Papers in the March 17 EMBO Molecular Medicine and the April 20 Neuron have reignited a simmering debate about the role of presenilin mutations in familial Alzheimer’s disease. Both studies dispute a claim made previously by Jie Shen, Ray Kelleher, and colleagues that some familial AD PS mutations completely lack γ-secretase activity in vivo. This claim is at the heart of Shen and Kelleher’s presenilin hypothesis of Alzheimer’s disease, which challenges the more widely supported hypothesis that long Aβ species are the central factor in Alzheimer’s disease.

In the EMBO paper, researchers led by Harald Steiner at Ludwig-Maximilians-University, Munich, report that, when expressed in cells, one of these PS variants, L435F, makes Aβ43. This is an aggregating and neurotoxic peptide that only recently started drawing the attention of AD researchers (Saito et al., 2011). Steiner and colleagues also report that Aβ43 accumulates in amyloid plaques in people harboring the L435F presenilin-1 mutation.

In a “Matters Arising” paper in Neuron, researchers led by Bart De Strooper and Lucia Chávez-Gutiérrez at KU Leuven, Belgium, report that L435F—and two other PS mutations purportedly causing loss of function—in fact generate Aβ43 from amyloid precursor protein in vitro, as well.

“This was the missing piece of evidence from the debate, and it clearly supports the amyloid hypothesis,” said Steiner. Together, the papers explain how a small amount of aberrant activity retained by mutant presenilin causes dementia in people who harbor one good and one bad copy of the gene. The good copy, and presumably two copies of presenilin 2, ensure that γ-secretase adequately processes Notch and other substrates that are essential for normal development and biology, while the bad copy over time accrues enough Aβ43 to seed plaques. “Since it has been shown that γ-secretase endopeptidase activity is normal in PS FAD patients, they cannot have major [Notch] signaling defects. However, they have consistent changes in ratios of longer vs. shorter Aβ, and now we find that this happens in [L435F] patients as well,” noted Steiner.

Physiological Relevance.

Plaques containing Aβ43 abound in the brain of an L435F patient, but not in a patient with a different FAD mutation (bottom). [Image courtesy Harald Steiner, EMBO Molecular Medicine.]

Both Steiner’s and De Strooper’s groups used mouse embryonic fibroblasts (MEFs) to test PS variants for their carboxypeptidase activity. This is the processive chopping of substrate into Aβ peptides of decreasing length, not the endopeptidase activity this enzyme complex also performs. The scientists expressed the protease together with APP harboring the Swedish mutation. This variant accelerates BACE cleavage of APP, giving the researchers a better chance of detecting γ-secretase activity. Both groups used carboxypeptidase null mutations (missing catalytic aspartates) as negative controls, and both measured activity in the absence of wild-type PS1 or PS2 to avoid any ambiguity about what enzyme is doing the catalysis. This is because Shen and colleagues had previously reported a “trans-dominant” effect, whereby PS1 loss-of-function alleles interfere with wild-type PS1 activity in MEFs to modulate activity and generate peptides longer than Aβ40 (see Heilig et al., 2013). 

In Munich, joint first authors Benedikt Kretner and Johannes Trambauer found that L435F PS1 made about four times as much Aβ43 as Aβ40. In Leuven, first author Sarah Veugelen found that all three mutants tested (R278I, C410Y, and L435F) reconstitute the γ-secretase complex in PS1/2-negative MEFs, and that they made various forms of Aβ. All three generated very little Aβ40 compared to wild-type PS. R278I and C410Y made about 30 percent less Aβ42 than did wild-type, while L435F made about 90 percent less. However, all three mutants made much more Aβ43 than did normal presenilin. Both R278I and L435F made more Aβ43 than Aβ42.

Responding in the same issue of Neuron, Shen, from Brigham and Women’s Hospital, and Kelleher, from Massachusetts General Hospital, both in Boston, maintain presenilin loss of function is “the most coherent explanation for all existing, reproducible human and mouse data.” First author Dan Xia had reported last year that L435F and C410Y PS knock-ins make no Aβ40/42 in the brain, but he had not checked Aβ43 (see Mar 2015 news). Now he addresses this, reporting three-month-old L435F heterozygous knock-ins made about half as much Aβ43 as did wild-type mice, while finding no Aβ43 in homozygous C410Y embryos.

Xia and colleagues insist that different experimental procedures explain the discrepancies. “It is worth noting the potential for cell type-specific or technical differences in detection of Aβ production between immortalized HEK or MEF clones overexpressing mutant PS1 and APP in culture (Steiner and De Strooper’s work) and the adult brain expressing mutant PS1 at normal endogenous levels (our work),” Kelleher wrote to Alzforum.

De Strooper disagreed. “Our goal was to determine if these variants have any activity,” he told Alzforum. “In my opinion a cell-based assay is a perfectly good way to measure the activity of a membrane protease,” he said, while noting that Veugelen also used knock-in cell models to avoid overexpression. He pointed out that the detergents Xia and colleagues used to make brain extracts may have altered the intra-membrane protease in such a way as to alter its activity and producing artifactual results.

