Uncovering a neurological protein signature for severe COVID-19

Coronavirus disease of 2019 (COVID-19), caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), has sparked a global pandemic with severe complications and high morbidity rate. Neurological symptoms in COVID-19 patients, and neurological sequelae post COVID-19 recovery have been extensively reported. Yet, neurological molecular signature and signaling pathways that are affected in the central nervous system (CNS) of COVID-19 severe patients remain still unknown and need to be identified. Plasma samples from 49 severe COVID-19 patients, 50 mild COVID-19 patients, and 40 healthy controls were subjected to Olink proteomics analysis of 184 CNS-enriched proteins. By using a multi-approach bioinformatics analysis, we identified a 34-neurological protein signature for COVID-19 severity and unveiled dysregulated neurological pathways in severe cases. Here, we identified a new neurological protein signature for severe COVID-19 that was validated in different independent cohorts using blood and postmortem brain samples and shown to correlate with neurological diseases and pharmacological drugs. This protein signature could potentially aid the development of prognostic and diagnostic tools for neurological complications in post-COVID-19 convalescent patients with long term neurological sequelae.


Introduction
COVID-19 exhibits a highly heterogeneous clinical profile with no, mild or severe symptoms, the latter associated with the need of life support in intensive care units (ICU) or fatal outcomes. Neurologic manifestations are common features of severe COVID-19 disease including neurological and neuropsychiatric symptoms of infection of the central nervous system (CNS) such as dizziness, headache, impaired consciousness, ataxia, altered mental status, anosmia, chronic fatigue, memory impairment, and depression (Varatharaj et al., 2020;Paterson et al., 2020;Tsivgoulis et al., 2020;Helms et al., 2020;Rogers et al., 2020;Leonardi et al., 2020;Achar and Ghosh, 2020;Mao et al., 2020;Eden et al., 2021a;Oliveira et al., 2021;Sinanovic et al., 2020;Bodnar et al., 2021). Moreover, a variety of CNS disorders; encephalopathies, seizures, ataxia and strokes, have been found particularly in severe COVID-19 patients (Chaumont et al., 2020;Logmin et al., 2020;Princiotta Cariddi et al., 2020;Zachariadis et al., 2020;Virhammar et al., 2020;Virhammar et al., 2021), and perturbations related to neurological/neurodegenerative diseases have been reported in severe COVID-19 brains (Yang et al., 2021;Fullard et al., 2021;Emmi et al., 2023;Schwabenland et al., 2021). Several emerging reports have demonstrated that different mechanisms may contribute to some or all aspects of the neuro-invasive potential of SARS-CoV-2 (Achar and Ghosh, 2020;Mao et al., 2020). These include (i) direct viral interaction with neuronal and supportive cells in a retrograde axonal transport manner, primarily in the olfactory mucosa; (ii) indirect effect of the systemic immune response leading to activation of CNS-resident immune cells; and (iii) viral interaction with cells of the vasculature including pericytes and/or endothelial cells and CNS invasion through the blood-brain barrier (BBB) (Eden et al., 2021a). Likely, different combinations of these mechanisms could contribute to the clinical disease complications of individual patients. At the cellular level, neuronal injury and glial activation have been observed in patients with severe COVID-19 using CNS biomarkers of axonal and astrocytic damage 23, (Hirzel et al., 2022). CNS biomarkers such as Neurofilament light chain (NEFL), Glial fibrillary acidic protein (GFAP), ubiquitin carboxy-terminal hydrolase L1 (UCH-L1), and Tau have been shown to reflect CNS damage in traumatic brain injury (Virhammar et al., 2021;De Lorenzo et al., 2021;Bacioglu et al., 2016;McMahon et al., 2015;Wang et al., 2021). These biomarkers have been extensively used to assess COVID-19 neuropathogenesis in ICU admitted severe patients and post-COVID-19 convalescent patients as well (Hanson et al., 2022;Eden et al., 2021b;Sun et al., 2021). Nevertheless, there is a need to explore other CNS proteins and decipher the molecular pathways underlying neurological/neuropsychiatric dysfunctions in severe COVID-19 cases.
Furthermore, as a significant number of case reports and studies have described different types of neurological complications in severe COVID-19 patients; early predictive and diagnostic tools to classify the risk of developing neurological complications and assess the neurological status of post-COVID-19 convalescent patients are crucial for clinical management. In this study, we hypothesized that alterations in plasma proteins could offer predictive and diagnostic neurological protein signatures that can help to predict the risk of developing complications and diagnose neurological complications in convalescent patients. To address these questions, we used high throughput Olink proteomics analysis of plasma samples from patients with severe complications versus mild symptoms and control subjects and assessed changes in neurological protein profiles and pathways associated with severe COVID-19.

