J. Neurochem. (2012) 120 (Suppl. 1), 3–8.
In this study, we review our finding of APP mutations in Alzheimer’s disease in 1990–1991 with the benefit of 20 years’ perspective. We discuss the historical context in which we made the finding, its immediate and continuing effects on research activity and our hopes for successful clinical testing of the hypothesis. We also briefly discuss the effects finding APP mutations has had on our own careers and those of our colleagues from 1991.
β-amyloid precursor protein
It is 20 years since our group first identified mutations in the β-amyloid precursor protein (APP) gene, which were co-inherited with Alzheimer’s disease (AD) (20 November 1990) (Goate et al. 1991). Our initial paper has been cited nearly 3000 times (Fig. 1) since it was published. Related papers from that time detailing other mutations (Chartier-Harlin et al. 1991; Murrell et al. 1991; Mullan et al. 1992) and the subsequent amyloid hypothesis (Hardy and Allsop 1991; Hardy and Higgins 1992) have also been very highly cited. Our purpose in writing this review is threefold: first to detail the historical context in which our finding was made (see also Ashall and Goate 1994; Goate 2006), second to review the predictions we made at that time and the effects on Alzheimer research since then, and third to look forward to where we need to get to before our understanding of AD pathogenesis can be translated into treatment.
Glenner and Wong (1984a,b) were the first to isolate Aβ (called by them, beta amyloid) from the meningeal vessels of first a late onset sporadic AD case and then, based on the universal occurrence of AD in trisomy 21 (Olson and Shaw 1969) from a Down syndrome brain. It is worth quoting the abstract of the second paper from Glenner and Wong (1984b) since it has proved extremely prescient, both in terms of the genetics of APP and the implied pathogenesis of AD.
The cerebrovascular amyloid protein from a case of adult Down’s syndrome was isolated and purified. Amino acid sequence analysis showed it to be homologous to that of the beta protein of Alzheimer’s disease. This is the first chemical evidence of a relationship between Down’s syndrome and Alzheimer’s disease. It suggests that Down’s syndrome may be a predictable model for Alzheimer’s disease. Assuming the beta protein is a human gene product, it also suggests that the genetic defect in Alzheimer’s disease is localized on chromosome 21.
This abstract is perhaps the first statement of the amyloid hypothesis (see Fig. 2 for different conceptualisations of the amyloid hypothesis in 1992, 1998 and 2002). The gene encoding the Aβ sequence was cloned three years later and was indeed localised to chromosome 21 (Goldgaber et al. 1987; Kang et al. 1987; Robakis et al. 1987). It seemed that all of Glenner’s predictions were being immediately fulfilled. Cloning of this gene also demonstrated that the Aβ peptide was synthesized as a much larger precursor protein (β-amyloid precursor protein – APP or βAPP). It was unclear at this point whether Aβ was solely a disease-associated protein or whether it was also a normal product of APP processing.
At this stage, two mistakes were made in the feverish excitement of these discoveries. The first was that APP gene duplications were reported in sporadic AD cases from France (Delabar et al. 1987): this was not correct in those cases in which it was reported (but see below) and the second was that the APP locus was reported to be genetically close to a locus for early onset AD (St George-Hyslop et al. 1987; Tanzi et al. 1987a). This also was not correct in the families on which the report was based (but, again, see below). A further error was the ruling out of the APP gene as a locus for early onset AD based on the notion that the disease was homogenous and that there was a single locus on chromosome 21 (Tanzi et al. 1987b; Van Broeckhoven et al. 1987). Thus, for ∼3 years (from late 1987 to early 1991) it seemed as if the amyloid hypothesis did not have genetic support.
However, with the realisation that early onset AD was genetically heterogeneous (Schellenberg et al. 1988; St George-Hyslop et al. 1990), and with the identification of APP mutations as a cause of amyloid deposition in hereditary cerebral hemorrhage with amyloidosis Dutch type (van Duinen et al. 1987; Levy et al. 1990; Van Broeckhoven et al. 1990), our group realised that the previous analyses which had ruled out the APP gene had been based on a flawed assumption (genetic homogeneity) and that APP mutations could cause amyloid deposition in some families. With this background, we determined to analyse our families one at a time, starting with the family (F23) in which we had good evidence for chromosome 21 linkage (Goate et al. 1989). This strategy, devised by our group in a series of brainstorming sessions in the summer of 1990, immediately began to pay dividends as we started to find APP mutations (Chartier-Harlin et al. 1991; Goate et al. 1991).
Ironically, given the above history: the families in which chromosome 21 linkage was first reported, were later shown to have presenilin 1 mutations (on chromosome 14: Sherrington et al. 1995) and APP duplications were later found, reliably, in other French families with an inherited form of the disease with cerebral amyloid angiopathy (Rovelet-Lecrux et al. 2006).