The debate does not hinge on cell data alone. In 2011, researchers led by Takaomi Saido at RIKEN Brain Science Institute, Wako, Japan, reported that the R278I mutation generates Aβ43 in a knock-in mouse model and that this highly amyloidogenic peptide generated extensive plaques (see Jul 2011 news). Last year, Chávez-Gutiérrez and colleagues reported that in brain tissue from 22 familial AD patients who carried various PS1 mutations, γ-secretase’s overall endopeptidase activity was similar between mutant and control tissue, a challenge to the loss-of-function hypothesis (Szaruga et a., 2015). Now, Steiner’s group caps it off by showing that two patients with confirmed L435F genotypes formed extensive Aβ43-containing plaques in the frontal cortex (see image above).

Where does all this leave the presenilin loss-of-function hypothesis? Kelleher insists none of this changes anything. “Our hypothesis does not rely on complete absence of any detectable Aβ production by mutant PS1. Rather, it proposes that loss of presenilin function plays an important role in FAD pathogenesis and provides a more coherent explanation for available data than Aβ oligomerization or deposition alone.”

“That’s a far cry from their earlier claims ‘that FAD mutations in the PSEN genes can cause complete loss of its function and γ-secretase activity in vivo,’” noted De Strooper. “In fact, the partial loss of function, and the possibility that this also contributes to the disease phenotype, has been discussed many times by us and others in the field,” he added. “The point is that the common denominator in Alzheimer’s disease is amyloid production, and loss of function on its own is not sufficient to cause that.”—Tom Fagan


  1. I would like to start my comment by stating the obvious—science is about finding the truth, and ultimately the truth prevails.

    Since our response paper published in Neuron has already thoroughly addressed the issues, including the technical limitations and internally inconsistent results, I would refer the interested readers to our original papers (see below for references).

    For those who would like to read a brief, relevant response from us, I include the last two paragraphs of our latest Neuron paper, where our views are not distorted.

    "Given the prevalence of AD and the lack of effective therapies, open-minded debates about the mechanisms of AD pathogenesis are needed and should lead to more productive therapeutic development. Unfortunately, the arguments presented by Veugelen et al. are undermined by numerous technical and conceptual limitations and inconsistencies, which raise significant uncertainties regarding the validity and relevance of their results. It is also regrettable that they present a distorted view of the conclusions advanced in our recent study. In particular, Veugelen et al. claim that we proposed that ‘Aβ is an epiphenomenon in FAD (Xia et al., 2015),’ even though the paper contains no such statement or conclusion. On the contrary, we performed extensive analysis focused on Aβ production and accumulation, and reported that heterozygosity for the FAD PSEN1 mutation increased the Aβ42/40 ratio and exacerbated Aβ deposition via a loss-of-function mechanism mediated by a greater reduction in steady-state levels of Aβ40 than Aβ42 (Xia et al., 2015). Consistent with this interpretation, the PSEN1 L435F mutation promotes Aβ deposition in the FAD brain in the form of cotton wool plaques, which display strong Aβ42 immunoreactivity but scant Aβ40 immunoreactivity (Heilig et al., 2010). Based on evidence that Aβ40 is protective against Aβ deposition (Kim et al., 2007; Wang et al., 2006), we suggest that preferential loss of Aβ40 can explain Aβ deposition in the brains of mouse models and human FAD patients without any need to invoke excessive production of longer Aβ species. While Aβ43 deposition may occur in the FAD brain via a similar loss-of-function mechanism, the assertion that Aβ43 is ‘highly amyloidogenic’ and ‘likely pathogenic’ warrants critical examination. Aβ43 differs from Aβ42 by an additional C-terminal threonine, a polar residue which should in principle reduce its hydrophobicity relative to Aβ42; indeed, recent biophysical analysis has shown that Aβ43 is less aggregation-prone and seeds amyloid formation less efficiently than Aβ42 (Chemuru et al., 2016). 

    “The fundamental difference between our views and those of Veugelen et al. is how PSEN mutations cause neurodegeneration and dementia in FAD. Veugelen et al. believe that ‘The hypothesis that β amyloid is triggering the neurodegeneration in FAD patients remains the most parsimonious and consistent explanation for all experimental data.’ While we acknowledge a significant role of Aβ in FAD, our genetic findings point to a more important causal role for loss of essential functions of Presenilin in the pathogenesis of neurodegeneration and dementia in FAD. First, Presenilin is essential for learning and memory, synaptic function, and age-dependent neuronal survival (Saura et al., 2004; Wines-Samuelson et al., 2010; Yu et al., 2001; Zhang et al., 2009; Zhang et al., 2010). Second, partial loss of Presenilin function is also sufficient to cause age-dependent neurodegeneration, though at a later age of onset and with lesser severity than observed with complete loss of Presenilin function (Watanabe et al., 2014). Third, FAD mutations in PSEN1 cause loss of γ-secretase activity, though the severity of impairment varies among mutations (Heilig et al., 2013Heilig et al., 2010; Saito et al., 2011; Xia et al., 2015). Lastly, our recent study showed that clinical PSEN1 mutations promote the development of key FAD-related phenotypes, including synaptic and cognitive impairment, amyloid deposition, and cerebral cortical neurodegeneration with accompanying inflammatory changes, through a loss-of-function mechanism (Xia et al., 2015). Thus, the loss-of-function mechanism that we have demonstrated for FAD-linked PSEN1 mutations provides the most coherent explanation for all existing, reproducible human and mouse data. Based on these insights, we have proposed that restoration of normal Presenilin function may offer the most direct approach to devise effective therapies that can combat dementia and neurodegeneration in FAD patients (Shen and Kelleher, 2007; Xia et al., 2015)."


