1. Immunology and Inflammation
  2. Medicine
Download icon

Cytokine ranking via mutual information algorithm correlates cytokine profiles with presenting disease severity in patients infected with SARS-CoV-2

  1. Kelsey E Huntington
  2. Anna D Louie
  3. Chun Geun Lee
  4. Jack A Elias
  5. Eric A Ross
  6. Wafik S El-Deiry  Is a corresponding author
  1. Brown Experimentalists Against COVID-19 (BEACON) Group, Brown University, United States
  2. Laboratory of Translational Oncology and Experimental Cancer Therapeutics, Warren Alpert Medical School, Brown University, United States
  3. The Joint Program in Cancer Biology, Brown University and Lifespan Health System, United States
  4. Cancer Center at Brown University, Warren Alpert Medical School, Brown University, United States
  5. Department of Pathology and Laboratory Medicine, Warren Alpert Medical School, Brown University, United States
  6. Pathobiology Graduate Program, Warren Alpert Medical School, Brown University, United States
  7. Department of Molecular Microbiology and Immunology, Brown University, United States
  8. Department of Surgery, Lifespan Health System and Warren Alpert Medical School, Brown University, United States
  9. Biostatistics and Bioinformatics Facility, Fox Chase Cancer Center, Temple University Health System, United States
  10. Hematology-Oncology Division, Department of Medicine, Lifespan Health System and Warren Alpert Medical School, Brown University, United States
Short Report
  • Cited 2
  • Views 1,332
  • Annotations
Cite this article as: eLife 2021;10:e64958 doi: 10.7554/eLife.64958

Abstract

Although the range of immune responses to severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is variable, cytokine storm is observed in a subset of symptomatic individuals. To further understand the disease pathogenesis and, consequently, to develop an additional tool for clinicians to evaluate patients for presumptive intervention, we sought to compare plasma cytokine levels between a range of donor and patient samples grouped by a COVID-19 Severity Score (CSS) based on the need for hospitalization and oxygen requirement. Here we utilize a mutual information algorithm that classifies the information gain for CSS prediction provided by cytokine expression levels and clinical variables. Using this methodology, we found that a small number of clinical and cytokine expression variables are predictive of presenting COVID-19 disease severity, raising questions about the mechanism by which COVID-19 creates severe illness. The variables that were the most predictive of CSS included clinical variables such as age and abnormal chest x-ray as well as cytokines such as macrophage colony-stimulating factor, interferon-inducible protein 10, and interleukin-1 receptor antagonist. Our results suggest that SARS-CoV-2 infection causes a plethora of changes in cytokine profiles and that particularly in severely ill patients, these changes are consistent with the presence of macrophage activation syndrome and could furthermore be used as a biomarker to predict disease severity.

Introduction

In December 2019, severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), the origin of coronavirus disease 2019 (COVID-19), emerged in Wuhan, China (Zhu et al., 2020). Although many COVID-19 patients remain asymptomatic, there exists a subset of patients who present with severe illness. Early treatment with dexamethasone appears to improve outcomes in these patients. However, it is not always initially clear which patients would benefit from this therapy (The RECOVERY Collaborative Group, 2020). Moreover, COVID-19 infection can be accompanied by a severe inflammatory response characterized by the release of pro-inflammatory cytokines, an event known as cytokine storm (CS) (Tang et al., 2020; Ragab et al., 2020). Thus far, this COVID-19-associated CS has predominantly been characterized by the presence of IL-1β, IL-2, IL-17, IL-8, TNF, CCL2, and most notably IL-6 (Tang et al., 2020; Merad and Martin, 2020; McGonagle et al., 2020; Wan et al., 2020; Otsuka and Seino, 2020). Severe cases of CS can be life threatening, and early diagnosis as well as treatment of this condition can lead to improved outcome. We hypothesize that cytokine profiles combined with clinical information can predict disease severity, potentially giving clinicians an additional tool when evaluating patients for preemptive intervention.