Study cohorts
We used two cohorts for this study; one cohort of COVID-19 patients and one cohort of healthy subjects(Al-Nesf et al., 2022) (Table 1, Supplementary data 1). All participants (patients and controls) provided written informed consent prior to enrolment in the study (reference MRC-05-003).

Qatar COVID-19 cohort
This cohort included 99 COVID-19 patients from Qatar who were admitted to Hamad Medical Corporation hospitals. All recruited patients had confirmed SARS-CoV-2 positive RT-PCR results of sputum and throat swabs, of which 49 patients had severe COVID-19 disease and were admitted to intensive care unit (ICU) and 50 patients had mildmoderate COVID-19 disease. Peripheral blood was collected within 5-7 days post-admittance and plasma was stored at -80 • C until use. Ethical approvals for this cohort were obtained from the Hamad Medical Corporation Institutional Review Board Research Ethics Committee (reference MRC-05-003), and the Qatar Biomedical Research Institute-Institutional Review Boards (Reference QBRI-IRB 2020-06-19).

Healthy control cohort
Age-and gender-matched healthy subjects (n = 40) with no history of prior COVID-19 infection and with normal oxygen saturation and vital signs were recruited by the Anti-Doping Laboratory-Qatar. Individuals with medical history or with cognitive disabilities were excluded.

Olink proteomic assays
Plasma samples were profiled using the Olink proximity extension assays (PEA), 92-plex immunoassay (Uppsala, Sweden) (Assarsson et al., 2014) following the standard protocol. Quality control and data normalization were carried out using the Normalized Protein eXpression (NPX) software and every run was checked and validated by the Olink support team in Uppsala, Sweden. Two different protein panels focusing on CNS and neurodegenerative diseases were used in our study; the Neurology target panel and the NeuroExploratory target panel. Protein expression values were calculated as log2(NPX), and Olink data that did not pass quality control were excluded from the analyses (Supplementary data 2).

Simoa assay
Using single molecule array (Simoa) technology (Quanterix), we measured four neurology biomarkers in human plasma samples from mild and severe cases of COVID-19 as well as healthy controls. We used the Human Neurology 4 Plex B (N4PB) kit and the assay was run on the Quanterix HD-X system. The four targets are total tau, neurofilament light (NEFL), glial fibrillary acidic protein (GFAP), and ubiquitin carboxyl-terminal hydrolase L1 (UCH-L1) and measurements were done according to manufacturer's instructions.

Correlation analysis
Illumina's Basespace™ Correlation Engine (BSCE; Illumina, San Diego, CA, USA) was used to correlate the 34 neurological protein signature of severe COVID-19 to available SARS-CoV-2 host infection transcriptome data. BSCE contains 166,000 Bioset lists of highly curated, statistically significant genes from over 25,100 omic-scale Curated Studies (Kupershmidt et al., 2010) as of 1 Oct 2022. Curated Studies in BSCE are processed in standard, platform-specific pipelines to generate gene sets and measures. For each study, available Test group vs. Control group results plus experimental metadata are generated. Second, BSCE has robust data-driven correlation, aggregation, and machine learning applications to exploit the consistent processing and curation of omic-scale studies from international public repositories. The BSCE Pharmaco Atlas and Disease Atlas applications were used to respectively categorize, and rank compounds or diseases based on signature correlations across all BSCE Curated Studies.

