Post-COVID-19 mental symptoms linked to ongoing support cell activity in the brain's emotion system

People with LC symptoms were prospectively recruited from the Neurocenter Outpatient Clinic of Turku University Hospital, Turku, Finland and from the Long COVID Outpatient Clinic at Helsinki University Hospital, Helsinki, Finland in 2021–2023. Inclusion criteria were age ≥ 18 years, neurological symptoms like fatigue or cognitive symptoms that interfered with work or study capacity, symptoms that had lasted for at least 2 months and were present at 3 months after the SARS-CoV-2 infection, and a confirmation of COVID-19 condition by a positive polymerase chain reaction (PCR) or antigen test during the acute infection phase. The exclusion criteria for LC participants were another condition causing symptoms associated with LC, pregnancy or breastfeeding, corticosteroid treatment within the past 4 weeks before PET/MRI, claustrophobia or history of severe anxiety or panic attacks, exposure to experimental radiation within the past 12 months, and intolerance to PET or MRI scans.

Age- and sex-matched HCs with no known neurological symptoms or diseases and patients with MS were used as control groups for imaging parameters and biomarker assays. HCs were prospectively recruited for this study from Turku University Hospital staff and members of the local community through staff networks (2021–2022). MS participants were originally recruited for other studies at the MS Neurocenter Outpatient Clinic of Turku University Hospital based on a confirmed MS diagnosis using current diagnostic criteria and willingness to participate in PET studies (2019–2022) and were retrospectively included in this study. The MS PET data presented in this study have not been used in other publications.

All study participants underwent PET and MRI scans and blood sampling. Clinical neurological examination was performed for individuals with LC for assessment of neurological symptoms and for patients with MS to assess the Expanded Disability Status Scale (EDSS) scores. The LC participants freely self-reported unstructured descriptions of their medical history, LC symptoms and hospitalisation details related to COVID-19 during a medical consultation and answered several mental health questionnaires.

Imaging parameters and biomarker assays were compared between the LC, MS, and HC groups to determine whether LC is associated with overall elevated neuroinflammation relative to a healthy reference group and a neuroinflammatory disease control. Mental health questionnaires were completed by LC participants only, while the HC and MS groups served exclusively as imaging and biomarker reference groups. Within the LC group, associations between questionnaire scores and imaging parameters were examined in an exploratory manner to evaluate the relationship between subjective symptom severity and neuroinflammation.

LC participants filled in Finnish versions of two quality of life questionnaires (European Health Interview Survey—Quality of Life [EuroHIS-8], RAND Corporation 36-item Short Form Health Survey [RAND SF-36]) [18, 19], two fatigue questionnaires (Fatigue Severity Scale [FSS], Modified Fatigue Impact Scale [MFIS]) [20, 21], anxiety (General Anxiety Disorder-7 [GAD-7]) [22] and depression questionnaires (Patient Health Questionnaire 9 [PHQ-9]) [23], insomnia (Insomnia Severity Index [ISI]) [24] and finally, health and disability questionnaires (WHO Disability Assessment Schedule 2.0 [WHODAS 2.0]) [25]. Questionnaire scores were calculated according to instructions as either the average of the total scores (EuroHIS-8, FSS), the sum of the total scores (MFIS, GAD-7, PHQ-9, ISI, and WHODAS 2.0) or the average of subsegments (RAND SF-36). For HC and MS participants, questionnaires were not obtained.

Serum NfL and GFAP concentrations were measured with a single molecule array assay (Neurology 2 plex B, Simoa® Technology, Quanterix, Billerica, MA, U.S.) from serum samples collected as previously described [26].

