Accumulating evidence suggests dysregulated immune checkpoint (IC) signaling can exacerbate COVID-19 severity, but the role of these molecules in the pathogenesis of fatal COVID-19-related diffuse alveolar damage (DAD) remains elusive. Understanding how IC proteins influence acute lung injury due to COVID-19 can provide insights into potential therapeutic strategies to modulate immune responses and improve patient outcomes. Here, in a single-center autopsy cohort, we determined the cellular localization of ICs in lung tissue from cases of fatal COVID-19, DAD-comparators, and non-fibrotic controls by using immunohistochemistry, and investigated their association with clinical outcomes. We expanded our findings by performing analyses of publicly available single-cell RNA sequencing datasets from patients with fatal COVID-19 and non-fibrotic controls. We demonstrated the presence of protein-protein interaction networks of ICs in the lung cellular niche by performing transcriptomic profiling of lung tissue-derived RNA counts from patients with fatal COVID-19. Further, we leveraged data from the largest international, multi-center COVID-19 autopsy cohort and validated that, among patients with fatal COVID-19, those with higher PD-L1/CD274 expression in lung endothelial cells had more rapid clinical deterioration. Lastly, in a cohort of individuals with early COVID-19, IC plasma protein levels were elevated in those with persistent SARS-CoV-2 RNAemia and adverse clinical outcomes. Collectively, our data provide unique pathological insights into the role of IC dysregulation and differential disease severity in COVID-19.
Article
Immune checkpoint dysregulation in COVID-19 pathogenesis and disease severity.
https://doi.org/10.21203/rs.3.rs-3508654/v1
This work is licensed under a CC BY 4.0 License
Version 1
posted
You are reading this latest preprint version
Immune checkpoints
COVID-19
disease severity
Acute lung injury (ALI) due to COVID-19 results from the interplay between immune, endothelial, stromal, and lung epithelial cells following SARS-CoV-2 infection[1]. Accumulating evidence suggests immunosuppressive signaling can exacerbate COVID-19 severity[2][3][4] and that soluble inhibitory immune checkpoint (IC) receptors and their ligands, antigen-independent regulators that help maintain immune system homeostasis, may serve as prognostic biomarkers in COVID-19[3][5][6][7][8]. Some studies have shown a higher expression of IC markers in peripheral T-cells from patients with severe COVID-19 compared with healthy controls and patients with milder disease[3][7][8][9][10][11][12], suggesting a propensity for T-cell exhaustion during severe SARS-CoV-2 infection. However, the role of these molecules in the pathogenesis of SARS-CoV-2-related diffuse alveolar damage (DAD) and their cellular localization within the lung microarchitecture have not been investigated to date [7]. An improved understanding of the host immune responses governing lung injury is crucial to the development of novel therapeutics to prevent DAD in the context of COVID-19, particularly due to unequal global vaccination coverage and the constant threat of new variants of concern that may possess increased transmissibility, virulence, or immune evasion[13], especially with reinfections[14].
In this study, we analyzed tissue protein cellular expression/distribution of IC molecules — PD-L1/CD274, HAVCR2/TIM-3, and PD-1— in a single-center autopsy cohort including lung tissue from fatal COVID-19 cases, pre-pandemic fatal cases of lethal respiratory failure, and healthy controls. We cross-checked these findings with publicly available single-cell and single-nucleus RNA sequencing (sc/snRNA-Seq) from COVID-19 decedents and healthy controls. We performed transcriptomic profiling of formalin-fixed paraffin-embedded (FFPE) tissue-derived RNA counts, gene expression correlations, and potential interaction/co-expression patterns of markers of interest. We then investigated the association of IC tissue protein expression/distribution with clinical outcomes in two COVID-19 autopsy cohorts, including the world’s largest international COVID-19 autopsy tissue repository. Lastly, we performed proteomics analyses to evaluate the relationship between IC plasma levels, SARS-CoV-2 RNAemia, and disease severity in patients with early COVID-19 presenting to the emergency department (ED) (Fig. 1).
Lung parenchymal expression of immune checkpoints in COVID-19
To study the microarchitectural localization of candidate ICs during SARS-CoV-2 infection, we performed immunohistochemical (IHC) cellular analyses in lung tissue (Fig. 2A-E) obtained from an autopsy cohort of patients with COVID-19 (n = 20; referred to as the MGH autopsy discovery cohort as described in Methods) matched with pre-pandemic autopsies (n = 21) from patients with lethal acute respiratory failure and histologic evidence of DAD due to both viral and non-viral etiologies (referred to as DAD controls). The groups were similar in demographics, clinical characteristics (e.g., hypoxemia severity within 24 hours of intubation), and therapies received during hospitalization, and there were no significant differences in immunosuppression (Supplementary table 1). We also studied IHC expression in non-fibrotic lung parenchyma without evidence of ALI obtained from age-matched, healthy individuals (n = 11, referred to as healthy controls) that underwent resection of benign tumor-like noncancerous malformations (pulmonary hamartoma). Clinical characteristics of the controls are listed as Supplementary Information (Supplementary tables 2 and 3).
First, we studied IHC expression of PD-L1/CD274, a transmembrane IC protein that plays a major role on adaptive immune system suppression[15]. Regarding protein cellular expression, most compelling was the distinct difference in PD-L1/CD274 expression in endothelial cells (p = 0.029, Fig. 2B and F). COVID-19 decedents exhibited multifocal-to-diffuse PD-L1/CD274 endothelial immunostaining (Supplementary Fig. 1) and a significantly higher mean PD-L1/CD274 endothelial IHC score compared to healthy controls (p = 0.008). The difference in PD-L1/CD274 endothelial IHC expression between COVID-19 and DAD controls approached statistical significance (p = 0.059) but was non-significant. We observed focal-to-patchy/diffuse PD-L1/CD274 alveolar macrophage immunostaining (Fig. 2F) in tissues from nearly almost all decedents with no significant differences between etiologies (p = 0.51, Fig. 2F). PD-L1/CD274 expression in epithelial cells (Fig. 2D) was rare across all cases. While overall differences in epithelial immunostaining between etiologic groups were not significant (p = 0.057, Fig. 2F), DAD controls had significantly higher epithelial PD-L1/CD274 expression compared to COVID-19 decedents (p = 0.02) and healthy controls (p = 0.02). No significant differences were found between individual viral and non-viral DAD etiologies (p = 0.32). Next, we investigated the cellular expression of PD-1, an immune regulatory surface receptor on T and B cells that binds to PD-L1 and PD-L2[16]. Lung parenchymal PD-1 immunostaining was less prominent, but significantly higher in COVID-19 patients compared to all DAD comparators (p < 0.001; non-viral DAD p = 0.004; viral DAD p = 0.04) and healthy controls (p = 0.01) (Fig. 2G). Next, we studied HAVCR2/TIM-3 (Fig. 2E), a cell surface receptor that mediates T-cell exhaustion through proliferation of inhibitory cytokines in chronic viral infections[17]. We observed significant differences in HAVCR2/TIM-3 expression in macrophages (p = 0.002) and lymphocytes (p < 0.0001) across all etiologies (Fig. 2H). Specifically, COVID-19 cases had a higher mean HAVCR2/TIM-3 macrophage immunostaining score compared to all DAD controls (all DAD comparators p = 0.0006; non-viral DAD p = 0.01; viral DAD p = 0.002) but not healthy controls (p = 0.07). COVID-19 cases had a higher mean HAVCR2/TIM-3 + lymphocyte immunostaining score compared to non-viral DAD (p = 0.038) and healthy controls (p < 0.0001), but not viral DAD controls (p = 0.65)(Supplementary Tables 4 and 5).
To corroborate these findings, we interrogated publicly available single-cell RNA sequencing (scRNA-Seq) datasets[18][19] including data from patients with COVID-19 and from control nonfibrotic lung tissue obtained from organs declined for donation, to compare the expression patterns of genes encoding these three proteins (CD274, HAVCR2, PDCD1) as well as other ICs of interest (LGALS9, CTLA4, LAG3, TIGIT, and IL2RA). These two datasets were merged and SCT-transformed. The cell identity was projected to the COVID-19 autopsy dataset[18]. While the IC gene expression patterns were similar in between COVID-19 and healthy control samples, certain genes were expressed at higher levels in COVID-19 (Extended Data Fig. 1). In particular, COVID-19 patients showed significantly higher expression of CD274, gene encoding PD-L1/CD274, in endothelial, epithelial, and myeloid cells (all p < 0.01); higher expression of HAVCR2, gene encoding HAVCR2/TIM-3, in B/plasma, myeloid, epithelial, fibroblasts, and mast cells, (all p < 0.05); higher expression of PDCD1, gene encoding PD-1, in epithelial, myeloid, and in T/natural killer (NK) cells (all p < 0.01); higher expression of LGALS9, gene encoding HAVCR2/TIM-3’s ligand, galectin-9, in endothelial, ciliated, myeloid, vascular contractile, T/NK, and mast cells (all p < 0.05); higher expression of IDO-1, gene encoding indoleamine 2,3-dioxygenase (IDO-1), in endothelial, epithelial, and myeloid cells (all p < 0.0001); LAG3, gene encoding LAG3, in ciliated, epithelial, secretory, and in T/NK cells (all p < 0.05); as well as higher expression of genes encoding other ICs and chemokines in different lung parenchymal and inflammatory cells, including CTLA4, TIGIT, IL2RA, PDCD1LG2, HLA-DRB, HMGB1, CXCL9, and GZMB (Extended Data Fig. 2). We also observed IC expression in analysis of scRNA-Seq datasets of bronchoalveolar lavage and peripheral blood mononuclear cells from patients hospitalized for COVID-19[20][21](Supplementary Fig. 2).
We evaluated cytokine gene expression in lung-derived myeloid, T/NK, and endothelial cells stratified by their expression of CD274 and HAVCR2, in scRNA-Seq datasets of fatal COVID-19 and healthy non-fibrotic controls[18][19]. Genes encoding for certain cytokines, including CXCL10 and CXCL9 were significantly upregulated in CD274-expressing myeloid cells from both COVID-19 lung tissues and healthy non-fibrotic control tissues; while IL18, IL1B, IL6, TGFB1, and TNF were only significantly upregulated in CD274-expressing myeloid cells from COVID-19 lung tissues. Both CD274-expressing lung endothelial cells and CD274-expressing T/NK cells from COVID-19 patients but not from healthy controls showed a statistically significant upregulation of the CXCL10 gene. The latter encodes a chemotactic protein that is secreted by several cell types in response to interferon-γ (IFN-γ)[22]. CD274-expressing T/NK cells from COVID-19 but not from healthy controls showed a statistically significant upregulation of IL18 expression. In parallel, CXCL10, IL10, IL18, IL6, and TGFB1 genes were upregulated in HAVCR2-expressing myeloid cells from both COVID-19 lung tissues and healthy lung tissues; while IL6, TGFB1 were only upregulated in HAVCR2-expressing myeloid cells from COVID-19 lung tissues. HAVCR2-expressing T/NK cells from patients with COVID-19 showed statistically significant upregulation of genes encoding multiple cytokines that were not upregulated in HAVCR2-expressing T/NK cells from healthy controls, including IFNG, CCL7, IL10, and TGFB2 (Extended Data Fig. 3).