Study design and cohort specific information
In the present work, SARS-CoV-2 infected patients of a local Qatar cohort were evaluated for their proteomics profiling. The cohort was recruited at HMC hospital in Doha, Qatar and included 49 subjects with severe and 50 subjects with mild disease; and 40 healthy control subjects (Table 1, Supplementary data 1). COVID-19 patients were admitted to ICU and classified as severe cases using World Health Organization (WHO) severity scoring for COVID-19. Severity was defined by low oxygen saturation, severe pneumonia, and/or signs of severe respiratory distress (accessory muscle use, inability to complete full sentences, respiratory rate > 30 breaths per minute). All severe cases were neurologically unaffected at the time of enrolment with no prior medical history of neurological conditions. Non-severe COVID-19 group were admitted to the medical floor with no WHO severity criteria and classified as mild-moderate cases. All the 50 mild COVID-19 patients recovered completely without developing a severe form of the disease.
The population of Qatar with SARS-CoV-2 infection was predominantly males and young during the early wave of the COVID-19 pandemic between March and April 2020. Our studied cohort of SARS-CoV-2 infected patients was collected in the early pandemic and in the same time frame as the nationwide cohort of the first consecutive 5000 patients with COVID-19 in Qatar (Al-Nesf et al., 2022;Omrani et al., 2020), thus had similar characteristics of most of the SARS-CoV-2 infected patients (88.7%) being males, with a median age of 35. Hospitalizations were predominately under 65 years of age; 95.7% for non-ICU and 85.2% for ICU patients. ICU patients in Qatar were further enriched for males (92.6%), and only 16 patients (14.8%) in the ICU were above 65 years or older (Dazzo et al., 2016). Our studied cohort has a limitation for being biased toward male and younger subjects with patients aged between 18 and 65 and being mainly males with only 8% females. This cohort selection was based on local COVID-19 demographics during the early first wave of the pandemic(Al-Nesf et al., Omrani et al., 2020).

Olink analysis revealed differential neurological plasma protein profile in patients with severe COVID-19 complications
To examine the effects of SARS-CoV-2 infection on neurological protein expression, we first performed a differential expression analysis from 99 COVID-19 patients (49 severe and 50 mild) and 40 healthy control subjects using two different Neurology Olink panels (Olink target 96 NeuroExploratory panel and Olink target 96 Neurology panel) (Wang et al., 2021). For a comprehensive molecular view, we carried out the analysis after combining the data from the two Olink panels in a single dataset (Supplementary data 3). Unsupervised hierarchical clustering revealed that the combined panels could differentiate severe disease from mild disease and controls as illustrated in the heatmap (Fig. 1A). Principal components analysis (PCA) of protein NPX data from the three groups demonstrated a clear separation of COVID-19 severe cases from COVID-19 mild and control samples (Fig. 1A). The number of differentially expressed proteins (DEPs) is summarized in Fig. 1B and Supplementary data 4. A total of 100 (54.3%) and 81 (44%) proteins were differentially expressed in cases with severe disease compared to control subjects or cases with mild disease, respectively. When comparing patients with mild disease to control subjects, 38 (20.6%) proteins were found to be differentially expressed. As shown in the volcano plots 36% of proteins were upregulated in severe cases, while only 8% and 18% were downregulated in Severe vs. Mild and Severe vs. controls, respectively, with a minimum fold change of 1.25 and FDR cutoff of 0.05. For instance, NGF, NEFL, PLXNB1, EFNA4, and others were significantly upregulated in severe cases, while GDNFR3, ADAM23 and others were significantly downregulated in severe cases (Fig. 1C). Interestingly, axonal injury NEFL protein profile of severe cases was validated by Simoa assay. NEFL was significantly highly expressed in severe cases compared to mild and control groups, showing a strong correlation with Olink data (r = 0.76, p < 00.1) (Sup Fig. 1A and B). Also, other known biomarkers for axonal injury such as GFAP and UCHL-1 proteins were found upregulated in severe cases (Sup Fig. 1A).

