The experimental procedure of this study
Postmortem tissue samples were collected during the autopsy of 3 patients who were deceased from respiratory failure caused by SARS-CoV-2 infection at Wuhan Jinyintan Hospital (Fig. 1a). We collected the samples of lung and muscle from Patient 1, the samples of lung, heart, liver, spleen, kidney, intestine, brain and muscle from Patient 2, and the samples of lung, heart, liver, spleen, kidney, brain and muscle from Patient 3 (Fig. 1a). Besides, lung paracancerous tissue samples from two lung cancer patients were collected for comparison. For each tissue sample, total proteins were extracted and processed by trypsin, and the resulting peptides were subjected to tandem mass tag (TMT) labeling and analyzed by liquid chromatography with tandem mass spectrometry (LC-MS/MS) (Fig. 1b).
We analyzed the pathology of pulmonary autopsy specimens from patients 2 and 3. The main pathological change of the post-mortem lung tissues from two patients was diffuse alveolar damage (Fig. 1c) which is similar with that caused by SARS14. The histology was represented mainly by a widespread destruction of pulmonary architecture, with extensive fibromyxoid exudate, alveolar haemorrhage, formation of hyaline membranes, and interstitial thickening. In addition, the ultrastructure of these lung tissue samples under transmission electron microscopy revealed several virion- like particles in alveolar epithelial cells (Fig. 1d). These virion-like particles were approximately 80-120 nm in diameter, with spiky-like projections on the surface and typical electron lucent center, which display typical coronavirus morphology of SARS-CoV-2 virion15. Furthermore, the immunofluorescent staining assays were performed to detect the presence of SARS-CoV-2 nucleocapsid protein (NP) in lung tissue samples. As shown in Fig. 1e, green fluorescence of NP protein was observed, showing the presence of SARS-CoV-2 antigens in the lung tissues of these patients.
A protein atlas of eight COVID-19 postmortem tissue types
From the LC-MS/MS analysis, we obtained 49,815 non-redundant peptides, with a number ranged from 36,046 to 37,855 peptides in 3 lung, 2 kidney, 2 liver, 1 intestine, 2 brain, 2 heart, 3 muscle and 2 spleen samples of COVID-19 postmortem tissues, as well as 2 normal lung samples (Fig. 2a). These peptides were mapped to their corresponding protein sequences, and we used the reporter ion MS2 module of the MaxQuant software package for protein quantification 16. From the results, we observed that 5346 human proteins were quantified in at least one sample (Table S1), with protein numbers ranged from 4776 to 5000 (Fig. 2b). Furthermore, we evaluated the quality of the proteomic data by checking the original MS/MS data, and found that the average spectral counts of all peptides were 2.66, with 29,618 peptides (59.5%) matched by ≥ 2 spectral counts (Fig. 2c). Thus, our results indicate that the proteomic profiling is highly reliable at the peptide level. Both human and SARS-CoV-2 protein sequences were included for database search, while no viral proteins were detected in any tissue samples, probably due to the background of large amount of host proteins. To eliminate the sample loading difference, a z-score plus median centering method was used to individually normalize the proteomic data for each sample, and the normalized protein expression (NPE) value was determined for each protein (Fig. 2d, Table S2). Using NPE values of proteins, a principal component analysis (PCA) revealed that different COVID-19 postmortem tissue types could be roughly separated (Fig. 2e). Moreover, we used an entropy-based method17,18 to identify 226 potential tissue-specific proteins (TSPs), including 158 TSPs in brain and 68 TSPs in other tissues, respectively (Fig. 2f, Table S3). This result is consistent with the existing knowledge, since brain is one of the most specialized organs in the human body. Thus, it’s not surprised that brain has most potential TSPs.
Next, a hierarchical clustering was conducted for all proteins in the eight tissue types, and the result was visualized by a software package named Heatmap Illustrator (HemI)19. Obviously, different tissue types had distinct molecular signatures, and potential TSPs can be directly recognized from the heatmap (Fig. 2g). Based on the annotations of GeneCards (https://www.genecards.org/)20, a comprehensive database for human genes, several TSPs were picked out and shown for each tissue sample (Fig. 2g and Fig. S1-S2). For example, UMOD/uromodulin, a known kidney-specific protein 21, exhibited much higher expressions in two kidney samples than other tissues (Fig. 2h). Also, S100A8 and its partner S100A9, the Ca2+ binding proteins that play important roles in regulating pro-inflammatory response22, are only highly expressed in lung samples (Fig. 2h and Fig. S1). In addition, NRGN/Neurogranin, a critical regulator in neurodevelopment and cognition23,24, exhibited higher expressions only in two brain samples. Taken together, our proteomic profiling revealed a landscape of differential protein expressions in COVID-19 postmortem tissue types.
Proteomic alterations reveal that human tissues are differentially affected in response to COVID-19
To probe the protein changes upon SARS-CoV-2 infection, we downloaded the proteomic datasets of six normal human tissues from the Human Proteome Map (HPM) 25, with a number of quantified proteins ranged from 12,007 to 16,868 (Fig. 3a). Compared to HPM, > 96.0% of proteins quantified in this study were also covered by HPM, indicating the high quality of our proteomic profiling (Fig. 3b). To enable an unbiased comparison between COVID-19 and normal samples, the same z-score plus median centering method was used to individually normalize each dataset (Fig. 3c).
