In Alzheimer’s research, finding a way to safely lower Aβ levels in the human brain remains a central goal. Thus, scientists were excited by a 2010 Nature paper in which researchers led by Nobel laureate Paul Greengard at Rockefeller University in New York City reported that the cancer drug imatinib, trade name Gleevec, suppressed Aβ42 production through a selective mechanism involving a protein they dubbed γ-secretase activating protein (GSAP) (see Sep 2010 news story). The finding suggested a way to target Aβ while sparing other substrates. Many labs began work in this area; however, in the last three years none have published replication of the key findings on GSAP, and some groups have since abandoned the work. A handful of papers presented mixed results that suggested the story may be more complicated than it first appeared.

Chemical Model of Imatinib

Now a December 10, 2013, Alzheimer’s and Dementia paper openly challenges the value of the original findings for human therapy development. Researchers led by Bob Olsson and senior author Kaj Blennow at the University of Gothenburg, Sweden, found no effect of imatinib on blood Aβ42 levels in human cancer patients or on Aβ production in several cell-culture models. When asked about the discrepancies, Greengard suggested that Olsson and colleagues may have been using different experimental conditions, and said he and his colleagues continue to investigate the therapeutic potential of GSAP and imatinib. 

Henrik Zetterberg at the University of Gothenburg, a co-author with Olsson and Blennow, wrote to Alzforum, “We believe there are many researchers who have tried different aspects of imatinib as an anti-amyloid agent but failed to get supportive data. Those negative results are difficult to publish.” 

Failure to replicate major findings has plagued biomedical science, including Alzheimer’s disease research. Some research groups have reported that attempts to replicate published data fail more often than not (see Prinz et al., 2011). The journal Nature recently issued new guidelines to make methods more transparent in hopes of improving reproducibility (see May 2013 news story), and Nature Biotechnology called for more efforts to replicate published research (see its editorial). The Swedish researchers were unable to persuade Nature to publish their data contradicting the original paper.

Does Cancer Drug Show Promise for Alzheimer’s?

Gamma-secretase has long represented a central target for AD therapy development because this enzyme finishes the sequential proteolysis of amyloid precursor protein (APP) to release the Aβ peptide. However, γ-secretase cuts many substrates, and side effects of direct inhibition have scuttled clinical trials (see Aug 2010 news storyApr 2011 conference storyAug 2011 conference story). Greengard and first author Bill Netzer previously reported that imatinib inhibited γ-secretase cleavage of APP without affecting other critical substrates such as Notch (see Oct 2003 news story). Exploring the mechanism, the authors reported that imatinib bound the 16 kDa protein GSAP, which is itself the cleavage product of a larger protein. In in-vitro and cellular model systems, GSAP interacted with a subunit of γ-secretase and the β-C-terminal fragment (β-CTF) of APP to facilitate cleavage of that fragment, the authors reported in 2010. In addition, knockdown of GSAP was reported to lower Aβ levels and plaque load in a mouse model of AD, suggesting this mechanism could be targeted therapeutically in people.

Olsson and colleagues wanted to find out if the imatinib results would translate to humans. Besides validating mouse findings on a potential AD treatment strategy, this would confirm the importance of GSAP as a target. To do this, they examined plasma samples from 51 patients who took imatinib to treat chronic myeloid leukemia. If the drug worked as published, the scientists reasoned, Aβ levels in the blood should drop. However, they saw no decrease over 12 months of treatment. To more directly test imatinib’s effects, Olsson and colleagues then applied the drug to three types of cell culture: human embryonic kidney cells that overexpress APP, induced cortical neurons made from people with Down’s syndrome who produce excess Aβ due to having three copies of APP, and mouse primary neurons. In every case, imatinib had no effect on Aβ, even at doses up to 10 μM. The results suggest that imatinib does not reliably lower Aβ in people, casting doubt on the usefulness of pursuing this pathway therapeutically, Olsson said. 

Other researchers lauded the work. “This is a cleverly designed human study. The data look very convincing,” Todd Golde at the University of Florida, Gainesville, wrote to Alzforum. He noted that kinase inhibitors such as imatinib can have distinct effects in different cell types. “Under some circumstances, imatinib may have some modulatory effects on Aβ. However, that is not going to be sufficient to warrant clinical testing,” he suggested. 

The study adds to an already contradictory literature on imatinib. Some papers have reported that imatinib treatment can lower Aβ, but some details conflict with Netzer and Greengard’s findings. For example, researchers led by Ellen Kilger at the University of Tübingen, Germany, saw a drop in secreted Aβ after administering imatinib to neuroglioma cells, but they traced the mechanism to enhanced degradation of Aβ, rather than inhibition of γ-secretase (see Eisele et al., 2007). A German group using 10 μM imatinib as a control in a study on a different topic recently reported that the drug lowered Aβ in cultured induced human neurons (see Mertens et al., 2013). Researchers led by Dirk Beher at Asceneuron SA, an Alzheimer’s-oriented spinoff of Merck Serono in Geneva, reported that imatinib lowered Aβ levels in cell culture, but did not affect plasma Aβ when administered to rats (see Hussain et al., 2013). “The new paper nicely extends our imatinib findings [from rat to human]. Together these studies put some question marks on the mechanism of action of imatinib, and whether associated targets are worth pursuing for Alzheimer’s disease,” Beher told Alzforum. 

