BQC19 is a provincial-wide biobank established in March 2020 to enable collection, storage, and sharing of samples and data of people affected by COVID-19 disease in Quebec, Canada (see bqc19.ca) (18).

Briefly, participants were recruited at the hospital and invited to participate between March 2020 and August 2021. Patients with SARS-CoV-2 diagnosis validated by RT-PCR or disease presentation were included. Medical history and clinical data were obtained from questionnaires and medical charts, and blood was collected at several timepoints.

After selecting for data from adult participants, we considered samples, clinical and proteomics data from 926 adults diagnosed with COVID-19 and 285 hospitalized controls. COVID-19 was diagnosed using RT-qPCR testing. All controls had negative COVID-19 tests. Controls were hospitalized for various reasons such as infectious respiratory symptoms (not linked to COVID-19), acute cardiovascular events, emboly, digestive symptoms not requiring surgery, uncontrolled diabetes. For participants with multiple blood samples collected during hospitalization, the samples collected at the earliest time point related to symptom onset were used to reflect the proteome of acute COVID-19, and to avoid use of measures during the recovery phase. No participant had received COVID-19 vaccination at the time samples were collected.

COVID-19 participants were categorized into mild, moderate or severe infections based on criteria from the WHO Working Group on the Clinical Characterisation and Management of COVID-19 (19). Briefly, mild or ambulatory disease included patients with asymptomatic presentation while having detectable SARS-CoV-2, or symptomatic independent, or requiring low assistance. Moderate diseases encompass hospitalized patients not requiring oxygen therapy or requiring oxygen by mask or nasal prongs. Severe disease refers to hospitalized patients requiring oxygen, intubation or mechanical ventilation, dialysis or ECMO. Participants who died during their hospitalization were included in the “fatal” infection group.

A total of 15 recovered COVID-19 participants were recruited at the McGill University Health Centre. Inclusion criteria included being adult, having received a diagnostic of COVID-19 at more than 14 days ago, and not having COVID-19 symptoms. People presenting with long-COVID-19 symptoms were excluded. Sample from controls were collected before December 2019 in Quebec, Canada, or from January 2020 to January 2021, with no history of COVID-19 tests or symptoms.

Whole blood was obtained through venipuncture using acid-citrate-dextrose vacutainer tubes or EDTA tubes for recovered participants, and plasma was separated by centrifugation at 750g, 10 min at room temperature. Isolated plasma was aliquoted and stored at −80°C until analysis.

Blood samples from a total of 1211 BQC19 participants were included for the plasma proteomics analysis. Proteomic profiles were assessed at SomaLogic using the SomaScan v4.0 proteomic platform that provides measurements on 4701 unique human circulating proteins using 4987 Slow Off-Rate Modified Aptamers (SOMAmer reagents) and quantifies protein levels in the form of relative fluorescence units (RFUs) (20). Experimental process and data normalization including hybridization control normalization, intraplate median signal normalization, and plate scaling and calibration were performed as previously described (20, 21).

When samples were available, plasma levels of human GDF-15 were measured in duplicates using the enzyme-linked immunosorbent assay (ELISA) assay (R&D systems, MN, USA) as per the manufacturer’s instructions.

Spike-specific IgG levels were quantified using a cell-based ELISA (CBE) method as previously described (22). Briefly, SARS-CoV-2 Spike-expressing HOS cells were washed and incubated with diluted plasma (1:250). After wash, anti-human IgG antibody coupled to horseradish peroxidase (HRP) were incubated. After wash and substrate addition, light emission was detected using a luminometer.

Statistical analyses were conducted using GraphPad Prism 9.0 (GraphPad, CA, USA). Spearman’s rank correlation test identified associations between 2 continuous variables. Mann-Whitney U test and student t-test were used to compare levels of continuous variables between two independent groups, as appropriate. Kruskal-Wallis one way ANOVA test was used to compare levels of continuous variables in more than 2 independent study groups. Paired analysis was performed by Wilcoxon signed-rank test. P-values < 0.001 were considered significant for samples with n>250, and p<0.05 for samples less<250. Logistic regression univariable models were used to generate Receiver operating characteristic (ROC) curve. Multivariable analysis was performed using SPSS.

