Ethics statement
Patients and controls in FinnGen provided informed consent for biobank research, based on the Finnish Biobank Act (https://www.finngen.fi/fi). All DNA samples and data were pseudonymised. Participants of the RECOVID-study [protocol approved by Ethical board of Helsinki University Hospital (HUS/1949/2020)] provided written informed consent. Clinical autopsies were performed at the Department of Pathology, HUS Diagnostic Center, Helsinki University Hospital according to Finnish legislation, with consent granted by the next of kin. The protocol covering COVID-19 autopsies was approved by the Ethics Committee of Helsinki University Hospital (HUS/1238/ 2020).
FinnGen biobank resource
We extracted COVID-19 cases and controls from FinnGen (Release (R) 7, cutoff date March 27th 2021), which comprises prospective epidemiological cohorts (initiated in 1992), disease-based cohorts, and hospital biobank samples. All cases were extracted based on the same criteria: cases were FinnGen samples who according to the National Infectious Disease Register had had COVID-19 and from whom information on disease severity (as per level of care: home, hospital ward or intensive care unit) was available. The genotype data is integrated via a unique national personal identification number with national registries such as the hospital discharge registry, national death registry and medication reimbursement registry. Information on COVID-19 is piped into FinnGen via the national register of infectious diseases. FinnGen R7 included 3469 individuals who had a diagnosis of COVID-19. We excluded 858 cases lacking information on possible hospitalisation. We divided the remaining 2,611 cases into three groups based on the level of care as a surrogate for the severity of disease: home-isolation, hospitalisation but no intensive care, and intensive care. We selected controls from the large nationwide population-based cohorts (FINRISK1992-2012 [15], FINHEALTH2017 [16], HEALTH2000/2012 [17]) included in FinnGen. These cohorts included 43,884 individuals from whom we excluded five individuals without phenotype data and 340 individuals who had had COVID-19, leaving 43,539 individuals as controls. Then we used the MatchIt v.4.1.0 package [18] as implemented in R [19] to match all three case groups (at-home-treatment, hospitalisation, intensive care) separately to controls. Matching was performed based on age (age at death/loss-of-follow up or current age) and sex using default parameters, except that five controls were matched to every case. In a complementary analysis, we added a composite phenotype variable of any cardiovascular disease (FinnGen R7 phenotype code FG_CVD that includes hypertensive diseases, ischaemic and other heart diseases, pulmonary embolism, cerebrovascular diseases, diseases of arteries/arterioles/capillaries, diseases of veins/lymphatic vessels and lymph nodes, and other cardiovascular diseases) as an additional matching term.
APOE genotype was derived from the single-nucleotide polymorphisms rs429358 and rs7412 that were genotyped in FinnGen using a microarray or imputed using a Finnish reference panel. The APOE ε1/ε3 genotype is indistinguishable from the ε2/ε4 genotype but since the ε1 is exceedingly rare, the ε2/ε4 genotype was used for all individuals with the above allele combination.
Recovery after critical COVID-19 infection (RECOVID) study design and subjects
Participants of the RECOVID-study were asked for separate written informed consent for APOE analysis. All participants were adult (≥ 18 years of age) native Finnish-speakers. Prior major neurological disease (e.g., traumatic brain injury, dementia, stroke or Parkinson´s disease) or developmental disability were exclusion criteria. All hospitalised COVID-19 patients with a diagnosis between March 1 and December 31, 2020 were recruited within three months from hospital discharge by mailed invitation or personal contact during a follow-up visit. Home-isolated patients with confirmed SARS-CoV-2 infection (RT-PCR or antibody test) from the same time period were recruited within three months from diagnosis by press and online announcements. Non-infected controls were recruited by online announcements in early 2021. Demographic data and blood samples were collected from all participants, and clinical data of the inpatient period was recorded from hospital-treated cohorts. One-hundred and fifty-six study participants (of whom 108 were COVID-19 patients and 48 controls) completed written questionnaires including the multi-dimensional fatigue scale (MFI-20) at six months post-discharge from hospital or recovery from acute COVID-19 (Additional file 1: Supplementary Fig. 1).