Results

Analysis was performed for 36 PCR-confirmed COVID-19 (+) and 36 (−) human plasma samples (Figure 1—source data 1). The COVID-19 Severity Score (CSS) was developed to categorize patients based on their status upon presentation to the emergency department. CSS is graded as follows: 0 = COVID (−), no symptoms, healthy control (n = 24); 1 = COVID (−), symptoms (n = 12); 2 = COVID (+), discharged from emergency room (n = 15); 3 = COVID (+), admitted, but who did not require supplemental oxygen (n = 7); 4 = COVID (+), admitted and required any amount of supplemental oxygen or positive pressure ventilation (n = 8); and 5 = COVID (+), admitted to ICU/step-down (n = 6) (Figure 1). CSS was used as the outcome variable for a mutual information minimum-redundancy maximum-relevance algorithm (Kratzer and Furrer, 2018; Figure 1), with the goal of selecting a subset of variables most predictive of CSS. The algorithm confirmed the predictive value of clinical variables such as age and chest x-ray abnormality and also ranked the information gain provided by each of 15 cytokines tested. Several cytokines were able to add unique predictive value to the mutual information model in addition to what was provided by clinical factors such as age or patient comorbidities. This algorithm also deprioritized factors when their predictive value was redundant with the most predictive variables. Macrophage colony-stimulating factor (M-CSF) was ranked second after age as it was the factor that added the most predictive power to the algorithm with minimal redundancy with age. It ranked ahead of abnormalities on chest x-ray because while both were relevant in predicting COVID severity, part of the predictiveness of chest x-ray abnormality was also explained by age differences (Figure 2). The top four cytokines combined with age were predictive of the most severe CSS (4–5) and had a receiver operating characteristic (Figure 2), with an area under the curve of 0.86. Multiple cytokines, including M-CSF (p<0.01), interferon-inducible protein 10 (IP-10) (p<0.01), interleukin 18 (IL-18) (p<0.01), and interleukin-1 receptor antagonist (IL-1RA) (p<0.01), were more relevant in predicting CSS than more frequently characterized cytokines in the context of COVID-19 such as IL-6 (p<0.01). These cytokines showed a statistically significant difference in their profiles when segregated by CSS (Figure 3), yet the mutual information algorithm prioritized them differently than would be expected based on univariate analyses. This indicates that the mutual information algorithm is prioritizing cytokines whose predictive value for COVID-19 severity cannot be fully explained by other clinical variables such as age or medical comorbidities.

Mutual information COVID Severity Score (CSS) relevancy matrix.

(A) Comprehensive matrix of relevancy to CSS of all variables assessed by mutual information algorithm, relevancy scores computed for not-yet selected variables are shown in each column, and variables are ordered to place maximum local scores on the diagonal, yielding a list in decreasing order from the upper left of variable relevancy. Warmer colors indicate higher relevancy, while cooler colors indicate higher redundancy. (B) COVID Severity Score Table with breakdown of categories as well as sample size per category.

Age, macrophage colony-stimulating factor (M-CSF), and chest x-ray are the most predictive variables for COVID Severity Score (CSS).

(A) The y-axis is CSS, and the x-axis is age in years with points colored by chest x-ray status. (B) The y-axis is M-CSF concentration in pg/mL, and the x-axis is age in years, with points colored based on CSS. See individual legends below graphs. (C) Receiver operating characteristic (ROC) curve predicting CSS 4–5 using age, M-CSF, IP-10, IL-18, and IL-1RA, and ROC curve predicting CSS 4–5 using only age.

Violin plot representations of cytokine expression levels ordered by COVID Severity Score (CSS).

Cytokines ordered by row from upper left corner based on mutual information relevancy matrix (upper left being most relevant and lower right being least relevant). The x-axis is CSS, and the y-axis is analyte concentration in pg/mL. One-way ANOVA F values and p values are listed on each plot. Source code 1. R code with sections to apply the varrank package to the source data, create a receiver operating characteristic curve, and calculate analysis of variance for each cytokine.

Figure 3—source data 1

Raw data.

Source Data Table 1. Mutual algorithm criteria table. *Documented obesity or overweight for height < 99th percentile. Source Data Table 2. Patient information.

