Relationship of Heterologous Virus Responses and Outcomes in Hospitalized COVID-19 Patients

This article is featured in Top Reads, p. 1171

The relationship of COVID-19 outcomes and Ab responses to other viruses has been incompletely examined. Although T cell responses to seasonal coronaviruses may be protective against severe COVID-19 disease (1), a conclusive effect of Ab responses to seasonal coronaviruses has not been shown. For influenza, vaccination was found to be slightly protective against severe COVID-19 in children (2), but more detailed relationships have yet to be uncovered. For CMV, there is evidence that seropositivity may predispose to more severe disease (3), although the mechanism is unclear.

The National Institutes of Health IMPACC (Immunophenotyping Assessment in a COVID-19 Cohort) study enrolled >1000 hospitalized COVID-19 patients with follow-up for up to 1 y postadmission, with frequent sampling during the first 28 d of hospitalization and quarterly thereafter (4). Multiple immunological and virological assays were performed over time, yielding correlates of disease severity. To quantify disease severity, patients were grouped into five clinical trajectory groups, from most mild to fatal (5).

To assess Ab responses to a wide variety of viruses, we created a Luminex-based serology assay to measure Abs to several SARS-CoV-2 Ags as well as Ags from seasonal coronaviruses, influenza, parainfluenza, respiratory syncytial virus (RSV), CMV, EBV, and measles (Table I). We then applied this assay to sera from a large subset of IMPACC patients (498 patients assayed at Stanford via Olink). Samples were taken during acute illness (up to 28 d postadmission) and convalescence (up to 1 y later). We sought to answer the following questions: 1) How do initial Ab responses to heterologous viruses correlate with the clinical response to COVID-19? 2) How do responses to heterologous viruses change over the course of COVID-19? 3) What trends do these Ab responses show with age and sex? Our findings demonstrate (to our knowledge) new relationships between heterologous virus responses and COVID-19, and they reinforce some findings from previous studies.

A total of 1699 serum samples were drawn from 498 patients in the IMPACC study (4), which corresponded to all samples sent to the Stanford Olink core. Clinical characteristics are summarized in Supplemental Table I. Samples were stored at −80°C after an initial thaw for Olink immunoassay. They were later thawed and diluted for Luminex as described below.

Ags of interest (Table I) were coupled/conjugated to barcoded beads as recommended by the manufacturer (Luminex, Austin, TX).

Assay Chex control beads by Radix BioSolutions (Georgetown, TX) were added to the panel of beads and included in each well. Samples were diluted 1:400 in PBS + 0.5% Triton X-100, and 25 μl of diluted serum or plasma was added to an assay plate containing the Ag-coupled beads and incubated for 2 h at room temperature, shaking on an orbital shaker at 500–600 rpm. This step supports the binding of Abs to their Ags, creating multiple unique sandwich assays on beads.

The plate was washed in a BioTek ELx405 magnetic washer (BioTek, Winooski, VT) to remove serum/plasma from the previous step. A secondary Ab goat anti-human IgG (Fc fragment) or anti-IgM Fc coupled to PE was diluted and added to the plate. Incubation was performed for 30 min at room temperature with shaking as above (anti-IgG, catalog no. NC9822979; anti-IgM, catalog no. 501941614; anti-IgA, catalog no. NC0631234). The plate was washed in a BioTek ELx405 magnetic washer (BioTek, Winooski, VT) to remove excess secondary Ab.

Wash buffer (130 μl) was added to wells and read on a Luminex Flex three-dimensional instrument with a lower bound of 50 beads per target Ag. The output CSV file was analyzed with MasterPlex QT software by MiraiBio/Hitachi. Data are presented as median fluorescence intensity (MFI). The PBS with Tween 20 buffer was measured for background, and a control serum sample was included as a negative control (Sigma-Aldrich, serum purchased prior to COVID-19 pandemic).

Data were detrended by regressing the log MFI for each viral Ag and isotype on covariates using ordinary least squares regression (6). Regression residuals (unexplained variation) were retained and added back to the regression intercept. All continuous covariates were mean-centered for detrending.

