|Home | About | Journals | Submit | Contact Us | Français|
Alzheimer’s disease is characterized by abnormal elevation of Aβ peptide and abnormal hyperphosphorylation of the tau protein. The “amyloid hypothesis,” which is based on molecular defects observed in autosomal-dominant early-onset Alzheimer’s disease (EOAD), suggests a serial model of causality, whereby elevation of Aβ drives other disease features including tau hyperphosphorylation. Here, we review recent evidence from drug trials, genetic studies, and experimental work in animal models that suggests that an alternative model might exist in late-onset AD (LOAD), the complex and more common form of the disease. Specifically, we hypothesize a “dual pathway” model of causality, whereby Aβ and tau can be linked by separate mechanisms driven by a common upstream driver. This model may account for the results of recent drug trials and, if confirmed, may guide future drug development.
Approximately 80 years after Alois Alzheimer described amyloid plaques and neurofibrillary tangles in the brain of a demented patient, the constituents of these hallmark histological features were finally isolated. Aβ peptide, a cleaved product of the amyloid precursor protein (APP), turned out to be the core biochemical element of plaques (Glenner and Wong, 1984), while hyperphosphorylated tau, a microtubule binding protein, was at the core of tangles (Grundke-Iqbal et al., 1986). Once isolated, an inevitable, and for many years intractable, debate ensued over whether Aβ and tau abnormalities were linked, and whether either represented an upstream pathogenic cause driving the disease.
Remarkably, mutations in both genes encoding APP and tau were found to cause dementing illness, and characterizing the clinical syndromes and molecular effects of these mutations informed this debate. Mutations in APP, which accelerate Aβ production, were found to underlie some cases of a monogenic form of the disease, early-onset Alzheimer’s disease (EOAD) (Goate et al., 1991). Although extremely rare, EOAD recapitulates the histological profile of plaques and tangles, and the disease’s clinical phenotype characterized by hippocampal-predominant dysfunction and dysfunction in other neocortical sites. In contrast, mutations in the gene encoding tau, although found to cause tau hyperphosphorylation, do not lead to AD. Rather, these mutations result in a different type of dementia, frontotemporal dementia (FTD) (Hutton et al., 1998), which, as indicated by its name, is characterized by frontal-predominant dysfunction. Moreover, FTD is histologically distinct from AD in that it is characterized by tangles that occur first in extrahippocampal sites, and most importantly, the brain is devoid of plaques. Further clarifying the debate, other disease-causing mutations of EOAD were found in presenilin1 and 2 (Levy-Lahad et al., 1995; Rogaev et al., 1995; Sherrington et al., 1995), and these mutations have a primary effect on Aβ processing and plaque formation.
This series of genetic and experimental findings form the basis of the “amyloid hypothesis” (Hardy and Selkoe, 2002; Tanzi and Bertram, 2005). This hypothesis proposes a serial model of causality, in which elevation of Aβ is the prime pathogenic driver, leading to tau hyperphosphorylation and other histological and clinical features of AD (Figure 1). Because EOAD and late-onset AD (LOAD) phenocopy each other clinically and histologically, the amyloid hypothesis—although based on molecular defects isolated in EOAD—was plausibly proposed to underlie all forms of the disease.
Since the amyloid hypothesis was first formulated over 15 years ago, two developments informing its relationship to LOAD have occurred. First, a crop of amyloid-reducing pharmacological agents has been successfully developed, allowing assumptions of the hypothesis to be tested in LOAD patients. Although, as discussed below, many factors can influence the results of human drug trials, it is nevertheless newsworthy that to date the results of these studies have been largely disappointing. Second, genetic and molecular investigations into LOAD itself have uncovered a separate set of molecular defects whose mechanisms of action suggest that additional models linking Aβ and tau are at least biologically plausible.
