Circulating proteins to predict adverse COVID-19 outcomes

Chen‐Yang Su, Sirui Zhou, Edgar Gonzalez‐Kozlova, Guillaume Butler‐Laporte, Elsa Brunet‐Ratnasingham, Tomoko Nakanishi, Wonseok Jeon, David Morrison, Lætitia Laurent, Jonathan Afilalo, Marc Afilalo, Danielle Henry, Yiheng Chen, Julia Carrasco-Zanini, Yossi Farjoun, Maik Pietzner, Nofar Kimchi, Zaman Afrasiabi, Nardin Rezk, Meriem Bouab, Louis Petitjean, Charlotte Guzman, Xiaoqing Xue, Chris Tselios, Branka Vulesevic, Olumide Adeleye, Tala Abdullah, Noor Almamlouk, Yara Moussa, Chantal DeLuca, Naomi Duggan, Erwin Schurr, Nathalie Brassard, Madéleine Durand, Diane M. Del Valle, Ryan C. Thompson, Mario A. Cedillo, Eric E. Schadt, Kai Nie, Nicole W. Simons, Konstantinos Mouskas, Nicolas Zaki, Manishkumar Patel, Hui Xie, Jocelyn Harris, Robert Marvin, Esther Cheng, Kevin Tuballes, Kimberly Argueta, Ieisha Scott, Celia M.t. Greenwood, Clare Paterson, Michael Hinterberg, Claudia Langenberg, Vincenzo Forgetta, Joëlle Pineau, Vincent Mooser, Thomas U. Marron, Noam D. Beckmann, Ephraim Kenigsberg, Seunghee Kim‐Schulze, Alexander W. Charney, Sacha Gnjatic, Daniel E. Kaufmann, Miriam Mérad, J. Brent RichardsmedRxiv 2021 · DOI

Abstract

Predicting the clinical trajectory of COVID-19 patients remains a significant challenge due to the disease's heterogeneous presentation. While clinical risk factors provide some guidance, they often fail to capture the underlying biological state driving severe outcomes. In this study, researchers utilized a high-throughput proteomic platform to measure 4,701 circulating proteins in two independent cohorts (n=986). By applying LASSO regularized regression, a 92-protein signature was identified that predicts severe COVID-19 with an Area Under the Curve (AUC) of 86% in an external validation cohort, significantly outperforming standard clinical models. These findings highlight the importance of non-immune pathways in disease progression and suggest that the circulating proteome can serve as a powerful tool for early patient risk stratification.

Introduction

The COVID-19 pandemic has been characterized by extreme variability in patient outcomes. Individuals with similar clinical profiles—matched for age, sex, and known comorbidities—often experience vastly different disease courses, ranging from mild respiratory symptoms to multi-organ failure and death. This "wildcard" nature complicates hospital triage and resource allocation.

Current clinical risk scores rely heavily on demographic data and pre-existing conditions. However, these factors are distal to the immediate physiological response to the virus. To improve predictive accuracy, researchers have turned to the proteome—the complete set of proteins expressed by an organism—which provides a real-time snapshot of biological activity.

Methods and Study Design

The study employed the SomaScan platform, a proteomic technology using Slow Off-Rate Modified Aptamers (SOMAmers) to quantify 4,701 unique human proteins from blood samples. To handle the high dimensionality of the data, the team used LASSO (L1 regularized logistic regression) to select the most predictive features while preventing overfitting.

The model's generalizability was tested using a rigorous two-cohort design:

  1. Training Cohort (BQC19): 417 individuals from the Biobanque Québécoise de la COVID-19 in Montreal, Canada.
  2. Testing Cohort (Mount Sinai): 569 individuals from Mount Sinai Hospital in New York City.

The primary objective was to predict "severe COVID-19" (defined by the need for supplemental oxygen or death) and "critical COVID-19" (intubation or death) based on samples collected within an average of seven days from symptom onset.

Results: Superior Predictive Accuracy

The proteomic model demonstrated a marked improvement over traditional clinical predictors. A baseline model using only age and sex achieved an AUC of 65% in the NYC testing cohort. In contrast, the 92-protein model reached an AUC of 86% for predicting severe outcomes.

Key performance metrics in the external validation cohort included:

  • Severe COVID-19 Prediction: 86% AUC.
  • Critical COVID-19 Prediction: 80% AUC.
  • Positive Predictive Value (PPV): 88.9% for severe outcomes, indicating high reliability in identifying high-risk patients.

Notably, the model's performance remained stable across the two geographically distinct cohorts, suggesting that the identified protein signatures reflect fundamental biological processes associated with COVID-19 severity rather than site-specific artifacts.

