Preprint Immunological and non-immunological features specific to Long-COVID: multiple overlapping aetiologies and potential biomarkers, 2025, Ponchel

[forestglip]

Immunological and non-immunological features specific to Long-COVID: multiple overlapping aetiologies and potential biomarkers

Frederique C Ponchel

[Line breaks added]

Abstract
Long-COVID (LC) is a serious clinical condition characterised by debilitating fatigue together with a wide array of symptoms that significantly reduce the quality of life of patients. Currently no holistic or even symptom specific treatment options are available, likely due to both a lack of insight into the disease processes that drive LC symptoms and an extreme heterogeneity in patients’ profiles.

We characterised patients and post infection controls, with respect to their immunological profiles with a non-exhaustive panel of biomarkers rationalised based on their potential role in driving symptoms.

We observed that the patients’ symptoms could be grouped into 4 clusters suggesting possible stratification. Systemic inflammation persisted and did not normalize over time in LC. This was not related to persistent SARS-CoV-2 infection, as the presence of circulating N-protein was detected similarly in both patients and controls.

No obvious deviation in B-cells and monocytes profiles were observed with minor changes for NK-cells (CD62L+/CD16+/HLA-DR+). Major changes affected CD4+T-cells (and to a lesser extent CD8+T-cells) with respect to exhaustion (PD1+/LAG3+/CD44+), regulation (Treg) and differentiation (naïve/memory-CD62L+).

Several candidate biomarkers (cytokines, microRNAs, phosphate metabolism) were present more frequently in LC at high levels and provided information on underlying disease processes. While frequencies of candidate autoantibody+ participants were not different, levels of some antibodies were higher in LC. Yet none of these candidates stood out as a universal biomarker for LC, with the exception of CRP (73% cases), and loss of Treg (50%). However, we confirmed that several overlapping underlying aetiologies may be involved in this complex disease.

Specific groups of biomarkers also associated with the 4 cluster of patients. Although to be taken with caution due to small numbers, 3 biomarkers discriminated controls from patients (Treg/CD4+PD1+/CD4+CD161+), others were associated with symptoms recovery (low IL10/IL12/IL4 and high miR766) or deterioration (high CD4+CD38+/ CD8+naiveCD62L+/low IL2) over 12 months.

This study provides rational for developing targeted therapeutic strategies as well as biomarkers to stratify LC patients most likely to respond.

Web | PDF | Preprint: MedRxiv | Open Access

 

[Utsikt]

Statistical analysis​

Data exploration used univariate analysis (Mann-Whitney-U, Chi2). To assess collinearity between variables we used the Cluster-3 algorithm (Stanford University 1998-99) based on Spearman correlations on Log-transformed data. Results are displayed using heat maps of biomarker levels (TreeView). This algorithm is not relaying on statistical significance. Correction for multiple testing was not applied, and p-values (<0.05) were considered trends in most analysis. SPSS V29 and GraphPad Prism V9 were used. Logistic regressions to discriminate LC from NFC, is described in supplementary material.

Is this just a fishing expedition?

Only one author is also a bit unusual, but nothing to hold against them.

 

[SNT Gatchaman]

Only one author is also a bit unusual

That's the corresponding author. It's a consortium involving researchers in UK, Ireland, The Netherlands, France and US. Patient numbers were relatively small though.

One group (n=93) was recruited specifically for biomarker investigations in Leeds (REC-21/EM0112) via the dedicated Leeds long-COVID service/pathway set up in NHS-England. […] A second group (n=20) was recruited in Saint-Etienne from a regular chronic fatigue clinic (ID NCT05899595). University Staff who had fully recovered from COVID-19 without experiencing fatigue were included as a no-fatigue controls (NFC, n=41).

 

[V.R.T.]

Major changes affected CD4+T-cells (and to a lesser extent CD8+T-cells) with respect to exhaustion (PD1+/LAG3+/CD44+), regulation (Treg) and differentiation (naïve/memory-CD62L+).

Is this possibly relevant to the BTN2A2 hit in DecodeME?

others were associated with symptoms recovery (low IL10/IL12/IL4 and high miR766) or deterioration (high CD4+CD38+/ CD8+naiveCD62L+/low IL2) over 12 months.

High CD38 was associated with worsening- would that fit with what we were discussing in the dara threads?

I believe @DMissa said CD38 was down in his cell lines - does that put a damper on our speculations?

@JonathanEdwards what do you think about this paper? Do you know the author? They appear to be a rheumatologist or a researcher in rheumatology.

What about you @jnmaciuch? Does this appear to be a fishing expedition as suggested above?

 

[Jonathan Edwards]

@JonathanEdwards what do you think about this paper? Do you know the author? They appear to be a rheumatologist or a researcher in rheumatology.

I don't know the authors. The higher CD38 is of interest. As usual the paper is written with far too much interpretation as it goes along making it very difficult to read in any other frame of reference.

The patients had raised CRP, which is difficult to relate to ME/CFS, although Beentjes found a few raised CRP levels.

 

[jnmaciuch]

What about you @jnmaciuch? Does this appear to be a fishing expedition as suggested above

I’m having a hard time following the structure of the analysis, maybe I’m just too brain foggy today.

In general, when you’re in the initial stages of feature selection, it’s fine to not do p-value correction since the goal at that point is to try to isolate signal amongst the noise as much as possible. But when you’re getting to the point of making statements about which specific features are associated with your phenotype of interest, you do need to apply that stringency. Without that, it does become a fishing expedition, and a more unhelpful one at that if you’re just testing whether any features are associated with any outcome measures.