58. The hallmarks of endocrine differentiation, namely secretory granules, are displaced and significantly reduced. However, a more comprehensive TEM-based analysis across a complete viral replication cycle in human islets, as well as more samples from infected patients, is required to draw definite conclusions. Nevertheless, the TEM observations are in line with the affected insulin secretion observed in our study, even though we faced experimental variations across the four investigated islet preparations. Of note, β-cells infected by enterovirus display decreased GSIS and loss of Golgi body structure88. Furthermore, dedifferentiation of β-cells mimicking reversal to a progenitor state accompanied by decreased β-cell-specific gene transcription may occur after viral89 but also chemical90 injury. Our RNA-seq, confocal microscopy analysis and TEM data would be in line with both hypotheses, namely ER stress followed by β-cell degranulation and dedifferentiation. However, pancreatic virus-induced injury can also be a self-potentiating damage driver due to inflammation, recruitment of bystander immune cells and potentially development of autoimmunity, which is specific to β-cells67. In fact, SARS-CoV-2 infection provoked a broad signature of cytokines and ISGs attributed to type I and II IFN responses in human islets. We recently showed that IFN-induced transmembrane (IFITM) proteins promote SARS-CoV-2 infection of human lung cells91. Of note, IFITM1–3 ranged top among upregulated transcripts in SARS-CoV-2-infected human islets. Similar GO terms have been reported in gut-derived organoids after SARS-CoV-2 infection58, identifying such intrinsically triggered immune responses as a general feature across distinct organs during COVID-19. These global studies of islet transcriptomes are, however, limited by the bulk design, preventing access to cell-type-specific, single-cell-resolved transcriptomes92. Finally, our results show that viral replication in ex vivo infected islets was efficiently inhibited by remdesivir used as a control to prevent SARS-CoV-2 replication. This inhibition of viral replication was associated with neither an entire rescue in β-cell function nor full restoration of transcriptomes. This is most probably due to a delay in full β-cell recovery, which cannot be reached in the present experimental setting due to the deterioration of islets on prolonged ex vivo culture. However, the effects of remdesivir suggest that observed changes are specific to SARS-CoV-2 infection.

Investigation of pancreata of COVID-19 deceased patients revealed a scattered distribution of infected cell clusters across the pancreas in all four patients, most visible in the exocrine compartment, but with a closeness to the islets of Langerhans. Such a pattern could indicate spread to neighbouring pancreatic cells originating from a few infected cells, potentially reached by viral particles directly via the bloodstream during temporary viraemia, typically occurring in severe COVID-19 (refs. 93,94). Kusmartseva et al. also investigated N-protein expression in COVID-19 deceased patients, but could not detect co-expression with insulin13. In our study, detection of N-/insulin-double-positive cells was technically challenging and rare in frequency. However, building on our observations in ex vivo cultured islets, co-staining of viral N protein with the β-cell lineage label NKX6.1 confirmed SARS-CoV-2 infection in cell clusters expressing NKX6.1 in all investigated patients with COVID-19. Notably, pancreatic NKX6.1 expression is unique because no other transcription factor is restricted exclusively to β-cells within the adult pancreas55. As we also observed those hormone-negative cells in the human islet preparations, this suggests that SARS-CoV-2 infection might perturb hormone positivity by cytokine and/or ER stress, followed by β-cell degranulation and dedifferentiation. Further analysis is necessary to fully understand the underlying mechanism and it is important to note that hormone loss on infection complicates these experimental evaluations. Specifically, insulin might not be a suitable marker to show colocalization with viral proteins. Thus, estimating infection rates of the respective cells, within islets or pancreatic tissues, remains a challenge. The precise SARS-CoV-2 infection pattern in human islet subpopulations is still warranted, and whether the number of SARS-CoV-2-infected cells and associated identity loss of β-cells suffice to affect endocrine function in healthy and diseased islets remain to be determined. Accordingly, our results prompt the question of whether SARS-CoV-2 directly perturbs β-cell integrity and potentially leads to endocrine dysregulation and causes autoantibody-negative T1DM, as reported in recent clinical studies after SARS-CoV-2 infection9,17,18,19,20,21,22,23,24,25,26,27,28,29,30. Alternatively, SARS-CoV-2 infection might be a precipitating factor of autoimmune-mediated diabetes mellitus arising even years after recovery1,22, underpinning the necessity for long-term follow-up of patients with COVID-1933. Although, for example, Coxsackie B4 or congenital rubella virus infections as such can trigger development of T1DM67,95,96, the virus-mediated β-cell insult presents heterogeneously across different viruses, for instance, enterovirus infection of β-cells can lead to: (1) cell death accompanied by increased proliferation in neighbouring non-infected β-cells, (2) impaired insulin production and secretion, or (3) β-cell dedifferentiation89,97. Collectively, whether SARS-CoV-2 infection triggers a detrimental immune response or directly reduces β-cell function, thus affecting the endocrine system, needs to be evaluated in future studies.