https://cdn.elifesciences.org/articles/64958/elife-64958-fig3-data1-v2.zip

Discussion

We found that a small number of clinical variables when combined with cytokine expression are predictive of presenting COVID-19 disease severity. Cytokines singled out for relevance by the mutual information algorithm shared a connection to macrophage activation syndrome (MAS), raising questions about the mechanism by which SARS-CoV-2 creates severe illness in a subset of patients. First, we examined the significant contribution of IP-10 to CSS. IP-10 is secreted by monocytes, fibroblasts, and endothelial cells in response to interferon gamma (IFN-γ), which is secreted by T cells (mainly, Th1), macrophages, mucosal epithelial cells, and natural killer (NK) cells (Liu et al., 2011). This release of IFN-γ induces several cell types to produce IP-10, which consequently recruits more Th1 cells, contributing to a positive feedback loop. IP-10 is also chemoattractant to CXCR3-postitive cells such as macrophages, dendritic cells, NK cells, and T cells. It has been proposed that macrophages recruited by IP-10, in the presence of persistent IFN-γ production, can lead to MAS (Merad and Martin, 2020; McGonagle et al., 2020; Otsuka and Seino, 2020). MAS is characterized as a state of systemic hyperinflammation often accompanied by CS, which, without intervention, can lead to severe tissue damage and, in extreme cases, death (Otsuka and Seino, 2020).

Moreover, the cytokine most relevant in predicting CSS was M-CSF, which is secreted by eukaryotic cells in response to viral infection and stimulates hematopoietic stem cells to differentiate into macrophages. Currently, there are three separate immune stages that describe the progression of COVID-19. The first stage is characterized by a potent induction of interferons that marks the early activation of the immune system that is important in the viral response, and the second stage is characterized by a delayed interferon response (Merad and Martin, 2020). These stages may prime the body for a third stage comprised of detrimental hyperinflammation characterized by CS and MAS (Merad and Martin, 2020). This excessive macrophage activation could explain the increase in IL1-RA that we observed, a cytokine abundantly produced by macrophages.

Steroids have shown a survival benefit for COVID-19, likely by suppressing such detrimental hyperinflammation (The RECOVERY Collaborative Group, 2020). Our analysis identified a pattern of cytokine alterations on presentation associated with COVID-19 severity. The ability to identify a cytokine pattern less redundant with known clinical factors such as age and chest x-ray could help better identify patients in need of immunomodulatory treatment without the confounders of current models where the measured cytokines correlate as much with age as with severity (Pierce et al., 2020). Further studies should be conducted to clarify the mechanistic role that these cytokines and macrophages play in the various stages of COVID-19 and correlate them with other hematologic parameters that were not collected in this database. The results of these future studies could identify more targeted immunomodulatory strategies beyond steroid administration such as treatment with MEK inhibitors (Zhou et al., 2020), as well as the ideal timing of these interventions to maximize therapeutic efficacy. Future studies could also address the size limitations of this study, which was not powered to explore race- or ethnicity-related differences in COVID-19 severity. Finally, we present the application of this mutual information algorithm as a way to evaluate the dataset as a whole and elucidate the most important cytokines in predicting the presenting severity of COVID-19. COVID-19 severity is influenced by many clinical factors, such as age, and this algorithm is able to identify cytokines that contribute information not present in the tested clinical variables. Identifying the most important variables for severe presentation of COVID-19 within a more complete cytokine profile may help determine global immune mechanisms of disease severity.

Materials and methods

Key resources table
Reagent type
(species) or resource
DesignationSource or referenceIdentifiersAdditional
information
Biological samples (Human)Human plasmaLifespan Brown COVID-19 BiobankSee data table48 unique patients
Biological samples (Human)Human plasmaLee BioSolutions991–58-PS24 unique patients
Commercial assay or kitMILLIPLEX MAP Human Immunology Multiplex AssayMillipore Sigma15-Plex # HCYTA-60K
Software, algorithmRR Project for Statistical ComputingRRID:SCR_001905
OtherLuminex 100/200 System assay platformThermo Fisher Scientifichttps://www.luminexcorp.com/luminex-100200

Biobank samples

Request a detailed protocol

COVID-19 (+) and (−) human plasma samples were received from the Lifespan Brown COVID-19 Biobank from Brown University at Rhode Island Hospital (Providence, RI). All biobank samples were collected on patients’ arrival in the Emergency Department at Rhode Island Hospital. All patient samples were deidentified but included the available clinical information as described in Results. It is unknown if any patients were blood relatives. The IRB study protocol ‘Pilot Study Evaluating Cytokine Profiles in COVID-19 Patient Samples’ did not meet the definition of human subjects research by either the Brown University or the Rhode Island Hospital IRBs. All samples were thawed and centrifuged at 14,000 rpm for 10 min following the manufacturer protocol included with the Luminex kit to remove cellular debris immediately before the assay was run.