Autopsies
Clinical autopsies were performed at the Department of Pathology, HUS Diagnostic Center according to Finnish legislation, with consent granted by the next of kin. The modified full autopsies were carried out in an autopsy room appropriate for handling infectious decedents, with appropriate personal protective equipment. All autopsies included the complete exploration of the visceral cavity and craniotomy. Modifications included extended neuropathological sampling (samples from 3–12 brain areas) in those 10 cases where a full neuropathological examination was not performed.
Neuropathological examination
Full neuropathological examination of the brain was performed on 11 cases. Each brain was dissected after at least 10 days of formaldehyde-fixation and at least 15 samples from different brain areas were collected. H&E stainings were performed using standardised protocols at the Department of Pathology, Helsinki University Hospital.
Immunohistochemistry (IHC)
Beta-amyloid and CD68 immunostainings were performed at the Department of Pathology, Helsinki University Hospital using standard protocols (details in Additional file 1: Supplementary Table 1).
Histopathological analysis of perivascular haemorrhage (“Bleed Grade” and “Bleed Score”)
Bleed Grade
The extent of perivascular haemorrhage was graded in brain samples from midbrain, pons, medulla, cerebellum, frontal cortex and basal ganglia by two pathologists (JK, HP), reaching consensus using a consultation microscope (Nikon Eclipse i80; eye-pieces CFI 10x/22; objectives: Nikon Plan Fluor 20x/0.50, Nikon Plan Fluor 10x/0.30, Nikon Plan Fluor 4x/0.13, Nikon Plan UW 2x/0.06). Grades were defined as follows: 3 = unequivocal perivascular haemorrhage in two or more foci as seen in one field-of-view (FOV) using a 2 × objective, 2 = unequivocal perivascular haemorrhage as seen in one FOV using a 2 × objective, 1 = unequivocal perivascular haemorrhage as seen in one FOV using a 4 × objective, 0 = no unequivocal perivascular haemorrhage in any FOV as seen using a 4 × objective. The pathologists graded the samples blinded to the information on genotype.
Bleed Score
The number of haemorrhagic foci were determined in brain samples from midbrain, pons, medulla, cerebellum, frontal cortex and basal ganglia by one pathologist (JK) using the same microscope as in the Bleed Grade. Scoring was performed as follows: four parenchymal hot spots for haemorrhage (total FOV 95 mm2) were scored using the 4 × objective and counting the number of haemorrhagic foci. Perivascular haemorrhage was defined as an enlarged perivascular space with red blood cells or enlarged perivascular space with debris and pigmented macrophages. The border of the perivascular space also had to show evidence of neuropil degeneration or vacuolation. Meningeal haemorrhage was not counted. The pathologist scored the samples blinded to information on genotype.
Analysis of cerebral amyloid angiopathy (CAA)
CAA was determined by a neuropathologist (LM) based on beta-amyloid-immunopositivity in at least one vessel on sections from frontal cortex and cerebellum.
Histopathological analysis of microglial reactivity
CD68-stained sections from midbrain, pons and medulla were assessed by two pathologists (JK, HP) using a modified version of the method described by Poloni and colleagues [20]. Briefly, the CD68-stained slide was assessed for representative areas using a 4 × objective, and five FOVs were scored according to the 4-grade scheme using a 10 × objective (FOV 3.8 mm2), with confirmatory use of higher magnification where needed. The 4-point scale (0–3) used was identical to the work by Poloni: 0 = absence of both perivascular infiltrate and microglial nodules and < 20 amoeboid cells/reactive microglial cells; 1 = presence of at least one perivascular infiltrate or 1 micronodule or > 20 amoeboid cells/reactive microglial cells; 2 = presence of 2–4 microglial nodules; and 3 = presence of > 4 microglial nodules. The pathologists graded the samples blinded to information on genotype, and grade was agreed upon by consensus at the consulting microscope described above.
In addition, a count of stained foci was determined by one pathologist (JK) in 3 hot spot areas using a 20 × objective (FOV 0.95 mm2). Where a single cell could be identified, the stained cells were counted. The pathologist scored the samples blinded to information on genotype.
