Distinguishing features of Long COVID identified through immune profiling

Jon Klein, Jamie Wood, Jillian R. Jaycox, Peiwen Lu, Rahul M. Dhodapkar, Jeff Gehlhausen, Alexandra Tabachnikova, Laura Tabacof, Amyn A. Malik, Kathy Kamath, Kerrie Greene, Valter Silva Monteiro, Mario A. Peña-Hernández, Tianyang Mao, Bornali Bhattacharjee, Takehiro Takahashi, Carolina Lucas, Julio Silva, Dayna McCarthy, Erica Breyman, Jenna Tosto‐Mancuso, Yile Dai, Emily S. Perotti, Koray Akduman, Tiffany J. Tzeng, Lan Xu, İnci Yıldırım, Harlan M. Krumholz, John Shon, Ruslan Medzhitov, Saad B. Omer, David van Dijk, Aaron M. Ring, David Putrino, Akiko IwasakimedRxiv 2022 · DOI

Introduction

SARS-CoV-2 infection can result in a constellation of persistent sequelae known as post-acute sequelae of COVID-19 (PASC), or Long COVID. Individuals affected by this condition frequently report debilitating symptoms, including unremitting fatigue, post-exertional malaise, and various cognitive and autonomic dysfunctions. Despite its prevalence, Long COVID has remained a "shadow" illness, lacking standardized diagnostic tests and a clear understanding of its underlying biological mechanisms.

In an exploratory, cross-sectional study, researchers utilized multi-dimensional immune phenotyping and machine learning to move beyond patient-reported symptoms and identify objective, measurable biomarkers. By comparing the biological profiles of Long COVID patients against various control groups, the study aimed to determine if a distinct "biological fingerprint" exists for the condition.

Study Design and Methodology

The study utilized the Mount Sinai-Yale Long COVID (MY-LC) cohort, which included 215 individuals categorized into four groups:

  1. Long COVID (LC): Individuals with persistent symptoms following infection.
  2. Convalescent Controls (CC): Individuals with prior infection who fully recovered.
  3. Healthy Controls (HC): Individuals with no prior SARS-CoV-2 infection.
  4. Healthcare Workers (HCW): A historical group of previously infected but healthy individuals.

To capture a comprehensive snapshot of the participants' biology—averaging over 400 days post-infection—the researchers employed a multi-omic approach. This included flow cytometry to map immune cell populations (T cells, B cells, and monocytes) and plasma proteomics to measure over 100 circulating proteins, including cytokines and hormones. Additionally, high-throughput platforms (REAP and SERA) were used to detect antibodies against thousands of human proteins and viral epitopes. Finally, machine learning methods, including LASSO regression and K-Nearest Neighbor (KNN) classifiers, were applied to identify the most predictive features of Long COVID status.

Key Biological Insights

The Cortisol Deficit

The most significant individual predictor identified by the machine learning models was a dramatic reduction in circulating cortisol levels. Participants with Long COVID exhibited median cortisol levels of 47.01 ng/mL, roughly half the levels found in healthy controls (90.32 ng/mL). As cortisol is a primary stress hormone responsible for regulating metabolism and inflammation, this state of "hypocortisolism" may provide a physiological explanation for the profound fatigue and malaise reported by patients.

Viral Reactivation and Immune Exhaustion

The study revealed unexpected increases in antibody responses directed against non-SARS-CoV-2 viral pathogens. Specifically, elevated responses to Epstein-Barr Virus (EBV) and Varicella-Zoster Virus (VZV) suggest that the initial COVID-19 infection may reactivate latent viruses dormant in the body.

Furthermore, Long COVID patients showed higher levels of "exhausted" T cells, marked by the expression of PD-1 and Tim-3. This suggests that the immune system remains in a state of chronic activation, potentially responding to persistent viral antigens or the aforementioned reactivated latent viruses. Notable shifts were also observed in myeloid populations, including an increase in non-classical monocytes, which are typically associated with vascular repair and anti-inflammatory responses.

Absence of Broad Autoimmunity

While some theories suggest Long COVID is driven by the body attacking itself, this study did not find a universal surge in broad autoantibodies across the Long COVID group. This indicates that while autoimmunity may play a role in specific individuals, it is likely not the primary driver for the entire patient population.