Differential gene expression profiles and protein-protein interaction analyses
Previously, we reported that vascular injury and increased alveolar vascular congestion contributes to increased pulmonary dead space ventilation and poor outcomes in both ARDS due to COVID-19 as well as alternative etiologies[23]. To better understand the molecular basis of this phenomenon, we used lung tissue-derived NanoString mRNA counts from patients with COVID-19 (n = 10) to identify unique transcriptional changes in patients with pulmonary vasculopathy based on a previously described novel morphometric statistic [23], the CVasc, a machine-learning-derived quantitative index of the degree of pulmonary vascular congestion in the alveolar septae (as defined in Methods, Supplementary Fig. 3). Our analysis revealed that CD274, IDO1, and SELL (gene encoding for L-selectin, a cell surface adhesion molecule on leukocytes) were upregulated, and that CCL18 (gene encoding CCL18, chemokine regulating the adaptive immune system) was downregulated in patients with COVID-19-related vasculopathy (Extended Data Fig. 4). We also analyzed NanoString mRNA counts from lung tissue with low and high levels of SARS-CoV-2 RNA to identify transcriptional changes associated with active viral replication. This revealed 20 differentially expressed genes and highlighted significant enrichment of IFN-γ and IFN-α signaling pathways, along with other inflammatory genes, in patients with high parenchymal SARS-CoV-2 RNA levels (Extended Data Fig. 5A-B).
Weighted gene co-expression network analysis (WGCNA) of lung mRNA gene counts in patients with fatal COVID-19 demonstrated that CD274 was highly correlated to five genes (HLA-DRB1, IDO1, PDCD1LG2, GZMB, and CXCL9) encoding central immune regulatory molecules including 1) the β-subunit of the major histocompatibility complex (MHC) Class II molecule, HLA-DRB1; 2) the potent immunosuppressive enzyme, IDO-1; 3) PD-1’s ligand/IC receptor, PD-L2; 4) a potent serine protease involved in target cell apoptosis and present in the cytotoxic granules of cytotoxic T/NK cells, Granzyme B; and 5) a chemokine leading to differentiation, multiplication, and tissue extravasation of leukocytes, CXCL9. (Extended Data Fig. 5C-D). In a publicly available COVID-19 sc/snRNA-Seq dataset[18], CD274 was co-expressed with three of the genes identified in the WGCNA: HLA-DRB1, IDO1, and PDCD1LG2. The co-expression with these three proteins was seen mainly in endothelial and myeloid cells (Extended Data Fig. 6).
Clinical outcomes and PD-L1/CD274 and HAVCR2/TIM-3 immunostaining status
Given the differential lung cellular expression of immune checkpoints in patients with COVID-19, we examined whether the degree of PD-L1/CD274 and HAVCR2/TIM-3 expression in lung cell subpopulations correlated with the timing of progression to death in two autopsy cohorts. Cohorts baseline characteristics are summarized in Table 1 and Supplementary Table 6.
In survival analyses of an extended development autopsy cohort referred to as the MGH COVID-19 autopsy development cohort (as described in Methods), patients with high-density endothelial PD-L1/CD274 immunostaining had a significantly faster progression to death from symptom onset, hospital admission, intensive care unit (ICU) admission, and initiation of oxygen therapy or invasive mechanical ventilation compared to those with low-density endothelial immunostaining (p ≤ 0.05 for all endpoints; Figs. 2I-J and S8). In survival analyses of an international autopsy cohort referred to as the International COVID-19 autopsy cohort (as described in Methods), patients with high-density endothelial PD-L1/CD274 immunostaining had a significantly faster progression to death from symptom onset and from hospital admission compared to patients with low-density endothelial immunostaining (p ≤ 0.05 for all endpoints; Fig. 2K-L).
Cox proportional hazard models using data from cases of the International COVID-19 autopsy cohort (Supplementary Fig. 4) in which immunohistochemical data was available (n = 264) and for which COVID-19 was determined to be the underlying cause of death by the pathologists who performed the autopsies (n = 228/264), revealed significant associations between high-density endothelial PD-L1/CD274 immunostaining and faster clinical deterioration across all five time-to-event models, after adjusting for age, sex, BMI, and antiviral/corticosteroid therapies (Table 2). In similar models, the same pattern of immunostaining was associated with faster clinical deterioration across all five time-to-event models in both the internal MGH COVID-19 autopsy development cohort (n = 37; Supplementary Table 7) and the external international cohort, even after correcting for the presence of cases from MGH (n = 191; Table 2).
We did not find any statistically significant differences in survival when comparing groups of non-COVID-19 DAD controls with differences in expression density of PD-L1/CD274 and HAVCR2/TIM-3 in different lung cell subpopulations. Extended survival analyses are available in the Supplementary Information (Supplementary Figs. 5 and 6, Supplementary Tables 8–18).
Soluble immune checkpoint markers associate with severe COVID-19 and persistent SARS-CoV-2 RNAemia
Having found that elevation of IC markers in lung tissue associate with worse clinical outcomes, we next evaluated the association between soluble IC markers and clinical outcomes in participants 18 years or older with a clinical concern for COVID-19 upon ED arrival (n = 126, further enrollment details are described in Methods). Individuals who presented to an urban ED with severe COVID-19 had increased plasma normalized protein expression (NPX) levels of HAVCR2/TIM-3 (p < 0.0001), PD-L1/CD274 (p < 0.0001), galectin-9/LGALS9 (p = 0.0005), LAG3 (p = 0.003), IL2RA (p < 0.0001), and PD-1 (p < 0.005) compared to patients with non-severe COVID-19 (Fig. 3A). Sparse partial least squares discriminant analysis (sPLS-DA) identified a plasma proteomic profile that differentiates individuals with severe and non-severe COVID-19 (Fig. 3B), driven by an enrichment of elevated plasma levels of HAVCR2/TIM-3 and PD-L1/CD274 (Fig. 3C), with excellent accuracy to distinguish severe and non-severe COVID-19 (area under the receiver operating characteristic [AUROC] curve = 0.87, DeLong 95%CI 0.81–0.94) (Fig. 3D).
SARS-CoV-2 RNAemia is associated with increased inflammation, tissue damage signals, disease severity, and mortality[24]. In our cohort, participants with persistent SARS-CoV-2 RNAemia had increased plasma NPX levels of HAVCR2/TIM-3 (p < 0.0001), PD-L1/CD274 (p = 0.0001), galectin-9/LGALS9 (p = 0.002), LAG3 (p = 0.01), and IL2RA (p = 0.02), but not PD-1 (p = 0.6), compared to those with viral clearance (Fig. 3E). sPLS-DA identified proteomic and clinical features able to separate patients with persistent RNAemia and those who cleared their RNAemia (Fig. 3F). The major contributing factors for Component 1 included elevated plasma levels of HAVCR2/TIM-3, PD-L1/CD274, and galectin-9/LGALS9, along with the presence of any chronic kidney disease (inclusive of dialysis dependence) (Fig. 3G). These factors had excellent accuracy to distinguish individuals with and without persistent RNAemia (AUROC = 0.83, DeLong 95%CI 0.74–0.93) (Fig. 3H). Furthermore, ICs selected by sPLS-DA were highly correlated with other tissue damage and proinflammatory markers (Fig. 3I). Soluble PD-L1/CD274 was positively correlated to most endothelial cell and immune cell markers, and the degree of correlation with these markers increased over time from hospital day 0 to 7 (Extended Data Fig. 7).
We developed an exhaustion score (ES1) using plasma proteomic data from HAVCR2/TIM-3, PD-L1/CD274, and galectin-9/LGALS9. Statistically significant differences in SARS-CoV-2 RNAemia levels (log copies/mL) on hospital day 0, rates of SARS-CoV-2 RNAemia persistence or severe disease, and 28-day mortality were observed across the four ES1 quartiles (p ≤ 0.01 for all endpoints; Extended Data Fig. 8).
We amassed the largest international multi-center cohort study of COVID-19 autopsy cases and associated both cellular and plasma IC expression with clinical outcomes across three independent cohorts. Lung tissue analyses from 228 COVID-19 decedents revealed that PD-L1/CD274 endothelial cell expression is associated with more severe disease and more rapid progression to death. In patients presenting to the ED with COVID-19, upregulation of six different soluble ICs in the blood associated with persistent SARS-CoV-2 RNAemia and more severe disease. Our analyses reinforce the results of prior studies suggesting a prognostic value of ICs in COVID-19[3][5][6][7][8], underscore the importance of understanding IC mechanisms in the pathogenesis of severe COVID-19, and provide strong histopathological evidence that IC dysregulation is linked to the severity of disease and poor clinical outcomes in COVID-19 patients, particularly acute lung injury.
In whole-tissue transcriptomic analyses from COVID-19 decedents, CD274 was implicated in contributing to pulmonary vasculopathy, an important driver of respiratory impairment and worse clinical outcomes in fatal ARDS due to COVID-19[23]. This is in line with animal models of ARDS, in which upregulation of endothelial cell PD-L1/CD274 expression is associated with loss of alveolar-capillary barrier function and increased leukocyte PD-1 expression. Indeed, genetic ablation of PD-L1/CD274 in murine ALI models resulted in increased survival and decreased endothelial barrier permeability via regulation of the angiopoietin/TIE2 axis[25][26][27]. In COVID-19, endothelial injury markers, like angiopoietin-2, are increased in the later phase of severe disease[2] and are associated with sustained SARS-CoV-2 RNAemia [24]. IDO1, the gene encoding for the rate-limiting enzyme in tryptophan catabolism, was also upregulated in patients with COVID-19-related vasculopathy and co-expressed with CD274 in endothelial and myeloid cells. IDO-1 increases in response to IFN-γ signaling[28], and exerts potent immune suppressive effects on T/NK-lymphocytes by depleting tryptophan[29], and driving an accumulation of tryptophan catabolites in the paracellular space[30]. SELL, the gene encoding L-selectin, a protein that facilitates leukocyte-endothelial adhesion, was also upregulated in patients with COVID-19-related vasculopathy. Sc/snRNA-Seq data showed that SELL was highly expressed in lymphoid and myeloid cells (Supplementary Fig. 7), indicating increased inflammatory/immune cell trafficking in the injured lung. In addition, CCL18, the gene encoding the chemokine CCL18, was downregulated in patients with COVID-19-related vasculopathy. CCL18 is constitutively and highly expressed in the lungs, where it plays a role in maintaining homeostasis by attracting cells of adaptive immune system[31]. The presence of IFN-γ signaling is known to dampen the production of CCL18[32].
