Four Years in the Skull

In the meninges and skull bone of COVID survivors, spike protein remains detectable for up to four years after infection, independent of direct viral infection of the brain and likely sustained by viral reservoirs. Meta-analyses covering more than four million patients find memory problems in 27.8% of people with Long COVID, while PET scans still show elevated microglial activation two years after infection. All of this is unfolding without epidemiological surveillance, without a national monitoring program, and inside a public debate in which mainstream institutions avoid the persistence findings to protect the vaccination ledger while vaccine critics downplay the quantitative dominance of infection in order to keep the vaccine in the role of principal culprit.

Spike in the skull bone, four years after infection

#

In 2024, a group of 27 researchers from 17 institutes, coordinated by Helmholtz Munich and LMU Munich, published an analysis in Cell Host & Microbe: SARS-CoV-2 spike protein persists in skull bone marrow and the meninges for up to four years after infection.¹ The accumulation is not a laboratory artifact. It tracks with elevated inflammatory markers in adjacent brain tissue. ACE2 receptors are particularly dense in the skull bone and meninges, which helps explain why the protein accumulates there. mRNA vaccination significantly reduces that accumulation; press materials from Helmholtz Munich put the reduction at roughly 50 percent.² The full paper remains paywalled.

That finding has not entered German public communication. Not as part of the surveillance framework of the Robert Koch Institute, Germany’s federal disease control agency (RKI). Not as the basis for any monitoring program under Federal Health Minister Nina Warken. And in the polarized argument over vaccine injury and Long COVID, it has disappeared into a peculiar void: useless to both camps.

What the imaging shows

#

Douaud et al. analyzed brain scans from 785 UK Biobank participants in Nature in 2022, 401 with SARS-CoV-2 infection and 384 controls, with the crucial methodological advantage that scans existed from both before and after infection.³ The result was a thinner gray matter layer in the orbitofrontal cortex and parahippocampal gyrus, greater tissue damage in olfactory-connected regions, and measurable declines in processing speed. These effects held even after hospitalized cases were excluded. The paper was describing mild to moderate infections.

In 2025, a TSPO-PET study from Amsterdam UMC reinforced the signal: microglial activation in Long COVID patients, measured two years after infection.⁴ TSPO is a marker of activated microglia, the brain’s resident immune cells. In 33 symptomatic Long COVID patients, TSPO binding in global gray matter was significantly higher than in infected but asymptomatic controls (0.80 ± 0.34 vs. 0.65 ± 0.17, p = 0.036). The scans were performed two years after infection. The authors point to structural parallels with neuroinflammatory patterns seen in multiple sclerosis and Alzheimer’s disease.

There are methodological limits. TSPO binding is not a pure microglia marker, since astrocytes and macrophages also express TSPO. The sample contains 47 participants. And there is no healthy uninfected control group. What the study establishes is an inflammatory subtype, not a universal Long COVID signature.

The scale at population level

#

A systematic review covering more than four million patients (BMC Neurology, 2025) reports pooled prevalences among Long COVID patients with at least six months of follow-up: memory problems in 27.8 percent (95% CI: 20.1 to 37.1), cognitive impairment in 27.1 percent (95% CI: 20.4 to 34.9), concentration problems in 23.8 percent, fatigue in 43.3 percent.⁵ Heterogeneity across the underlying studies is high.

The UK’s Office for National Statistics offers a more concrete population snapshot. In Wave 4, covering February to March 2024, 3.3 percent of the population in England and Scotland reported Long COVID, about two million people. Of those, 39.4 percent reported difficulty concentrating. 74.7 percent reported limitations in daily life.⁶

Applied to Germany, with an estimated infection rate of 70 to 80 percent by the end of 2024 and a population of 84 million, the British prevalence rate implies a rough order of magnitude of 2.7 million people living with Long COVID symptoms. How many of them have cognitive impairment is not recorded. The RKI scaled back systematic antibody monitoring in 2023 and does not run a continuing Long COVID surveillance system with cognitive subgroups. As of 2026, Germany has no nationwide baseline for post-SARS-CoV-2 cognitive population health.

