Brain and cognitive changes in patients with long COVID compared with infection-recovered control subjects

Abstract

Between 2.5% and 28% of people infected with SARS-CoV-2 suffer long COVID or persistence of symptoms for months after acute illness. Many symptoms are neurological, but the brain changes underlying the neuropsychological impairments remain unclear. This study aimed to provide a detailed description of the cognitive profile, the pattern of brain alterations in long COVID and the potential association between them.

To address these objectives, 83 patients with persistent neurological symptoms after COVID-19 were recruited, and 22 now healthy control subjects chosen because they had suffered COVID-19 but did not experience persistent neurological symptoms. Patients and controls were matched for age, sex and educational level. All participants were assessed by clinical interview, comprehensive standardized neuropsychological tests and structural MRI.

The mean global cognitive function of patients with long COVID assessed by Addenbrooke’s Cognitive Examination-III screening test [overall cognitive level (OCL_z) = −0.39 ± 0.12] was significantly below the infection recovered-controls (OCL_z = +0.32 ± 0.16, P < 0.01). We observed that 48% of patients with long COVID had episodic memory deficit, with 27% also with impaired overall cognitive function, especially attention, working memory, processing speed and verbal fluency. The MRI examination included grey matter morphometry and whole brain structural connectivity analysis. Compared to infection recovered controls, patients had thinner cortex in a specific cluster centred on the left posterior superior temporal gyrus. In addition, lower fractional anisotropy and higher radial diffusivity were observed in widespread areas of the patients’ cerebral white matter relative to these controls. Correlations between cognitive status and brain abnormalities revealed a relationship between altered connectivity of white matter regions and impairments of episodic memory, overall cognitive function, attention and verbal fluency.

This study shows that patients with neurological long COVID suffer brain changes, especially in several white matter areas, and these are associated with impairments of specific cognitive functions.

long COVID syndrome, neuroimaging, neuropsychological assessment, cognitive impairment, white matter integrity, SARS-Cov-2 infection

Introduction

Between 2.8% and 27.6% of subjects infected with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) do not fully recover.¹ Their post-viral syndrome, referred to as long COVID, is characterized by a variety of symptoms that include fatigue and cognitive difficulties (‘brain fog’). Long COVID describes persons with a history of probable or confirmed SARS-CoV-2 infection who maintain symptoms 3 months after infection for at least 2 months and cannot be explained by an alternative diagnosis.² The number of people living with long COVID worldwide is estimated at ∼65 million. The UK Office for National Statistics has reported that 1.9 million COVID-19 infected people in the UK may experience long COVID.³ Similarly, in Spain, ∼10% of 11 million COVID-19 sufferers may undergo persistent symptoms, i.e. ∼1 million.⁴ The consequential socioeconomical and health consequences need to be addressed responsibly for which accurate medical information is essential.

Long COVID may be newly onset after recovery from an acute episode of COVID-19 or may persist from the initial infection.² Regarding neurological symptoms, an association with disruption of the nervous system in the acute phase has been proposed, because of direct viral invasion, uncontrolled effects of peripheral inflammation resulting in impaired blood–brain barrier (BBB) function and neuroinflammation, peripheral organ dysfunction (lung, kidney and liver) and/or cerebrovascular endothelial injury.^5,6 Very recently, a prospective cohort study found an association between elevated fibrinogen and D-dimer during SARS-CoV-2 infection and the presence of cognitive deficits in the post-acute phase 6–12 months after acquiring COVID-19 infection. These elevated blood biomarkers could reflect COVID-19-associated coagulopathy causing cerebral microthrombi or pulmonary emboli leading to cerebral hypoxia.⁷

There is also evidence that cognitive deficits⁸ and brain changes⁹ may occur at the time of the acute infection. Executive dysfunction is commonly reported, especially in patients over 60 years of age with moderate-to-severe acute symptoms. In addition, infected patients have consistently demonstrated impairment in memory tasks.¹⁰ Moreover, reduced white matter integrity, assessed by diffusion tensor imaging (DTI), has even been observed 1 year after infection in specific areas, including the anterior cingulate cortex and thalamus.¹⁰ White matter hyperintensities, strategically located in the bilateral frontal subcortical and bilateral periventricular areas, have also been described in patients shortly after COVID infection.^11,12 Striking evidence for brain-related changes in COVID-19 comes from a large longitudinal study (pre-post infection) by the UK Biobank, with a sample of 401 patients who tested positive for SARS-Cov-2 infection and 384 uninfected controls.⁹ The comparative analysis of structural MRI data acquired before the pandemic and after infection showed a decreased cortical thickness in the olfactory and limbic areas together with cognitive impairment after, on average, 141 days post-infection.⁹

Regarding long COVID syndrome, several important studies have characterized the dysfunction as mainly involving attention and processing speed, episodic memory and executive functions.¹³ However, there is only a limited literature directly addressing the associations between cognitive deficits and brain changes in long COVID. Hypometabolism in the cerebellum has repeatedly been found to be associated with deficits in memory, attention and executive functioning between 1 and 6 months after COVID-19.^14,15 Another glucose PET study was unable to replicate such metabolic brain changes but, to the contrary, reported that increased cerebellar metabolism was specifically associated with more severe deficits in working memory and executive processing in long COVID patients.¹⁶ A multimodal neuropsychological/brain MRI study conducted on a Spanish sample of long COVID patients, but with only uninfected controls, found reduced functional connectivity, reduced corticolimbic grey matter volume and altered white matter signal. These changes were partially associated with alterations in cognitive performance, especially attention, processing speed and working memory in patients.¹⁷ However, with so few published studies on the subject of long COVID, further investigation seemed worthwhile.

In this context, a key issue concerns the appropriate control group. The present study compared patients with persistence of neurological symptoms more than 14 months after infection with a control group of healthy subjects who had also been infected with SARS-Cov-2 but did not experience neurological long COVID. Our aim was to assess cognitive performance with an extensive neuropsychological battery and look for associations with quantitative features of the neuroimaging profile. Given that neuropsychological and neuroradiological alterations have been reported to be associated with COVID-19 disease, the existence of differences between patients with long COVID and infection recovered controls not suffering from long COVID could more confidently be attributed to mechanisms underlying the persistence of neurological symptoms.

Materials and methods

Participants

This is a single-centre, cross-sectional study with clinical cases and controls. Patients who had acquired COVID-19 during the first wave (March–April 2020) and had persistent neurological symptoms for a period of more than 2 months and up to the time of evaluation were consecutively recruited trough the Department of Neurology at the University Hospital Complex of Albacete with inclusion criteria: (i) 18 years of age or over; (ii) clinical diagnosis of neurological long COVID (including evidence of past infection with COVID-19); (iii) agreement to undergo blood and CSF tests (reported separately), brain MRI and neuropsychological evaluation; and (iv) informed consent. Infection recovered control subjects matched for age, sex and education were recruited with the inclusion criteria as above, except (i) evidence of past COVID-19 infection (PCR or antigen positive); and (ii) absence of current COVID-19-related neurological symptoms. Exclusion criteria for all participants were: (i) diagnosis of another neurological, psychiatric or systemic illness that could explain the neurological symptoms; (ii) previous diagnosis of chronic fatigue or fibromyalgia; (iii) immunosuppression and/or corticosteroid treatment prior to the blood or CSF test; (iv) educational level below primary school level; and (v) absence of informed consent.

