The FDA approved aducanumab based on biomarker data hoping the drug will eventually improve clinical outcomes. Like many in the field, the agency reasons that because brain amyloid deposition precedes cognitive decline, removing amyloid will slow that slide. Time will tell, but relying on unproven biomarkers can be a costly misfire. Consider the HDL debacle: Pharmaceutical companies spent billions pursuing drugs that boost plasma levels of this “good” high-density lipoprotein, only to realize they had barked up the wrong tree. The drugs did increase HDL levels all right, but they did not reduce heart disease and, in some cases, made patients worse. Could the Alzheimer’s field be heading down a similar path?   

  • Scientists remain largely empty-handed in their hunt for neurodegenerative disease treatment.
  • Are their underlying hypotheses misguided?
  • Mendelian randomization could help validate targets.

“The disconnect between pharmaco-epidemiology and benefits in therapeutic trials is a remarkable phenomenon that has vexed the cardiovascular field as well as the Alzheimer one,” noted David Knopman, Mayo Clinic, Rochester, Minnesota.

For neurodegenerative diseases, this epidemiological disconnect has long loomed large. Early in the field, approved drugs were repurposed for AD trials because epidemiological studies had suggested a neuroprotective effect. Think statins, NSAIDs, insulin—all were duds.

Targets or Decoys? Some scientists worry that using key neurodegenerative disease proteins as targets, and also as surrogate biomarkers, can mislead drug developers. [Modified from wikimediacommons under CC-BY-4.0.]

The amyloid hypothesis has not been bountiful, either. No anti-amyloid treatments tested in trials to date, even those that moved Aβ biomarkers in the expected direction, have meaningfully improved clinical outcomes. BACE inhibitors worsened cognition. Immunotherapies that lower tau in the CSF have not yet helped people with primary tauopathies. Ditto for drugs to boost urate, a protective marker linked to Parkinson’s. A trial of an antisense oligonucleotide that reduces huntingtin in the CSF recently failed because it held more risk than benefit. Are scientists putting too much faith in the cherished assumption that treating even a well-substantiated biomarker of disease will benefit the patient?

Some scientists see a rush to develop drugs before the biology behind neurodegenerative diseases is sufficiently understood. “The field has been challenged by a poor understanding of the consequences of the genetic lesions that protect from, or increase risk for, disease, and too often has relied on a relatively superficial understanding of biomarkers without understanding the molecular mechanisms involved,” said Zachary Sweeney, who recently started a multi-omics company after five years of overseeing discovery research at Denali Therapeutics, San Francisco.

The HDL saga offers a cautionary tale. Decades of research had shown that people with high levels of this marker in their blood were less likely to have cardiovascular disease. Ergo, scientists expected that increasing HDL with a drug would achieve the same. Pfizer, Roche, Eli Lilly, Merck, and others jumped in, developing small molecules to block the activity of cholesteryl ester transfer protein. CETP passes cholesteryl esters from HDL to triglyceride-rich lipoproteins, including the “bad” low-density kind. Blocking CETP increased plasma HDL in animal models of atherosclerosis and protected them from heart disease. It all looked like it would work in humans,

Alas, it was not to be. In a Phase 3 trial of more than 15,000 people at high risk for cardiovascular disease, Pfizer’s torcetrapib failed to protect. More people on the drug had heart attacks and even died, despite a 72 percent bump in plasma HDL. Pfizer axed torcetrapib in 2006. That shocker was followed by another six years later, when Roche stopped its Phase 3 trial of dalcetrapib in nearly 16,000 people for lack of efficacy, again despite increases in HDL. And in 2014, Eli Lilly aborted evacetrapib when a Phase 3 in more than 12,000 people proved futile. Refusing to call off the hunt, Merck upped the ante, testing its CETP inhibitor in more than 30,000 people at high risk for vascular disease in a four-year-long Phase 3 trial. While anacetrapib modestly reduced risk for major cardiovascular events, the effect seemed to stem more from its lowering LDL than boosting HDL. Merck pulled the plug on anacetrapib in 2017 (for a review see Tall and Rader, 2018). 

What had happened?

“Part of the problem stems from not really understanding the basic pathogenesis of the disease,” said Knopman. “But an almost certain contributor is indication bias toward healthy people that confounds pharmaco-epidemiological inferences of causality,” he added. In other words, people who have high plasma HDL may be healthy for a variety of reasons, and it is those other factors, not HDL, that hold cardiovascular disease at bay.