Infection of pancreatic ducts and acinar cells further raises the question of whether there is a correlation with observed lipase level elevations and acute oedematous pancreatitis in patients with COVID-19 with SARS-CoV-2-associated pneumonia. Acute pancreatitis has also been reported to occur in COVID-19 (refs. 5,22,23,98). Recent studies reported acute pancreatitis in 12.6% of the entire cohort and 32.5% in critical patients23. However, it remains unclear whether acute pancreatitis, a potentially fatal disease because it in particular causes deterioration in critically ill patients, occurs as a side effect or a direct consequence of pancreatic SARS-CoV-2 infection.

Analysis of other organs of COVID-19 deceased patients revealed robust staining of the kidneys, whereas the spleen was not consistently infected. Lung tissue showed different infection patterns, which might correspond to differences in temporal and spatial disease presentation71,99. We cannot exclude infection in other lung or spleen regions due to the sampling bias during the postmortem examination and the low sample size. In particular, the question of renal tropism remains under discussion. Of note, there are reports that kidneys are highly susceptible to SARS-CoV-2 infection3,6,71,73,74,75,76,77,78,79, whereas others failed to detect viral elements in kidneys78,100,101,102,103. These discrepancies reveal that our knowledge about (1) the frequencies and (2) the association with pre-existing diseases, and (3) the clinical impact of extrapulmonary infections by SARS-CoV-2 remains incomplete. There are several explanations for these conflicting findings such as spatial and temporal heterogeneity of viral spread, technical constraints such as sample preservation and the detection assays used. Time-resolved, ultrasound-guided tissue biopsies might circumvent the limitations of endpoint analysis in postmortem studies.

Collectively, we demonstrate that the exocrine and endocrine compartments of the pancreas are susceptible to productive SARS-CoV-2 infection, which can perturb β-cell integrity. The mechanism of virus-induced damage and whether infection has a direct consequence for glucose homoeostasis or might even trigger diabetes mellitus remain under discussion and deserve future studies.

Methods

Drugs

Remdesivir was obtained from Selleck Chemicals (catalogue no. S8932).