Although inconsistencies exist within the wide range of clinical and experimental studies related to LOAD, we believe that as a whole there is currently sufficient evidence to hypothesize an alternative “dual pathway” linking Aβ and tau abnormalities. Specifically, as developed here, we hypothesize that common upstream drivers cause both elevation in Aβ and tau hyperphosphorylation through independent but parallel mechanisms (Figure 1). Notably, this alternative model is not presented as a refutation of the amyloid hypothesis; disease-causing mutations in APP and subsequent studies in animal models (reviewed in Duff, 2001) leave little doubt that the serial model can underlie the disease in certain cases. Rather, the hypothesized dual pathway model suggests an additional mechanism linking Aβ and tau in LOAD, which if confirmed has important therapeutic implications.
According to the amyloid hypothesis, a pharmacological agent that reduces brain Aβ levels should act as an effective drug against the disease. Of the peptide’s two variants, Aβ40 and Aβ42, the latter is thought to be more pathogenic, and decreasing its production or increasing its clearance has motivated drug development. Guided by this rationale, a number of pharmacological agents have been developed and have been evaluated in LOAD patients. Aβ42 immunization is perhaps the most exciting pharmacological development (Wisniewski and Konietzko, 2008), and two methods have been evaluated in patients: active immunization, in which endogenous anti-Aβ42 antibodies are generated by injecting patients with the peptide itself, or passive immunization achieved by directly injecting patients with exogenously generated anti-Aβ42 antibodies. Preclinical studies in mouse models have established that both immunization regimens significantly reduce soluble and insoluble forms of brain Aβ. A third agent developed and tested in LOAD patients is tarenflurbil (Flurizan™, Myriad Pharmaceuticals), which acts by decreasing the production of Aβ42 (Aisen, 2008). Finally, tramiprosate (Alzhemed™, Neurochem Inc.) is a fourth pharmacological agent tested in LOAD patients and worth mentioning (Aisen et al., 2007). While its proposed mode of action is blocking Aβ aggregation, not reducing Aβ per se, its development was in part based on the amyloid hypothesis.
Although at different stages of development, to date, these human clinical trials have been largely disappointing. There are, of course, many reasons for failures in human drug trials and, in any case, negative results do not necessarily refute the serial model underlying the amyloid hypothesis. It is easy to defend the amyloid hypothesis by invoking the plausible assumption that Aβ acts as a “trigger” of downstream tau hyperphosphorylation, so that once initiated, pathology might progress even when Aβ levels are significantly reduced. Because AD has a long incubation period, it is possible therefore that the Aβ-reducing agents tested in LOAD patients were ineffective simply because they were given too late, when secondary events like tau hyperphosphorylation had already begun.
In this regard, one recent study documenting the long-term effects of active Aβ42 immunization is worth highlighting (Holmes et al., 2008). Although the phase I trial was halted because active immunization caused lethal menigoencephalitis in a small number of subjects, this study followed a group of subjects free of the drug’s untoward effects prospectively for over a 6 year period. In fact, during this period, eight consenting patients came to autopsy, allowing the rare opportunity to document drug efficacy by postmortem analysis. In clinical assessments, the dozens of treated patients were compared to a placebo group (AD patients who received a low dose of the peptide that did not generate an immune response), and the results revealed no significant effect of immunization on any clinical measure. Because no subject in the placebo group came to autopsy, the postmortem assessment was performed by comparing the brains’ of the treated group to an age-matched group of AD brains harvested from unimmunized patients. In agreement with preclinical animal studies, immunized patients were found to have significantly lower levels of Aβ. Remarkably, correlating with degree of the immune response, two patients with the highest levels of anti-Aβ antibodies were found to have a near complete reduction of Aβ pathology.
Although the results are based on a small and potentially biased sample, this particular study is notable for its long 6 year followup (most clinical studies are 18 months or less) and the histological evidence documenting the drug’s Aβ-reducing effects. Because of these features, the cognitive profile of clinical decline and the anatomical pattern of pathology reported in the study are extremely informative. At baseline a number of the treated subjects were only mildly affected, as indicated, for example, by a modified mini-mental exam score above 20 (maximum score = 30). Nevertheless, nearly all the subjects who came to autopsy had a profound decline in their cognition over the 6 year period. Seven out of the eight patients were found to have a mini-mental exam score of 0 prior to death, indicating that the patients were so thoroughly impaired that they could not even engage in a simple cognitive assessment.