Biological Insights: Beyond the Cytokine Storm

While much of the early COVID-19 research focused on "cytokine storms," this broad proteomic analysis revealed a more complex landscape. Cytokines and chemokines represented only approximately 5% of the proteins selected by the model.

More than half of the enriched biological pathways were not directly related to the immune response. Significant pathways included protein phosphorylation and glycosaminoglycan binding, suggesting that systemic physiological disruptions beyond inflammation are critical drivers of severity. Furthermore, the study found that adding clinical comorbidity data (e.g., diabetes, heart disease) to the protein model did not improve its performance. This implies that the circulating proteome acts as a molecular proxy, already capturing the physiological impact of these underlying conditions.

Discussion and Limitations

The ability to predict adverse outcomes early in the disease course (within 14 days of symptoms) offers a potential window for proactive clinical intervention. However, several hurdles remain before this approach can be implemented in a clinical setting. The current SomaScan panel is resource-intensive and not optimized for the rapid turnaround required in emergency departments. Future work must focus on developing targeted, low-cost assays for the specific 92-protein signature identified.

Additionally, while the study's multi-center design is a strength, further validation in more ancestrally diverse populations is necessary. The researchers also noted that the control groups consisted of symptomatic individuals with other respiratory illnesses; thus, the model identifies general "severity of illness" in a clinical context rather than being a specific diagnostic tool for COVID-19.

Conclusion

This research demonstrates that a high-dimensional proteomic approach can significantly outperform standard clinical factors in predicting COVID-19 severity. By uncovering the role of non-immune pathways and providing a robust predictive signature, this study paves the way for more precise patient triaging and a deeper understanding of the systemic nature of severe COVID-19.

Summary updated

Abstract

Abstract Predicting COVID-19 severity is difficult, and the biological pathways involved are not fully understood. To approach this problem, we measured 4,701 circulating human protein abundances in two independent cohorts totaling 986 individuals. We then trained prediction models including protein abundances and clinical risk factors to predict adverse COVID-19 outcomes in 417 subjects and tested these models in a separate cohort of 569 individuals. For severe COVID-19, a baseline model including age and sex provided an area under the receiver operator curve (AUC) of 65% in the test cohort. Selecting 92 proteins from the 4,701 unique protein abundances improved the AUC to 88% in the training cohort, which remained relatively stable in the testing cohort at 86%, suggesting good generalizability. Proteins selected from different adverse COVID-19 outcomes were enriched for cytokine and cytokine receptors, but more than half of the enriched pathways were not immune-related. Taken together, these findings suggest that circulating proteins measured at early stages of disease progression are reasonably accurate predictors of adverse COVID-19 outcomes. Further research is needed to understand how to incorporate protein measurement into clinical care.
BibTeX
@misc{su2021circulating,
  author = {Chen‐Yang Su and Sirui Zhou and Edgar Gonzalez‐Kozlova and Guillaume Butler‐Laporte and Elsa Brunet‐Ratnasingham and Tomoko Nakanishi and Wonseok Jeon and David Morrison and Lætitia Laurent and Jonathan Afilalo and Marc Afilalo and Danielle Henry and Yiheng Chen and Julia Carrasco-Zanini and Yossi Farjoun and Maik Pietzner and Nofar Kimchi and Zaman Afrasiabi and Nardin Rezk and Meriem Bouab and Louis Petitjean and Charlotte Guzman and Xiaoqing Xue and Chris Tselios and Branka Vulesevic and Olumide Adeleye and Tala Abdullah and Noor Almamlouk and Yara Moussa and Chantal DeLuca and Naomi Duggan and Erwin Schurr and Nathalie Brassard and Madéleine Durand and Diane M. Del Valle and Ryan C. Thompson and Mario A. Cedillo and Eric E. Schadt and Kai Nie and Nicole W. Simons and Konstantinos Mouskas and Nicolas Zaki and Manishkumar Patel and Hui Xie and Jocelyn Harris and Robert Marvin and Esther Cheng and Kevin Tuballes and Kimberly Argueta and Ieisha Scott and Celia M.T. Greenwood and Clare Paterson and Michael Hinterberg and Claudia Langenberg and Vincenzo Forgetta and Joëlle Pineau and Vincent Mooser and Thomas U. Marron and Noam D. Beckmann and Ephraim Kenigsberg and Seunghee Kim‐Schulze and Alexander W. Charney and Sacha Gnjatic and Daniel E. Kaufmann and Miriam Mérad and J. Brent Richards and J Brent Richards},
  title = {{Circulating proteins to predict adverse COVID-19 outcomes}},
  year = {2021},
  doi = {10.1101/2021.10.04.21264015},
  howpublished = {medRxiv},
}
Figure