Cell culture

Vero E6 (Cercopithecus aethiops–derived epithelial kidney, American Type Culture Collection (ATCC)) cells were grown in Dulbecco’s modified Eagle’s medium (DMEM, Gibco) which was supplemented with 2.5% heat-inactivated fetal calf serum (FCS), 100 U ml−1 of penicillin, 100 μg ml−1 of streptomycin, 2 mM l-glutamine and 1 mM sodium pyruvate. Calu-3 (human epithelial lung adenocarcinoma, kindly provided by Professor Frick, Ulm University) cells were cultured in minimum essential medium Eagle (MEM, Sigma-Aldrich, catalogue no. M4655) supplemented with 10% FCS, 100 U ml−1 of penicillin, 100 μg ml−1 of streptomycin, 1 mM sodium pyruvate and 1× non-essential amino acids. EndoC-βH1 cells (Univercell Biosolutions) were cultured on coated culture wells (2 µg ml−1 of fibronectin, 1% extracellular matrix) in OPTIβ1 medium (Univercell Biosolutions) according to the manufacturer’s protocol. Human umbilical vein endothelial cells were cultivated in endothelial cell growth medium (Merck). Cells were grown at 37 °C in a 5% CO2 humidified incubator. For differentiation of stem-cell-derived intestinal organoids, human embryonic stem cell (hESC) line HUES8 (Harvard University) was used with permission from the Robert Koch Institute according to the ‘79. Genehmigung nach dem Stammzellgesetz, AZ 3.04.02/0084’. For differentiation, 300,000 cells per well were seeded on 24-well plates coated with growth-factor-reduced Matrigel (Corning) in mTeSR1 with 10 µM ROCK inhibitor. The next day differentiation was started by washing with phosphate-buffered saline (PBS) and adding d0 differentiation medium (BE1 medium (MCDB131 (Invitrogen), 2 mM l-glutamine (Gibco), 1.174 g l−1 of sodium bicarbonate (Sigma-Aldrich), 0.8 g l−1 of glucose (Sigma-Aldrich), 0.1% fatty-acid-free bovine serum albumin (BSA, Proliant)) with 100 ng ml−1 of Activin A (R&D/PeproTech) and 2 µM CHIR99021 (Axon MedChem). After 24 h and 48 h the medium was changed to d1/d2 medium (BE1 with 100 ng ml−1 of Activin A). From day 3 on, the medium was changed daily to fresh mid-hindgut formation medium (RPMI-1640 (Gibco), 2% FCS (Biochrom), 2 mM l-glutamine (Gibco), 1% penicillin–streptomycin (Sigma-Aldrich), 10 ng ml−1 of BMP4 (Peprotech), 100 ng ml−1 of basic fibroblast growth factor (Novoprotein), 3 µM CHIR99021 (Cayman Chemical Co.)). From day 7 structures started floating in the medium and were collected and plated on 48-well Nunclon plates (Sigma-Aldrich) in Matrigel domes (20–40 µl of Matrigel per dome). Then, 350 µl of intestinal growth medium (DMEM F12 (Gibco), 1× B27 supplement (Thermo Fisher Scientific), 2 mM l-glutamine, 1% penicillin–streptomycin (Sigma-Aldrich), 40 mM 4-(2-hydroxyethyl)-1-piperazine-ethanesulfonic acid (Hepes) solution (Sigma-Aldrich), 3 µM CHIR99021, 200 nM LDN-193189 (Sigma-Aldrich) and 100 ng ml−1 of human epithelial growth factor (Novoprotein)) was added per well. The medium was changed twice a week.

Virus strains and virus propagation

Viral isolate BetaCoV/Netherlands/01/NL/2020 (catalogue no. 010V-03903) of the pandemic D614G variant was obtained from the European Virus Archive global and propagated on Vero E6 cells. To this end, 70% confluent cells in 75-cm² cell culture flasks were inoculated with 100 µl of SARS-CoV-2 isolate in 3.5-ml serum-free medium containing 1 µg ml−1 of trypsin. Cells were incubated for 2 h at 37 °C, before adding 20 ml of medium containing 15 mM Hepes. Cells were incubated at 37 °C and supernatant harvested when a strong cytopathic effect was visible. Supernatants were centrifuged for 5 min at 1,000g to remove cellular debris, and then aliquoted and stored at −80 °C as virus stocks. The infectious virus titre was determined as plaque-forming units on Vero E6 cells, which was used to calculate the multiplicity of infection (MOI).

Isolation of RNA and RT-qPCR

Viral RNA from cells was isolated using the QIAGEN RNeasy Plus Mini Kit (catalogue no. 74136) and RNA from supernatants using the QIAGEN Viral RNA Mini Kit (catalogue no. 52906) as described by the manufacturer. Cells were lysed in 600 μl RLT Plus buffer containing 1% β-mercaptoethanol, vortexed for 30 s and then frozen at −20 °C until further isolation. For supernatants, 140 μl was mixed with 560 μl of AVL buffer, vortexed and frozen as above. Reverse transcription (RT)-qPCR from cells or supernatants was performed with primer sets targeting SARS-CoV-2 N (nucleocapsid) or ORF1b-nsp14 (refs. 108,109) using TaqMan Fast Virus 1-Step Master Mix (Thermo Fisher Scientific, catalogue no. 4444436) and a StepOnePlus Real-Time PCR System (96-well format, fast mode). Expression of ACE2 and TMPRSS2 was evaluated using TaqMan assays (Thermo Fisher Scientific, catalogue nos. Hs01085333_m1 and Hs01122322_m1, respectively). Synthetic SARS-CoV-2-RNA (Twist Bioscience, catalogue no. 102024 or ATCC, catalogue no. VR-3276SD) was used as a standard to obtain viral copy numbers for quantification of viral RNA in supernatants. For RNA isolated from cells, glyceraldehyde 3-phosphate dehydrogenase (GAPDH) was used as an endogenous control (Applied Biosystems, catalogue no. 4310884E) to calculate relative expression. The threshold for detection was set at threshold cycle, Ct, of 35. All reactions were run in duplicate.