    . Presenilin-1 knockin mice reveal loss-of-function mechanism for familial Alzheimer's disease. Neuron. 2015 Mar 4;85(5):967-81. PubMed.

    . Loss of Aβ43 Production Caused by Presenilin-1 Mutations in the Knockin Mouse Brain. Neuron. 2016 Apr 20;90(2):417-22. PubMed.

  2. The present dispute concerns discrepancies between our finding that FAD-linked Psen1 knock-in (KI) mutations cause a loss of Aβ43 production in the adult mouse brain when mutant PS1 is expressed at normal levels and with the normal spatiotemporal pattern, and claims from the De Strooper and Steiner groups that the same mutations may not fully inactivate Aβ43 production when mutant PS1 is overexpressed using viral transduction in immortalized MEF cell clones. The reader may judge which approach is more physiologically relevant to the pathogenesis of brain phenotypes such as dementia and neurodegeneration in FAD. As described below and in our recent Neuron article, the reports from the De Strooper and Steiner groups are fraught with potential for non-physiological effects, troubling inconsistencies and inadequate controls, and conclusions that are not supported by their data. And far from unexpected, Aβ43 deposition in the brain of FAD patients heterozygous for the PSEN1 L435F mutation is in fact predicted by our finding that heterozygosity for the identical KI mutation in mice promotes Aβ42 deposition in the adult brain via a loss-of-function mechanism (Xia et al., 2015; see below). 

    We proposed the Presenilin hypothesis nearly 10 years ago (Shen and Kelleher, 2007). At the time, γ-secretase inhibition was being vigorously pursued as a therapeutic strategy for AD in an effort to counter the presumed hyperactivity of γ-secretase and excessive production of Aβ envisioned by the then-current version of the amyloid hypothesis. By contrast, the Presenilin hypothesis posits that PSEN mutations cause FAD through a loss-of-function pathogenic mechanism. We showed in 2004 that modeling loss of Presenilin function by inactivating Presenilins in the adult brain caused an array of brain phenotypes resembling AD, including cognitive decline and age-dependent neurodegeneration accompanied by inflammatory changes (Saura et al., 2004). Several years later, clinical trials of γ-secretase inhibitors were halted due to worsening of cognition in treated AD patients, consistent with predictions of the Presenilin hypothesis. More recently, we found that clinical PSEN1 mutations can produce a spectrum of AD-related phenotypes in the mouse brain via a loss-of-function pathogenic mechanism, providing direct experimental support for the Presenilin hypothesis (Xia et al., 2015; Xia et al., 2016).

    Contrary to the distortions of our views offered by De Strooper and Steiner, the Presenilin hypothesis does not depend on complete inactivation of Presenilin or γ-secretase activity by pathogenic PSEN mutations, nor have we suggested that all PSEN mutations completely inactivate these functions. Indeed, we proposed the Presenilin hypothesis before we subsequently discovered that some clinical PSEN1 mutations cause virtually complete inactivation of γ-secretase. As we have proposed and demonstrated in our published studies over the past decade, PSEN mutations in FAD cause a general impairment of Presenilin and γ-secretase activity, with the severity of this impairment varying among mutations (Shen and Kelleher, 2007; Heilig et al., 2010; Heilig et al., 2013; Xia et al., 2015; Xia et al., 2016). 

    The L435F and C410Y mutations analyzed in our recent work represent particularly severe examples that are instructive because they enable a clear assessment of the impact of loss of Presenilin function caused by pathogenic mutations on brain phenotypes relevant to AD. Consistent with our finding that the L435F mutation inactivates PS1 function, the atomic structure of human γ-secretase localized the Leu435 residue in close proximity to the Asp catalytic dyad in PS1 (Bai et al., 2015), suggesting that substitution of a bulky aromatic residue would result in orthosteric inhibition of catalytic activity.

    The amyloid hypothesis has undergone a series of revisions as the tenets of previous iterations have later proven inaccurate, starting with the notion that FAD mutations increased production of all Aβ peptides, followed by the contention that FAD mutations invariably increased the production of Aβ42, which led to the development of γ-secretase inhibitors to block this presumed hyperactivity of γ-secretase. The long and well-documented history of development and pursuit of γ-secretase inhibitors effectively refutes the notion advanced by De Strooper that FAD PSEN mutations were known to cause a loss of function. Indeed, these agents continued in clinical trials even after we showed in 2004 that inactivation of γ-secretase in the mouse brain caused cognitive decline and neurodegeneration (Saura et al., 2004). This latest iteration shifts the focus to Aβ43, but the assertions that Aβ43 is pathogenic, or even that it initiates Aβ deposition, are lacking in experimental support. Notably, Aβ43 would be predicted to be less hydrophobic and aggregation-prone than Aβ42, and this prediction has been substantiated by biophysical analysis (Chemuru et al., 2016). A more important question to be addressed by future studies is the role of Aβ43, or other Aβ species, in the cognitive decline and neurodegeneration caused by PSEN1 mutations in FAD.

