When Lara Hawthorne, an illustrator in Bristol, UK, began developing strange symptoms after having COVID-19, she hoped that they weren’t due to the virus. Her initial illness had been mild. “I’ve been triple vaccinated. I felt quite protected,” she says. But months later, she was still sick with a variety of often debilitating symptoms: earaches, tinnitus, congestion, headaches, vertigo, heart palpitations, muscle pain and more. On some days, Hawthorne felt so weak that she could not get out of bed. When she finally saw her physician, the diagnosis was what she had been dreading: long COVID.
Unable to find relief, she became increasingly desperate. After reading an opinion piece in The Guardian newspaper about how blood clots might be to blame for long COVID symptoms, Hawthorne contacted a physician in Germany who is treating people with blood thinners and a procedure to filter the blood. She hasn’t heard back yet — rumour has it that people stay on the waiting list for months — but if she has the opportunity to head there for these unproven treatments, she probably will. “I don’t want to wait on my health when I’m feeling so dreadful,” she says.
Researchers are baffled by long COVID: hundreds of studies have tried to unpick its mechanism, without much success. Now some scientists, and an increasing number of people with the condition, have been lining up behind the as-yet-unproven hypothesis that tiny, persistent clots might be constricting blood flow to vital organs, resulting in the bizarre constellation of symptoms that people experience.
Proponents of the idea (#teamclots, as they sometimes refer to themselves on Twitter) include Etheresia Pretorius, a physiologist at Stellenbosch University in South Africa, and Douglas Kell, a systems biologist at the University of Liverpool, UK, who led the first team to visualize micro-clots in the blood of people with long COVID. They say that the evidence implicating micro-clots is undeniable, and they want trials of the kinds of anticoagulant treatment that Hawthorne is considering. Pretorius penned the Guardian article that caught Hawthorne’s attention.
But many haematologists and COVID-19 researchers worry that enthusiasm for the clot hypothesis has outpaced the data. They want to see larger studies and stronger causal evidence. And they are concerned about people seeking out unproven, potentially risky treatments.
When it comes to long COVID, “we’ve now got little scattered of bits of evidence”, says Danny Altmann, an immunologist at Imperial College London. “We’re all scuttling to try and put it together in some kind of consensus. We’re so far away from that. It’s very unsatisfying.”
Cascade of clots
Pretorius and Kell met about a decade ago. Pretorius had been studying the role of iron in clotting and neglected to cite some of Kell’s research. When he reached out, they began chatting. “We had a Skype meeting and then we decided to work together,” Pretorius says. They observed odd, dense clots that resist breaking down for years in people with a variety of diseases. The research led them to develop the theory that some molecules — including iron, proteins or bits of bacterial cell wall — might trigger these abnormal clots.
Blood clotting is a complex process, but one of the key players is a cigar-shaped, soluble protein called fibrinogen, which flows freely in the bloodstream. When an injury occurs, cells release the enzyme thrombin, which cuts fibrinogen into an insoluble protein called fibrin. Strands of fibrin loop and criss-cross, creating a web that helps to form a clot and stop the bleeding.
Under a microscope, this web typically resembles “a nice plate of spaghetti”, Kell says. But the clots that the team has identified in many inflammatory conditions look different. They’re “horrible, gunky, dark”, Kell says, “such as you might get if you half-boiled the spaghetti and let it all stick together.” Research by Kell, Pretorius and their colleagues suggests that the fibrin has misfolded1, creating a gluey, ‘amyloid’ version of itself. It doesn’t take much misfolding to seed disaster, says Kell. “If the first one changes its conformation, all the others have to follow suit”, much like prions, the infectious misfolded proteins that cause conditions such as Creutzfeldt–Jakob disease.
Pretorius first saw these strange, densely matted clots in the blood of people with a clotting disorder2, but she and Kell have since observed the phenomenon in a range of conditions1 — diabetes, Alzheimer’s disease and Parkinson’s disease, to name a few. But the idea never gained much traction, until now.
When the pandemic hit in 2020, Kell and Pretorius applied their methods almost immediately to people who had been infected with SARS-CoV-2. “We thought to look at clotting in COVID, because that is what we do,” Pretorius says. Their assay uses a special dye that fluoresces when it binds to amyloid proteins, including misfolded fibrin. Researchers can then visualize the glow under a microscope. The team compared plasma samples from 13 healthy volunteers, 15 people with COVID-19, 10 people with diabetes and 11 people with long COVID3. For both long COVID and acute COVID-19, Pretorius says, the clotting “was much more than we have previously found in diabetes or any other inflammatory disease”. In another study4, they looked at the blood of 80 people with long COVID and found micro-clots in all of the samples.
So far, Pretorius, Kell and their colleagues are the only group that has published results on micro-clots in people with long COVID.
But in unpublished work, Caroline Dalton, a neuroscientist at Sheffield Hallam University’s Biomolecular Sciences Research Centre, UK, has replicated the results. She and her colleagues used a slightly different method, involving an automated microscopy imaging scanner, to count the number of clots in blood. The team compared 3 groups of about 25 individuals: people who had never knowingly had COVID-19, those who had had COVID-19 and recovered, and people with long COVID. All three groups had micro-clots, but those who had never had COVID-19 tended to have fewer, smaller clots, and people with long COVID had a greater number of larger clots. The previously infected group fell in the middle. The team’s hypothesis is that SARS-CoV-2 infection creates a burst of micro-clots that go away over time. In individuals with long COVID, however, they seem to persist.
Dalton has also found that fatigue scores seem to correlate with micro-clot counts, at least in a few people. That, says Dalton, “increases confidence that we are measuring something that is mechanistically linked to the condition”.
In many ways, long COVID resembles another disease that has defied explanation: chronic fatigue syndrome, also known as myalgic encephalomyelitis (ME/CFS). Maureen Hanson, who directs the US National Institutes of Health (NIH) ME/CFS Collaborative Research Center at Cornell University in Ithaca, New York, says that Pretorius and Kell’s research has renewed interest in a 1980s-era hypothesis about abnormal clots contributing to symptoms. Pretorius, Kell and colleagues found amyloid clots in the blood of people with ME/CFS, but the amount was much lower than what they’ve found in people with long COVID5. So clotting is probably only a partial explanation for ME/CFS, Pretorius says.
Where these micro-clots come from isn’t entirely clear. But Pretorius and Kell think that the spike protein, which SARS-CoV-2 uses to enter cells, might be the trigger in people with long COVID. When they added the spike protein to plasma from healthy volunteers in the laboratory, that alone was enough to prompt formation of these abnormal clots6.
Bits of evidence hint that the protein might be involved. In a preprint7 posted in June, researchers from Harvard University in Boston, Massachusetts, reported finding the spike protein in the blood of people with long COVID. Another paper8 from a Swedish group showed that certain peptides in the spike can form amyloid strands on their own, at least in a test tube. It’s possible that these misfolded strands provide a kind of template, says Sofie Nyström, a protein chemist at Linköping University in Sweden and an author of the paper.
A California-based group found that fibrin can actually bind to the spike. In a 2021 preprint9, it reported that when the two proteins bind, fibrin ramps up inflammation and forms clots that are harder to degrade. But how all these puzzle pieces fit together isn’t yet clear.
If the spike protein is the trigger for abnormal clots, that raises the question of whether COVID-19 vaccines, which contain the spike or instructions for making it, can induce them as well. There’s currently no direct evidence implicating spike from vaccines in forming clots, but Pretorius and Kell have received a grant from the South African Medical Research Council to study the issue. (Rare clotting events associated with the Oxford–AstraZeneca vaccine are thought to happen through a different mechanism (Nature596, 479–481; 2021).)
Raising safety concerns about the vaccines can be uncomfortable, says Per Hammarström, a protein chemist at Linköping University and Nyström’s co-author. “We don’t want to be over-alarmist, but at the same time, if this is a medical issue, at least in certain people, we have to address that.” Gregory Poland, director of the Mayo Clinic’s vaccine research group in Rochester, Minnesota, agrees that it’s an important discussion. “My guess is that spike and the virus will turn out to have a pretty impressive list of pathophysiologies,” he says. “How much of that may or may not be true for the vaccine, I don’t know.”
Dearth of data
Many researchers find it plausible and intriguing that micro-clots could be contributing to long COVID. And the hypothesis does seem to fit with other data that have emerged on clotting. Researchers already know that people with COVID-19, especially severe disease, are more likely to develop clots. The virus can infect cells lining the body’s 100,000 kilometres of blood vessels, causing inflammation and damage that triggers clotting.
Those clots can have physiological effects. Danny Jonigk, a pathologist at Hanover Medical School in Germany, and his colleagues looked at tissue samples from people who died of COVID-19. They found micro-clots and saw that the capillaries had split, forming new branches to try to keep oxygen-rich blood flowing10. The downside was that the branching introduces turbulence into the flow that can give rise to fresh clots.
Several other labs have found signs that, in some people, this tendency towards clotting persists months after the initial infection. James O’Donnell, a haematologist and clotting specialist at Trinity College Dublin, and his colleagues found11 that about 25% of people who are recovering from COVID-19 have signs of increased clotting that are “quite marked and unusual”, he says.
What is less clear is whether this abnormal clotting response is actually to blame for any of the symptoms of long COVID, “or is it just, you know, another unusual phenomenon associated with COVID?” O’Donnell says.
Alex Spyropoulos, a haematologist at the Feinstein Institutes for Medical Research in New York City, says the micro-clot hypothesis presents “a very elegant mechanism”. But he argues that much more work is needed to tie the lab markers to clinical symptoms. “What’s a little bit disturbing is that these authors and others make huge leaps of faith,” Spyropoulos says.
Jeffrey Weitz, a haematologist and clotting specialist at McMaster University in Hamilton, Canada, points out that the method Pretorius’s team is using to identify micro-clots “isn’t a standard technique at all”. He adds: “I’d like to see confirmation from other investigators.” Micro-clots are difficult to detect. Pathologists can spot them in tissue samples, but haematologists tend to look for markers of abnormal clotting rather than the clots themselves.
Other, larger studies of long COVID have failed to find signs of clotting. Michael Sneller, an infectious-disease specialist, and his colleagues at the NIH in Bethesda, Maryland, thoroughly examined 189 people who had been infected with SARS-CoV-2, some with lingering symptoms and some without, and 120 controls12. They did not specifically look for micro-clots. But if micro-clots had been clogging the capillaries, Sneller says, they should have seen some evidence — tissue damage in capillary-rich organs such as the lungs and kidneys, for example. Micro-clots might also damage red blood cells, leading to anaemia. But Sneller and his colleagues found no signs of this in any of the lab tests.
Kell and Pretorius argue that just because this study didn’t find any evidence of micro-clots doesn’t mean they aren’t there. One of the key issues with long COVID is that “every single test comes back within the normal ranges”, Pretorius says. “You have desperately ill patients with no diagnostic method.” She hopes that other researchers will read their papers and attempt to replicate their results. “Then we can have a discussion,” she says. The ultimate causal proof, she adds, would be people with long COVID feeling better after receiving anticoagulant therapies.
There is some limited evidence of this. In an early version of a preprint, posted in December 2021, Kell, Pretorius and other researchers, including physician Gert Jacobus Laubscher at Stellenbosch University, reported that 24 people who had long COVID and were treated with a combination of two antiplatelet therapies and an anticoagulant experienced some relief13. Participants reported that their main symptoms resolved and that they became less fatigued. They also had fewer micro-clots. Pretorius and Kell are working to gather more data before they try to formally publish these results. But other physicians are already using these medications to treat people with long COVID. Some are even offering a dialysis-like procedure that filters fibrinogen and other inflammatory molecules from the blood. To O’Donnell, such treatment feels premature. He accepts that some people with long COVID are prone to clots, but leaping from a single small study to treating a vast number of people is “just not going to wash in 2022 in my book”, he says. Sneller agrees. “Anticoagulating somebody is not a benign thing. You basically are interfering with the blood’s ability to clot,” he says, which could make even minor injuries life-threatening.
Kell says he’s tired of waiting for a consensus on how to treat long COVID. “These people are in terrible pain. They are desperately unwell,” he says. Altmann understands that frustration. He gets e-mails almost daily, asking: “Where are the drug trials? Why does it take so long?” But even in the midst of a pandemic, he argues, researchers have to follow the process. “I’m not rubbishing anybody’s data. I’m just saying we’re not there yet,” he says. “Let’s join up the dots and do this properly.”