Both observational data in human viral infections[33][34] and animal models of viral lung injury demonstrate PD-L1/CD274 is robustly induced in lung and peripheral-blood cells through IFN-γ and IFN-α responses, impairing CD8 + cytotoxic T lymphocyte (TCD8) function[35][36][37][38][39][40]. PD-L1/PD-1 axis blockade results in virus-specific TCD8 expansion, accelerated viral clearance, and disease recovery[41]. CD274 is also co-expressed with PDCD1LG2 and HLA-DRB1 in endothelial and myeloid cells (Extended Data Fig. 5). The co-expression of CD274 and HLA-DRB1, the gene encoding the β-subunit of MHC Class II molecule, HLA-DR, suggests that the axis may play a critical role in both antigen presentation and T-cell suppression in COVID-19, and possibly in other infectious processes. By co-expressing with PDCD1LG2, which encodes for the PD-1 ligand, PD-L2[42], the T-cell suppressing function of PD-L1/CD274 + cells[43] is likely augmented in COVID-19.
In line with limited prior observations that plasma IC levels associate with acute SARS-CoV-2 infection and progression from prodromal stages to severe COVID-19 [3][8], we observed high plasma levels of multiple ICs in patients with severe disease and persistent RNAemia. Plasma levels of three ICs (HAVCR2/TIM-3, PD-L1/CD274, and galectin-9/LGALS9) accurately distinguished individuals with differences in disease progression and severity. Lung expression of genes encoding ICs were correlated and showed interactions with genes regulating other ICs, T-cell immunosuppression, TCD8/NK-cell-driven cytotoxicity, and inflammation (Extended Data Fig. 4C-D). Although compensatory upregulation of alternative checkpoints has been reported following IC blockade in the setting of oncologic therapy[44], our findings are consistent with other experimental and observational studies in COVID-19 and chronic viral infections[8][10][45][46], demonstrating that co-expression of multiple distinct ICs is associated with greater T-cell exhaustion, higher viral loads, and higher pro-inflammatory cytokines (Extended Data Fig. 2B), all which can lead to more severe disease. Poor viral control due to a suboptimal immune response may be itself causing immune exhaustion in a step-wise and progressive manner during chronic infections[47]. Further, the observation that HAVCR2/TIM-3 plasma levels correlate with COVID-19 outcomes rather than lung tissue levels, suggests that the contribution of ICs to the mechanisms of disease may be more systemic, rather than local. Studies evaluating IC cellular expression in other compartments participating in the immune response to viral infections (e.g., lymph nodes) could help unravel pathomechanisms. By overcoming systemic immunosuppressive mechanisms, IC inhibitors could potentially improve immune function in severe viral infections. Ex-vivo blockade of PD-1 nearly normalized T-cell-mediated responses to SARS-CoV-2 peptides in T-cells obtained from patients who recovered from COVID-19[45].
Although this study is limited by the observational nature of the data, our findings were consistent across three varied clinical patient cohorts, including the world’s largest international multi-center autopsy cohort enrolled to date. For most autopsies, clinical data was collected retrospectively, contributing to missing data. Further, we only studied autopsies before the emergence of novel SARS-CoV-2 variants and widespread SARS-CoV-2 vaccination. Therefore, our pathological findings can be considered unconfounded by the effects on host adaptive immunity. We did not use IHC protein co-localization techniques (e.g., multiplex IHC), so the IHC cellular localization may be limited by the subjectivity of the pathologists’ interpretation. However, the pathologists who performed the IHC review were blinded to clinical data and were not involved in the survival analyses. Our analyses imply PD-L1/CD274 endothelial expression is highly relevant in the most extreme phenotype of COVID-19. As PD-L1/CD274 is expressed in the endothelium of non-fatal COVID-19-related lung injury[48], and in other diseases causing vessel inflammation (e.g., vasculitides)[37][49], further studies are needed to assess the pathogenic role of this marker in non-fatal phenotypes with vascular injury. We found no association between cellular IC expression and poor clinical outcomes in our DAD comparators. Given the heterogeneity of DAD[50] further research should investigate the role of ICs in the pathogenesis of DAD in general. Lastly, our analysis using generalized estimating equations (GEE) models did not reveal any statistical associations between various clinical characteristics/comorbidities (e.g., smoking, hypertension, autoimmune disease) with PD-L1 lung endothelial expression (Supplementary Tables 19). It is important to note that the lack of findings may be attributed to a limited sample size, and further studies may be needed to confirm these results.
Our data support the re-analysis of completed clinical trials of immunomodulators using dysregulated IC status as a treatable trait and sets the stage for future clinical trials to test the role of IC blockade in patients with severe COVID-19 and dysregulated IC status.
Reviewed and approved by Partners IRB [Protocol No. 2020P001146].
Analysis of patient autopsy material was reviewed and approved by the Mass General Brigham (formerly known as Partners) Institutional Review Board (IRB): Protocol #: 2020P001146. The analyses of human plasma samples were also approved by the Mass General Brigham IRB, with an IRB-approved waiver of informed consent (protocol 2017P001681).
Ethics statement
The study protocol was reviewed and approved by by the Mass General Brigham (formerly known as Partners) Institutional Review Board (IRB) under the protocol No. 2020P001146. The analyses of patient autopsy material was reviewed and approved by the Mass General Brigham (formerly known as Partners) IRB under the protocol No. 2020P001146. The analyses of human plasma samples were also approved by the Mass General Brigham IRB, with an IRB-approved waiver of informed consent (protocol 2017P001681).
Inclusion and ethics in global research
The research process that involved data from the International COVID-19 autopsy cohort included local researchers throughout the study different steps: study design, study implementation, data ownership, and authorship of publication. All the roles and responsibilities were agreed amongst collaborators ahead of the research. This study does not result in stigmatization, incrimination, discrimination or otherwise personal risk to participant.
Uniform Biological Material Transfer Agreement (UBMTAs) were executed in all situations when biological human materials were transferred between the local researchers of the COVID-19 International lung autopsy consortium and the researchers at MGH. The standard UBMTA used in this study was the UBMTA master agreement published in the Federal Register on March 8, 1995. The legal offices between the institutions of the provider scientists and the recipient scientists drafted the UBMTAs. The UBMTAs defined the rights of the provider scientist (local researchers) and the recipient scientist (MGH researchers) with respect to the biological human materials.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Data availability
All requests for raw data of the three autopsy cohorts will be promptly reviewed by the corresponding authors to determine whether the request is subject to any confidentiality obligations. Any data that can be shared will be released via a data sharing agreement. In collaboration with Olink, the proteomics data (>1.3M protein measurements) and essential associated clinical parameters of the MGH ED cohort are now being made available online for everyone in the global scientific community via the Olink website (https://olink.com/application/mgh-covid-19-study/). Additional data supporting the findings in this work are available in the main text, figures, extended data and supplementary files. Any additional information required to reanalyze the data reported in this work paper is available from the lead contact upon request, but costs might be incurred and a data-sharing agreement would be necessary.
Code availability
All codes used to analyze the data in this study will be available before publication.
- Leisman DE, Mehta A, Thompson BT, Charland NC, Gonye ALK, Gushterova I, et al. Alveolar, Endothelial, and Organ Injury Marker Dynamics in Severe COVID-19. Am J Respir Crit Care Med. 2022 Mar 1;205(5):507-519.
- Dioverti V, Salto-Alejandre S, Haidar G. Immunocompromised Patients with Protracted COVID-19: a Review of "Long Persisters". Curr Transplant Rep. 2022;9(4):209-218.
- Schultheiß C, Paschold L, Simnica D, Mohme M, Willscher E, von Wenserski L, et al. Next-Generation Sequencing of T and B Cell Receptor Repertoires from COVID-19 Patients Showed Signatures Associated with Severity of Disease. Immunity. 2020 Aug 18;53(2):442-455.e4.
- Sumida TS, Dulberg S, Schupp JC, Lincoln MR, Stillwell HA, Axisa PP, Comi M, et al. Type I interferon transcriptional network regulates expression of coinhibitory receptors in human T cells. Nat Immunol. 2022 Apr;23(4):632-642.
- Sabbatino F, Pagliano P, Sellitto C, Stefanelli B, Corbi G, Manzo V, De Bellis E, Liguori L, Salzano FA, Pepe S, Filippelli A, Conti V. Different Prognostic Role of Soluble PD-L1 in the Course of Severe and Non-Severe COVID-19. J Clin Med. 2023 Oct 27;12(21):6812.
- Kong Y, Wang Y, Wu X, Han J, Li G, Hua M, Han K, Zhang H, Li A, Zeng H. Storm of soluble immune checkpoints associated with disease severity of COVID-19. Signal Transduct Target Ther. 2020 Sep 7;5(1):192.
- Al-Mterin MA, Alsalman A, Elkord E. Inhibitory Immune Checkpoint Receptors and Ligands as Prognostic Biomarkers in COVID-19 Patients. Front Immunol. 2022 Mar 31;13:870283.
- Diao B, Wang C, Tan Y, Chen X, Liu Y, Ning L, Chen L, Li M, Liu Y, Wang G, Yuan Z, Feng Z, Zhang Y, Wu Y, Chen Y. Reduction and Functional Exhaustion of T Cells in Patients With Coronavirus Disease 2019 (COVID-19). Front Immunol. 2020 May 1;11:827.
- Bellesi S, Metafuni E, Hohaus S, Maiolo E, Marchionni F, D'Innocenzo S, La Sorda M, Ferraironi M, Ramundo F, Fantoni M, Murri R, Cingolani A, Sica S, Gasbarrini A, Sanguinetti M, Chiusolo P, De Stefano V. Increased CD95 (Fas) and PD-1 expression in peripheral blood T lymphocytes in COVID-19 patients. Br J Haematol. 2020 Oct;191(2):207-211.
- Zheng HY, Zhang M, Yang CX, Zhang N, Wang XC, Yang XP, Dong XQ, Zheng YT. Elevated exhaustion levels and reduced functional diversity of T cells in peripheral blood may predict severe progression in COVID-19 patients. Cell Mol Immunol. 2020 May;17(5):541-543.
- Jeannet R, Daix T, Formento R, Feuillard J, François B. Severe COVID-19 is associated with deep and sustained multifaceted cellular immunosuppression. Intensive Care Med. 2020 Sep;46(9):1769-1771.
- Shahbaz S, Xu L, Sligl W, Osman M, Bozorgmehr N, Mashhouri S, Redmond D, Perez Rosero E, Walker J, Elahi S. The Quality of SARS-CoV-2-Specific T Cell Functions Differs in Patients with Mild/Moderate versus Severe Disease, and T Cells Expressing Coinhibitory Receptors Are Highly Activated. J Immunol. 2021 Aug 15;207(4):1099-1111.
