Brain functional connectivity alterations associated with neuropsychological performance 6–9 months following SARS‐CoV‐2 infection

Abstract Neuropsychological deficits and brain damage following SARS‐CoV‐2 infection are not well understood. Then, 116 patients, with either severe, moderate, or mild disease in the acute phase underwent neuropsychological and olfactory tests, as well as completed psychiatric and respiratory questionnaires at 223 ± 42 days postinfection. Additionally, a subgroup of 50 patients underwent functional magnetic resonance imaging. Patients in the severe group displayed poorer verbal episodic memory performances, and moderate patients had reduced mental flexibility. Neuroimaging revealed patterns of hypofunctional and hyperfunctional connectivities in severe patients, while only hyperconnectivity patterns were observed for moderate. The default mode, somatosensory, dorsal attention, subcortical, and cerebellar networks were implicated. Partial least squares correlations analysis confirmed specific association between memory, executive functions performances and brain functional connectivity. The severity of the infection in the acute phase is a predictor of neuropsychological performance 6–9 months following SARS‐CoV‐2 infection. SARS‐CoV‐2 infection causes long‐term memory and executive dysfunctions, related to large‐scale functional brain connectivity alterations.

default mode, somatosensory, dorsal attention, subcortical, and cerebellar networks were implicated. Partial least squares correlations analysis confirmed specific association between memory, executive functions performances and brain functional connectivity. The severity of the infection in the acute phase is a predictor of neuropsychological performance 6-9 months following SARS-CoV-2 infection. SARS-CoV-2 infection causes long-term memory and executive dysfunctions, related to large-scale functional brain connectivity alterations. refers to a multisystem condition that occurs in individuals with a history of probable or confirmed SARS-CoV-2 infection, usually 3 months after onset of COVID-19, with symptoms that last for at least 2 months and cannot be explained by an alternative diagnosis. To date, at least 52 clinical or biological signs have been listed (Tran et al., 2021), impacting eight different systems: pulmonary, cardiovascular, hematological, renal, endocrine, gastrointestinal, dermatological, and neuropsychiatric (Nalbandian et al., 2021).
This constellation of symptoms persists well after the acute phase of the infection and includes cognitive disorders (for a review, see Vanderlind et al., 2021). Observations suggest impairment of various cognitive functions up to 3 months following COVID-19, with disruption of global cognitive efficiency Amalakanti et al., 2021;Beaud et al., 2021;Blazhenets et al., 2021;De Lorenzo et al., 2020;Ferrucci et al., 2021;Kas et al., 2021;Negrini et al., 2021;Ortelli et al., 2021;Pirker-Kees et al., 2021;Pistarini et al., 2021;Raman et al., 2021;Solaro et al., 2021;Udina et al., 2021), memory functions (Almeria et al., 2020;Hampshire et al., 2021;Jaywant et al., 2021;Whiteside et al., 2021;Woo et al., 2020), attention Almeria et al., 2020;Hampshire et al., 2021), executive functions Tay et al., 2021;Whiteside et al., 2021;Woo et al., 2020), logical reasoning (Hampshire et al., 2021), and language Almeria et al., 2020;Whiteside et al., 2021). The etiopathogenesis of these disorders remains subject to debate, but three hypotheses have already been postulated. To date, the most plausible according to the literature seems to be an indirect/mediated damage may result from an excessive immune or inflammatory reaction. This is supported by evidence of hyperinflammation with features of cytokine storm syndrome (Cron et al., 2021), and by studies showing a link between neuropsychiatric symptoms and immune data , as well as recent evidences from wide histopathological cohorts, suggesting an extensive glia activation and infiltration of CD4/8pos lymphocytes within the perivascular spaces (Matschke et al., 2020;Schwabenland et al., 2021;Thakur et al., 2021). That said, all three hypotheses can be supported by positron emission tomography (PET) studies revealing patterns of hypometabolism in the olfactory, frontal and limbic systems (Delorme et al., 2020;Guedj et al., 2021;Hosp et al., 2021). There is also the potential impact of the post-resuscitation / intensive care unit (ICU) syndrome in patients whose symptoms were sufficiently severe to require such treatment. Cognitive