Distinct Metabolic Profile Associated with a Fatal Outcome in COVID-19 Patients during the Early Epidemic in Italy

ABSTRACT In one year of the coronavirus disease 2019 (COVID-19) pandemic, many studies have described the different metabolic changes occurring in COVID-19 patients, linking these alterations to the disease severity. However, a complete metabolic signature of the most severe cases, especially those with a fatal outcome, is still missing. Our study retrospectively analyzes the metabolome profiles of 75 COVID-19 patients with moderate and severe symptoms admitted to Fondazione IRCCS Ca’ Granda Ospedale Maggiore Policlinico (Lombardy Region, Italy) following SARS-CoV-2 infection between March and April 2020. Italy was the first Western country to experience COVID-19, and the Lombardy Region was the epicenter of the Italian COVID-19 pandemic. This cohort shows a higher mortality rate compared to others; therefore, it represents a unique opportunity to investigate the underlying metabolic profiles of the first COVID-19 patients in Italy and to identify the potential biomarkers related to the disease prognosis and fatal outcome. IMPORTANCE Understanding the metabolic alterations occurring during an infection is a key element for identifying potential indicators of the disease prognosis, which are fundamental for developing efficient diagnostic tools and offering the best therapeutic treatment to the patient. Here, exploiting high-throughput metabolomics data, we identified the first metabolic profile associated with a fatal outcome, not correlated with preexisting clinical conditions or the oxygen demand at the moment of diagnosis. Overall, our results contribute to a better understanding of COVID-19-related metabolic disruption and may represent a useful starting point for the identification of independent prognostic factors to be employed in therapeutic practice.

Despite the fact that this retrospective epidemiological analysis was only conducted for a short time and had a limited sample size, the topic is still relevant and important. in the observation section : Comment 1 The multivariate logistic regression must be detailed, that is, how you proceeded to the selection of these explanatory variables (univariate analysis: a simple logistic regression for each explanatory variable). Comment 2 Because the statistical purpose of this study is to construct an explanatory final model, the initial model must appear next to the final model in table 1, together with the raw OR and final OR, their confidence intervals, and the P value (we propose have another table just for the regression). Comment 3 The method for choosing the final model (stepwise regression method) must be mentioned in the paragraph (forward selection or backward selection, with a parsimony index such as the Akaike criterion (Akaike Information Criteria "AIC"). Comment 4 The "VIF" (Variance Inflation Factor) test for detecting collinearity between the selected variables must be specified in order to verify the lack of collinearity, which is a major problem in regressions. Comment 5 A model fit test should also be mentioned, eg Hosmer -Lemeshow test.
In "Metabolites associated with COVID-19 related in-hospital mortality" section : Comment 6 You mentioned that 35 metabolites were significantly associated with COVID-19 mortality (p<0.05), but you only discussed one of them (cAMP) in your explanation, whereas there are other metabolites reported in your study that are important and deserve to be discussed to make the manuscript more relevant and robust.
Reviewer #2 (Comments for the Author): Please brief up the BMI imputation using MICE and specifically mention the biomarkers considered in that case and also specify the markers considered for each case (e.g line no 135, 142, etc.).

Reviewer #3 (Comments for the Author):
The manuscript by Saccon et al presents data regarding the metabolic profile of 75 samples from COVID-19 patients with mild to severe disease. The data is relevant since it sets the path for a more in-depth investigation regarding the host response against SARS-CoV-2 infection that can predict the severity outcome of COVID-19.
Staff Comments:

Preparing Revision Guidelines
To submit your modified manuscript, log onto the eJP submission site at https://spectrum.msubmit.net/cgi-bin/main.plex. Go to Author Tasks and click the appropriate manuscript title to begin the revision process. The information that you entered when you first submitted the paper will be displayed. Please update the information as necessary. Here are a few examples of required updates that authors must address: • Point-by-point responses to the issues raised by the reviewers in a file named "Response to Reviewers," NOT IN YOUR COVER LETTER. • Upload a compare copy of the manuscript (without figures) as a "Marked-Up Manuscript" file. • Each figure must be uploaded as a separate file, and any multipanel figures must be assembled into one file. Please return the manuscript within 60 days; if you cannot complete the modification within this time period, please contact me. If you do not wish to modify the manuscript and prefer to submit it to another journal, please notify me of your decision immediately so that the manuscript may be formally withdrawn from consideration by Microbiology Spectrum.
If you would like to submit an image for consideration as the Featured Image for an issue, please contact Spectrum staff.
If your manuscript is accepted for publication, you will be contacted separately about payment when the proofs are issued; please follow the instructions in that e-mail. Arrangements for payment must be made before your article is published. For a complete list of Publicat ion Fees, including supplemental material costs, please visit our website.
Corresponding authors may join or renew ASM membership to obtain discounts on publication fees. Need to upgrade your membership level? Please contact Customer Service at Service@asmusa.org.
Thank you for submitting your paper to Microbiology Spectrum.
Despite the fact that this retrospective epidemiological analysis was only conducted for a short time and had a limited sample size, the topic is still relevant and important.
in the observation section :

Comment 1
The multivariate logistic regression must be detailed, that is, how you proceeded to the selection of these explanatory variables (univariate analysis: a simple logistic regression for each explanatory variable).

Comment 2
Because the statistical purpose of this study is to construct an explanatory final model, the initial model must appear next to the final model in table 1, together with the raw OR and final OR, their confidence intervals, and the P value (we propose have another table just for the regression).

Comment 3
The method for choosing the final model (stepwise regression method) must be mentioned in the paragraph (forward selection or backward selection, with a parsimony index such as the Akaike criterion (Akaike Information Criteria "AIC").