    Importantly, we do not discount a role for Aβ in FAD, but we have suggested based on a variety of lines of evidence that loss of Presenilin function is more proximate to the pathogenesis of dementia and neurodegeneration in FAD. This evidence includes the absence of neurodegeneration in human mutant APP transgenic mouse models despite accumulation of Aβ42 to >6,000-fold greater than normal levels (Irizarry et al., 1997; Irizarry et al., 1997; Mucke et al., 2000; Saura et al., 2004), and the earlier age of onset conferred by PSEN1 mutations relative to APP mutations in FAD (Ryman et al., 2014). Notably, our recent study demonstrating that the L435F and C410Y mutations inactivate PS1 function included an extensive analysis of Aβ production and deposition. This analysis revealed that heterozygosity for the L435F mutation promotes Aβ deposition in the brain through a loss-of-function mechanism mediated by relatively greater reduction in levels of Aβ40 than Aβ42. This finding is consistent with evidence that Aβ40 is protective against Aβ deposition (Wang et al., 2006), and implies that the Aβ42 deposited in amyloid plaques is the product of co-expressed wild-type PS1. Importantly, our results predict that heterozygosity for the L435F mutation would similarly promote Aβ43 deposition via the same loss-of-function mechanism that we identified for Aβ42.

    The communication from De Strooper and colleagues is rife with problems, including missing controls, internal inconsistencies, and glaring discrepancies with well-established published findings (Veugelen et al., 2016). These defects are too numerous to elaborate here, but a few illustrative examples will suffice. In “solubilized reconstituted cells,” they find that Aβ43 accounts for ~70 percent of Aβ production by wild-type PS1 (!)—this result is highly inconsistent with published data on the subject, making it impossible to draw any meaningful or reliable conclusions from this line of investigation. By contrast, our findings on the pattern and levels of Aβ peptides produced by wild-type PS1 are entirely consistent with the literature.

    Contradicting the claim that their data is in complete agreement with those of Saido et al. (2011), the results presented by De Strooper and colleagues display a number of concerning discrepancies (Veugelen et al., 2016). First, the high level of secreted Aβ43 detected in heterozygous R278I KI (“WT/KI”) MEFs is inconsistent with the undetectable level of secreted Aβ43 previously reported for this MEF genotype (Saito et al., 2011). Second, Saito et al. reported a strong R278I KI allele dosage-dependent reduction in de novo Aβ43 production, as shown by ~50 percent reduction in heterozygous MEFs and ~90 percent reduction in homozygous MEFs, using the same in vitro γ-secretase assay employed in our analysis (Fig. S10c, Saito et al., 2011). Third, Saito et al. reported that the R278I mutation severely impaired AICD and NICD production in both cell-based and cell-free assays (Fig. 1e and Fig. S10, Saito et al., 2011). These results indicate that endoproteolytic cleavage of APP and Notch at the ε site is severely impaired by the R278I mutation, consistent with our prior findings for the same mutation (Heilig et al., 2013; Heilig et al., 2010), and in direct conflict with De Strooper’s claim that this mutation does not alter the endopeptidase activity of PS1.

    The report from Steiner and colleagues has a curious backstory—we were contacted by one of the study’s senior authors last summer due to concerns that they had been unable despite multiple attempts to confirm the L435F mutation in the brain samples they were analyzing. Assuming that the samples are in fact correct, it is not surprising that the brains of individuals with this mutation would display Aβ43 deposition, since we have previously shown that these brains display abundant Aβ42 deposition. Our recent analysis of Psen1 KI mice revealed that heterozygosity for the L435F mutation promotes Aβ42 deposition in mice through a loss-of-function mechanism: Inactivation of PS1 bearing the L435F mutation causes an overall reduction in Aβ40 and Aβ42, but the relatively greater loss of Aβ40 results in a modest increase in the Aβ42/Aβ40 ratio, thereby promoting deposition of Aβ42 produced by wild-type PS1. In light of these findings, deposition of Aβ43 in the brains of FAD patients heterozygous for the L435F mutation is entirely predictable and likely to be explained by the same loss-of-function mechanism that we have demonstrated for Aβ42 deposition. 

    The comparisons of the activity of wild-type, D385A, and L435F PS1 conducted by Kretner et al. in HEK cells are flawed because it is unclear to what extent stable transduction with exogenous PS1 in different cell lines resulted in replacement of endogenous PS1/PS2 (Kretner et al., 2016). This is a key confounding issue when trying to draw conclusions regarding trace amounts of PS1 endoproteolysis or Aβ production. However, assuming that the majority of endogenous PS1 is replaced, it is difficult to reconcile the differences in Aβ production observed for L435F PS1 between HEK cells and PS1/2 KO MEFs. In addition, the manner in which data are presented—i.e., consistently overloading western blots and mass spectrometry analysis by five- to 10-fold for L435F PS1 relative to wild-type PS1, showing only relative and not absolute levels of Aβ43 produced in PS1/2 KO MEFs, etc.—obscures the fact that the levels of Aβ43 production detected from L435F PS1 are minuscule.