deficits after ICU, associated with mechanical ventilation, have been demonstrated in other pathologies and are increasingly recognized (Jackson et al., 2007;Kohler et al., 2019) (for a recent review, see Sakusic et al., 2018). Interestingly, this review (Sakusic et al., 2018) found that the factors that predicted impaired cognition and structural brain damage after hospitalization in ICU were delirium and its duration. Based on these reviews, medication (sedatives and analgesics), mechanical ventilation, extracorporeal membrane oxygenation, trophic feeding, intraoperative hypotension, and hypoxia appear not to influence the likelihood of long-term cognitive impairment (Sakusic et al., 2018).
The impact of respiratory severity in the acute phase of COVID-19 on chronic neuropsychological symptoms has yet to be clarified/determined. Nevertheless, some studies using validated neuropsychological testing approaches to explore the consequences of SARS-CoV-2 infection have shed some light on this issue. For example, Woo et al. (2020) and Almeria et al. (2020) compared patients who benefited from oxygen therapy with those who did not. Woo et al. (2020) found no differences, whereas Almeria et al. (2020) reported significant differences on verbal memory, visual memory, working memory, processing speed, executive function, and global cognition. Reduced performances for executive functions were only observed in ICU patients. Alemanno et al. (2021) observed better cognitive scores among patients who had been under sedation and ventilated in the ICU, compared to patients who had been hospitalized without oxygen therapy. Nevertheless, the presence of methodological limitations reduces the extent to which inferences can be drawn about the potential impact of respiratory severity in the acute phase on chronic neuropsychological deficits. Moreover, only a small number of studies have simultaneously assessed chronic neuropsychological symptoms and carried out neuroimaging. In particular, to date, few studies have investigated functional connectivity in patients in longterm following SARS-CoV-2 infection (>3 months postinfection), or only in the acute phase (Benedetti et al., 2021;Esposito et al., 2022;Fischer et al., 2022;Yildirim et al., 2022) and considering psychiatric (Benedetti et al., 2021) or olfactory symptoms (Esposito et al., 2022;Yildirim et al., 2022). Nevertheless, Fu et al. (2021) identified pattern of functional connectivity, associated with post-traumatic stress disorder symptoms, revealing modifications in the sensorimotor and visual networks. Zhang et al. (2022) focused on analysis of intraconnectivity and interconnectivity of the default mode network (DMN) and revealed a higher interconnectivity of the DMN in patients reporting long-term symptoms following SARS-CoV-2 infection. To date, no study has assessed brain functional connectivity in relation with neuropsychological performances as function of the severity of the acute infection.
In this context, the objective of the present study was to test whether differences in neuropsychological performances at 6-9 months postinfection were associated with modifications in functional brain networks, considering the severity of the respiratory symptoms in the acute phase. To this end, patients without clinical history that could induce neuropsychological deficits prior to infection with SARS-CoV-2 underwent a comprehensive assessment that probed multiple cognitive domains, emotion recognition, psychiatric symptoms, dyspnea, and olfaction. They were divided into three groups according to the respiratory severity of the disease in the acute phase: severe (ICU hospitalization; n = 24), moderate (conventional hospitalization; n = 42), and mild (no hospitalization; n = 44). Of these patients, 50 agreed to undergo MRI, for which structural visual and functional connectivity analyses were performed.
In view of our objectives, we developed two hypotheses. First, we expected differences in neuropsychological performances and modifications of the cerebral functional connectivity to be a function of disease severity in the acute phase (Hampshire et al., 2020), although moderate and mild patients might also exhibit deficits Woo et al., 2020). Second, based on previous observation of altered connectivity patterns in the long-term following SARS-CoV-2 infection (Fu et al., 2021;Zhang et al., 2022), we suspected that relationships between neuropsychological scores and changes in functional brain connectivity could be observed as a function of severity.

