SARS-CoV-2 is associated with changes in brain structure in UK Biobank

Abstract

There is strong evidence of brain-related abnormalities in COVID-191-13. However, it remains unknown whether the impact of SARS-CoV-2 infection can be detected in milder cases, and whether this can reveal possible mechanisms contributing to brain pathology. Here we investigated brain changes in 785 participants of UK Biobank (aged 51-81 years) who were imaged twice using magnetic resonance imaging, including 401 cases who tested positive for infection with SARS-CoV-2 between their two scans-with 141 days on average separating their diagnosis and the second scan-as well as 384 controls. The availability of pre-infection imaging data reduces the likelihood of pre-existing risk factors being misinterpreted as disease effects. We identified significant longitudinal effects when comparing the two groups, including (1) a greater reduction in grey matter thickness and tissue contrast in the orbitofrontal cortex and parahippocampal gyrus; (2) greater changes in markers of tissue damage in regions that are functionally connected to the primary olfactory cortex; and (3) a greater reduction in global brain size in the SARS-CoV-2 cases. The participants who were infected with SARS-CoV-2 also showed on average a greater cognitive decline between the two time points. Importantly, these imaging and cognitive longitudinal effects were still observed after excluding the 15 patients who had been hospitalised. These mainly limbic brain imaging results may be the in vivo hallmarks of a degenerative spread of the disease through olfactory pathways, of neuroinflammatory events, or of the loss of sensory input due to anosmia. Whether this deleterious effect can be partially reversed, or whether these effects will persist in the long term, remains to be investigated with additional follow-up.

Conflict of interest statement

R.C. has been seconded from the University of Oxford as chief executive and principal investigator of UK Biobank, which is a charitable company. N.A. is chief scientist for UK Biobank. P.M.M. acknowledges consultancy fees from Novartis and Biogen; he has received recent honoraria or speakers’ honoraria and research or educational funds from Novartis, Bristol Myers Squibb and Biogen. P.M.M. serves as the honorary chair of the UK Biobank Imaging Working Group and as an unpaid member of the UK Biobank Steering Committee; he is chair of the UKRI Medical Research Council Neurosciences and Mental Health Board.

© 2022. The Author(s).

Figures

Fig. 1. The most significant longitudinal group comparison results from the hypothesis-driven approach.

a–d, The top four regions consistently showing longitudinal differences across the three models comparing SARS-CoV-2 cases and controls demonstrated either a significantly greater reduction in grey matter thickness and intensity contrast, or an increase in tissue damage (largest combined |Z| across Models 1–3). All three models pointed to the involvement of the parahippocampal gyrus (a), whereas Models 1 and 2 also showed the significant involvement of the left orbitofrontal cortex (b) and of the functional connections of the primary olfactory cortex (c, d). For each region, the IDP’s spatial region of interest is shown at the top left in blue, overlaid either on the FreeSurfer average inflated cortical surface, or the T1 template (left is shown on the right). For each IDP, the longitudinal percentage changes are shown with age for the two groups (control participants in blue, participants with infection in orange), obtained by normalising ΔIDP using the values for the corresponding IDPs across the 785 participants’ scans as the baseline. These are created using a 10-year sliding window average, with s.e.m. values shown in grey. The counter-intuitive increase in thickness in the orbitofrontal cortex in older controls has been previously consistently reported in studies of ageing,. The difference in cortical thickness, intensity contrast or diffusion indices between the two time points is shown for the 384 controls (blue) and 401 infected participants (orange), enabling a visual comparison between the two groups in a binary manner (therefore underestimating the effects estimated when modulating with age; see the ‘Main longitudinal model, deconfounding’ section in the Methods). The 15 hospitalised patients are indicated (red circles). ISOVF, isotropic volume fraction; OD, orientation dispersion. All y axes represent the percentage change.

Fig. 2. Vertex-wise and voxel-wise longitudinal group differences in grey matter thickness and mean diffusivity changes.