The clinical progression of AD, as exemplified in this study, reflects two underlying events: the worsening of AD pathology within a targeted brain region, and the anatomical spread of AD pathology to new and relatively unaffected brain regions (Braak et al., 2006) (Figure 2). The relatively long duration of the immunization study and the documented pattern of cognitive decline strongly suggest that both events occurred in the immunized subjects. This interpretation agrees with the observation that at death all treated subjects had tau pathology heavily disseminated across the cortex, as evidenced by a Braak stage of VI in seven out of the eight subjects (the eighth subject, who died 4 months after initiation of treatment, had a Braak stage of V).
The Braak staging system semiquantifies the degree of tau hyperphosphorylation and/or aggregated tangles within a given region (Braak et al., 2006), and, more importantly, the anatomical distribution of tau pathology across the brain. While imperfect, the Braak staging system is considered a good anatomical map of AD-related brain dysfunction, and in general agrees with neuropsychological and imaging indicators of disease (Whitwell et al., 2007). During early stages of the disease, Braak stage I–II, dysfunction is observed primarily in the archiocortex of the entorhinal area, and then, during Braak stage III–V, spreads to tertiary and secondary neocortex within the temporal, occipital, and frontal lobes. Typically, it is only during the last stage (Braak et al., 2006), Braak stage VI, that AD pathology spreads to the primary motor and sensory cortex (Figure 2). While there is some mismatch between anatomical patterns of plaques and tangles, maps of plaque burden agree that primary cortex is typically affected only late in the disease process (Braak and Braak, 1991). Notably, most studies would suggest that based on their mini-mental exam scores, the immunized subjects would most likely be at a Braak stage no greater than IV–V (Jack et al., 2002) at baseline. Although there are rare exceptions to this rule (Knopman et al., 2003), it is nevertheless fair to assume that the primary cortex of the treated group would have been relatively free of tau abnormalities prior to receiving treatment (Figure 2).
The “trigger” idea invoked to defend the serial model of the amyloid hypothesis could have easily explained clinical progression in the immunized subjects had the evidence suggested that AD pathology worsened within tertiary or secondary neocortex. However, we believe that the serial model of causality, even with the trigger caveat, is strained by the evidence suggesting that tau hyperphosphorylation spread throughout the brain, reaching as far as primary cortex.
Besides the issue of timing, a range of additional possibilities have been generated to try and explain the incontrovertibly negative results of this study—in the paper itself (Holmes et al., 2008) and in an accompanying editorial (St George-Hyslop and Morris, 2008). For example, because Aβ peptides aggregate to form dimers and oligomers of different sizes, the possibility was raised that the “truly” neurotoxic form of Aβ might not have been reduced. Additionally, the small number of subjects and the high dropout rate were flagged as potential sources of bias, although, even if considered just a descriptive study, case-studies are typically sufficient to test a “strong version” of the hypothesis. Indeed, the two patients who had a nearly complete reduction of Aβ pathology nevertheless progressed from a baseline mental status score in the low 20 s to 0.
Post hoc explanations are always possible. Nevertheless, we believe that the results from this and other drug studies, at the very least, open up the possibility that an alternative model of causality linking Aβ and tau in LOAD is worthy of consideration.
As mentioned, the amyloid hypothesis emerged out of molecular defects found to be causative in EOAD, though not necessarily in LOAD. In contrast to EOAD, LOAD is a complex, not a monogenic, disorder. Indeed, numerous molecular pathways have been linked to LOAD, and consistent with its etiological complexity, it is often difficult to confirm and validate whether a defective pathway associated with LOAD plays a causative role. Nevertheless, even without the equivalent of “Koch’s postulates,” the pathogenic likelihood of an implicated pathway is increased if it has some degree of genetic confirmation, if expressing defective elements of the pathway in animal models phenocopies features of the disease, and if specific mechanisms linking the pathway to Aβ or tau have been established.