Protein pathways analysis uncovered upregulated and downregulated neurological pathways in severe COVID-19 patients
GO and KEGG pathways analysis were performed to identify the functional contribution of DEPs in biological processes. We found that 13 pathways associated with nervous system functions were significantly enriched in patients with severe disease as compared to healthy subjects (Neurotrophin signaling pathway, regulation of synapse assembly, nervous system development, neuron projection, axon guidance receptor activity, Semaphorin receptor complex and axon guidance) ( Fig. 2A-D, Supplementary data 5). Additionally, peripheral nervous system (PNS) functions were affected in severe vs. mild cases with significantly downregulated PNS development pathway (Sup Fig. 2A). Overall, severe vs. mild cases showed 2 downregulated pathways and 3 upregulated pathways with 2 non-overlapping pathways with severe vs. controls (PNS development pathway and semaphoring receptor activity) (Sup Fig. 2A-D, Supplementary data 6). Such enrichment with neurological pathways was absent in mild vs. controls (Sup Fig. 2E, Supplementary data 7), suggesting that mild and control groups might be devoid of neurological complications. Next, we ran IPA pathway analysis for severe cases versus controls, and we found 7 neurological pathways significantly affected in severe COVID-19 patients, such as axonal guidance signaling, synaptogenesis signaling, semaphorin neuronal repulsive signaling, neuroprotective role of THOP1 in Alzheimer's disease, GDNF family ligand receptor interaction, neuroinflammation signaling and Ephrin A signaling (Sup Fig. 3,  Supplementary data 8).

Neurological protein signature of severe COVID-19
To predict the risk of neurological complications of severe COVID-19, we set out to create a blood-based neurological protein signature. For this purpose, we ran WGCNA analysis for 184 tested proteins to identify network modules with co-expressed and interacting proteins. A network of 155 proteins was identified and divided into 4 modules ( Fig. 3A, Supplementary data 9). The turquoise module had the highest number of proteins (n = 54), and was significantly enriched in neurological processes, functions, and cellular components such as neuron projection development, axon development, synapse, neuron differentiation, ephrin receptor activity, etc. (Fig. 3B), whereas the other modules did not include any neurological pathways. In addition, STRING analysis revealed that the 54 proteins form a dense protein-protein interaction (PPI) network with multiple interactions (Sup Fig. 4A). To define a neurological protein signature for severe COVID-19 patients that comprises only co-expressed and interacting proteins, we assessed which of the severe vs control DEPs (n = 100) overlap with the 54-protein network (Fig. 1B). As such, we identified a severe COVID-19 neurological signature of 34 proteins, 26 upregulated and 8 downregulated, that form multiple interactions within PPI networks (Fig. 3C,  Sup Fig. 4B).

Severe COVID-19 neurological protein signature is validated in different independent cohorts at the protein and RNA levels
Next, we validated the 34-neurological protein signature using the Massachusetts General Hospital (MGH cohort) Olink dataset (Filbin et al., 2021) (Supplementary data 10). More specifically, we used the Olink proteomic data from 139 COVID-19 MGH samples that were collected at day 7 after admittance. We considered cases under acuities (death at 28 days, and intubated, ventilated, supplemented oxygen, n = 71) as severe COVID-19 cases, and cases under acuities (hospitalized with supplemented O 2 and hospitalized without supplemented O 2 , n = 68) as mild COVID-19 cases (Supplementary data 11). Using unsupervised hierarchical clustering, we revealed that the 34-neurological protein signature could differentiate the MGH severe COVID-19 from mild COVID-19 cases as illustrated in the heatmap (Fig. 4A), and PCA Fig. 1. Differential protein analysis of COVID-19 patients. Differentially expressed proteins (DEPs) were identified from two Olink Neurology panels and the combined dataset (184 unique proteins), which was defined as DEPs with more than 1.25-fold change with a P-value <0.05 and FDR < 0.05. A. Hierarchical clustering based on all 184 proteins assayed using the two Olink panels showed a separation between patients with severe complications compared to mild cases and controls. The heatmap shows z-scores with cut-off 4, and principal component analysis (PCA) confirmed the separation of the severe cases based on the expression profiles of all proteins. B. Histogram displaying number of DEPs across the patients' groups. C. Volcano plot summarizing DEPs based on log 2 fold changes across the patient groups. Red and blue circles show significantly differential-expressed proteins ≥0.6 log 2 fold change (upregulated) or ≥ − 0.6 log 2 fold change (downregulated), and p value ≥0.05. The number and percentage of the DEPs relevant to all proteins assayed are stated in each panel. Differential expression analysis addressed severity as the main effect and included different factors to correct for interaction with severity, such as obesity, age, sex, nationality, SpO 2 , heart rate, and blood pression. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) demonstrated a clear separation of COVID-19 severe samples from COVID-19 mild samples (Fig. 4A). Out of the 34 proteins, we found that 21 proteins were significantly upregulated and 1 protein was significantly downregulated in MGH severe COVID-19 patients compared to the mild cases ( Fig. 4B and C, Supplementary data 12). Interestingly, all the 22 MGH cohort proteins positively correlated with Qatar cohort severe COVID-19 protein signature profile, with 21 upregulated proteins, and 1 downregulated protein (Fig. 4 D, Supplementary data 13). Interestingly, comparison of protein expression in patients with severe COVID-19 and symptomatic non COVID-19 related respiratory infections confirmed 21 out of the 22 severity-associated proteins to be significantly altered in patients with severe COVID-19 disease only (Sup Fig. 5, Supplementary data 14).
We further correlated the 34 neurological protein signature to transcriptomics data from prefrontal cortex autopsy of COVID-19 patients, including 9525 differentially expressed RNAs (Pujadas et al., 2021). We found 14 genes overlap between the two biosets, 9 genes with positive correlation and 5 genes with negative correlations to the neurological protein signature (Fig. 5A). More interestingly, using RNA data from whole blood of COVID-19 patients critical in ICU with ARDS  Fig. 6A-B), we detected most of the overlapping genes correlated positively to the neurological protein signature, while fewer genes had negative correlations (Supplementary data 15).