Titers for each human coronavirus (HCoV) Ag (HCov-NL63 S1, HCov-OC43 HE, HCoV-229E S1, HCoV-HKU1 S1) were split at the baseline visit detrended (6) median. Then, two groups of participants were selected: those who fell below all four medians and those who fell above all four medians. For each SARS-CoV-2 Ag (spike membrane prtein [SM], receptor-binding domain [RBD], nucleocapsid protein [NP]), temporal trends in covariate detrended (6) data were plotted for these two groups within same graph. The detrended (6) MFI was regressed on a B-spline (7) over elapsed time, the baseline seasonal HCoV status, and their interaction using generalized linear mixed models (8). A random intercept per participant was included to account for longitudinal structure. Agreement of residuals and random effects with normal distribution was marginal to fair as assessed by quantile–quantile plots.

The logarithm of the MFI for each Ag was regressed on age, sex, logarithm of the MFI of nonspecific binding, plate/batch, and trajectory group using ordinary least squares (6) with allowance for variance heterogeneity (9). All p values were adjusted for multiple comparisons using the method of Holm (10), separately for each combination of Ag type (SARS-CoV-2, HCoV) and Ab isotype. All means comparisons are against trajectory group 1. Data in plots are detrended (6) for covariates.

The logarithm of the MFI for each Ag was regressed on age, sex, logarithm of the MFI of nonspecific binding, plate/batch, and trajectory group using ordinary least squares (6) with allowance for variance heterogeneity (9). All p values were adjusted (11, 12) to control the false discovery rate at 5%, separately for each Ab isotype and trajectory group. All means comparisons are against trajectory group 1. Data in plots are detrended (6) for covariates.

IgG CMV seropositivity status (0, negative; 1, positive) was regressed on age, sex, logarithm of the MFI of nonspecific binding, plate/batch, and trajectory group using binomial generalized linear models (13). All means comparisons are against trajectory group 1.

The logarithm of the MFI for each Ag and Ab isotype was regressed on age, sex, logarithm of the MFI of nonspecific binding, plate/batch, and trajectory group using ordinary least squares (6) with allowance for variance heterogeneity (9). All p values were adjusted (11, 12) to control the false discovery rate at 5%, by each combination of age or sex and isotype. Data in plots are detrended (6) for covariates.

The logarithm of the MFI of each Ag was regressed on age, sex, logarithm of the MFI of nonspecific binding, batch/plate, trajectory group, square root of elapsed days, the interaction of the trajectory group and square root of elapsed days, and a quadratic (curvilinear) term for square root of elapsed days using linear quantile mixed models (LQMMs) (14). LQMMs included a random intercept per participant to account for longitudinal structure. All p values were adjusted for multiple comparisons using Holm (10), separately for each combination of Ag type (SARS-CoV-2, HCoV), Ab isotype, and trajectory group. All comparisons of slopes over time are against the slope in trajectory group 1. Data in plots are detrended (6) for covariates.

Patients were classified as CMV or EBV seropositive at the first visit based on empiric cutoffs. IgG MFI data detrended (6) for nonspecific binding and plate/batch were regressed on a B-spline basis (7) for elapsed days, initial CMV or EBV seropositivity, and their interaction using generalized linear mixed models (8). A random intercept per participant was included to account for longitudinal structure. The interaction term was tested to assess whether longitudinal trends varied with initial CMV or EBV seropositivity. Agreement of residuals and random effects with normal distribution was poor to good as assessed by quantile–quantile plots.

Olink serum cytokine data have been previously reported for the IMPACC study (15). All records in the cytokine data that did not pass quality control or were missing observations for all cytokines were excluded from analyses. Remaining missing cytokine values were imputed using the knn.impute function in R package impute (16). Cytokines were excluded from the analyses when their dynamic range (maximum minus minimum) fell at or below the 10th percentile of all cytokines’ dynamic ranges. Cytokine data were detrended for plate effects, and Ab data were detrended for plate/batch and nonspecific binding effects using ordinary least squares (6). All Ab MFI data were logarithmically transformed for analyses. A Kendall correlation (17) was calculated for all pairs of Ags and proteins. All p values were adjusted (11, 12) to control the false discovery rate at 5%.