    Kretner et al. also make erroneous claims about our view of the role of trans-dominant negative effects of PSEN mutations vis-à-vis the Presenilin hypothesis. We have proposed and published evidence that PS1 bearing FAD mutations can exert a dominant-negative effect on wild-type PS1 (Shen and Kelleher, 2007; Kelleher and Shen, 2010; Heilig et al., 2013); despite their claims, the results presented by Kretner et al. do nothing to refute this evidence, and in fact, the experiments presented do not even address this issue. Instead, they investigate only the “cis” effect of PSEN1 mutations when mutant PS1 is expressed alone, and they do not investigate the “trans” effect of the mutation when mutant PS1 is co-expressed with wild-type PS1. Put simply, dominant-negative effects cannot be studied when the mutant protein is expressed in the absence of the wild-type protein. Thus, the key issue of the impact of PSEN1 mutations in heterozygosity, as is the case in FAD patients, is left unexamined by Kretner et al.

    De Strooper appears to argue that results obtained using viral overexpression in immortalized MEF cell clones is sufficient to draw reliable conclusions regarding pathogenic mechanisms in FAD. Although it should not be necessary to point this out, there are many examples of cell-type-specific differences between brain and other tissues, not to mention the well-established potential of overexpression systems to yield non-physiological results. If one were to submit a grant application or manuscript describing investigation of the mechanisms of FAD pathogenesis in the skin of mouse embryos, such a study would be roundly criticized as lacking in disease relevance, and the authors would be advised to investigate disease mechanisms in the appropriate physiological context, i.e., the brain. Conducting such investigations in MEFs is one step further removed from disease relevance, and it is therefore vital to validate any findings from such analysis in the brain. Our initial analysis of the impact of the L435F and C410Y mutations was conducted in PS1/2 KO MEFs, with the key difference that we developed a sensitive assay system relying on physiological expression levels of PS1. Nevertheless, we quickly moved on to generate novel lines of KI mice and perform in vivo analysis of these mutations in the brain. Our results suggest that the Aβ43 production detected in MEFs from the L435F and C410Y mutations by the Steiner and De Strooper groups is likely to be non-physiological, whether due to cell type-specific differences and/or overexpression artifacts, as we do not detect significant Aβ43 production by mutant PS1 bearing these mutations in the brains of homozygous KI mice.

    Considering the growing evidence in support of the Presenilin hypothesis, the sort of response exemplified by the De Strooper and Steiner reports and their accompanying comments is reminiscent of nothing so much as the following quote attributed to Gandhi:

    “First they ignore you, then they laugh at you, then they fight with you, then you win.”


    . An atomic structure of human γ-secretase. Nature. 2015 Sep 10;525(7568):212-7. Epub 2015 Aug 17 PubMed.

    . C-Terminal Threonine Reduces Aβ43 Amyloidogenicity Compared with Aβ42. J Mol Biol. 2015 Jun 26; PubMed.

    . Trans-dominant negative effects of pathogenic PSEN1 mutations on γ-secretase activity and Aβ production. J Neurosci. 2013 Jul 10;33(28):11606-17. PubMed.

    . A presenilin-1 mutation identified in familial Alzheimer disease with cotton wool plaques causes a nearly complete loss of gamma-secretase activity. J Biol Chem. 2010 Jul 16;285(29):22350-9. PubMed.

    . APPSw transgenic mice develop age-related A beta deposits and neuropil abnormalities, but no neuronal loss in CA1. J Neuropathol Exp Neurol. 1997 Sep;56(9):965-73. PubMed.

    . Abeta deposition is associated with neuropil changes, but not with overt neuronal loss in the human amyloid precursor protein V717F (PDAPP) transgenic mouse. J Neurosci. 1997 Sep 15;17(18):7053-9. PubMed.

    . Genetics. Gamma-secretase and human disease. Science. 2010 Nov 19;330(6007):1055-6. PubMed.

    . Generation and deposition of Aβ43 by the virtually inactive presenilin-1 L435F mutant contradicts the presenilin loss-of-function hypothesis of Alzheimer's disease. EMBO Mol Med. 2016 May 2;8(5):458-65. PubMed.

    . High-level neuronal expression of abeta 1-42 in wild-type human amyloid protein precursor transgenic mice: synaptotoxicity without plaque formation. J Neurosci. 2000 Jun 1;20(11):4050-8. PubMed.

    . Symptom onset in autosomal dominant Alzheimer disease: a systematic review and meta-analysis. Neurology. 2014 Jul 15;83(3):253-60. Epub 2014 Jun 13 PubMed.

    . Potent amyloidogenicity and pathogenicity of Aβ43. Nat Neurosci. 2011 Aug;14(8):1023-32. PubMed.

    . Loss of presenilin function causes impairments of memory and synaptic plasticity followed by age-dependent neurodegeneration. Neuron. 2004 Apr 8;42(1):23-36. PubMed.

    . The presenilin hypothesis of Alzheimer's disease: evidence for a loss-of-function pathogenic mechanism. Proc Natl Acad Sci U S A. 2007 Jan 9;104(2):403-9. PubMed.