| Participants
Patients were selected among all the patients from the Geneva University Hospitals (HUG) that showed evidence of a SARS-CoV-2 infection (between March 2020 and May 2021) either by positive polymerase chain reaction (PCR) from nasopharyngeal swab and/or by positive serology while being included according to the exclusion criteria (see below). Patients were divided into 3 groups and included to the study at 223.07 ± 41.69 days postinfection: 24 patients who had been admitted to ICU during the acute phase of the infection (severe), 42 patients who had been hospitalized but did not require mechanical ventilation (moderate), and 44 patients who had tested positive but had not been hospitalized (mild). Of these patients, 50 agreed to undergo MRI scans (severe: n = 9, moderate: n = 21, mild: n = 20) (see Table 1).
The required number of participants in each group was determined by a power analysis involving the comparison of two means.
This analysis was based on the literature evaluating the short-term neuropsychological effects of COVID-19 in mild patients (Woo et al., 2020). To achieve the desired statistical power (1 À β) of 90% and risk of Type I error (α) of 0.05, results indicated that for a onesided hypothesis, 13 participants would be needed in each group and for a bilateral hypothesis 18. As we planned to perform nonparametric analyses, we had to increase the sample size by 15% (Lehmann, 2012), resulting in a minimum of 15 participants per group in the case of one-sided hypothesis and 21 participants per group in the case of bilateral hypothesis.
The mild and moderate groups were matched during the screening-inclusion process to the severe group for median age (mild = 57.50 years; moderate = 56.50 years; severe = 60 years), sociocultural level, and clinical variables (except for chronic renal failure) due to a limited number of available patients who were in ICU and met our exclusion criteria. Participants (n = 50) who underwent MRI were not matched during the screening-inclusion process, and all patients that agreed for the MRI study were included. Nevertheless, the groups were still comparable on sociodemographic characteristics (except gender) and severity. Participants were recruited via CoviCare program (Nehme et al., 2021) following patients with post-COVID symptoms in Geneva, Switzerland (MN, OB, and IG), as well as from registers from another study (LB). For each patient, we carried out a medical file review, followed by a telephone call inviting the patient to take part in the study, if all the eligibility criteria were met. Exclusion criteria were a history of neurological issues, psychiatric disorders (two of the included participants had had an episode of depression more than 10 years before their SARS-CoV-2 infection), cancer (to exclude possible chemotherapy-and radiotherapy-related cognitive impairment (Cascella et al., 2018)), neurodevelopmental pathologies, pregnancy, and age above 80 years (see Figure 1).

| General procedure and ethics
A flowchart displaying the successive stages of the study according to the eligibility criteria for each experimental group is provided in Figure 1.
After being given a full description of the study, participants provided their written informed consent. The study was conducted in accordance with the Declaration of Helsinki, and the study protocol was approved by the cantonal ethics committee of Geneva (CER-02186).

| Neuropsychological assessment and other clinical outcomes
The experimental design and tests used are comparable to those used in a previous published study .

| Executive functions
The Stroop task, Trail Making Test, and categorical and lexical verbal fluency from the GREFEX battery (Roussel & Godefroy, 2008) were administered to evaluate inhibition, shifting, and updating, in accordance with Miyake et al. (2000). Verbal working memory and visuospatial working memory were assessed with the backward digit span (Drozdick et al., 2018) and backward Corsi tests (Kessels et al., 2000).
We also administered computer-based tasks designed to gauge focused attention, divided attention, phasic alertness, working memory, and incompatibility, using version 2.1 of the Test for Attentional Performance (Zimmermann & Fimm, 2002).

| Memory systems
Short-term memory was assessed with forward digit spans (Drozdick et al., 2018) and the Corsi test (Kessels et al., 2000). Verbal episodic memory was assessed with the 16-item Grober and Buschke free/cued recall (RL/RI 16) paradigm (Grober & Buschke, 1987), as it distinguishes between the cognitive subprocesses of encoding, storage, and recall ( Van der Linden et al., 2004). Visual episodic memory was assessed with the delayed recall of the Rey-Osterrieth Complex Figure test (Meyers & Meyers, 1995

| Instrumental functions
Language was assessed with the BECLA battery (Macoir et al., 2016), ideomotor praxis with a short validated battery (Mahieux-Laurent et al., 2009), visuoconstructive abilities with the Rey-Osterrieth Complex Figure test (Meyers & Meyers, 1995), and visuoperceptual functions with four subtests from the Visual Object and Space Perception battery (Warrington & James, 1991) that measured object perception (fragmented letters, object decision) and spatial perception (localization of numbers, analysis of cubes).