The mechanism: spike as a CNS-active agent

#

The Charité finding from Radke et al. (2024, Nature Neuroscience) matters methodologically: no direct neuronal infection by SARS-CoV-2 could be demonstrated.⁷ The neuroinflammation is not running through direct spike contact with neurons. It is driven by systemic inflammatory cascades, activated in the vagal region of the brainstem and propagated through immune-mediated signaling pathways.

The molecular cascades themselves have been identified across several papers:

NLRP3 inflammasome: Spike protein activates the NLRP3 inflammasome in primary human microglia and triggers a pro-inflammatory cytokine cascade, demonstrated in cell culture experiments (Molecular Psychiatry, 2022).⁸

Blood-brain barrier: Radiolabeled spike S1 protein crosses the BBB in mouse models via adsorptive transcytosis (Rhea et al., Nature Neuroscience, 2021).⁹ Cell-culture studies also show BBB dysfunction after spike exposure.¹⁰ Those models do not translate directly into human infection-level concentrations.

MMP-9 from microglia: Kempuraj et al. report significantly elevated matrix metalloproteinase-9 in serum in a small study of 13 Long COVID patients and 13 matched controls, along with dose-dependent MMP-9 release from cultured human microglia after exposure to spike protein.¹¹ MMP-9 is widely used as a marker of BBB disruption and neuroinflammatory activity.

These three mechanisms reinforce one another. BBB dysfunction allows spike entry. NLRP3 activation launches inflammatory cascades. MMP-9 worsens BBB damage. The Charité result, no direct neuronal infection, does not weaken that model. It sharpens it. The damage appears to run through immune cells, not infected neurons.

The source question: where is the spike protein coming from?

#

This is the point no political camp fully states.

Spike protein from vaccination is measurable. In patients with post-vaccination myocarditis, free, antibody-unbound spike has been found in the bloodstream (Yonker et al., Circulation, 2023, n = 16 myocarditis patients; in asymptomatic vaccinated individuals it was not detectable).¹² Yale’s 2025 study on post-vaccination syndrome reports spike persistence beyond 700 days in a clinically conspicuous minority; there is no comparable finding in healthy vaccinated controls.¹³

Spike protein from infection is measurable too, and persists longer in broader cohorts. Pereira de Melo et al. (Viruses, 2025) report spike in 65 percent of Long COVID plasma samples up to 12 months after infection, with viral reservoirs in the lung, heart, kidney, CNS, GI tract, and lymph nodes.¹⁴

The plausible conclusion, and it remains a conclusion because no direct quantitative comparison exists, is that infection is the dominant spike source at population scale. The reasons are straightforward: in vivo viral replication over weeks, persistent tissue reservoirs, and the fact that nearly the entire population has been infected, whereas vaccine-derived spike is produced for a limited period and within a much narrower anatomical range, chiefly the injection site and draining lymph nodes.

One important piece of indirect evidence points in the same direction. A meta-analysis in Nature Communications (2025) found that vaccination reduces Long COVID risk in the Omicron era.¹⁵ If vaccine-derived spike were the primary driver of neuroinflammation, vaccination should worsen Long COVID. What has been observed is the opposite.

The missing study is obvious. There is still no single-cohort comparison of CNS spike burden across naturally infected people without Long COVID, infected people with Long COVID, asymptomatically vaccinated people, and patients with post-vaccination syndrome. That comparison does not exist.

Spikeopathy: findings versus label

#

One term circulates in parts of the literature and across alternative media: “spikeopathy.” It is meant to gather pathologies that treat spike protein, whether from virus or vaccine, as a common causal driver.