    . Presenilin-1 knockin mice reveal loss-of-function mechanism for familial Alzheimer's disease. Neuron. 2015 Mar 4;85(5):967-81. PubMed.

    . Loss of Aβ43 Production Caused by Presenilin-1 Mutations in the Knockin Mouse Brain. Neuron. 2016 Apr 20;90(2):417-22. PubMed.

    . Familial Alzheimer's Disease Mutations in Presenilin Generate Amyloidogenic Aβ Peptide Seeds. Neuron. 2016 Apr 20;90(2):410-6. PubMed.

    . Wild-type presenilin 1 protects against Alzheimer disease mutation-induced amyloid pathology. J Biol Chem. 2006 Jun 2;281(22):15330-6. PubMed.

  3. I am very interested in the debate on the pathological effects of PS1 inactive mutations. I rather agree that we should not underestimate the potential role of Aβ43 in the formation of Aβ aggregates. Aβ43 possibly acts as a seed to recruit other Aβ species including Aβ42, as proposed in previous studies, including one in which fly models were used (Burnouf et al., 2015).


    . Aβ43 is neurotoxic and primes aggregation of Aβ40 in vivo. Acta Neuropathol. 2015 Jul;130(1):35-47. Epub 2015 Apr 11 PubMed.

  4. It is fascinating to follow this debate between the teams of De Strooper/Chávez-Gutiérrez and Shen/Kelleher. The technical arguments presented by both teams in the discussion sections of their recent papers in Neuron are valuable information. However, there is a third alternative in this argument that does not attempt to view the PSEN mutant data from an Aβ perspective. The majority of mutations causing familial Alzheimer’s disease (fAD) occur in the PSEN genes and these mutations, as a group, cause an average age of disease onset that is earlier than fAD mutations in APP (Ryman et al., 2014), suggesting (to quote from the Xia et al., 2016, paper) that “PS1 dysfunction is more proximate to disease pathogenesis.” We suggest (Jayne et al., 2016) that a useful strategy for resolving the pathological functions of fAD mutations may be to view the issue from, initially, an exclusively PSEN perspective. In other words, rather than attempting to fit the PSEN fAD mutations into an Aβ story we can focus on the non-γ-secretase functions of the PSENs and then see how we can fit APP and Aβ into that. In that sense our approach is aligned with the Presenilin hypothesis of AD as presented by Shen and Kelleher in 2007. We argue that a parsimonious framework for the PSEN genetic data is that γ-secretase activity is not the critical function of the PSENs affected in AD. Instead, the dominant genetic behavior of the PSEN fAD mutations may be exerted through multimerisation of PSEN holoproteins and the effect that has on non-γ-secretase functions such as the role of PSEN1 holoprotein in N-glycosylation of the V0a1 subunit of v-ATPase (Lee et al, 2010). Our paper is “in press” in the Journal of Alzheimer’s Disease although the PDF of this open access paper may not be downloadable until late June. Anyone who would like a preprint can send me an email,


    . Evidence For and Against a Pathogenic Role of Reduced γ-Secretase Activity in Familial Alzheimer's Disease. J Alzheimers Dis. 2016 Apr 4;52(3):781-99. PubMed.

    . Lysosomal proteolysis and autophagy require presenilin 1 and are disrupted by Alzheimer-related PS1 mutations. Cell. 2010 Jun 25;141(7):1146-58. PubMed.

    . Symptom onset in autosomal dominant Alzheimer disease: a systematic review and meta-analysis. Neurology. 2014 Jul 15;83(3):253-60. Epub 2014 Jun 13 PubMed.

    . The presenilin hypothesis of Alzheimer's disease: evidence for a loss-of-function pathogenic mechanism. Proc Natl Acad Sci U S A. 2007 Jan 9;104(2):403-9. PubMed.

    . Loss of Aβ43 Production Caused by Presenilin-1 Mutations in the Knockin Mouse Brain. Neuron. 2016 Apr 20;90(2):417-22. PubMed.

  5. The PS1 gene appears to be a FAD mutation hot spot: More than 200 mutations have been identified (PSEN1 mutations). Does anyone know why? The age of onset varies from 30 to 60. In comparison, the number of APP mutations is smaller (APP mutations), creating the erroneous perception that FAD mutation carriers in the PS1 gene suffer from the disease earlier. The age of onset of the APP Beyreuther/Iberian/Tuebingen mutation is 28-30. In addition, Down’s syndrome patients exhibit AD-like pathology and symptoms between 30 and 60. So overall, PS1 mutation patients are not necessarily younger-onset than APP mutation patients.

    What matters more than the amount of total Aβ, presumably, is the ratio between the longer and shorter forms of Aβ. To me, the debate over the molecular mechanism being loss-of-function or gain-of-function sounds trivial, because it would not affect the strategy for preventing and treating sporadic AD.

    To my disappointment, in their 2015 Neuron paper, Xia et al. totally neglected our Nat Neurosci paper (Saito et al., 2011) on the PS1 R278I mutation knock-in mice. This was a pity, because I have been a big fan of Jie Shen for her attitude of being independent, fair, different, and outspoken. It was also sad and surprising to see that Neuron published a paper whose data were quite similar to ours, which had been published in Nat Neurosci four years before. 