- Tao K, Tzou PL, Nouhin J, Gupta RK, de Oliveira T, Kosakovsky Pond SL, et al. The biological and clinical significance of emerging SARS-CoV-2 variants. Nat Rev Genet. 2021 Dec;22(12):757-773.
- Hadley E, Yoo YJ, Patel S, Zhou A, Laraway B, Wong R, Preiss A, Chew R, Davis H, Brannock MD, Chute CG, Pfaff ER, Loomba J, Haendel M, Hill E; N3C and RECOVER consortia; Moffitt R. Insights from an N3C RECOVER EHR-based cohort study characterizing SARS-CoV-2 reinfections and Long COVID. Commun Med (Lond). 2024 Jul 11;4(1):129.
- Dong H, Zhu G, Tamada K, Chen L. B7-H1, a third member of the B7 family, co-stimulates T-cell proliferation and interleukin-10 secretion. Nat Med. 1999 Dec;5(12):1365-9.
- Latchman Y, Wood CR, Chernova T, Chaudhary D, Borde M, Chernova I, et al. PD-L2 is a second ligand for PD-1 and inhibits T cell activation. Nat Immunol. 2001 Mar;2(3):261-8.
- Wolf Y, Anderson AC, Kuchroo VK. TIM3 comes of age as an inhibitory receptor. Nat Rev Immunol. 2020 Mar;20(3):173-185.
- Delorey TM, Ziegler CGK, Heimberg G, Normand R, Yang Y, Segerstolpe Å, et al. COVID-19 tissue atlases reveal SARS-CoV-2 pathology and cellular targets. Nature. 2021 Jul;595(7865):107-113.
- Habermann AC, Gutierrez AJ, Bui LT, Yahn SL, Winters NI, Calvi CL, Peter L, Chung MI, Taylor CJ, Jetter C, Raju L, Roberson J, Ding G, Wood L, Sucre JMS, Richmond BW, Serezani AP, McDonnell WJ, Mallal SB, Bacchetta MJ, Loyd JE, Shaver CM, Ware LB, Bremner R, Walia R, Blackwell TS, Banovich NE, Kropski JA. Single-cell RNA sequencing reveals profibrotic roles of distinct epithelial and mesenchymal lineages in pulmonary fibrosis. Sci Adv. 2020 Jul 8;6(28):eaba1972.
- Wilk AJ, Rustagi A, Zhao NQ, Roque J, Martínez-Colón GJ, McKechnie JL, et al. A single-cell atlas of the peripheral immune response in patients with severe COVID-19. Nat Med. 2020 Jul;26(7):1070-1076.
- Liao M, Liu Y, Yuan J, Wen Y, Xu G, Zhao J, et al. Single-cell landscape of bronchoalveolar immune cells in patients with COVID-19. Nat Med. 2020 Jun;26(6):842-844.
- Luster AD, Unkeless JC, Ravetch JV. Gamma-interferon transcriptionally regulates an early-response gene containing homology to platelet proteins. Nature. 1985 Jun 20-26;315(6021):672-6.
- Villalba JA, Hilburn CF, Garlin MA, Elliott GA, Li Y, Kunitoki K, et al. Vasculopathy and Increased Vascular Congestion in Fatal COVID-19 and Acute Respiratory Distress Syndrome. Am J Respir Crit Care Med. 2022 Oct 1;206(7):857-873.
- Li Y, Schneider AM, Mehta A, Sade-Feldman M, Kays KR, Gentili M, et al. SARS-CoV-2 viremia is associated with distinct proteomic pathways and predicts COVID-19 outcomes. J Clin Invest. 2021 Jul 1;131(13):e148635.
- Lomas-Neira J, Monaghan SF, Huang X, Fallon EA, Chung CS, Ayala A. Novel Role for PD-1:PD-L1 as Mediator of Pulmonary Vascular Endothelial Cell Functions in Pathogenesis of Indirect ARDS in Mice. Front Immunol. 2018 Dec 20;9:3030.
- Lomas-Neira J, Venet F, Chung CS, Thakkar R, Heffernan D, Ayala A. Neutrophil-endothelial interactions mediate angiopoietin-2-associated pulmonary endothelial cell dysfunction in indirect acute lung injury in mice. Am J Respir Cell Mol Biol. 2014 Jan;50(1):193-200.
- Morrell ED, Holton SE, Wiedeman A, Kosamo S, Mitchem MA, Dmyterko V, Franklin Z, Garay A, Stanaway IB, Liu T, Sathe NA, Mabrey FL, Stapleton RD, Malhotra U, Speake C, Hamerman JA, Pipavath S, Evans L, Bhatraju PK, Long SA, Wurfel MM, Mikacenic C. PD-L1 and PD-1 Are Associated with Clinical Outcomes and Alveolar Immune Cell Activation in ARDS. Am J Respir Cell Mol Biol. 2024 Jul 1. doi: 10.1165/rcmb.2024-0201OC. Epub ahead of print. PMID: 38950166.
- Schalper KA, Carvajal-Hausdorf D, McLaughlin J, Altan M, Velcheti V, Gaule P, Sanmamed MF, Chen L, Herbst RS, Rimm DL. Differential Expression and Significance of PD-L1, IDO-1, and B7-H4 in Human Lung Cancer. Clin Cancer Res. 2017 Jan 15;23(2):370-378. doi: 10.1158/1078-0432.CCR-16-0150. Epub 2016 Jul 20.
- Munn DH, Mellor AL. Indoleamine 2,3 dioxygenase and metabolic control of immune responses. Trends Immunol. 2013 Mar;34(3):137-43.
- Frumento G, Rotondo R, Tonetti M, Damonte G, Benatti U, Ferrara GB. Tryptophan-derived catabolites are responsible for inhibition of T and natural killer cell proliferation induced by indoleamine 2,3-dioxygenase. J Exp Med. 2002 Aug 19;196(4):459-68.
- Chenivesse C, Chang Y, Azzaoui I, Ait Yahia S, Morales O, Plé C, et al. Pulmonary CCL18 recruits human regulatory T cells. J Immunol. 2012 Jul 1;189(1):128-37.
- Vulcano M, Struyf S, Scapini P, Cassatella M, Bernasconi S, Bonecchi R, et al. Unique regulation of CCL18 production by maturing dendritic cells. J Immunol. 2003 Apr 1;170(7):3843-9.
- Valero-Pacheco N, Arriaga-Pizano L, Ferat-Osorio E, Mora-Velandia LM, Pastelin-Palacios R, Villasís-Keever MÁ, et al. PD-L1 expression induced by the 2009 pandemic influenza A(H1N1) virus impairs the human T cell response. Clin Dev Immunol. 2013;2013:989673.
- Bengsch B, Seigel B, Ruhl M, Timm J, Kuntz M, Blum HE, et al. Coexpression of PD-1, 2B4, CD160 and KLRG1 on exhausted HCV-specific CD8+ T cells is linked to antigen recognition and T cell differentiation. PLoS Pathog. 2010 Jun 10;6(6):e1000947.
- Erickson JJ, Gilchuk P, Hastings AK, Tollefson SJ, Johnson M, Downing MB, et al. Viral acute lower respiratory infections impair CD8+ T cells through PD-1. J Clin Invest. 2012 Aug;122(8):2967-82.
- Rutigliano JA, Sharma S, Morris MY, Oguin TH 3rd, McClaren JL, Doherty PC, et al. Highly pathological influenza A virus infection is associated with augmented expression of PD-1 by functionally compromised virus-specific CD8+ T cells. J Virol. 2014 Feb;88(3):1636-51.
- Rodig N, Ryan T, Allen JA, Pang H, Grabie N, Chernova T, et al. Endothelial expression of PD-L1 and PD-L2 down-regulates CD8+ T cell activation and cytolysis. Eur J Immunol. 2003 Nov;33(11):3117-26.
- Dimou P, Wright RD, Budge KL, Midgley A, Satchell SC, et al. The human glomerular endothelial cells are potent pro-inflammatory contributors in an in vitro model of lupus nephritis. Sci Rep. 2019 Jun 6;9(1):8348.
- Kim J, Myers AC, Chen L, Pardoll DM, Truong-Tran QA, Lane AP, et al. Constitutive and inducible expression of b7 family of ligands by human airway epithelial cells. Am J Respir Cell Mol Biol. 2005 Sep;33(3):280-9.
- Pardoll DM. The blockade of immune checkpoints in cancer immunotherapy. Nat Rev Cancer. 2012 Mar 22;12(4):252-64.
- Barber DL, Wherry EJ, Masopust D, Zhu B, Allison JP, Sharpe AH, et al. Restoring function in exhausted CD8 T cells during chronic viral infection. Nature. 2006 Feb 9;439(7077):682-7.
- Latchman Y, Wood CR, Chernova T, Chaudhary D, Borde M, Chernova I, et al. PD-L2 is a second ligand for PD-1 and inhibits T cell activation. Nat Immunol. 2001 Mar;2(3):261-8.
- Sharpe AH, Wherry EJ, Ahmed R, Freeman GJ. The function of programmed cell death 1 and its ligands in regulating autoimmunity and infection. Nat Immunol. 2007 Mar;8(3):239-45.
- Koyama S, Akbay EA, Li YY, Herter-Sprie GS, Buczkowski KA, Richards WG, et al. Adaptive resistance to therapeutic PD-1 blockade is associated with upregulation of alternative immune checkpoints. Nat Commun. 2016 Feb 17;7:10501.
- Loretelli C, Abdelsalam A, D'Addio F, Ben Nasr M, Assi E, Usuelli V, et al. PD-1 blockade counteracts post-COVID-19 immune abnormalities and stimulates the anti-SARS-CoV-2 immune response. JCI Insight. 2021 Dec 22;6(24):e146701.
- Jin HT, Anderson AC, Tan WG, West EE, Ha SJ, Araki K, Freeman GJ, Kuchroo VK, Ahmed R. Cooperation of Tim-3 and PD-1 in CD8 T-cell exhaustion during chronic viral infection. Proc Natl Acad Sci U S A. 2010 Aug 17;107(33):14733–14738.
- Kahan SM, Wherry EJ, Zajac AJ. T cell exhaustion during persistent viral infections. Virology. 2015 May;479-480:180-93.
- Doglioni C, Ravaglia C, Chilosi M, Rossi G, Dubini A, Pedica F, et al. Covid-19 Interstitial Pneumonia: Histological and Immunohistochemical Features on Cryobiopsies. Respiration. 2021;100(6):488-498.
- Zhang H, Watanabe R, Berry GJ, Vaglio A, Liao YJ, Warrington KJ, et al. Immunoinhibitory checkpoint deficiency in medium and large vessel vasculitis. Proc Natl Acad Sci U S A. 2017 Feb 7;114(6):E970-E979.