Neurological diseases and pharmacological drugs correlating with the neurological proteins signature of severe COVID-19
To gain insight into the role of the severe COVID-19 neurological proteins, we conducted correlation analyses for the 34 proteins using the BSCE Disease Atlas. We found 15 brain and nervous system disorders positively correlated with the COVID-19 protein signature, such as Nerve injury, Parkinson's disease, Alzheimer's disease, Schizophrenia, and Amyotrophic Lateral Sclerosis (16, 9, 6, 6 and 3 studies, respectively), and 3 neurological diseases negatively correlated with the proteins signature, such as Meningitis and Epilepsy (Table 2, Supplementary data 16). In addition, using the BSCE Pharmaco Atlas we analyzed protein signature-drug correlations in independent studies and, identified 42 and 58 drugs respectively to be positively and negatively correlated with the 34-protein signature (Supplementary data 17).

Discussion
There is an increasing interest in understanding the mechanisms underlying the neurological complications seen in some COVID-19 patients, as well as revealing new health issues arising from COVID-19 neurological complications. Several papers have described different types of neurological complications in severe COVID-19, and have demonstrated the neuro-invasive potential of SARS-CoV-2 1, 2, 4, 7, (Mao et al., 2020). It has been reported that most of the severe COVID-19 patients have developed neurological complications and are among Fig. 4. Validation of severe COVID-19 neurological protein signature in MGH cohort. 22 DEPs were identified from the 34 neurological Protein signature with more than 1.25-fold change with a P-value <0.05 and FDR < 0.05. A. Hierarchical clustering based on all 34 proteins assayed showed a separation between patients with severe complications compared to mild cases. The heatmap shows z-scores with cut-off 4, and principal component analysis (PCA) confirmed the separation of the severe cases based on the expression profiles of all proteins. B. Histogram displaying number of 22 DEPs, and Volcano plot summarizing DEPs based on log 2 fold changes across the patient groups. Red and blue circles show significantly differential-expressed proteins ≥0.6 log 2 fold change (upregulated) or ≥ − 0.6 log 2 fold change (downregulated), and p value ≥0.05. The number and percentage of the DEPs relevant to all proteins assayed are stated in each panel. Differential expression analysis addressed severity as the main effect and included different factors to correct for interaction with severity, such as age and BMI. C. MGH severe COVID-19 signature protein list. D. Scatter plot showing positive correlation of severe COVID-19 signature protein between Qatar cohort and MGH cohort. P-value = 0.001. the most debilitating manifestations (Taquet et al., 2021) With more than 48 million documented infections since the beginning of the pandemic, it is likely that more than 10 million individuals have been or are currently affected by neurological sequelae in the world (Baig, 2021;Nalbandian et al., 2021). Longitudinal studies are needed to evaluate persistent CNS damage in patients with severe COVID-19 neurological sequelae over time and their impact on cognitive functions, neurodegenerative and mental disorders in large populations. Developing bloodbased neurological protein signature can serve as a predictive biomarker tool to assess neurological complications in post-severe COVID-19 convalescent individuals.
In this current study, we explored changes in neurological protein expression between severe COVID-19, mild COVID-19, and healthy controls. Severe COVID-19 patients were all neurologically unaffected prior infection and at the time of enrolment. We conducted proteomics profiling of 184 CNS proteins in plasma samples using the Olink neurology and NeuroExploratory panels. Here, we identified a high differential protein expression in plasma patients with severe COVID-19 compared to controls, with 36% DEPs being upregulated and 18% being downregulated. We have observed that most upregulated proteins were related to severe disease but not to mild disease or control subjects.