Primer sets:

Target N:

Forward primer (HKU-NF): 5′-TAATCAGACAAGGAACTGATTA-3′

Reverse primer (HKU-NR): 5′-CGAAGGTGTGACTTCCATG-3′

Probe (HKU-NP): 5′-FAM-GCAAATTGTGCAATTTGCGG-TAMRA-3′

Target ORF1b-nsp14:

Forward primer (HKU-ORF1b-nsp14F): 5′-TGGGGYTTTACRGGTAACCT-3′

Reverse primer (HKU- ORF1b-nsp14R): 5′-AACRCGCTTAACAAAGCACTC-3′

Probe (HKU-ORF1b-nsp141P): 5′-FAM-TAGTTGTGATGCWATCATGACTAG-TAMRA-3′.

TCID50 endpoint titration

To determine the tissue culture infectious dose 50 (TCID50), supernatant samples were serially diluted and used to inoculate Vero E6 cells. To this end, 20,000 Vero E6 cells were seeded per well in 96-well, flat-bottomed plates in 100 µl of medium and incubated overnight before 62 µl of fresh medium was added. Next, 18 µl of five- or tenfold titrated sample was used for inoculation in triplicate. Cells were then incubated for 4–6 d and monitored for cytopathic effect. The TCID50 per ml was calculated according to Reed and Muench110.

Islet culture

Four preparations of human pancreatic islets were obtained from the Alberta Diabetes Institute IsletCore at the University of Alberta, headed by P. E. MacDonald. Islet material was derived from excess donor organs used for clinical transplantation and with the written informed consent for research (approval no. Pro00013094). The first donor (95% purity), female, aged 31 years, had a body mass index (BMI) of 20.3 and no history of diabetes (glycated haemoglobin (HbA1c): 4.8%). The second donor (50% purity), female, aged 55 years, had a BMI of 29.1 and no history of diabetes (HbA1c: not given). The third donor (40% purity), male, aged 58 years, had a BMI of 31 and no history of diabetes (HbA1c: 5.6%). The fourth donor (90% purity), female, aged 53 years, had a BMI of 33.2 and no history of diabetes (HbA1c: 6.1%). This study was also approved by ‘Ethikkommission TUM’ no. 394/20S. After shipping, islets were washed in culture medium and seeded into ultra-low attachment, 12-well plates (Corning) at a density of 400 islets per well. Infection experiments were started 48 h after recovery. Pancreatic islets were cultivated in CMRL1066 (Gibco) supplemented with 10% human serum (Sigma-Aldrich), 2 mM l-glutamine (Sigma-Aldrich), 1% penicillin–streptomycin (Sigma-Aldrich) and 25 mM Hepes buffer (Sigma-Aldrich). The medium was replaced every second day.

Islet infection

Pancreatic islets of donor 1 in 800 µl of medium were infected by adding 200 µl of virus inoculum, resulting in an MOI of ~1. To achieve higher infection rates, islets of donors 2–4 were pre-incubated with TrypLE for 5 min, the reaction stopped by adding BSA-containing DMEM/F12, the supernatant discarded and the islets infected with an MOI of ~2.5. After 3 h of incubation, islets were washed three times and cultured in 1 ml of medium per 24 wells or 4 ml per 6 wells, which was in one setting supplemented with 5 µM remdesivir. At indicated time points, 0.5 or 2 ml of medium supernatant was collected for RT-qPCR and TCID50 analysis and medium was refilled respectively.

