Public proteomics data
The data used in the present study are publicly available from the Massachusetts General Hospital Emergency Department COVID-19 Cohort (Filbin, Goldberg, Hacohen) with Olink Proteomics (https://www.olink.com/mgh-covid-study/). We named this the MGH COVID-19 cohort. The patients in this cohort were diagnosed as having mild-to-critical COVID-19 according to the definitions of the National Institutes of Health COVID-19 Treatment Guidelines : Mild: varied symptoms but no shortness of breath, dyspnea or abnormal imaging; Severe: SpO2 <94%, PaO2/FiO2 <300, respiratory rate >30 breaths/minute, or lung infiltration >50%; and Critical: respiratory failure, septic shock and/or multi-organ dysfunction. The present study was performed in an urban academic hospital in Boston from March 2020 to April 2020. This observational study included 306 COVID-19 patients. The diagnosis of COVID-19 was based on a positive SARS-CoV-2 RT-PCR result. The patients were classified according to their acuity levels (A1–A5) on days 1, 4, 8 and 29 (based on the World Health Organization ordinal outcomes scale : A1, dead; A2, intubated, survived; A3, hospitalized with oxygen; A4, hospitalized without oxygen; A5, discharged). The following clinical data were collected: age, body mass index (BMI), comorbidities, and maximal acuity (Acuitymax). Blood samples on day 1 were considered to be those obtained when the initial clinical blood draw was performed in the Emergency Department. Samples were obtained on days 4 and 8 if the patient remained in hospital.
Four panels of proteins (inflammation, oncology, cardiometabolic and neurology proteins) were used to evaluate 1,472 plasma proteins, including 1,463 unique proteins (Olink® Explore 1536). We assessed the relationship between resistin, which was included in the 1,463 proteins, and the prognosis of COVID-19 patients. The Olink multiplex proximity extension assay is a dual-recognition immunoassay. In this assay, two matched antibodies labeled with unique DNA oligonucleotides are used to bind the target proteins. Upon binding, these oligonucleotides come into close proximity and hybridize. This is followed by extension, with the generation of a unique sequence used for the digital identification of the specific protein assay (www.olink.com). Plasma (2.8 µl) was incubated overnight with oligonucleotide-labeled antibody pairs to form specific DNA duplexes. Extension and pre-amplification were then performed together, and the measurement of individual protein markers was performed using a Nova-Seq 6000 system. These counts were normalized to an extension control and an inter-plate control and then adjusted using a correction factor to calculate the normalized protein expression value (NPX) in log2 scale.
Study design and patient characteristics of the Osaka cohort
A prospective observational multicenter study of COVID-19 patients was conducted at the Department of Traumatology and Acute Critical Care Medicine, Osaka University Graduate School of Medicine and the Osaka Prefectural Nakakawachi Emergency and Critical Care Center from August 2020 to December 2020. All of the study patients were diagnosed as having SARS CoV-2 (based on a positive RT-PCR test) and pneumonia (based on computed tomography imaging). This group was designated the Osaka COVID-19 cohort. Patients with sepsis who were admitted to the Department of Traumatology and Acute Critical Care Medicine, Osaka University Graduate School of Medicine between February 2014 and July 2015 were included as an ICU control group. The patients with sepsis were all over 18 years of age, and all patients met the Sepsis-3 criteria. Outpatients who were recruited from public poster advertisements were included as a healthy control population.
Demographic variables (age, sex, BMI), Acute Physiology and Chronic Health Evaluation (APACHE) II score  and Sequential Organ Failure Assessment (SOFA) score , comorbid conditions (hypertension, diabetes and hyperlipidemia) and clinical variables (day of discontinuation of MV and discharge status) were extracted from electronic medical records by the investigators.