Netzer said that differences in experimental technique could explain the discrepancies between his results and those of Olsson et al. He noted that in Olsson’s study the plasma concentrations of imatinib in the cancer patients averaged 1.5 μM, far lower than the 5 μM needed to see an effect on Aβ in his cell culture studies. Imatinib is not very potent at lowering Aβ, Netzer noted. Drugs that are developed for chronic use generally act in the nanomolar range. In addition, imatinib poorly penetrates the brain, so if plasma Aβ originates from the brain, as some studies suggest (see Demattos et al., 2002), peripheral administration of the drug would not be expected to affect levels, Netzer told Alzforum. “In our hands, imatinib is so reliable that we use it as a positive control for lowering Aβ,” he said. He said he has not tested the drug in the cell cultures used by Olsson and colleagues, but has seen it lower Aβ in rat primary neurons and several cell lines. Netzer thinks that compounds related to imatinib, but more potent and brain-penetrant, will be promising therapeutics for AD. 

Some researchers interviewed for this article pointed to a 2005 paper from Michael Wolfe and colleagues at Brigham and Women’s Hospital, Boston, for a possible explanation of the discrepancies. Wolfe and colleagues analyzed imatinib preparations by high-performance liquid chromatography, and traced the γ-secretase inhibitory activity Netzer and Greengard had reported in 2003 to a contaminant present in some preparations (see Fraering et al., 2005). This contaminant might represent a breakdown product of imatinib, Wolfe speculated, but he added that since 2005 he has been unable to identify the active compound. 

GSAP’s Mysteries Deepen

Reports on GSAP similarly conflict. While some researchers have found an effect on Aβ, no one has confirmed cleavage of GSAP or replicated a direct interaction between it and the other players: APP, γ-secretase, and imatinib. Beher and colleagues found that knockdown of GSAP in cell culture lowered Aβ, but overexpression had no effect. Beher saw no interaction between GSAP and the C-terminal fragment of APP in co-immunoprecipitation studies, nor any effect of GSAP on Aβ in in-vitro γ-secretase assays. “When we looked at direct effects of GSAP on γ-secretase, everything was negative,” Beher told Alzforum. He suggested that GSAP may affect Aβ production or secretion through some indirect mechanism. He is not pursuing the research further. 

Likewise, researchers led by Chuck Sanders at Vanderbilt University, Nashville, Tennessee, expressed recombinant GSAP in bacteria, and found that the purified protein did not bind either imatinib or APP β-CTF (see Deatherage et al., 2012). “It’s not the definitive word on GSAP, but it was negative enough that we think the story must be more complicated than we thought originally,” Sanders told Alzforum. He has also stopped work on the protein. He noted that although the published papers show varied data, “In all cases, the results are not what you would have expected based on the original work.” 

In addition to Hussain et al. and Deatherage et al., Lawrence Rajendran at the University of Zurich, whose lab recently published RNAi-based screens on APP processing (see Udayar et al., 2013Bali et al., 2012), told Alzforum that he has been unable to replicate the key findings but declined to comment on the details as his paper is still under consideration at Nature. Rajendran believes there may be an alternate explanation for GSAP’s effect on Aβ. 

Other researchers declined to be identified but privately told Alzforum they could not reproduce the results and had abandoned this line of research. In total, eight γ-secretase research groups confirmed that they were unable to replicate aspects of the data. Olsson said he tried to collaborate with other labs, but all the groups he contacted told him they had canceled research in this area. Golde wrote to Alzforum, “As far as I know, there has been no high-quality independent replication of the GSAP data.”

Greengard suggested that the problem may arise from the chemical characteristics of GSAP. The protein is extremely hydrophobic and aggregates easily, he told Alzforum. In his own lab, researchers see inconsistent results in in-vitro assays based on whether GSAP has aggregated or not, whereas knockdown of the protein in cells always lowers Aβ, he said. He considers the key findings valid, and is currently looking for other proteins that bind GSAP and might make attractive therapeutic targets.

The Challenge of Reproducibility

How to make sense of the conflicting reports? Gerhard Multhaup at McGill University, Montreal, noted that small differences in experimental techniques can greatly affect results. The current papers compare apples to oranges, he told Alzforum. In order to refute a study, “You have to try to reproduce the data by using the exact same methods,” he said. For GSAP and imatinib, this has not yet been done, he added. 

