A distinct metabolic profile associated with a fatal outcome in COVID-19 patients during early epidemic in Italy

Leveraging the unique biological resource based upon the initial COVID-19 patients in Policlinico di Milano (Italy), our study provides the first metabolic profile associated with a fatal outcome. The identification of potential predictive biomarkers offers a vital opportunity to employ metabolomics in a clinical setting as diagnostic tool of disease prognosis upon hospital admission.


Introduction 48
As of February 2021, more than 100 million cases of coronavirus disease 2019 (COVID-19), caused 49 by the novel severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection, had been 50 confirmed globally, with more than 2 million related deaths (https://covid19.who.int/). Although most 51 individuals remain asymptomatic or display mild symptoms, 15-20% of patients exhibit severe 52 symptoms, specifically respiratory distress, often requiring mechanical ventilation or/and intensive 53 care (ICU) admission [1], with a mortality rate after ICU admission estimated around 40% [2]. 54 Multiple studies have identified profound underlying conditions that demonstrate increased 55 susceptibility to a more severe prognosis and a higher risk of fatality, including the male gender, old 56 age [3], or certain underlying medical conditions, such as hypertension, cardiovascular diseases, 57 diabetes or obesity [4]. Additionally, patients infected with SARS-CoV-2 present metabolic 58 dysregulation, possibly due to immune-triggered inflammation or other changes in the host 59 physiology, and that these alterations often reflect the disease severity [1,5,6]. For instance, levels of 60 particular amino acids positively correlated with severe COVID-19 cases [1,7]. Moreover, 61 perturbations in energy metabolisms, TCA and urea cycle [6] and lipid metabolism [1,8] are correlated 62 to disease prognosis. Thus, it is essential to assemble a complete metabolic signature correlated to 63 disease severity to identify a set of biomarkers strongly associated with the patient outcome, with the 64 final goal of employing them for diagnostics and therapeutic purposes. 65 Our study retrospectively analyzes the metabolome profile of 75 COVID-19 patients with moderate 66 and severe symptoms admitted to Policlinico di Milano (Lombardy region, Italy) following SARS-67 CoV-2 infection between March and April 2020. Italy was the first Western country to experience 68 COVID-19 disease, and the Lombardy region was the epicenter of the Italian COVID-19 pandemic. 69 . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) Logistic regression was used to model associations of each biomarker with COVID-19-related in-84 hospital mortality adjusting for age, gender, and body mass index (BMI). As BMI was missing for 85 many patients, multivariate imputation by chained equations (MICE) was used (via the MICE R-86 package) to impute BMI. Imputation was done in two stages, first including no biomarkers, and then 87 including those biomarkers that were found to be significant under the original imputation. The 88 resulting pooled estimates and inference were obtained using Rubin's rules. P-values were not adjusted 89 in this analysis. 90 Metabolomics data were log2 transformed and plotted using histograms with normal distribution 91 superimposed. R package LIMMA was applied for differential abundance analysis between different 92 mask types (Nasal cannula/VentMask/CPAP), outcome (survivors/non-survivors), and severity 93 (moderate/severe). Adjustment for multiple testing was assessed using false discovery rate (FDR) 94 <0.05. Heatmap was built using the R package ComplexHeatmap. Uniform Manifold Approximation 95 and Projection (UMAP) representations were done using the R package UMAP. Metabolites with 96 variance equal to zero were removed and positive significant pairwise correlations after Bonferonni 97 . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 15, 2021. In a logistic regression analysis (adjusted to age, gender, and BMI), 35 metabolites, among the >1000 119 tested, were significantly associated with COVID-19 mortality (p<0.05, unadjusted), among which 10 120 biomarkers were significant at the unadjusted 0.025 level ( Figure 1A). Interestingly, cyclic adenosine 121 monophosphate (cAMP) is significantly increased in non-survivors compared to survivors (OR: 7.4 122 95% CI 1.5 -37). cAMP is a well-known intracellular messenger that functions as a regulator of 123 various cellular activities, including cell growth and differentiation, gene transcription, protein 124 expression, and is intimately involved in mitochondrial dynamics [11]. As cAMP plays a role in 125 . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

SARS-CoV-2 endocytosis in the initial phases of the infection [12], its involvement in disease 126
progression is worthy of further investigations as a potential biomarker. 127 Using UMAP, we observed that the distribution of the patients in the biomarker enrichment showed a 128 separation between survivors and non-survivors ( Figure 1B). No specific pattern or clustering was 129 observed for comorbidities (Fig 1C), the number of comorbidities (Fig 1D), or diabetes (Fig 1E), 130 indicating that these metabolite sets only differentiate related to the clinical outcome. 131 To further understand the patterns of metabolic changes related to COVID-19, we performed a 132 weighted correlation network analysis on the metabolomics dataset, using significant positive 133 correlations (Spearman, adjusted p<0.00001). We identified 10 metabolites communities highly 134 connected in a network of 916 nodes (metabolites) and 11453 edges ( Figure 1F). Six predicted 135 biomarkers (out of the 10 previously identified with p<0.025, unadjusted) were highly correlated and 136 were present in the network and belong to the lipid pathways. These metabolites are also known to be 137 associated with peroxisomal fatty acid oxidation disorders (3-hydroxysebacate) [13] or insuline 138 resistance (5-dodecenate (12:1n7), tetradecadienoate (14:2)* and myristoleate (14:1n5)) [14]. 139 Finally, to further discriminate the subset of metabolites significantly associated with a fatal clinical 140 outcome, we selected all the biomarkers and their first neighbors in the network analysis previously 141 described (238 metabolites). Based on these data, we found a clear clustering of non-survivors in 142 opposition to survivors (Fig 1G). No clustering according to mask type was observed, indicating that 143 the metabolic signature associated with mortality appears to be independent of the oxygen demand at 144 the moment of hospitalization, providing the first identified correlation between a metabolite profile 145 and disease severity in COVID-19 patients. 146 It is important to consider that when this study was planned (March-April 2020), there was little 147 knowledge about the COVID-19. Therefore, several clinical data were missing. Despite that, this is the 148 first set of biomarkers identified from high throughput metabolomics data that are associated with 149 mortality and are not confounded by other preexisting conditions. 150 151

Conclusions 152
. CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted April 15, 2021. ; https://doi.org/10.1101/2021.04.13.21255117 doi: medRxiv preprint Our analysis has identified metabolic biomarkers that in our data differentiate between COVID-19 153 survivors and non-survivors and that may be predictive of death from COVID-19, from the early stage 154 of the epidemic, independently from oxygen demand at the moment of diagnosis. Our results on high 155 throughput metabolomics contribute to a better understanding of COVID-19-related metabolic 156 disruption and may represent a useful starting point for the identification of independent prognostic 157 factors to be employed in the therapeutic practice. The authors would also like to thank all patients involved in this study, as well as the dedicated 177 medical and research staff who fight against SARS-CoV-2. We also like to acknowledge the work of 178 the "COVID-19 NETWORK" working group: 179 Responsible person: Prof. Silvano Bosari 180 . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 15, 2021. ; https://doi.org/10.1101/2021.04.13.21255117 doi: medRxiv preprint . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 15, 2021. ; https://doi.org/10.1101/2021.04.13.21255117 doi: medRxiv preprint