Ethics

This work was carried out in accordance with the Declaration of Helsinki for human research, the EU Regulation 2016/679, the Spanish the Organic Law 3/2018 on protection of personal data, and approved by the local Medical Research ethics committee (Acta 02/2021).

Neuropsychological assessment

The evaluation, performed by a neurologist and two trained neuropsychologists, began with a clinical interview about the medical history, acute COVID-19 infection symptomatology, symptoms that persisted and/or appeared, functional status of the patient, and demographic data.

Participants were subject to the cognitive screening Addenbrooke’s Cognitive Examination (ACE III). This provides scores for attentional, verbal fluency, visuospatial processing, language, learning and memory domains and the overall cognitive level (OCL).

This was followed by the vocabulary subtest from the Weschler as a measure of premorbid intelligence quotient. The protocol continued with a thorough assessment of long-term memory, executive functions, language, anxiety and depression (Supplementary Table 1). Briefly, language comprehension and production were assessed with the Boston Naming Test. Long-term memory with verbal list learning (Buschke Selective Reminding Test, FCRST), the Rey-Osterrieth Complex Figure and the Rivermead Behavioural Memory Test (RMBT). The RMBT assesses episodic memory with 14 subscales that tap into prospective memory, immediate and delayed free recall and recognition (visual and verbal) and spatial memory.

Executive functions were assessed with verbal phonologic and semantic fluency tests, the forward and backward Digit Span test, Corsi block-tapping test, Stroop Task, Trail Making Test, Symbol Search, Baddeley’s dual attention task, Wisconsin Sorting Card Test (presented by the software PsyToolkit) and the Tower of Hanoi. The assessment protocol for executive functions used the Spanish version by Tirapu et al.¹⁸

Three questionnaires were filled in by a partner or family/friend: (i) the Spanish version the Sunderland memory questionnaire regarding episodic/spatial memory during the past 3 months; (ii) educational level; and (iii) anxiety and depression (Supplementary Table 1).

Behavioural scores were transformed from standard/scale scores to typical z-scores using the standardized NEURONORMA for Spanish population and other normative data from the tests (Supplementary Table 1). Scores from the tests/versions adapted for this study were not standardized and therefore, patients scores were compared with controls with two independent samples t-test or chi square as appropriate.

Neuroimaging analysis

Participants were scanned with an Optima MR 450 W General Electric 1.5 T system with: (i) anatomical: SPGR 3D, repetition time: 8.5 ms, echo time: 3.2 ms, inversion time: 400 ms, thickness: 1 mm, gap: 0, isotropic matrix: 256 × 256 × 256, acquisition time = 7.28 min; (ii) diffusion: T₂-weighted fluid attenuation inversion recovery (FLAIR) imaging; and (iii) DTI: 20 directions, repetition time: 9425 ms, echo time: 103 ms, angle: 90, thickness: 3 mm, gap: 0, voxel size: 3 × 3 × 3, acquisition time = 7.25 min. DICOM images were converted to NIfTI format using the dcm2niix tool from MRIcron¹⁹ and organized in Brain Imaging Data Structure format (BIDS).^20,21

Brain grey matter was analysed with T₁ images in FreeSurfer (v.7.1.1).²² Briefly, the automated stream consists of motion correction and averaging, removal of non-brain tissue and skull stripping, automated Talairach transformation and intensity normalization, subcortical white and grey matter segmentation, spherical registration and cortical parcellation according to the Desikan-Killiany atlas.²³ Accuracy of automated segmentation was manually edited when necessary according to FreeSurfer guidelines. Cortical thickness was calculated as the average of the distance from the white matter surface to the closest point on the pial surface and from that point back to closest point on the white matter surface.²⁴ FreeSurfer’s general linear model (GLM) was used with a 10-mm full-width at half-maximum (FWHM) Gaussian kernel to compare cortical thickness between controls and patients and to correlate cortical thickness with neuropsychological variables. Age, gender and intracranial volume were treated as nuisance variables. Monte-Carlo simulation was used to correct for multiple comparisons.

Diffusion data processing was conducted with the FMRIB’s Diffusion Toolbox (FDT, FSL, FMRIB Software Library, Oxford, UK v.6.0).²⁵ We generated a brain mask with the FSL BET tool²⁶ with the non-diffusion-weighted (b0) DTI image. Images were registered to the (b0) image to minimize artefacts due to eddy currents distortions. Fractional anisotropy (FA), mean diffusivity (MD) and radial diffusivity (RD) values were calculated by using the FSL DTI-FIT to fit all the tensors. λ2 and λ3 values were averaged to obtain RD maps. The tract-based spatial statistics (TBSS)²⁷ group analysis protocol was used to perform non-linear registration of FA images to MNI standard space, and to generate a mean FA map and mean FA skeleton (centre of all tracts common to the group) thresholded at FA > 0.25. Each subject’s aligned FA data were then projected onto this skeleton to carry out voxel-wise statistics. We used a randomizing tool to correct for multiple comparisons. Threshold-free cluster enhancement (TFCE) was used to carry out non-parametric permutation-based statistical comparisons of patient versus control with FA, MD and RD maps (5000 permutations) with intracranial volume, age and gender as covariates. Spatial correlations with cognitive variables and FA were run for patients and controls separately. Significant differences were taken if P < 0.05 corrected.

Statistics

Statistical analyses (SPSS v.28) started with normality of data measured with Kolmogorov–Smirnov.

Demographic and clinical variables

Chi square, Mann-Whitney U-test or multivariate analysis ANOVA or ANCOVA were used for group comparisons, as appropriate.

Cognitive profile of impairment and severity of cognitive deficit

First, a one-sample t-test determined significant distances below the standard norms (reference mean of 0). Second, patients and controls were compared with two independent samples t-test. Participants were then classified into three different categories for each of the cognitive tests according with the guidelines of the American Academy of Clinical Neuropsychology²⁸ (Supplementary Table 2) as follows: (i) average score: z > −0.71 equivalent (eq.) to >24th percentile; (ii) low average score: z ≤ −0.71 eq. ≤ 24th percentile; and (iii) below average score: z ≤ −1.40 eq. ≤ 8th percentile. Scores z ≤ −0.71 eq. ≤ 24th percentile were taken as flag for possible cognitive deficit and z ≤ −1.40 eq. ≤ 8th percentile as significant cognitive impairment. We compared patients and controls in all the cognitive variables (Fig. 1A), but to identify potential phenotypes in terms of degree of cognitive deficit, we used the same classification criteria as before (i.e. above and below average) to build Fig. 1B.

