A new Long COVID study lead by Yale Immunology Professor Dr Akiko Iwasaki, PhD that involved 215 individuals from Mount Sinai Hospital in New York City and Yale New Haven Hospital in Connecticut has interesting found that Epstein-Barr Virus (EBV) reactivation and low cortisol levels are common in many Long COVID patients.
The study which also had many other interesting findings is the first study using multi-dimensional immune phenotyping in conjunction with machine learning methods to identify key immunological features distinguishing Long COVID.
The study findings were published on a preprint server and is currently being peer reviewed.
Persistent sequelae are a prominent and debilitating consequence of infection with SARS-CoV-21, 3, 29. Our exploratory analyses identified key significant immunological differences relative to demographically matched control populations at >400 days post infection.
A number of significant changes in circulating leukocytes, including increases in non-classical monocytes, activated B cells, double-negative B cells, exhausted T cells, and IL-4/IL-6 secreting CD4 T cells, and decreases in conventional DC1 and central memory CD4 T cells were identified.
In addition, antibodies to SARS-CoV-2 antigens and herpesvirus lytic antigens were elevated in participants with Long COVID. In contrast, no significant differences were found for autoantibodies to human exoproteome. Most strikingly among participants with Long COVID, levels of plasma cortisol were roughly half of those found in healthy or convalescent controls.
Based on machine learning, cortisol levels alone were the most significant predictor for Long COVID classification, as well as for estimation of Long COVID Propensity Score. Multiple hypotheses have been proposed for Long COVID pathogenesis, including persistent virus/virus remnants, autoimmunity, dysbiosis, latent viral reactivation and unrepaired tissue damage 18, 32–38.
Our data suggest the involvement of persistent antigen, reactivation of latent herpesviruses, and chronic inflammation, and are less consistent with the autoantibodies to extracellular antigens.
Immune phenotyping of PBMC populations revealed notable elevations in circulating non-classical monocytes among the Long COVID group. Non-classical monocytes are frequently associated with anti-inflammatory responses programs; however, they are also engaged in maintenance of vascular homeostasis, Th1 anti-viral response polarization, regulation of immune complex deposition, and various chronic inflammatory and autoimmune conditions 39–41, 52.
Significant decreases in levels of circulating cDC1 populations were also observed among participants with Long COVID, which are classically associated with antigen presentation and cytotoxic T cell priming during viral infection 57, 58. In parallel with perturbations in circulating myeloid populations, significant reductions in the CD4+ TCM cells and elevations in exhausted CD4+ and CD8+ T cells were observed.
Analysis of intracellular cytokine production following stimulation displayed significant increases in the production of inflammatory IL-2 and IL-6 among CD4+ and CD8+ T cells, and a specific elevation of IL-4 CD4+ T cells. Unexpectedly, subsets of Long COVID participants also had polyfunctional IL-4 / IL-6 co-expressing CD4+ T cells which correlated with antibody reactivity against EBV lytic antigens, but not SARS-CoV-2 antigens. In aggregate, these findings are consistent with chronic immune engagement against reservoirs of viral antigen among participants with Long COVID.
With respect to humoral immunity, SARS-CoV-2 specific IgG against Spike and S1 were elevated in Long COVID participants as compared to vaccination-matched controls. Linear peptide profiling of antibody binding across SARS-CoV-2 Spike demonstrated both exaggerated magnitude and unique binding targets among participants with Long COVID, most notably at residues 682-690 comprising the furin cleavage site.
These findings are consistent with results from analysis of PBMC populations, such as TEX increases suggesting chronic immune responses directed against viral antigens within the Long COVID group. Intriguingly, circulating Spike protein has also been observed in Long COVID participants, but not in convalescent controls 53.
Furthermore, these findings are supported by prior reports of persistent viral antigen in intestinal biopsies of convalescent COVID-19 individuals, and suggest persistent antigen might drive the continuous elevation in antibody responses among people with Long COVID33–36.
In parallel to the analysis of humoral responses and circulating immune effector populations, systematic profiling of soluble immune mediators found numerous significant differences among MY-LC study groups. Participants with Long COVID demonstrated striking decreases in systemic cortisol levels – decreases which remained significant after accounting for differences in individual demographic factors and sample collection times.
This hypocortisolism was not associated with a significant perturbation in ACTH levels, suggesting an inappropriately blunted compensatory response by the hypothalamic-pituitary axis. The significance of the magnitude and prevalence of hypocortisolism in individuals with Long COVID is highlighted in that low levels of cortisol were also the dominant feature driving the accurate separation of Long COVID participants in machine learning models.
Prior reports have associated low cortisol levels during the early phase of COVID-19 in patients that develop respiratory Long COVID symptoms37. Thus, our current finding of persistently decreased cortisol production in participants with Long COVID more than a year following acute infection warrants expanded investigation.