Pharmaceutical company researchers Alzforum spoke with about this issue tended to disagree that a failure to replicate is valid only if the exact procedures were repeated with the exact same models and reagents. They maintained that in order for a finding to be of practical use in a therapy development setting, it has to be robust enough that it can be replicated in several different models and with slight variations in technique and reagents.

Given the difficulties with reproducing GSAP and imatinib data, some researchers wondered why Nature has not published a Brief Communication Arising on the original paper, a preferred format for follow-up data. When asked that, Olsson told Alzforum he indeed had submitted his paper first to Nature as a BCA, but that the journal rejected it after soliciting comments from Greengard and two independent reviewers. According to Blennow, the original authors argued in their review that the Swedish paper must be wrong because it was unable to replicate the original cell and animal results in patients. "We see it the other way round," Blennow said. The AD field abounds with examples of therapies that looked promising in mice but failed in clinical trials.

A spokeswoman from Nature declined to discuss how this paper was handled, citing journal policy of not commenting on individual manuscripts. She noted that Nature’s policy is to send BCAs first to the lab that produced the original paper. The BCA and the original authors’ response are then sent to independent referees. “The decision to publish a BCA is based solely on whether the independent referees believe that the BCA presents a conclusive challenge to an aspect of the original paper,” the spokeswoman wrote to Alzforum. She added that the process can be customized and decisions are evaluated on a case-by-case basis. The same editor who handled the original paper generally handles the BCA.

Scientists frequently bellyache about the peer review process, but usually in private. It is rare that complaints around publication of follow-up data or replication attempts spill into the public. In 2010, this happened for a different Aβ paper, when a follow-up study prompted a comment by Adriano Aguzzi

“It is crucial for the sake of science that deviating opinions are heard, and preferably at the same level as the original research,” Bart De Strooper at VIB and KU Leuven, Belgium, wrote to Alzforum. He serves on the editorial board of EMBO Molecular Medicine and eLife. “We should, as a community, accept debate, and discussion should prevail over dogma. The journal Science did a wonderful job when publishing the opposing views on the bexarotene paper,” he added (see May 2013 news story). He believes the GSAP story should be handled similarly. “Nobody has been able to confirm the GSAP data. Had negative data been published earlier, it would have saved a lot of time, money, and expectations,” he said.—Madolyn Bowman Rogers


Make a Comment

To make a comment you must login or register.

Comments on this content

  1. We welcome a discussion on the problem with irreproducibility of published research that we recently encountered, as outlined in the Alzforum news story. In this context, the recent editorial guidelines from Nature (1) might represent one step to reduce this problem. A checklist for methodological and statistical information, together with unlimited space for methods, statistical description, and access to raw data, are commendable. In our mind, however, at least two problems remain. One is related to the difference in the requirement of independent replication between clinical and preclinical research, another to the technical requirements on the methods used for quantifying biochemical changes.

    Due to the issue with poor replication of results, Nature Genetics stipulated that novel genetic associations must be verified in at least two independent cohorts (2). Correspondingly, replication in independent cohorts is a core feature of any high-quality publication in clinical biomarker research (3). In contrast, in high-impact basic research journals, publication is common with very small numbers of inbred animals or cell experiments, without replication in independent sets. One example is the imatinib/GSAP paper (4), in which many critical conclusions were based on n=3 experiments, a number that actually is too low to do statistical calculations. We appreciate that the variability in animal models may be lower than in clinical samples, but at the same time these models may be highly skewed and not representative of the clinical disease they are supposed to reflect (5). To reduce the problem with irreproducibility, any finding should be replicated in two independent disease models, or at least two independent experiments with a number of cell cultures or animals that allows proper statistical testing.

    Another bias is related to the methodology used for quantification. Clinical studies often include validated methods and provide data on assay performance, e.g., between-run variability. In contrast, it is common in basic research to use non-validated semi-quantitative methods, e.g., Western blot, without presenting any information on method precision, variability, or specificity. For this, again, the imatinib/GSAP paper can serve as an example (4). In our view, such low-quality quantitative data may underlie irreproducibility, yet high-impact journals are full of them.

    We believe that addressing these two problems, in addition to the actions presented in the recent editorial, will solve a large part of the problem with irreproducible data from preclinical research.


    Announcement: Reducing our irreproducibility. Nature 496, 398 (2013).

    Framework for a fully powered risk engine. Nat Genet. 2005 Nov;37(11):1153. PubMed.

    . Towards complete and accurate reporting of studies of diagnostic accuracy: the STARD initiative. Fam Pract. 2004 Feb;21(1):4-10. PubMed.

    . Gamma-secretase activating protein is a therapeutic target for Alzheimer's disease. Nature. 2010 Sep 2;467(7311):95-8. PubMed.

    . Drug development: Raise standards for preclinical cancer research. Nature. 2012 Mar 28;483(7391):531-3. PubMed.