Figure 1

Neuropsychological profile in long COVID is characterized by cognitive dysfunction especially in memory, attention and verbal fluency. (A) Differences in neuropsychological scores between patients and controls . (B) Differences in neuropsychological scores between patients with overall cognitive level (OCL) below low average ≤24th percentile (pc) and above (>24th pc) and controls. The Weschler’s Vocabulary subtest was taken as measure of premorbid IQ. See P-values in Table 3. The 24th percentile is indicated with dashed grey line in A and B. D = delayed; Dir. = direct; GMI = general memory index; I = immediate; RBMT = Rivermead Behavioural Memory Test; SD = standard deviation; TMT A, B = Trail Making Test versions A and B.

OCL (ACE III) correlated significantly with >70% of the tests and was therefore taken as a measure of general cognitive status. This dimension was used to identify severity of cognitive deficit and its potential association with physical factors during COVID-19 and long COVID. To determine this, physical symptoms were compared in two subgroups of patients: the unimpaired long COVID group (i.e. with above average OCL level z > 0.71, > 24th percentile) and another one with below average OCL (z ≤ −0.71; ≤ 24th percentile).

Multiple linear regression (backward, 95% confidence interval) was used to explore the association of the following independent variables: (i) medical antecedents; (ii) current medical conditions; (iii) COVID-19 symptoms; and (iv) long COVID symptoms with the dependent variables: cognitive dimensions for which patients obtained below average scores (z ≤ −1.41; equivalent to ≤ 8th percentile). Values of P ≤ 0.05 were considered significant for all analyses.

Results

Demographic and clinical characteristics

This study included 83 patients with neurological long COVID and 22 COVID-19 recovered controls not suffering from persistent neurological symptoms. Demographic variables are shown in Table 1. Patients and controls were similar in age (50.50 ± 2.59 versus 50.82 ± 1.03 years), educational level (years of schooling) and time (months) elapsing from acute infection to date of neuropsychological and MRI assessments (15 ± 2 versus 16 ± 1). The patient group had more females (71%) than the infection-recovered control group (36%), but a regression analysis showed no effects of sex. Almost all participants were vaccinated after their acute COVID-19 infection (pre-omicron) and showed similar vaccination rates prior to the cognitive assessment and MRI scan (patients 93%; controls 99%; X₁ = 0.15, P = 0.70). The time elapsing from vaccination to examinations was significantly longer in the control group (months, 13.8 ± 1.7) than patients (6.6 ± 1) [t(82) = 3.86, P < 0.001], but this period of time did not predict OCL as assessed with multiple regression [F(1,83) = 0.658, P = 0.419; Stand(β) = 0.089, t = 0.90, P = 0.419]. Patients and controls had similar medical comorbidities before acute COVID-19 infection (Supplementary Table 3). Moreover, 10 of 11 of solved antecedents and the current medical conditions showed no correlation with OCL. The exception was varicella zoster infection, which provided protection against cognitive deficit (comparison of the two group patients classified according to OCL; ≤ 24th percentile versus >24th percentile; X₁= 13.60 P < 0.001) (Supplementary Table 3).

Table 1

Demographic and clinical symptoms during the acute COVID 19 infection in relation to the cognitive outcome

	Controls n = 22	Patients n = 83	P	OCLa ≤ 24 pc (z ≤ −0.71) n = 22	OCL ≥ 24 pc (z > −0.71) n = 61	P
Demographics
Age, years, mean ± SEM	50.50 ± 2.59	50.82 ± 1.03	0.455	50 ± 2.06	51.13 ± 1.19	0.312
Age range, minimum−maximum	24–66	23–72	–	23–61	25–72	–
Sex, female n (%)	8 (36)	59 (71)	0.003*	16 (69.6)	43 (71.7)	0.850
Educational level scale 0–8, mean ± SEM	4.95 ± 0.53	4.73 ± 0.23	0.671	4.14 ± 0.45	4.96 ± 0.26	0.108
Days of sick leave, n (%)	2 (10)	57 (70)	0.001**	42 (51)	8 (13)	0.598
Months from the acute infection	15 ± 2	16 ± 1	0.674	15 ± 2	16 ± 1	0.551
Infection at home in isolation, n (%)	19 (86)	65 (78)	<0.001**	19 (83)	46 (77)	0.750
Hospitalization, n (%)	1 (5)	11 (13)	<0.001**	2 (9)	9 (15)	0.750
Intensive care unit, n (%)	1 (5)	7 (8)	<0.001**	2 (9)	5 (8)	0.750
Vaccinated, n (%)	21 (99)	70 (93)	0.700	19 (83)	51 (83)	0.513
Acute COVID 19 infection
Fever	8 (36)	59 (71)	0.009*	17 (77)	42 (69)	0.725
Asthenia	9 (41)	74 (89)	<0.001**	20 (91)	54 (89)	0.690
Arthralgia	2 (9)	20 (24)	0.167	5 (23)	15 (24)	0.756
Myalgia	1 (5)	38 (46)	<0.001**	13 (59)	25 (41)	0.224
Cough	5 (23)	25 (30)	0.651	6 (27)	19 (31)	0.620
Dyspnoea	2 (9)	51 (61)	<0.001**	12 (55)	39 (64)	0.283
Chest pain	0	16 (19)	0.033*	4 (18)	12 (19)	0.787
Other respiratory	2 (9)	14 (17)	0.447	3 (14)	11 (17)	0.563
Vomiting	1 (5)	14 (17)	0.177	6 (27)	8 (12)	0.165
Constipation	0	3 (4)	0.388	1 (5)	2 (2)	0.825
Diarrhoea	3 (14)	33 (40)	0.037*	12 (53)	21 (34)	0.152
Abdominal pain	1 (5)	8 (10)	0.510	3 (14)	5 (7)	0.515
Anosmia	5 (23)	55 (66)	<0.001**	16 (71)	39 (64)	0.694
Ageusia	5 (23)	54 (65)	<0.001**	16 (71)	38 (62)	0.594
Headache	7 (32)	58 (70)	0.004*	16 (71)	42 (69)	0.969
Other neurological symptomsb	0	16 (19)	0.033*	5 (23)	11 (17)	0.725
Insomnia	0	14 (17)	0.048*	4 (18)	10 (16)	0.937

pc = percentile; SEM = standard error of the mean.

^aOCL = overall cognitive level (ACE-III).