Analysis of resistin, inflammatory cytokines and endothelial damage markers
ELISAs (R&D Systems, Minneapolis, MN, USA) were performed to measure the plasma levels of resistin, IL-6, IL-8, IL-10, MCP-1, intercellular adhesion molecule 1 (ICAM-1) and vascular cell adhesion molecule-1 (VCAM-1). After thawing of frozen plasma samples, measurement was conducted according to the manufacturer’s protocol. A microplate reader (SH-9000Lab; Corona Electric Co., Ltd., Ibaraki, Japan) was used to measure absorbance. The minimum detectable levels were as follows: resistin, IL-8, IL-10 and ICAM-1: 31.2 pg/ml; MCP-1 and VCAM-1: 15.6 pg/ml; and IL-6: 9.4 pg/ml.
mRNA expression of resistin in whole blood
Total RNA was isolated from day 1 leukocyte samples from 10 COVID-19 patients and 5 healthy controls in the Osaka cohort using the PAXgene™ Blood RNA System (BD Bioscience, San Jose, CA, USA). After blood collection, all collection tubes were stored until further use at -30°C. We performed library preparation using a TruSeq stranded mRNA sample prep kit (Illumina, San Diego, CA, USA) according to the manufacturer’s instructions. Sequencing was performed with an Illumina NovaSeq 6000 platform in 101-base paired-end mode. Sequenced reads were mapped to the human reference genome sequences (hg19) using the TopHat (version 2.0.13) software program, with Bowtie 2 (version 2.2.3) and SAMtools (version 0.1.19). We calculated the fragments per kilobase of exon per million mapped fragments using the Cufflinks software program (version 2.2.1). FeatureCounts was used to determine the gene-level expression raw read counts. Raw data from this study were submitted for future access under Gene Expression Omnibus accession number GSE179850.
Values are reported as n (%) and the median value (quartile 1–3) if the data distribution was skewed, or as the mean ± SE unless stated otherwise.
In the MGH COVID-19 cohort, the patients were divided into four groups based on the quartiles of the day 1 resistin NPX. The proportion of Acuitymax was calculated for each group. The difference in the proportion was compared using the chi-squared test. The Wilcoxon rank-sum test was used to evaluate the differences between survivors and non-survivors on day 1, day 4 and day 8. To investigate whether the day 1 resistin value is useful as a prognostic biomarker, a receiver operating characteristic (ROC) curve was created. The area under the curve (AUC), accuracy, sensitivity and specificity were also determined.
In the Osaka COVID-19 cohort, resistin, inflammatory cytokines and endothelial damage markers were transformed to logarithmic values to normalize the data distribution before the analyses. Dunnett’s test was used to evaluate the difference of each value between patients and healthy controls. Because the median day of weaning off MV was 12 days after intubation in the validation cohort, and that of tracheostomy was 12 days after intubation in a large observational study in Spain , treatment with MV for ≤12 days was defined as early recovery, whereas MV >12 days and hospital death were defined as late recovery or death, respectively. The patients were divided into two groups in the acute phase (day 1, days 2–3, and days 6–8): survivors and non-survivors or early recovery and late recovery or death. Wilcoxon rank-sum tests were used to evaluate differences between two groups on each day. A Cox proportional hazards analysis was performed to estimate hazard ratios and their 95% confidence intervals (CIs) for the time to wean off MV. The levels of resistin, inflammatory cytokines and endothelial damage markers were analyzed as time-dependent covariates (i.e., the value was updated prospectively for each sample taken). We used the Z score of the log-transformed level in each biomarker to the strength of association for comparison among biomarkers. Recognizing that the clinical course may be affected by various factors, we extended this model to adjust for multiple potential confounders, including sex and age, which were found to be associated with the prognosis in COVID-19 [19–21]. We also calculated the median time to wean off MV in patients divided into two groups according to the quartile of the resistin level determined in a Kaplan-Meier analysis with time as a dependent covariate.
Correlations between resistin, and inflammatory cytokines, and endothelial damage markers were evaluated by a hierarchical clustering analysis based on Spearman correlation coefficients. A network analysis was performed with Cytoscape® software (www.cytoscape.org) version 3.8.0. Log2 fold changes were calculated by dividing the average mediator levels in COVID-19 or sepsis patients by the average levels in healthy controls.
The mRNA expression of resistin was compared between COVID-19 patients and healthy controls by a Wilcoxon rank-sum test.
Statistical analyses were performed using the R software program (version 4.0.2; R Foundation for Statistical Computing, Vienna, Austria). Data are presented using the GraphPad Prism software program (version 8.4.3, GraphPad Software, La Jolla, CA). Statistical significance was defined as P <0.05.