Static GSIS

Static GSIS was performed 3 d post-infection. Pancreatic islets were washed in Krebs–Ringer bicarbonate buffer (KRBH) containing 0.1% BSA and incubated for 1 h in KRBH with 2 mM glucose. After 1 h, islets were washed in KRBH/BSA and resuspended in a buffer containing 2 mM glucose. After 1 h of incubation, the supernatant was taken and stored for insulin quantification; the islets were washed in KRBH and incubated in KRBH with 20 mM glucose for another hour. The supernatant was taken and stored for insulin quantification. The GSIS procedure was performed for one-islet sample (donor 1) and three-islet samples (donors 2 and 4). The supernatants from static GSIS were analysed in duplicate with the ultrasensitive insulin ELISA Kit (Alpco), according to the manufacturer’s instructions. Of note, stimulation of one uninfected islet sample (donor 2) with 2 mM glucose did not result in detectable insulin secretion. Values were normalized to cell numbers or protein content (quantified by BCA assay, Thermo Fisher Scientific, catalogue no. A53227) and the fold induction of insulin secretion (20 mM glucose stimulation compared with 2 mM glucose) was calculated for each sample.

TEM of islets

Sample preparation for TEM was done as described previously111. To this end, pancreatic islets were washed once with PBS and fixed with 2.5% glutaraldehyde containing 1% saccharose in phosphate buffer, pH 7.3. Samples were washed with PBS and post-fixed in 2% aqueous osmium tetroxide. After dehydrating the samples in a graded series of 1-propanol, they were stained in 2% uranyl acetate and embedded in Epon. Ultra-thin sections (80 nm) were collected on copper TEM grids, contrasted with 0.3% lead citrate for 1 min and imaged in a Jeol TEM 1400 at 120 kV.

Histology of healthy pancreatic tissue sections and ex vivo culture human islets

Sections of human pancreas were provided by the pathology department of Ulm University. Non-neoplastic pancreatic tissue integrity was approved by a board-certified pathologist (T.F.E.B.). Experiments were conducted in accordance with the guidelines of the Ethics Committee of the Federal General Medical Council and approved by the Ethics Committee of the University of Ulm (vote for usage of archived human material 03/2014). Sections were deparaffinized, rehydrated and subjected to heat-mediated antigen retrieval in Tris buffer, pH 9, or citrate buffer, pH 6. Tissue was permeabilized with 0.5% Triton X-100 for 30 min at room temperature, and stained overnight with primary antibodies (Extended Data Fig. 1) in antibody diluent (Zytomed) in a wet chamber at 4 °C. After washing with PBS–Tween 0.05% (PBS-T), slides were incubated with secondary antibodies (Alexa Fluor IgG H + L, Invitrogen, 1:500) and 500 ng ml−1 of 4′,6-diamidino-2-phenylindole (DAPI) in antibody diluent for 90 min in a wet chamber at room temperature. After washing with PBS-T and water, slides were mounted with Fluoromount-G (Southern Biotech). For antibody-blocking experiments, ACE2 primary antibody with blocking peptide (Extended Data Fig. 1) at sevenfold excess protein amount was incubated for 30 min at room temperature in antibody diluent before staining.

For histological examination, pancreatic islets were fixed in 4% paraformaldehyde (PFA) overnight. Then, they were incubated in 1 M sucrose in PBS overnight and embedded in O.C.T. freezing compound (Tissue-Tek). Cryoblocks were sectioned at 7 µm and slides stored at −80 °C. Immunofluorescence staining was performed similarly to paraffin sections, except for the deparaffinization, rehydration and antigen retrieval steps, which were omitted. Washing in between staining steps was performed with 0.1% Triton X-100 in PBS.