- Hariri L, Hardin CC. Covid-19, Angiogenesis, and ARDS Endotypes. N Engl J Med. 2020 Jul 9;383(2):182-183.
|
Clinical and laboratory characteristics |
MGH Development cohort (n = 39) |
International COVID-19 cohort (n = 329†) |
|---|---|---|
|
Age Median age; years (IQR) Distribution; n (%) < 20 years 20–39 years 40–59 years ≥ 60 years |
67 (54.7–76.2) 0 3/39 (8) 11/39 (28) 25/39 (64) |
70 (58–80) 0 21/329 (6) 73/329 (22) 235/329 (71) |
|
Sex and body mass index (BMI) Female; n (%) Median BMI (kg/m²; IQR; n) Overweight-Obesity (BMI ≥ 25 kg/m²) |
13/39 (33) 30.7 (26.9–36.3; 39) 32/39 (82) |
118/329 (36) 28.2 (24.9–33.2; 278) 210/280 (75) |
|
Pre-existing Conditions Cardiovascular disease Cerebrovascular disease Hypertension Diabetes Neoplastic disease Chronic lung disease Chronic renal disease Immunocompromised states Connective tissue or autoimmune disorders Transplantation Solid organ Hematopoietic cell HIV infection Other conditions‡ |
17/39 (44) 4/39 (10) 26/39 (67) 16/39 (41) 8/39 (21) 20/39 (51) 7/39 (18) 4/39 (10) 2/39 (5) 0 1/39 (3) 35/39 (90) |
125/300 (42) 15/300 (5) 173/302 (57) 107/301 (36) 55/304 (18) 81/301 (27) 54/298 (18) 22/311 (7) 10/311 (3) 1/312 (0.3) 3/312 (1) 158/305 (52) |
|
Smoking History Never smoked Former smoked Current smoker |
16/37 (43) 16/37 (43) 5/37 (14) |
142/258 (55) 49/258 (19) 67/258 (26) |
|
Duration of Illness Time to death from symptom onset, days, median (IQR; n) |
18 (12–26; 39) |
17 (11 − 28; 262) |
|
Hospitalization Hospital admittance Hospital length-of-stay, days, median (IQR; n) Intensive care unit (ICU) admittance ICU length-of-stay, days, median (IQR; n) Mechanical ventilatory support Invasive Duration, days, median (IQR; n) Non-invasive Duration, days, median (IQR; n) Extracorporeal membrane oxygenation (ECMO) Duration, days, median (IQR; n) Mechanically ventilated patients meeting Berlin definition for ARDS ARDS patients Nadir PaO2/FIO2, mm Hg, median (IQR; n) Hypoxemia severity Mild - PaO2/FIO2, 201–300 mm Hg Moderate - PaO2/FIO2, 101–200 mm Hg Severe - PaO2/FIO2, ≤ 100 mm Hg Prone positioning Vasopressor support Renal replacement therapy Duration, days, median (IQR; n) |
36/39 (92) 13.5 (7–22.3; 36) 28/39 (72) 10.5 (6.8–22.3; 28) 25/39 (64) 9 (7–17; 25) 5/39 (13) 4 (3–5.5; 5) 3/39 (8) 18 (11–20; 3) 23/25 (92) 121 (77–160; 23) 1/22 (5) 11/22 (50) 10/22 (45) 21/39 (54) 28/39 (72) 10/39 (26) 3 (1.3–7; 10) |
311/329 (95) 11 (5–21; 311) 220/329 (67) 10 (5–21; 220) 174/303 (57) 10 (5–20; 172) 59/292 (20) 5 (2–7.3; 36) 16/305 (5) 20 (12.5–29.5; 16) 109/118 (93) 82 (59–120; 109) 4/109 (4) 37/109 (34) 69/109 (63) 123/263 (47) 109/200 (55) 52/292 (18) 4.5 (2–10; 34) |
|
Laboratory Findings (closest to death) SARS-CoV-2 rRT-PCR positive Hemoglobin, median (g/dL; IQR; n) C-reactive protein, median (mg/L; IQR; n) Creatinine, median (mg/dL; IQR; n) D-dimer, median (ng/mL; IQR; n) Lactate dehydrogenase, median (U/L; IQR; n) Procalcitonin, median (ng/mL; IQR; n) IL-6, median (pg/mL; IQR; n) Ferritin, median (ug/L; IQR; n) Erythrocyte sedimentation rate, median (mm/h; IQR; n) Absolute Neutrophil count, median (K/uL; IQR; n) Absolute Lymphocyte count, median (K/uL; IQR; n) Red Cell Distribution Width, median (%; IQR; n) Aspartate aminotransferase, median (U/L; IQR; n) Alanine aminotransferase, median (U/L; IQR; n) |
39/39 (100) 10.4 (8.2–11.6; 36) 130.3 (46.8–201.9; 36) 1.57 (0.98–2.6; 36) 3,610 (1737–5,577; 34) 435 (334.3–570.3; 36) 0.26 (0.11–3.47; 34) 103.9 (29.0–353.0; 15) 960.5 (549.8–1,871.3; 36) 64 (36–96; 33) 13.1 (8.4–16.3) 1.2 (0.6–2.3; 36) 15.7 (14.8–16.7; 36) 50 (35.6–120.5; 35) 35 (20–67; 34) |
329/329 (100) 10.4 (8.8–12.6; 245) 116.5 (62.9–228.3; 224) 1.39 (0.77–2.83; 244) 2,010 (1,010–5,000; 189) 440 (338–657.8; 192) 0.76 (0.22–3.23; 128) 210.5 (70.1–780.8; 70) 934 (480–1,869; 139) 32 (22–65; 65) 9.4 (6.3–15.1; 219) 0.9 (0.6–1.7; 222) 15.9 (14.4–17.8; 155) 65 (34–107; 185) 46 (22.5–90; 207) |
|
Treatments During Hospital Stay Antibiotics Remdesivir Systemic Corticosteroids Dexamethasone Hydroxychloroquine / Chloroquine Anticoagulants Prophylactic dose Therapeutic dose |
36/39 (92) 3/39 (8) 16/39 (41) 3/39 (8) 16/39 (41) 34/39 (87) 22/34 (65) 12/34 (35) |
247/281 (88) 36/307 (12) 122/291 (42) 28/291 (10) 108/291 (37) 254/306 (83) 137/254 (54) 116/254 (46) |
|
Data are number of patients/total number of patients (%), median (IQR), or otherwise specified. IQR: interquartile range. †Formalin-fixed paraffin-embedded (FFPE) lung tissue-blocks were collected in all patients of the international cohort (535/535), and in most patients (n = 329/535), they were available for ancillary testing and are stored in a common international autopsy repository at MGH. The final number of samples included in the IHC-derived survival analyses was lower (n = 264) due to the exclusion of samples with limited quality, FFPE tissue-block exhaustion, or incomplete clinicopathologic/survival annotation. ‡Other pre-existing conditions are enumerated in Supplementary Table 6. |
||
|
Models testing |
COVID-19 related deaths determined on autopsy (n = 228†) |
|||||
|---|---|---|---|---|---|---|
|
Hazard ratio |
95% CI |
p-value |
||||
|
Time-to-death from symptom onset (days) |
||||||
|
Unadjusted |
1.56 |
1.04–2.33 |
0.03 |
|||
|
Adjusted for |
||||||
|
Age, sex, BMI, corticosteroid and Remdesivir use |
1.70 |
1.08–2.69 |
0.021 |
|||
|
Age, sex, BMI, corticosteroid and Remdesivir use, cases from Massachusetts General Hospital |
2.07 |
1.21–3.57 |
0.008 |
|||
|
Hospital length-of-stay (LOS) (days) |
||||||
|
Unadjusted |
1.49 |
1.02–2.19 |
0.04 |
|||
|
Adjusted for |
||||||
|
Age, sex, BMI, corticosteroid and Remdesivir use |
1.67 |
1.06–2.63 |
0.026 |
|||
|
Age, sex, BMI, corticosteroid and Remdesivir use, cases from Massachusetts General Hospital |
2.13 |
1.28–3.56 |
0.004 |
|||
|
Intensive care unit (ICU) LOS (days) |
||||||
|
Unadjusted |
1.48 |
0.91–2.39 |
0.10 |
|||
|
Adjusted for |
||||||
|
Age, sex, BMI, corticosteroid and Remdesivir use |
1.89 |
1.02–3.48 |
0.042 |
|||
|
Age, sex, BMI, corticosteroid and Remdesivir use, cases from Massachusetts General Hospital |
2.26 |
1.13–4.51 |
0.02 |
|||
|
Length-of-oxygen-therapy (days) |
||||||
|
Unadjusted |
1.44 |
0.96–2.16 |
0.08 |
|||
|
Adjusted for |
||||||
|
Age, sex, BMI, corticosteroid and Remdesivir use |
1.64 |
1.04–2.58 |
0.035 |
|||
|
Age, sex, BMI, corticosteroid and Remdesivir use, cases from Massachusetts General Hospital |
2.20 |
1.32–3.66 |
0.003 |
|||
|
Length-of-invasive mechanical ventilation (days) |
||||||
|
Unadjusted |
1.45 |
0.87–2.44 |
0.16 |
|||
|
Adjusted for |
||||||
|
Age, sex, BMI, corticosteroid and Remdesivir use |
1.94 |
0.99–3.82 |
0.055 |
|||
|
Age, sex, BMI, corticosteroid and Remdesivir use, cases from Massachusetts General Hospital |
2.39 |
1.12–5.13 |
0.03 |
|||
|
†The cox proportional models included autopsy from all cohorts in which the underlying cause of death was found to be directly related to COVID-19 by the individual pathologists who performed the autopsies. Cases in which COVID-19 was deemed to be a concomitant disease were not included in these analyses. BMI: body mass index. |
||||||
1. Study cohorts, data collection, and biologic samples
To identify biomarkers of disease severity, we evaluated IC protein and gene expression using complementary methodologies in three COVID-19 autopsy cohorts and in a cohort of early COVID-19 patients in the ED.
1.1. Study cohorts
1.1.1. Massachusetts General Hospital autopsy cohorts:
Two autopsy cohorts were prospectively enrolled at the Massachusetts General Hospital (MGH) in Boston, Massachusetts. These cohorts are referred as to the MGH autopsy discovery cohort and the MGH COVID-19 autopsy development cohort.
1.1.1.1 MGH autopsy discovery cohort
The cohort consisted of prospectively enrolled consecutive decedents with respiratory failure and DAD due to SARS-CoV-2 infection (n=20) during the first wave of the pandemic in Northeastern U.S. (March–May 2020) and retrospective consecutive decedents with respiratory failure and DAD (n=21) prior to the SARS-CoV-2 pandemic. This cohort has been reported previously [23]. Baseline clinical and laboratory characteristics of the cohort are described in Supplementary Tables 1 and 2. SARS-CoV-2 nuclei acid was not identified by specific in-situ hybridization in the pre pandemic autopsy lung tissues of any of the non-SARS-CoV-2 control cases as expected (Supplementary Table 20). Patients were unvaccinated against SARS-CoV-2 and did not receive COVID-19-specific therapeutics.