A team of international researchers may have uncovered the cause of the neurological conditions seen in patients with long-Covid, such as brain fog. The team from Swinburne University of Technology and La Trobe University in Australia and Luxembourg University in Luxembourg revealed that fragments of proteins from the SARS-CoV-2 virus can form amyloid clumps in the brain that look similar to the amyloids found in patients with neurodegenerative diseases such as Alzheimer’s and Parkinson’s.
Further, the study published in the journal Nature Communications showed that these amyloids are highly toxic to brain cells. To understand, the team designed, performed and analysed the biochemical flow cytometry assays used to determine the mechanism of brain cell death triggered by the amyloids and assisted with physical characterisation of the amyloids at the Australian Synchrotron.
“If further studies are able to prove that the formation of these amyloids is causing long-Covid then anti-amyloid drugs developed to treat Alzheimer’s might be used to treat some of the neurological symptoms of long-Covid,” said Dr Mirren Charnley, a postdoctoral researcher at Swinburne.
Long-Covid is marked by neurological symptoms, such as memory loss, sensory confusion, severe headaches, and even stroke.
These neurological symptoms are similar to the early stages of neurodegenerative diseases such as Alzheimer’s and Parkinson’s, which are characterised by the presence of clumps of ordered proteins a” known as amyloids – in the brain.
The long-Covid symptoms can persist for months after the infection is over. While there is evidence that the virus can enter the brain of infected people, the precise mechanisms causing these neurological symptoms are unknown.
COVID-19 is primarily known as a respiratory disease caused by SARS-CoV-2. However, neurological symptoms such as memory loss, sensory confusion, severe headaches, and even stroke are reported in up to 30% of cases and can persist even after the infection is over (long COVID). These neurological symptoms are thought to be produced by the virus infecting the central nervous system, however we don’t understand the molecular mechanisms triggering them. The neurological effects of COVID-19 share similarities to neurodegenerative diseases in which the presence of cytotoxic aggregated amyloid protein or peptides is a common feature. Following the hypothesis that some neurological symptoms of COVID-19 may also follow an amyloid etiology we identified two peptides from the SARS-CoV-2 proteome that self-assemble into amyloid assemblies. Furthermore, these amyloids were shown to be highly toxic to neuronal cells. We suggest that cytotoxic aggregates of SARS-CoV-2 proteins may trigger neurological symptoms in COVID-19.
The disease caused by viral infection with severe acute respiratory syndrome (SARS)-COV-2 is known as COVID-19 and whilst predominantly a respiratory disease affecting the lungs it has a remarkably diverse array of symptoms. These include a range of moderate to severe neurological symptoms reported in as many as 30% of patients, which can persist for up to 6 months after infection1. These symptoms include memory loss, sensory confusion (e.g., previously pleasant smells become fixed as unpleasant), cognitive and psychiatric issues, severe headaches, brain inflammation and haemorrhagic stroke1,2,3,4,5. COVID-19-related anosmia and phantosmia have been shown to correlate with a persistence of virus in the olfactory mucosa and in the olfactory bulb of the brain, and with persistent inflammation; however, negative evidence for continuing viral replication has also been shown for long-term anosmia6. Furthermore, there is evidence that SARS-CoV-2 is neuroinvasive with either the full virus7,8 or viral proteins8 being found in the CNS of mouse models and the post-mortem brain tissue of COVID-19 patients. Whilst the neuroinvasiveness of SARS-CoV-2 is apparent the molecular origin of the associated neurological symptoms is as yet unknown, although they are similar to hallmarks of amyloid-related neurodegenerative diseases such as Alzheimer’s (AD)9,10, and Parkinson’s11. For instance, impaired olfactory identification ability and mild cognitive impairment have also been reported in the early stages of AD and prodromal AD12.
A number of in vitro studies have shown that proteins from SARS-CoV-2 can detrimentally affect a variety of cell types including kidney, liver and immune cells13,14. Furthermore, experiments on brain organoids show that SARS-CoV-2 can infect neuronal cells resulting in cell death15. Combined these papers point to a potential cytotoxic cause of neurological symptoms in COVID-19.
Proteins from the Zika virus16 and also the coronavirus responsible for the SARS outbreak in 2003 (SARS-COV-1)17 have been shown to contain sequences that have a strong tendency to form amyloid assemblies. As the proteome of SARS-CoV-1 and SARS-CoV-2 possess many similarities18, we propose amyloid nanofibrils formed from proteins in SARS-CoV-2 may be implicated in the neurological symptoms in COVID-19. Therefore, amyloid-forming proteins from the SARS-CoV-2 virus in the CNS of COVID-19 infected patients could have similar cytotoxic and inflammatory functions to amyloid assemblies that are the molecular hallmarks of amyloid-related neurodegenerative diseases such as AD (Aβ, Tau) and Parkinson’s (α-synuclein). The worst-case scenario given the present observations is that of the progressive neurological amyloid disease being triggered by COVID-19. To the authors’ knowledge, there has so far been no documented example of this; however, it has been noted that up-regulation of Serum amyloid A protein driven by inflammation in COVID-19 seems like a probable trigger for the systemic (non-neurological) amyloid disease AA amyloidosis19, which is already known to be a concomitant of inflammatory disease in general.
If the proteome of SARS-COV-2 does contain amyloid-forming sequences, this raises the question, what is their function? It is known that viral genomes evolve rapidly and are highly constrained by size; therefore, every component is typically functional either to help the virus replicate or to impede the host immune system. To this end, there are several potential roles for amyloid assemblies in pathogens generally20 and specifically in coronaviruses such as SARS-CoV-2. The simplest is that amyloid is an inflammatory stimulus21, and proinflammatory cytokines can up-regulate the expression of the spike protein receptor ACE-2 such that intercellular transmissibility of SARS-CoV-2 is increased. Alternatively Tayeb-Fligelman et al.22 found that the nucleocapsid protein in SARS-CoV-2, which is responsible for packaging RNA into the virion, contains a number of highly amyloidogenic short peptide sequences within its intrinsically disordered regions22. It has been shown that the self-assembly of these peptides is enhanced in the presence of viral RNA, during liquid–liquid phase separation (LLPS is an important stage in the viral replication cycle)23,24. These findings suggest amyloids may play an important role in RNA binding and packaging during the viral replication cycle. Finally, it is also possible that amyloid assemblies in coronaviruses might have a role in inhibiting the action of the host antiviral response similar to a discussed role for amyloid in other viruses. Pham et al.25 observed that amyloid aggregates from murine cytomegalovirus can interfere with RIPK3 kinase activation and potentially inhibit its antiviral immune signalling capabilities.
In this study, we choose to focus on a selection of proteins from the SARS-CoV-2 proteome known as the open reading frames (ORFs). These ORF proteins were chosen as they have no obvious roles in viral replication26, perhaps freeing them up to have yet uncharacterised roles in disrupting the host antiviral responses. By sequence and length, they appear to be largely unstructured, making them good candidates for amyloid formation in vivo. We performed a bioinformatic screening of the ORF proteins to look for potential amyloidogenic peptide sequences. This analysis was used to select two sub-sequences, one each from ORF6 and ORF10, for synthesis. The synthesised peptides were both found to rapidly self-assemble into amyloid assemblies with a variety of polymorphic morphologies. Cytotoxicity assays on neuronal cell lines showed these peptide assemblies to be highly toxic at concentrations as low as 0.0005% (0.04 mg mL−1).
Since commencing this work, others have found that ORF6 is the most cytotoxic single protein of the SARS-CoV-2 proteome, showing localisation to membranes when overexpressed in human and primate immune cell lines13. In contrast, ORF10 has been reported as an unimportant gene with very low expression and no essential role in virus replication26; however, the functions of immune suppression or inflammation promotion via amyloid formation would be non-essential, if present, and should not necessarily require transcription in large volumes, making ORF10 an intriguing second candidate for the present study. It is also interesting that ORF10 and ORF8 are the only two coded proteins present in SARS-CoV-2 which do not have a homologue in SARS-CoV-127, perhaps suggesting a unique amyloid etiology for COVID-19. While long-term consequences from SARS-CoV-1 infection were severe, including tiredness, depression, and impaired respiration, few or zero unequivocally neurological post-viral symptoms were recorded from the (admittedly quite small) set of documented cases28.
Results and discussion
Amyloid aggregation prediction algorithms identified two short peptides from ORF6 and ORF10 that are likely to form amyloids
Figure 1 shows selected output from bioinformatics tools applied to predict the amyloidogenicity of peptide sequences within larger polypeptides. Application of the ZIPPER tool to ORF6 provides more than ten choices of six-residue windows of the sequence predicted to be highly amyloidogenic (Fig. 1a), while ORF10 shows only three such highly amyloidogenic sequence windows (Fig. 1b). To narrow down our search for candidate peptides we also used the TANGO algorithm, for ORF6 there are two regions that are predicted to be highly aggregation prone, I14LLIIMR and D30YIINLIIKNL. The region I14-R20 overlaps almost perfectly with the hexapeptide I14LLIIM identified by ZIPPER. The region 30–40 also contains multiple hits in ZIPPER, but as this study was limited to two candidate peptides we chose ILLIIM as our first candidate as it closely resembles the sequence ILQINS from Hen Egg White Lysozyme that has previously been seen to be highly amyloidogenic (the mutation TFQINS in human lysozyme is disease-linked)29,30,31. Looking now at the TANGO plots for ORF10 the main aggregation-prone sequence is residues F11TIYSLLLC, although there are no high stability hexapeptides in this sequence predicted by ZIPPER. The octapeptide R24NYIAQVD was chosen due to its zwitterionic residue pair R-D which should strongly enhance interpeptide association, despite being too far apart in the sequence to trigger the highly local bioinformatics algorithms. Encouragingly ZIPPER also predicts the hexapeptide NYIAQV contained within RNYIAQVD to be highly amyloidogenic. Based on the outputs from ZIPPER and TANGO and also on the experience in making and studying amyloid, we selected RNYIAQVD and ILLIIM to be synthesised and their amyloid-forming capability investigated.
Nanoscale imaging reveals both peptide sequences self-assemble into polymorphic amyloid assemblies
Atomic force microscopy (AFM) imaging of the two peptide assemblies revealed that both peptides assembled at 37 °C almost immediately at 1 mg mL−1 (Supplementary Figs. 1 and 2) into a highly polymorphic mixture of nanofibrous and crystalline structures (Supplementary Fig. 3). For both peptides, the dominant polymorph was needle-like crystalline assemblies as seen in Fig. 2. In an attempt to ensure any observed polymorphism was not due to a heterogeneous mixture of insoluble peptide seeds we added warm PBS (90 °C) to the lyophilised peptides, and maintained this elevated temperature for at least 3 h in order dissolve as much of the monomeric peptide as possible. Self-assembly was subsequently initiated by slowly reducing the temperature, using a previously developed protocol32. This method produced less polymorphism resulting in the needle-like crystalline polymorph being overwhelmingly dominant, however a number of twisted fibrillar polymorphs were still present for RNYIAQVD (Fig. 2i and Supplementary Fig. 4a). To facilitate repeatable quantitative analysis of the biochemical and biophysical properties of the assemblies we used the slow cooling assembly method for all further experiments.
AFM and transmission electron microscopy (TEM) imaging of assemblies formed at either 1 or 5 mg mL−1 for 24 h revealed that assemblies from both peptides tend to stack on top of each other forming multi-laminar structures (Fig. 2a–d and Supplementary Figs. 5 and 6). Evidence of lateral assembly of the needles was also observed but this appears to happen more frequently in the ILLIIM assemblies (Fig. 2g) compared to RNYIAQVD. ILLIIM tends to form very large (2–3 μm in width) multi-laminar crystalline assemblies (Fig. 2g), whereas RNYIAQVD predominantly forms long linear needle-like structures. The apparent lower tendency of RNYIAQVD to form large two-dimensional lateral assemblies can be explained by the polymorphism seen in this peptide, which does not promote translational symmetry (i.e., extended crystals). Figure 2i and Supplementary Fig. 4a both clearly show that in addition to the flat needle-like crystals seen elsewhere, RNYIAQVD can also form non-planar partially twisted fibrillar assemblies. This polymorphism observed in RNYIAQVD assemblies may reduce the ability of the crystals to laterally associate and stack into multi-laminar species, simply because of a mismatch in planarity between two adjacent assemblies, the molecular basis for this polymorphism is briefly discussed in the next section of the manuscript. AFM was further used to investigate the height of the individual assemblies of both ILLIIM and RNYIAQVD. Figure 2a, b shows a line section through a multi-laminar RNYIAQVD assembly with two distinct layers with a step height of 5.5 nm between each layer. Similarly for ILLIIM (Fig. 2c and Supplementary Fig. 5), we see multi-laminar stacking with individual step heights that vary between 4 and 12 nm. Turning to RNYIAQVD we see single crystals with step heights varying between 5 and 20 nm (Supplementary Fig. 6). Together this heterogeneous distribution of crystal heights provides further evidence for the polymorphic nature of both the ILLIIM and RNYIAQVD assemblies.