  2. I have not tried to replicate He et al. 2010 myself, so I can only comment based on my readings about this topic. I am particularly interested in the regulation of γ-secretase, and carefully followed the research about GSAP. Two papers report a lack of interaction between GSAP and APP-CTFβ, also called C99 (Hussain et al., 2013, and Deatherage et al., 2012). Moreover, Hussain et al. report that GSAP has no effect on γ-secretase mediated ε cleavages of both APP and Notch (Fig. 5 and 6). Finally, Hussain et al. report that imatinib does not lower Aβ in rats in vivo.

    However, while we were particularly interested in the findings by He et al., we mentioned in a recent review (Barthet et al., 2012) some discrepancies found in the paper. Most notably, the authors reported that GSAP stimulates production of Aβ while it reduces AICD, the product of γ-secretase cleavage of APP at the ε site, a phenomenon that is difficult to conceive.

    Finally, the paper by Olsson et al. in Alzheimer’s & Dementia reports that imatinib does not reduce Aβ in humans. All together, three independent groups report evidence that raises doubts about the effects of imatinib and GSAP on γ-secretase.


    . The Role of γ-Secretase Activating Protein (GSAP) and Imatinib in the Regulation of γ-Secretase Activity and Amyloid-β Generation. J Biol Chem. 2013 Jan 25;288(4):2521-31. PubMed.

    . Purification and characterization of the human γ-secretase activating protein. Biochemistry. 2012 Jun 26;51(25):5153-9. PubMed.

    . Cellular mechanisms of γ-secretase substrate selection, processing and toxicity. Prog Neurobiol. 2012 Aug;98(2):166-75. PubMed.

    . Imatinib treatment and Aβ42 in humans. Alzheimers Dement. 2013 Dec 10; PubMed.

  3. I did not try to reproduce He et al. 2010 myself but agree with the shortcomings of this data as reported in this story. If patient data suggest there will be no efficacy, I question whether it is worth resolving the remaining data discrepancies at this point.

  4. Unfortunately, many studies in the AD field have failed to be replicated and it is then difficult to get these data published. Journals tend to be reluctant to publish out of concern that failed replication studies might tarnish their image. I have been moderately successful in publishing failed replications, but it takes an enormous effort. Without such publication, however, the original paper continues to get cited, often with a reference to the effect of “but these findings are somewhat controversial."
    We should also acknowledge a real fear of retaliation in our community. Given very competitive NIH funding and promotions on the line, not everyone wants to antagonize a colleague who might be a future reviewer, even if it is the right thing to do.

  5. The irreproducibility of the literature data also impacts the speed and cost of discovering new therapies. The commitment to start an industry drug-discovery program is an expensive proposition, and risky if based only on the scientific literature because of the knowledge gaps created by the bias against publishing negative and contradictory findings. The biology is rarely as straightforward as presented in the first few publications on a new target. Thus, companies have to conduct internal research to fill in the knowledge gaps, and for competitive reasons, that knowledge is not made public immediately.

    The shrinking of industry drug-discovery groups results in fewer scientists working on target identification and a movement toward well-documented targets, which in turn results in homogenization of industry pipelines. Innovation and the discovery of game-changing therapies slows.

    Journals could help to address this problem by acknowledging the value of work that both replicates the data of others as well as work that may bring into question conclusions of previous papers. With the societal costs and human suffering that results from Alzheimer's disease, this field in particular needs to acknowledge the issue of publication biases.

  6. Robust enough for drug discovery?

    From an industry perspective, we follow the emergence of new biology with the hope that the new findings point to a potential new target or pathway for us to work toward producing candidate drugs. It is generally recognized, however, that a new finding regularly needs to be investigated further before we are ready to put the machinery in motion to design new drugs to a new target.

    In deciding whether a new target is ready, industry highly values replication of the original finding. But absolute replication can be challenging, and Madolyn Bowman Rogers’ article outlines just a few of the reasons why replication may not give the same results. I support the growing culture of sharing detailed experimental protocols, however, even the most detailed and accurate protocol may not work the first time, or initially give you the same results in a new setting. In industry, transferring new techniques or protocols to the labs of colleagues, or collaborators having researchers discuss the methods, and visit each other including doing the work together, is commonplace.

    In deciding if a target is robust enough, we also look to generate additional experiments that scope around the original finding and generate broader, hopefully supportive knowledge. In this stage we may find the original finding does not translate well. For example, some of us have been unlucky enough to have worked on targets that were subsequently found not to be expressed in the relevant tissues in humans, or have a markedly different pharmacological profile across species. We spend time on these and other such questions, and it is a platform of evidence that suggests whether a target is robust enough. If you look at new therapies that have come to market, you will find that they often have decades of biology research behind them—the hedgehog molecular signaling pathway and Genentech’s new skin cancer drug is one recent such story (see Bazurto article).