1.1.1.2 MGH COVID-19 autopsy development cohort
An extended cohort of consecutive COVID-19 autopsies from MGH (n=39) was prospectively enrolled for this study from March to July 2020 (Table 1). COVID-19 autopsy cases included in the MGH autopsy discovery cohort (n=20) were also part of the MGH COVID-19 autopsy development cohort.
Autopsy protocols and sample processing (MGH autopsy cohorts)
Autopsies were carried out and tissues were collected following Centers for Disease Control and Prevention (CDC) guidelines for biosafety and infection control practices. The specimen collection was performed as previously described[23]. All autopsies were performed at the MGH Department of Pathology following consent of the legal next of kin or healthcare proxy as appropriate.
Two representative FFPE-tissue blocks of the MGH autopsy cohorts were selected by a board-certified pathologist, and hematoxylin and eosin (H&E)-stained slides, chromogenic immunohistochemistry (IHC) for different markers, and chromogenic SARS-CoV-2 RNA in situ hybridization (RNA-ISH; RNAscope™, Advanced Cell Diagnostics, Inc., Newark, CA) were performed in each selected FFPE whole-tissue section. As described in our prior report[23], the two slides included the most congested and the least congested lung parenchymal region out of all histological sections. All selected slides contained sections of at least three different generations of airways (from both the conducting and respiratory zones), which could include: lobar bronchi, segmental bronchi, subsegmental bronchi, bronchioles, lobular bronchioles, terminal bronchioles, respiratory bronchioles, alveolar ducts, alveolar sacs, and alveoli.
In addition, we performed machine-learning derived morphometrics of the vascular congestion in the alveolar septa in autopsy cases of the MGH COVID-19 autopsy discovery cohort in which Nanostring RNA data was available (n=10), as detailed in section 2.2.
1.1.2 International COVID-19 autopsy cohort
The COVID-19 international lung autopsy consortium was established between March 1, 2020, and February 28, 2021 as a research collaboration between pathology, pulmonary, critical care, and infectious diseases physicians and investigators from 47 research sites in 16 countries. The consortium aimed to inform the international medical community with a standardized and systematic description of the histopathological findings and pathomechanisms of SARS-CoV-2 infection in the lung. This cohort is referred as to the international cohort.
The consortium amassed the largest international, multi-center autopsy cohort including a clinical and pathological metadata registry of 535 decedents with confirmed SARS-CoV-2 infection. The cohort includes both autopsies of hospitalized and non-hospitalized COVID-19 deceased persons. The cohort also included autopsy cases from MGH. The fact that vaccination became available worldwide after December 8, 2020, for the elderly and selected subpopulations (e.g., healthcare workers) is reflected in the absence of post-vaccination cases. Because participants were unvaccinated against SARS-CoV-2, and most did not receive targeted therapies, the cohort is suitable to study the natural history of COVID-19.
Individual-participant data and a pre-specified list of 269 variables were requested from investigators at eligible research sites and retrieved from the medical records or the laboratory information system at each participating institution. The underlying cause of death was determined in each case by the pathologist who performed the autopsy. A centralized histopathological review of all cases was performed by board-certified subspecialized thoracic pathologists at MGH blinded to clinical data. Additional details about site enrollment and data collected are included in Supplementary Tables 21-23.
Search strategy and selection criteria of collaborating research sites (International cohort)
Briefly, we searched for prospective or retrospective cohort, case-control studies, observational studies, randomized controlled trials (RCTs), and case reports describing autopsies of patients with laboratory confirmed SARS-CoV-2 infection published between Dec 31, 2019, and September 20, 2020, in MEDLINE, and Embase electronic databases, to identify potential collaborative research sites.
Research site eligibility was determined based on the content of scientific publications from the respective sites. Since the pathological description of COVID-19 autopsies was our primary study outcome, we excluded studies not including COVID-19 autopsy cohorts, population-based studies, outbreak reports, editorials, letters to editor without sufficient pathological description of autopsy cases, reviews, theoretical models, and any other studies describing COVID-19 in which autopsies were not performed.
The 9-month timeframe was chosen on the basis of expected availability of clinical and pathological autopsy data after the Wuhan Municipal Health Commission reported the first cluster of cases of SARS-CoV-2 related pneumonia in Wuhan, Hubei Province, in China on December 31, 2019[51][52]. We also included unpublished data of relevant studies available as preprints in medRxiv, bioRxiv, and arXiv databases between Dec 31, 2019, and September 20, 2020.
We included the following terms in our search strategy: (“autopsy”, OR “biopsy”, OR “pathology”, OR “pathological findings”, OR “histopathology”, OR “histopathological findings”) AND (“SARS-CoV-2”, OR “COVID-19”, OR “coronavirus disease”). We restricted our search to studies in humans. The search strategy was performed by one reviewer (YL). We additionally reviewed reference lists of all relevant studies to identify additional studies.
One reviewer (JV) reviewed articles in two stages to select potential collaborative research sites: evaluation of titles and abstracts followed by full-text review. Relevant articles were subject to a full text review by the reviewer. The senior investigator (MMK) was consulted in certain situations regarding research site inclusion. If research site eligibility could not be assessed from the full-text manuscript because of missing information, we contacted authors for clarification. We evaluated eligible articles for duplication of data on the same individuals and research autopsy cohorts and excluded manuscripts if necessary.
Reviewers (International cohort)
JV is a board-certified pathologist in anatomic pathology, clinical pathology, and medical microbiology, with fellowship training in pulmonary pathology. JV has extensive clinical and research expertise in pulmonary and infectious disease pathology and epidemiology. JV was a clinical fellow in pathology at the MGH and Harvard University when the study started. He is currently the lead of the human diagnostic pathology team at the Infectious Diseases Pathology Branch of the Centers for Disease Control and Prevention (CDC), and an adjunct assistant professor of the Department of Pathology and Laboratory Medicine at Emory University School of Medicine. YL is a board-certified infectious disease physician with extensive research and clinical expertise in COVID-19, HIV, and hepatitis B virus epidemiology and pathogenesis. YL was a clinical fellow at Brigham and Women’s Hospital and Harvard University when the study started. He is currently a clinical assistant professor of Medicine at the University of Pittsburgh Medical Center (UPMC). MMK is a board-certified pathologist in anatomic pathology with over 20 years of experience in pulmonary pathology. MMK is a Professor of Pathology at Harvard Medical School, the vice chair for Anatomic Pathology, and the director of Pulmonary Pathology at the MGH.
Selection criteria of research sites (International cohort)
To be eligible for inclusion, a research site found through a published study needed to comply with the following criteria:
(1) Include at least one autopsy case of a deceased patient with laboratory-confirmed SARS-CoV-2 infection, (2) either women or men, (3) from hospital or non-hospital settings, in which (4) the cause of death was determined by a pathologist, in which (5) minimal demographic data (age, gender, date of death) was available, and (6) availability to share pathological material from the lungs obtained during the autopsies with the main research site (The Massachusetts General Hospital). The material shared could include scanned whole slide images (WSIs) if physical material could not be sent to MGH for different reasons. Appropriate ethics approval was also an inclusion criterion for all research sites within the consortium.
Exclusion criteria (International cohort)
We excluded sites that were unable to demonstrate laboratory confirmation of SARS-CoV-2. We also excluded research sites in which COVID-19 autopsy cases were evaluated as consultations and referrals for second opinions since the clinical data from those cases was often not available.
Eligible research sites, and site enrollment (International cohort)
After the search strategy was performed, a total of 103 research sites around the world were eligible to be contacted and included in the COVID-19 International lung autopsy consortium.
We contacted the corresponding authors of the selected studies and invited them to participate in the COVID-19 International lung autopsy consortium. A list of the references of scientific publications from all research sites that were contacted and invited to join the COVID-19 International lung autopsy consortium is included in the Supplementary Information. E-mails of the authors were obtained from the published manuscripts or online scientific platforms (e.g., https://www.researchgate.net/). The authors were contacted in English, with few exceptions in which we used their language of origin, based on the official language of the country of origin of the institution they listed in the main manuscript. Authors who did not respond on at least two occasions were deemed to have not replied. In multi-institutional studies we contacted authors from different institutions. If their e-mails were not listed in the manuscript, we contacted the corresponding authors and asked for the contact information of the authors affiliated to other institutions. In addition, we contacted authors with unpublished COVID-19 autopsy cohorts via our collaborators and included their cohorts if inclusion criteria were met.
Participating research sites (International cohort)
A total of 47 research sites in 16 countries and 4 continents were included in the COVID-19 International lung autopsy consortium. A list of all included sites is shown in SupplementaryTable 21.
We endeavored to contact the authors of the other 56 studies, but 11 research groups declined our request, 12 research groups had initially accepted and then were unresponsive after four attempts, three research groups mentioned that the pathological material was unavailable as it was not collected in their studies, 29 research groups were unresponsive after two attempts, and one research group mentioned that their cases could not be shared as they belonged to a third party that had sent them to this institution for second opinion and ancillary testing. A detailed list of the 103 research sites that were contacted, and their studies (when applicable), can be found in Supplementary Table 22.
Autopsy protocols and selection of pathological material in each research site (International cohort)
Briefly, autopsies were performed following internal protocols at each participating site. A selected anatomic pathologist in each site evaluated all available postmortem lung tissue. The underlying cause of death was determined in each case by the pathologist who performed the autopsy examination. Pathologists of the participating research site were asked to select the tissue slides, sections or blocks that were most representative of the pathological process occurring within the lung of each autopsy.
The research sites were asked to share pathological material with the MGH. The acceptable lung pathological material included: a) H&E-stained slides or b) whole slide images (WSIs) of H&E-stained slides from the lung parenchyma, and c) their corresponding FFPE lung tissue blocks, or d) their corresponding unstained FFPE tissue sections. There were no limits imposed, and the pathologist in each research site could submit as many FFPE tissue blocks, unstained tissue sections, or tissue slides that were deemed necessary. As for example if the pathological process seen in one COVID-19 autopsy included only acute diffuse alveolar damage (DAD) with hyaline membranes, the pathologist in the research site selected the most representative tissue slide/block. If there were two processes occurring in the lungs of one autopsy the pathologist was invited to select the two tissue slides/blocks that were most representative. In occasions, two pathological processes were present in a single tissue slide/block. In addition, some research sites also provided pathological material from premortem lung biopsies of patient with laboratory-confirmed SARS-CoV-2 infection.
Sample processing, and histopathological review (International cohort)
Selected H&E-stained slides, and their corresponding unstained FFPE tissue sections and/or FFPE tissue blocks were sent to MGH and archived at the international autopsy biorepository. Unstained FFPE tissue sections were stained with H&E in those cases in which H&E-stained slides were not provided by the research sites. The H&E-stained slides of each autopsy were reviewed by board-certified subspecialized thoracic pathologists. Different histopathological features were systematically recorded from all available H&E-stained tissue slides, and one representative tissue section from each case was stained with immunohistochemical markers (section 2.1).