^bOther neurological symptoms include: hyperalgesia, neuropathic pain in limbs, dysautonomia, loss of sensation limbs, hand clumsiness, numbness in hands and/or fingers, facial hemiparesis, fasciculations, tinnitus, dysphonia, audition loss, vision loss, red eyes, dry eye, eye itchiness, eye pain, absences due to memory failures, spatial and time disorientation, nightmares, dizziness.

P-value refers to Chi-square for dichotomic variables or two-independent samples t-test for continuous variables, as appropriate. **P ≤ 0.01, *P ≤ 0.05.

Regarding the characteristics of acute COVID-19 infection, patients required significantly more days of sick leave and a higher percentage hospitalized in the intensive care than controls (Table 1). In addition, neurological complaints, headache, ageusia, insomnia, fever and some respiratory alterations were more commonly reported, with statistical significance, among patients with long COVID. Prevalence of these symptoms during acute infection did not differentiate patients with OCL below and those that scored above the 24th percentile (−0.71). However, a multiple linear regression analysis (backward) performed within the patient group showed that, amongst all the acute COVID-19 symptoms shown in Table 1, neurological symptoms, including memory failures, disorientation, dizziness, hyperalgesia, as well as ageusia during the acute infection was associated with poorer attention [ACE III: F(6,81) = 10.29, P = 0.003; Stand(β) = −0.33, t = −3.19, P = 0.002] and episodic memory [RMBT: F(6,81) = 2.42, P = 0.034; Stand(β) = −0.36, t = −2.21, P = 0.030], respectively.

Reported long COVID symptoms (Table 2) included memory failure and poor concentration (90%), asthenia (82%), headache (72%), anxiety (58%), limbs paresis (57%), insomnia (57%) and depression (46%). Some of the long COVID symptoms, such as anosmia, ageusia and dyspnoea, improved over the months, but asthenia, myalgia and headache were the most persistent (Supplementary Table 4). None of the long COVID symptoms predicted patients’ OCL. A multiple linear regression with long COVID symptoms showed that only insomnia was associated significantly with episodic memory deficit [RMBT: F(12,81) = 1.98, P = 0.043; Stand(β) = −0.29, t = −2.2, P = 0.014].

Table 2

Long COVID symptoms

Long COVID symptoms, n (%)	Patients n = 83	OCLa ≤ 24 pc (z ≤ −0.71) n = 22	OCL ≥ 24 pc (z > −0.71) n = 61	P
Memory failure	75 (90)	21 (95)	54 (89)	0.857
Poor concentration	74 (89)	21 (95)	53 (87)	0.697
Asthenia	68 (82)	19 (86)	49 (59)	0.597
Headache	60 (72)	15 (68)	45 (54)	0.406
Arthralgia	24 (29)	7 (32)	17 (20)	0.625
Limbs paresis	47 (57)	16 (72)	31 (51)	0.141
Other neurological symptomsb	31 (37)	14 (64)	17 (27)	0.006**
Depression	38 (46)	15 (68)	23 (37)	0.028**
Anxiety	48 (58)	15 (68)	33 (54)	0.399
Other mental health symptoms	6 (7)	2 (9)	4 (6)	0.749
Hair loss	7 (8)	1 (5)	6 (9)	0.407
Skin problems	8 (10)	1 (5)	7 (11)	0.312
Neck pain	37 (45)	10 (45)	27 (44)	0.901
Insomnia	47 (57)	14 (64)	33 (54)	0.629
Other persistent symptoms	33 (40)	9 (41)	24 (39)	0.942
Anxiety/depression (mean ± SEM)	17.53 ± 0.96	19.22 ± 1.86	15.19 ± 0.96	0.130

pc = percentile; SEM = standard error of the mean. P-values refer to two-independent sample t-test.

^aOCL = overall cognitive level (ACE-III).

^bOther neurological symptoms include: hyperalgesia, neuropathic pain in limbs, dysautonomia, loss of sensation limbs, hand clumsiness, numbness in hands and/or fingers, facial hemiparesis, fasciculations, tinnitus, dysphonia, audition loss, vision loss, red eyes, dry eye, eye itchiness, eye pain, absences due to memory failures, spatial and time disorientation, nightmares, dizziness.

It is important to note that the prevalence of affective disorders (anxiety or depression) prior to COVID-19 infection did not differ between patients and controls; as expected, however, long COVID patients exhibited higher rate and scores in anxiety and depression (Supplementary Table 3). These currents conditions did not correlate with cognitive impairment.

Neuropsychological findings

It is important to clarify that to more easily interpret the results from the neuropsychological assessment, patients (n = 83) were grouped into three categories based on the score and the corresponding percentile rank achieved for each evaluation as follows: patients presenting severe impairment (z-value −1.41, ≤ 8th percentile), relatively mild deficit (z ≤ −0.71, >8th ≤24th percentile) or average performance (z ≥ −0.70, ≥25th percentile) in the specific neurocognitive domain tested (Table 3, see more details in Supplementary Table 2 and ‘Statistics’ in the ‘Materials and methods’ section).

Table 3

Results from the neuropsychological assessment expressed in typical z-scores (mean ± SEM), whereby negative z-values indicate cognitive loss