    Ultimately whether a new target does in disease what we think it does comes with human clinical trials. But here on the first clinical test we need to be aware that there are two questions remaining—does this target operate in humans the way it has in preclinical experiments, and does the molecule that we are using achieve the target engagement needed in order to see the expected effect? In the discussion on GSAP, the field has had the boon of a clinical-ready compound, enabling us to skip the process of designing a molecule so we can interrogate the human question directly. But clinical experiments, just like preclinical experiments, are designed to ask a question and the conditions of the experiment need to support asking that question.

    When thinking of the Olsson et al. imatinib study, my first question is whether there was enough target engagement: In cellular experiments with imatinib the potency has been reported around the 5 uM range, while the human studies had mean plasma levels of 1.46 uM. Even if translation for this mechanism is linear across systems, the clinical exposure is a total concentration and lower. Translation of potency across species and systems is rarely linear, and so a deep understanding of the pharmacology and pharmacokinetics is needed to underpin whether a drug is in sufficient exposure to have an effect, or to decide if the mechanism doesn’t translate. Indeed, in a review of 44 programs to reach Phase 2, Pfizer scientists determined that demonstrating target engagement (showing that the drug occupies the desired target sufficiently to elicit its pharmacological effect) is crucial for success (Morgan et al., Drug Discovery Today 2011). Time and dose response studies in vivo that measure effects in the same (plasma) compartment are needed.

    As a field searching for potential therapies for Alzheimer’s disease, we need the biology discoveries to continue. We need to expand our knowledge around any new target. We need to openly discuss and share findings. Comments to this article and in wider groups highlight that the current culture of publication is perhaps not best supporting these aims—aims which I am certain each of these labs share.

    Many years back AlzForum pioneered the Alzgene database, which addressed issues with replication or strength of linkage for the genetics of Alzheimer’s disease. Perhaps a similar representation of targets and the associated biology findings would help display the biology associated to potential targets.

  7. Co-authored by William Netzer and Paul Greengard, The Rockefeller University; Yueming Li and Darren Veach, Memorial Sloan-Kettering Cancer Center; Wenjie Luo, Weill Cornell Medical College; Fred Gorelick, Yale University School of Medicine; and Dongming Cai and Sam Gandy, Icahn School of Medicine at Mount Sinai, New York.

    Madolyn Rogers’ news article, and some of the comments in response to that article, draw attention to interlaboratory variability in the replication of two studies published by our research group.  In our original paper on this topic (Netzer et al., 2003), we reported that Gleevec (imatinib) reduced Aβ generation in cell culture and in vivo in guinea pig brain.  In the second paper (He et al., 2010), we described the identification of γ–secretase-activating protein (GSAP), an endogenous regulator of Aβ formation.  Here we focus primarily on studies published subsequent to Netzer et al. that have evaluated the Aβ-lowering effects of imatinib.

    Studies of imatinib in intact cell systems

    Netzer et al. reported that imatinib lowered Aβ40 levels in rat primary neurons and lowered Aβ40 and 42 in N2a (mouse neuroblastoma) cells stably transfected with APP 695 or PS1ΔE9/APP Swe.  Subsequent to that publication, we tested imatinib in Chinese hamster ovary cells transfected with wild-type human APP695 (CHO-hAPP), in 3T3 fibroblasts transfected with wild-type human APP695, and in mouse embryonic fibroblasts (MEFs) transfected with Swedish mutant human APP695.  In each of these three cell lines, imatinib reduced total Aβ by 40 to 80 percent (Netzer, unpublished results).  The observation that imatinib reduces Aβ generation in five different cell types—namely rat primary neurons, N2a, CHO, 3T3, and MEF cells—does not exclude the possibility that some cell types or cell lines might be resistant to imatinib-induced reduction in Aβ generation (see below).

    Independent confirmation of the imatinib effect in the cell types used by Netzer et al. (2003) 

    Hussain et al. confirmed in N2a cells our published observations (in that same cell line) that imatinib causes a 50 to 60 percent reduction in Aβ40 and 42 secretion.

    Arslanova et al. tested 100 μM imatinib in CHO-hAPP cells in culture and demonstrated that secretion of Aβ40 and 42 was reduced by 30 percent and 80 percent, respectively.  The published results of Arslanova et al.  were confirmed in the unpublished studies of Netzer mentioned above.

    Extension of the studies of Netzer et al. (2003) to three additional cell types

    Eisele et al. extended the list of imatinib-responsive cells to include H4-APPwt cells in their demonstration of an effect of imatinib on Aβ generation over a drug concentration range of 3.7 to 20μM.  A 50 percent reduction in secretion of Aβ40 and 42 was observed in the presence of 10μM imatinib. 

    Hussain et al. tested imatinib in SHSY5Y-SPA4CT cells and found that 10μM imatinib reduced secretion of Αβ40 and 42 by about 40 percent, adding yet another cell line to the roster of those that display imatinib-responsive Αβ generation.

    Mertens et al. reported that 10μΜ imatinib lowered Aβ40 and 42 secretion by 70 percent and 75 percent, respectively, in human differentiated neurons in culture, consistent with our observations using rat primary neuronal cultures. 

