The Amyloid Paradox: What Can Alzheimer’s Disease Drug Development Learn from Oncology?
By Dennis Chang
This week, the FDA approved Biogen and Eisai’s anti-amyloid antibody aducanumab (Aduhelm) for Alzheimer’s disease (AD).
The controversy embroiling this decision has been and will likely continue to be heated. However, regardless of whether or not one believes the FDA decision was the correct one, it is apparent that aducanumab is at best an incremental advance; it is not a transformative treatment for AD. If it were, then the EMERGE and ENGAGE trials would not have been terminated for futility, and the controversy around aducanumab’s approval would not exist. Even aducanumab’s advocates acknowledge this; the essence of their argument is that AD is such a devastating and hard-to-treat condition that even an incremental advance is worth making available to patients as soon as possible. On the other hand, aducanumab’s detractors argue that the evidence has not yet established whether the drug has any benefit, and thus the drug toxicity and financial costs are not justified. The specific arguments made on both sides of the debate have been laid out previously1, 2, 3, 4, 5 and will continue to be discussed in the wake of aducanumab’s approval; we will not repeat them here in detail.
Instead, we reflect that the aducanumab controversy continues to highlight an apparent paradox in anti-amyloid drug development. On one hand, the drug class and mechanism are grounded in a mountain of biological evidence so vast and diverse that it seems unimpeachable. On the other hand, the track record of empirical failure—of drug after drug, trial after trial, year after year—is so long that it seems to have reached the point of sheer folly.
|Biological evidence in support of|
anti-amyloid therapy for AD55, 6, 7
|Examples of clinical trial failures|
of amyloid-based therapies for AD55, 8, 9
|• Autopsy studies: Amyloid plaques are a defining feature of AD pathology|
• Human biomarker studies: e.g., a rise in amyloid beta (Aβ) peptides in CSF precedes AD development
• Human genetics: e.g., familial forms of AD are associated with genes encoding amyloid precursor protein (APP) and presenilins involved in APP processing
• Mouse genetics: e.g., transgenic mice recapitulate the findings of human genetics and respond to anti-amyloid therapy
• Biochemical studies: e.g., Aβ peptides are directly neurotoxic in animal models and cultured neurons
• Clinical studies: Multiple anti-amyloid clinical studies have shown hints of slowing the progression of mild AD
• Bapineuzumab [Pfizer/JNJ]: targets N-terminus of Aβ; discontinued in ph 3
Small-molecule inhibitors of BACE1 (beta-secretase) involved in Aβ peptide generation
• Verubecestat [Merck]: discontinued in ph 3
• Tarenflurbil [Encore/Myriad]: gamma-secretase inhibitor; discontinued in ph 3
This paradoxical pattern has arisen for multiple drug mechanisms in oncology. The lessons from those cancer therapies may define a path forward for amyloid-based therapies and for AD drug development more broadly.
The Case Study of PI3K in Oncology
PI3K (phosphoinositide-3 kinase) inhibition is a case in point. The evidence for an oncogenic role of PI3K (in solid tumors primarily the α and β isoforms) is extremely well established through diverse lines of evidence. And yet for years, clinical development of PI3K inhibitors resulted in failure after failure.
|Biological evidence of|
an oncogenic role for PI3K1010, 11, 12
|Clinical trial outcomes for PI3K inhibitors8, 11, 13|
|• Human genetics: PIK3CA (the gene encoding PI3Kα) is mutated in >10% of all cancers; PTEN (a natural inhibitor of PI3K) is mutated in >5% of cancers|
• Prognostic studies: Both PIK3CA activating mutations and PTEN loss are associated with poor prognosis in multiple cancer types
• Transgenic mouse studies have confirmed oncogenic role of PI3K pathway activation
• Preclinical studies of PI3K pathway inhibitors including cell line, mouse, and patient-derived xenograft studies, for dozens of inhibitors starting with wortmannin in the 1980s
• Clinical studies: Efficacy signals reported for many inhibitors
|PI3K/mTOR dual inhibitors|
• DS-7423 [Daiichi]: discontinued after ph 1
Broad-spectrum PI3K inhibitors
• PX-866 [Cascadian]: discontinued in ph 2
• AZD8186 [AstraZeneca]: discontinued in ph 1
• Taselisib [Roche]: discontinued in ph 3
The pattern looks remarkably similar to amyloid in AD but with one key difference: the PI3K inhibitor alpelisib was not just successfully approved, it showed definitive, unambiguous clinical benefit in its pivotal trial. There have also been prior approvals of inhibitors of mTOR, a kinase acting downstream of PI3K, as well as multiple inhibitors of PI3Kδ for B-cell lymphomas. How did PI3K drug development break the pattern?