Analysis and graphing were performed in SAS (SAS Institute, Cary, NC) and R (https://www.r-project.org) under R Studio (https://www.rstudio.com) with R packages dplyr (18), impute (16), ggplot2 (19), mutoss (20), Kendall (21), sandwich (22), and lqmm (23).

A total of 1699 serum samples from 498 IMPACC patients (4) were collected at admission (initial samples) and at 1–28 d later (longitudinal samples). These were assessed via Luminex assay for their responses to SARS-CoV-2 Ags and Ags from seasonal coronaviruses as well as other respiratory and nonrespiratory viruses (Table I). IgM responses were analyzed in a subset (192 samples from 85 patients). We first asked whether initial responses to SARS-CoV-2 Ags varied by clinical trajectory group (5). As shown in Fig. 1, the first visit titers for four SARS-CoV-2 Ags were quite heterogeneous, but they tended to increase slightly from the mildest cases (group 1) to the most severe nonfatal cases (group 4), with fatal cases (group 5) being more similar to group 1. The difference between group 4 and group 1 reached statistical significance for RBD and S-M. We further asked whether longitudinal trends in SARS-CoV-2 Ab titers during the acute and convalescent period (up to 1 y postadmission) differed by clinical trajectory group. As seen in Fig. 1B, trajectory group 4 showed a significant difference from group 1 for the NP Ag of SARS-CoV-2. Longitudinal trends for the other SARS-CoV-2 Ags showed similar patterns, but they did not reach statistical significance (data not shown).

We performed a similar analysis for initial titers to the four seasonal coronaviruses, that is, 229E, HKU1, OC43, and NL63. Fig. 2A shows that these titers tended to be lower with increasing disease severity, reaching statistical significance for groups 4 and/or 5 depending on the seasonal coronavirus strain.

When examining longitudinal trends for seasonal coronaviruses, we found little change during the course at 28 d postadmission, except for several cases of spiking IgM responses (Fig. 2B). We interpreted these to be due to cross-reactivity of SARS-CoV-2 Ags with seasonal coronaviruses, because it is unlikely that these patients experienced concomitant seasonal coronavirus infections.

Given the suspected cross-reactivity mentioned above, we asked whether pre-existing seasonal coronavirus titers might cross-react with SARS-CoV-2 to influence the outcome of COVID-19. However, we found no difference in outcome groups for those with high versus low initial titers to the seasonal coronaviruses (Fig. 2C).

We next asked whether initial titers to influenza and parainfluenza Ags varied by COVID-19 trajectory group. We assessed responses to a common influenza A Ag as well as to hemagglutinins from the 2019 influenza vaccine strains, plus parainfluenza 2 and parainfluenza 3. As shown in Fig. 3, virtually all of these showed a similar trend as the seasonal coronavirus titers, with lower average titers as the COVID-19 trajectory group increased. The differences relative to group 1 reached statistical significance for groups 2, 3, 4, and/or 5 depending on the Ag. The only exception was the influenza B strain, B/Phuket/3073/2013, which had no significant differences between trajectory group 1 and other groups, but for which titers were overall relatively low.

We next asked whether decreasing titers for the above viruses with COVID-19 severity were related to a serum cytokine signature. To examine this, we tested for correlations between Ab titers for these viruses and serum cytokines as determined by an Olink proximity extension assay. We found a number of significant correlations (Table II), both positive and negative. Many of these were significant for multiple viruses, for example, positive correlations of CCL8, CCL25, DNER, IL-12β, and KITLG, and negative correlations of CCL4, OSM, S100A12, and TNFSF14 with titers for multiple viruses.