Negative controls were performed using immunoglobulin (Ig)G controls or irrelevant polyclonal serum (anti–Mycobacterium tuberculosis) for polyclonal antibodies, respectively. The absence of background staining confirmed the specificity of the primary antibodies. Laser scanning confocal images were acquired using the Zeiss LSM710. Alternatively, some images shown in Figs. 1 and 2 were acquired using a Leica TCS SP8 equipped with a HC PL APO CS2 63×/1.2 WATER immersion objective. Images were acquired in sequential scan mode as single confocal airy sections using HyD-detectors and the following detection range: Ex: 405 nm, Em: 430–470 nm; Ex: 488 nm, Em: 500–535 nm; Ex: 561, Em: 571–620 nm. Colocalization was quantified using the Coloc2 quantification tool of Fiji and standard settings. The fraction of SARS-CoV-2 N-protein-positive cells (Fig. 2c) was quantified in infected islets of donors 2 and 3 (days 3 and 5), stained for SARS-CoV-2 N protein. N-protein-positive cells were counted in seven (day 3) and six (day 5) islet clusters positive for N protein and the fraction of the total cell population is shown in Fig. 2c. Uninfected islets from each donor (n = 4) served as a control.

Images of SARS-CoV-2-infected islets of donor 3 (day 5) stained for NKX6.1 and SARS-CoV-2 N protein (Extended Data Fig. 6d) were acquired using a Zeiss ApoTome. NKX6.1-positive cells and N-protein-positive cells were quantified in 11 islet clusters positive for N protein and the fraction of NKX6.1 and N-protein double-positive cells of the total NKX6.1-positive cell population is shown in Extended Data Fig. 6e.

Immunohistology of tissue sections of SARS-CoV-2-infected patients

Tissue samples were provided by the tissue bank of the German Center for Infection Research (DZIF, Heidelberg, Germany, approval S242/2020) and of the Institute of Pathology, Ulm University. The characteristics of the patients are reported in Table 1. Immunohistochemical double stainings were performed by sequential incubation with antibodies from different species against nucleocapsid (N protein) of SARS-CoV-2 (mouse, 1:50 in situ, 1:100 in vitro) and against insulin as a marker for endocrine β-cells (rabbit, 1:10,000 in situ, 1:5,000 in vitro), NKX6.1 (mouse, 1:75 in situ) and CK19 (mouse, 1:100 in situ) as ductal markers. Antigen retrieval was carried out by treatment of the slides in citrate buffer, pH 6.1 in a steamer for 25 min. Specific antigen binding was ensured by incubation of all used antibodies for 30 min at room temperature. Detection of antibodies was by alkaline phosphatase/RED Detection system (Dako REAL Detection System) for N protein and DAB Chromogen System (Dako EnVision HRP DAB System) for insulin and NKX6.1. For imaging, pictures were taken in a bright field using a Zeiss Axiophot microscope with an included CCD-camera (JVC Digital Camera KY-F75U) connected to Diskus Viewer software (v.5.0).

The distance from N-protein-positive cell clusters to insulin-positive cells (vicinity score) was quantified using Leica LAS-X imaging software and classified in cells with a distance <100 µm or ≥100 µm. To allow statistical testing, a reference distance was generated (expected numbers of cells in islet proximity). For this, cells were randomly selected using LAS-X imaging software and the respective distance of such reference cells to islets was quantified. For statistical testing, Fisher’s exact test was applied to the absolute numbers in the contingency tables. For illustration only the fraction of cells, not the absolute numbers of cells, with a proximity <100 µm to C-pep-positive islets is displayed (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001).

Under the assumption that clusters of NKX6.1-positive cells (>10 positive nuclei in close proximity, with both preserved and disturbed islet morphology) represent or are derived from the endocrine compartment, we estimated the number of infected clusters in non-necrotic areas of the specimens (defined as preserved acinar morphology). To this end, 20 randomly assigned clusters were evaluated for N-protein expression and clusters harbouring more than five double-positive cells were assumed to be infected by SARS-CoV-2. A more detailed analysis and quantification were hindered by large areas of necrotic tissue in postmortem pancreatic tissue (percentage of necrotic area given in Extended Data Fig. 10b).

RNA-seq

NGS library preparation

For the preparation of next-generation sequencing (NGS) libraries from low-input samples of cellular RNA isolates, we followed the Smart-seq2 protocol112. Library concentrations were quantified with the Qubit 2.0 Fluorometric Quantitation system (Life Technologies) and the size distribution was assessed using the Experion Automated Electrophoresis System (Bio-Rad). For sequencing, samples were diluted and pooled into NGS libraries in equimolar amounts.

