Highlights d AI system that can diagnose COVID-19 pneumonia using CT scans d Prediction of progression to critical illness d Potential to improve performance of junior radiologists to the senior level d Can assist evaluation of drug treatment effects with CT quantification

It was recently brought to our attention that our paper was missing information regarding when the patient chest computed tomography (CT) scans were obtained and that there were some discrepancies in the clinical metadata, associated with the very large image dataset, that we made publicly available through the China National Center for Bioinformation (http://ncov-ai.big.ac.cn/ download?lang=en). All of the chest CT and clinical metadata used in our prognostic analysis were collected from patients at the time of hospital admission, and we have now added this statement to the STAR Methods section of our paper. We believe that the errors in the clinical metadata were introduced when the chest CT images, clinical metadata, and codes were transferred to the web server, and we have now corrected the errors manually. Although these corrections do not alter any of the conclusions made in the paper, we do apologize for these errors and any confusion that they may have caused.

BackgroundRegulatory B (Breg) cells represent one of the B cell subsets that infiltrate solid tumors and exhibit distinct phenotypes in different tumor microenvironments. However, the phenotype, function and clinical relevance of Breg cells in human hepatocellular carcinoma (HCC) are presently unknown.MethodsFlow cytometry analyses were performed to determine the levels, phenotypes and functions of TIM-1+Breg cells in samples from 51 patients with HCC. Kaplan-Meier plots for overall survival and disease-free survival were generated using the log-rank test. TIM-1+Breg cells and CD8+ T cells were isolated, stimulated and/or cultured in vitro for functional assays. Exosomes and B cells were isolated and cultured in vitro for TIM-1+Breg cell expansion assays.ResultsPatients with HCC showed a significantly higher TIM-1+Breg cell infiltration in their tumor tissue compared with the paired peritumoral tissue. The infiltrating TIM-1+Breg cells showed a CD5highCD24−CD27−/+CD38+/high phenotype, expressed high levels of the immunosuppressive cytokine IL-10 and exhibited strong suppressive activity against CD8+ T cells. B cells activated by tumor-derived exosomes strongly expressed TIM-1 protein and were equipped with suppressive activity against CD8+ T cells similar to TIM-1+Breg cells isolated from HCC tumor tissue. Moreover, the accumulation of TIM-1+Breg cells in tumors was associated with advanced disease stage, predicted early recurrence in HCC and reduced HCC patient survival. Exosome-derived HMGB1 activated B cells and promoted TIM-1+Breg cell expansion via the Toll like receptor (TLR) 2/4 and mitogen-activated protein kinase (MAPK) signaling pathways.ConclusionsOur results illuminate a novel mechanism of TIM-1+Breg cell-mediated immune escape in HCC and provide functional evidence for the use of these novel exosomal HMGB1-TLR2/4-MAPK pathways to prevent and to treat this immune tolerance feature of HCC.Electronic supplementary materialThe online version of this article (10.1186/s40425-018-0451-6) contains supplementary material, which is available to authorized users.

Human UC-MSCs are regarded as an attractive alternative to BM-MSCs for clinical applications due to their easy preparation, higher proliferation and lower immunogenicity. However, the mechanisms underlying immune suppression by UC-MSCs are still unclear. We studied the mechanism of inhibition by UC-MSCs during the differentiation of monocytes into DCs and focused on the specific source and the role of the involved cytokines. We found that UC-MSCs suppressed monocyte differentiation into DCs and instructed monocytes towards other cell types, with clear decreases in the expression of co-stimulatory molecules, in the secretion of inflammatory factors and in allostimulatory capacity. IL6, HGF and IL10 might be involved in this process because they were detected at higher levels in a coculture system. UC-MSCs produce IL-6 and HGF, and neutralization of IL-6 and HGF reversed the suppressive effect of UC-MSCs. IL10 was not produced by UC-MSCs but was exclusively produced by monocytes after exposure to UC-MSCs, IL-6 or HGF. In summary, we found that the UC-MSC-mediated inhibitory effect was dependent on IL6 and HGF secreted by UC-MSCs and that this effect induced monocyte-derived cells to produce IL10, which might indirectly strengthen the suppressive effect of UC-MSCs.

Common lung diseases are first diagnosed via chest X-rays. Here, we show that a fully automated deep-learning pipeline for chest-X-ray-image standardization, lesion visualization and disease diagnosis can identify viral pneumonia caused by Coronavirus disease 2019 (COVID-19), assess its severity, and discriminate it from other types of pneumonia. The deep-learning system was developed by using a heterogeneous multicentre dataset of 145,202 images, and tested retrospectively and prospectively with thousands of additional images across four patient cohorts and multiple countries. The system generalized across settings, discriminating between viral pneumonia, other types of pneumonia and absence of disease with areas under the receiver operating characteristic curve (AUCs) of 0.88–0.99, between severe and non-severe COVID-19 with an AUC of 0.87, and between severe or non-severe COVID-19 pneumonia and other viral and non-viral pneumonia with AUCs of 0.82–0.98. In an independent set of 440 chest X-rays, the system performed comparably to senior radiologists, and improved the performance of junior radiologists. Automated deep-learning systems for the assessment of pneumonia could facilitate early intervention and provide clinical-decision support.