| Logical reasoning
This was assessed using the Puzzle and Matrices subtests of the Wechsler Adult Intelligence Scale-Fourth Edition (Wechsler, 2008 (Roth et al., 2005). To quantify anosognosia, we calculated a self-appraisal discrepancy (SAD) score for each memory and executive domain evaluated by the QPC and BRIEF-5 (Leicht et al., 2010;Rosen et al., 2010;Tondelli et al., 2018). First, we calculated standardized scores for the cognitive complaints, dividing the raw scores of the self-report questionnaires into four categories: 0 = normal behavior, 1 = limited influence on daily life, 2 = noticeable influence on daily life, and 3 = substantial influence on daily life. We then subtracted each of these standardized scores from the standardized score for the relevant function. For example, if a patient reported no memory disorders (QPC score = 3), but performed very poorly on the RL/RI 16 delayed free recall test (score = 0), he or she would be deemed to exhibit anosognosia for memory dysfunction: 0-3 = À3. SAD scores could therefore range from À3 to 3, and any score below 0 indicated anosognosia.

| Other clinical outcomes
We collected patients' sociodemographic data and medical history.
Psychiatric data (including current fatigue, insomnia, and somnolence), dyspnea, and data on olfactory abilities at the time of the interview were also collected. Finally, a neurological assessment of CNS and peripheral nervous system functions and walking was carried out by two certified neurologists (FA and GA). Inventory-Second edition (Beck et al., 1996), anxiety with the State-Trait Anxiety Inventory (Spielberg et al., 1993), apathy and its distinct subtypes with the Apathy Motivation Index (Ang et al., 2017), PTSD with the Posttraumatic Stress Disorder Checklist for DSM-5 (Ashbaugh et al., 2016), manic symptoms with the Goldberg Mania

| Sociodemographic and clinical data
Inventory (Goldberg, 1993), dissociative symptoms in the patient's daily life with the Dissociative Experience Scale (Carlson & Putnam, 1993, current stress perception with the Perceived Stress Scale -14 items (Lesage et al., 2012), cognitive reappraisal of an emotional episode and expressive emotional suppression abilities with the Emotion Regulation Questionnaire (Gross & John, 2003), and susceptibility to others' emotions with the Emotional Contagion Scale (Doherty, 1997). Finally, fatigue was assessed with the French version of the Fatigue Impact Scale (Debouverie et al., 2007), potential sleeping disorders with the Insomnia Severity Index (Morin, 1993), and symptoms of sleepiness in daily life with the Epworth Sleepiness Scale (Johns, 1991).

Motion-correcting transformations, BOLD-to-T1w transformation and
T1w-to-template (MNI) warp were concatenated and applied in a single step using antsApplyTransforms (ANTs v2.1.0), configured with Lanczos interpolation. Framewise displacement (Power et al., 2014) was calculated for each functional run using Nipype and volumes with a framewise displacement greater than 0.7 mm were excluded (SI 3).
Many internal operations of fMRIPrep use Nilearn (Abraham et al., 2014), principally within the BOLD-processing workflow. For more details of the pipeline, see the section corresponding to workflows in the fMRIPrep documentation.
In addition, the preprocessed fMRI timeseries were detrended and the first five lowest frequency basis of the discrete cosine trans- 2.6.5 | Neuroimaging statistical analysis

