From early limbic inflammation to long COVID sequelae

We read with great interest the study of Hosp and colleagues1 on altered cerebral glucose metabolism in the subacute stage of COVID-19 using 18F-FDG PET. The authors established a spatial covariance pattern using voxel-wise principal component analysis (PCA) in comparison to control patients in whom brain diseases were excluded, and reported its correlation with Montreal Cognitive Assessment (MoCA) scores. This pattern includes negative weights within parietal and frontal association cortices, and positive weights within the brainstem, cerebellum, white matter, mesial temporal lobe structures and medial orbitofrontal region. The authors interpreted negative weights as hypometabolism, and positive weights as artefacts of relatively preserved metabolism. This covariance approach was compared to SPM analysis using activity normalization on white matter, and neocortical hypometabolism was described.1 We believe that this study constitutes an important contribution to the understanding of brain damage in COVID-19. It provides detailed clinical-imaging correlations, including structural MRI and post-mortem examination of one patient. We would like to discuss similarities and discrepancies with respect to our 18F-FDG PET results in long COVID,2 and propose an interpretation linking the findings of both reports. Beyond differences in initial severity and stage of evaluation, we are struck by the similar pattern of positive weights of covariance in the subacute stage reported in Freiburg1 and the pattern of relative hypometabolism in long COVID in Marseille.2 Using distinct approaches, both studies identified brain areas including the medial orbitofrontal region (and probably olfactory pathway, at z = −25 in Fig. 3 of Hosp et al.1), limbic medial temporal region, thalamus/hypothalamus, brainstem and cerebellum. In 35 patients with long COVID, we showed relative hypometabolism of olfactory bulbs as well as remote areas including limbic/paralimbic regions (amygdala, hippocampus, thalamus/hypothalamus), the brainstem (pons/medulla) and cerebellum. These findings were obtained in patients with proven SARS-CoV-2 infection and persistent/recurrent complaints, occurring on average 3 months after mild-to-moderate initial symptoms. Brain abnormalities and their association with symptoms were identified at the group and individual levels in comparison to 44 healthy subjects. We notice that the most consistent cross-aetiology findings on a recent systematic review of 90 neuroimaging and neuropathological studies on COVID-19 were the impairment of white matter, brainstem, and fronto-temporal areas, and especially of olfactory and orbitofrontal regions.3 Given the demonstration of viral persistence and inflammation in human olfactory epithelium, with diffusion to brain cortex, brainstem and cerebellum after intranasal inoculation in hamsters,4 we interpreted the results in long COVID as hypometabolic dysfunction of a brain network connected to the olfactory pathway, through a propagation mechanism previously proposed for other coronaviruses, across the lamina cribrosa, from nose to olfactory bulb,5 where ACE2 receptors are strongly expressed. Of note, both RNA and protein of the virus have been found in neurons of COVID-19 patients, with T cell inflammatory infiltration and microglial activation in macaques after apparent remission (for a review, see Guedj et al.6). Hypometabolism of deeper connected regions in the limbic system, brainstem and cerebellum could at least partially correspond to a diaschisis phenomenon in relation to olfactory pathway deafferentation. An alternative explanation would be the diffusion of inflammation/infection to these more posterior regions through a trans-synaptic mechanism previously reported for other coronaviruses,4 possibly in patients with more severe presentation or those with little recovery after rehabilitation. Further multi-modal imaging studies will be needed to determine the possible contributions of each hypothesis. 18F-FDG PET might also be used to identify hypermetabolism within this network at earlier stages, since this feature has already been reported during the acute phase of COVID-19 (either in the presence or absence of encephalitis), especially by Kas et al.,7 and may more directly reflect inflammation/infection.7,8 The question about possible artefactual origin of positive weights is a legitimate issue when discussing regions of relative hypermetabolism in 18F-FDG PET. Hosp and colleagues1 justified their artefactual interpretation by unsignificant hypermetabolism using SUV measures, but their findings were also unsignificant for hypometabolism using the same approach. Consequently, we believe these findings should not be set aside before methodological and pathophysiological implications have been evaluated. The exact interpretation of such a ‘red’ component has indeed been addressed in the field of neurodegenerative diseases by the group who developed the ScAnVP/SSMPCA toolbox,9 which colleagues of Freiburg used in their study. By separating positive and negative weights, Spetsieris and Eidelberg9 showed that positive ‘red’ disease-related pattern areas could not be entirely interpreted as spurious findings of preserved regions induced by global normalization to balance ‘blue’ cortical deficits. They concluded that effects are expressed in both ‘red’ and ‘blue’ networks, and that the former quantify the disease process at least as well as the latter.9 Accordingly, hypermetabolism should also be discussed in terms of pathophysiological meaning, especially in diseases where brain inflammation can also be expected or in disorders still in the course of clinical and pathological definition (as previously done for autoimmune encephalitis and now for COVID-19). Of note, in the paper from Hosp and colleagues,1 ‘red’ weights were not excluded from the spatial covariance pattern when correlations with symptoms and between-group differences were evaluated, and no demonstration is provided that the relevant performance of proposed pattern