Finally, we assessed initial titers to other unrelated viruses (CMV, EBV, measles, and RSV) by COVID-19 trajectory group. As seen in Fig. 4A, there were no significant differences relative to trajectory group 1 for any of these viruses. Furthermore, the longitudinal plots of titers were generally flat, with the exception of a few patients who showed a spike in IgM for CMV or EBV during the course of COVID-19 infection (Fig. 4B).

Because CMV has been associated with an increased risk of hospitalization for COVID-19 (3), we asked whether our hospitalized cohort had an overrepresentation of CMV or EBV seropositive individuals. To determine this, we calculated an expected rate of CMV or EBV positivity based on published studies in healthy adults. This was aided for CMV by a large study of CMV seropositivity rates by age in United States adults (24). From this, we calculated an age-adjusted expected frequency of CMV seropositives of 0.752, which compared with an observed frequency of 0.844 in our cohort (Supplemental Fig. 1A). For EBV, the observed rate of 0.965 in our cohort was very similar to the expected rate of 0.94–0.96 from other studies (25, 26). We determined seropositivity for CMV and EBV using an empiric cutoff as shown in Supplemental Fig. 1B.

CMV seropositivity was not associated with trajectory group (data not shown). We did not see a difference in CMV (or any other virus) titers between males and females (Supplemental Fig. 1C). We also did not see a difference in longitudinal trend of titers to SARS-CoV-2 Ags in those who were CMV or EBV seropositive versus seronegative (Supplemental Fig. 1D, 1E).

We also checked our cohort for associations of age with viral titers. We saw a downward trend with age for SARS-CoV-2 titers, which was significant for all three SARS-CoV-2 Ags tested (p = 0.001–0.038; Supplemental Fig. 2A). Interestingly, all other significant age associations were upward trends with age. These included one of the seasonal coronaviruses (OC43), as well as measles, parainfluenza 2 and 3, and several influenza Ags (Supplemental Fig. 2B, 2C). We did not see significant age associations for CMV or EBV.

Finally, we looked for correlations between titers for different Ags tested. Not surprisingly, we saw strong correlations between titers for the three SARS-CoV-2 Ags tested, with the strongest being between S-M and RBD (with the latter being a subunit of the former) (r = 0.97, p < 0.01; Supplemental Fig. 3A). The correlations in titers for the four seasonal coronaviruses were more modest but still significant (r = 0.19–0.51, p < 0.01; Supplemental Fig. 3B). For all other virus titers, there was low or no correlation (Supplemental Fig. 3C), except for a good alignment between titers for the two influenza B strains (B/Colorado/06/2017 and B/Phuket/3073/2013; r = 0.84, p < 0.01).

In this study, we described the relationships of viral titers for different viruses in a subset of approximately half of the IMPACC cohort of hospitalized patients with COVID-19 (4). These titers were quite heterogeneous, possibly due in part to differing time since symptom onset, which we did not account for. Nevertheless, we showed upward trends with increasing clinical trajectory group for baseline SARS-CoV-2 titers (except for group 5, which was fatal cases). Conversely, we found downward trends with increasing clinical trajectory group for other respiratory virus baseline titers, including seasonal coronaviruses, influenza, and parainfluenza. No significant differences with trajectory group were seen for titers to CMV, EBV, measles, or RSV. With the exception of RSV, these latter are all nonrespiratory viruses, whereas those with significant downward trends with increasing clinical trajectory group (seasonal coronaviruses, influenza, and parainfluenza) are all respiratory viruses. A possible hypothesis for this finding is that some individuals intrinsically mount poor responses to respiratory viruses, and these individuals then fare more poorly when they contract COVID-19. For influenza, there are also data to suggest a positive benefit of vaccination for COVID-19 outcome, but this was in children rather than adults (2).