Cognitive assessment	Control n = 22	Patients n = 83	Patients versus controls	≤8pc (≤−1.41z)	>8pc ≤24pc (>−1.40 ≤ −0.71z)	>24pc (>−0.71z)
Tests and cognitive dimensions	Mean ± SEM	Mean ± SEM	P	n (%)	n (%)	n (%)
Addenbrooke III (ACE-III)
Overall cognitive level	0.32 ± 00.16	−0.39 ± 0.12	0.002*	10 (12)	12 (15)	61 (73)
Visuospatial	0.35 ± 00.15	−0.15 ± 0.11	0.005*	16 (20)	10 (12)	57 (69)
Language	0.42 ± 00.06	0.22 ± 0.08	0.025*	2 (2)	7 (8)	74 (89)
Fluency	0.33 ± 00.18	−0.35 ± 0.13	0.002*	13 (16)	14 (17)	56 (68)
Memory	0.06 ± 00.23	−0.54 ± 0.14	0.022*	17 (21)	9 (11)	57 (69)
Attention	0.16 ± 00.24	−0.37 ± 0.17	0.040*	16 (19)	15 (18)	55 (66)
Premorbid IQ (WAIS III Vocabulary)	2.30 ± 0.19	1.60 ± 0.10	0.004*	2 (2)	1 (1)	81 (98)
Boston Naming test	1.30 ± 0.21	0.49 ± 0.14	0.004*	4 (5)	6 (7)	67 (81)
Long-term memory
REY figure copy speed	−0.03 ± 0.15	0.07 ± 0.10	0.308	4 (5)	6 (7)	72 (87)
REY figure copy precision	1.22 ± 0.29	1.38 ± 0.13	0.299	1 (1)	2 (2)	80 (97)
REY figure delayed free recall	0.41 ± 0.16	−0.33 ± 0.12	<0.001**	7 (8)	19 (23)	56 (68)
Retention index (FCSRT,16 words list learning)	0.97 ± 0.21	0.35 ± 0.15	0.011*	9 (11)	13 (16)	59 (71)
Free recall trial 1	0.40 ± 0.26	−0.13 ± 0.11	0.016*	2 (2)	21 (25)	58 (70)
Total recall	0.89 ± 0.30	0.02 ± 0.13	0.003**	7 (8)	8 (10)	66 (80)
Delayed free recall	0.90 ± 0.24	−0.04 ± 0.13	0.001**	9 (11)	8 (10)	64 (77)
Delayed total recall	1.70 ± 0.31	0.62 ± 0.20	0.007**	10 (12)	8 (10)	63 (76)
Total free recall	0.68 ± 0.23	−0.24 ± 0.43	0.152	7 (8)	5 (6)	69 (83)
RMBT episodic memory index	0.69 ± 0.20	−0.59 ± 0.13	<0.001**	20 (24)	20 (24)	42 (52)
Names and last names	0.22 ± 0.13	−0.27 ± 0.12	0.003**	11 (13)	18 (22)	54 (65)
Belongings	−0.10 ± 0.18	−0.62 ± 0.12	0.022*	14 (17)	17 (21)	52 (63)
Appointments delayed recall	0.30 ± 0.14	−0.19 ± 0.09	0.002**	4 (5)	11 (13)	68 (82)
Objects delayed recognition	−0.02 ± 0.19	−0.89 ± 0.14	0.002**	31 (37)	16 (19)	36 (43)
Story immediate	0.24 ± 0.16	−0.37 ± 0.08	<0.001**	3 (4)	25 (30)	55 (66)
Story delayed	0.24 ± 0.11	−0.36 ± 0.09	<0.001**	6 (7)	17 (21)	60 (72)
Faces delayed recognition	0.14 ± 0.22	−0.58 ± 0.13	<0.005*	20 (24)	10 (12)	53 (64)
Spatial route immediate	0.15 ± 0.13	−0.50 ± 0.13	<0.001**	18 (22)	6 (7)	59 (71)
Spatial route delayed	0.18 ± 0.12	−0.56 ± 0.13	<0.001**	17 (21)	11 (13)	55 (66)
Messages immediate memory	0.35 ± 0.02	−0.01 ± 0.9	<0.001**	7 (8)	7 (8)	69 (83)
Messages delayed memory	0.33 ± 00.02	−0.01 ± 0.9	<0.001**	6 (7)	9 (11)	69 (83)
Orientation and date	0.09 ± 00.16	−0.19 ± 0.10	0.071	10 (12)	6 (7)	67 (81)
Visuospatial puzzle immediate	0.48 ± 00.15	0.10 ± 0.11	0.023*	5 (6)	20 (24)	70 (84)
Visuospatial puzzle delayed	0.39 ± 00.08	−0.04 ± 0.11	0.005**	9 (11)	9 (11)	65 (78)
Every day/spatial memory	0.36 ± 0.25	−1.21 ± 0.26	<0.001**	26 (31)	8 (10)	29 (35)
Executive function
WAIS-III Digit Span Direct	0.13 ± 0.20	−0.29 ± 0.12	0.039*	17 (21)	3 (4)	63 (76)
WAIS III Digit Span Inverse	0.66 ± 0.20	0.10 ± 0.11	0.011*	4 (5)	4 (5)	75 (90)
CORSI blocks direct	1.60 ± 0.13	0.84 ± 0.11	<0.001**	1 (1)	3 (4)	79 (95)
Stroop	0.89 ± 0.20	0.39 ± 0.15	0.055	7 (8)	7 (8)	62 (75)
TMT A	0.14 ± 0.25	−0.31 ± 0.11	0.042*	8 (10)	15 (18)	60 (72)
TMT B	0.23 ± 0.24	−0.18 ± 0.11	0.052	8 (10)	11 (13)	63 (76)
WAIS III Symbol Search	0.61 ± 0.20	0.10 ± 0.10	0.013*	2 (2)	10 (12)	65 (78)
Phonologic verbal fluency	0.53 ± 0.19	−0.09 ± 0.10	0.003*	7 (8)	10 (12)	65 (78)
Semantic verbal fluency	0.96 ± 0.27	0.31 ± 0.11	0.006*	5 (6)	15 (18)	72 (87)

FCSRT = Free and Cued Selective Reminding Test; pc = percentile; RMBT = Rivermead Behavioural Memory Test; SEM = standard error of the mean; TMT = Trail Making Test; WAIS = Wechsler Adult Intelligence Scale.

P-value refers to independent samples t-tests. Significance is indicated as **P < 0.01; *P < 0.05.

The overall mean OCL of patients was significantly below standard norms and controls [OCL_z < 0, t(82) = −3.52, P < 0.001] (Table 3) with 27% in the impaired range (OCL_z ≤ −0.71; ≤24th percentile). The cognitive severity of these patients spanned from severe impairment (12%) to relatively mild deficit (15%) (Fig. 1A and Table 3).

Participants then underwent the extensive neuropsychological assessment to thoroughly analyse several individual cognitive domains. Patients achieved average scores relative to the norms (≥24th percentile) in ACE-III Language, premorbid IQ (WAIS-IV Vocabulary), Boston Naming Test and Rey Figure visual perception and visuomotor processing, but nevertheless, patients often obtained significantly worse test scores compared to controls (Table 3).

A key finding is that the domain most affected, and clinically relevant, was episodic memory and the memory index of patients evaluated through RBMT and the ACE-III tests, respectively. In the RBMT, patients performed poorly relative to the standard normal mean [General Memory Index (RBMT_z) < 0, t(82) = −4.64, P < 0.001]. Indeed, 48% of all patients scored below the 24th percentile in RBMT episodic memory (global index) and 50% of them exhibited a severe deficit (Fig. 1A and Table 3). Specifically, of the 14 RBMT subtests, eight were significantly below normal, including delayed object recognition [z < 0, t(83) = −6.47 P < 0.001], delayed face recognition (z < 0, t = −4.75, P < 0.001), delayed spatial route recall (z < 0, t = −4.18, P < 0.001) and delayed story recall [z < 0, t(83) = −3.89, P < 0.001]. Measures below average scores (≤−0.71; ≤ 24th percentile) in these RBMT subtests of recognition memory for objects and faces were registered in 36% and 56% of total patients, respectively. Very interestingly, severe impairment in these subtests (z ≤ −1.41; ≤ 8th percentile) was detected in 24–37% of the entire group of patients. Consistent with these observations, delayed visual free recall was impaired in patients [Rey Figure z < 0, t(82) = −2.73, P < 0.008]. The Memory index in ACE-III was also significantly below normal population level [memory index ACE III_z < 0, t(82) = −3.67, P < 0.001] with 21% in the severe impaired range. Delayed free recall was also lower in both the visual (Rey Figure memory) and verbal domains (FCSRT) (Table 3 and Fig. 1A). However, even though patients had lower rates in free recall in the 16 words list learning test (FCSRT), the retention index remained within the normal range. Family member ratings on patients’ episodic and spatial memory were below population level norms [t(62) = −4.72, P < 0.001] and rated as impaired in 31% of the patients.