expression score would be reinforced by the exclusion of this alleged artefact. Indeed because of the similarity of regions involved in the previously reported hypometabolic pattern,2 we believe that the positive weights found by the Freiburg group might not necessarily represent an artefactual finding. Using proportional scaling, we would have found artefactual hypermetabolism of these ‘preserved’ regions at the latter stage, and not hypometabolism as we actually found.2 In contrast, the transition from hypermetabolism to hypometabolism after the subacute stage (or recovery) is in agreement with the evolution of PET findings in autoimmune encephalitis and viral encephalitis.10 The difference in interpretation of this positive component has a major impact on methodological choices when analysing brain PET metabolism with Statistical Parametric Mapping (SPM). Indeed, Hosp and colleagues1 used white matter as a region of preserved metabolism to normalize the PET activity. This methodological choice has been seldom used in comparison to the most classical approach of normalization to the pons or cerebellum,7 or as we proposed with global proportional scaling.2,8 This choice might also be responsible for some of the differences with respect to the pattern found in the patients analysed in Marseille,2 as this whole network covaries with white matter in their own PCA results.1 Normalizing to a relatively hypermetabolic region may also artefactually enhance true neocortical hypometabolism. The choice of such normalization by the white matter might also be suboptimal given evidence of astrogliosis in the post-mortem tissue of these patients and previous abnormalities found of these regions on neuroimaging studies.3 In this line, Hosp and colleagues1 interestingly described such white matter abnormalities in the one patient who had a post-mortem examination. As a first preliminary clarification, we would like to share the results of the analysis of our previous report2 comparing 35 patients with long COVID and 44 healthy subjects of similar age using the same spatial covariance approach proposed by Hosp and colleagues.1 The second PC accounted for 8.4% of the variance (for a predefined threshold of 5%) with a difference between patients and healthy subjects (P < 0.001, t-test), and correlated with the symptom scores we previously defined2 (r = 0.371, P = 0.028). Figure 1A presents the spatial covariance pattern, and as expected, the network we previously found has clearly negative weights (i.e. the limbic system, brainstem and cerebellum), with an inverse cold/hot pattern between bottom/up regions than those reported by the Freiburg group at later stages.1 This overlap might suggest a transition from positive to negative weights from subacute to long COVID. 8F-FDG PET group analysis of spatial covariance patterns. Results are presented from the 18F-FDG PET group analysis of spatial covariance patterns in comparison to 44 healthy subjects similar in age: (A) 35 patients with long COVID; (B) 73 patients with autoimmune encephalitis; and (C) six patients with subacute COVID-19. Voxels with negative region weights are in cool colours, and regions with positive region weights are in hot colours. Second, to test the hypothesis that the spatial covariance pattern found at the subacute stage of COVID-19 could correspond to inflammatory metabolic changes similar to those found in autoimmune encephalitis, we used this same covariance method to analyse 18F-FDG PET scans of 73 patients with autoimmune encephalitis acquired before the SARS-CoV-2 outbreak (mean age: 55 years; 56% female). The first PC accounted for 18.9%, with a difference between patients and healthy subjects of similar age (P < 0.001, t-test). This profile is presented in Fig. 1B, with a quite similar pattern as the one reported by the Freiburg group,1 which is inverse to those we found in long COVID.2 Interestingly, the two covariance patterns were very highly negatively correlated in healthy subjects (r = −0.93, P < 0.001; without statistical difference for age), demonstrating that the same regions were involved in long COVID and encephalitis-like covariance patterns with opposite bottom/up weights from hypermetabolism to hypometabolism. Finally, a spatial covariance pattern analysis was performed in six subjects retrospectively explored at the subacute stage of COVID-19 (four from Genoa and two from Marseille; mean age: 66 years; four females; mean interval with initial symptoms: 3 weeks (1–4) ; two patients with acute infectious symptoms, two with isolated hyposmia, two with various functional complaints, and none with encephalopathy/encephalitis). The first PC accounted for 16.3%, with a difference between patients and healthy subjects (P < 0.001, t-test), and a similar global pattern to those reported by Hosp and colleagues1 (Fig. 1C). Interestingly, and as expected, this subacute COVID-19 component was negatively correlated with those identified in long COVID (r = −0.49; P = 0.001) and positively correlated with those identified in autoimmune encephalitis (r = 0.37; P = 0.013). Further studies are required to determine the possible correct sequence between hypermetabolism and hypometabolism in subacute and long COVID, with absolute quantification of glucose consumption representing the ideal gold standard in comparison to healthy subjects. In this framework, other PET tracers targeting neuroinflammation through microglial activation/reactive astrocytes might be particularly relevant. The confirmation of positive weights, as a true feature of subacute COVID-19, with an encephalitis-like pattern in the same regions showing hypometabolism at later stages of COVID-19 could reinforce the hypothesis of network disruption through early diffusion of inflammation/infection followed by hypometabolic sequelae in long COVID, with therapeutic implications for early intervention. Data are available from the corresponding author on reasonable request. This work was funded by APHM (NCT00484523). The authors report no competing interests.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.