Quantitative analysis of the distribution of assembly heights and contour lengths was performed using the freely available software tool FiberApp33. The analysis of assembly height distribution was taken from the z-axis (height) of the AFM images, both peptide assemblies show a heterogeneous distribution of fibril heights due to the previously observed tendency of both assemblies to form polymorphic multi-laminar stacks (Fig. 2f, h). Analysis of the distribution of the contour length of the two assemblies showed a biphasic distribution of lengths for both fibrils with two broad sub-populations centred around 1 and 3 µm (Fig. 2j, l). The sub-population at 3 µm was seen to be much larger for the RNYIAQVD peptide (Fig. 2j) compared to the ILLIIM (Fig. 2l). This population of longer fibrils correlates with the observation from Fig. 2 that for RNYIAQVD longer, thinner assemblies are favoured (self-assembly via fibril extension) over the wider shorter assemblies more commonly seen for ILLIIM (assembly via lateral association of protofilaments). Analysis of the persistence length of the fibrils (Supplementary Fig. 7) showed that whilst both peptides formed very straight linear assemblies, the persistence length of RNYIAQVD (λ = 41.92 µm) is greater than that of ILLIIM (λ = 31.96 µm).
To further investigate the polymorphic nature of the assembly of these peptides, we investigated the structures formed from 1:1 mixtures of the two peptides. Interestingly when mixed prior to assembly the peptides form a wide range of polymorphic structures exceeding that of either peptide assembled individually. Supplementary Fig. 8 shows a selection of some of the polymorphs formed, especially interesting are the large flat structures with well-defined edges that seem almost to interlock (Supplementary Fig. 8c). Such well-ordered 2D crystals were never observed for either peptide individually, and provide clear evidence of co-crystallisation. At this stage we have no evidence for the biological relevance of this co-crystallisation; however, as the ORF proteins from which these peptides are identified are themselves very small proteins (ORF6 is 61 amino acids in length), it is feasible that these small proteins may undergo similar co-crystallisation during their viral replication cycle facilitating an as yet unknown biological function.
X-ray scattering, spectroscopic characterisation, fluorescent microscopy and molecular modelling confirm the amyloid nature of the assemblies
Figure 3a shows the radially averaged 1D small-angle X-ray scattering (SAXS) plots for ILLIIM and RNYIAQVD at the lower concentration studied (at the higher concentration, sedimentation made recording X-ray scattering spectra impossible). In the central part of the scattering curve, the ILLIIM assemblies produced a slope with a q−2 dependence which is consistent with the form factor of an infinite 2D surface30, and is most likely arising due to the broader lateral dimensions observed by AFM and TEM for ILLIIM compared to RNYIAQVD. RNYIAQVD, however, displays a q−4 dependence in the central part of the scattering curve, appearing more towards the high-q limit. Porod’s law indicates that q−4 scaling (at high q but still less than 0.1 Å−1) is consistent with any aggregates having sharp surfaces but does not otherwise specify shape34. The data from the SAXS plots provide supporting evidence that the laterally associated amyloid assemblies seen by AFM and TEM are not artefacts induced either by the dehydration (AFM), applying vacuum conditions (TEM) or the mica (AFM) or carbon (TEM) substrates used, but a genuine structure observed also in bulk.
Figure 3b shows the CD spectra of mature assemblies, of both peptides. Assemblies of ILLIIM display a quite simple spectrum indicating the dominance of β-sheets, with a minimum between 225 and 230 nm and a strong maximum at 205 nm29. The CD spectra of RNYIAQVD possess a well-defined minimum at 203 nm and a distinct shoulder at around 215 nm.
To further investigate the predicted secondary structure of both peptide assemblies we employed the secondary structure analysis software BeStSel (Supplementary Table 1)35,36. As expected from the classic shape of the spectra the predicted secondary structure of ILLIIM at 5 mg mL−1 is exclusively made up of β-sheets (41.8%) and β-turns (58.2%). At these high concentrations, the composition of these β-sheets is shown to be exclusively left twisted, whilst at lower concentrations (1 mg mL−1) a more complex mixture of right, left and non-twisted (relaxed) β-sheets are predicted. At both concentrations, the CD spectra of RNYIAQVD again suggest the secondary structure is dominated by β-sheets; however, now they appear to be exclusively in the form of higher energy right-twisted β-sheets, similar to that observed in the highly strained structure of other amyloidogenic ultra-short peptides30. This additional strain introduced by β-sheets opposing the left-handed chirality seen in natural amino acids may explain the additional polymorphism and twisted microstructures seen in the AFM and TEM images of the RNYIAQVD assemblies (Fig. 2i and Supplementary Fig. 4a)29. The BeStSel fitting algorithm predicted the remainder of the RNYIAQVD secondary structure is composed of α-helical structure and further backbone conformations that could not be assigned (Supplementary Table 1). Part-helical CD spectra do not necessarily imply helical structure, especially considering that a single octapeptide cannot literally be 19% helix (two residues). Backbone conformation as reported by CD correlates through sheet structure to the assembled tertiary structure but no single level of organisation exclusively dictates any other, this is especially true in the case of coupling the twist of a peptide strand to the overall twist of the aggregate, which can relax to meet surface and shape-driven constraints through intersheet and interchain as well as intrachain degrees of freedom29.
To further investigate the conformation of the amyloid assemblies formed we utilised the conformation-specific antibody A11 and the fluorescent probe thioflavin T (ThT). The former binds specifically to non-fibrillar amyloid oligomers and the latter is a commonly used molecular probe that becomes highly fluorescent when binding to amyloid assemblies37. As expected, when ILLIIM and RNYIAQVD assemblies were stained with ThT both demonstrated clearly visible fluorescent emission at 590 nm, providing further evidence of their amyloid nature (Supplementary Fig. 9a, b). Conversely, A11 exhibited no positive binding; specifically, the level of fluorescence observed was similar to the background staining seen in the negative controls, as confirmed by similar fluorescent intensities for both assemblies (Supplementary Fig. 9e, f) and the negative controls (Supplementary Fig. 9d, h). Higher levels of A11 binding were seen for the positive control that consisted of phenylalanine assemblies known to form oligomeric species38 (Supplementary Fig. 9g), with fluorescent intensities for these assemblies over 4 times greater than for ILLIIM or RNYIAQVD. Combined these data confirm the amyloid nature of the two ORF peptide fragments and suggest that non-fibrillar oligomeric amyloid species are absent.
The amyloid nature of the two assemblies is further confirmed by the wide-angle X-ray scattering (WAXS) spectra (Fig. 3d) of the peptide assemblies which possessed a number of strongly diffracting Bragg peaks. Both peptides have a clear peak at 1.38 Å−1 corresponding to a d-spacing of 4.6 Å which is indicative of an amyloid assembly composed of extended β-sheets39. It is worth noting that the apparent intrastrand spacing of ILLIIM assemblies is very slightly lower than the typically reported values (4.7–4.8 Å). This may be explained by the BeStCell analysis of the ILLIIM assemblies at 1 mg mL−1, which suggests that the β-sheets in these assemblies are composed of a complex mixture of left-handed, strained right-handed and relaxed β-sheets (Supplementary Table 1); therefore, it is perhaps not surprising that the observed average intersheet spacing very slightly differs from that which is commonly reported. Furthermore, Lomont et al.40 report that the observed intrastrand spacing from a range of amyloid crystal structures can vary by as much as 0.45 Å. ILLIIM also has a very strong Bragg reflection at 0.58 Å−1 (11 Å) corresponding to a typical intersheet spacing given moderately bulky hydrophobic sidechains forming a steric zipper. RNYIAQVD has a number of well-defined Bragg peaks between 0.3 and 0.75 Å−1 that are consistent with a mixture of first and second-order reflections corresponding to an amyloid-like 3D symmetry. Typical reflections arising from the combinations of the longer two axes of the unit cell of short peptide amyloid crystals arise in the 0.3–0.75 Å−1 region with a qualitatively similar appearance to the pattern from RNYIAQVD, although in this case the peaks could not be individually assigned.
Discovery of sub-Å resolution structures from solution WAXS is highly challenging; however, given the simple nature of the scattering from the ILLIIM system, it was possible to produce an atomistic model matching the positions of the observed peaks, although not their sharpness. Physically, peak sharpness increases with the ordering length scale, indicating that some structures in the solution were larger than could be managed in the simulation. The sheet-like shape factor and the presence of peaks at roughly 2π/4.6 and 2π/11 Å−1 indicate assembly in solution dominated by the hydrogen bonding axis (with the typical parallel β sheet period of ≈ 4.7 Å) and by a sidechain-sidechain hydrophobic zipper interface. A metastable candidate structure of size 6 × 50 × 1 peptides was constructed following this geometry (see Methods) and found to reproduce the observed WAXS and to fully exclude water at the steric zipper (Fig. 4). The q−2 dependence of ILLIIM scattering at low q in solution (Fig. 3a) is consistent with a 2D sheet-like structure similar to that produced in the modelling. Initial assembly into sheets is also consistent with the eventual formation of multi-laminar structures as shown in the AFM (Fig. 2), as well as with the tendency of ILLIIM in particular to form lateral assemblies of needle microcrystals (Fig. 2g, k vs. e, i). Atomistic details of the interaction of the 2D sheet-like oligomer structure of Fig. 4 with neuronal cell membrane are difficult to predict and would be an interesting subject for further work. However, the juxtaposition of hydrophobic sidechains with polar termini in the ILLIIM fragment (or with titratable residues in the longer fragment E13ILLIIMR, which unfortunately could not be synthesised) has a length of approximately 10 Å, comparable to the polar-hydrophobic-polar length scale of 40 Å for the two leaflets of the eukaryotic cell membrane, indicating a potential for planar aggregates of, in particular, four sheets in thickness (four peptides end-to-end, linked in the middle by salt bridges) to disrupt the cell membrane.
The ThT stain, which becomes highly fluorescent upon binding to β-sheet rich amyloid assemblies was used to assess the assembly kinetics of both ILLIIM and RNYIAQVD (Fig. 3c). Both peptides show rapid kinetics with significant assembly occurring almost instantaneously and reaching a plateau after 30–60 min. Longer amyloidogenic polypeptides typically show a distinct lag phase in their assembly kinetics; however, this was not observed in these sequences. This apparent lack of a lag phase in the assembly kinetics behaviour is typical of amyloidogenic short peptides, which have been previously seen to assemble very rapidly29,41. This is highly likely to be due to a lack of additional non-amyloidogenic amino acid sequences acting as a kinetic barrier to amyloid formation. The ThT signal for ILLIIM at 5 mg mL−1 plateaus at about 300 a.u., this is slightly stronger than the maximum signal generated from mature fibrils of the somewhat homologous peptide ILQINS, which was around 250 a.u29,30,31,42, suggesting that the amyloidogenicity of the two peptides is comparable. RNYIAQVD, whilst showing similar ThT values at low concentrations (1 mg mL−1), generated a ThT signal nearly 3 times as large at 5 mg mL−1 suggesting that the assembly of this peptide is highly concentration dependent. For reasons yet unknown, it seems that RNYIAQVD appears to reach a maximum ThT value and then begin to drop, this can be seen at both concentrations but is most obvious at the higher concentration. This could be due to a reversible self-assembly as seen in other functional amyloids39,41,43,44, or to self-quenching of the amyloid bound aromatic ThT molecules, or simply to a reduction of exposed ThT-binding sites as larger aggregates with a lower surface area to volume ratio come to dominate the solution.