Sequencing and raw data processing

Expression profiling libraries were sequenced on a HiSeq 4000 instrument (Illumina) in 50-bp, single-end mode. Base calls, provided by the real-time analysis (RTA, v.2.7.7) software (Illumina), were subsequently converted into multiplexed, unaligned BAM format before demultiplexing into sample-specific, unaligned BAM files. For raw data processing off the instruments, customized programs, based on Picard tools (v.2.19.2)113, were used.

Transcriptome analysis

NGS reads were mapped to the Genome Reference Consortium GRCh38 assembly via ‘Spliced Transcripts Alignment to a Reference’ (STAR, v.2.7.5a)114 using the ‘basic’ Ensembl transcript annotation from v.e100 (April 2020) as reference transcriptome. As the hg38 assembly flavour of the University of California, Santa Cruz (UCSC) Genome Browser was preferred for downstream data processing with Bioconductor packages for entirely technical reasons, Ensembl transcript annotation had to be adjusted to UCSC Genome Browser sequence region names. STAR was run with options suggested by the ENCODE project. Aligned NGS reads overlapping Ensembl transcript features were counted with the Bioconductor (3.11) GenomicAlignments::summarizeOverlaps() function (1.24.0)115, taking into account that the Smart-seq2 protocol is not strand specific. Transcript-level counts were aggregated to gene-level counts and the Bioconductor DESeq2 package (v.1.28.1)104 was used to test for differential expression based on a model using the negative binomial distribution.

An initial exploratory analysis included principal component analysis, multidimensional scaling, sample distance and expression heatmap plots, all annotated with variables used in the expression modelling (CRAN ggplot2 v.3.3.2 (ref. 116), Bioconductor ComplexHeatmap v.2.4.3 (ref. 117)) as well as volcano plots (Bioconductor EnhancedVolcano v.1.6.0 (ref. 118)). Sample distance heatmaps were obtained via the CRAN pheatmap package119 by calculating the Euclidean distance matrix (via R stats::dist()) of DESeq2 normalized count values after variance-stabilizing transformation in model-aware mode (via DESeq2::varianceStabilizingTransformation) and subsequent hierarchical clustering with the complete linkage method (via R stats::hclust()). Biologically meaningful results were extracted from the model, log2(fold-change) (log2(FC)) values were shrunk with the CRAN ashr package (v.2.2-47)120 and P values were adjusted using the Bioconductor Independent Hypothesis Weighting (IHW, v.1.16.0) package121. The resulting gene lists were annotated, filtered for significantly differentially up- and downregulated genes (adjusted P < 0.1 and log2(FC) > |1|), and were independently subjected to overrepresentation analyses. While g:Profiler [v.: e100_eg47_p14_7733820]105 was applied to test enrichment of significantly differentiated genes against gene sets in common databases, EnrichR allowed expansion to COVID-19-related disease terms (Enrichr)106,122. For GSEA, the gsea desktop tool107 was used (GSEA v.4.0.3) with setting ‘Permutation Type’ to ‘gene_set’ and ‘Metric for ranking genes’ to ‘Signal2Noise’. Enrichment was tested against hallmark gene sets from MSigDB 7.2 (ref. 123) and an additional β-cell gene cluster derived from Ackermann et al.66. Enrichment scores were cross-checked with GSEA Preranked tool using log2(FC) as ranking parameter, yielding highly similar results.

Whole cell and tissue lysates

To determine expression of cellular proteins, frozen pancreatic tissue was pulverized with a liquid nitrogen-cooled mortar and pestle, and directly transferred into a tube on dry ice to reduce protein degradation. Pulverized tissue samples as well as cell pellets were washed in PBS and lysed in western blot lysis buffer (150 mM NaCl, 50 mM Hepes, 5 mM ethylenediaminetetraacetic acid, 0.1% NP-40, 500 μM Na3VO4 and 500 μM NaF, pH 7.5) supplemented with protease inhibitor (Roche). After 5 min (cell lysate) or 20 min (tissue lysate) of incubation on ice, the samples were centrifuged (4 °C, 20 min, 14,000 r.p.m.) to remove cell debris. The supernatant was transferred to a fresh tube, the protein concentration was measured using the BCA assay (Thermo Fisher Scientific, catalogue no. A53227) and the concentration was adjusted using western blot lysis buffer.