Top, the main analysis (Model 1): the thresholded map (|Z| > 3) shows that the strongest, localised reductions in grey matter thickness in the 401 infected participants compared with the 384 controls are bilaterally in the parahippocampal gyrus, anterior cingulate cortex and temporal pole, as well as in the left orbitofrontal cortex, insula and supramarginal gyrus. Similarly, the strongest longitudinal differences in mean diffusivity (|Z| > 3, left is shown on the right) could be seen in the orbitofrontal cortex and anterior cingulate cortex, as well as in the left insula and amygdala (top). Bottom, secondary analysis (Model 4): the thresholded cortical thickness map (|Z| > 3) demonstrated longitudinal differences between the 15 hospitalised and 386 non-hospitalised SARS-CoV-2-positive cases in the orbitofrontal frontal cortex and parahippocampal gyrus bilaterally, right anterior cingulate cortex, as well as marked widespread differences in fronto-parietal and temporal areas, especially in the left hemisphere. We show the voxel-wise or vertex-wise longitudinal effects for illustrative purposes, avoiding any thresholding based on significance (as this would be statistically circular), similar to our previous analyses.

Fig. 3. Significant longitudinal differences in cognition.

a, b, The percentage longitudinal change for SARS-CoV-2-positive cases and controls in the duration to complete trails A (a) and B (b) of the UK Biobank Trail Making Test. The absolute baseline (used to convert longitudinal change into percentage change) was estimated across the 785 participants. These curves were created using a ten-year sliding window across cases and controls (s.e. values are shown in grey).

Extended Data Fig. 1. Age distributions for SARS-CoV-2 positive participants and controls at each time point do not differ significantly.

Two-sample Kolmogorov-Smirnov was used to compute the P values for age comparisons, since age for each group was not normally distributed (Lilliefors P = 1e-03 for each group, and both age at Scan 1 or Scan 2). This showed no significant difference in age distribution between SARS-CoV-2 participants and controls at Scan 1: P = 0.15 or at Scan 2: P = 0.08.

Extended Data Fig. 2. Histograms showing the well-matched distributions of Scan 1 - Scan 2 intervals for case and control groups.

The below IDP reproducibility Extended Data Fig. 3 shows, for comparison against the cases and controls, reproducibility from around 3,000 (2,943) UK Biobank participants who had returned for a second scan prior to the pandemic; hence we also show here the interscan intervals for this “3k” group, with tighter control over this interval (we have normalised each of those 3 groups to have a peak of 1, to make the relative comparison easier).

Extended Data Fig. 3. Scan-rescan reproducibility for all 2,047 IDPs used in the main modelling.

Each dot represents a single IDP, arranged into different classes of IDPs. For each IDP, the vector of values for each subject (i.e., 785x1 vector) from the first scan was correlated with the equivalent vector of IDP values from the second scan. The y axis shows the resulting correlation coefficient. These calculations are made separately for the pre-pandemic scan-rescan datasets ("3k DPUK"), and for cases and controls, demonstrating highly similar distributions within each IDP class for all 3 subject groups.

Extended Data Fig. 4. QQ plot for −log10[Puncorr] against the theoretical null distribution.

The black line at y=x shows the expected plot if no effects were present in the data. Orange points reflect ΔIDPs where the case-control effect passes FDR significance, and blue reflects those that do not.

Extended Data Fig. 5. Model Z-statistics (one point per IDP, arranged in IDP classes) for the 4 main models.

Note that these are model Z-statistics, not raw effect size. Some IDP classes (e.g., cortical thickness and grey-white intensity contrast) show consistent group-difference effect directions across most IDPs (i.e., different brain regions), and all 4 models.

Extended Data Fig. 6. Examples of percentage change in some of the most significant longitudinal group comparison results from the exploratory approach.

Four amongst the top IDPs consistently showing longitudinal differences between SARS-CoV-2 cases and controls. All demonstrate either a greater reduction in local or global brain thickness and volume, or an increase in CSF volume. For each four IDP are the percentage changes with age for the two groups, obtained by normalising ΔIDP using as baseline the values for the corresponding IDPs across the 785 scans (created using a 10-year sliding window across cases and controls, with standard errors in grey). The counterintuitive increase in thickness in the rostral anterior cingulate cortex in older controls has been previously consistently reported in studies of ageing, together with that of the orbitofrontal cortex,.