Other pathways that fit this paradoxical pattern include CDKs (cyclin-dependent kinases), FGFR (fibroblast growth factor receptor), and HDAC (histone deacetylase). But the PI3K example is sufficiently representative to extract key lessons, which are as follows.
Lesson 1. Distinguish Between “Drivers” and “Passengers”
The oncology field has resolved the paradox by recognizing that for any given cancer—which may have many different altered targets or pathways—one or a few alterations may be much more important than others. I.e., some mutations/pathways are “drivers” while the rest are “passengers”14.
In the example of HR+ (hormone receptor positive) breast cancer, we know that the main driver of cancer growth is estrogen signaling. This is well established by the strong efficacy of anti-estrogen therapy across multiple lines of therapy as well as an abundance of basic biology and preclinical studies.15 Even if a PIK3CA mutation is present, a PI3K inhibitor on its own—in the absence of estrogen inhibition—would have little effect. There is not universal consensus on the definitions of drivers and passengers—e.g., there may be debate over whether PIK3CA is a passenger or a weak driver—but the concept is still broadly useful.
We would also caveat that the classification may shift depending on the context. One specific context dependency in AD may be the timing of intervention: there is a hypothesis that trials have failed because they were being run too late in the course of disease. If that is the case, then amyloid may transiently be a driver during initial stages of disease but still should not be considered a driver of disease progression later. In oncology, this is analogous to the distinction between preventing carcinogenesis and treating cancer; if a target/mechanism is critical for the initiation of cancer but is not a driver later, then if we are trying to treat disease that has already developed, then a drug for that target is unlikely to be very effective.
In any case, if a pathway is clearly and universally involved in disease pathophysiology, but single-agent drug therapy lacks (much) efficacy, this can be explained by the importance of other pathways. In Alzheimer’s disease, the amyloid pathway fits this “passenger” pattern of evidence remarkably well. The obvious question then is, if amyloid is just a “passenger” then what other pathway or pathways are most important? This remains an unresolved question, but the Alzheimer’s research community is actively testing many hypotheses, including tau dysregulation, neuroinflammation and microglial dysfunction, autophagy and metabolic impairments, synaptic plasticity modulation, and others.5 These efforts are crucial: if amyloid is a “passenger” then mechanisms beyond amyloid are essential to transforming disease treatment.
Lesson 2. Pursue Combination Therapy
A corollary lesson is that if your pathway is a passenger, then a drug targeting that pathway will have the most benefit not as a single agent but in combination with another drug, ideally one targeting the driver (if there is one, and it is known). However, what do you do if the optimal combination partner is unknown?
This situation also occurs frequently in oncology, and one of the approaches used in this situation is a multi-arm platform study. Typically, a disease and a “backbone” drug are chosen, and a variety of combination partners are assessed in different study arms. Selected examples include the following trials16:
- ADPT01C (NCT04294160) [Novartis]: Dabrafenib-based combinations for BRAF V600-mutant colorectal cancer
- KEYMAKER-U01 (NCT04165798) [Merck]: Pembrolizumab-based combinations for non-small cell lung cancer (NSCLC)
- MORPHEUS-TNBC (NCT03424005) [Roche]: Atezolizumab-based combinations for triple-negative breast cancer
- ORCHARD (NCT03944772) [AstraZeneca]: Osimertinib-based combinations in EGFR-mutant NSCLC after prior osimertinib treatment
- PORTER (NCT03835533) [Parker Institute]: Nivolumab-based immunotherapy combinations for castrate-resistant prostate cancer
- Precision Promise (NCT04229004) [PanCAN]: Chemo-based combinations for pancreatic adenocarcinoma
The last two examples are notable in that they were created and are run by a non-profit entity that can broker collaborations across multiple companies, facilitating combination of drugs owned and developed by different manufacturers. Most of the combination umbrella trials listed above are run by a single company using only drugs by that one manufacturer, which limits the potential scope of exploration. To the credit of Roche, the MORPHEUS trials are exceptions, and do involve cross-company collaboration. Clarion has written previously about the need for greater cross-company collaboration in combination clinical trials in oncology, and the value of non-profit-led initiatives to run those trials17.
Lesson 3. Make Precision Medicine Fundamental
Another critical feature of the PI3K story is the use of precision medicine. During the last 17 years—which is the time span since the last drug before aducanumab approved for Alzheimer’s disease, Namenda (memantine) [AbbVie] in October 2003—there were approximately 400 FDA approvals for oncology therapeutics, and none of them were approved simply for “cancer” or “advanced cancer”. Every single one of the hundreds of approvals have been for a subtype of cancer, as defined by site of origin, histology, molecular biomarkers, stage of disease, line of therapy, or (often) a combination of multiple such criteria18. The alpelisib approval is a case in point: it was specifically for breast cancer that was HR-positive, HER2-negative, and PIK3CA-positive, in the advanced/metastatic setting, after progression on prior endocrine therapy. This is a stark contrast to AD, where clinical trials tend to pursue a broad AD indication—selecting patients mainly based on severity level (mild, moderate, severe) as a proxy for stage of disease.