Human genetics research bears this out. Over the past decade, Mendelian randomization has become a powerful tool to discern whether the relationship between a trait and a disease is causal or merely correlative. This is helpful, especially when it comes to biomarkers, because Mendelian randomization eliminates variables that confound epidemiological data, such as socioeconomics and lifestyle factors such as diet, exercise, and others. Lo and behold, a large Mendelian randomization published in 2012 did not support the premise that simply boosting HDL would decrease a person's risk for myocardial infarction.

The study showed that people who inherit a serine-for-asparagine point mutation at amino acid 396 of endothelial lipase have consistently higher plasma HDL—enough to reduce their incidence of myocardial infarcts by 13 percent if one goes by the epidemiology data. And yet, the study also showed that among 116,320 people across 20 studies, carriers of the serine variant had just as much risk for infarcts as those with the asparagine variant. Likewise, a genetic score based on 14 single-nucleotide polymorphisms that exclusively associate with HDL suggested that an increase in the lipoprotein by one standard deviation does not reduce the risk for myocardial infarction as suggested by epidemiology (Voight et al., 2012). 

The neurodegenerative disease field is not immune from similar pitfalls. In the case of Parkinson’s, epidemiology data had associated high blood levels of the antioxidant urate with lower incidence and slower progression of PD (de Lau et al., 2005; Chen et al., 2009; Gao et al., 2016). After a Phase 2 trial suggested that the precursor inosine could increase levels of urate in the plasma, a Phase 3 trial began in 2016. Although the treatment group had a 50 percent increase in plasma urate over the course of 19 months, their disease worsened just as fast as in the placebo group. In 2018, as in the case of HDL, Mendelian randomization studies found no association between genetic loci that influence plasma urate level and PD incidence or prevalence (Kobylecki et al., 2018; Kia et al., 2018). The clinical data from the negative Phase 3 trial were published this month (Parkinson Study Group SURE-PD3 Investigators, 2021). 

“Mendelian randomization predicted that drugs that would modulate urate would not have a beneficial effect on PD,” said Sweeney. “This really has to be the approach. If you can’t validate your target, then you have a challenge.”

How does this bode for other neurodegenerative diseases? Recent MR studies found no causal link between levels of serum interleukin-6 and risk for AD, PD, or ALS, or between dietary polyunsaturated fatty acids and AD (Zhang et al., 2020; Tomata et al., 2019). MR found no relation between rheumatoid arthritis and AD risk, and predicted that anti-TNF-α drugs, commonly and effectively used to treat RA, would have no benefit in AD (Andrews and Goate, 2019). 

More concerningly, MR suggests that FDA-approved inhibitors of PCSK9, a key protease in the regulation of cholesterol and lipid metabolism, might put people at risk for AD. Gain-of-function mutations in PCSK9 cause a form of familial hypercholesterolemia, while loss-of-function mutations and monoclonal antibodies that target the protease lower plasma levels of LDL-cholesterol. If that drops by one standard deviation, then AD risk increases by almost 50 percent, according to one MR analysis. The same study finds no genetic support for repurposing certain statins to ward off AD (Williams et al., 2020). 

Aside from these examples, AD drug development research has been slow to adopt MR as a way of validating its biomarkers and drug targets. This might be because for Alzheimer’s, the genetic case for the involvement of Aβ has always been strong. Dozens of pathogenic AD mutations all shift proteolytic processing of the peptide, increasing the proportion of longer, more amyloidogenic forms, while the A673T mutation in APP that limits processing protects against AD (July 2012 news). 

And yet the field has long struggled to explain exactly how Aβ leads to dementia. The correlation between amyloid plaques and cognition is weak, plus many other factors appear to play a role. Many scientists contend the data does not support the FDA’s decision to accept brain amyloid reduction as an approvable surrogate for subsequent clinical benefit (Hershey and Tarawneh, 2021; Planche and Villain, 2021; Alexander et al., 2021). 