A parallel study on the full IMPACC cohort (15) found no such associations of SARS-CoV-2 titers and trajectory group, using a highly comprehensive VirScan assay. However, more severe disease (group 5) was associated with increased seroreactivity to the N terminal domain of S and decreased Ab seroreactivity to the LINK domain of N (adjusted p = 0.023) (as shown in supplemental figure 2D in Ref. 15). We could not corroborate the domain-specific findings, as our study used only whole protein Ags or subunits (RBDs). It is possible that our Luminex assay was more quantitative than the VirScan assay, and thus was able to distinguish differences using the whole-protein Ags that were not apparent in the latter.

Many factors may contribute to the strength of Ab responses to heterologous viruses, including age of exposure, number of exposures, and exposure through infection versus vaccination. Although we do not know all of these variables for the IMPACC participants, we can make some inferences. For example, measles immunity is presumably almost all vaccine derived, whereas CMV and EBV are entirely infection derived; however, all of these show no trend with COVID-19 severity. Thus, the difference between viruses, whose titers trend with COVID-19 severity and those that do not, cannot be due simply to the variable of whether immunity was derived from infection versus vaccination.

It is also unlikely that immunosuppressive therapies caused the decreasing titers with severity seen for most of the respiratory viruses, because 1) the lower titers were seen on admission (visit 1), and 2) only respiratory viruses were affected. Similarly, conditions such as immune deficiency or cancer would not be expected to affect respiratory viruses preferentially. In fact, we tested for association of viral titers with malignant neoplasm and found none (all adjusted p = 1, data not shown).

We tracked titers to SARS-CoV-1 as a negative control, but instead saw positive responses in a subset of IMPACC patients (data not shown). We assume these reflected cross-reactivity with SARS-CoV-2 epitopes, and that only some COVID-19 patients mounted Abs to the cross-reactive epitopes.

Patients with high titers to seasonal coronaviruses did not show a significant longitudinal difference in their titers to SARS-CoV-2 Ags compared with those with low seasonal coronavirus titers (Fig. 2C). This suggests no advantage in the development of humoral immunity to SARS-CoV-2 based on previous exposure to seasonal coronaviruses. Although other studies have suggested potential benefits for COVID-19 patients from cross-reactive immunity to seasonal coronaviruses, this has mostly been on the side of cellular immunity (1, 27), so our results for cross-reactive Abs are not too surprising.

A previous study suggested cross-reactivity of T cell responses to CMV with SARS-CoV-2 (28). However, we did not find differences in the longitudinal development of Abs to SARS-CoV-2 in those who were CMV seropositive versus those who were not. We did, however, see an overrepresentation of CMV seropositives in our hospitalized COVID-19 cohort compared with the expected frequency in an age-adjusted healthy population (24). This is consistent with other studies suggesting that CMV seropositivity may be associated with increased risk of hospitalization for COVID-19 (3, 29).

We saw suggestions of CMV and EBV reactivation during COVID-19 in the form of spikes for IgM against CMV or EBV during acute COVID-19. Reactivation of these viruses in the setting of COVID-19 has been previously described, and it may be either an epiphenomenon or a driver of more severe inflammatory disease (30). In any case, it is interesting that we only saw IgM spikes in a small minority of COVID-19 patients, despite that these patients were all severe cases necessitating hospitalization.

We tested IgM and IgA responses in a subset (∼17%) of participants and did not find significant differences in titer by severity group. This could be because of the smaller sample size, resulting in lower statistical power, or it could be related to the higher prevalence and longer half-life of serum IgG relative to IgM and IgA.

We wondered whether the decreasing titers with COVID-19 severity for most respiratory viruses might be associated with a particular serum cytokine signature. Indeed, there were repeated correlations found between certain serum cytokine levels and multiples of these respiratory viruses (Table II). Although the pattern of positive and negative correlations was consistent, it did not suggest a simple conclusion, such as an increased proinflammatory environment associated with higher titers.

In examining the relationship of age and viral titers, we were surprised to see so many viruses for which there were significantly increased titers with age. These included one seasonal coronavirus (OC43), influenza, parainfluenza 2 and 3, and measles. Most of these are explainable by repeated exposure and/or vaccination during the course of the lifespan. However, the measles vaccine is generally only given in early childhood, and circulating virus is currently rare in the United States. The high titers for measles in older patients is thus likely a reflection of natural immunity rather than vaccine-acquired immunity seen in young patients.