    Failure to replicate the imatinib effect in three previously untested cell types

    Olsson et al. reported the only published failure to reproduce the imatinib effect in intact cells.  They reported that imatinib does not lower Aβ42 secretion from either: (i) HEK293 cells (stably transfected with human APP Swe); (ii) human Down’s syndrome embryonic stem cell-derived cortical projection neurons; or (iii) mouse primary cortical neurons.  The protocol published by Olsson et al. provided no obvious explanation for why their results differed from those derived from multiple other cell types in the hands of multiple other investigators.  As of this writing, we are unaware of any other laboratory that has tested any of the cell types used by Olsson et al.  At present, we cannot determine whether the Olsson results are technical in nature or whether all the cell types they studied are imatinib-resistant.

    In summary, four of the five studies published since Netzer et al. have confirmed the Aβ-lowering effect of imatinib in cell culture.  We have been unable to identify any obvious explanation for this disparity in careful scrutiny of the published protocols from those failing to replicate the imatinib effect.  In an effort to resolve the issue as efficiently as possible, we would welcome visits to our laboratory by any scientists seeking a detailed demonstration of our methods of measuring the effect of imatinib on Aβ secretion by cultured cells.

    Demonstration of the imatinib effect in a cell-free system

    In addition to the results obtained in intact cell systems, a study by Fraering et al. reported that 75μM imatinib reduced Aβ40 and 42 formation by 50 percent in a cell-free γ–secretase assay.

    Studies of imatinib in intact rodents and in humans

    Netzer et al. reported that continuous intrathecal infusion of imatinib by osmotic pump reduced Aβ levels in the intact guinea pig brain. Intrathecal infusion was chosen as the route of administration because imatinib is rapidly pumped out of the brain by the p-glycoprotein pump (PGP) at the blood-brain-barrier (see Dai et al.).  As a result of robust PGP action, imatinib that is peripherally injected or ingested will appear in brain only to a negligible extent (Leis et al.).  For this reason, Netzer et al. explicitly stated that imatinib administered peripherally would not be useful as a therapeutic for Alzheimer’s disease (AD).  Nevertheless, both Hussain et al. (i.p. imatinib in rats) and Olsson et al. (oral imatinib in humans) studied the effects of peripherally injected or orally ingested imatinib.  Since most plasma Αβ is derived from the CNS (DeMattos et al.; Ghersi-Egea et al.; Shibata et al.), the failures of Hussain et al. and Olsson et al. to observe Αβ–reducing effects of peripherally injected or orally ingested imatinib on brain or plasma Aβ levels were the predicted results.  Therefore, the data of Hussain et al. and Olsson et al. are irrelevant to any assessment of the potential for therapeutic success using a CNS imatinib-like compound or a GSAP-modulating compound.

    Since we remain interested in developing an AD therapeutic based on imatinib, the Olsson et al. study in humans merits particular attention.  These investigators measured plasma Aβ42 levels in chronic myelogenous leukemia patients (n = 51) before and during treatment with imatinib.  They estimated that plasma concentrations of imatinib in these patients was 1.46 μM (the concentration of imatinib in the brain would have been much lower).  We have only observed reductions in Aβ secretion by imatinib for imatinib concentrations of 5μΜ or greater (Netzer, unpublished results).  Thus, the dose employed by Olsson et al. would not be expected to have an impact on peripheral generation of Aβ (if any such Aβ generation exists).   

    Studies of GSAP

    With respect to GSAP, four members of our laboratory and one publication from another laboratory (Hussain et al.) have confirmed the observation reported in He et al. that GSAP knockdown reduces Aβ secretion by cells in culture.  There have been no contradictory publications reporting a failure of GSAP knockdown to lower Aβ secretion in intact cells.  Two additional groups reported higher expression levels of GSAP in human tissues (including AD brain) that correlated with increased expression of γ-secretase components and increased Aβ deposition (Satoh et al.; Nogalska et al.).  We continue to actively investigate the molecular mechanism(s) by which GSAP modulates Aβ accumulation.  

    Two research groups (Hussain et al. and Deatherage et al.) have reported failure to reproduce some of the results reported by He et al. We are investigating the basis for these discrepancies.


    . Gleevec inhibits beta-amyloid production but not Notch cleavage. Proc Natl Acad Sci U S A. 2003 Oct 14;100(21):12444-9. PubMed.

    . Gamma-secretase activating protein is a therapeutic target for Alzheimer's disease. Nature. 2010 Sep 2;467(7311):95-8. PubMed.

    . The Role of γ-Secretase Activating Protein (GSAP) and Imatinib in the Regulation of γ-Secretase Activity and Amyloid-β Generation. J Biol Chem. 2013 Jan 25;288(4):2521-31. PubMed.