The PET/MR imaging was performed at the Turku PET Centre (Turku, Finland) with GE SIGNA™ PET/MR scanner (GE HealthCare, Chicago, IL, U.S.). Beginning at radioligand injection, a 30-min 3.0 T brain MRI and a dynamic 60-min whole brain [¹¹C]-PK11195 PET were performed simultaneously. MRI included axial T2, 3D fluid-attenuated inversion recovery (FLAIR) and 3D T1 sequences. The [¹¹C]PK11195 radioligand radiochemical synthesis was performed as previously described [27]. The mean (SD) injected radioligand was 443.5 (63.4) for HC, 445.4 (47.15) for MS, and 429.6 (56.3) for LC participants, with no significant differences between groups.

T1 MR images were used for auto segmentation of regions of interest (ROIs) with the FreeSurfer image analysis suite v7.2.0. (http://surfer.nmr.mgh.harvard.edu/↗). ROIs used in this study were the whole brain, white matter (WM), cortical gray matter (GM), amygdala, hippocampus, thalamus, putamen, pallidum, cingulate cortex, corpus callosum and brain stem. In the MS participants, the WM ROI was converted to a normal-appearing white matter (NAWM) ROI by subtracting manually curated MS lesion masks from it. For the LC and HC participants, the WM and NAWM ROIs were identical. ROI volumes were calculated from MR images based on segments created in FreeSurfer as described previously [27]. Intracranial volumes (ICVs) were calculated using SPM12. ROI volumes are reported as parenchymal fractions (PF), defined as the ratio of brain region parenchymal volume to total intracranial volume.

MR and PET image co-registration was performed using SPM12 (Wellcome Trust Center for Neuroimaging) running on Matlab 2017a (The MathWorks, Natick, MA, U.S.). PET images were reconstructed using Bayesian penalised likelihood algorithm Q.Clear that incorporates a time-of-flight model and point spread function correction. A penalization factor β = 350, 17 time frames, and voxel dimensions 1.172 × 1.172 × 2.78 mm (radial, tangential, axial) were used for reconstruction. Mutual information realignment in SPM12 was used to correct possible displacements between frames. PET had an approximate spatial resolution of 4.4 × 4.1 × 5.5 mm as determined by NEMA 2012 1 cm sphere standard test with ¹⁸F isotope. All images were resliced to the MRI voxel size of 1 × 1 × 1 mm. Specific binding of [¹¹C]PK11195 radioligand was quantified as distribution volume ratio (DVR) in different ROIs using the Logan method within a time interval of 20–60 min. A supervised clustering algorithm with four predefined kinetic tissue classes was used with the Matlab SuperPK software package to acquire the time activity curve corresponding to a reference region devoid of specific radioligand binding [28, 29]. The kinetic classes describing binding in normal GM (GM cortex), normal WM (centrum semiovale), and the blood activity (95% threshold mask from the sum from the first five PET frames) were defined using data from the HCs. The kinetic class for the high-binding GM was extracted from the core of the thalamus of the participants with secondary progressive MS.

Statistical analyses were conducted in R programming environment (R v4.3.2). Non-normality of data was determined using Shapiro–Wilk test (stats v4.3.2, shapiro.test), histograms and quantile–quantile plots (ggplot2 v3.5.1, geom_qq). For parameters non-normally distributed within any group, the median with first and third quantile limits (Q1–Q3) and Wilcoxon test (stats v4.3.2, wilcox.test) for group-wise comparisons is reported. Otherwise mean with standard deviation and two-sample Welch’s t-test (stats v4.3.2, t.test) is reported. In group-wise comparisons of sex proportions Fisher’s exact test (stats v4.3.2, fisher.test) is reported. Corrected p-values were adjusted for within-parameter multiple comparisons using Benjamini–Hochberg method (stats v4.3.2, p.adjust). For correlation analyses, Spearman correlation coefficient (ρ), 95% confidence interval and raw p-value are reported (psych v2.4.6.26, corr.test). Trendlines shown in correlation scatterplots are based on linear regression and are not derived from the associated Spearman correlation coefficients. Enrichment of significant correlations in limbic regions was assessed using an exact permutation test, enumerating all 165 possible combinations of 3 ROIs from the 11 examined. Analysis was restricted to non-dependent symptom scores WHODAS total, MFIS total, RAND-36 subscales, FSS total, GAD-7 total, PHQ-9 total, ISI total, and EUROHIS-QOL total to ensure independence of symptom measures. Fold enrichment was calculated relative to the null expectation of random proportional distribution across all 11 ROIs. Cronbach’s alpha test (psych v2.4.6.26, alpha) for covariance and Spearman–Brown corrected mean even–odd split-half reliability (psych v2.4.6.26, splitHalf) of all possible splits for internal reliability consistency among questionnaire replies were calculated for final and subsection scores across the corresponding individual questions. All tests were two-tailed and p-values under 0.05 were considered statistically significant for all analyses.