In summary, our study both reinforces other studies and adds (to our knowledge) new hypotheses about the relationship of responses to heterologous viruses and the course of SARS-CoV-2 infection. Perhaps most intriguingly, we find a trend, for many respiratory viruses, of decreasing titers with increasing COVID-19 severity, suggesting possible common defects in response to multiple respiratory pathogens. These could include defects in cell trafficking or differentiation in respiratory mucosa, or other unknown differences when encountering respiratory pathogens.

Our study is limited by only measuring Ab (not T cell) responses, and by only sampling hospitalized (and thus more severe) COVID-19 patients. We also analyzed data using predefined trajectory groups as the major outcome; the choice of an ordinal scale alone could have impacted our results. Still, the inclusion of ∼500 patients from across the United States, with multiple time point sampling, is a strength. Further dissection of the role of immunity to heterologous viruses, especially CMV and influenza, is encouraged by these findings.

Al Ozonoff, Joann Diray-Arce, Carly E. Milliren, Kerry McEnaney, Brenda Barton, Claudia Lentucci, Mehmet Saluvan, Ana C. Chang, Annmarie Hoch, Marisa Albert, Tanzia Shaheen, Alvin T. Kho, Shanshan Liu, Sanya Thomas, Jing Chen, Maimouna D. Murphy, Mitchell Cooney, and Caitlin Syphurs (Clinical and Data Coordinating Center (CDCC), Boston Children’s Hospital, Boston, MA 02115, USA)

Arash Nemati Hayati, Robert Bryant, and James Abraham (Research Computing, Boston Children’s Hospital, Boston, MA 02115, USA)

Joanna Schaenman, Elaine F. Reed, Ramin Salehi-Rad, David Elashoff, Jenny Brook, Estefania Ramires-Sanchez, Megan Llamas, Adreanne Rivera, Claudia Perdomo, Dawn C. Ward, Clara E. Magyar, Jennifer Fulcher, Harry C. Pickering, and Subha Sen (David Geffen School of Medicine at the University of California Los Angeles, Los Angeles CA 90095, USA)

Naresh Doni Jayavelu, Matthew C. Altman, Scott Presnell, Bernard Khor, and Tomasz Jancsyk (Benaroya Research Institute, University of Washington, Seattle, WA 98101, USA)

Alejandra Jauregui, Aleksandra Leligdowicz, Alexander Beagle, Alexandra Tsitsiklis, Alyssa Ward, Ana Gonzalez, Andrew W. Schroeder, Andrew Willmore, Arjun Rao, Austin Sigman, Bonny Alvarenga, Bushra Samad, Carolyn Leroux, Carolyn M. Hendrickson, Carolyn S. Calfee, Charles R. Langelier, Christina Love, Cindy Curiel, Cole Shaw, David J. Erle, Deanna Lee, Eran Mick, Gabriela Fragiadakis, Gayelan Tietje-Ulrich, Jayant Rajan, Jeff Milush, Jonathan Singer, Joshua J. Vasquez, Kevin Tangv, Kirsten N. Kangelaris, Legna Betancourt, Lekshmi Santhosh, Lenka Maliskova, Logan Pierce, Luz Torres Altamirano, Maria Tercero Paz, Matthew F. Krummel, Michael Adkisson, Michael Matthay, Michael R. Wilson, Neeta Thakur, Nicklaus Rodriguez, Nicole Sutter, Norman Jones, Pratik Sinha, Prescott G. Woodruff, Priya Prasad, Rajani Ghale, Raphael Lota, Ravi Dandekar, Ravi Patel, Sadeed Rashid, Saurabh Asthana, Sharvari Bhide, Sidney A. Carrillo, Suzanna Chak, Tasha Lea, Viet Nguyen, Walter Eckalbar, Estella Sanchez Guerrero, and Yumiko Abe-Jones (University of California, San Francisco, CA, USA)