The debate is reinvigorating criticism of Aβ as the key therapeutic target. “The main premise on which Alzheimer’s and all neurodegenerative diseases are conceived, is essentially the idea that proteins are toxic. It should end,” Alberto Espay, University of Cincinnati, told Alzforum. Espay believes plaques are problematic only because they remove Aβ42 from the brain. In the June eClinical Medicine, Espay and Kariem Ezzat at the Karolinska Institute, Stockholm, revive the old idea that it is this loss of soluble protein, not the presence of plaques, that causes neurodegeneration and cognitive decline. If they were right, ongoing Aβ-targeting investigational therapies would be heading in the wrong direction.

The authors reached their conclusion by analyzing data from 598 ADNI participants. They asked what distinguishes cognitively healthy people with amyloid plaques from cognitively impaired people with similar levels of amyloid plaques. “We found that people who have normal cognition have higher soluble Aβ42 no matter how much plaque they have,” Ezzat told Alzforum.

Aβ42 vs. Plaques. Among almost 600 people in ADNI with a positive amyloid PET scan, those with higher CSF Aβ42 levels are more likely to have normal cognition and hippocampal volume. [Image courtesy Karolinska Institute.]

Others found the paper provocative but unconvincing. “I would have to see the relative loss of Aβ42 versus Aβ40,” said Henrik Zetterberg, University of Gothenburg, Sweden. The former is stickier and gets pulled more easily into amyloid plaques, and as it does, the Aβ42/Aβ40 ratio in the CSF falls. “If the people who are still cognitively intact and resistant to amyloid have a normal ratio, then the authors may be onto something,” Zetterberg said. “But if the ratio is low, then they might have just found people who produce high levels of Aβ and who are simply in an early stage of degeneration.”

Knopman is with Zetterberg on this one, writing, “I’d like to see how Aβ40 fares in the same comparison.” He also noted that the analysis might have stacked the deck in favor of Aβ42. While the authors’ conclusion came down to a stronger correlation between cognition and CSF Aβ42 than between cognition and global amyloid PET, Knopman noted that people with low amyloid were excluded, whereas people with high CSF Aβ42 were not. “Generally, one would expect that exclusion of all of the low-amyloid PET people would reduce the apparent strength of the association: a clear bias against amyloid PET,” he wrote. Nevertheless, Knopman applauded the authors for putting the idea forward. “I hope they will flesh out their argument in the future with stronger analyses,” he added.

Espay and Ezzat want their findings to inspire a paradigm shift on how we view disease markers. “Our key message is that neurodegenerative diseases, in general, are associated with loss of protein,” said Espay. He contends that yes, aggregates accumulate, but total soluble protein goes down and that is what leads to disease. Tau falls in tauopathies, synuclein in Parkinson’s, Aβ in AD, progranulin in FDD/ALS. “There is no individual with a neurodegenerative disease who has higher levels of any of these [soluble] proteins, so it’s hard to sustain the idea that they are toxic when they are being depleted,” he said. “We should change ‘proteinopathy’ to ‘proteinopenia’ because levels and function are being lost,” he concluded.

Most current therapeutic approaches aim to reduce protein levels. “We’ve already done the experiment in people,” said Espay. “In the case of verubecestat, we’ve lowered soluble Aβ to such an extent that it led to cognitive loss.” Other scientists argue for trying these drugs again at much lower doses (McDade et al., 2021). 

Where does that leave similar strategies targeting tau, huntingtin, synuclein, C9ORF72, and other neurodegenerative disease proteins? Trials thus far look underwhelming. In the case of tau, Biogen’s gosuranemab lowered N-terminal fragments completely in the CSF without affecting tangles, cognitive decline in AD, or clinical progression in PSP (June 2021 news; Dec 2019 news). Roche’s semorinemab reduced CSF total tau and phospho-tau but not cognitive decline in people with prodromal or mild AD in the Phase 2 Tauriel trial, though it did seem to slow decline in some patients in the Phase 2 LAURIET trial (Mar 2021 conference news; Sep 2021 news). AbbVie’s tilavonemab nearly halved CSF tau, without any effect on progression in a Phase 2 PSP trial (Hoglinger et al., 2021; Jul 2019 news). 

Is reducing tau in the brain the right approach? “I don’t see how targeting normal tau, which we all have in our brains, will work,” said Zetterberg. “From my point of view, the tau part of the neurodegeneration story is a loss of tau function at the microtubules, so I’m afraid that downregulating approaches, be it antisense oligonucleotides or antibodies, might not be good.”