Although a number of molecular pathways implicated in LOAD meet these pathogenic criteria, our goal is not to provide an exhaustive review. Rather, in order to establish the plausibility of an alternative model, we will focus on a select few simply to demonstrate the principle that mechanisms exist by which Aβ and tau can be regulated by single upstream drivers. When generating a testable hypothesis in complex disorders, we believe that some, not all, molecular defects are sufficient for establishing biological plausibility. More than just meeting some of these pathogenic criteria and illustrating the plausibility of the model, the three molecular pathways we will focus on have the additional benefit of being mechanistically interconnected. As we emphasize below, each molecular pathway has its own limitations and weaknesses, so that the overall model (Figure 3) is strengthened when viewed as a whole.
Of the molecular factors linked to LOAD, apolipoprotein E (APOE) genotype is the one that is most robustly implicated in LOAD pathogenesis, although it has been shown to influence EOAD as well (Pastor et al., 2003). The APOE gene encodes three variants: APOE2, APOE3, and APOE4. In 1993, the APOE4 genotype was found to profoundly increase LOAD risk compared with the other variants (Corder et al., 1993; Saunders et al., 1993).
Although inconsistencies do exist, several postmortem studies have shown that inheritance of the APOE4 genotype is associated with an increase in both plaque and tangle load (as shown and reviewed in Tiraboschi et al., 2004). As a corollary in living subjects, cerebral spinal fluid (CSF) studies have also found an association between APOE4 genotype and both Aβ (Galasko et al., 1998; Sunderland et al., 2004) and phospho-tau levels (Glodzik-Sobanska et al., 2007; Golombowski et al., 1997). Although FTD caused by tau mutations does not appear to be robustly influenced by APOE4 genotype (Houlden et al., 1999; Pickering-Brown et al., 2000), it is plausible that the mutations override the impact of APOE4, similar to the lack of APOE4’s effect in AD patients with presenilin mutations (Van Broeckhoven et al., 1994). In support of this, in one study relatively young nondemented individuals (mean = 44), unlikely to have plaques or Aβ pathology, were found to have more tangles in the entorhinal area if they were APOE4 positive (Ghebremedhin et al., 1998).
Experimental confirmation of APOE’s impact on the pathological pathways seen in AD comes from a wide range of cell and animal model studies showing that APOE can affect both Aβ/plaque accumulation (Bales et al., 2002; DeMattos et al., 2004; Dodart et al., 2005) and tau/tangle accumulation (Bi et al., 2001; Brecht et al., 2004; Genis et al., 2000; Harris et al., 2003, 2004; Huang et al., 2001; Ohkubo et al., 2003; Tesseur et al., 2000a, 2000b). Nevertheless, as will be reviewed below, current work in animal models provides a stronger link between APOE4 and Aβ/plaque accumulation. Importantly, a range of findings suggests that APOE affects Aβ and tau through independent mechanisms (Figure 3).
Although a number of mechanisms linking APOE4 to Aβ have been proposed (Ye et al., 2005), the dominant view suggests that APOE released from astroglia plays a role in clearing Aβ from the extracellular brain parenchyma, and that compared to other isoforms, APOE4 causes increased Aβ by decreasing its clearance (Bales et al., 2002; DeMattos et al., 2004; Fagan et al., 2002). From an anatomical perspective is it worth noting that neurons of the entorhinal cortex send their axons to the molecular layer of the dentate gyrus, a layer observed to have high concentration of Aβ deposition (Reilly et al., 2003), and severing these axons causes decreased deposition (Buxbaum et al., 1998; Lazarov et al., 2002). A number of studies have documented that APOE4 causes increased deposition of Aβ in this outflow site of the entorhinal cortex (Irizarry et al., 2000a, 2000b).