Charles B. Cairns, Elias K. Haddad, Mariana Bernui, Debra L. Powell, James N. Kim, Brent Simmons, I. Michael Goonewardene, Cecilia M. Smith, Mark Martens, Michele A. Kutzler, Carolyn Edwards, Jennifer Connors, Edward Lee, Edward Lin, Brett Croen, Nicholas Semenza, Brandon Rogowski, Nataliya Melnyk, Kyra Woloszczuk, Gina Cusimano, Matthew Bell, Sara Furukawa, Renee McLin, Pam Schearer, Julie Sheidy, George P. Tegos, and Crystal Nagle (Drexel University, Tower Health Hospital, Philadelphia, PA 19104, USA)

Vicki Seyfert-Margolis (MyOwnMed Inc, Bethesda, MD, USA)

Monica Kraft, Christian Bime, Jarrod Mosier, Hiroki Kimura, Michelle Conway, Dave Francisco, Allyson Molzahn, Heidi Erickson, Connie Cathleen Wilson, Ron Schunk, Trina Hughes, and Bianca Sierra (University of Arizona, Tucson AZ 85721, USA)

Lindsey R. Baden, Ofer Levy, Kinga K. Smolen, Michael Desjardins, Amy C. Sherman, Stephen R. Walsh, Simon van Haren, Xhoi Mitre, Jessica Cauley, Xiofang Li, Alexandra Tong, Bethany Evans, Christina Montesano, Jose Humberto Licona, Jonathan Krauss, Nicolas C. Issa, Jun Bai Park Chang, Natalie Izaguirre, Hanno Steen, Patrick van Zalm, Benoit Fatou, Kevin Mendez, Jessica Lasky-Su, Meenakshi Jha, Arthur Viode, and Rebecca Rooks (Harvard Medical School, Boston, MA 02115, USA)

Scott R. Hutton, Greg Michelotti, and Kari Wong (Metabolon Inc, Morrisville, NC 27560, USA)

Albert C. Shaw, Omkar Chaudhary, Andreas Coppi, Charles S. Dela Cruz, Denise Esserman, Shelli Farhadian, John Fournier, David A. Hafler, Akiko Iwasaki, Albert I. Ko, Subhasis Mohanty, Ruth R. Montgomery, M. Catherine Muenker, Allison Nelson, Khadir Raddassi, Michael Rainone, William Ruff, Syim Salahuddin, Wade L. Schulz, Pavithra Vijayakumar, Haowei Wang, Elsio A. Wunder Jr., H. Patrick Young, Yujiao Zhao, Leying Guan, Steven H. Kleinstein, Jeremy P. Gygi, Shrikant Pawar, Anderson Brito, Jessica Rothman, Anna Konstorum, Ernie Chen, Chris Cotsapas, Nathan D. Grubaugh, Xiaomei Wang, Leqi Xu, and Hiromitsu Asashima (Yale School of Medicine and/or Public Health, New Haven, CT 06510, USA)

Florian Krammer, Ana Fernandez Sesma, Viviana Simon, Harm Van Bakel, Miti Saksena, Deena Altman, Erna Kojic, Komal Srivastava, Lily Q Eaker, Maria C Bermúdez-González, Katherine F Beach, Levy A Sominsky, Arman R. Azad, Juan Manuel Carreño, Gagandeep Singh, Ariel Raskin, Johnstone Tcheou, Dominika Bielak, Hisaaki Kawabata, Temima Yellin, Miriam Fried, Leeba Sullivan, Sara Morris, Lubbertus CF Mulder, Giulio Kleiner, Adeeb Rahman, Daniel Stadlbauer, Jayeeta Dutta, Hui Xie, MS, Seunghee Kim-Schulze, Ana Silvia Gonzalez-Reiche, Adriana van de Guchte, Jingjing Qi, Brian Lee, Geoffrey Kelly, Manishkumar Patel, and Kai Nie (Icahn School of Medicine at Mount Sinai, New York, NY 10029, USA)