    One way scientifically to settle this debate is to cross-breed our APP knock-in mice with different PS1 mutation knock-in mice carrying different properties (Saito et al., 2014; Apr 2014 webinar). We have three lines: NL, NL-F, and NL-G-F. The NL line would be the best as the Aβ sequence is humanized, because the Swedish mutation only affects β-cleavage and because γ-cleavage is not affected. We would be happy to perform an international collaboration for this purpose if our colleagues are interested. The fastest way to achieve our common goals together is if the groups who have the best ideas, tools, and animals work together.

    For perspective, in Japan, we often have to apply for grants as a group from different organizations. When accepted we have to exchange unpublished data in meetings of that group, based on a shared trust that we will not scoop each other. Dr Nishimura, Shiga University Medical School, actually presented the artificial R278I mutation in one such meeting in the year 2000, five years before he published it, and we started to collaborate. I believe science progresses best when we collaborate and acknowledge each other’s contributions.

    Finally, I have one comment on the threonine 43 being hydrophilic. The free amino acid and the amino acid residue in a protein can differ from each other. The hydrophilicity/hydrophobicity of an amino acid residue depends on the 3D structure of the peptide/protein, so it is dangerous to claim that threonine 43 is hydrophilic. Do you know what is the most hydrophobic amino acid residue when there is no water in close vicinity? It is D/E (aspartic acid and glutamic acid).


    . Presenilin-1 knockin mice reveal loss-of-function mechanism for familial Alzheimer's disease. Neuron. 2015 Mar 4;85(5):967-81. PubMed.

    . Potent amyloidogenicity and pathogenicity of Aβ43. Nat Neurosci. 2011 Aug;14(8):1023-32. PubMed.

    . Single App knock-in mouse models of Alzheimer's disease. Nat Neurosci. 2014 May;17(5):661-3. Epub 2014 Apr 13 PubMed.

  6. One should add to the debate that the recent Chemuru et al., 2016, paper cited in Jie Shen and Ray Kelleher's comments as showing reduced amyloidogenicity of Aβ43 compared to Aβ42 is controversial. Another recent study, which was not cited, reported exactly the opposite (see Conicella and Fawzi, 2014). Thus, while both Aβ42 and Aβ43 can act as toxic Aβ aggregation seeds, which one of them is more potent has not been settled by biophysical in vitro studies. However, as APP transgenic mice deposit Aβ43 earlier than Aβ42 (Zou et al., 2013), Aβ43 is likely a primary Aβ-aggregation factor and trigger of pathogenesis in vivo (Zou et al., 2013Saito et al., 2011). 

    With respect to the PS1 L435F mutation carriers, the genetic status of our PS1 L435F patient brain samples was confirmed from corresponding frozen tissue (see Fig. EV2 of Kretner et al., 2016, freely available online). Immunohistochemical analysis of Aβ in the frozen tissue gave the same results as the formalin-fixed material shown in our paper. In addition, the brain bank registration numbers of the PS1 L435F cases that we analyzed were confirmed by Ray Kelleher.

    The key point of our study is that the PS1 L435F mutant, on its own (that is, in the absence of any wild-type presenilin), generates Aβ43 as the predominant Aβ species. Among other points already discussed that are inconsistent with the presenilin loss of function hypothesis, it is also important to realize that total γ-secretase activity defects observable for PS FAD mutants in cell culture models are compensated in PS FAD mutation carriers by the remaining wild type PS1 and PS2 genes (Szaruga et al., 2015). Altogether, the interpretation that the primary pathogenic trigger of FAD for the PS1 L435F mutant is Aβ (in this case particularly Aβ43), rather than a presenilin loss-of-function effect on total γ-secretase activity, makes more sense. The observation of robust deposition of the toxic Aβ43 species in the brains of the PS1 L435F mutation carriers is entirely consistent with this view. 


    . C-Terminal Threonine Reduces Aβ43 Amyloidogenicity Compared with Aβ42. J Mol Biol. 2015 Jun 26; PubMed.

    . The C-terminal threonine of Aβ43 nucleates toxic aggregation via structural and dynamical changes in monomers and protofibrils. Biochemistry. 2014 May 20;53(19):3095-105. Epub 2014 May 7 PubMed.

    . Qualitative changes in human γ-secretase underlie familial Alzheimer's disease. J Exp Med. 2015 Nov 16;212(12):2003-13. Epub 2015 Oct 19 PubMed.

    . Generation and deposition of Aβ43 by the virtually inactive presenilin-1 L435F mutant contradicts the presenilin loss-of-function hypothesis of Alzheimer's disease. EMBO Mol Med. 2016 May 2;8(5):458-65. PubMed.

    . Potent amyloidogenicity and pathogenicity of Aβ43. Nat Neurosci. 2011 Aug;14(8):1023-32. PubMed.

    . Aβ43 Is the Earliest-Depositing Aβ Species in APP Transgenic Mouse Brain and Is Converted to Aβ41 by Two Active Domains of ACE. Am J Pathol. 2013 Jun;182(6):2322-31. PubMed.