Structural MRI inspection
First, the neuroimaging data were visually analyzed to look for noticeable brain lesions such as microbleeds and WM damages. Groups were compared on the total number of microbleeds and impact on WM, with the Wahlund scale (Wahlund et al., 2001). Second, voxelbased morphometry (VBM) analyses (Ashburner & Friston, 2000;Mechelli et al., 2005) were performed by computing the proportion of grey and WM voxels within the whole brain mask or within the fMRI parcellation (see below) and by comparing the outcome between the groups. Statistical differences were assessed with ANCOVA and considering age, gender and sociocultural level as covariates.

fMRI statistical analysis
The processed functional time courses were averaged into 156 regions of interest (100 cortical regions (Schaefer et al., 2018) that can be associated with 17 resting-state networks (Yeo et al., 2011), 34 cerebellar regions (Diedrichsen et al., 2009) and 22 subcortical regions (Amunts et al., 2013)) to perform functional connectivity analyses considering the whole brain. Measures of functional connectivity were converted into z scores with the Fisher z transformation and compared using two-sample t tests to investigate differences between groups. The normality of functional connectivity measures was con- group comparison were considered in this analysis along with the whole brain functional connectivity. Three PLSC analyses were conducted. First, the data were observed in the whole group to identify general associations between behavioral and neuroimaging data.
Then, a group-PLS analysis was performed considering the group based on the severity. Finally, we repeated the analyses within each individual group to confirm the results from the group-PLS approach.

| Neuropsychological symptoms as a function of disease severity
The three groups differed significantly on (

Memory encoding
Moderate patients scored significantly higher on the RL/RI 16-Immediate recall than severe patients after FDR correction (z = À2.43, p = .015), but the other two pairwise comparisons were not significant after FDR correction.

Long-term episodic verbal memory
Mild patients scored significantly higher on the RL/RI 16-Sum of three free recalls than severe patients after FDR correction (z = À2.95, p = .003), but the other two pairwise comparisons were not significant after FDR correction. Mild patients scored significantly higher on the RL/RI 16-Delayed free recall than severe patients after FDR correction (z = À3.26, p = .001), but the other two pairwise comparisons were not significant after FDR correction (see Figure 2).