Cytotoxicity of both studied peptides also began to drop slightly (without statistical significance) at the highest concentrations tested (vide infra, Fig. 5), together with the drop in ThT response this supports the existence of a kinetically or thermodynamically available aggregate structure with reduced ‘amyloid activity’. This is reminiscent of strongly amyloid correlated diseases such as AD, where the toxicity of amyloid can vary dramatically, with the relationship between the amount of amyloid deposited to the progress of the disease being idiosyncratic and highly non-linear45. Combined the CD spectroscopy (Fig. 3b), the ThT spectroscopy (Fig. 3c) and confocal microscopy (Supplementary Fig. 9), the presence of the Bragg peaks corresponding to the intra- and inter-β-sheet spacings (Fig. 3d) and the molecular modelling (Fig. 4) confirm beyond doubt the β-sheet rich, amyloid nature of these two fragments.
ILLIIM and RNYIAQVD peptide assemblies are both highly toxic towards the neuroblastoma cell line SH-SY5Y
Given the physical evidence and the discussions referred to in the introduction of various means by which SARS-CoV-2 and other viral infections could enhance their fitness (to the detriment of the host) by the production of amyloidogenic peptides, we hypothesised that the SARS-CoV-2 viral transcript fragments ILLIIM and RNYIAQVD are toxic to human neurons. This is in particular supported by the previously reported neuroinvasive capabilities of SARS-CoV-27,8, the noted similarities of the symptoms to a (hopefully transient form of) AD5 and the previous detection of amyloid assemblies driven by other viruses20. To investigate this further we performed a number of cytotoxicity assays of the two peptide sequences against a human-derived neuroblastoma cell line (SH-SY5Y) often used as a model cell line for studying Parkinson’s and other neurodegenerative diseases46. Using an MTT assay we found that cells grown in the presence of both peptide assemblies possessed much lower viability after 48 h incubation. Concentrations as low as 0.04 and 0.03 mM (for RNYIAQVD and ILLIIM, respectively) were seen to reduce the viability of cultured cells after 48 h to <50% (IC50) compared to the cells cultured without the peptides (Fig. 5a, b). This toxicity in relation to concentration is similar to that reported for Aβ4247 although expression levels and time-scales (sudden for COVID versus chronic for AD) are likely to be very different.
To gain further insight into the mechanism of cell death occurring in the peptide exposed cells, we performed a detailed flow cytometry analysis using the apoptotic stain Annexin V and the viability dye 7-AAD. Figure 5c shows representative flow cytometry plots; cells can be identified as viable (bottom left quadrant), viable but undergoing early apoptosis (bottom right), non-viable and necrotic (top left) or non-viable due to late-stage apoptosis (top right). The percentages of cells in these quadrants are roughly equal for all conditions tested except in the case of late-stage apoptosis where we see a large increase in the cells exposed to the peptide assemblies (a 6.25-fold increase for RNYIAQVD at 2.5 mg mL−1). Quantification over a range of concentrations showed that on average cells exposed to both ILLIIM and RNYIAQVD had a 3–5-fold increase in late-stage apoptosis compared to SH-SY5Y cells cultured in the absence of peptide assemblies (Fig. 5d, e). No evidence of increasing necrosis was seen in any of the samples, suggesting that the amyloid assemblies are triggering programmed cell death via an apoptotic pathway. This triggering of late-stage apoptosis in the cells was more pronounced for ILLIIM than for RNYIAQVD, showing statistically significant increases in apoptotic cells at concentrations as low as 0.04 mg mL−1 for ILLIIM compared to 0.15 mg mL−1 for RNYIAQVD. This increase in apoptosis down to low concentration provides convincing evidence, especially for ILLIIM, that the amyloid aggregates are responsible for this toxicity, as at these low concentrations we would expect very little un-assembled peptide to exist. The mechanisms of cell death in neurodegenerative diseases are complex and can vary between different diseases48, and here we provide evidence that induction of apoptosis may be an important mechanism of neuronal death in COVID-19. Intriguingly, the conserved protein ORF6 from SARS-CoV-1 (not SARS-CoV-2) has previously been shown to induce apoptosis49. Furthermore, we performed a series of cell counting experiments and demonstrated that after 48 h incubation we saw statistically significant decreases in cell number for both peptides at concentrations as low as 0.04 mg mL−1 for ILLIIM and 0.32 mg mL−1 for RNYIAQVD. These results confirm that in addition to the cytotoxic nature of the peptide assemblies, they significantly reduce cell number especially in the case of ILLIIM. The significant increase in apoptosis and reduction in cell number seen for ILLIIM correlates with the work of Lee et al. who have previously shown that the ORF6 protein (that contains the ILLIIM sequence) is the most cytotoxic protein in the proteome of SARS-CoV-213. Combined with our data, this suggests that this toxicity might be due to the amyloidogenic nature of this short protein.
Previous research has shown that the polymorphism, size distribution and the morphology of amyloid aggregates can have a large influence on their cytotoxicity. Marshall et al.50 showed that a range of crystal-forming assemblies formed from short peptide sequences show surprisingly little toxicity to the same neuroblastoma cell line used in this study. Our TEM and AFM images (Fig. 2) confirm that the assemblies formed by the sequences identified from ORF6 and ORF10 look very similar to the assemblies in Marshall et al.50 but the SARS-CoV-2-related peptides are significantly more toxic, suggesting a specific mechanism of toxicity for these assemblies. Xue et al.51 showed that shorter amyloid assemblies from a range of different proteins/peptides have increased the ability to disrupt the bilayer of unilamellar vesicles and provide a greater cytotoxic effect on neuroblastoma cells. Mocanu et al.52 showed a dose-dependent cytotoxic effect in epithelial cells for long-thin lysozyme amyloid fibrils, and a threshold dependent mechanism for the larger laterally associated fibrils. We see similar effects to both Xue et al. and Mocanu et al. suggesting that the observed toxicity of the assemblies may be related to their aspect ratio. We observed that ILLIIM assemblies are both more toxic, wider (Fig. 2h) and shorter than their RNYIAQVD counterparts (Fig. 2k), this is shown schematically in Fig. 6. Similarly to Mocanu et al.52 we see that the long-thin RNYIAQVD fibrils show a clear dose-dependent increase in apoptosis (Fig. 5e), and the laterally associated ILLIIM fibrils show similarly high levels of apoptosis induction at all concentrations above a threshold of 0.04 mg mL−1 (Fig. 5d).
There is a wealth of literature suggesting that in neurodegenerative diseases like Alzheimer’s and Parkinson’s amyloid oligomers are the main toxic culprits and mature amyloid fibrils are a more inert assembly end-point. This is seemingly at odds with our data; however, there have also been multiple studies that show mature assemblies can also display significant toxicity37,53,54. Alternatively, it may be the nature of the amyloids species seen here that differs from amyloids in neurodegenerative diseases; the amyloids seen here appear to be largely crystalline (especially in the case of ILLIIM). Previous work has shown that amyloid crystals are deeper in the free energy landscape compared to twisted protofilaments and amyloid ribbons30,55, representing a global energy minima. AFM and TEM data have shown that these stable amyloid crystals are the dominant polymorph for ILLIIM (Fig. 2c, g, k) and that RNYIAQVD shows examples of higher energy (partially) twisted fibrils (Fig. 2i and Supplementary Figs. 4a and 6). Therefore, we hypothesise that the low energy ILLIIM crystalline assemblies are more slowly metabolised and cells are exposed for longer timeframes to the cytotoxic effect compared to RNYIAQVD assemblies. To date, there have been few investigations into the toxicity of amyloid crystals compared to other more commonly reported amyloid species. For the reasons above, the toxic nature of these amyloid assemblies warrants further investigations into the potential presence of amyloid aggregates from SARS-CoV-2 in the CNS of COVID-19 patients, and the potential role of amyloids in the neurological symptoms observed.
In conclusion, using a bioinformatics approach we identified two strongly amyloidogenic sub-sequences from the ORF6 and ORF10 sections of the SARS-COV-2 proteome. Nanoscale imaging, X-ray scattering, molecular modelling, spectroscopy and kinetic assays revealed that these self-assembled structures are amyloid in nature, and screening against neuronal cells revealed that they are highly toxic (approximately as toxic as the toxic amyloid assemblies in AD) to a cell line frequently used as a neurodegenerative diseases model. The neuroinvasive nature of SARS-COV-2 has been established previously7,8; therefore, it is entirely plausible that amyloid assemblies either from these ORF proteins or other viral proteins could be present in the CNS of COVID-19 patients. The cytotoxicity and protease-resistant structure of these assemblies may result in their persistent presence in the CNS of patients post-infection that could partially explain the lasting neurological symptoms of COVID-19, especially those that are novel in relation to other post-viral syndromes such as that following the original SARS-CoV-1. The outlook in relation to triggering of progressive neurodegenerative disease remains uncertain. Given the typically slow progress of neurodegenerative disease if such a phenomenon exists, it will most probably take some time to become evident epidemiologically.
Amyloid prediction algorithms
The online amyloid prediction algorithms TANGO and ZIPPER were used to predict peptide sequences with a tendency to form β-rich amyloid assemblies. TANGO is an algorithm that predicts aggregation nucleating regions in unfolded polypeptide chains56. It works on the assumption that the aggregating regions are buried in the hydrophobic core of the natively folded protein. ZIPPER is an algorithm that predicts hexapeptides within larger polypeptide sequences that have a strong energetic drive to form the two complementary β-sheets (known as a steric zipper) that give rise to the spine of an amyloid fibril57. Both methods are physically motivated but rely on statistically determined potentials.
Self-assembly of peptides
NH2-ILLIIM-CO2H and Ac-RNYIAQVD-NH2 (>95% pure) were purchased from GL Biochem Ltd (Shanghai, China). Ideally, it would have been preferred to have both peptides capped (N-terminus: Acetyl and C-terminus: Amide), as they would better represent small fragments of a larger peptide sequence. Due to the fact ILLIIM contains no charged sidechains, synthesising capped sequences to high purity would have been very challenging, therefore only the RNYIAQVD sequence remained capped and the ILLIIM sequence had regular carboxyl and amino termini. To ensure that all peptide seeds were fully dissolved before self-assembly was initiated the peptides were solubilised in warmed PBS (90 °C) at either 1 or 5 mg mL−1 The warmed peptide solutions were vortexed vigorously and held at 90 °C for 3 h to ensure maximum dissolution. After the second round of vortexing, the peptide suspensions were cooled slowly. This protocol has been previously used to maximise a homogenous starting population of monomeric peptide32. Alternatively, self-assembly was carried out at a constant temperature of 37 °C without pre-solubilising the peptides in hot PBS.
Atomic force microscopy (AFM)
AFM imaging was performed on a Bruker Multimode 8 AFM and a Nanoscope V controller. Tapping mode imaging was used throughout, with antimony (n)-doped silicon cantilevers having approximate resonant frequencies of 525 or 150 kHz and spring constants of either 200 or 5 Nm−1 (RTESPA-525, Bruker or RTESPA-150, Bruker). No significant differences were observed between cantilevers. 50 µL aliquots of the peptide (either at 1 or 5 mg mL−1) were drop cast onto freshly cleaved muscovite mica disks (10 mm diameters) and incubated for 20 min before gently rinsing in MQ water and drying under a nitrogen stream. All images were flattened using the first order flattening algorithm in the nanoscope analysis software and no other image processing occurred. Statistical analysis of the AFM images was performed using the open-source software FiberApp33 from datasets of no less than 900 fibres.
Transmission electron microscopy (TEM)
Copper TEM grids with a formvar-carbon support film (GSCU300CC-50, ProSciTech, Qld, Australia) were glow discharged for 60 s in an Emitech k950x with k350 attachment. Then, 5 µL drops of sample suspension were pipetted onto each grid, allowed to adsorb for at least 30 s and blotted with filter paper. Two drops of 2% uranyl acetate were used to negatively stain the particles with excess negative stain removed by blotting with filter paper after 10 s each. Grids were then allowed to dry before imaging. Grids were imaged using a Joel JEM-2100 (JEOL (Australasia) Pty Ltd) transmission electron microscope equipped with a Gatan Orius SC 200 CCD camera (Scitek Australia).