Importantly, cortisol plays a critical role in mediating homeostatic stress responses and hypocortisolism shares substantial clinical overlap with Long COVID symptoms54. Low levels of cortisol have also been reported for myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS)55 and treatment with hydrocortisone is reported to elicit modest improvement in symptoms. However, adrenal suppression has ultimately precluded its widespread clinical use for this indication55 and additional clinical trials may be needed to optimize glucocorticoid replacement therapies for Long COVID and ME/CFS.
The multi-dimensional immune profiling of Long COVID participants also revealed elevated humoral immune responses to non-SARS-CoV-2 viral antigens, particularly EBV. EBV viremia was previously reported during acute COVID-1937, 56 in hospitalized patients and predicted the development of persistent symptoms in the post-acute COVID-19 period 37.
The observation of elevated IgG against EBV lytic antigens in this study suggests recent reactivation of latent herpesviruses (EBV and VZV) may be a common feature of Long COVID. Additionally, concordant analysis of EBV IgG responses by REAP and SERA found significant positive correlations between reactivity against EBV p23 and TEMRA and IL- 4+/IL-6+ CD4 T cell populations, as well as correlations between reactivity against EBV gp42 and IL-4+/IL-6+ double positive CD4 T cell populations among participants with Long COVID.
These results suggest that herpesvirus reactivation is not incidental following SARS-CoV-2 infection, and instead that non-SARS-CoV-2 viral pathogens may alternatively mediate, aggravate, or exploit the persistent changes observed in circulating immune effector populations. Whether EBV reactivation may also predispose people with Long COVID to the development or exacerbation of autoimmune pathologies, as has been recently reported for people with multiple sclerosis57, 58, will require extensive longitudinal monitoring and surveillance of people with Long COVID.
Extensive autoantibody profiling indicated that there were no stereotypical or shared extracellular autoantibodies that could differentiate participants with Long COVID from controls. Furthermore, there was no correlation between the degree of autoantibody reactivity and Long COVID propensity, nor was there any disproportionate targeting of functional pathways to distinguish Long COVID.
In context of prior hypotheses suggesting that autoantibodies may contribute to the pathogenesis of Long COVID18, 38, our results suggest they play a more limited role in disease pathology. However, the present results suggest autoantibodies may associate with particular features of Long COVID symptoms in subsets of affected individuals, such as tinnitus and nausea.
Given the generally sparse, private nature of the detected autoreactivities, a prospective study with additional statistical power will be required to validate and detect more subtle patterns in autoantibody reactivity amongst participants with Long COVID that could play a disease-modifying role. Additionally, whether autoantibodies may be associated with other adverse clinical outcomes following COVID-19 merits future study.
Finally, machine learning models identified multiple significant predictors of Long COVID status relative to convalescent and healthy control populations. While cortisol was the most robust individua predictor of Long COVID status, maintaining its excellent specificity for Long COVID diagnoses outside of the MY-LC study is unlikely given its known pleiotropic role in a variety of diverse disease processes.
Instead, it is proposed that a minimal set of soluble biomarkers identified in this study (decreased cortisol, increased IL-8 and galectin-1) may serve as more specific diagnostic biomarkers for Long COVID. Additionally, classification accuracy using solely immunological data obtained from Long COVID participants strongly agreed with Long COVID classifications using LCPS scoring (Cohen’s Kappa – 0.865), demonstrating that both PROs and immunological analyses are highly concordant in diagnosing Long COVID.
Importantly, this study has several limitations. Primary among these considerations is the relatively small number of participants that were extensively immunophenotyped. While broad in its coverage of diverse biological features, this study lacked the thousands of independent observations that traditional machine learning methods rely upon to robustly train and optimize classification models.
Instead, this study leveraged machine learning to parse hundreds of individuals, each with tens of thousands of data points, to identify a suite of candidate immunological features important in the accurate classification of Long COVID. Naturally this approach limits broad applicability without external validation, and the reported observations should serve primarily to guide and inform future studies investigating mediators of Long COVID pathogenesis.
Beyond sample size, this study also prioritized analysis of peripheral (circulating) immune factors from study participants. As Long COVID often presents with organ-system specific dysfunctions, a greater emphasis on analysis of local – as well as systemic – immune features would serve as a critical extension of the findings presented in the current study.
Further, our analysis of autoantibodies was restricted only to the exoproteome. Whether antibodies to intracellular autoantigens and non-protein autoantigens play a role in Long COVID pathogenesis was not tested. Lastly, analysis of cortisol levels among participants also displayed excellent specificity for a Long COVID diagnosis within the MY-LC study; however, the reported accuracies do not account for other disease processes that phenotypically resemble Long COVID or other disease processes where hypocortisolism is prominent.
In summary, significant biological differences have been identified between participants with Long COVID and demographically and medically matched convalescent and healthy control groups, validating the extensive reports of persistent symptoms by various Long COVID advocacy groups. Unbiased machine learning models further identified both putative biomarkers of Long COVID, as well as potential mediators of Long COVID disease pathogenesis. Our study provides a basis for future investigations into the immunological underpinnings driving the genesis of Long COVID.