Patrice M. Becker, Alison D. Augustine, Tatyana Vaysman, Steven M. Holland, Lindsey B. Rosen, and Serena Lee (National Institute of Allergy and Infectious Diseases, National Institute of Health, Bethesda, MD 20814, USA)

Amer Bechnak, Andrew Cheng, Aneesh Mehta, Arun K. Boddapati, Bernadine Panganiban, Brandi Johnson, Caroline R. Ciric, Chistopher Huerta, Christine Spainhour, David Cowan, Elizabeth Beagle, Erin Carter, Erin M. Scherer, Evan J. Anderson, Greg K. Tharp, Hady Samaha, Jacob Usher, Jonathan E. Sevransky, Kathryn L. Pellegrini, Kieffer Hellmeister, Laila Hussaini, Laurel Bristow, Lauren Hewitt, Nadine Rouphael, Nina Mcnair, Steven E. Bosinger, Susan Pereira Ribeiro, Sydney Hamilton, and Thomas Hodder (Emory School of Medicine, Atlanta, GA 30322, USA)

Alexandra S Lee, Andrea Fernandes, Angela Rogers, Bali Pulendran, Catherine Blish, Hena Naz Din, Holden T. Maecker, Iris Chang, Jonasel Roque, Linda Geng, Maja Artandi, Mark M. Davis, Monali Manohar, Natalia Sigal, Neera Ahuja, Kari C. Nadeau, Samuel S Yang, Sharon Chinthrajah, and Thomas Hagan (Stanford University School of Medicine, Stanford, CA 94304, USA)

Catherine L. Hough, William B. Messer, Amanda E. Brunton, Sarah A.R. Siegel, Peter E. Sullivan, Matthew Strnad, Zoe L. Lyski, Felicity J. Coulter, Zhengchun Lu, and Courtney Micheleti (Oregon Health Sciences University, Portland, OR 97239, USA)

Bjoern Peters, James A. Overton, Randi Vita, and Kerstin Westendorf (La Jolla Institute for Immunology, La Jolla, CA 92037, USA)

Nelson I Agudelo Higuita, Jordan P. Metcalf, John L. Booth, Lauren A. Sinko, Douglas A. Drevets, and Brent R. Brown (Oklahoma University Health Sciences Center, Oklahoma City, OK 73104, USA)

Matthew L. Anderson, and Brittany Borrensen (University of South Florida, Tampa FL 33620, USA)

Mark A. Atkinson, Scott C. Brakenridge, Lyle Moldawer, Brittney Roth-Manning, and Ricardo F. Ungaro (University of Florida, Gainesville, FL 32611, USA)

Jordan Oberhaus, and Faheem W Guirgis (University of Florida, Jacksonville, FL 32218, USA)

David B. Corry, Farrah Kheradmand, Li-Zhen Song, and Ebony Nelson (Baylor College of Medicine and the Center for Translational Research on Inflammatory Diseases, Houston, TX 77030, USA)

Lauren I. R. Ehrlich, Esther Melamed, Rama V Thyagarajan, Justin Rousseau, Dennis Wylie, Todd A. Triplett, Nadia Siles, Cole Maguire, Janelle Geltman, and Kerin Hurley (University of Texas, Austin, TX 78712, USA)

Grace A. McComsey, Rafick Sekaly, Scott Sieg, Slim Fourati, Heather Tribout, Paul Harris, Mary Consolo, and George Yendewa (Case Western Reserve University and University Hospitals of Cleveland, Cleveland, OH 44106, USA)

Casey P. Shannon, and Scott J. Tebbutt (Prevention of Organ Failure (PROOF) Centre of Excellence, University of British Columbia, Vancouver, BC V6T 1Z3, Canada)

The authors have no financial conflicts of interest.

We thank the IMPACC clinical sites for recruiting patients and processing and banking samples, and the IMPACC patients for contributing to the study.