A separate and distinct group of mechanisms has been proposed to link APOE with tau hyperphosphorylation (Brecht et al., 2004; Harris et al., 2003; Huang et al., 2001), but one in particular has been supported by a growing number of studies. Specifically, a wide range of studies have established a link between APOE and glycogen synthase kinase 3 (GSK3), one of the main kinases that phosphorylates tau (Hong et al., 1997; Lovestone et al., 1996) (Figure 3). For example, studies have shown that APOE regulates GSK3 activity (Hoe et al., 2006; Ohkubo et al., 2003), that APOE4 has the greatest effect in activating GSK3 (Cedazo-Minguez et al., 2003), and that compared with other isoforms, APOE4 is least likely to bind tau at GSK3 binding sites (Gibb et al., 2000). Because most of these studies were performed in cell culture, these effects are likely to have occurred independent of the effect APOE has on the extracellular concentrations of Aβ in the intact brain.
Further strengthening the link between and APOE and GSK3, several mechanisms have been postulated to account for these empirical observations. One relies on the suggestion that APOE might bind tau (Huang et al., 1995), thereby blocking phosphorylation sites. Because it appears to bind less avidly than other isoforms (Strittmatter and Roses, 1995), studies suggest that APOE4 increases the likelihood of GSK3-mediated tau hyperphosphorylation (Gibb et al., 2000). An unresolved issue, however, is how much APOE is actually expressed in neurons, and whether APOE or its fragments are in the correct intracellular compartment to bind endogenous tau.
A more compelling mechanism linking APOE, GSK3, and tau phosphorylation does not necessarily require neuronal expression of APOE or a direct interaction with tau; rather, it is based on cell surface LDL and LDL receptor-related proteins (LRPs) to which APOE binds (Beisiegel et al., 1989; Herz and Bock, 2002). Although other members of this family of receptors might play a role (Herz and Bock, 2002; Zilberberg et al., 2004), LRP5 and LRP6 are the ones that most directly regulate GSK3 activity, as part of the wnt signaling pathway (De Ferrari and Moon, 2006). Studies suggest that APOE directly binds and interacts with LRP5 (Kim et al., 1998; Magoori et al., 2003), although a functional relationship between APOE and LRP6 has also been suggested (Caruso et al., 2006). At the same time, we should be point out that many experiments use poorly lipidated forms of APOE. LRP5 and LRP6 together with members of the frizzled family of proteins make up a receptor complex to which the extracellular glycoprotein wnt binds. GSK3 is the dominant kinase that mediates the intracellular wnt signaling pathway, such that kinase activity is suppressed when wnt binds its receptor complex (De Ferrari and Moon, 2006). A recent study found that compared with other APOE isoforms, APOE4 inhibits this pathway via LRP5 and LRP6, leading to an upregulation of GSK3 activity (Caruso et al., 2006). Furthermore, a recent genetic finding in LOAD patients suggests a genetic link between APOE and LRP6 (De Ferrari et al., 2007) and provides evidence that the identified LRP6 genetic variant affects wnt signaling via increased GSK3 activity. Additional indirect evidence linking APOE, GSK3, and tau comes from an interesting gene expression study showing that APOE and GSK3 are both differentially expressed in the entorhinal cortex (Liang et al., 2007), which, as mentioned, is the site where tau pathology begins.
Nevertheless, as previously stated, on balance the current evidence establishes a stronger link between APOE4 and Aβ/plaques. In particular, we acknowledge that although in vitro studies indicate that tau phosphorylation can be mediated by GSK3 in the wnt signaling pathway (Asuni et al., 2006; Mercado-Gomez et al., 2008; Mi et al., 2006; Scali et al., 2006), and APOE4 mice do show some degree of tau hyperphosphorylation (Brecht et al., 2004; Harris et al., 2003; Huang et al., 2001), in vivo evidence documenting an APOE-mediated effect on wnt signaling and tau phosphorylation is still lacking.