BQC19 received ethical approval from the institutional review board (IRB) of the Jewish General Hospital and the Centre Hospitalier de l’Université de Montréal (CHUM) in Montréal, QC, Canada. All participants gave informed consent. Secondary analysis of BQC19 data and additional analyses for this project were approved by the research ethics board of the McGill University Health Centre (MUHC).

The demographic and clinical characteristics of all study participants are presented in Table 1 . In total, there were 926 RT-qPCR positive COVID-19 cases (median age: 60; range 18-99), encompassing 46% female and 54% male. In addition, 285 RT-qPCR COVID-19 negative hospitalized participants were included as controls (median age 56, range 20-99): 47% were female and 53% were male. In the RT-qPCR confirmed COVID-19 positive cohort (n=926), 256 were classified as mild, 384 as moderate, 235 as severe and 24 as fatal using WHO criteria. Clinical outcomes and comorbidities were extracted from clinical charts and assessed for all participants including those who had died.

Proteomics assay measured plasma protein levels in all COVID-19 infected and control adult participants. Plasma GDF-15 levels were significantly higher by 1.7-fold in COVID-19 infected participants compared to controls (p<0.0001) ( Figure 1A ). In COVID-19 positive patients, as COVID-19 severity worsened from mild to fatal, the plasma GDF-15 levels increased with the highest median GDF-15 levels detected in the fatal group (p<0.0001) ( Figure 1B ). Severe COVID-19 groups also had significantly greater plasma GDF-15 elevation compared to moderate and mild COVID-19 groups (p=0.0049, p<0.0001 respectively). Multivariable analysis showed that age, but not sex nor presence of comorbidities, was a confounding factor of the association between GDF-15 and disease severity ( Table 2 ), because GDF-15 was associated with age (Manova p < 0.0001) but not Severity (Manova p = 0.89) ( Table 2 ) Age-adjusted regression analysis increased association between plasma GDF-15 and COVID-19 severity (unadjusted β: 6.10⁶, adjusted β: 0.222). ROC analysis was conducted to evaluate the predictive ability of plasma GDF-15 to discriminate mild and severe COVID-19 states (including fatal cases) ( Figure 1C ). The area under the curve (AUC) was estimated for observed GDF-15 levels and their predicted values by fitting regression models. Observed GDF-15 levels were able to predict COVID-19 disease severity (AUC=0.82 ± 0.02, p<0.0001). These results support the potential of plasma GDF-15 levels in predicting COVID-19 severity.

Plasma levels of other 5200 markers were compared in patients with mild and severe/fatal disease presentation. GDF-15 was found as one of the 461 proteins associated with severity, with the most significant q and p value (-log₁₀ 118.2; p<0.00001, Supplementary Figure 1 ). Q value was less significant for IL-6 and CRP (-log₁₀ 35.81 and 108.3, respectively, p<0.00001 for both), two common biomarkers of severity.

We compared plasma GDF-15 levels between the initial wave of COVID-19, spanning from March to July 2020, to those of the second wave from august 2020 to August 2021 in the Province of Quebec. Patients in both waves were mostly affected by the ancestral SARS-CoV-2 strains in Quebec. No participants received priori vaccination even though vaccination campaign started in 2021 in Quebec. Plasma levels of GDF-15 were notably higher during the first wave compared to those observed during the second COVID-19 wave (p<0.001) (see Figure 1D ). The variation was independent of comorbidities, and sex in the two collection periods. However, the age of the patients during the second wave was lower than the age of patients in the first wave (57.4 vs 66.9, p<0.001).