14 participants with LC symptoms, 11 HC participants and 13 MS participants were included in the study. Six of the 14 LC participants had been hospitalised during the acute SARS-CoV-2 infection but did not require intensive care. In the HC group, five participants had been diagnosed with and had fully recovered from COVID-19 and were free of any LC-like symptoms prior to study participation. The recovered controls did not differ from other controls in demographic, serum biomarker or PET parameters (Table S1). The MS group contained six relapsing–remitting MS (RRMS) and seven secondary progressive MS (SPMS) patients with a median disease duration of 20 years and a median EDSS of 3.5. Age and weight did not differ significantly between the LC, MS and HC groups. The mean BMI of the LC participants was higher compared to HCs, but the difference was not significant after multiple comparison correction (Table 1).

LC duration from SARS-CoV-2 diagnosis ranged from 7 to 33 (median 15) months at the time of imaging. The most commonly reported symptom was post-exertional malaise, defined as the worsening of symptoms after physical or mental exertion, reported by 11 (79%) LC participants. Fatigue and “brain fog” were also common, reported by nine (64%) participants (Fig. 1).

All 14 LC participants underwent a clinical neurological examination. Seven participants (50%) showed a fully normal examination, while the remaining seven exhibited some neurological abnormalities. Four (29%) participants had impaired olfaction—two (14%) of whom also experienced loss of taste. Four (29%) had decreased sensation in the lower extremities (mainly impaired sense of vibration), and one (7%) had difficulties in toe-and-heel walking, jumping on one leg and squat sitting. Additionally, one participant had dysmetria (7%), and another presented a bilateral Babinski’s sign (7%).

The symptoms reported by the LC participants, apart from anosmia, were rarely accompanied by corresponding clinical examination abnormalities. For example, of the five participants reporting weakness or numbness of the limbs, three exhibited clinical symptoms upon the neurological examination: one reported impaired sensation, one had dysmetria, and one reported impaired olfaction. The remaining two had a fully normal neurological examination.

EuroHIS-8, FSS, GAD-7, MFIS, ISI, PHQ-9, and WHODAS 2.0 questionnaire final scores are shown in Table 2. RAND-36 questionnaire scores by subsection, and subsection scores for the other questionnaires, are shown in Table S2. All questionnaire final scores were internally consistent at Cronbach’s α ≥ 0.70 and mean split-half reliability r ≥ 0.80.

The LC participant PHQ-9 and GAD-7 mean scores were 8.4 and 12.5, indicating mild depression and moderate anxiety, respectively. Two participants demonstrated moderately severe depression with a PHQ-9 score of 17. Three participants had a GAD-7 score of 15 or higher, indicating severe anxiety.

Questionnaire scores did not correlate with duration of LC, weight, BMI or age, except for correlations between older age and higher fatigue questionnaire scores (FSS ρ = 0.68, p = 0.022 and MFIS ρ = 0.78, p = 0.005).

There were no significant differences in [¹¹C]PK11195 DVRs between LC and HC participants in any examined brain region (Table 3; Fig. 2A–F). Compared to MS patients, LC participants had lower DVRs across most brain regions with significant differences in the putamen, thalamus, amygdala and pallidum. Compared to HC participants, the MS group had higher DVRs in the thalamus (1.26 vs. 1.20, p < 0.001), and SPMS patients (Table S5) additionally in the NAWM (1.08 vs. 1.05, p = 0.023) (Fig. 2G–H, Table S6).