Precision medicine “aims to ensure the delivery of the right treatment to the right patient at the right time”19. This principle is fundamental to oncology but should be embraced by all diseases that are biologically heterogeneous. Differences in biology imply the need for different treatments. And AD is certainly heterogeneous. Studies have shown variations in biomarkers such as phospho-tau levels or the ratios of different Aβ peptides, the involvement of different genotypes (e.g., at the APOE locus; also separating familial early-onset AD from sporadic AD cases), variations in the brain regions most severely affected (as assessed by PET scan), and variations in immune/inflammatory biomarkers. One may also want to consider the cardiovascular health/history of the patient due to the potential for concomitant vascular dementia or stroke risk, other co-morbidities, and other clinical factors.
In oncology, a precision medicine strategy may identify patient subpopulations where a drug has exceptional efficacy, even though it lacks meaningful single-agent activity in broader indications. For example, the CSF1R inhibitor pexidartinib [Daiichi] is approved as monotherapy for tenosynovial giant cell tumors, a disease defined by rearrangements in the CSF1R gene. Pexidartinib and other CSF1R inhibitors have minimal efficacy in cancers more broadly but continue to be explored in combination regimens20, 16.
Similarly, it may be that a subpopulation of AD could benefit highly from aducanumab or other anti-amyloid therapies, even though the broader population does not. Further study of the best responders versus patients with poor responses in the phase 3 trials may provide insight on this question.
Our goal in writing this perspective was not to opine on whether or not the FDA made an error in approving aducanumab. That debate is an important one but has been covered in detail elsewhere, and it remains uncontested that AD is still in need of a breakthrough. Our focus here is on what is ultimately needed to achieve that breakthrough. We draw on the experience of oncology because of the abundance of case studies of both success and failure. Based on the patterns seen in those case studies, our proposal is to view the beta-amyloid pathway as analogous to “passenger” mutations in cancer.
There are those who would continue to hypothesize that beta-amyloid is the “driver” of AD and would suggest other reasons for the limited efficacy of amyloid-targeted therapies, including the degree of brain penetration, the presence of resistance pathways, and the need to continue to fine-tune which specific amyloid species are targeted and how7. There may be some truth to these arguments, though others have critiqued them5. Rather than reiterate what has already been written, we would simply note that this pattern—of a target that has many drug failures but is finally unmasked as a driver when strong monotherapy efficacy is achieved by just the right drug with just the right properties—has to our knowledge never been observed in oncology. Occasionally there is a driver that has a few failed attempts, e.g., with compounds that have poor pharmacology or off-target toxicity, but it is more typical to see meaningful efficacy with even a rough, first-generation driver-targeted therapy. Subsequent generations of drugs then improve on the benefit. Thus, our pattern-fit exercise led us to the passenger hypothesis.
If amyloid is just a passenger, the oncology experience tells us that the path forward must include studying other pathways, testing combination therapies, and putting precision medicine at the center of AD drug development. Of course, we are not the first to recognize these research needs: globally hundreds of research projects on these topics are already underway in academia—funded by the NIH, Alzheimer’s Association, Alzheimer’s Research UK, and others—and in industry. We applaud these efforts and consider them to be validation of our approach to apply learnings from case studies in a cross-disciplinary manner. Continuing to assess case studies and associated patterns in oncology may yield further insights—for example, insights from successes and failures in immuno-oncology may be relevant for assessing immunomodulatory approaches in AD and other conditions.
3 Washington Post, May 31, 2021
4 Kuller and Lopez 2021 Alzheimers Dement 17:692
5 Mullane and Williams 2020 Biochem Pharmacol 177:113945
6 Selkoe and Hardy 2016 EMBO Mol Med 8:595
7 Tolar et al. 2020 Alzheimers Dement 16:1553
9 Panza et al. 2014 Expert Opin Biol Ther 14:1465
10 Project GENIE
11 Mayer and Arteaga 2016 Annu Rev Med 67:11
12 Wright et al. 2021 Cancers 13:1538
14 Pon and Marra 2015 Annu Rev Pathol 10:25
15 Brufsky and Dickler 2018 Oncologist 23:528
17 Scarlett et al. 2016 Cancer Discov 6:956
19 NASEM 2016, Biomarker Tests for Molecularly Targeted Therapies