In addition, it should be noted that specific cognitive functions and executive domains were significantly below the population norms and controls: attention [ACE-III z < 0, t(82) = −3.35, P < 0.001], verbal fluency [ACE-III z < 0, t(82) = −1.36, P < 0.006], processing speed [TMT-A z < 0, t(82) = −2.92, P = 0.005] and verbal working memory [WAIS IV Digit span z < 0, t(82) = −2.45, P < 0.017] (Table 3). These evaluations revealed that 24–34% of patients had some degree of deficit (≤24th percentile) and, more interesting, up to 20% of them showed severe impairment (≤8th percentile). However, patients’ z-scores obtained in other executive dimensions, including visuospatial working memory (Corsi Blocks), inhibition (Stroop) and some aspects of processing speed and attentional switch (WAIS IV Symbol Search) were above the 24th percentile, although evaluation values were slightly, but significantly, lower than those reached by controls (Fig. 1A and Table 3). The visuospatial index of the ACE-III scores were frequently (20%) below 8th percentile; however, on average, this score was below the control group but no different from the normal standard.

The assessment of higher order executive functions with tests that evaluate problem solving, divided attention and perseverance error [Tower of Hanoi, Baddeley dual task and Wisconsin Card Sorting Test (WSCT)] showed statistically significant difference for Tower of Hanoi performance in patients relative to that of controls (Supplementary Table 3). In addition, we noted that the subgroup of patients who scored below the 8th percentilein OCL assessed with ACE-III (12%, n = 10/83) were significantly lower in these cognitively demanding executive tasks (Supplementary Table 5). The Tower of Hanoi task with four disks and the Baddeley divided attention task were particularly impaired, with WCST perseverance error rate at borderline significance. These results indicate, therefore, that 88% of long COVID patients, who achieved measures above the OCL 8th percentile, did not differ from controls in the higher order executive tasks.

Finally, and with the objective of providing a detailed profile of cognitive deficit, we divided the patient group into two groups based on the z-score in the OCL (ACE-III) as follows: subgroup OCL_z > −0.71 or subgroup OCL_z ≤ −0.71, which is the equivalent to over/below 24th percentile (blue and red in Fig. 1B) and then analysed the performance in all the cognitive variables for each patient subgroup. As shown in Fig. 1B, this analysis highlighted a subgroup of patients, that corresponded to the patients who scored below 24th percentile in OCL (27%), with poorer cognitive outcome, specifically in executive function and episodic memory. Interestingly, even in the higher performing group with OCL_z > −0.71, who had close to normal scores in most cognitive functions, there was still a specific loss in episodic memory relative to the normative mean and control group levels, but not in executive function and retention of word lists (Fig. 1B).

Taken together, these results indicate that 48% of patients had selective cognitive impairment with a profile characterized by significantly poor episodic memory, including verbal and visual memory domains. These deficits were observed against a background of relatively preserved premorbid IQ, language, visuospatial perception and fine motor control.

Neuroimaging findings and association with cognitive outcome

The long COVID patients had significantly thinner cortex, relative to controls, in a cluster located in the posterior part of the left superior temporal gyrus (STG) (whole-brain vertex-wise analysis) (Fig. 2, STG, P < 0.01, in orange). This also extended towards the middle temporal gyrus and continued caudally towards the inferior temporal gyrus (P < 0.05, in yellow, Monte Carlo simulations correction). This thinning of cortex occurred along with white matter changes measured with diffusion DTI (lower FA and higher RD), indicative of poorer white matter integrity located just deep of the cluster of thinner cortex in the STG (Fig. 2). However, no differences in cortical thickness were observed between the subgroups of patients classified by degree of overall cognitive impairment (over or below 24th percentile), no correlations between cortical thickness and any of the cognitive outcome variables. Volume-based comparisons by FreeSurfer revealed no differences between patients and controls in any of the subcortical nuclei.

Figure 2

Differences in cortical thickness and underlying white matter between patients and controls in the left posterior superior temporal region. (A) Compared to the control group who had COVID-19 but no long COVID, the patients with long COVID had significantly thinner cortex in a cluster located in the posterior part of the left superior temporal gyrus. This thinning of cortex is shown with two different levels of statistical significance: light orange cluster (FreeSurfer whole-brain vertex-wise analysis, P < 0.05) and dark orange cluster (P < 0.01). (B) Poorer integrity of the white matter of patients with long COVID was found just deep to the thinner cortex of the caudal portion of the left superior temporal region as shown by the overlapped (purple) decreased fractional anisotropy (FA, red) and increased radial diffusivity (RD, blue) relative to the fully recovered COVID-19 control group. Comparisons have been corrected for intracranial volume, age and gender.

Significant differences between groups in FA (Fig. 3A, red) and RD (Fig. 3B, blue) in the white matter (P < 0.05, corrected for multiple comparisons) are shown in the skeleton (green). There were significantly lower FA values in the patient group in both hemispheres in the white matter underlying the dorsolateral, orbitofrontal and medial frontal cortices. In addition, decreased FA was observed in the cingulum bundle, rostrum and genu of the corpus callosum, uncinate fasciculus, superior and inferior longitudinal fasciculus including the white matter of the anterior part of the temporal lobe (temporal stem), parts of the arcuate fasciculus, splenium of the corpus callosum, and medial and lateral occipitotemporal white matter (Fig. 3A).

Figure 3

Loss of white matter integrity in the patient group relative to controls. (A) White matter regions with significant decreased fractional anisotropy (FA, red) in the patient group were observed in frontal regions, including the white matter in the medial frontal, anterior cingulum, dorsolateral and uncinate fasciculus regions. Superior and inferior longitudinal fasciculi, cingulum bundle, corpus callosum and occipital regions had also significant decreased in FA. (B) White matter regions with significant increased radial diffusivity (RD, blue) in the patient group. Note FA and RD overlap substantially, suggesting that the loss of FA could be associated, at least in part, to increased RD. Both FA and RD, after correction for intracranial volume, age and gender (P < 0.05).

None of the patients’ white matter areas showed significantly higher FA or lower RD relative to controls. Areas with significantly higher RD values (Fig. 3B, blue) overlapped with lower FA, but were spatially more distributed in the skeleton than those with lower FA.