Small- and wide-angle X-ray scattering (SAXS/WAXS)
SAXS/WAXS experiments were performed at room temperature on the SAXS/WAXS beamline at the Australian Synchrotron. Peptide assemblies in PBS prepared at both 1 and 5 mg mL−1 were loaded into a 96-well plate held on a robotically controlled x–y stage and transferred to the beamline via a quartz capillary connected to a syringe pump. Data from the 5 mg mL−1 assemblies were discarded due to sedimentation of the assemblies preventing reliable sample transfer into the capillaries. The experiments used a beam wavelength of λ = 1.03320 Å−1 (12.0 keV) with dimensions of 300 µm × 200 µm and a typical flux of 1.2 × 1013 photons per second. 2D diffraction images were collected on a Pilatus 1M detector. SAXS experiments were performed at q ranges between 0.002 and 0.25 Å−1 and WAXS experiments were performed at a q range between 0.1 and 2 Å−1. These overlapping spectra provide a total q range of 0.002–2.2 Å−1. Spectra were recorded under flow (0.15 mL min−1) to prevent X-ray damage from the beam. Multiples of approximately 15 spectra were recorded for each time point (exposure time = 1 s) and averaged spectra are shown after background subtraction against PBS in the same capillary.
Circular dichroism spectroscopy
CD spectroscopy was performed using an AVIV 410-SF CD spectrometer. Spectra were collected between 190 and 260 nm in PBS using 1 mm quartz cuvettes with a step size of 0.5 nm and 2 s averaging time. Data were analysed using the BeStSel (Beta Structure Selection) method of secondary structure determination35.
Atomistic models were constructed using the Nucleic Acid Builder58. Simulations were run in explicit water (TIP3P59) using the ff15ipq forcefield60 and the pmemd time integrator61. In order to hold the unit cell geometry to values consistent with the observed scattering, the alpha carbon of the central residue of each chain was subjected to a restraining force with spring constant 2 kcal mol−1 Å−2. Periodic boundaries were applied to the system such that it formed a truncated octahedron, which was relaxed during equilibration to a volume of 10,310 nm3, giving a density of 0.973 reference with 323,535 water molecules. The system state after 10 ns of equilibration was stripped of water molecules more than 10 Å from any non-hydrogen solute atom, and passed to CRYSOL3 for calculation of orientationally averaged scattering profile given the example state (including the ordered waters from the explicit solvent shell, and also including an approximate treatment of ordered water beyond this shell)42,62.
Thioflavin T amyloid kinetic assays
Peptide assemblies were made up to concentrations of 1 or 5 mg mL−1 suspensions containing 25 µM ThT in PBS. The first fluorescence measurement (t = 0) was recorded immediately after sample preparation. All the samples were then stored at room temperature and fluorescence intensity was recorded at different time points. Measurements were performed in triplicate using a ClarioStar fluorimeter equipped with a 96-well plate reader (excitation wavelength: 440 nm, emission wavelength: 482 nm).
Cell line and cultures
Human-derived neuroblastoma cells (SH-SY5Y, ATCC Product Number: CRL-2266) were cultured in DMEM-F12 (Invitrogen) medium supplemented with 10% (v/v) foetal calf serum (FCS), 100 UmL−1 penicillin and 100 µgmL−1 streptomycin (Invitrogen, Carlsbad, CA). Cells were cultured at 37 °C in a humidified atmosphere containing 5% CO2.
Immunofluorescent and thioflavin T microscopy
ThT staining was performed by incubating the amyloid assemblies with a 25 µM solution of ThT in PBS for 15 min, in an 8-well Labtek II chamber (Nunc). For the antibody stain the same assemblies were incubated in a 1:200 dilution of the A11 polyclonal antibody raised in rabbit (Invitrogen, REF: AHB0052, LOT:VF299837) in 2% BSA in PBS for 1 h. Following this the wells were carefully washed in PBS and a solution of the 2° antibody (Goat-Anti Rabbit IgG-Alexa Fluor 647, Product A21244, Lot #: 1871168, Molecular Probes) at 1:1000 dilution in PBS was incubated with the peptides for 1 h. Finally the assemblies were once again washed in PBS before being imaged via laser scanning confocal microscopy using a FV3000 microscope (Olympus) and 60× objective lens (1.35 NA Oil Plan Apochromat) using the following settings: ThT channel λex = 450 nm, λem = 490 nm, Alexa Fluora 647 channel λex = 650 nm, λem = 665 nm. The same imaging settings were used for all samples and the negative controls (peptide assemblies + 2° antibody only) were used to determine the level of background fluorescence. For positive controls, amyloid assemblies of phenylalanine were used under concentrations known to readily form oligomeric amyloid assemblies38, which were shown here to bind strongly to the A11 antibody (Supplementary Fig. 9e, f).
Cell viability assay
Cells were seeded into 96-well plates at 1 × 105 cells per mL and incubated for 24 h to ensure good attachment to the surface. A stock solution of peptide assemblies (10 mg mL−1) was serially diluted into DMEM-F12 media 2.5–0.02 mg mL−1 or 3.3–0.027 mM for ILLIIM and 2.45–0.02 mM for RNYIAQVD) and seeded onto the SH-SY5Y cells and incubated for 48 h, cell viability was determined using 3-(4,5-dimethylthiazol-2-yl)2,5-diphenyltetrazolium bromide (MTT) (Sigma-Aldrich) as described previously63. Equivalent MTT assays were performed on cells cultured in the same ratios of PBS to media, but in the absence of peptide, assemblies to confirm that the culture conditions were non-toxic (Supplementary Fig. 10). Absorbance readings of untreated control wells in 100% cell culture media were designated as 100% cell viability. Statistical analysis was performed by one-way ANOVA tests with Tukey comparison in the software GraphPad (Prism) ***p < 0.001. Flow cytometry assays to determine cell viability were performed in a similar manner to the MTT assays. Briefly, to determine the effect of the peptides on cellular viability SH-SY5Y cells were cultured in the presence of the peptides for 48 h, harvested and stained with the apoptosis stain Annexin V for 10 min on ice (Cat No. 550474, BD Biosciences, 5 µL in 100 µL of 2% FCS in PBS). Samples were diluted with 300 µL of 2% FCS in PBS and stained with the viability dye 7-AAD (559925 BD Biosciences, 5 µL per sample) and analysed using flow cytometry (FACS Aria III; BD Biosciences). Cell counts were performed manually using a hemocytometer, with tryphan blue to differentiate non-viable cells.
All code used in this study is either free (NAB 1.3, pymol 2, TANGO 2.2 and ZIPPER) or commercially available (pmemd 19, CRYSOL3 3.0.3).
Frontera, J. A. et al. A prospective study of long-term outcomes among hospitalized COVID-19 patients with and without neurological complications. J. Neurol. Sci.426, 117486 (2021).CASPubMedPubMed CentralArticleGoogle Scholar
de Melo, G. D. et al. COVID-19-related anosmia is associated with viral persistence and inflammation in human olfactory epithelium and brain infection in hamsters. Sci. Transl. Med.13, eabf8396 (2021) .
Saumya, K. U., Gadhave, K., Kumar, A. & Giri, R. Zika virus capsid anchor forms cytotoxic amyloid-like fibrils. Virology560, 8–16 (2021).CASPubMedArticleGoogle Scholar
Ghosh, A. et al. Self-assembly of a nine-residue amyloid-forming peptide fragment of SARS corona virus E-protein: mechanism of self aggregation and amyloid-inhibition of hIAPP. Biochemistry54, 2249–2261 (2015).CASPubMedArticleGoogle Scholar
Tayeb-Fligelman, E. et al. Inhibition of amyloid formation of the Nucleoprotein of SARS-CoV-2. Preprint at bioRxiv 2021.
Chen, H. et al. Liquid–liquid phase separation by SARS-CoV-2 nucleocapsid protein and RNA. Cell Res.30, 1143–1145 (2020).CASPubMedArticleGoogle Scholar
Savastano, A., Ibáñez de Opakua, A., Rankovic, M. & Zweckstetter, M. Nucleocapsid protein of SARS-CoV-2 phase separates into RNA-rich polymerase-containing condensates. Nat. Commun.11, 6041 (2020).ADSCASPubMedPubMed CentralArticleGoogle Scholar
Pham, C. L. et al. Viral M45 and necroptosis-associated proteins form heteromeric amyloid assemblies. EMBO Rep.20, e46518 (2019).PubMedArticleCASGoogle Scholar
Xu, J. et al. Systematic comparison of two animal-to-human transmitted human coronaviruses: SARS-CoV-2 and SARS-CoV. Viruses 12, 244 (2020).
Ngai, J. C. et al. The long-term impact of severe acute respiratory syndrome on pulmonary function, exercise capacity and health status. Respirology15, 543–550 (2010).PubMedPubMed CentralArticleGoogle Scholar
Lara, C. et al. ILQINS hexapeptide, identified in lysozyme left-handed helical ribbons and nanotubes, forms right-handed helical ribbons and crystals. J. Am. Chem. Soc.136, 4732–4739 (2014).CASPubMedArticleGoogle Scholar
Zaguri, D. et al. Nanomechanical properties and phase behavior of phenylalanine amyloid ribbon assemblies and amorphous self-healing hydrogels. ACS Appl. Mater. Interfaces12, 21992–22001 (2020).CASPubMedArticleGoogle Scholar
Usov, I. & Mezzenga, R. FiberApp: an open-source software for tracking and analyzing polymers, filaments, biomacromolecules, and fibrous objects. Macromolecules48, 1269–1280 (2015).ADSCASArticleGoogle Scholar
Sinha, S. K., Sirota, E. B., Garoff, S. & Stanley, H. B. X-ray and neutron scattering from rough surfaces. Phys. Rev. B38, 2297–2311 (1988).ADSCASArticleGoogle Scholar
Micsonai, A. et al. BeStSel: a web server for accurate protein secondary structure prediction and fold recognition from the circular dichroism spectra. Nucleic Acids Res.46, W315–W322 (2018).CASPubMedPubMed CentralArticleGoogle Scholar
Adler-Abramovich, L. et al. Phenylalanine assembly into toxic fibrils suggests amyloid etiology in phenylketonuria. Nat. Chem. Biol.8, 701–706 (2012).CASPubMedArticleGoogle Scholar
Do, T. D., Kincannon, W. M. & Bowers, M. T. Phenylalanine oligomers and fibrils: the mechanism of assembly and the importance of tetramers and counterions. J. Am. Chem. Soc.137, 10080–10083 (2015).CASPubMedArticleGoogle Scholar
Dharmadana, D., Reynolds, N. P., Conn, C. E. & Valéry, C. Molecular interactions of amyloid nanofibrils with biological aggregation modifiers: implications for cytotoxicity mechanisms and biomaterial design. Interface Focus7, 20160160 (2017).PubMedPubMed CentralArticleGoogle Scholar
Lomont, J. P. et al. Not all β-sheets are the same: amyloid infrared spectra, transition dipole strengths, and couplings investigated by 2D IR spectroscopy. J. Phys. Chem. B121, 8935–8945 (2017).CASPubMedPubMed CentralArticleGoogle Scholar
Dharmadana, D., Reynolds, N. P., Conn, C. E. & Valéry, C. pH-dependent self-assembly of human neuropeptide hormone GnRH into functional amyloid nanofibrils and hexagonal phases. ACS Appl. Bio Mater.2, 3601–3606 (2019).CASPubMedArticleGoogle Scholar
Zanjani, A. A. H. et al. Amyloid evolution: antiparallel replaced by parallel. Biophysical J.118, 2526–2536 (2020).ADSCASArticleGoogle Scholar
Dharmadana, D. et al. Heparin assisted assembly of somatostatin amyloid nanofibrils results in disordered precipitates by hindrance of protofilaments interactions. Nanoscale10, 18195–18204 (2018).CASPubMedArticleGoogle Scholar
Dharmadana, D. et al. Human neuropeptide substance P self-assembles into semi-flexible nanotubes that can be manipulated for nanotechnology. Nanoscale12, 22680–22687 (2020).CASPubMedArticleGoogle Scholar
Morris, G. P., Clark, I. A. & Vissel, B. Inconsistencies and controversies surrounding the amyloid hypothesis of Alzheimer’s disease. Acta Neuropathologica Commun.2, 135 (2014).Google Scholar
Xicoy, H., Wieringa, B. & Martens, G. J. M. The SH-SY5Y cell line in Parkinson’s disease research: a systematic review. Mol. Neurodegeneration12, 10 (2017).ArticleCASGoogle Scholar
Krishtal, J. et al. Toxicity of amyloid-β peptides varies depending on differentiation route of SH-SY5Y cells. J. Alzheimers Dis.71, 879–887 (2019).CASPubMedArticleGoogle Scholar
Kopecky-Bromberg, S. A. et al. Severe acute respiratory syndrome coronavirus open reading frame (ORF) 3b, ORF 6, and nucleocapsid proteins function as interferon antagonists. J. Virol.81, 548–557 (2007).CASPubMedArticleGoogle Scholar
Mocanu, M.-M. et al. Polymorphism of hen egg white lysozyme amyloid fibrils influences the cytotoxicity in LLC-PK1 epithelial kidney cells. Int. J. Biol. Macromolecules65, 176–187 (2014).CASArticleGoogle Scholar
Novitskaya, V., Bocharova, O. V., Bronstein, I. & Baskakov, I. V. Amyloid fibrils of mammalian prion protein are highly toxic to cultured cells and primary neurons. J. Biol. Chem.281, 13828–13836 (2006).CASPubMedArticleGoogle Scholar
Adamcik, J. & Mezzenga, R. Amyloid polymorphism in the protein folding and aggregation energy landscape. Angew. Chem. Int. Ed.57, 8370–8382 (2018).CASArticleGoogle Scholar
Fernandez-Escamilla, A.-M., Rousseau, F., Schymkowitz, J. & Serrano, L. Prediction of sequence-dependent and mutational effects on the aggregation of peptides and proteins. Nat. Biotechnol.22, 1302–1306 (2004).CASPubMedArticleGoogle Scholar
Macke, T. J. & Case, D. A. Modeling unusual nucleic acid structures. In Molecular Modeling of Nucleic Acids (eds Leontis, N. B. & SantaLucia, J.) Vol. 682, 379–393 (American Chemical Society, 1997).