Donor samples

Request a detailed protocol

Normal, healthy, COVID-19 (−) samples were commercially available from Lee BioSolutions (991–58-PS-1, Lee BioSolutions, Maryland Heights, MO). All samples were thawed and centrifuged at 14,000 rpm for 10 min following the manufacturer protocol included with the Luminex kit to remove cellular debris immediately before the assay was run.

Cytokine and chemokine measurements

Request a detailed protocol

A MilliPlex MILLIPLEX MAP Human Cytokine/Chemokine/Growth Factor Panel A – Immunology Multiplex Assay (HCYTA-60K-13, Millipore Sigma, Burlington, MA) was run on a Luminex 200 Instrument (LX200-XPON-RUO, Luminex Corporation, Austin, TX) according to the manufacturer’s instructions. Plasma levels of granulocyte colony-stimulating factor (G-CSF), IFN-γ, interleukin one alpha (IL-1α), interleukin-1 receptor antagonist (IL-1RA), IL-2, IL-6, IL-7, IL-12, IP-10, monocyte chemoattractant protein-1 (MCP-1), M-CSF, macrophage inflammatory protein-1 alpha (MIP-1α), and tumor necrosis factor alpha (TNF-α) were measured. Data pre-processing: values below limit of detection were re-coded as half the limit of detection. A single extreme outlier value in IFN-y levels was removed after confirming outlier status via Hampel and Grubbs outlier testing (both p<0.01).

Clinical variables

Request a detailed protocol

Available deidentified clinical variables were collected from patients and from chart review during their time in the emergency department. Clinical variables were categorized to create combined variables such as the number of chronic conditions or the number of presenting symptoms. The full breakdown of clinical variable categorization can be found in Figure 2—source data 1.

Data analysis

Request a detailed protocol

Data analysis and visualization were generated using R (R Development Core Team, 2020). The varrank package (Kratzer and Furrer, 2020) was used to apply a minimum-redundancy maximum-relevance mutual information algorithm. The algorithm classifies the amount of information each cytokine and clinical variable can provide about the outcome variable, CSS. Each cytokine variable was discretized into two clusters – either high or low analyte concentration in pg/mL – using k-means clustering to minimize within-variable entropy and, thus, over-fitting. This algorithm partitions each data point into the cluster (high or low analyte concentration) with the nearest mean. Clinical variables and cytokine levels were used to predict CSS. The first variable was selected for local optimum relevance by a greedy algorithm. All subsequent variables were ordered to maximize relevancy and minimize redundancy. The ordering was robust to leave-one-out cross-validation. For each cytokine, one-way ANOVA with Tukey’s honest significant difference test and Šidák correction for multiple comparisons was used to compare plasma cytokine levels among CSS groups.

Data availability

Source data and source code files have been provided.

References

  1. Software
    1. Kratzer G
    2. Furrer R
    (2020) Varrank, version 0.3
    Heuristics Tools Based on Mutual Information for Variable Ranking.
  2. Software
    1. R Development Core Team
    (2020) R: A Language and Environment for Statistical Computing
    R Foundation for Statistical Computing, Vienna, Austria.

Decision letter

  1. Jameel Iqbal
    Reviewing Editor; James J. Peters Veterans Affairs Medical Center, United States
  2. Mone Zaidi
    Senior Editor; Icahn School of Medicine at Mount Sinai, United States

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

It's now known that SARS-CoV-2 infection can cause numerous changes in cytokines, and it has been hypothesized that in critically ill patients these changes are correlated with the severity of disease. This paper is of interest to the overall COVID-19 community in that presents an analysis of cytokines that identifies Macrophage Activation Syndrome as a correlate with disease severity.

Decision letter after peer review:

Thank you for submitting your article "Cytokine ranking via mutual information algorithm correlates cytokine profiles with disease severity in COVID-19" for consideration by eLife. Your article has been reviewed by two peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by Mone Zaidi as the Senior Editor. The reviewers have opted to remain anonymous.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

We would like to draw your attention to changes in our revision policy that we have made in response to COVID-19 (https://elifesciences.org/articles/57162). Specifically, we are asking editors to accept without delay manuscripts, like yours, that they judge can stand as eLife papers without additional data, even if they feel that they would make the manuscript stronger. Thus the revisions requested below only address clarity and presentation.