Sanger sequencing
We genotyped APOE using Sanger sequencing. All consented participants of the RECOVID study gave venous blood samples from which we extracted DNA using standard methods. In the AUTOPSY cohort, DNA was isolated from frozen tissues collected at autopsy using standard protocols. We amplified DNA fragments containing rs429358 and rs7412 using previously published primers [21] that amplify rs429358 and rs7412 on the same DNA fragment. The amplification reaction contained 1X DreamTaq Buffer, 0.35 mM of dNTP mix, 0.6 μM of forward and reverse primers, 1 M of Betaine, 1.25U of DreamTaq DNA Polymerase (Applied Biosystems, ThermoFisher Scientific, Waltham, MA, USA), approximately 50 ng of template DNA and nuclease-free water to attain a total volume of 20 μl. The cycling conditions have been previously published [21]. We then purified the amplicons using Exonuclease I—Shrimp Alkaline Phosphatase Clean Up (Applied Biosystems) according to the manufacturer’s protocol. We then sequenced the purified amplicons using BigDye Terminator v3.1 Cycle sequencing kit (Applied Biosystems) according to the manufacturer’s protocol with the forward primer. Sequences were analysed on a ABI3730XL DNA Analyzer (Applied Biosystems) at the Finnish Institute of Molecular Medicine (FIMM). We determined the genotypes of rs429358 and rs7412 using Sequencher (Sequencher® version 5.1 DNA sequence analysis software, Gene Codes Corporation, Ann Arbor, MI USA).
Fatigue testing (Multidimensional Fatigue Inventory MFI-20)
We evaluated chronic fatigue at a 6-month follow-up using the Multidimensional Fatigue Inventory (MFI-20) score [22]. The questionnaire was translated into Finnish and was a part of a larger set of questions concerning neuropsychological recovery and mental health. MFI-20 measures five subscales of fatigue (General Fatigue, Physical Fatigue, Mental Fatigue, Reduced Motivation, Reduced Activity), each of which is probed with four questions. The subject specifies the extent to which each statement relates to her/him on a 5-point Likert scale, ranging from “Yes, that is true” to “No, that is not true”. The range of possible scores is 4–20 in each fatigue subscale, and 20–100 in total fatigue, with a higher score signifying a higher level of fatigue.
Statistical analysis of the AUTOPSY and RECOVID cohorts
All statistical analyses were performed using GraphPad Prism (version 8.4.2 for Windows, GraphPad Software, San Diego, California USA, www.graphpad.com), SPSS (IBM Corp. Released 2020. IBM SPSS Statistics for Windows, Version 27.0. Armonk, NY: IBM Corp), and the statistical software R [19].
We used Fisher’s exact test as implemented in R’s stats package to test if carrying at least one APOE ε4 allele associated with each COVID-19 case group. We used linear mixed-effect model as implemented in R’s nlme (Pinheiro J, Bates D, DebRoy S, Sarkar D, R Core Team (2021). nlme: Linear and Nonlinear Mixed Effects Models. R package version 3.1–153, https://CRAN.R-project.org/package=nlme) and cumulative link mixed model as implemented in R’s ordinal (Christensen, R. H. B. (2019). ordinal—Regression Models for Ordinal Data. R package version 2019.12–10, https://CRAN.R-project.org/package=ordinal) package to determine the statistical significances of perivascular bleed score and bleed grade depending on APOE4 carriership, respectively. Additionally, we used exact Pearson’s chi-square test and Mann–Whitney U test as implemented in SPSS to determine the statistical significance of perivascular bleed grade and bleed score depending on APOE4 carriership specifically in different brain areas, respectively. All patients who filled in the MFI-20 questionnaire were included in the analysis. We used a negative binomial regression model as implemented in the MASS [23] package in R to analyse factors associated with the MFI-20 fatigue score in both univariate and multivariate analyses (Tables 2 and 3). Negative binomial regression was chosen over Poisson regression to compensate for the overdispersion in data. COVID-19 severity was classified into three categories, as done in a previously published analysis of fatigue associated with long-COVID [24]. The categories were: 0 – noninfected controls, 1 – home-isolated COVID-19, 2 – hospitalised COVID-19. As COVID-19 severity was the only non-binary categorical variable in multivariate binomial regression model and reached statistical significance in the model in one but not all factor levels, its independence was further analysed by comparison (ANOVA) with the same model with severity excluded.