Jorgensen, W. L. et al. Comparison of simple potential functions for simulating liquid water. J. Chem. Phys.79, 926–935 (1983).ADSCASArticleGoogle Scholar
Debiec, K. T. et al. Further along the road less traveled: AMBER ff15ipq, an original protein force field built on a self-consistent physical model. J. Chem. Theory Comput.12, 3926–3947 (2016).CASPubMedPubMed CentralArticleGoogle Scholar
Salomon-Ferrer, R. et al. Routine microsecond molecular dynamics simulations with AMBER on GPUs. 2. Explicit solvent particle mesh Ewald. J. Chem. Theory Comput.9, 3878–3888 (2013).CASPubMedArticleGoogle Scholar
Varrette, S., Bouvry, P., Cartiaux, H. & Georgatos, F. Management of an academic HPC cluster: the UL experience. In 2014 International Conference on High Performance Computing & Simulation (HPCS), 21–25 July 2014, 959–967 (2014).
N.P.R. would like to acknowledge The La Trobe Institute of Molecular Sciences (LIMS) for the receipt of a Nicholas Hoogenraad fellowship, and the CASS foundation for partially funding this work through a philanthropic grant (#10053, ‘Determining the role of protein aggregation in COVID-19’). N.P.R. would also like to acknowledge that Fig. 6 was created using Biorender.com. The authors thank Dr Susi Seibt for assistance on the SAXS/WAXS beamline at the Australian Synchrotron. This research was undertaken, in part, on the SAXS/WAXS beamline at the Australian Synchrotron, part of ANSTO. Molecular dynamics calculations made use of the HPC service of the University of Luxembourg64. The project was part-funded by grant C20/MS/14588607 of the Fonds Nationale de la Recherche, Luxembourg.
The mechanisms that lead to cognitive impairment associated with COVID-19 are not well understood.
Brain lysates from control and COVID-19 patients were analyzed for oxidative stress and inflammatory signaling pathway markers, and measurements of Alzheimer’s disease (AD)-linked signaling biochemistry. Post-translational modifications of the ryanodine receptor/calcium (Ca2+) release channels (RyR) on the endoplasmic reticuli (ER), known to be linked to AD, were also measured by co-immunoprecipitation/immunoblotting of the brain lysates.
We provide evidence linking SARS-CoV-2 infection to activation of TGF-β signaling and oxidative overload. The neuropathological pathways causing tau hyperphosphorylation typically associated with AD were also shown to be activated in COVID-19 patients. RyR2 in COVID-19 brains demonstrated a “leaky” phenotype, which can promote cognitive and behavioral defects.
COVID-19 neuropathology includes AD-like features and leaky RyR2 channels could be a therapeutic target for amelioration of some cognitive defects associated with SARS-CoV-2 infection and long COVID.
1.1 Contextual background
Patients suffering from COVID-19 exhibit multi-system organ failure involving not only pulmonary1 but also cardiovascular,2 neural,3 and other systems. The pleiotropy and complexity of the organ system failures both complicate the care of COVID-19 patients and contribute, to a great extent, to the morbidity and mortality of the pandemic.4 Severe COVID-19 most commonly manifests as viral pneumonia-induced acute respiratory distress syndrome (ARDS).5 Respiratory failure results from severe inflammation in the lungs, which arises when SARS-CoV-2 infects lung cells. Cardiac manifestations are multifactorial and include hypoxia, hypotension, enhanced inflammatory status, angiotensin-converting enzyme 2 (ACE2) receptor downregulation, endogenous catecholamine adrenergic activation, and direct viral-induced myocardial damage.6, 7 Moreover, patients with underlying cardiovascular disease or comorbidities, including congestive heart failure, hypertension, diabetes, and pulmonary diseases, are more susceptible to infection by SARS-CoV-2, with higher mortality.6, 7
In addition to respiratory and cardiac manifestations, it has been reported that approximately one-third of patients with COVID-19 develop neurological symptoms, including headache, disturbed consciousness, and paresthesias.8 Brain tissue edema, stroke, neuronal degeneration, and neuronal encephalitis have also been reported.2, 8–10 In a recent study, diffuse neural inflammatory markers were found in >80% of COVID-19 patient brains, processes which could contribute to the observed neurological symptoms.11 Furthermore, another pair of frequent symptoms of infection by SARS-CoV-2 are hyposmia and hypogeusia, the loss of the ability to smell and taste, respectively.3 Interestingly, hyposmia has been reported in early-stage Alzheimer’s disease (AD),3 and AD type II astrocytosis has been observed in neuropathology studies of COVID-19 patients.10
Systemic failure in COVID-19 patients is likely due to SARS-CoV-2 invasion via the ACE2 receptor,9 which is highly expressed in pericytes of human heart8 and epithelial cells of the respiratory tract,12 kidney, intestine, and blood vessels. ACE2 is also expressed in the brain, especially in the respiratory center and hypothalamus in the brain stem, the thermal center, and cortex,13 which renders these tissues more vulnerable to viral invasion, although it remains uncertain whether SARS-CoV-2 virus directly infects neurons in the brain.14 The primary consequences of SARS-CoV-2 infection are inflammatory responses and oxidative stress in multiple organs and tissues.15–17 Recently it has been shown that the high neutrophil-to-lymphocyte ratio observed in critically ill patients with COVID-19 is associated with excessive levels of reactive oxygen species (ROS) and ROS-induced tissue damage, contributing to COVID-19 disease severity.15
Recent studies have reported an inverse relationship between ACE2 and transforming growth factor-β (TGF-β). In cancer models, decreased levels of ACE2 correlated with increased levels of TGF-β.18 In the context of SARS-CoV-2 infection, downregulation of ACE2 has been observed, leading to increased fibrosis formation, as well as upregulation of TGF-β and other inflammatory pathways.19 Moreover, patients with severe COVID-19 symptoms had higher blood serum TGF-β concentrations than those with mild symptoms,20 thus further implicating the role of TGF-β and warranting further investigation.
Interestingly, reduced angiotensin/ACE2 activity has been associated with tau hyperphosphorylation and increased amyloid beta (Aβ) pathology in animal models of AD.21, 22 The link between reduced ACE2 activity and increased TGF-β and tau signaling in the context of SARS-CoV-2 infection needs further exploration.
Our laboratory has shown that stress-induced ryanodine receptor (RyR)/intracellular calcium release channel post-translational modifications, including oxidation and protein kinase A (PKA) hyperphosphorylation related to activation of the sympathetic nervous system and the resulting hyper-adrenergic state, deplete the channel stabilizing protein (calstabin) from the channel complex, destabilizing the closed state of the channel and causing RyR channels to leak Ca2+ out of the endoplasmic/sarcoplasmic reticulum (ER/SR) in multiple diseases.23–29 Increased TGF-β activity can lead to RyR modification and leaky channels,30 and SR Ca2+ leak can cause mitochondrial Ca2+ overload and dysfunction.29 Increased TGF-β activity31 and mitochondrial dysfunction32 are also associated with SARS-CoV-2 infection.
Here we show that SARS-CoV-2 infection is associated with adrenergic and oxidative stress and activation of the TGF-β signaling pathway in the brains of patients who have succumbed to COVID-19. One consequence of this hyper-adrenergic and oxidative state is the development of tau pathology normally associated with AD. In this article, we investigate potential biochemical pathways linked to tau hyperphosphorylation. Based on recent evidence that has linked tau pathology to Ca2+ dysregulation associated with leaky RyR channels in the brain,3, 33 we investigated RyR2 biochemistry and function in COVID-19 patient brains.
Systematic review: The authors reviewed the literature using PubMed. While the mechanisms that lead to cognitive impairment associated with COVID-19 are not well understood, there have been recent reports studying SARS-CoV-2 infection and brain biochemistry and neuropathology. These relevant citations are appropriately cited.
Interpretation: Our findings link the inflammatory response to SARS-CoV-2 infection with the neuropathological pathways causing tau hyperphosphorylation typically associated with Alzheimer’s disease (AD). Furthermore, our data indicate a role for leaky ryanodine receptor 2 (RyR2) in the pathophysiology of SARS-CoV-2 infection.
Future directions: The article proposes that the alteration of cellular calcium dynamics due to leaky RyR2 in COVID-19 brains is associated with the activation of neuropathological pathways that are also found in the brains of AD patients. Both the cortex and cerebellum of SARS-CoV-2–infected patients exhibited a reduced expression of the Ca2+ buffering protein calbindin. Decreased calbindin could render these tissues more vulnerable to cytosolic Ca2+ overload. Ex vivo treatment of the COVID-19 brain using a Rycal drug (ARM210) that targets RyR2 channels prevented intracellular Ca2+ leak in patient samples. Future experiments will explore calcium channels as a potential therapeutic target for the neurological complications associated with COVID-19.
1.2 Study conclusions and disease implications
Our results indicate that SARS-CoV-2 infection activates inflammatory signaling and oxidative stress pathways resulting in hyperphosphorylation of tau, but normal amyloid precursor protein (APP) processing in COVID-19 patient cortex and cerebellum. There was reduced calbindin expression in both cortex and cerebellum rendering both tissues vulnerable to Ca2+-mediated pathology. Moreover, COVID-19 cortex and cerebellum exhibited RyR Ca2+ release channels with the biochemical signature of ‘‘leaky’’ channels and increased activity consistent with pathological intracellular Ca2+ leak. RyR2 were oxidized, associated with increased NADPH oxidase 2 (NOX2), and were PKA hyperphosphorylated on serine 2808, both of which cause loss of the stabilizing subunit calstabin2 from the channel complex promoting leaky RyR2 channels in COVID-19 patient brains. Furthermore, ex vivo treatment of COVID-19 patient brain samples with the Rycal drug ARM210, which is currently undergoing clinical testing at the National Institutes of Health for RyR1-myopathy (ClinicalTrials.gov Identifier: NCT04141670), fixed the channel leak. Thus, our experiments demonstrate that SARS-CoV-2 infection activates biochemical pathways linked to the tau pathology associated with AD and that leaky RyR Ca2+ channels may be a potential therapeutic target for the neurological complications associated with COVID-19.
The molecular basis of how SARS-CoV-2 infection results in ‘‘long COVID’’ is not well understood, and questions regarding the role of defective Ca2+ signaling in the brain in COVID-19 remain unanswered. A recent comprehensive molecular investigation revealed extensive inflammation and degeneration in the brains of patients that died from COVID-19,34 including in patients with no reported neurological symptoms. These authors also reported overlap between marker genes of AD and genes that are upregulated in COVID-19 infection, consistent with the findings of increased tau pathophysiology reported in the present study. We propose a potential mechanism that may contribute to the neurological complications caused by SARS-CoV-2: defective intracellular Ca2+ regulation and activation of AD-like neuropathology.