Diagnostic Potential and Machine Learning

By integrating these diverse immune features, the researchers developed machine learning models capable of distinguishing Long COVID patients from controls with high accuracy, achieving an Area Under the Curve (AUC) of 0.96. The study also introduced a "Long COVID Propensity Score" (LCPS) based on symptoms, which correlated strongly with the biological findings. This correlation validates that patient-reported outcomes are tied to measurable physiological changes, bridging the gap between clinical symptoms and biological reality.

Limitations and Future Directions

As a cross-sectional study, these findings represent a snapshot in time and cannot definitively prove that low cortisol or viral reactivation causes Long COVID; these features could be consequences of the chronic illness. Longitudinal studies are required to track how these markers evolve over time. Additionally, while the study was deep in its profiling, the sample size of 215 individuals necessitates validation in larger, more diverse cohorts.

Future research may also need to look beyond peripheral blood to tissue-specific samples (such as the gut or brain) to fully understand the localized impacts of the disease. Finally, while these findings suggest potential therapeutic avenues—such as cortisol replacement or anti-viral treatments—the authors emphasize that such interventions require rigorous clinical trials to ensure safety and efficacy.

Conclusion

This research provides a critical step toward defining the pathobiology of Long COVID. By identifying low cortisol, T cell exhaustion, and latent viral reactivation as key features of the disease, the study offers a foundation for developing objective biomarkers and targeted therapies for millions of patients worldwide.

Summary updated

Abstract

SARS-CoV-2 infection can result in the development of a constellation of persistent sequelae following acute disease called post-acute sequelae of COVID-19 (PASC) or Long COVID <sup>1-3</sup> . Individuals diagnosed with Long COVID frequently report unremitting fatigue, post-exertional malaise, and a variety of cognitive and autonomic dysfunctions <sup>1-3</sup> ; however, the basic biological mechanisms responsible for these debilitating symptoms are unclear. Here, 215 individuals were included in an exploratory, cross-sectional study to perform multi-dimensional immune phenotyping in conjunction with machine learning methods to identify key immunological features distinguishing Long COVID. Marked differences were noted in specific circulating myeloid and lymphocyte populations relative to matched control groups, as well as evidence of elevated humoral responses directed against SARS-CoV-2 among participants with Long COVID. Further, unexpected increases were observed in antibody responses directed against non-SARS-CoV-2 viral pathogens, particularly Epstein-Barr virus. Analysis of circulating immune mediators and various hormones also revealed pronounced differences, with levels of cortisol being uniformly lower among participants with Long COVID relative to matched control groups. Integration of immune phenotyping data into unbiased machine learning models identified significant distinguishing features critical in accurate classification of Long COVID, with decreased levels of cortisol being the most significant individual predictor. These findings will help guide additional studies into the pathobiology of Long COVID and may aid in the future development of objective biomarkers for Long COVID.
BibTeX
@misc{klein2022distinguishing,
  author = {Jon Klein and Jamie Wood and Jillian R. Jaycox and Peiwen Lu and Rahul M. Dhodapkar and Jeff Gehlhausen and Alexandra Tabachnikova and Laura Tabacof and Amyn A. Malik and Kathy Kamath and Kerrie Greene and Valter Silva Monteiro and Mario A. Peña-Hernández and Tianyang Mao and Bornali Bhattacharjee and Takehiro Takahashi and Carolina Lucas and Julio Silva and Dayna McCarthy and Erica Breyman and Jenna Tosto‐Mancuso and Yile Dai and Emily S. Perotti and Koray Akduman and Tiffany J. Tzeng and Lan Xu and İnci Yıldırım and Harlan M. Krumholz and John Shon and Ruslan Medzhitov and Saad B. Omer and David van Dijk and Aaron M. Ring and David Putrino and Akiko Iwasaki},
  title = {{Distinguishing features of Long COVID identified through immune profiling}},
  year = {2022},
  doi = {10.1101/2022.08.09.22278592},
  howpublished = {medRxiv},
}
Figure