Investigation of liquid biopsy analytes in peripheral blood of individuals after SARS-CoV-2 infection

Abstract

Background: Post-acute COVID-19 syndrome (PACS) is linked to severe organ damage. The identification and stratification of at-risk SARS-CoV-2 infected individuals is vital to providing appropriate care. This exploratory study looks for a potential liquid biopsy signal for PACS using both manual and machine learning approaches.

Methods: Using a high definition single cell assay (HDSCA) workflow for liquid biopsy, we analysed 100 Post-COVID patients and 19 pre-pandemic normal donor (ND) controls. Within our patient cohort, 73 had received at least 1 dose of vaccination prior to SARS-CoV-2 infection. We stratified the COVID patients into 25 asymptomatic, 22 symptomatic COVID-19 but not suspected for PACS and 53 PACS suspected. All COVID-19 patients investigated in this study were diagnosed between April 2020 and January 2022 with a median 243 days (range 16-669) from diagnosis to their blood draw. We did a histopathological examination of rare events in the peripheral blood and used a machine learning model to evaluate predictors of PACS.

Findings: The manual classification found rare cellular and acellular events consistent with features of endothelial cells and platelet structures in the PACS-suspected cohort. The three categories encompassing the hypothesised events were observed at a significantly higher incidence in the PACS-suspected cohort compared to the ND (p-value < 0.05). The machine learning classifier performed well when separating the NDs from Post-COVID with an accuracy of 90.1%, but poorly when separating the patients suspected and not suspected of PACS with an accuracy of 58.7%.

Interpretation: Both the manual and the machine learning model found differences in the Post-COVID cohort and the NDs, suggesting the existence of a liquid biopsy signal after active SARS-CoV-2 infection. More research is needed to stratify PACS and its subsyndromes.

Funding: This work was funded in whole or in part by Fulgent Genetics, Kathy and Richard Leventhal and Vassiliadis Research Fund. This work was also supported by the National Cancer InstituteU54CA260591.

Keywords: COVID-19; Liquid biopsy; Long COVID; Post-COVID sequelae; Post-acute COVID-19 syndrome (PACS); SARS-CoV-2.

Fig. 1

Gallery of rare cellular events found in the PB of Post-COVID patients. (A) DAPI positive, Vim positive, and CD45/CD31 positive with varying morphology; (B) Elongated cells with CD45/CD31 foci at both ends; (C) large DAPI positive and CD45/CD31 positive events. Blue: DAPI, Red: CK, White: VIM, Green: CD45/CD31. Images taken at 100× magnification. Scale bar = 10 μm.

Fig. 2

Gallery of rare acellular events found in the PB of Post-COVID patients. Blue: DAPI, Red: CK, White: VIM, Green: CD45/CD31. Images taken at 100× magnification. Scale bar = 10 μm.

Fig. 3

Rare event detection using HDSCA3.0 in PB samples collected from PACS Suspected (n = 5) and ND (n = 5). (A) Enumeration and (B) frequency of each rare event by channel-type specification for PACS Suspected patient samples. (C) Box plot of the channel-type rare events/mL between PACS Suspected and ND samples ordered by degree of statistical significance. The median value is depicted by the midline of the boxplot, the third and first quartiles are the upper and lower bounds of the box, and the whiskers cover 1.5 times the IQR. Channel-type specifications that were statistically significant across the two classifications are highlighted (p < 0.05).

Fig. 4

Enumeration profile and feature importance of the ten top clusters out of a total of 306 for the classifier built on the ND (N = 19) and Post-COVID (N = 100) cohorts. (A) Box plot showing the cellular and acellular cluster enumeration profile of the ND and Post-COVID cohorts. The median value is depicted by the midline of the boxplot, the third and first quartiles are the upper and lower bounds of the box, and the whiskers cover 1.5 times the IQR. (B) Feature importance of top ten clusters quantified by the mean decrease in impurity obtained from the Random Forest component of the classifier model.