Summary:

The premise behind this manuscript is important and timely both for scientists focused on better understanding the variable immune response against the severe acute respiratory syndrome-coronavirus-2 (SARS-CoV-2) in different individuals and for physicians intent on improving the outcomes for their SARS-CoV-2 infected patients. The paper correlates the COVID-19 Severity Score (CSS) with a cytokine array along with demographic factors and a chest x-ray to see which components are predictive of severe COVID-19. The analysis is well presented and the experiments are straight forward in nature. Overall, the data are collected, controlled, and analyzed properly, and support the conclusions of the manuscript within the current context. If replicated in follow-on studies of independent patients, the findings may ultimately help to inform the development of improved diagnostics and therapeutics.

Essential revisions:

Overall the comments from the reviewers were positive, and suggest that the manuscript can be further strengthened by the revisions that are mostly explanatory in nature as indicated below.

1) Please include data on common lab parameters if available. For example, if there is data on CRP or fibrinogen, that would be interesting to add. Likewise, since the absolute monocyte count and the relative percentage of monocytes in the total white blood cell count are standard clinical assays obtained routinely in all SARS-CoV-2 infected patients presenting to hospital EDs for evaluating the possibility of progression in the less severely affected subset of subjects as well as in the subset of more severely affected subjects at presentation, can you discuss what you found when these two immunologically-relevant cellular parameters were included in the analysis you performed. If you did not analyze them, can you explain why not? Either way, a sentence or two should be included at this point in the Discussion section that these routinely measured cellular hematologic parameters would be a logical biomarker to explore in light of your main findings and conclusions, which include M-CSF as one of the four cytokines you identified that increase predictiveness.

2) Please state when the patient-derived blood samples were obtained. This is critical information and should be described here. Based on your use of "presenting" in the title and other places in the text, we are quite sure that the patient samples were collected on their arrival in the Emergency Dept of the Rhode Island Hospital, which would also have to be the case if the samples from all of the subjects (other than the normal subjects whose samples were purchased) were supposed to have been drawn at the same clinically-relevant point-in-time, i.e. as some of SARS-CoV-2 infected patients were asymptomatic and not admitted to the hospital, etc. Please also mention whether all of the patients were unrelated or if any were related, as this would then require accounting for the non-independence of the data generated from them, although this is mainly important for genetic association studies.

3) Please describe the centrifugation parameters employed including the g-force and length of time of the spin and mention what was the exact sample that was obtained. Was it, for example, citrated plasma samples used for coagulation studies? If so, why were the tubes needing to be spun again as whole blood samples centrifuged to remove aliquots of citrated plasma are spun at high speed two times to ensure platelet poor plasma which would also ensure cellular debris would have been removed.

4) The text of the manuscript, both in the Materials and methods section and elsewhere, creates the impression you performed a multiplex assay to measure the levels of the 15 cytokines listed in each study subject's plasma sample. But here, in the Materials and methods section, you mention (for the first time) that it is each subject's "production of" cytokines in the culture supernatant" which was used for the multiplex measurement. Please provide details as to: (i) what blood cell type or types (e.g., PBMCs) were used; (ii) how you isolated the blood cell(s); (iii) how you cultured them, i.e. in which medium and under what atmosphere they were cultured, and for how long, etc.; and (iv) how the conditioned medium was collected, processed and assayed.

5) Please explain briefly if and how in these statistical analyses you accounted for the issue of multiple testing (i.e., for the multiple cytokine measures).

6) It is important to include information on the well-known race- and ethnicity-based disparities in case fatality rates (CFRs) in African Americans and Mexican Americans. Please also include it as one of your "clinical variables" in your analysis. If you did incorporate it in your analysis, please add text to discuss it. If you did not, please discuss why not?

7) Figure 3 is cited in the text of the manuscript proper before Figure 2 is cited. Switch the numbering please.

8) In the legend for Figure 3B the labels for the y-axis ("years") and x-axis ("M-CSF concentration in pg/mL") are reversed to what is in the actual figure. Please correct this discrepancy.

9) It would be helpful to also show in Figure 3C the ROC curve(s) for predicting CSS 4-5 using age alone, and possibly, age together with only M-CSF levels.

10) Please correct the stylistic points and/or typos which we have highlighted yellow and/or struck through in the marked up word version of your manuscript.

https://doi.org/10.7554/eLife.64958.sa1

Author response

Essential revisions:

Overall the comments from the reviewers were positive, and suggest that the manuscript can be further strengthened by the minor revisions that are mostly explanatory in nature as indicated below.