Several lines of evidence have implicated GSK3 in LOAD, independent of its association with APOE, although a recent study has also implicated GSK3 in FTD (Schaffer et al., 2008). Studies linking GSK3 to LOAD include human studies showing that active-state GSK3 is bound to tangles in the human entorhinal area and other brain regions (Pei et al., 1999; Shiurba et al., 1996), and studies in animal models showing that overexpressing GSK3 causes tau hyperphosphorylation and hippocampal dysfunction in transgenic mice (Engel et al., 2006; Hernandez et al., 2002; Lucas et al., 2001) and enhanced tau pathology in transgenic flies (Jackson et al., 2002).
In addition, a number of genetic studies have either directly (Mateo et al., 2006) or indirectly (De Ferrari et al., 2007) linked GSK3 activity to LOAD. As mentioned, besides APOE4, all other genetic findings linked to LOAD have either small effect sizes or have not been universally found in all studies. Because a genetic variant with a small effect size might nevertheless be pathogenic in complex disorders, as a general point, when evaluating the validity of genetic findings in LOAD a number of questions should be considered. In how many independent populations has the link been found? Because most studies are based on linkage analysis, has an actual “functional” polymorphism in the gene been isolated? Is there is a plausible mechanism suggesting how the genetic defect can play a pathogenic role in the disease?
Based on current knowledge, we believe that the genetic evidence indirectly linking GSK3 activity to LOAD is more compelling than the limited studies that suggest variance in the GSK3 gene itself (Mateo et al., 2006). In the recent study that established a link between genetic variance in LRP6 and LOAD (De Ferrari et al., 2007), this finding was replicated in three separate populations, and a functional polymorphism was identified that increased GSK3 activity through the wnt signaling pathway. Interestingly, genetic variance in T-catenin (Bertram et al., 2007), which is also part of the wnt signaling pathway, has also been linked to LOAD. Additionally, it is worth mentioning that presenilin itself is also part of the wnt signaling pathway indirectly affecting GSK3 activity (Kang et al., 2002; Koo and Kopan, 2004).
As discussed above, there is no question that increasing GSK3 activity will lead to hyperphosphorylation of tau (Figure 3). In a parallel fashion, however, a growing number of studies suggest that GSK3 overactivity might also lead to elevation of Aβ, likely by affecting the enzymatic processing of APP (Phiel et al., 2003; Su et al., 2004) (Figure 3). GSK3 has two isoforms—GSK3α and GSK3β—and currently the most compelling evidence implicates GSK3α activity in APP processing (Phiel et al., 2003), although evidence for GSK3β also exists (Su et al., 2004). In any case, the high sequence homology between the two isoforms suggest functional overlap (Woodgett, 2001), and more importantly, there is now growing evidence to suggest that both isoforms are regulated by similar upstream pathways including wnt signaling (Asuni et al., 2006).
The retromer sorting pathway is made up of multimeric coat complex, which transports a transmembrane retromer-binding receptor (Seaman, 2005). The mammalian complex is made up of VPS35, VPS26, VPS29, and SNX1/2, all highly expressed in the brain (Haft et al., 2000). Mouse studies have shown that elements of the retromer complex are differentially expressed in the hippocampal formation (Lein et al., 2007), and a human study found that within the hippocampal formation, VPS35 and VPS26 are differentially expressed in the entorhinal cortex (Small et al., 2005). Among the many possible candidates, two VPS10-containing proteins, sorLA and sortilin, have been shown to act as retromer-binding receptors in neurons (Muhammad et al., 2008). In yeast, in which the retromer sorting pathway was first described, VPS10 acts as a major retromer-binding receptor, suggesting functional homology.