Together with the Simoa results, our Olink data confirm previously published findings that patients hospitalized with severe COVID-19 have elevated biomarkers of CNS injury in their blood (Kanberg et al., 2020;Ameres et al., 2020;Chung et al., 2021;Geis et al., 2021;Sutter et al., 2021;Kanberg et al., 2021), such as NEFL, GFAP, UCH-L1, reflecting CNS damage and traumatic brain injury (De Lorenzo et al., 2021;Bacioglu et al., 2016;McMahon et al., 2015;Wang et al., 2021). This finding therefore supports the body of literature showing that COVID-19 neurological complications are accompanied by damage to the CNS. Beside the previously reported CNS injury biomarkers, here we identified proteins that have not yet been reported in COVID-19 severity, including NGF, GDNFR3, EFNA4, PLXNB1, and ADAM23, etc. They were found deregulated and may play an important role in neurological complications of severe COVID-19 patients. These proteins are mainly neuronal survival and guidance factors that play a major role in axonal growth, neurogenesis, neuroprotection, neural differentiation, and synaptogenesis (Laussu et al., 2014) (Markus-Koch et al., 2017Guthrie, 2007;Limoni and Niquille, 2021;Nordvall et al., 2022). Upregulated neuronal survival factors such as NGF is an indication of a healing process taking place in brain tissue (Aloe et al., 2015), as a consequence of brain damage in severe COVID-19 patients. These factors may have long term neurological impact. Although explorative study of bloodbased neurological proteins in the blood may add further insight into CNS damage and COVID-19 neurological complications, future studies should use cerebrospinal fluid (CSF) sampling as it reflects alterations occurring more accurately at the level of the CNS.
To identify the functional contribution of DEPs in biological processes, we ran GO, KEGG and IPA pathway enrichment analysis and found relative enrichment of neurological signaling pathways in severe COVID-19 disease, including nervous system development, synaptogenesis, neuron projection, axon guidance, Semaphorin receptor complex, Neurotrophin signaling pathway, neuroprotective role of THOP1 in Alzheimer's disease, GDNF family ligand-receptor interaction, neuroinflammation signaling, and others. Using WGCNA analysis, we narrowed down the DEPs number from 100 proteins to 34 co-expressed and co-interacting proteins, establishing a neurological protein signature for severe COVID-19. To validate our neurological protein signature in an independent cohort, we compared our data to an independent Olink study generated from MGH cohort plasma samples of severe COVID-19. We validated 22 proteins among the 34 neurological protein signature in MGH cohort of severe COVID-19 patients. 21 out of 22 proteins are specific to severe SARS-CoV-2 infection and not to other respiratory infections. Our Qatar cohort is ethnically diverse, which include Middle Eastern, Africans and Asian populations. Here, we provide evidence that our severe COVID-19 neurological protein signature is detected and validated in an independent multi-ethnic cohort and using the same Olink technology.
We then explored other cohorts of severe COVID-19 patients to validate the neurological protein signature in different biological samples and at the RNA level. More importantly, we first sought whether our protein signature can also be detected in brain tissue of severe COVID-19 postmortem patients. Fortunately, 9 genes were found positively correlating with our neurological protein signature (PLXNB1, RGMB, ADAM22, ADAM23, DKK4, SIGLEC1, TDGF1, ACVRL1, and PHOS-PHO1), meaning that these molecular changes in expression profile are similarly affected at the protein and RNA levels in both peripheral blood and brain severe COVID-19 samples. These findings suggest that these proteins may have an impact on brain functions, and most probably lead to neurological complications. Some of these proteins are known as neuronal guidance molecules such as PLXNB1 and RGMB. Guidance molecules play an important role during CNS development and are emerging as important factors in neurological diseases and brain injury (Yaron and Zheng, 2007). Deregulation of neuronal guidance molecules might be associated with predisposition to epilepsy, and it was also suggested that guidance molecules play roles in the ability of the adult nervous system to recover and repair after injury (Yaron and Zheng, 2007).