Comment 4
The "VIF" (Variance Inflation Factor) test for detecting collinearity between the selected variables must be specified in order to verify the lack of collinearity, which is a major problem in regressions.

Comment 5
A model fit test should also be mentioned, eg Hosmer -Lemeshow test.
In "Metabolites associated with COVID-19 related in-hospital mortality" section :

Comment 6
You mentioned that 35 metabolites were significantly associated with COVID-19 mortality (p<0.05), but you only discussed one of them (cAMP) in your explanation, whereas there are other metabolites reported in your study that are important and deserve to be discussed to make the manuscript more relevant and robust.

Reviewer #1 (Comments for the Author):
Despite the fact that this retrospective epidemiological analysis was only conducted for a short time and had a limited sample size, the topic is still relevant and important.

Response:
We appreciate the reviewer's positive comment regarding the impact of our study.

In the observation section Comment 1
The multivariate logistic regression must be detailed, that is, how you proceeded to the selection of these explanatory variables (univariate analysis: a simple logistic regression for each explanatory variable). Response: We are thankful to the reviewer for the comment. Using one biomarker at a time, we adjusted the analysis for a set of confounders that were identified based upon prior reports in the literature. No variable selection was done for confounders. We have modified the text to include an additional sentence that provides greater clarity (line 84): "One biomarker was included in each model, but all other covariates were pre-specified for inclusion based on previous literature suggesting they are potential confounders." Due to limited word count for observation, we mentioned the confounders in the figure legends As follows "All biomarkers that were significant at the 0.05 level, after adjustment of age, gender, and BMI, were included in the plot and order by the value of the odds ratio."

Comment 2 Because the statistical purpose of this study is to construct an explanatory final model, the initial model must appear next to the final model in table 1, together with the raw OR and final OR, their confidence intervals, and the P value (we propose have another table just for the regression).
Response: we are thankful to the reviewer for the point. As stated above, there are no earlier models, as model selection was not used. The same pre-specified covariates were used in each model with a single biomarker. The only earlier models were those used to determine which biomarkers to include in the imputation. Please see response to reviewer #2 below for more detail.
Within the text, we stated the goal of this work was "to identify a set of biomarkers strongly associated with the patient outcome". While an explanatory model is a potential outcome, we believe that identifying COVID-associated biomarkers is an important stand-alone outcome.

Comment 3
The method for choosing the final model (stepwise regression method) must be mentioned in the paragraph (forward selection or backward selection, with a parsimony index such as the Akaike criterion (Akaike Information Criteria "AIC"). Response: As stated above there was no selection done for the "final model". We selected covariates based on previous literature indicating what may be confounding factors. We included a single biomarker in each regression with these covariates, in each model the same covariates were used.

Comment 4
The "VIF" (Variance Inflation Factor) test for detecting collinearity between the selected variables must be specified in order to verify the lack of collinearity, which is a major problem in regressions.

Response:
We did not do variable selection, rather, we pre-specified the model based on previous literature and only include one biomarker with those covariates. Therefore, although it is possible that there is some collinearity in the pre-specified variables, we are not able to change it based on a test of collinearity without increasing the type one error rate. Pre-specification, in this case, is more important than any possible minor collinearity. As we are not doing any form of model selection and each model contains the same three variables (age, BMI, and gender) in addition to one biomarker, this allows for a fair comparison of the biomarkers' association with the outcome. This provides an appropriate context to compare the relative importance of the individual biomarkers, regardless of any collinearity. Finally, there were no indications of non-convergence of the models, i.e. extremely large or extremely small SE, that would indicate near prefect collinearity. In the case of perfect collinearity, R automatically removes one of the variables.

Comment 5
A model fit test should also be mentioned, eg Hosmer -Lemeshow test.
Response: As we are not doing prediction based on the logistic model, the overall goodness of fit is not directly relevant to these analyses. We have a prespecified set of confounders (or potential confounders) that we include in each model. We are using these models to estimate associations with each of the biomarkers after adjustment for these known and pre-specified set of confounders, not for prediction of the outcome using the full model.
In "Metabolites associated with COVID-19 related in-hospital mortality" section: Comment 6 You mentioned that 35 metabolites were significantly associated with COVID-19 mortality (p<0.05), but you only discussed one of them (cAMP) in your explanation, whereas there are other metabolites reported in your study that are important and deserve to be discussed to make the manuscript more relevant and robust. Response: We agree that the identified metabolites deserve a longer discussion for their know involvement in different cell functions and/or pathologies. Unfortunately, due to the required word limit (1200), we chose to focus on cAMP due to its possible involvement in SARS-CoV-2 infection and four others (3-hydroxysebacate, 5-dodecenate (12:1n7), tetradecadienoate (14:2)* and myristoleate (14:1n5)) for their strong association with the patients' outcome.

Reviewer #2 (Comments for the Author):
Please brief up the BMI imputation using MICE and specifically mention the biomarkers considered in that case and also specify the markers considered for each case (e.g line no 135, 142, etc.).

Response:
We are thankful to the reviewer for the comment. We have added the following text to the paper (line no 86-91): "As BMI was missing for many patients, multiple imputation by chained equations (MICE) was used (via the MICE R-package) to impute BMI using the predictive mean matching method. Imputation was done in two stages, first including no biomarkers, and then including those biomarkers that were found to be significant under the original imputation (see table for list of biomarkers). Fifty imputed datasets were created, and the resulting pooled estimates and inference were combined using Rubin's rules. P-values were not adjusted for multiple comparisons in this analysis."

Reviewer #4
The manuscript by Saccon et al presents data regarding the metabolic profile of 75 samples from COVID-19 patients with mild to severe disease. The data is relevant since it sets the path for a more in-depth investigation regarding the host response against SARS-CoV-2 infection that can predict the severity outcome of