1) Please include data on common lab parameters if available. For example, if there is data on CRP or fibrinogen, that would be interesting to add. Likewise, since the absolute monocyte count and the relative percentage of monocytes in the total white blood cell count are standard clinical assays obtained routinely in all SARS-CoV-2 infected patients presenting to hospital EDs for evaluating the possibility of progression in the less severely affected subset of subjects as well as in the subset of more severely affected subjects at presentation, can you discuss what you found when these two immunologically-relevant cellular parameters were included in the analysis you performed. If you did not analyze them, can you explain why not? Either way, a sentence or two should be included at this point in the Discussion section that these routinely measured cellular hematologic parameters would be a logical biomarker to explore in light of your main findings and conclusions, which include M-CSF as one of the four cytokines you identified that increase predictiveness.

One of the biggest limitations of our data set is that the deidentified data of this biobank lacks certain parameters, such as common lab parameters like complete blood count, CRP or fibrinogen. While privacy issues limit our ability to obtain that data for this study, we have updated our Discussion to include this avenue for future research.

2) Please state when the patient-derived blood samples were obtained. This is critical information and should be described here. Based on your use of "presenting" in the title and other places in the text, we are quite sure that the patient samples were collected on their arrival in the Emergency Dept of the Rhode Island Hospital, which would also have to be the case if the samples from all of the subjects (other than the normal subjects whose samples were purchased) were supposed to have been drawn at the same clinically-relevant point-in-time, i.e. as some of SARS-CoV-2 infected patients were asymptomatic and not admitted to the hospital, etc. Please also mention whether all of the patients were unrelated or if any were related, as this would then require accounting for the non-independence of the data generated from them, although this is mainly important for genetic association studies.

In the Materials and methods section we have clarified when the patient-derived blood samples were obtained. As intuited by the reviewers, all samples (other than purchased normal samples) were indeed obtained on the patient’s arrival to the emergency department of Rhode Island Hospital. Unfortunately, due to the limited data present in the deidentified data set, we are unable to determine if any of the patients were related.

3) Please describe the centrifugation parameters employed including the g-force and length of time of the spin and mention what was the exact sample that was obtained. Was it, for example, citrated plasma samples used for coagulation studies? If so, why were the tubes needing to be spun again as whole blood samples centrifuged to remove aliquots of citrated plasma are spun at high speed two times to ensure platelet poor plasma which would also ensure cellular debris would have been removed.

The centrifugation parameters including g-force (14,000 rpm) and length of spin (10 minutes) have been added to the Materials and methods section. Although samples were platelet-poor plasma, and thus should already have been free of debris, the Luminex operating instructions suggest that all samples, regardless of their initial processing, be spun again immediately before being run to ensure they are free of debris and prevent potential clogs of the instrument.

4) The text of the manuscript, both in the Materials and methods section and elsewhere, creates the impression you performed a multiplex assay to measure the levels of the 15 cytokines listed in each study subject's plasma sample. But here, in the Materials and methods section, you mention (for the first time) that it is each subject's "production of" cytokines in the culture supernatant" which was used for the multiplex measurement. Please provide details as to: (i) what blood cell type or types (e.g., PBMCs) were used; (ii) how you isolated the blood cell(s); (iii) how you cultured them, i.e. in which medium and under what atmosphere they were cultured, and for how long, etc.; and (iv) how the conditioned medium was collected, processed and assayed.

Thank you for pointing out this inaccuracy in the Materials and methods section. The multiplex assay measured the levels of the 15 cytokines on each study subjects’ plasma sample. The Materials and methods have been updated to accurately reflect the experiment performed.

5) Please explain briefly if and how in these statistical analyses you accounted for the issue of multiple testing (i.e., for the multiple cytokine measures).

A Sidak correction was applied to correct for multiple testing for the multiple cytokine measures. The text and figures have been updated to reflect this.

6) It is important to include information on the well-known race- and ethnicity-based disparities in case fatality rates (CFRs) in African Americans and Mexican Americans. Please also include it as one of your "clinical variables" in your analysis. If you did incorporate it in your analysis, please add text to discuss it. If you did not, please discuss why not?

Race was included as a clinical variable in the mutual information algorithm. Unfortunately, the small sample size of our data set does not give it the power to identify race-based differences in COVID-19 severity. Indeed, the mutual information algorithm found the information provided by race was redundant after considering other factors such as age, symptoms, insurance status, number of abnormal vital signs, and chest x-ray abnormalities. We have updated our Discussion section to include discussion of race and COVID-19. In Author response image 1 is a race/ethnicity scatter plot if it is of interest to the reviewers.