TGF-β belongs to a family of cytokines involved in the formation of cellular fibrosis by promoting epithelial-to-mesenchymal transition, fibroblast proliferation, and differentiation.35 TGF-β activation has been shown to induce fibrosis in the lungs and other organs by activation of the SMAD-dependent pathway. We have previously reported that TGF-β/SMAD3 activation leads to NOX2/4 translocation to the cytosol and its association with RyR channels, promoting oxidization of the channels and depletion of the stabilizing subunit calstabin in skeletal muscle and in heart.28, 30 Alteration of Ca2+ signaling may be particularly crucial in COVID-19-infected patients with cardiovascular/neurological diseases due, in part, to the multifactorial RyR2 remodeling after the cytokine storm, increased TGF-β activation, and increased oxidative stress. Moreover, SARS-CoV-2–infected patients exhibited a hyperadrenergic state. The elevated expression of glutamate carboxypeptidase 2 (GCPII) in COVID-19 brains reported in the present study could also contribute directly to increased PKA signaling of RyR2 by reducing PKA inhibition via metabotropic glutamate receptor 3 (mGluR3).36 Hyperphosphorylation of RyR2 channels can promote pathological remodeling of the channel and exacerbate defective Ca2+ regulation in these tissues. The increased Ca2+/cAMP/PKA signaling could also open nearby K+ channels which could potentially weaken synaptic connectivity, reduce neuronal firing,36 and could activate Ca2+ dependent enzymes.
Interestingly, both the cortex and cerebellum of SARS-CoV–2-infected patients exhibited a reduced expression of the Ca2+ buffering protein calbindin. Decreased calbindin could render these tissues more vulnerable to the cytosolic Ca2+ overload. This finding is in accordance with previous studies showing reduced calbindin expression levels in Purkinje cells and the CA2 hippocampal region of AD patients37–39 and in cortical pyramidal cells of aged individuals with tau pathology.33, 40 In contrast to the findings in the brains of COVID-19 patients in the present study, calbindin was not reduced in the cerebellum of AD patients, possibly protecting these cells from AD pathology.39, 41
Leaky RyR channels, leading to increased mitochondrial Ca2+ overload and ROS production and oxidative stress, have been shown to contribute to the development of tau pathology associated with AD.3, 23–29, 33 Recent studies of the effects of COVID-19 on the central nervous system have found memory deficits and biological markers similar to those seen in AD patients.42, 43 Our data demonstrate increased activity of enzymes responsible for phosphorylating tau (pAMPK, pGSK3β), as well as increased phosphorylation at multiple sites on tau in COVID-19 patient brains. The tau phosphorylation observed in these samples exhibited some differences from what is typically observed in AD, occurring in younger patients and in areas of the brain, specifically the cerebellum, that usually do not demonstrate tau pathology in AD patients. Taken together, these data suggest a potential contributing mechanism to the development of tau pathology in COVID-19 patients involving oxidative overload-driven RyR2 channel dysfunction. Furthermore, we propose that these pathological changes could be a significant contributing factor to the neurological manifestations of COVID-19 and in particular the “brain fog” associated with long COVID, and represent a potential therapeutic target for ameliorating these symptoms. For example, tau pathology in the cerebellum could explain the recent finding that 74% of hospitalized COVID-19 patients experienced coordination deficits.44 The data presented also raise the possibility that prior COVID-19 infection could be a potential risk factor for developing AD in the future.
The present study was limited to the use of existing autopsy brain tissues at the Columbia University Biobank from SARS-CoV-2–infected patients. The number of subjects is small and information on their cognitive function as well as their brain histopathology and levels of Aβ in cerebrospinal fluid and plasma are lacking. Furthermore, we did not have access to a suitable animal model of SARS-CoV-2 infection in which to test whether the observed biochemical changes in COVID-19 brains and potential cognitive and behavioral deficits associated with the brain fog of long COVID could be reversed or attenuated by therapeutic interventions. The design of future studies should include larger numbers of subjects that are age- and sex-matched. The cognitive function of SARS-CoV-2–infected patients who presented cognitive symptoms should be assessed and regularly monitored. Moreover, it is important to know whether the observed neuropathological signaling is unique to SARS-CoV-2 infection or are common to all other viral infections. Previous studies have reported cognitive impairment in Middle East respiratory syndrome45 as well as Ebola46, 47 patients. Retrospective studies comparing the incidence and the magnitude of cognitive impairments caused by these different viral infections would improve our understanding of these neurological complications of viral infections.
2 CONSOLIDATED RESULTS AND STUDY DESIGN
There were increased markers of oxidative stress (glutathione disulfide [GSSG]/ glutathione [GSH]) in the cortex (mesial temporal lobe) and cerebellum (cerebellar cortex, lateral hemisphere) of COVID-19 tissue. Kynurenic acid, a marker of inflammation, was increased in COVID-19 cortex and cerebellum brain lysates compared to controls, is in accordance with recent studies showing a positive correlation between kynurenic acid and cytokines and chemokine levels in COVID-19 patients.48–50
To determine whether SARS-CoV-2 infection also increases tissue TGF-β activity, we measured SMAD3 phosphorylation, a downstream signal of TGF-β, in control and COVID-19 tissue lysates. Phosphorylated SMAD3 (pSMAD3) levels were increased in COVID-19 cortex and cerebellum brain lysates compared to controls, indicating that SARS-CoV-2 infection increased TGF-β signaling in these tissues. Interestingly, brain tissues from COVID-19 patients exhibited activation of the TGF-β pathway, despite the absence of the detectable (by immunohistochemistry and polymerase chain reaction, data not shown) virus in these tissues. These results suggest that the TGF-β pathway is activated systemically by SARS-CoV-2, resulting in its upregulation in the brain, as well as other organs. In addition to oxidative stress, COVID-19 brain tissues also demonstrated increased PKA and calmodulin-dependent protein kinase II association domain (CaMKII) activity, most likely associated with increased adrenergic stimulation. Both PKA and CaMKII phosphorylation of tau have been reported in tauopathies.51, 52
The hallmarks of AD brain neuropathology are the formation of Aβ plaques from abnormal APP processing by BACE1, as well as tau ‘‘tangles’’ caused by tau hyperphosphorylation.53 Brain lysates from COVID-19 patients’ autopsies demonstrated normal BACE1 and APP levels compared to controls. The patients analyzed in the present study were grouped by age (young ≤ 58 years old, old ≥ 66 years old) to account for normal, age-dependent changes in APP and tau pathology. Abnormal APP processing was only observed in brain lysates from patients diagnosed with AD. However, AMPK and GSK3β phosphorylation were increased in both the cortex and cerebellum in COVID-19 brains. Activation of these kinases in SARS-CoV-2–infected brains leads to a hyperphosphorylation of tau consistent with AD tau pathology in the cortex. COVID-19 brain lysates from older patients showed increased tau phosphorylation at S199, S202, S214, S262, and S356. Lysates from younger COVID-19 patients showed increased tau phosphorylation at S214, S262, and S356, but not at S199 and S202, demonstrating increased tau phosphorylation in both young and old individuals and suggesting a tau pathology similar to AD in COVID-19–affected patients. Interestingly, both young and old patient brains demonstrated increased tau phosphorylation in the cerebellum, which is not typical of AD.
RyR channels may be oxidized due to the activation of the TGF-β signaling pathway.30 NOX2 binding to RyR2 causes oxidation of the channel, which activates the channel, manifested as an increased open probability that can be assayed using 3[H]ryanodine binding.54 When the oxidization of the channel is at pathological levels, there is destabilization of the closed state of the channel, resulting in spontaneous Ca2+ release or leak.27, 30 To determine the effect of the increased TGF-β signaling associated with SARS-CoV-2 infection on NOX2/RyR2 interaction, RyR2 and NOX2 were co-immunoprecipitated from brain lysates of COVID-19 patients and controls. NOX2 associated with RyR2 in brain tissues from SARS-CoV-2–infected individuals were increased compared to controls.
Given the increased oxidative stress and increased NOX2 binding to RyR2 seen in COVID-19 brains, RyR2 post-translational modifications were investigated. Immunoprecipitated RyR2 from brain lysates demonstrated increased oxidation, PKA phosphorylation on serine 2808, and depletion of the stabilizing protein subunit calstabin2 in SARS-CoV-2–infected tissues compared to controls. This biochemical remodeling of the channel is known as the ‘‘biochemical signature’’ of leaky RyR223, 55, 56 that is associated with destabilization of the closed state of the channel. This leads to SR/ER Ca2+ leak, which contributes to the pathophysiology of a number of diseases including AD.23, 24, 26, 30, 55–57 RyR channel activity was determined by binding of 3[H]ryanodine, which binds only to the open state of the channel. RyR2 was immunoprecipitated from tissue lysates and ryanodine binding was measured at both 150 nM and 20 μM free Ca2+. RyR2 channels from SARS-CoV-2–infected brain tissue demonstrated abnormally high activity (increased ryanodine binding) compared to channels from control tissues at physiologically resting conditions (150 nM free Ca2+), when channels should be closed. Interestingly, cortex and cerebellum of SARS-CoV-2–infected patients also exhibited a reduced expression of the Ca2+ binding protein calbindin. Calbindin is typically not reduced in the cerebellum of AD patients, possibly providing some protection against AD pathology. The low calbindin levels in the cerebellum of COVID-19 brains could contribute to the observed tau pathology in this brain region. An additional atypical finding in the COVID-19 brains studied in this investigation is an increased level of GCPII. This could contribute to the observed RyR PKA phosphorylation by increasing cAMP and inhibiting the metabotropic glutamate receptor type 3.36
3 DETAILED METHODS AND RESULTS
3.1.1 Human samples
De-identified human heart, lung, and brain tissue were obtained from the COVID BioBank at Columbia University. The cortex samples were from the mesial temporal lobe and the cerebellum samples were from the cerebellar cortex, lateral hemisphere. The Columbia University BioBank functions under standard operating procedures, quality assurance, and quality control for sample collection and maintenance. Age- and sex-matched controls exhibited absence of neurological disorders and cardiovascular or pulmonary diseases. Sex, age, and pathology of patients are listed in Table 1.TABLE 1. Sex, age, and pathology of COVID-19 patients
Acute hypoxic-ischemic injury in the hippocampus, pons, and cerebellum.
Hypoxic/ischemic injury, acute to subacute, involving hippocampus, medulla and cerebellum. Mild atherosclerosis. Mild arteriolosclerosis
Hypoxic-ischemic encephalopathy, acute, global, mild to moderate. Diffuse Lewy body disease, neocortical type, consistent with Parkinson disease dementia. Atherosclerosis, severe. Arteriolosclerosis, mild.
Lysate preparation and Western blots
Tissues (50 mg) were isotonically lysed using a Dounce homogenizer in 0.25 ml of 10 mM Tris maleate (pH 7.0) buffer with protease inhibitors (Complete inhibitors from Roche). Samples were centrifuged at 8000 × g for 20 minutes and the protein concentrations of the supernatants were determined by Bradford assay. To determine protein levels in tissue lysates, tissue proteins (20 μg) were separated by 4% to 20% sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) and immunoblots were developed using the following antibodies: pSMAD3 (Abcam, 1:1000), SMAD3 (Abcam, 1:1000), AMPK (Abcam, 1:1000), tau (Thermo Fisher, 1:1000), pTauS199 (Thermo Fisher, 1:1000), pTauS202/T205 (Abcam, 1:1000), pTauS262 (Abcam, 1:1000), GSK3β (Abcam, 1:1000), pGSK3βS9 (Abcam, 1:1000), pGSK3βT216 (Abcam, 1:1000), APP (Abcam, 1:1000), BACE1 (Abcam, 1:1000), GAPDH (Santa Cruz Biotech, 1:1000), CTF-β (Santa Cruz Biotechnology, Inc., 1:1000), Calbindin (Abcam, 1:1000), and GCPII (Thermo Fisher, 1:4000).