While no correlations between FA and cognitive outcome were found in the COVID infection recovered control group, significant correlations of FA with OCL verbal fluency, attention and episodic memory (verbal and visual domains) were observed in the patient group (Fig. 4). Lower OCL was significantly associated with lower FA values in widely distributed areas of the white matter of the frontal lobe, thalamus, fornix, along the STG white matter and temporal stem, cingulum bundle, internal capsule, parietal and occipital white matter (Fig. 4A).

Figure 4

Spatial positive correlation in the patient group of FA values with neuropsychological scores. (A) Overall cognitive level (OCL) correlated with widespread decreased fractional anisotropy (FA) that distributed especially in the temporal lobe, subcortical white matter (internal capsule), parietal lobe, inferior and superior longitudinal fasciculus and corpus callosum. (B) Fluency associated decreased FA was even more widespread than with OCL and affected white matter in the frontal, temporal and parietal lobes in addition to the corpus callosum and subcortical white matter beyond the internal capsule into the diencephalic white matter. (C) Lower attention scores associated with lower FA in the corpus callosum, superior and inferior longitudinal fasciculus, internal capsule, anterior commissure, white matter of the posterior superior temporal gyrus, posterior inferior temporal gyrus, occipital white matter and brainstem. (D) Rey memory association with decreased FA had a similar widespread distribution to fluency, but unlike it, it affected more to the temporal lobe and less to the inferior parietal lobe. (E) Objects recognition association with decreased FA was more restricted compared with other functions, affecting primarily the inferior longitudinal fasciculus, including the temporal lobe white matter and the posterior occipital region. (F) Orientation date scores were associated with FA nearby subcortical regions, basal ganglia and diencephalon as well as caudal occipital and corpus callosum. All correlations have been corrected for intracranial volume, age and gender (P < 0.05).

Similarly, lower FA scores correlated significantly with lower measures verbal fluency (ACE-III, Fig. 4B) and attention (ACE-III, Fig. 4C) and measures of memory such as delayed visual free recall (Rey Memory, Fig. 4D), object recognition (Fig. 4E) and time and spatial orientation (Fig. 4F).

There was a substantial overlap of white matter areas associated with outcome in different cognitive domains; however, there was some specificity. For example, scores in attention were correlated with white matter of the STG, scores from the Rey memory test had a specifically high correlation cluster in the temporal stem, an area of white matter important for memory (Fig. 4D). Object recognition scores correlated with FA in the cingulum bundle, splenium of corpus callosum, right temporal stem and posterior optic radiations (Fig. 4E). Orientation in time and place correlated primarily with FA in the internal capsule and the splenium of the corpus callosum (Fig. 4F).

Discussion

This study reports brain changes associated with cognitive deficits in a cohort of 83 patients with neurological long COVID compared to 22 infection-recovered control subjects who did not experience persistent neurological symptoms. The key findings are: (i) 48% of patients with neurological long COVID were impaired in episodic memory, including visual and verbal, recognition and recall functions, and 27% of patients also showed deficits in attention, working memory, processing speed and verbal fluency; (ii) neurological complaints and ageusia (loss of taste) during the acute SARS-Cov-2 infection and insomnia observed during long COVID were associated with episodic memory deficits in long COVID; (iii) long COVID patients exhibited cortical thinning in a cluster located in the left STG and altered connectivity compared to controls; and (iv) reduced integrity of white matter regions distributed in frontal and parietal lobes, the STG, temporal stem and several fasciculi were found to be correlated with lower OCL, deficits in visual memory and orientation, as well as verbal fluency and attention.

The neuropsychological results are consistent with the findings of other major studies of long COVID, which reported that working and episodic memory, attention, verbal learning and executive skills are the most frequently altered domains.^13,17,29-35 It should be noted that our results also indicated two degrees of severity of cognitive impairment among our patients: 27% of patients with long COVID scored <24th percentile in overall cognitive level and, of these, 12% presented even lower values corresponding to the 8th percentile. The more impaired subgroup presented significant deficits in episodic memory (RBMT), attention and verbal fluency (ACE-III), processing speed (TMT-A) and verbal working memory (Digit span). In contrast, the less affected group also had impaired episodic memory but, interestingly, spared the former cognitive functions and word list learning. This finding points to the possible existence of two different levels of cognitive impairment in neurological long COVID. This possibility was also raised in a recent report, which found a variety of different cognitive domains affected, with attention and processing speed being the most impaired in patients with long COVID.¹³ In agreement with the rate of severe impairment observed in our patients with chronic neurological symptoms, a recent meta-analysis reported that 22% of patients with long COVID presented significant cognitive deficits.³⁶

Emerging evidence suggests an association between illness variables in the acute COVID infection, including respiratory symptoms, and the cognitive and functional impairments that persist in long COVID.³⁵ Furthermore, a previous study showed that even a mild COVID-19 disease might affect cognitive function.¹⁰ Specifically, clinical variables observed in acute and post-acute phase were correlated with the magnitude of later cognitive deficits. In particular, neurological symptoms and ageusia during infection was predictive of poor performance in both attention and episodic memory. In the same vein, dysgeusia and hyposmia have been associated with increased vulnerability to memory deficits in long COVID.¹⁰ As expected, we also observed a relationship between insomnia during the chronic phase and impaired episodic memory, whereas neither depression nor anxiety were statistically associated with cognitive dysfunction in line with previous studies.³³ Also consistent with a similar previous study,¹⁷ neither the timing of first vaccination nor days from first vaccine to cognitive assessment were related to cognitive outcome.

Regarding the neuroimaging findings, in a multimodal MRI analysis of patients with neurological manifestations at 11 months after infection, decreased thickness of the left STG was observed, along with thinning of the parahippocampal area and the anterior cerebellum.¹⁷ Impaired attention and processing speed were further found to be correlated with reduced cortical thickness of such cortical regions.¹⁷ In the present study, however, cortical thinning in the STG was not associated with any of the cognitive measures, but we did find a significant association between decreased integrity of white matter underlying this left posterior gyrus and verbal fluency and attention deficits. The cluster of left posterior STG includes functionally supramodal association areas, in addition to Wernicke’s language region, whose disruption could explain, in part, some of the high order processing deficits observed in our patients with long COVID.