| Structural MRI results as a function of disease severity
No substantial structural damage could be observed. The intergroup structural analysis failed to reveal any significant differences between groups on WM lesions using the mean score on the Wahlund scale.
Concerning microbleeds, a single patient had two microbleeds, 18 patients had one microbleed, and 25 had no microbleeds. A significantly higher proportion of mild (55%) patients had at least one microbleed, compared with the moderate (18.50%) and severe (12.50%) patients (see SI 5 and 6). First of all, our results support previous reports of cognitive deficits in the absence of structural brain lesions in COVID-19 (Khoo et al., 2020;Manganelli et al., 2020;Mohamud et al., 2020;Pilotto et al., 2020). They also suggest that the severity of the initial impairment is a risk factor for the development of long-term neuropsycho-  (Matschke et al., 2020;Thakur et al., 2021) and in DMI-MRI (Rau et al., 2022). Finally, the PLSC approach revealed stable associations between episodic verbal memory and the dorsal attentional networks suggesting that lower connectivity was coherent with worse memory performances, which is consistent with neuroimaging studies of episodic memory in healthy individuals (see Jeong et al., 2015;Rugg & Vilberg, 2013). Our results also question whether these cognitive effects are solely due to ICU/mechanical ventilation, and perhaps suggest a potential direct or indirect effect of a SARS-CoV-2 infection on long-term neuropsychological consequences. Although the moderate patients were not admitted in ICU and did not undergo mechanical ventilation, they still showed reduced cognitive performance, with reduced performances in mental flexibility in comparison to the mild patients. This corroborates previous behavioral observations by Alemanno et al. (2021), who observed significantly reduced executive scores in patients who received oxygen therapy different than mechanical ventilation but is in contradiction with a recent histopathological study who has observed that changes after COVID-19 were delimited by those caused by the extracorporeal respiratory assistance treatments . An interesting hypothesis that could encompass the results obtained with the three groups could be a potential alteration of local and global connectivity following a neurological disturbance, in this case, SARS-CoV-2 infection. Recent studies in acquired neurological (e.g., cranio-cerebral trauma), neuroimmunological (e.g., multiple sclerosis) or neurodegenerative (e.g., mild cognitive impairment or Alzheimer's disease) pathologies have highlighted patterns of both higher and lower connectivity (for review, see Hillary et al., 2015). Authors have suggested that hyperconnectivity is a common response following a neurological disruption, but the subsequent depletion of neural resources leads to a rapid decrease in connectivity (Hillary et al., 2015). The presence of compensatory mechanisms inducing patterns of higher connectivity in the short-term following SARS-  (Zakzanis et al., 2005), but also the cerebellum (Moll et al., 2002). As discussed above, our neuroimaging results showed increased activation patterns in the temporal cortical networks and in the cerebellum in moderate patients, but no patterns in the frontal lobes. Thus, despite the fact that these networks are involved in the processing of mental flexibility, it is possible that these regions are currently compensating for the other neuropsychological deficits which could therefore induce deficits for mental flexibility by a slowing down the processing speed. According to the literature, such phenomenon could be an important side effect of hyperconnectivity patterns following a neurological disturbance (Hillary et al., 2015). In the case of memory which was significantly reduced in the severe group, neuroimaging studies on healthy subjects have suggested distributed networks of brain regions have been associated with process of encoding, consolidation and retrieval for verbal episodic memory (for review, see Jeong et al., 2015;Rugg & Vilberg, 2013). Interestingly, the majority of studies have demonstrated the involvement of mediotemporal lobe regions (involving hippocampal or parahippocampal structures) in the different processes of verbal episodic memory. However, studies have also shown the involvement of subregions of frontal networks (e.g., dorsolateral prefrontal cortex for the encoding process or medial prefrontal cortex for retrieval). Nevertheless, despite the observed results, the neuropsychological and neurological long-term effects following SARS-CoV-2 are currently unknown, which narrows the scope of interpretation.
Our study has several limitations. By enrolling volunteers, we may have selected the most severe cases, although a significant proportion of our sample did not report any complaints, as confirmed by the very low mean score on the self-report QPC. This study was only performed on patients who were infected with SARS-CoV-2, and these patients had no known clinical history, posing two limitations for generalization of results. Here, we did not include a control group because the aim of the present study was to investigate differences in cognition and brain connectivity as function of the severity of the acute infection. Moreover, with the high rates of infection, it has become more difficult to recruit subjects that have never been infected with SARS-CoV-2.
Therefore, we cannot exclude that the mild group also exhibits reduced neuropsychological scores in comparison to a control group as has been described in the literature. That said, a recent study by our group did not show a significant accumulation of deficits in the group of mild patients compared to a simulated normative population, while the moderate and severe groups presented a significantly greater accumulation of neuropsychological deficits . Moreover, our moderate and severe groups are potentially not representative of the population of hospitalized SARS-CoV-2 patients because of their lack of comorbidities. It is important to highlight the considerable variance observed in the moderate group, as this could explain the small number of significant differences between groups. The cognitive and psychiatric, as well as functional connectivity (as described above) of the moderate group were extremely heterogeneous, suggesting that some patients presented deficits while others had none, leading to nonsignificant results.
The statistical comparison of behavioral data and functional connectivity revealed an imbalance between the groups and the small number of severe participants who underwent MRI may limit the generalization of this group's neuroimaging data. Nevertheless, to the best of our knowledge, studies on functional connectivity in acute and long-term following SARS-CoV-2 infection presented results, on average, from 31 participants with an imbalance in group as function of severity. The acquisition of field maps was not part of the MRI protocol and correction for susceptibility distortion was not performed. Finally, the generalizability of PLS methods has been criticized and, while results stay informative about multivariate correlations within the data, the correlations from PLSC should be validated with techniques such as crossvalidation.

| CONCLUSION
Our study confirms the presence of long-term neuropsychological effects in patients who had moderate-to-severe symptoms in the

CONFLICT OF INTEREST
The authors declare no conflict of interest.

DATA AVAILABILITY STATEMENT
At the end of the COVID-COG project, nonsensitive data will be made available in open access on a dedicated platform.