Biomarkers of COVID-19 Severity and Progression

The severity of COVID-19 can vary wildly from patient to patient: one may end up intubated in an intensive care unit and another may experience no symptoms at all. To better predict which patients will progress to severe disease, researchers have turned to nearly every tool in the omics toolkit—as well as artificial intelligence. The combination of systems-level biological analysis and machine learning has yielded a bevy of biomarkers for predicting COVID-19 severity and progression.

Unbiased snapshots of severe COVID-19

Like many infectious diseases, COVID-19 can follow different courses, ranging from asymptomatic to fatal. Identifying robust biomarkers of COVID-19 severity, therefore, requires using a severity grading scale to stratify patients. The World Health Organization (WHO) ordinal scale is a commonly used one. In this scale, high scores denote severe cases: an uninfected individual would be scored 0, a patient with severe disease 5, 6, or 7, and a fatal case 8. All severe cases require ventilation in some form or another.

Once patients are stratified, biomolecules can be measured and their relative abundance compared across severity groups. Researchers have left virtually no stone unturned in their search for biomarkers of COVID-19 severity, leveraging the throughput of omics to mine the transcriptome, proteome, and even metabolome for potential candidates. These approaches typically use next-generation sequencing to measure RNA or mass spectrometry to measure proteins and metabolites (lipids and other small molecules).

Despite variations in the methods applied and cohorts studied, commonalities have emerged from multiple omics studies. Proteomic studies have repeatedly identified coagulation proteins (KLKB1, KNG1, and SERPIND1), complement proteins (CRP, CFB, CFH, and CFI), serum amyloid proteins (SAA1 and SAA2), and protease inhibitors (ITIH1, ITIH3, ITIH4, and SERPINA3) as biomarkers of COVID-19 severity, with higher levels of these biomolecules correlating with higher disease severities.^1-5

Under similarly diverse circumstances, metabolomic studies have uncovered changes in lipid profiles associated with COVID-19 severity. In particular, multiple studies have reported changes in some phosphatidylcholines and lysophophatidylcholines, with most of these studies reporting a lower abundance of these lipids in more severe COVID-19 cases.^6-8 These lipids have been implicated in viral replication, host immunity, and organ damage.

In addition to these proteomic and metabolic biomarkers, several transcripts have been shown to be differentially expressed in severe versus mild COVID-19.^3,9 These include neutrophil extracellular trap (NET) markers MPO and PRNT3 and the pro-neutrophil markers CD24 and BPI, which have been previously shown to be upregulated in severe sepsis.

Enlisting AI

A key strength of omic-based approaches to biomarker discovery is their unbiased nature. Instead of focusing on a predefined set of usual suspects (cytokines, for example), omic technologies measure everything. However, the sheer volume of data produced by these technologies can make it difficult to determine which biomarkers, or combinations of biomarkers, are most predictive of disease severity. To help navigate these massive datasets, researchers have enlisted a form of artificial intelligence called machine learning (ML).

Simply put, ML leverages algorithms to identify patterns in large datasets. In COVID-19 biomarker discovery, these algorithms are trained to identify features associated with different patient populations. For example, Shu and colleagues⁵ used ML not only to distinguish between mild and severe COVID-19 cases, but also between severe cases with nonfatal and fatal outcomes. In this study, elevated acute-phase proteins, or those expressed early during infection (e.g., ORM1, ORM2, S100A9, CRP, and SERPINA3), were associated with severe and fatal outcomes. A separate time-resolved proteomic study confirmed CRP and SERPINA3 as predictors of future worsening in COVID-19 patients.¹

Looking before infection

Most biomarker studies have looked to the host response to predict COVID-19 severity and prediction. Few, by contrast, have looked for biomarkers that could be screened before infection to predict future COVID-19 outcomes. Yekelchyk and colleagues¹⁰ did just that, zeroing in on the cellular fitness marker Flower lose.

Cells use Flower lose to communicate their fitness to neighboring cells, essentially labeling themselves for elimination by the immune system. Given that Flower lose can accumulate over time in stressed cells, Yekelchyk and colleagues¹⁰ hypothesized that the marker would be elevated in patients with risk factors for severe COVID-19. The researchers found that lung tissue Flower lose expression was elevated in old age as well as in individuals with obesity, diabetes, chronic obstructive pulmonary disease, and cardiovascular disease. In COVID-19 patients, Flower lose also correlated with COVID-19 severity.

The marker’s expression in lung tissue offered an additional advantage for its use as a prognostic factor of COVID-19 severity: it could be detected in nasopharyngeal samples, the gold standard sample for COVID-19 diagnosis. Thus, Flower lose has the potential to simplify the triaging of COVID-19 patients by enabling simultaneous diagnosis and prognosis of COVID-19. What’s more, Yekelchyk and colleagues¹⁰ showed that the biomarker performs better than other COVID-19 biomarkers such as ferritin, CRP, D-dimer, and neutrophil-lymphocyte ratio.

Overall, the search for biomarkers of COVID-19 severity has done more to inform our understanding of COVID-19 pathophysiology than transform clinical practice. This is because easy-to-measure biomarkers are poorly predictive, and the combination of omics and machine learning has not been translated to the clinic. Moreover, the emergence of widely available COVID-19 treatments, as well as milder strains like Omicron, has substantially reduced the incidence of severe COVID-19. Nevertheless, the efforts to date offer a blueprint that will undoubtedly facilitate public health responses to future pandemics.

References

1. Demichev V, Tober-Lau P, Lemke O, et al. A time-resolved proteomic and prognostic map of COVID-19. Cell Syst. 2021;12(8):780-794.e7. 

2. Messner CB, Demichev V, Wendisch D, et al. Ultra-High-Throughput Clinical Proteomics Reveals Classifiers of COVID-19 Infection. Cell Syst. 2020;11(1):11-24.e4. 

3. Overmyer KA, Shishkova E, Miller IJ, et al. Large-Scale Multi-omic Analysis of COVID-19 Severity. Cell Syst. 2021;12(1):23-40.e7. 

4. Shen B, Yi X, Sun Y, et al. Proteomic and Metabolomic Characterization of COVID-19 Patient Sera. Cell. 2020;182(1):59-72.e15. 

5. Shu T, Ning W, Wu D, et al. Plasma Proteomics Identify Biomarkers and Pathogenesis of COVID-19. Immunity. 2020;53(5):1108-1122.e5. 

6. Delafiori J, Navarro LC, Siciliano RF, et al. Covid-19 Automated Diagnosis and Risk Assessment through Metabolomics and Machine Learning. Anal Chem. 2021;93(4):2471-2479. 

7. Sindelar M, Stancliffe E, Schwaiger-Haber M, et al. Longitudinal metabolomics of human plasma reveals prognostic markers of COVID-19 disease severity. Cell Rep Med. 2021;2(8):100369. 

8. Song JW, Lam SM, Fan X, et al. Omics-Driven Systems Interrogation of Metabolic Dysregulation in COVID-19 Pathogenesis. Cell Metab. 2020;32(2):188-202.e5. 

9. Schulte-Schrepping J, Reusch N, Paclik D, et al. Severe COVID-19 Is Marked by a Dysregulated Myeloid Cell Compartment. Cell. 2020;182(6):1419-1440.e23. 

10. Yekelchyk M, Madan E, Wilhelm J, et al. Flower lose, a cell fitness marker, predicts COVID-19 prognosis. EMBO Mol Med. 2021;13(11):e13714.