Author response image 1

7) Figure 3 is cited in the text of the manuscript proper before Figure 2 is cited. Switch the numbering please.

The numbering of these two figures has been corrected.

8) In the legend for Figure 3B the labels for the y-axis ("years") and x-axis ("M-CSF concentration in pg/mL") are reversed to what is in the actual figure. Please correct this discrepancy.

This discrepancy has been corrected.

9) It would be helpful to also show in Figure 3C the ROC curve(s) for predicting CSS 4-5 using age alone, and possibly, age together with only M-CSF levels.

The ROC curve for predicting CSS 4-5 found in Figure 2 has been updated to include a comparison curve predicting CSS 4-5 using age alone. This age ROC curve has an AUC of 0.74.

10) Please correct the stylistic points and/or typos which we have highlighted yellow and/or struck through in the marked up word version of your manuscript.

Stylistic points and typos have been addressed.

https://doi.org/10.7554/eLife.64958.sa2

Article and author information

Author details

  1. Kelsey E Huntington

    1. Brown Experimentalists Against COVID-19 (BEACON) Group, Brown University, Providence, United States
    2. Laboratory of Translational Oncology and Experimental Cancer Therapeutics, Warren Alpert Medical School, Brown University, Providence, United States
    3. The Joint Program in Cancer Biology, Brown University and Lifespan Health System, Providence, United States
    4. Cancer Center at Brown University, Warren Alpert Medical School, Brown University, Providence, United States
    5. Department of Pathology and Laboratory Medicine, Warren Alpert Medical School, Brown University, Providence, United States
    6. Pathobiology Graduate Program, Warren Alpert Medical School, Brown University, Providence, United States
    7. Department of Molecular Microbiology and Immunology, Brown University, Providence, United States
    Contribution
    Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing - original draft, Writing - review and editing
    Contributed equally with
    Anna D Louie
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-5810-8220
  2. Anna D Louie

    1. Brown Experimentalists Against COVID-19 (BEACON) Group, Brown University, Providence, United States
    2. Laboratory of Translational Oncology and Experimental Cancer Therapeutics, Warren Alpert Medical School, Brown University, Providence, United States
    3. The Joint Program in Cancer Biology, Brown University and Lifespan Health System, Providence, United States
    4. Cancer Center at Brown University, Warren Alpert Medical School, Brown University, Providence, United States
    5. Department of Surgery, Lifespan Health System and Warren Alpert Medical School, Brown University, Providence, United States
    Contribution
    Conceptualization, Formal analysis, Investigation, Visualization, Methodology, Writing - original draft, Writing - review and editing
    Contributed equally with
    Kelsey E Huntington
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-8394-6106
  3. Chun Geun Lee

    1. The Joint Program in Cancer Biology, Brown University and Lifespan Health System, Providence, United States
    2. Cancer Center at Brown University, Warren Alpert Medical School, Brown University, Providence, United States
    3. Department of Molecular Microbiology and Immunology, Brown University, Providence, United States
    Contribution
    Conceptualization, Resources, Investigation, Writing - review and editing
    Competing interests
    No competing interests declared
  4. Jack A Elias

    1. The Joint Program in Cancer Biology, Brown University and Lifespan Health System, Providence, United States
    2. Cancer Center at Brown University, Warren Alpert Medical School, Brown University, Providence, United States
    3. Department of Molecular Microbiology and Immunology, Brown University, Providence, United States
    Contribution
    Conceptualization, Resources, Investigation, Writing - review and editing
    Competing interests
    No competing interests declared
  5. Eric A Ross

    Biostatistics and Bioinformatics Facility, Fox Chase Cancer Center, Temple University Health System, Philadelphia, United States
    Contribution
    Formal analysis, Methodology, Writing - review and editing
    Competing interests
    No competing interests declared
  6. Wafik S El-Deiry