Predictions we made based on the findings of APP mutations
We made five predictions based on our identification of APP mutations and their position in the APP molecule:
- (i) Other kindreds with AD would be identified with APP mutations.
- (ii) That Aβ was a normal product of APP metabolism and that the mutations would influence Aβ production.
- (iii) Other Alzheimer genes would be identified and that these would influence Aβ production.
- (iv) Regulatory variants in APP would pre-dispose to late onset AD.
- (v) Aβ deposition is the central event in the pathogenesis of AD.
Predictions 1–3 have been largely borne out (Citron et al. 1992; Haass et al. 1992; Mullan et al. 1992; Seubert et al. 1992; Cai et al. 1993; Suzuki et al. 1994; Scheuner et al. 1996), and the identification of presenilins as the other genes involved in the Mendelian disorder (St George-Hyslop and Petit 2005) together with the demonstration that presenilin is the catalytic subunit of the APP metabolising complex, γ secretase was powerful corroborating evidence for the accuracy of these predictions (De Strooper 2010). Additionally, the genetic demonstration that MAPT mutations were downstream in the pathologic cascade from APP in both humans and transgenic mice (Hutton et al. 1998; Poorkaj et al. 1998; Lewis et al. 2001) was also strong circumstantial evidence for their verity. Prediction (iv) has not yet been shown to be correct, although the identification of pathogenic APP duplications (Rovelet-Lecrux et al. 2006) is consistent with this idea. However, perhaps the most important prediction, prediction (v), with its implication that by reducing Aβ we might prevent AD in general will not be considered proven until Aβ lowering strategies are shown to be efficacious in treating the disease.
Although regulatory variants in APP have not been found that influence risk for LOAD the most common risk factor for LOAD, the ε4 allele of Apolipoprotein E (APOEε4), does influence risk by an Aβ dependent mechanism (Castellano et al. 2011). APOEε4 alleles cause a dose dependent increase in risk and decrease in age at onset of AD while APOEε2 alleles decrease risk and increase age at onset (Corder et al. 1993, 1994). Metabolic labeling experiments in humans have also shown that Aβ clearance is impaired in AD compared with non-demented controls (Mawuenyega et al. 2010). Although the role of genetic factors were not examined in this study a subsequent study has demonstrated that human APOE isoforms differentially affect Aβ clearance resulting in an APOE isoform-dependent pattern of Aβ accumulation later in life (Castellano et al. 2011). These studies provide strong support for the amyloid hypothesis by demonstrating that LOAD risk factors as well as FAD genes influence risk via an Aβ dependent mechanism, albeit clearance rather than production of Aβ. Recent genome wide association studies have provided evidence for at least eight novel risk alleles for LOAD (Harold et al. 2009; Hollingworth et al. 2011; Naj et al. 2011). These studies point to genetic variability in cholesterol metabolism and in the complement cascade as being crucial (Jones et al. 2010). The complement cascade is clearly a damage response pathway and this work suggests that variability in how we respond to Aβ deposition may be of great importance in risk to AD. Genetic analysis is ongoing and our labs and others are now sequencing exomes (all the protein coding regions) to find more of the genetic variability underpinning the disease process.
How finding mutations in APP changed AD research
Prior to the discovery of these mutations there were no laboratory animal models of AD. Molecular studies of AD were therefore focused on post-mortem human tissue, which made it impossible to determine whether the observed differences were causal or differences resulting from the disease process. After causal mutations were discovered transgenic mouse models were developed that allowed studies of the molecular pathogenesis of AD (Games et al. 1995; Hsiao et al. 1996; Sturchler-Pierrat et al. 1997). A consistent property of these animals is age-dependent Aβ deposition. Many of the models develop neuritic plaques, microglial activation, astrocytosis, evidence of oxidative damage, and changes in neuronal cytoskeletal proteins including tau (Masliah et al. 1996; Irizarry et al. 1997; Frautschy et al. 1998) and some, but not all develop behavioural impairment (Hsiao et al. 1996). These mice do not develop marked neuronal loss nor do they develop typical neurofibrillary tangles; however, it has been shown that there are significant functional synaptic abnormalities, which develop in the hippocampus and neocortex as Aβ deposition progresses and surrounding Aβ plaques (Busche et al. 2008). Another, striking observation coming from these mice is that over-expression of Aβ42 leads to parenchymal Aβ deposition, such as that seen in AD, while over-expression of Aβ40 leads to Aβ deposition primarily in the cerebral vessels (Herzig et al. 2004) thus mutations that increase total Aβ, such as APP duplication show substantial parenchymal and vascular Aβ deposition and exhibit clinical symptoms of cerebral hemorrhage and dementia while mutations altering Aβ42 but not total Aβ have largely parenchymal deposition and dementia only. These animals thus model many aspects of ‘pre-clinical’ or ‘pre-symptomatic’ AD in humans (Perrin et al. 2009).