The underlying findings it points to are real enough: NLRP3 activation, BBB disruption, MMP-9 release, spike persistence in tissue. In 2025, Andreas Posa published a review article in Annals of Anatomy on the neurological damage associated with spike protein, assembling those mechanisms in one place.¹⁶

The term itself is not solid. “Spikeopathy” has no agreed diagnostic criteria, no ICD-11 code, no standardized biomarker panel, and no standing in WHO, NIH, or EMA terminology. The criticism, including from McGill’s Office for Science and Society, is methodologically sound: the core papers repeatedly rely on formulations like “potential” and “could,” and a catch-all label for heterogeneous mechanisms suggests more conceptual unity than the evidence can currently sustain.

This essay names the findings. It does not rely on the label.

What helps: intervention evidence

#

None of the following interventions has yet produced a completed randomized controlled trial for Long COVID neuroinflammation. That is the present state of the evidence.

Low-dose naltrexone (LDN): Three randomized controlled trials are ongoing, including NCT05430152 in British Columbia with a brain-MRI sub-study. Observational data from case series and cohort studies are positive. Mechanistically, LDN is plausible because it modulates microglia through Toll-like receptor 4 antagonism. A Frontiers paper from 2025 reports restoration of TRPM3 ion channel function in NK cells from Long COVID patients treated with LDN.¹⁷ Evidence grade: IV, with RCT results still pending.

Omega-3 (EPA/DHA): No Long COVID-specific RCT was identified. The mechanism is nevertheless well grounded, since EPA and DHA modulate microglia through specialized pro-resolving mediators. A 2025 meta-analysis reports positive signals in Alzheimer’s subgroups. Evidence grade for Long COVID: indirect, III to IV.

Sleep: Sleep deprivation drives microglia into a pro-inflammatory state and raises CRP, TNF-alpha, IL-1, and IL-6 (Nature Communications, 2022). Sleep disturbance is common in Long COVID and can amplify pre-existing neuroinflammation. Evidence grade: III, based on cohort work and animal models.

Melatonin: A non-enzymatic CNS antioxidant that reduces pro-inflammatory cytokines in preclinical models. A meta-analysis in Brain, Behavior and Immunity (2021) found a significant reduction in inflammatory markers across clinical studies. No Long COVID-specific RCT exists. Evidence grade for Long COVID: indirect, III.

The first federally funded RCT for cognitive Long COVID symptoms, NIH RECOVER-NEURO (JAMA Neurology, 2025), tested three non-drug interventions and found no superior efficacy for any one of them.¹⁸

The political silence

#

The RKI scaled back systematic antibody monitoring in 2023. It has no ongoing national program tracking cognitive subgroups among Long COVID patients. Germany cannot quantify the epidemiology of cognitive damage after SARS-CoV-2, not because the data would be impossible to gather, but because no collection system exists.

Germany’s Federal Ministry of Health, under Nina Warken, has announced no monitoring program for post-COVID cognitive population health. COVIDOM+ (EUR 4.9 million from the Federal Ministry of Health in 2025) and Charité’s NKSG program (EUR 10 million from the Federal Ministry of Education and Research) fund research, not surveillance.

The debate that exists instead looks like this:

Mainstream institutions: Long COVID as a physical disease entity is now established in German journalism. Spike persistence as a neuroinflammatory issue barely appears there. The implicit logic is easy to see: openly communicating spike as a persistent neuroinflammatory agent would place the vaccination ledger uncomfortably close to the problem, even if the source question is answered in favor of infection.

The vaccine-critical sector: The focus is on post-vaccination syndrome and vaccine-derived spike. The quantitative dominance of infection as a spike source, and the evidence that vaccination reduces Long COVID risk, rarely fit that frame. Both silences are narratively convenient.

If 2 to 3 percent of a population lives with cognitive impairment, that share becomes statistically invisible inside productivity and health data as long as no attribution system exists. The baseline shifts before anyone measures it. The RKI knows this and still does not run the system.

Findings

#

Four statements are true independently of one another.

Spike protein persists in skull bone and meninges. Microglia remain activated two years after infection. Memory problems affect millions at population scale. No national surveillance system tracks it.

They are connected by a common mechanism. The combination is what is missing, and that absence is not a gap. It is a decision.