Sweeney agrees. He thinks that once the field started to focus on targeting extracellular tau, research became driven by a narrative more than data. “We are not trying sufficiently to understand the exact functional consequences of tau coding and non-coding genetic variants, how, for example, the tau coding variants change native tau protein interactions. And we are not using all the tools we now have available to validate our therapeutic hypotheses,” he said. Indeed, once tau leaves its home base on microtubules, it does not merely form neurofibrillary tangles. Rather, it can interfere broadly with cell biology, including lysosomal degradation, nucleocytoplasmic transport, synapse and mitochondrial function (Sep 2019 conference news; Jan 2019 news; Tracy et al., 2019). 

Other neurodegenerative protein targets are not looking much snappier. Last March, Roche halted a Phase 3 trial of its antisense oligonucleotide (ASO) therapy against huntingtin. Though tominersen reduced CSF mHtt by half, treated patients fared no better than those on placebo (Mar 2018 news; Endpoints news). This came as no surprise to Espay. “The ASO did exactly what it was intended to do. It lowered htt in CSF, but the brain atrophy and behavior got worse.”

Still, the end of tominersen was an especially painful blow to the Huntington's community, because a week later, a Phase 1/2 ASO developed by Wave Life Sciences was also discontinued (Kwon, 2021). 

Ezzat and Espay suggest adding back these proteins, rather than trying to deplete them. “We want to replace peptides that are there with proteins that are less aggregation-prone,” said Ezzat. This type of approach is already being used to treat diabetes, where pramlintide, a less-amyloidogenic form of amylin, is injected along with insulin to reduce blood-sugar levels.

Whether the field is chasing the wrong biomarkers and targets will be hard to answer until a treatment improves clinical outcomes. “Only then we will be able to qualify a biomarker,” said Zetterberg. “For now, we are in limbo but I’m more optimistic after seeing recent data on donanemab and lecanemab. There looks to be some type of clinical benefit, and that’s why the Phase 3 trials will be super important.”

Knopman agreed that without a clinical outcome, the field is hamstrung trying to validate a marker, but he is less optimistic about targeting amyloid. “The mark of a sophisticated field is that it doesn’t make the same mistake 10 times before approaching epidemiological claims of therapeutic benefits with more skepticism,” he said.—Tom Fagan


  1. I read with great interest the paper by Espay’s lab, which favors the hypothesis of a loss of function of the Aβ protein in the onset of AD neurodegeneration. My laboratory has been studying the physiological functions associated with soluble Aβ monomers for over a decade (Giuffrida et al., 2009; Giuffrida et al., 2015; Santangelo et al., 2021) and, therefore, this paper is more than welcome for us.

    However, I would be cautious in stating that AD is a proteinopenia rather than a proteinopathy, as the two can co-exist. Indeed, the self-association of functional Aβ monomers into soluble toxic oligomers could determine both at the same time. At a certain point, soluble monomers could be depleted by direct interaction with preformed fibrils in secondary nucleation processes. Therefore, we need to understand what the relative contribution of amyloid proteinopenia and proteinopathy is across the neurodegenerative event. Until this is clarified, I agree with being cautious with Aβ-targeting therapeutic strategies. Meanwhile, neuroprotective strategies limiting oligomers toxicity, which are lagging behind, would not hurt.


    . Beta-amyloid monomers are neuroprotective. J Neurosci. 2009 Aug 26;29(34):10582-7. PubMed.

    . Monomeric ß-amyloid interacts with type-1 insulin-like growth factor receptors to provide energy supply to neurons. Front Cell Neurosci. 2015;9:297. Epub 2015 Aug 7 PubMed.

    . β-amyloid monomers drive up neuronal aerobic glycolysis in response to energy stressors. Aging (Albany NY). 2021 Jul 21;13(14):18033-18050. PubMed.

Make a Comment

To make a comment you must login or register.