Proving the Aβ hypothesis through successful treatment trials
There is strong evidence from pre-clinical studies in mice that prevention or reversal of Aβ and tau aggregation has strong potential to serve as a way to delay onset or slow progression of AD. In regard to Aβ, strategies have been devised to decrease soluble Aβ production, enhance soluble Aβ clearance, influence Aβ aggregation, neutralize Aβ toxicity, or remove existing Aβ aggregates. It is not the purpose of this article to discuss the detailed strategies and outcomes of completed or ongoing treatment trials (see Golde et al. 2010). However, while it has to be acknowledged that no trials have yet been successful, it is also clear that most of them have had serious and rather obvious flaws in their design or been prematurely halted by side effects. They have thus not been satisfactory tests of the hypothesis.
Much has been made of the fact that the first immunisation trials were successful in clearing the Aβ from the brain in humans, as they had done in mice, but had no large effect on the clinical symptoms suggesting that Aβ clearance in late stage disease is not a miraculous cure (Schenk et al. 1999; Holmes et al. 2008). Against this, it is worth noting that L-dopa was not shown immediately to have benefit in Parkinson’s disease and that mechanistic trials are intrinsically much more difficult to organise and interpret than symptomatic trials. Perhaps the most hopeful aspect of the immunisation trial was that CSF tau levels were significantly lower after treatment (this was, admittedly, a secondary analysis). It is however, difficult to conceive of a way that a reduction in CSF tau could occur unless the rate of brain damage had been reduced. Certainly, our view is that if this independent biomarker of neuronal damage shows signs of improvement in a large trial, we should be confident we are on the right road.
Longitudinal and cross-sectional biomarker studies in AD have shown that Aβ deposition and changes in CSF Aβ levels occur up to 15 years prior to symptom onset and show little change in symptomatic individuals while CSF tau levels increase shortly before symptom onset and continue to increase throughout the disease (reviewed in Holtzman et al. 2011). In cognitively normal elderly individuals, a high ratio of CSF tau/Aβ42 is a strong predictor of conversion to very mild dementia/mild cognitive impairment over a 3= to 4-year period (Fagan et al. 2007; Li et al. 2007; Craig-Schapiro et al. 2010; Jack et al. 2010). These studies strongly argue that anti-amyloid treatments in symptomatic individuals are unlikely to have a large impact on disease progression but could delay or halt disease onset if administered to pre-symptomatic individuals. It is therefore imperative that we reconsider clinical trial design if we are to truly test the amyloid hypothesis.
Several Alzheimer’s disease prevention trials are planned in the next five years (http://www.alzforum.org/new/detail.asp?id=2727). These trials will be largely performed in the very families our genetic studies first identified. These families have been shut out of previous clinical trials because of their early onset of disease. However, for prevention trials, particularly those based on anti-amyloid therapies (e.g. immunization, β or γ-secretase inhibition) they provide two crucial advantages: first the cause of disease in these families is clearly overproduction of Aβ peptides and second, pre-symptomatic individuals can be unambiguously identified by mutation screening enabling treatment to be administered in pre-symptomatic individuals with biomarker profiles indicative of ongoing disease pathology. Although initial studies will use biomarker endpoints to determine efficacy, it will be necessary to demonstrate functional endpoints before drugs are approved for treatment in other AD populations.
A personal view
We knew, when we saw APP mutations in F23, that our lives would be changed forever and so it has proved. Within 2 years, our group had broken up and moved on, tempted by the opportunities offered to us as well as driven by the poor economic climate in the UK. We separately moved to the United States, although one of us (JH) has recently returned. Behind nearly every really important finding, there is some drama, and this paper was not an exception. In our case, the mishandling of the commercialisation initiated a complex series of events which are still the subject of lawsuits (see http://www.alzforum.org/new/detail.asp?id=2472; Check Hayden 2011). However, most of us involved in the work are still in friendly contact with each other and our groups have collaborated extensively in the identification of presenilin mutations in 1995 (Clark et al. 1995), MAPT mutations in 1998 (Hutton et al. 1998) and most recently with Mike Owen (a third member of our group in 1988–1991) were part of the recent and very successful genome wide association studies for Alzheimer’s disease (Harold et al. 2009; Jones et al. 2010; Hollingworth et al. 2011; Naj et al. 2011). Although we both feel, we have done important work since 1991, there is no doubt that this paper, has dominated our careers.
conflict of Interests
Authors have not disclosed any potential conflicts.