Data request (International cohort)
Pre-specified individual-participant data and a pre-specified list of 269 variables were requested from authors of all eligible research sites. Data was retrieved from the medical records or the laboratory information system in each participating institution.
The variables included participant demographic characteristics, characteristics of the autopsy, gross and microscopic autopsy findings, underlying cause of death, characteristics of the submitted histopathological material, presenting symptoms and illness onset, characteristics of SARS-CoV-2 testing, pre-existing conditions, smoking history, characteristics of hospital stay, characteristics of intensive care unit care stay, clinical complications during hospitalization, microbiological testing during hospitalization and on autopsy, laboratory testing during hospitalization (hematology testing, inflammatory markers and acute phase reactants testing, clinical biochemistry testing, arterial blood gas testing), gas exchange and respiratory mechanics parameters, pharmacotherapy during disease course and hospitalization, radiographic findings and relevant clinical outcomes (e.g. oxygen- and ventilatory-support parameters, time to death from symptom onset, length of hospital stay). A detailed list of the requested data is described in the Supplementary Table 23.
This study follows PRISMA-IPD guidelines for individual-participant data reporting[53]. The consortium protocol is registered with PROSPERO (CRD42022237969)[54] and includes a pre-specified analytical plan. Data from eligible cohorts for aggregate data statistics were extracted by one author (JV).
Characteristics of the cohort (International cohort)
Of the participants with available FFPE tissue-blocks (n=329), 71% were at least 60 years old, 36% were female, the median time between symptom onset and death was 17 days (interquartile range [IQR] 11–28; n=262), and the median time of between las hospital admission and death was 11 days (IQR 5–21; n=313). A total of 311 of 329 participants (95%) were admitted to the intensive care unit, and 174 of 303 participants (57%) required invasive mechanical ventilation with a median duration of 10 days (IQR 5–20; n=172). During their hospitalization, 12% of patients (36/307) received Remdesivir, 10% (28/291) dexamethasone, and 42% (122/291) any type of corticosteroid therapy.
1.1.3 MGH ED cohort
To identify cellular and immune responses associated with outcome, clinicians and researchers at MGH enrolled longitudinally acutely ill patients in the Emergency Department (ED) in a large, urban, academic hospital in Boston, Massachusetts (with institutional review board approval) from March 24, 2020, to April 30, 2020, during the first peak of the COVID-19 surge in the U.S. Northeast. Participant enrollment is described in previously published reports[24][55].
The cohort included patients 18 years or older with a clinical concern for COVID-19 upon ED arrival, and with acute respiratory distress with at least one of the following:
1) tachypnea ≥ 22 breaths per minute;
2) oxygen saturation ≤ 92% on room air;
3) a requirement for supplemental oxygen; or
4) positive-pressure ventilation.
Participants with symptomatic SARS-CoV-2 infection confirmed by nucleic acid tests (n=126) were included in this study. Participants had blood drawn for proteomic analysis (using the Olink® platform) and for plasma SARS-CoV-2 viral mRNA load quantification on hospital days 0, 3, and 7. Clinical course was followed to 28 days after enrollment or until hospital discharge if that occurred beyond 28 days. Data from this longitudinal cohort is publicly available to the wider research community from the Olink® website[56].
Disease severity definitions (MGH ED cohort)
Enrolled participants of the MGH ED cohort who had a positive SARS-CoV-2 nasopharyngeal RT-PCR test (n=126) were categorized into five disease groups:
A1, death within 28 days;
A2, requiring mechanical ventilation and survival to 28 days;
A3, requiring hospitalization on supplemental oxygen within 28 days;
A4, requiring hospitalization without need of supplemental oxygen within 28 days; and
A5, discharge from ED and no subsequent requirement of hospitalization within 28 days.
Severe disease was defined as belonging to group A1 or A2. Data for classification was obtained from the electronic medical record.
RNAemia clearance definitions (MGH ED cohort)
For any participants enrolled in the MGH ED cohort whom the VL at admission (day 0) was greater than or equal to 3 log copies/ml, RNAemia clearance was defined as a VL at day 7 at least unquantifiable (<2 log copies/ml) or a VL at Day 3 at least unquantifiable (<2 log copies/ml if the VL at Day 7 was not available). For any participants in whom the VL at day 0 was lower than 3 log copies/ml, RNAemia clearance was defined as an undetectable VL at day 7 or a VL at day 3 at least unquantifiable (<2 log copies/ml if Day 7 VL not available).
Characteristics of the cohort (MGH ED cohort)
Half of the participants were at least 65 years old (63/126), 41.3% were female (52/126), and the median time between symptom onset and presentation to the ED was 7 days (IQR 4-11). 39 of 126 participants (31.0%) had a baseline SARS-CoV-2 viral load above the limit of quantification (2 log10 copies/mL), 76 (60.3%) had severe disease, 73 (57.9%) required invasive mechanical ventilation, and 19 (15.1%) died of COVID-19.
Details about plasma viral load quantification and proteomics analyses are described in sections 2.7, 2.8, and 2.9.
2. Procedures
2.1 Histopathological and immunohistochemistry analyses
Semiquantitative histological and IHC analyses were performed on cases from the three autopsy cohorts to characterize histopathological features and immunostaining patterns of selected ICs (Figure 2A-E; SupplementaryTable 24).
Histopathological features and semiquantitative scoring
Histological features that were recorded in all autopsy cohorts included the phases of lung injury (ALI, acute DAD, acute and organizing DAD, predominantly organizing DAD, and fibrotic DAD), and the presence of capillary congestion and dilatation.
Alveolar capillary vascular congestion and dilatation were scored in a semiquantitative way as described in Supplementary Table 25.
Immunohistochemistry
Chromogenic immunostains were performed in an automated platform (Leica BOND-III, Leica Biosystems, Deer Park, IL) and using antibodies against PD-L1/CD274, HAVCR2/TIM-3, PD-1, and CD8. Details about the antibodies and IHC protocols used in this study are included in Supplementary Table 24.
Semiquantitative immunohistochemistry scoring
All IHC-stained slides of the MGH autopsy discovery cohort were scored in a semiquantitative fashion on a scale of 0-3 by two board-certified thoracic pathologists [ARS, MMK] that were blinded to disease outcome and clinical characteristics of the patients. Aggregate measurements were compared across all cases. In the international cohort and the extended MGH autopsy cohort, all cases were scored in a semiquantitative fashion on a scale of 0-3 by one board-certified thoracic pathologist [ARS]. All data curation and statistical analyses performed with the immunohistochemical data from both cohorts was performed by another pathologist that was not involved in the semiquantitative review, and who was also blinded to clinical characteristics and disease outcomes of the patients. Semiquantitative scales used in the current study are described in Supplementary Table 26.
For survival analyses, participants were sub-grouped in two categories based on lung tissue density of PD-L1/CD274+ and HAVCR2/TIM-3+ cells as follows:
Lung tissue PD-L1/C274+ cell density
1- Participants with low-density of PD-L1/C274+ cells: participants with none-to-rare PD-L1/CD274 staining, which correspond to a semiquantitative score = 0
2- Participants with high-density of PD-L1/C274+ cells: participants with focal, multifocal, diffuse to DIP-like PD-L1/CD274 staining, which correspond to a semiquantitative score > 0
Lung tissue HAVCR2/TIM-3+ cell density
1- Participants with low-density of HAVCR2/TIM-3+ cells: participants with none to rare/weak HAVCR2/TIM-3 staining, which correspond to a score ≤ 1
2- Participants with high-density of HAVCR2/TIM-3+ cells: participants with moderate to strong/scattered to patchy HAVCR2/TIM-3 staining, which correspond to a semiquantitative score > 1
RNA in situ hybridization
Chromogenic RNA in situ hybridization (RNA-ISH) was performed on FFPE tissue sections of the MGH COVID-19 autopsy discovery cohort (in COVID-19 and non-COVID-19 DAD controls) using SARS‐CoV‐2 RNA‐specific probes RNAscope 2.5 LS Probe-V-nCoV2019-S [Advanced Cell Diagnostics (ACD), Newark, CA, USA; 848568] and RNAscope 2.5D DAB on an automated diagnostic Leica Biosystems BOND-III apparatus (Leica Biosystems, Danvers, MA, USA). The probe is specific for the S gene encoding the spike protein. RNA-ISH was performed on 5‐µm‐thick FFPE lung tissue sections. All of the steps from baking for 1 h at 60°C to counterstaining with hematoxylin were performed on the BondRx machine. RNA unmarking was performed with Bond Epitope Retrieval Solution 2 for 15 min at 95°C, and this was followed by protease treatment for 15 min and probe hybridization for 2 h. Signals were amplified by a series of signal amplification steps, and this was followed by color development. The reactivity was visualized as brown dots. Appropriate controls were used when performing RNA-ISH to confirm RNA preservation.
RNA-ISH-stained slides from autopsies of the MGH COVID-19 discovery cohort were scored in a semiquantitative fashion by two board-certified thoracic pathologists and aggregate measurements were compared across all cases. Each case had two RNA-ISH-stained slides which were scored on a scale of 0-3, as described in Supplementary Table 27.
2.2 Machine-learning derived morphometrics
Briefly, a random forest classifier was trained and used to label every pixel in each H&E-stained lung whole slide image (WSI) as belonging in one of three categories: stroma, red blood cell (RBC), or air. Two H&E-stained histological slides were selected by the same pathologist and scanned under 80X optical magnification at approximately 8,200 pixels per micron. The two H&E-stained slides included the most congested and the least congested lung parenchymal region out of all 50 different lung tissue histological slides per autopsy (10 H&E-stained sections per each lung lobe).
The alveolar septal CVasc represents the proportion of surface area occupied by RBC pixels in (dilated) vascular lumina in one cluster of 10 contiguous alveoli that were manually annotated by a board-certified pathologist in a WSI. We annotated three clusters per WSI, and a total of six clusters per autopsy case as two slides were selected. Intra-alveolar contents (e.g., inflammatory cells, hemorrhage, fibrin, and debris) were excluded from the alveolar manual annotations. Of note, annotations were not performed within areas of lung infarction and necrosis, areas with extensive intra-alveolar hemorrhage or inflammation, areas of fibrosis, or areas in which the lung microarchitecture was severely disrupted. Analyses were performed within QuPath[57], which uses the OpenCV library for pixel classification. The final classification was applied within the clusters of alveoli to produce detailed quantitative measurements. The quantitative data of the classification in the alveolar clusters were used for morphometrical determination of the CVasc, which represents the proportion of surface area occupied by (dilated) vascular lumina in the alveolar clusters of the lung parenchyma, defined as:
Morphometric data were obtained from all of the six alveolar clusters of contiguous alveoli that were manually annotated per case. From each autopsy, we also separately designated one individual annotation with the highest CVasc, also known as “interslide-MaxCVasc”, and another annotation with the lowest CVasc, also known as “interslide-MinCVasc”, out of the six annotations of alveolar clusters per autopsy case.