To identify differentially expressed proteins (DEPs), we used a tool named Model- based Analysis of Proteomic data (MAP) to analyze each pair of COVID-19 and normal tissues26. Muscle and spleen samples were not analyzed due to the lack of the corresponding normal tissues data in HPM. In contrast with conventional statistical methods, MAP did not estimate technical and systematic errors from technical replicates. Based on a hypothesis that technical and systematic errors might be approximately identical for quantified proteins within a small window, the standard normal distribution was adopted to model the proteomic data and directly calculate a p- value for each protein. In total, we identified 2604, 611, 212, 173, 51 and 42 potential DEPs in lung, kidney, liver, intestine, brain and heart tissue samples, respectively (Fig. 3d, Table S4, adjusted p-value < 0.05). In particular, we revealed numerous alterations of host proteins in different organ that may contribute to the pathogenesis of COVID-19. For example, the protein levels of APC, AKT1 and S100A8/A9 were significantly elevated in postmortem lung tissues (Fig. S3). Among them, S100A8 and S100A9 are acute phase proteins whose alterations are usually in response to inflammation, infection and injury, and can promote inflammatory cytokines release and immune cell migration22. Besides, APC can increase T lymphocyte activation through nuclear factor of activated T cells27, AKT1 is essential for microvascular permeability and leukocyte recruitment and extravasation during acute inflammation28, and RNF14 is a transcriptional co-activator involved in immune response and mitochondrial function29. Interestingly, we found that a number of TSPs such as S100A8/9, UMOD and NRGN were also DEPs in the same tissue types, and all these proteins were significantly up-regulated in COVID-19 postmortem tissues. In addition, we found that many important proteins were down-regulated in different postmortem tissues. For example, ALB and HBB, two fundamental proteins in plasma, were significantly down-regulated in all the six tissue types, especially in liver and kidney (Fig. 3d).
The count of potential DEPs across the six postmortem tissue types demonstrated that lung tissues harbored the greatest number of DEPs (70.5%, 2604/3693), following with kidney (16.5%), liver (5.7%), intestine (4.7%), brain (1.4%) and heart (1.1%) (Fig. 3E). In particular, overlaps were less observed through a comparison of different tissues, and 583 DEPs in lung were mutually shared by other tissues (Fig. 3f) and 57 DEPs were shared by at least four tissues (Fig. 3f and Fig. S4). Taken together, our results indicate that lungs presented the most significantly protein alterations in response to COVID-19 in all the tissues examined, implying that lung represents the major organ for SARS-CoV-2-host interactions, while other tissues are also affected by COVID-19.
Fundamental biological processes are distinctly impacted in different tissues of COVID-19 patients
To identify biological processes up- or down-regulated in the six COVID-19 postmortem tissue types, we used a tool named Gene Set Enrichment Analysis (GSEA), which was developed based on the Kolmogorov–Smirnov test30. No statistically over- represented processes were detected in postmortem kidney, intestine and heart samples, whereas 16, 3 and 15 enriched processes were identified from lung, liver and brain of COVID-19 tissues, respectively (Fig. 4a, Table S5). In postmortem lung tissues, up- regulated processes were mainly focused on immune response- and inflammation- related process, such as humoral immune response (GO:0006959), complement activation (GO:0006956), and B cell mediated immunity (GO:0019724), which were not significantly elevated in other tissue types (Fig. 4b and Fig. S5). On the other hand, the cell morphology maintenance-related pathways, such as establishment endothelial barrier, were only found to be downregulated in lung tissues (Fig. S5), supporting the hypothesis that lung is the potentially major virus-host battlefields of COVID-19. Three metabolism-related processes were up-regulated in liver, and a number of neuron- specific processes were enriched in brain (Fig 4b), indicating that organ-specific functions were altered upon SARS-CoV-2 infection. Top 10 mostly changed DEPs were separately shown for three processes including humoral immune response (GO:0006959) and complement activation (GO:0006956) in postmortem lung as well as nicotinamide adenine dinucleotide (NADH) metabolic process in postmortem kidney (Fig. 4c).
In particular, we observed that many fundamental processes involved in organ movement, respiration and metabolism were dramatically down-regulated in the six postmortem tissue types, with a number ranged from 6 (heart) to 72 (kidney) (Fig. 4a). In total, there were 15 basic processes, such as actin filament-based movement (GO:0030048), NADH metabolic process (GO:0006734) and glucose catabolic process (GO:0006007), were significantly down-regulated in ≥ 4 tissue types (Fig. 4d and Table S5). Based on these results, it could be proposed that brain and heart were less affected by COVID-19 in the aspects of the numbers of DEPs and altered processes, whereas in addition to lung, kidney and liver also significantly affected (Fig. 3e and 4a). Taken together, these results indicate that the responses of distinct tissues in response to COVID-19 are different in critically ill conditions.
A COVID-19-associated protein-protein interaction network
We sought to map the protein-protein interactions between SARS-CoV-2-encoded proteins and DEPs by using a published interactome data of SARS-CoV-2 proteins 31. We obtained 110 known virus-host protein-protein interactions (PPIs) between 23 viral proteins and 110 interacting DEPs differentially regulated in postmortem lung tissues (Table S6). Other lung DEPs were also included for modeling an integrative virus-host molecular network. As shown in Fig. 5, these interacting DEPs were classified into 6 groups according to their functions, including immune response, metabolic process, transcription/translation, cell signaling/development, transport, and cytoskeleton organization, which are participate in almost all the major biological functions in host. Moreover, Gene Ontology (GO) analysis showed that these DEPs were generally involved in several immune response-related processes, including Rab protein signal transduction, blood coagulation and neutrophil degranulation (Fig. S6 and Table S7), which are consistent with the previous findings that cytokine storm, alveolar macrophage activation, intravascular coagulation and microthrombosis are frequently presented in severe COVID-19 cases13,32. Together, these results suggest that SARS- CoV-2-encoded proteins might directly affected the functions of the interacting host proteins in infected lungs.