    . Phenotypic analysis of images of zebrafish treated with Alzheimer's gamma-secretase inhibitors. BMC Biotechnol. 2010;10:24. PubMed.

    . Gleevec increases levels of the amyloid precursor protein intracellular domain and of the amyloid-beta degrading enzyme neprilysin. Mol Biol Cell. 2007 Sep;18(9):3591-600. PubMed.

    . APP Processing in Human Pluripotent Stem Cell-Derived Neurons Is Resistant to NSAID-Based γ-Secretase Modulation. Stem Cell Reports. 2013;1(6):491-8. Epub 2013 Dec 5 PubMed.

    . Imatinib treatment and Aβ42 in humans. Alzheimers Dement. 2013 Dec 10; PubMed.

    . gamma-Secretase substrate selectivity can be modulated directly via interaction with a nucleotide-binding site. J Biol Chem. 2005 Dec 23;280(51):41987-96. PubMed.

    . Distribution of STI-571 to the brain is limited by P-glycoprotein-mediated efflux. J Pharmacol Exp Ther. 2003 Mar;304(3):1085-92. PubMed.

    . Central nervous system failure in patients with chronic myelogenous leukemia lymphoid blast crisis and Philadelphia chromosome positive acute lymphoblastic leukemia treated with imatinib (STI-571). Leuk Lymphoma. 2004 Apr;45(4):695-8. PubMed.

    . Plaque-associated disruption of CSF and plasma amyloid-beta (Abeta) equilibrium in a mouse model of Alzheimer's disease. J Neurochem. 2002 Apr;81(2):229-36. PubMed.

    . Fate of cerebrospinal fluid-borne amyloid beta-peptide: rapid clearance into blood and appreciable accumulation by cerebral arteries. J Neurochem. 1996 Aug;67(2):880-3. PubMed.

    . Clearance of Alzheimer's amyloid-ss(1-40) peptide from brain by LDL receptor-related protein-1 at the blood-brain barrier. J Clin Invest. 2000 Dec;106(12):1489-99. PubMed.

    . Immunohistochemical characterization of gamma-secretase activating protein expression in Alzheimer's disease brains. Neuropathol Appl Neurobiol. 2011 Jun 30; PubMed.

    . Activation of the γ-secretase complex and presence of γ-secretase-activating protein may contribute to Aβ42 production in sporadic inclusion-body myositis muscle fibers. Neurobiol Dis. 2012 Oct;48(1):141-9. PubMed.

    . Purification and characterization of the human γ-secretase activating protein. Biochemistry. 2012 Jun 26;51(25):5153-9. PubMed.

  8. We read with interest the response by Netzer, Greengard, and colleagues. It is clear that we have different views not only on the biological effect of GSAP but also on Αβ (patho)physiology in general.

    There are some strong claims in the response, e.g., "that most plasma Αβ is derived from the CNS." Netzer and Greengard cite some papers, which they claim support this statement. One of these (Ghersi-Egea et al., 1996) showed that a large portion of Αβ40 injected into lateral ventricle CSF of normal rats ends up in blood, while another (Shibata et al., 2000) showed that Αβ40 injected into brains of mice is cleared from it by the LRP-1 receptor. Although interesting in themselves, these articles do not tell us whether or not most plasma Αβ is derived from the CNS.

    There are many additional papers, not cited by Netzer and Greengard, that are highly relevant for the question of whether plasma Αβ is CNS-derived. Walker and co-workers showed that intravenous administration of a fusion protein of the Aβ-degrading enzyme neprilysin (NEP), which resembles the orientation of the native NEP in membranes, in mice resulted in a fast and very marked dose-dependent reduction in plasma Aβ without affecting either soluble or formic acid-extractable brain Aβ levels, or CSF Aβ levels (Walker et al., 2013). These results also have been confirmed using NEP engineered to have an extended plasma half-life and an increased Aβ degradation activity. In long-term studies on mice, rats, and monkeys, a marked decrease of Aβ was found in plasma without any change in brain or CSF Aβ levels (Henderson et al., 2014). This paper was also commented on at the AlzForum website.

    Further, pharmacodynamic studies in man on BACE1 inhibitors also show that the decrease in plasma Aβ is much faster, and also more pronounced, than the decrease in CSF Aβ (May et al., 2011). Similar results were found in a Phase 2 trial on a gamma-secretase inhibitor (Fleisher et al., 2008). 

    Taken together, neither the NEP studies in animals nor the clinical trials in man support the existence of any robust peripheral Aβ efflux sink, and thus do not support "that most plasma Αβ is derived from the CNS." 

    Last, it well known that Aβ is generated in considerable amounts in many organs and cell types outside the CNS, such as platelets, the neuromuscular junctions of skeletal muscle, and arterial walls (Li et al., 1998; Kuo et al., 2000; Roher et al., 2009). Undoubtedly, these peripheral organs contribute to the pool of circulating Aβ in plasma.