Subsequently, we stratified the cases into two morphometrically defined subgroups based on the median interslide-MaxCVasc and the median interslide-MinCVasc of the COVID-19 cases of the discovery cohort. We used these morphometrically defined subgroups to evaluate differences in transcriptomic data and to perform downstream differential gene expression analyses as described in section 2.3. The two morphometrically defined subgroups per each parameter were named as follows:
1- For interslide-MaxCVasc:
-Patients with an interslide-MaxCVasc less than the median interslide-MaxCVasc of the cohort were regarded as “cases with minimally-congested-alveolar-septa in highly congested areas of the lung”
-Patients with a mean interslide-MaxCVasc equal to or greater than the median interslide-MaxCVasc of the cohort were regarded as “cases with highly-congested-alveolar-septa in highly congested areas of the lung”.
2- For interslide-MinCVasc:
-Patients with an interslide-MinCVasc less than the median interslide-MinCVasc of the cohort were regarded as “cases with minimally-congested-alveolar-septa in minimally congested areas of the lung”
-Patients with a mean interslide-MinCVasc equal to or greater than the median interslide-MinCVasc of the cohort were regarded as “cases with highly-congested-alveolar-septa in minimally congested areas of the lung”.
2.3 Transcriptomic profiling, pathway analysis, and differential gene expression analysis
Downstream differential gene expression (DGE) and pathway enrichment analyses were performed for COVID-19 patients in the MGH COVID-19 autopsy discovery cohort whose formalin-fixed paraffin-embedded lung-tissue-derived RNA was available. We compared DGE in morphometrically defined participant groups classified by a machine-learning derived analysis of the pulmonary microvasculature (Supplementary Figure 3).
FFPE tissue-derived RNA counts generated by the NanoString nCounter® Coronavirus Panel Plus[58][59] underwent quality control and normalization with housekeeping genes using the nSolver software version 4.0 (NanoString Technologies). The NanoString nCounter® Coronavirus Panel Plus was added to the nCounter® Human Organ Transplant Panel created through a collaboration between NanoString and the Banff Foundation for Allograft Pathology. The latter includes 770 genes related to immune, inflammation and tissue damage [60]. The downstream analyses were performed on R using DESeq2 package[61] and RUVSeq package[62][63] was used to remove unwanted variation with internal reference genes defined in the Banff 2019 Meeting report[58]. Gene set enrichment analysis (GSEA) was conducted using the fgsea package[64] and gene expression was compared against Molecular Signatures Database Hallmark gene set[65]. We used the whole gene set for GSEA, taking their normalized differences as input (refer to https://stephenturner.github.io/deseq-to-fgsea/). Weighted correlation network analysis (WGCNA) was used to identify gene modules and gene adjacency[66].
We used the STRING v11.5 (Search Tool for the Retrieval of Interacting Genes/Proteins) knowledgebase and software tool to assess for known and predicted protein-protein interactions (PPI), co-expression, and association[67]. The top 100 adjacent genes were assessed in the STRING and we used K-means to cluster the genes. The cluster that contained gene of interest was singled out and we selected the gene interaction with high confidence to build the final interaction network.
2.4 Single-cell RNA sequencing (scRNA-Seq) data exploration
Four publicly available single-cell RNA sequencing (scRNA-Seq) datasets[18][19][20][21] were interrogated to evaluate expression patterns of genes encoding ICs (HAVCR2, PDCD1, CD274, LGALS9, CTLA4, LAG3, TIGIT, and IL2RA) and other proteins of interest. scRNA-Seq data from the publicly available datasets[18][19][20][21] were accessed and imported to Seurat v4.0 for further analyses. Doublets were removed. We used the same cell identity as labeled in the original publications. mRNA expression levels and percentage in different cell types were extracted from the dataset expression matrices. The mean expression level for each gene was scaled against its highest absolute mean expression level among different cell types. Uniform manifold approximation and projection (UMAP) graph and heatmaps were plotted using Seurat 4.0.
2.5 Survival analyses
We explored whether the degree of PD-L1/CD274 and HAVCR2/TIM-3 expression in different cell subpopulations correlated with the timing of progression to death in autopsy cases of our cohorts. We decided not to investigate PD-1 as a prognostic marker due to its rarity of cellular staining observed in the MGH discovery cohort. Time-to-events in the two COVID-19 autopsy cohorts were defined as time-to-death from symptoms onset, length of hospital stay, time-to-death from intensive care unit (ICU) admission, time-to-death from initiation of in-hospital oxygen therapy, and time-to-death from initiation of mechanical ventilation. No data was censored. Kaplan Meier analysis was used, and groups were compared using the log-rank (Mantel-cox) test.
We also performed Cox proportional hazard model analyses with data from the two autopsy cohorts to estimate hazard ratios and to test whether the association of chromogenic IHC semiquantitative expression of PD-L1/CD274 and HAVCR2/TIM-3, and different time-to-event outcomes would be affected by different clinical variables of the cohorts, including age, sex, body mass index (BMI), corticosteroid and remdesivir therapy.
2.6 Plasma SARS-CoV-2 viral load quantification
Plasma SARS-CoV-2 viral load (VL) measurement was performed with biological samples from the MGH ED cohort as reported in previous studies[68][24]. Briefly, SARS-CoV-2 VL was quantified using the US CDC 2019-nCoV_N1 primers and probe set[68]. The lower limit of SARS-CoV-2 N gene quantification was 100 copies/mL. Samples with a positive signal but VL lower than 100 copies/mL were denoted as detectable but unquantifiable.
2.7 Proteomics analyses
Proteomic analyses used in this study have been previously described in prior studies[1][55]. Briefly, proteins in plasma from participants of the MGH ED cohort were measured using the Olink® proximity extension assay, a high-dimensional multiplex oligonucleotide-labeled antibody assay for high-specificity measurement of low-abundance proteins. Signal amplification is performed by PCR. Therefore, measurements are expressed in normalized protein expression (NPX) units, reflecting relative abundance on a log2 scale, rather than absolute concentration. This assay allows within-analyte comparisons between different samples but not between-analyte comparisons. For example, HAVCR2/TIM-3 plasma levels between two different patients can be compared, but the HAVCR2/TIM-3 plasma level cannot be directly compared to interleukin-6 plasma level in the same patient. Analytical performance validation for each protein assay is available online (www.olink.com). Information regarding protein expression at the tissue and blood cell levels, protein function, and protein localization was derived from the Human Protein Atlas[69][70].
We have previously reported that SARS-CoV-2 RNAemia is associated with adverse outcomes in COVID-19[24][71]. To evaluate the role of immune checkpoints as immunologic mediators of viral bloodstream spread, we measured plasma NPX levels of immune checkpoints in participants presenting to an emergency department and who had viral ‘clearance’ or viral ‘persistence’ at hospital day 7 as defined above in section 1.1.3.
2.8 Sparse Partial Least Squares Discriminant Analysis (sPLS-DA)
In order to define a minimal set of the most important plasma proteins associated with severe COVID-19 and RNAemia persistence, we used sparse partial least squares discriminant analysis (sPLS-DA) to evaluate participants presenting to the ED with confirmed SARS-CoV-2 infection and available clinical data within 28 days of hospital admission (n=126).
Sparse PLS-DA was conducted using the “mixOmics” package in R 4.2.0[72][73]. We included 5 components for model tuning, with ten-fold cross validation, balanced error rate (BER) and centroids distance for tuning performance measurement. Each feature was scaled before performing Sparse PLS-DA. Receiver operating characteristic (ROC) curve was plotted using the “pROC” package. 95% confidence interval was calculated using the deLong method. Correlation network graph was plotted using the “igraph” and “ggraph” package in R. Heatmap was plotted using the “pheatmap” package.
3. Other analytic considerations
We summarized continuous variables using median and IQRs. Categorical variables were evaluated using the χ2 test or Fisher’s exact test. To compare multiple groups, a Kruskal–Wallis test was used with a Dunnett test correcting for multiple comparisons. We used Spearman’s rank correlation coefficient to evaluate correlation between different continuous variables, and the P values were corrected by Benjamini–Hochberg correction. Five distinct measures of disease progression were evaluated in the MGH COVID-19 autopsy development cohort and in the COVID-19 International cohort. These included time-to-death from symptom onset, hospital length-of-stay, intensive care unit (ICU) length-of-stay, length-of-oxygen-therapy, and length-of-invasive mechanical ventilation. Each outcome of disease progression was modeled independently, and unadjusted and adjusted models were compared. Survival analyses were performed using the Kaplan–Meier method, and groups were compared by log-rank test. Cox proportional-hazards models were used to estimate hazard ratios. Analyses were performed using R (version 4.1.0), PythonSciPy v.1.4.1, JMP 16.0 (SAS Institute Inc.), and GraphPad Prism 7.0 (GraphPadSoftware, Inc.).
Yes there is potential Competing Interest.
Julian A. Villalba is a federal employee of the U.S. government at the Division of High-Consequence Pathogens and Pathology at the Centers for Disease Control and Prevention. This work was not funded by CDC, and the data, results, and opinions included in this manuscript do not reflect the views or positions of this federal agency. Yijia Li reports receiving consultation fees from DynaMed as a topic editor. Arnav Mehta has served a consultant/advisory role for Third Rock Ventures, Asher Biotherapeutics, Abata Therapeutics, ManaT Bio, Flare Therapeutics, venBio Partners, BioNTech, Rheos Medicines and Checkmate Pharmaceuticals, is currently a part-time Entrepreneur in Residence at Third Rock Ventures, is an equity holder in ManaT Bio, Asher Biotherapeutics and Abata Therapeutics, and has received research funding support from Bristol-Myers Squibb. Giulia D’Amati has research funding support from NextGenerationEU (PNRR “HEAL ITALIA - Health Extended Alliance for Innovative Therapies, Advanced Lab-research, and Integrated Approaches of Precision Medicine” n. 3021, project: PE00000019, to Giulia d’Amati CUP: B53C22004000006). Moshe Sade-Feldman received funding from Calico Life Sciences, Bristol-Myers Squibb, and Istari Oncology and served as a consultant for Galvanize Therapeutics. James Stone is the Hub-Site PI for the NIH RECOVER Consortium, receives grant support from the NIH RECOVER Consortium, is the convening chair for the Autopsy Coordinating Committee of the NIH RECOVER Consortium, and is a member of the Steering Committee for the NIH RECOVER Consortium. Yin P Hung has received royalties from Elsevier for editorial work. Lida Hariri reports receiving consultation fees from Boehringer Interim, Pliant Therapeutics, Clario and Abbvie, and unrelated grant funding from NIH (R01HL152075), DoD (W81XWH2210713), and Boehringer Ingelheim. The remaining authors declare no competing interests.