News Citations

  1. Protective APP Mutation Found—Supports Amyloid Hypothesis
  2. Biogen Shelves Gosuranemab After Negative Alzheimer’s Trial
  3. Gosuranemab, Biogen’s Anti-Tau Immunotherapy, Does Not Fly for PSP
  4. N-Terminal Tau Antibodies Fade, Mid-Domain Ones Push to the Fore
  5. First Cognitive Signal that Tau Immunotherapy Works?
  6. AbbVie’s Tau Antibody Flops in Progressive Supranuclear Palsy
  7. Can Induced Neurons Identify Early Signs of Neurodegeneration?
  8. Invasion of the Microtubules: Mutant Tau Deforms Neuronal Nuclei
  9. Antisense Therapy Cuts Huntingtin Protein in CSF by Half

Paper Citations

  1. . Trials and Tribulations of CETP Inhibitors. Circ Res. 2018 Jan 5;122(1):106-112. Epub 2017 Oct 10 PubMed.
  2. . Plasma HDL cholesterol and risk of myocardial infarction: a mendelian randomisation study. Lancet. 2012 Aug 11;380(9841):572-80. Epub 2012 May 17 PubMed.
  3. . Serum uric acid levels and the risk of Parkinson disease. Ann Neurol. 2005 Nov;58(5):797-800. PubMed.
  4. . Plasma urate and Parkinson's disease in the Atherosclerosis Risk in Communities (ARIC) study. Am J Epidemiol. 2009 May 1;169(9):1064-9. Epub 2009 Mar 18 PubMed.
  5. . Prospective study of plasma urate and risk of Parkinson disease in men and women. Neurology. 2016 Feb 9;86(6):520-6. Epub 2016 Jan 13 PubMed.
  6. . Plasma urate and risk of Parkinson's disease: A mendelian randomization study. Ann Neurol. 2018 Aug;84(2):178-190. Epub 2018 Aug 25 PubMed.
  7. . Mendelian randomization study shows no causal relationship between circulating urate levels and Parkinson's disease. Ann Neurol. 2018 Aug;84(2):191-199. PubMed.
  8. . Effect of Urate-Elevating Inosine on Early Parkinson Disease Progression: The SURE-PD3 Randomized Clinical Trial. JAMA. 2021 Sep 14;326(10):926-939. PubMed.
  9. . Mendelian randomization study to evaluate the effects of interleukin-6 signaling on four neurodegenerative diseases. Neurol Sci. 2020 Oct;41(10):2875-2882. Epub 2020 Apr 24 PubMed.
  10. . Polyunsaturated fatty acids and risk of Alzheimer's disease: a Mendelian randomization study. Eur J Nutr. 2019 Nov 1; PubMed.
  11. . Mendelian randomization indicates that TNF is not causally associated with Alzheimer's disease. Neurobiol Aging. 2019 Dec;84:241.e1-241.e3. Epub 2019 Sep 9 PubMed.
  12. . Lipid lowering and Alzheimer disease risk: A mendelian randomization study. Ann Neurol. 2020 Jan;87(1):30-39. PubMed.
  13. . Clinical Efficacy, Drug Safety, and Surrogate Endpoints: Has Aducanumab Met All of Its Expectations?. Neurology. 2021 Sep 14;97(11):517-518. Epub 2021 Jul 7 PubMed.
  14. . US Food and Drug Administration Approval of Aducanumab-Is Amyloid Load a Valid Surrogate End Point for Alzheimer Disease Clinical Trials?. JAMA Neurol. 2021 Nov 1;78(11):1307-1308. PubMed.
  15. . Revisiting FDA Approval of Aducanumab. N Engl J Med. 2021 Aug 26;385(9):769-771. Epub 2021 Jul 28 PubMed.
  16. . The case for low-level BACE1 inhibition for the prevention of Alzheimer disease. Nat Rev Neurol. 2021 Nov;17(11):703-714. Epub 2021 Sep 21 PubMed.
  17. . Safety and efficacy of tilavonemab in progressive supranuclear palsy: a phase 2, randomised, placebo-controlled trial. Lancet Neurol. 2021 Mar;20(3):182-192. PubMed.
  18. . Tau interactome mapping reveals dynamic processes in synapses and mitochondria associated with neurodegenerative disease. bioRxiv. June 17, 2019 bioRxiv.
  19. . Failure of genetic therapies for Huntington's devastates community. Nature. 2021 May;593(7858):180. PubMed.

Other Citations

  1. mutations

External Citations

  1. wikimediacommons
  2. CC-BY-4.0
  3. Endpoints news

Further Reading

No Available Further Reading

Primary Papers

  1. . High cerebrospinal amyloid-β 42 is associated with normal cognition in individuals with brain amyloidosis. EClinicalMedicine. 2021 Aug;38:100988. Epub 2021 Jun 28 PubMed.