Analyses of ryanodine receptor complex
Tissue lysates (0.1 mg) were treated with buffer or 10 μM Rycal (ARM210) at 4°C. RyR2 was immunoprecipitated from 0.1 mg lung, heart, and brain using an anti-RyR2 specific antibody (2 μg) in 0.5 ml of a modified radioimmune precipitation assay buffer (50 mm Tris-HCl, pH 7.2, 0.9% NaCl, 5.0 mm NaF, 1.0 mm Na3VO4, 1% Triton X-100, and protease inhibitors; RIPA) overnight at 4°C. RyR2-specific antibody was an affinity-purified polyclonal rabbit antibody using the peptide CKPEFNNHKDYAQEK corresponding to amino acids 1367–1380 of mouse RyR2 with a cysteine residue added to the amino terminus. The immune complexes were incubated with protein A-Sepharose beads (Sigma) at 4°C for 1 hour, and the beads were washed three times with RIPA. The immunoprecipitates were size-fractionated on SDS-PAGE gels (4%–20% for RyR2, calstabin2, and NOX2) and transferred onto nitrocellulose membranes for 1 hour at 200 mA. Immunoblots were developed using the following primary antibodies: anti-RyR2 (Affinity BioReagents, 1:2500), anti-phospho-RyR-Ser(pS)-2808 (Affinity BioReagents 1:1000), anti- calstabin2 (FKBP12 C-19, Santa Cruz Biotechnology, Inc., 1:2500), and anti-NOX2 (Abcam, 1:1000). To determine channel oxidation, the carbonyl groups in the protein side chains were derivatized to DNP by reaction with 2,4-dinitrophenylhydrazine. The DNP signal associated with RyR2 was determined using a specific anti-DNP antibody according to the manufacturer using an Odyssey system (LI-COR Biosciences) with infrared-labeled anti-mouse and anti-rabbit immunoglobulin G (IgG; 1:5000) secondary antibodies.
RyR2 was immunoprecipitated from 1.5 mg of tissue lysate using an anti-RyR2 specific antibody (25 μg) in 1.0 ml of a modified RIPA buffer overnight at 4°C. The immune complexes were incubated with protein A-Sepharose beads (Sigma) at 4°C for 1 hour, and the beads were washed three times with RIPA buffer, followed by two washes with ryanodine binding buffer (10 mM Tris-HCl, pH 6.8, 1 M NaCl, 1% CHAPS, 5 mg/ml phosphatidylcholine, and protease inhibitors). Immunoprecipitates were incubated in 0.2 ml of binding buffer containing 20 nM [3H] ryanodine and either of 150 nM and 20 μm free Ca2+ for 1 hour at 37°C. Samples were diluted with 1 ml of ice-cold washing buffer (25 mm Hepes, pH 7.1, 0.25 m KCl) and filtered through Whatman GF/B membrane filters pre-soaked with 1% polyethyleneimine in washing buffer. Filters were washed three times with 5 ml of washing buffer. The radioactivity remaining on the filters is determined by liquid scintillation counting to obtain bound [3H] ryanodine. Nonspecific binding was determined in the presence of 1000-fold excess of non-labeled ryanodine.
GSSG/GSH ratio measurement and SMAD3 phosphorylation
Approximately 20 mg of tissue suspended in 200 μL of ice-cold phosphate-buffered saline/0.5% NP-40, pH6.0 was used for lysis. Tissue was homogenized with a Dounce homogenizer with 10 to 15 passes. Samples were centrifuged at 8000 × g for 15 minutes at 4°C to remove any insoluble material. Supernatant was transferred to a clean tube. Deproteinizing of the samples was accomplished by adding 1 volume ice-cold 100% (w/v) trichloroacetic acid (TCA) into five volumes of sample and vortexing briefly to mix well. After incubating for 5 minutes on ice, samples were centrifuged at 12,000 × g for 5 minutes at 4°C and the supernatant was transferred to a fresh tube. The samples were neutralized by adding NaHCO3 to the supernatant and vortexing briefly. Samples were centrifuged at 13,000 × g for 15 minutes at 4°C and supernatant was collected. Samples were then deproteinized, neutralized, TCA was removed, and they were ready to use in the assay. The GSSG/GSH was determined using a ratio detection assay kit (Abcam, ab138881). Briefly, in two separate assay reactions, GSH (reduced) was measured directly with a GSH standard and Total GSH (GSH + GSSG) was measured by using a GSSG standard. A 96-well plate was set up with 50 μL duplicate samples and standards with known concentrations of GSH and GSSG. A Thiol green indicator was added, and the plate was incubated for 60 minutes at room temperature (RT). Fluorescence at Ex/Em = 490/520 nm was measured with a fluorescence microplate reader and the GSSG/GSH for samples were determined comparing fluorescence signal of samples with known standards.
Kynurenic acid assay
Kynurenic acid (KYNA) concentration in brain lysates was determined using an enzyme-linked immunosorbent assay (ELISA) kit for KYNA (ImmuSmol). Briefly, samples (50 μl) were added to a microtiter plate designed to extract the KCNA from the samples. An acylation reagent was added for 90 minutes at 37°C to derivatize the samples. After derivatization, 50 μl of the prepared standards and 100 μl samples were pipetted into the appropriate wells of the KYNA microtiter plate. KYNA Antiserum was added to all wells and the plate was incubated overnight at 4°C. After washing the plate four times, the enzyme conjugate was added to each well. The plate was incubated for 30 minutes at RT on a shaker at 500 rpm. The enzyme substrate was added to all wells and the plate was incubated for 20 minutes at RT. Stop solution was added to each well. A plate reader was used to determine the absorbance at 450 nm. The sample signals were compared to a standard curve.
PKA activity assay
PKA activity in brain lysates was determined using a PKA activity kit (Thermo Fisher, EIAPKA). Briefly, samples were added to a microtiter plate containing an immobilized PKA substrate that is phosphorylated by PKA in the presence of ATP. After incubating the samples with ATP at RT for 2 hours, the plate was incubated with the phospho-PKA substrate antibody for 60 minutes. After washing the plate with wash buffer, goat anti-rabbit IgG horseradish peroxidase (HRP) conjugate was added to each well. The plate was aspirated, washed, and TMB substrate was added to each well, which was then incubated for 30 minutes at RT. A plate reader was used to determine the absorbance at 450 nm. The sample signals were compared to a standard curve.
CaMKII activity assay
CaMKII activity in brain lysates was determined using the CycLex CaM kinase II Assay Kit (MBL International). Briefly, samples were added to a microtiter plate containing an immobilized CaMKII substrate that is phosphorylated by CaMKII in the presence of Mg2+ and ATP. After incubating the samples in kinase buffer containing Mg2+ and ATP at RT for 1 hour, the plate was washed and incubated with the HRP conjugated anti-phospho-CaMKII substrate antibody for 60 minutes. The plate was aspirated, washed, and TMB substrate was added to each well, which was then incubated for 30 minutes at RT. A plate reader was used to determine the absorbance at 450 nm. The sample signals were compared to a standard curve.
Group data are presented as mean ± standard deviation. Statistical comparisons between the two groups were determined using an unpaired t-test. Values of P < .05 were considered statistically significant. All statistical analyses were performed with GraphPad Prism 8.0.
3.2.1 Oxidative stress and TGF-β, PKA, and CaMKII activation
Oxidative stress levels were determined in brain tissues (cortex, cerebellum) from COVID-19 patient autopsy tissues and controls by measuring the ratio of GSSG to GSH by an ELISA kit. COVID-19 patients exhibited significant oxidative stress with a 3.8- and 3.2-fold increase in GSSG/GSH ratios in cortex (Ctx) and cerebellum (CB) compared to controls, respectively (Figure 1A). High circulating levels of kynurenine have been reported in COVID-19.48–50 However, the expression of KYNA in COVID-19 brain tissue has not been examined. Levels in the Ctx and CB were measured using an ELISA kit. COVID-19 brains had a significant increase in the Ctx and CB compared to controls (Figure 1A). An additional marker of tissue inflammation is increased cytokine expression. SMAD3 phosphorylation, a downstream signal of TGF-β, was increased in COVID-19 Ctx and CB tissue lysates compared to controls (Figure 1B and 1C). Increased adrenergic activation in the brain of patients infected with SARS-CoV-2 was also demonstrated by measuring PKA activity in the Ctx and CB and CaMKII activity was increased as well (Figure 1D).
Activation of AD-linked signaling
Both PKA and CaMKII have been directly implicated in the increased phosphorylation of tau associated with AD.51, 52 Because COVID-19 brain lysates had increased PKA and CaMKII activity, AD-linked biochemistry was evaluated in the COVID-19 brain lysates. Normal APP processing was observed in COVID-19 brain lysates as demonstrated by normal BACE1 and APP levels compared to controls (Figure 2A and B). Abnormal APP processing was only observed in brain lysates from patients diagnosed with AD (see Table 1 for patient details). However, phosphorylation/activation of AMPK and GSK3β was observed in SARS-CoV-2–infected patient brain lysates. Activation of these kinases along with the activation of PKA and CaMKII (Figure 1) leads to a hyperphosphorylation of tau at multiple residues (Figure 2C and D). Tau hyperphosphorylation in the cerebellum is not typical of AD pathology. The CB tau pathology demonstrated in COVID-19 warrants further investigation.
RyR2 channel oxidation and leak
RyR2 biochemistry was investigated to determine whether RyR2 in COVID-19 brain tissues demonstrated a “leaky” phenotype. Increased NOX2/RyR2 binding was shown in Ctx and CB lysates from SARS-CoV-2–infected individuals compared to controls using co-immunoprecipitation (Figure 3A and B). In addition, RyR2 from SARS-CoV-2–infected brains had increased oxidation, increased serine 2808 PKA phosphorylation, and depletion of the stabilizing protein subunit calstabin2 compared to controls (Figure 3A and B). RyR channels exhibiting these characteristics can be inappropriately activated at low cytosolic Ca2+ concentrations resulting in a pathological ER/SR Ca2+ leak. 3[H]Ryanodine binding to immunoprecipitated RyR2 was measured at both 150 nM and 20 μM free Ca2+. Because ryanodine binds only to the open state of the channel under these conditions, 3[H]Ryanodine binding may be used as a surrogate measure of channel open probability. The total amount of RyR immunoprecipitated was the same for control and COVID-19 samples (data not shown). Increased RyR2 channel activity at resting conditions (150 nM free Ca2+) was observed in COVID-19 channels compared to controls (Figure 3C). Under these conditions, RyR channels should be closed. Rebinding of calstabin2 to RyR2, using a Rycal, has been shown to reduce SR/ER Ca2+ leak, despite the persistence of the channel remodeling. Indeed, calstabin2 binding to RyR2 was increased when COVID-19 patient brain tissue lysates were treated ex vivo with the Rycal drug ARM210 (Figure 3A and B). Abnormal RyR2 activity observed at resting Ca2+ concentration was also decreased by Rycal treatment (Figure 3C).
An interesting finding concerning the tau phosphorylation in brain lysates from SARS-CoV-2 patients was the increase of phosphorylation at multiple sites in the cerebellum. This is atypical of AD. One potential mechanism to explain this finding is the significantly decreased levels of calbindin expressed in COVID-19 cerebellum (Figure 3D, 3E). The decreased cerebellar calbindin levels could make this area of the brain more susceptible to Ca2+-induced activation of enzymes upstream of tau phosphorylation. Moreover, increased GCPII expression was observed in COVID-19 cortex and cerebellar lysates (Figure 3D, 3E), which would reduce mGluR3 inhibition of PKA signaling and could contribute to the PKA hyperphosphorylation of RyR2.
Model for the role for leaky RyR2 in the pathophysiology of SARS-CoV-2 infection
Our data indicate a role for leaky RyR2 in the pathophysiology of SARS-CoV-2 infection (Figure 4). In addition to the brain of COVID-19 patients, we observed increased systemic oxidative stress and activation of the TGF-β signaling pathway in lung, and heart, which correlates with oxidation-driven biochemical remodeling of RyR2 (Figure 3 and S1 in supporting inormation). This RyR2 remodeling results in intracellular Ca2+ leak, which can play a role in heart failure progression, pulmonary insufficiency, as well as cognitive dysfunction.23–26, 28 The alteration of cellular Ca2+ dynamics has also been implicated in COVID-19 pathology.58, 59 Taken together, the present data suggest that leaky RyR2 may play a role in the long-term sequelae of COVID-19, including the “brain fog” associated with SARS-CoV-2 infection which could be a forme fruste of AD,60 and could predispose long COVID patients to developing AD later in life. Leaky RyR2 channels may be a therapeutic target for amelioration of some of the persistent cognitive deficits associated with long COVID.