It has been reported that neuronal receptors ADAM22, and ADAM23 play major role in neuronal functions through LGI1 protein and have been associated to epilepsy disorder (Hivert et al., 2019;Dazzo et al., 2016). In our study, we report that both proteins are deregulated in severe COVID-19 patients, suggesting that they may lead to neurological complications in post-severe COVID-19 patients. We also found DKK4 protein deregulated in brain samples of severe COVID-19 patients, which has been shown to be significantly associated with schizophrenia (Zhang et al., 2011).
Glycosylation is vital for brain functions and an alteration in the process may lead to nervous system disorders. One of the types of glycosylation that is abundantly found in the brain is sialylation. SIGLEC proteins (Sialic acid-binding immunoglobulin-type lectins) have been shown to play roles in neuroinflammation and neurodegeneration (Freeze et al., 2015). Thus, having SIGLEC1 protein deregulated in severe COVID-19 patients may indicate a deficiency in brain functions. Also, PHOSPHO1 has been previously found altered in Parkinson's disease patients (Bartl et al., 2022), its protein expression profile is also affected in severe COVID-19 patients.
When exploring the correlation of neurological protein signature in RNA data from other cohorts of severe COVID-19 patients' whole blood or peripheral blood mononuclear cells, most of the common genes correlated positively with the neurological protein signature such as SIGLEC1, SCARB2, PLXNB1, CLEC10A, ACVRL1, DRAXIN, GGT5, TNFRSF12A, ASGR1, RGMB, PHOSPHOS1, CD38 CD300LF, and ADAM23 (Fig. 5, Sup Fig. 6 and Sup Fig. 7). Some of these findings are consistent with the above validations done on brain RNA samples such as PLXNB1, RGMB, ADAM23, SIGLEC1, and ACVRL1. On the other hand, we found NEFL, SCARB2, CD38, RGMA, TMPRSS5 proteins expression negatively correlated with RNA expression from severe COVID-19 brain samples (Fig. 5), meaning that peripheral blood and brain may slightly differ in neurological expression profile of severe COVID-19 patients. This interesting finding needs to be further investigated.
Furthermore, using BSCE Disease Atlas analyses to compare our protein signature to multiple studies, several neurological diseases and disorders were found positively correlating with our protein signature, such as Nerve Injury, Parkinson's disease, Alzheimer's disease, Amyotrophic lateral sclerosis, and Schizophrenia. Thus, it seems there is a risk of developing chronic neurological diseases in post-severe COVID-19 patients. The correlations to 'Nerve Injury' used RNA expression data from rodent induced-nerve injury models. As highlighted by TNFRSF12A, alike severe COVID-19 signature 12 out of 16 biosets had upregulated TNFRSF12A expression (Sup Fig. 8, Supplementary data 16) (Li et al., 2013;Omura et al., 2015;Chen et al., 2013;Wu et al., 2013). We next looked at Parkinson's disease curated studies, we found five out of six were SNP/GWAS: 2 studies had upregulated correlation of CD38, while SCARB2 correlated in 1 Study. TNFRSF12A and LAYN were concordantly upregulated in severe COVID-19 protein signature and RNA expression from Parkinson's disease iPS derived neurons bioset (Sup Fig. 9, Supplementary data 16) (Satake et al., 2009;Pankratz et al., 2009;Do et al., 2011;Woodard et al., 2014). In superior frontal gyrus, post central gyrus and prefrontal cortex of brain post-mortem Alzheimer's disease patients, PLXNB1 was upregulated while ADAM22 and ADAM 23 were downregulated similar to severe COVID-19 protein signature (Sup Fig. 10, Supplementary data 16) (Wang et al., 2016;Berchtold et al., 2013;Nativio et al., 2018;Stopa et al., 2018;Kant et al., 2018). Further relevance was evident for GGT5, CD300LF and MSR1 being upregulated in severe COVID-19 and in human ALS spinal cord RNA expression, as well as in a FUS1 overexpression mouse ALS model (Sup Fig. 11 and 12, Supplementary data 16) (Durrenberger et al., 2012;Durrenberger et al., 2015;Funikov et al., 