  7. "Genetic analysis measures the functional importance of genes, it can readily separate the essential from the optional, and the fundamental from the ornamental.”—M.O. Hengartner

    In general, loss-of-function mutations are found to be recessive. Almost by definition, a loss-of-function mutation should be perfectly mimicked by an experimental system in which one allele is knocked out, resulting in about 50 percent protein expression. Therefore, two wild-type alleles will not show an aberrant phenotype, one mutant allele can result in an aberrant phenotype (if half the protein is not sufficient for normal functioning), and two mutant alleles will show the aberrant disease phenotype. If the phenotype observed in a heterozygous condition is not identical to the total loss of one gene copy, then by definition we are dealing with a gain-of-function situation. Note that strictly following the definition, a dominant-negative mutation is a gain-of-function mutation. In such situations the mutant protein has gained the additional function of inhibiting the wild-type protein present in the cell. If the phenotype of the dominant-negative heterozygous condition is identical to the total knockout, in which no protein is expressed, it may be considered a loss-of-function mutation. However, if the phenotype of the dominant-negative heterozygous condition is different than the phenotype of the total knockout, then the mutation is a gain-of-function mutation (with or without a loss of some or all of the wild-type function).  Note that a gain-of-function mutation can occur perfectly together with even a 100 percent loss of the wild-type function. If an enzyme with activity A is mutated to an enzyme with activity B, the phenotype can result from loss of activity A or gain of activity B. If the total knock-out of activity A does not result in the phenotype, then it is the gain of function of activity B that is responsible for the phenotype, even if all enzyme A activity has disappeared. It is important to realize that having no enzyme A activity present is something different than having enzyme B activity present.

    Most scientists will agree that AD will not develop in a situation where one PSEN1 gene copy is intact and one allele is completely knocked out, resulting in a straightforward 50 percent protein expression. Similarly, no plaques will be formed in a total knockout with no PSEN1 expression. So, by definition, because the phenotype of the dominant FAD PSEN1 mutations are different than the one-knocked-out allele situation, no matter if 100 percent of the wild-type function of the defective FAD PSEN1 mutation has been lost, it is the additional gain-of-function of the mutation that causes the important hallmark of AD (amyloid plaques), and that gain of function explains that part of the phenotype not present in the one-knocked-out allele (and 50 percent protein expression) situation.

    So I guess both camps are right in the sense that FAD PSEN1 mutations often involve some loss of PSEN1 protein function, but FAD PSEN1 mutations also result in a clear gain of protein function not present in the wild-type protein. The correct way to express this would be to agree that FAD PSEN1 mutations are gain-of-function mutations (which should be undisputable from a genetics point of view) with a variable amount of (from 0-100 percent?) loss of wild-type function. How important the loss of PSEN1 function is for AD I will leave out of this discussion, but what seems undisputable is that FAD PSEN1 mutations show a gain of function that is crucial for the disease. AD cannot be mimicked by half of wild-type PSEN1 expression or even a total loss of PSEN1 expression, and the FAD PSEN1 mutations in humans are autosomal-dominant—altogether, a textbook example of a gain-of-function mutation. Mutant FAD PSEN1 protein does something that the wild-type protein just does not do, and results in a phenotype different than when there is no PSEN1 present, namely FAD PSEN1 protein processes APP differently than the wild-type protein does, resulting in amyloid plaques—i.e., enzyme A turned (at least partially) into enzyme B. Better molecular understanding of the FAD PSEN1 gain-of-function aspect seems crucial for our understanding of AD, as just studying half or no PSEN1 expression will not suffice. The supporters of a PSEN1 loss-of-function hypothesis are mostly studying a gain-of-function effect, but are not realizing it.

Make a Comment

To make a comment you must login or register.


Mutations Citations

  1. PSEN1 R278I
  2. PSEN1 C410Y
  3. PSEN1 L435F

News Citations

  1. Mutant Presenilin Knock-in Mice Mimic Knockouts, Stir Old Debate
  2. What’s Another Amino Acid? Aβ43 Drives Amyloid Pathology

Paper Citations

  1. . Potent amyloidogenicity and pathogenicity of Aβ43. Nat Neurosci. 2011 Aug;14(8):1023-32. PubMed.
  2. . Trans-dominant negative effects of pathogenic PSEN1 mutations on γ-secretase activity and Aβ production. J Neurosci. 2013 Jul 10;33(28):11606-17. PubMed.
  3. . Qualitative changes in human γ-secretase underlie familial Alzheimer's disease. J Exp Med. 2015 Nov 16;212(12):2003-13. Epub 2015 Oct 19 PubMed.

Further Reading

No Available Further Reading

Primary Papers

  1. . Generation and deposition of Aβ43 by the virtually inactive presenilin-1 L435F mutant contradicts the presenilin loss-of-function hypothesis of Alzheimer's disease. EMBO Mol Med. 2016 May 2;8(5):458-65. PubMed.
  2. . Familial Alzheimer's Disease Mutations in Presenilin Generate Amyloidogenic Aβ Peptide Seeds. Neuron. 2016 Apr 20;90(2):410-6. PubMed.
  3. . Loss of Aβ43 Production Caused by Presenilin-1 Mutations in the Knockin Mouse Brain. Neuron. 2016 Apr 20;90(2):417-22. PubMed.