Sodium dodecylsulfate–polyacrylamide gel electrophoresis and immunoblotting

For western blotting, whole cell lysates were mixed with 4× protein sample-loading buffer (LI-COR, at a final dilution of 1×) supplemented with 10% β-mercaptoethanol (Sigma-Aldrich), heated at 95 °C for 10 min, separated on NuPAGE 4–12% Bis–Tris Gels (Invitrogen) for 90 min at 110 V and blotted on to Immobilon-FL poly(vinylidene fluoride) membranes (Merck Millipore). The transfer was performed at a constant voltage of 30 V for 30 min. After the transfer, the membrane was blocked with 1% casein in PBS (Thermo Fisher Scientific, catalogue no. 37528). Proteins were stained using primary antibodies (Extended Data Fig. 1) against ACE2 (1:1,000, Abcam), TMPRSS2 (1:1,000, Abcam) and GAPDH (1:5,000, Abcam, Bio-Rad), and infrared dye-labelled secondary antibodies (LI-COR IRDye; Bio-Rad StarBright). Uncropped and unprocessed western blots are provided with Source data.

Immunofluorescence of EndoC-βH1 cells

EndoC-βH1 cells were seeded in an 8-µm IBIDI slide. After 48 h, cells were fixed with 1% PFA for 15 min at room temperature. Cells were treated with 0.1 M glycine for 10 min at room temperature, followed by permeabilization with 0.25% Triton X-100 in PBS for 10 min and subsequent blocking with 3% BSA in 0.1% Triton X-100 for 30 min at room temperature. After blocking, primary antibodies were added, diluted in 5% normal donkey serum in 0.1% Triton X-100 in PBS and incubated for 2 h at room temperature. Cells were washed twice and incubated with secondary antibodies and DAPI in 5% normal donkey serum in 0.1% Triton X-100 in PBS for 1 h at room temperature in the dark. For ACE2-blocking experiments, ACE2 primary antibody with and without blocking peptide (Extended Data Fig. 1) at sevenfold excess protein amount was incubated for 30 min at room temperature in the respective buffer.

Cell culture and immunofluorescence staining of HAECs

Differentiated HAECs were generated as described by Winkelmann et al.124. All experiments were performed with approval of the ethics committee of the Medical School Hannover (project no. 2701-2015). Briefly, 3.5 × 104 basal epithelial cells obtained from several donors were seeded on collagen-coated 6.5-mm Transwell filters (Corning Costar). On reaching confluency (~48 h), the apical side medium was removed (air lifting) and the basolateral side replaced with air–liquid interface differentiation medium (DMEM-H and LHC Basal (1:1), Thermo Fisher Scientific, supplemented with Airway Epithelial Cell Growth Medium Supplement Pack, Promocell). To avoid mucus accumulation, cells were washed with PBS every 3 d from day 14. Then 25–30 d after air lifting, cells were fixed with 4% PFA in PBS for 30 min and permeabilized for 10 min with 0.2% saponin and 10% FCS (Thermo Fisher Scientific) in PBS. Cells were stained with anti-ACE2 (1:750, Abcam) or anti-TMPRSS2 (1:250, Abcam), and anti-α-tubulin antibody (1:500, Thermo Fisher Scientific; Extended Data Fig. 1) diluted in PBS, 0.2% saponin and 10% FCS overnight at 4 °C. Subsequently, cells were washed twice with PBS and incubated for 1 h at room temperature in PBS, 0.2% saponin and 10% FCS containing Alexa Fluor-488-labelled anti-rabbit secondary antibody, Alexa Fluor-647-labelled anti-rat secondary antibody (1:500; Thermo Fisher Scientific) and DAPI (1:5,000; Thermo Fisher Scientific). Images were acquired on an inverted confocal microscope (Leica TCS SP5) using a 40× lens (Leica HC PL APO CS2 40×1.30 OIL). Images for the blue (DAPI), green (Alexa Fluor-488) and red (Alexa Fluor-647) channels were taken in sequential mode using appropriate excitation and emission settings.

Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary linked to this paper.