2018). Additionally, RGMB, ADAM23 and ADAM22 were downregulated in Schizophrenia hippocampus RNA expression and severe COVID-19 protein signature (Sup Fig. 13) (Lanz et al., 2019;Topol et al., 2015;Topol et al., 2016;Narla et al., 2017). GFRA3 was downregulated in severe COVID-19 protein signature and correlated with Schizophrenia associated SNP biosets (2681 cases, 2653 controls) detecting the presence of a European ancestry intron variant in rs33967909 (P = 0.001). DRAXIN was upregulated in severe COVID-19 protein signature as well as in schizophrenia-derived iPSC patients and Schizophrenia associated SNP study (Sup Fig. 13, Supplementary data 16) (Stefansson et al., 2009;Sullivan et al., 2008). Correlating biomarker signatures with neurological diseases can be used as a tool to predict the development of neurological complications in post-severe COVID-19 patients, thus these biomarkers might serve as a clinical prognostic tool. Overall, these analyses support a broad potential impact that could provide context to future studies and for understanding long COVID-19 disease with neurological complications. Proteomic profiling is widely used for biomarker discovery, but it can also be used to predict potential drugs for intervention. In our study, we analyzed protein signature-drugs correlations in several independent studies, and we found drugs that either correlated positively or negatively with the protein signature. These correlations can be considered for patients' clinical management, as drugs with positive correlation could be dismissed, since they may aggravate the complications, but those with negative correlation may alleviate the complications by reversing the deregulated protein profiles. Drugs that may reverse COVID-19 expression patterns included risperidone, a selective blocker of dopamine D2 and serotonin 5-HT2 receptors that has been shown to improve positive and negative symptoms of schizophrenia (Janssen et al., 1988). Meta-Analysis of 34 proteins in a risperidone RNA expression time course of human SK-N-SH neuroblastoma cells (GSE36678) showed high COVID-19 induced protein levels for LAYN, TNFRSF12A, EFNA2, CD302, EDA2R and NGF that correlated with downregulated RNA levels after 6, 24 and 48 h risperidone treatment (Mas et al., 2013). Furthermore, GFRA3 had an inverse relationship of downregulated protein and upregulated RNA (Sup Fig. 14, Supplementary data 17). There is a need for more comprehensive neurological cell line models of SARS-CoV-2 infection and compound treatment effects at both the protein and RNA levels to better understand the mechanistic potential for repurposing safe and effective drugs.

Conclusions
Our study identified deregulated neurological proteins in the plasma of severe COVID-19 patients that may inform therapeutic interventions. Our findings revealed novel neurological protein signature for severe COVID-19, affecting various neurological signaling pathways. This protein signature should play diverse roles though important functions in severe COVID-19 neurological complications and emphasize the value of neurological follow-up in recovered individuals. The neurological protein signature was validated in independent cohorts at both protein and RNA levels, and shown to correlate with neurological diseases and pharmacological drugs. Thus, it may serve as prognostic and diagnostic tools for post-severe COVID-19 patients with neurological sequalae, and prospectively help healthcare management and clinical decision-making to provide adequate drugs to prevent or slow down further neurological complications.

Funding
Research conducted on Qatar cohort was funded by the Infectious Disease Interdisciplinary Research Program (QB-10-ID-IDRP) at QBRI and a grant fund from Hamad Medical Corporation (fund number MRC-05-003).

Institutional review board statement
The study was conducted according to the Ministry of Public Health (MOPH) guidelines and approved by the Institutional Review Board Research Ethics Committee of the Hamad Medical Corporation (reference MRC-05-003). Informed consent was obtained from all participants.

Declaration of Competing Interest
None.

Data availability
Data will be made available on request.