    1. Brown Experimentalists Against COVID-19 (BEACON) Group, Brown University, Providence, United States
    2. Laboratory of Translational Oncology and Experimental Cancer Therapeutics, Warren Alpert Medical School, Brown University, Providence, United States
    3. The Joint Program in Cancer Biology, Brown University and Lifespan Health System, Providence, United States
    4. Cancer Center at Brown University, Warren Alpert Medical School, Brown University, Providence, United States
    5. Department of Pathology and Laboratory Medicine, Warren Alpert Medical School, Brown University, Providence, United States
    6. Pathobiology Graduate Program, Warren Alpert Medical School, Brown University, Providence, United States
    7. Hematology-Oncology Division, Department of Medicine, Lifespan Health System and Warren Alpert Medical School, Brown University, Providence, United States
    Contribution
    Conceptualization, Resources, Supervision, Funding acquisition, Investigation, Methodology, Writing - original draft, Project administration, Writing - review and editing
    For correspondence
    wafik@brown.edu
    Competing interests
    Senior Editor, eLife
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-9577-8266

Funding

Brown University

  • Wafik S El-Deiry

National Institute of General Medical Sciences (U54GM115677)

  • Kelsey E Huntington
  • Anna D Louie
  • Chun Geun Lee
  • Jack A Elias
  • Eric A Ross
  • Wafik S El-Deiry

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

The work was supported by a Brown University COVID-19 Seed Grant (to WSE-D). The COVID-19 Biobank through which plasma samples were obtained was supported by Institutional Development Award Number U54GM115677 from the National Institute of General Medical Sciences of the National Institutes of Health, which funds Advance Clinical and Translational Research (Advance-CTR). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. WSE-D is an American Cancer Society Research Professor.

Senior Editor

  1. Mone Zaidi, Icahn School of Medicine at Mount Sinai, United States

Reviewing Editor

  1. Jameel Iqbal, James J. Peters Veterans Affairs Medical Center, United States

Publication history

  1. Received: November 17, 2020
  2. Accepted: January 13, 2021
  3. Accepted Manuscript published: January 14, 2021 (version 1)
  4. Version of Record published: February 9, 2021 (version 2)

Copyright

© 2021, Huntington et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 1,332
    Page views
  • 168
    Downloads
  • 2
    Citations

Article citation count generated by polling the highest count across the following sources: PubMed Central, Crossref, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

  1. Further reading

Further reading

    1. Evolutionary Biology
    2. Immunology and Inflammation
    Srijan Seal et al.
    Review Article

    Researchers worldwide are repeatedly warning us against future zoonotic diseases resulting from humankind’s insurgence into natural ecosystems. The same zoonotic pathogens that cause severe infections in a human host frequently fail to produce any disease outcome in their natural hosts. What precise features of the immune system enable natural reservoirs to carry these pathogens so efficiently? To understand these effects, we highlight the importance of tracing the evolutionary basis of pathogen tolerance in reservoir hosts, while drawing implications from their diverse physiological and life-history traits, and ecological contexts of host-pathogen interactions. Long-term co-evolution might allow reservoir hosts to modulate immunity and evolve tolerance to zoonotic pathogens, increasing their circulation and infectious period. Such processes can also create a genetically diverse pathogen pool by allowing more mutations and genetic exchanges between circulating strains, thereby harboring rare alive-on-arrival variants with extended infectivity to new hosts (i.e., spillover). Finally, we end by underscoring the indispensability of a large multidisciplinary empirical framework to explore the proposed link between evolved tolerance, pathogen prevalence, and spillover in the wild.

    1. Immunology and Inflammation
    2. Microbiology and Infectious Disease
    Hao Gu et al.
    Research Article

    Vaccination strategies for rapid protection against multidrug-resistant bacterial infection are very important, especially for hospitalized patients who have high risk of exposure to these bacteria. However, few such vaccination strategies exist due to a shortage of knowledge supporting their rapid effect. Here, we demonstrated that a single intranasal immunization of inactivated whole cell of Acinetobacter baumannii elicits rapid protection against broad A. baumannii-infected pneumonia via training of innate immune response in Rag1-/- mice. Immunization-trained alveolar macrophages (AMs) showed enhanced TNF-α production upon restimulation. Adoptive transfer of immunization-trained AMs into naive mice mediated rapid protection against infection. Elevated TLR4 expression on vaccination-trained AMs contributed to rapid protection. Moreover, immunization-induced rapid protection was also seen in Pseudomonas aeruginosa and Klebsiella pneumoniae pneumonia models, but not in Staphylococcus aureus and Streptococcus pneumoniae model. Our data reveal that a single intranasal immunization induces rapid and efficient protection against certain Gram-negative bacterial pneumonia via training AMs response, which highlights the importance and the possibility of harnessing trained immunity of AMs to design rapid-effecting vaccine.