A total of 926 COVID-19 infected patients were included in this study. The most common comorbidities were cardiovascular disease (CVD) (n=488, 52.70%), diabetes (247, 26.67%), cancer (105, 11.34%), obesity (91, 9.83%), COPD (84, 9.07%) and HIV (7; 0.76%) ( Table 3 ). Compared to hospitalized controls, COVID-19 infected patients exhibited higher average plasma GDF-15 levels regardless of presence or absence of comorbidities. In all COVID-19 infected patients, GDF-15 levels were consistently higher in those with comorbidities groups. Diabetes, CVD, COPD, and cancer were the conditions significantly associated with higher GDF-15 levels (p<0.0001 for the three comparisons) ( Figure 2A ). Although higher levels of GDF-15 were noticed in obese (body mass index greater than 30) patients with COVID-19, this difference was not significant (p = 0.24). Studies have found a higher risk of hospitalization in Canadian PLWH (23, 24). We found no difference in GDF-15 levels between PLWH and HIV-negative patients, although the number of PLWH was small (n=7). Such observations indicate that elevated plasma GDF-15 levels in COVID-19 appears independent of patient’s comorbidities. We next examined the impact of these comorbidities on GDF-15 expression in severe and fatal COVID-19 patients. GDF-15 levels in these groups remained higher than that of hospitalized controls. Significantly higher levels of GDF-15 expression were observed in patients with CVD, COPD and cancer (p<0.0001, 0.0030 and <0.001 respectively) compared to severe cases without those conditions. Differences in obese or diabetic participants were not significantly different ( Figure 2B ). Multivariable analysis showed that increased levels of GDF-15 in patient with severe vs mild and moderate disease was independently associated with severity while adjusting for the presence of any comorbidity, sex and but was influenced by age ( Table 4 ).

To validate findings made using SOMAscan proteomics, we assessed plasma GDF-15 by ELISA. We compared GDF-15 levels during the acute phase, early after recovery (less than 30 days after hospitalization) or up to 6 months after recovery in different groups. We examined GDF-15 levels in left-over plasma samples from 21 disease-free control participants; 28 acute patients with 0 to 2 days of hospitalization; 10 early recovered patients with an average of 30 days after hospitalization and 19 late recovered patients with an average of 223 days of recovery duration ( Table 2 ). We performed cross-sectional analysis for control, acute and late recovered cohorts as well as early and late recovered cohorts. Acute patients had significant higher levels of GDF-15 compared to controls (p = 0.04) and late recovered patients (p = 0.03) ( Figure 3A ). Early recovered patients also had significantly higher plasma levels of GDF-15 compared to that of late recovered patients (p = 0.02) ( Figure 3A ).

Longitudinal assessment of 7 pairs of acute COVID-19 patients showed no trend in variations in plasma GDF-15 levels after 30 days follow up ( Figure 3B ). Hence, ELISA showed elevation of plasma GDF-15 levels during the acute phase of the infection, which lasted for at least 30 days, and decreased to levels observed in health controls after months of recovery.

Comparison of plasma GDF-15 levels with other inflammation markers (IL-6, CRP) and disease severity marker (neutrophil/lymphocyte ratio) was performed using clinical and proteomics data. GDF-15 levels were positively correlated with age, IL-6 and neutrophil/lymphocyte ratio ( Figure 4A ). However, we found no association between SARS-CoV-2 Spike specific IgG levels assessed by ELISA or CBE and plasma GDF-15 levels ( Supplementary Table 1 ; Figure 4B ). In recovered participants, GDF-15 were also not associated with anti-spike IgG levels ( Figure 4B ). Interestingly, we found a positive correlation between GDF-15 and Eotaxin, another marker associated with aging (25). We found a negative association between plasma GDF-15 and soluble angiotensin converting enzyme 2 (ACE2) levels, the main receptor of SARS-CoV-2.

Assessing levels of the known soluble receptors of GDF-15, we found a significant and strong correlation with soluble CD48 levels, but not erbb2, nor TGF-β receptors 1 and 2 ( Figures 4C, D ). As CD48 is shown to be the receptor and to promote the function of regulatory T-cells (Tregs), we looked for regulatory cytokines. We found a positive correlation between GDF-15 and TGF-β1 levels, with an inverse correlation between GDF-15 and IL-10 in COVID-19 patients ( Figure 4C ).