In contrast to the relatively restricted localization of brain changes in cortical grey matter, white matter alterations were more diffuse and were significantly associated with poor cognitive performance. It is worth noting that a potential limitation of cross-sectional analyses (i.e. patients versus controls) compared with pre-post analysis (i.e. within-subjects) is the influence of individual differences that may result in a low statistical significance of data. Despite this drawback, a significant correlation between changes in white matter and specific cognitive impairment was observed in the patient group while no association was found between neuroimaging alterations and cognitive status in the COVID infection recovered control group. Similar neuroimaging results have been found in patients with anosmia³⁷ and persistent headache.³⁸ Another previous Spanish study of long COVID patients also reported alterations of diffusivity in several white matter areas.¹⁷ However, an association with cognitive dysfunction of modest significance was reported. In our work, on the contrary, we found that lower FA and higher RD values in specific regions of white matter were significantly correlated with deficits in OCL, verbal fluency, attention, visual memory, object recognition, orientation and date in the patient group compared to infection-recovered subjects who did not suffer from long COVID. These cognitive functions depend on the interaction between different cortical areas for higher order processing and the frontal cortex and diencephalon via important white matter tracts that have been found to be altered in the patients: uncinate fasciculus, cingulum bundle, corpus callosum, temporal stem and fornix. In keeping with this argument, recent structural MRI data have revealed that executive functions recruit areas beyond the frontal lobe including long association pathways, like the cingulum bundle and corpus callosum.³⁹

We are not in a position to add pre-COVID data to rule out the presence of pre-existing cognitive neuroanatomical alterations in patients with long COVID. However, we believe that the existence of brain changes prior to COVID infection in patients is unlikely because, if this were the case, there would have been a substantial proportion of patients who have sought medical advice, given the involvement of the affected areas in relevant cognitive functions. The only significant difference between the groups in their previous medical history and the clinical interview conducted during the study was the rate of varicella zoster infection, but this was in fact more frequent in the controls than in the long COVID cases. Therefore, our results on reduced integrity of extended white matter areas support the hypothesis that specific cognitive dysfunctions experienced by patients with neurological long COVID are associated with alteration of distal anatomical pathways critical for attention, memory, verbal fluency and overall cognitive processing. Furthermore, we propose that these brain and cognitive changes found in the patients could represent the specific signature of long COVID syndrome, given that they emerge from the comparisons conducted between patients with persistent neurological symptoms and the control group, which had COVID-19 but did not experience long COVID.

In our study, ∼90% of the patients had subjective cognitive complaints. This is in the line with the persistent (6 months) subjective poor concentration and attention (78%) and memory (n = 1800; 73%) reported by the Spanish Society of General Practitioners.⁴⁰ While these subjective complaints may exaggerate facets of cognitive dysfunction, COVID-19 patients 3–4 months after hospital discharge revealed that such complaints correlated significantly with objective measurement of global cognitive impairment.³⁵ In fact, 59% of the patients presenting cognitive complaints were found to have objectively mild deficits, but 17% had severe cognitive impairment, similar to our study. Thus, clinicians should suspect presence of cognitive impairments in almost half of patients with long COVID. From a clinical point of view, these findings may warrant implementation, in routine practice, of specific cognitive tests to detect that subgroup of patients with objectively affected cognitive domains. The ACE-III test, a brief cognitive test, could be used for this purpose.

Several neurobiological factors, which are not mutually exclusive, have been proposed to be involved⁴¹ in potential pathophysiological mechanisms underlying cognitive and brain abnormalities in long COVID. Such mechanisms include neuroinflammation that can be accompanied by BBB disruption, dysregulation of brain cells, autoimmunity, brain coagulation abnormalities, cerebral endothelial dysfunction, metabolic aberrations, hypoxia during acute phase, reactivation of latent herpes virus and even persistent infection of CNS.^5-7,41,42 Animal experiments have shown that the presence of microglial reactivity in the white matter together with elevated cytokines in the brain, including CCL11, lead to persistent impairment of hippocampal neurogenesis and myelin loss.⁴³ In agreement, in vivo analyses have revealed hippocampal volume loss associated with cognitive dysfunction in long COVID is accompanied by altered blood markers, which reflect inflammation (elevated CCL11) and neuronal and myelin damage.⁴² On the other hand, elevated markers of thrombosis detected in patients with long COVID at 6–12 months after infection, along with memory impairments, suggests that reduced oxygen delivery to the brain may play a role in such deficits.

To the best of our knowledge, this is the first study involving cognitive testing and neuroimaging from both patients with neurological long COVID and previously SARS-Cov-2 infected subjects who no longer present persistent neurological symptoms (our control group). It allows us, therefore, to argue that the neuropsychological and brain alterations found in our long COVID patients are specific to this syndrome, and responsible for chronic cognitive dysfunction. However, some limitations should be considered. First, our cohort, as was the case in previous long COVID studies, did not undergo brain MRI and neuropsychological examinations prior to COVID infection because neither patients nor controls had previously reported neurological complaints. The absence of clinical and neuropsychological antecedents would make it unlikely the pre-existence of the neuroimaging findings associated with deficits in episodic memory, attention and verbal fluency in patients with long COVID. Second, the occurrence of false negatives when comparing unbalanced groups, would represent another limitation that we addressed by analysing the neuropsychological outcomes compared to standardized norms, and not only with our control group. On the other hand, neuroimaging studies require a minimum number of subjects to be consistent/reliable, which this study fulfils. Another limitation was heterogeneity in the sample of patients in terms of hospitalization during acute COVID phase, although our statistical analysis revealed no differences in cognitive outcome between patients who had been hospitalized and those who overcame the infection at home. Furthermore, as this study is observational, as opposed to a randomized intervention study, we cannot make claims of causality of brain alterations. It has, however, been possible to determine the association between cognitive impairments and brain changes in patients with long COVID.

In conclusion, our data indicate a specific profile of cognitive dysfunction in neurological long COVID characterized by impairment in episodic memory, attention, processing speed and verbal fluency. This altered cognitive performance is associated with reduced integrity of specific white matter regions involved in the interconnection of distal regions responsible for these cognitive functions. Finally, physical symptoms and cognitive complaints in long COVID had previously been attributed to the psychological trauma associated with the aftermath of the health crisis, rather than impairment of the CNS. In contrast to that idea, our study supports the now increasing evidence for there being a diverse manifestation of neurological symptoms associated with cognitive alterations and brain changes.

Data availability

Data are available upon request.

Acknowledgements

We offer our primary thanks to the participants and their families for taking part in this study. In addition, Javier Tirapu helped in elaborating the neuropsychological protocol, and the multidisciplinary team at the Association of Acquired Brain Damage of Albacete (ADACE) provided generous assistance in the cognitive rehabilitation of participants in this study. Francisco Chillerón provided technical help with the brain scanner, and students Diego Torres and Victoria López of the Medical School conducted honours projects (final degree study) that formed part of the clinical data analysis.

Funding

This research was primarily funded by the Spanish Instituto de Salud Carlos III Ref. PI21/00010. Part of the PhD of Víctor Serrano del Pueblo and the technical work of Pepa Piqueras was funded via the European Regional Development Fund, Cohesion Fund and REACT-EU funds for regional development, B Strange, Polytechnic University, regional government (Madrid, Spain), and a Welcome Trust (Advanced Investigator Grant to RGMM).

Competing interests

The authors report no competing interests.

Supplementary material

Supplementary material is available at Brain online.

Author notes

These authors contributed equally to this work.