LC participants imaged within 16 months of SARS-CoV-2 diagnosis (median LC duration 10 months, n = 8) had consistently higher DVRs in most regions compared to participants with longer time between diagnosis and imaging (median LC duration 26 months, n = 6) (Tables S7 and S8). The difference was significant for NAWM (1.05 vs. 1.02, p = 0.042) and trending to significance for the whole brain (1.08 vs. 1.06, p = 0.059) (Fig. 2I–J).

There were no significant differences between HC and LC participants in the volume parenchymal fractions of the examined brain regions (Table 4; Fig. S1A–F). The fractions were consistently higher in LC participants compared to MS patients, and the difference was significant in all brain areas except for the hippocampus and amygdala. MS parenchymal fractions of NAWM, thalamus, putamen, pallidum and brain stem were significantly lower than those in both LC and HC groups.

NfL serum concentrations showed no significant differences between any of the study groups (Table 1; Fig. 2K–L). No significant differences were found in GFAP concentration between LC participants and HCs. The MS patients had higher GFAP concentration compared to HCs (median 76 vs. 68 pg/ml, p = 0.043), but the difference did not remain significant after multiple comparison correction (p= 0.13).

There were several significant correlations between LC participant questionnaire scores and DVRs in deep gray matter structures thalamus, hippocampus and amygdala (Table 5, Figs. 3A–P and 4). Lower EuroHIS-8 scores indicating decreased quality of life correlated strongly with higher DVRs in all three regions (ρ = − 0.70, p = 0.03; ρ = −0.83, p < 0.01; ρ = −0.76, p = 0.01). Higher GAD-7 and PHQ-9 scores evaluating anxiety and depression correlated strongly with increased hippocampus and amygdala DVRs (ρ = 0.88, p< 0.01; ρ = 0.75, p = 0.01).

Of the fatigue questionnaires, an increased MFIS score, which includes questions about mental fatigue, correlated with higher DVRs in the hippocampus and thalamus. In contrast, the FSS questionnaire, in which mental and physical fatigue are not evaluated separately, did not correlate with any DVRs.

Overall health and disability questionnaire WHODAS 2.0 showed no correlation with DVRs in any region, but the cognitive section subscore correlated with increased hippocampal (ρ = 0.71, p= 0.02) and amygdala (ρ = 0.64, p = 0.04) DVRs (Tables S3, S4, Fig. 3C, G). Similarly, worse RAND-36 energy/fatigue-, pain- and emotional wellbeing subsection scores, but not general health subsection scores, correlated with higher DVRs in the hippocampus (Table S3, Fig. 3).

The sleep evaluation questionnaire ISI showed no significant correlation with DVRs in any studied region.

Across all comparisons, 16 of the 17 identified significant correlations were in the limbic system. A permutation enrichment analysis was used to assess whether this concentration exceeds the number expected by chance. The analysis was restricted to the questionnaire scores that were independent of each other by excluding questionnaire subsection scores. Of the 13 remaining correlations, 12 concentrated in the anatomically predefined limbic system, representing a more than threefold enrichment (permutation p = 0.006), inconsistent with a multiple-testing artefact. The one significant correlation outside the limbic system was between putamen DVR and RAND social functioning (ρ = − 0.70, p = 0.02) (Table S4).

The authors declare no competing interests.

This study involves human participants and was approved by the Ethics Committee of the Hospital District of Southwest Finland (reference numbers 56/1801/2021, 89/1800/2019 and 67/1801/2018). All study participants have provided written informed consent that covers the use of their data in this study. All procedures performed were in line with the principles of the 1964 Declaration of Helsinki and its later amendments.

Data are available on reasonable request. Anonymised data not published within the article will be shared on a request from a qualified investigator.