    In summary, the literature suggests that there is a (large) pool of Αβ in the plasma that is not derived from the CNS, and that compounds that suppress amyloidogenic processing of APP can shrink that pool. We found no such effect in patients treated with imatinib. We are happy to see continued discussion on the important topic of validating data from cell and animal experiments by translational studies in humans. However, the statement by Netzer and Greengard that "the data of Hussain et al. and Olsson et al. are irrelevant to any assessment of the potential for therapeutic success using a CNS imatinib-like compound or a GSAP-modulating compound" seems to be based on a misconception due to selective referencing of available data. Instead, an unbiased literature review tells us that the forecast of imatinib-like or GSAP-modifying compounds as Aβ-lowering therapies in AD does not look bright.


    . Enhanced proteolytic clearance of plasma Aβ by peripherally administered neprilysin does not result in reduced levels of brain Aβ in mice. J Neurosci. 2013 Feb 6;33(6):2457-64. PubMed.

    . Sustained peripheral depletion of amyloid-β with a novel form of neprilysin does not affect central levels of amyloid-β. Brain. 2014 Feb;137(Pt 2):553-64. Epub 2013 Nov 20 PubMed.

    . Robust central reduction of amyloid-β in humans with an orally available, non-peptidic β-secretase inhibitor. J Neurosci. 2011 Nov 16;31(46):16507-16. PubMed.

    . Phase 2 safety trial targeting amyloid beta production with a gamma-secretase inhibitor in Alzheimer disease. Arch Neurol. 2008 Aug;65(8):1031-8. PubMed.

    . Elevated abeta42 in skeletal muscle of Alzheimer disease patients suggests peripheral alterations of AbetaPP metabolism. Am J Pathol. 2000 Mar;156(3):797-805. PubMed.

    . Secretion of Alzheimer's disease Abeta amyloid peptide by activated human platelets. Lab Invest. 1998 Apr;78(4):461-9. PubMed.

    . Amyloid beta peptides in human plasma and tissues and their significance for Alzheimer's disease. Alzheimers Dement. 2009 Jan;5(1):18-29. PubMed.

    . Fate of cerebrospinal fluid-borne amyloid beta-peptide: rapid clearance into blood and appreciable accumulation by cerebral arteries. J Neurochem. 1996 Aug;67(2):880-3. PubMed.


News Citations

  1. There’s a GSAP for That: Novel APP Partner a New Therapeutic Target?
  2. Guidelines at Nature Aim to Stem Tide of Irreproducibility
  3. Lilly Halts IDENTITY Trials as Patients Worsen on Secretase Inhibitor
  4. Barcelona: Live and Learn—γ-Secretase Inhibitors Fade, Modulators Rise
  5. Paris: Semagacestat Autopsy and Other News of Trial Tribulations
  6. Gleevec for Alzheimer's?
  7. Bexarotene Revisited: Improves Mouse Memory But No Effect on Plaques

Paper Citations

  1. . Believe it or not: how much can we rely on published data on potential drug targets?. Nat Rev Drug Discov. 2011 Sep;10(9):712. PubMed.
  2. . Gleevec increases levels of the amyloid precursor protein intracellular domain and of the amyloid-beta degrading enzyme neprilysin. Mol Biol Cell. 2007 Sep;18(9):3591-600. PubMed.
  3. . APP Processing in Human Pluripotent Stem Cell-Derived Neurons Is Resistant to NSAID-Based γ-Secretase Modulation. Stem Cell Reports. 2013;1(6):491-8. Epub 2013 Dec 5 PubMed.
  4. . The Role of γ-Secretase Activating Protein (GSAP) and Imatinib in the Regulation of γ-Secretase Activity and Amyloid-β Generation. J Biol Chem. 2013 Jan 25;288(4):2521-31. PubMed.
  5. . Brain to plasma amyloid-beta efflux: a measure of brain amyloid burden in a mouse model of Alzheimer's disease. Science. 2002 Mar 22;295(5563):2264-7. PubMed.
  6. . gamma-Secretase substrate selectivity can be modulated directly via interaction with a nucleotide-binding site. J Biol Chem. 2005 Dec 23;280(51):41987-96. PubMed.
  7. . Purification and characterization of the human γ-secretase activating protein. Biochemistry. 2012 Jun 26;51(25):5153-9. PubMed.
  8. . A paired RNAi and RabGAP overexpression screen identifies Rab11 as a regulator of β-amyloid production. Cell Rep. 2013 Dec 26;5(6):1536-51. PubMed.
  9. . Role of genes linked to sporadic Alzheimer's disease risk in the production of β-amyloid peptides. Proc Natl Acad Sci U S A. 2012 Sep 18;109(38):15307-11. PubMed.

External Citations

  1. editorial
  2. comment by Adriano Aguzzi

Further Reading

Primary Papers

  1. . Imatinib treatment and Aβ42 in humans. Alzheimers Dement. 2013 Dec 10; PubMed.