Converging lines of evidence have implicated the retromer sorting pathway in LOAD: (1) studies in human brain tissue have found deficient levels of VPS35, VPS26, and sorLA (Scherzer et al., 2004; Small et al., 2005). (2) Studies in animal models have shown that deficiencies in VPS35 and VPS26 similar to those found in LOAD patients cause hippocampal dysfunction (Muhammad et al., 2008). (3) Studies in animal models and cell culture have shown that deficiencies in VPS35, VPS26, and sorLA cause an elevation in Aβ and Aβ aggregates (Andersen et al., 2005; Muhammad et al., 2008). (4) Studies in human patients have shown that genetic variance in sorLA (also called sorl1) is linked to LOAD (Rogaeva et al., 2007).
In evaluating the genetic link, it should be emphasized that it has now been replicated in most (though not all) subsequent studies, and according to Alzgene (www.alzforum.org/res/com/gen/alzgene/default.asp), positive linkage has been found in 13 out of 20 populations. Additionally, because sorLA could potentially be involved in numerous biological functions, the primary study that first reported linkage established that sorLA is indeed part of the retromer sorting pathway—showing, for example, that VPS35 binds sorLA and that downregulating either sorLA or VPS26 results in similar increases in Aβ (Rogaeva et al., 2007).
At the same time, it should be pointed out that deficiency in the sorLA protein has been reliably observed in a large number of LOAD brains (Sager et al., 2007; Scherzer et al., 2004), more commonly than would be expected from this rare genetic variant. This concern is partly allayed by a recent finding showing that a primary reduction in VPS35 leads to secondary reductions in sorLA (Nielsen et al., 2007), as would be expected for a retromer-binding receptor (Seaman, 2005). Thus, it is likely that sorLA deficiency observed in LOAD is driven in part by the genetic variant and secondarily by other factors that cause deficiencies in the retromer complex itself. Taken as a whole, we conclude that the evidence implicating the retromer sorting pathway in LOAD is compelling, even though the genetic evidence has a relatively weak effect size.
As mentioned, there is now clear evidence documenting that deficiencies in elements related to the retromer sorting pathways will similarly elevate Aβ. Because sorLA binds APP (Andersen et al., 2005; Spoelgen et al., 2006), the plausible mechanism to account for this effect is that defects in the retromer pathway will cause a mis-sorting of APP, increasing the resident time of APP within a given membrane compartment, thereby increasing the interaction between APP and its cleaving enzymes and leading to accelerated Aβ production (Small and Gandy, 2006) (Figure 3).
Interestingly, a separate and growing group of studies has established that the retromer sorting pathway also plays a role in the wnt signaling pathway. Because, to date, this observation is based on eight separate studies (Belenkaya et al., 2008; Coudreuse et al., 2006; Franch-Marro et al., 2008; George et al., 2007; Pan et al., 2008; Port et al., 2008; Prasad and Clark, 2006; Yang et al., 2008) (as reviewed in (Eaton, 2008) using a range of animal and cell models, the relationship between the retromer sorting pathway and wnt signaling is considered beyond dispute. After transcription and posttranscriptional modification in the endoplasmic reticulum and golgi apparatus, wnt binds and forms a complex with a transmembrane chaperone protein called wntless (WLS) in the transgolgi network (TGN) (Hausmann et al., 2007). The bound complex is then trafficked to the cell surface where wnt dissociates from WLS and is secreted into the extracellular space. The membrane-bound WLS is then endocytosed, and in the endosome WLS binds the retromer complex, which transports WLS back to the TGN. In the face of retromer deficiency, WLS, like sorLA, is sorted to the lysosome for degradation.
Thus, just as with sorLA, retromer deficiency causes a reduction in WLS, with the end result being decreased wnt secretion; a decrease in activation of the wnt receptor complex, frizzled, and LRP5 and LRP6; and an upregulation of GSK3 activity (Figure 3). Although not yet demonstrated empirically, we hypothesize that the upregulation of GSK3 activity caused by retromer deficiency should result in tau hyperphosphorylation (Figure 3), independent and in parallel of its established effects on APP processing.
As mentioned, the dual pathway hypothesis is offered as an additional model of causality, not as a replacement of the amyloid hypothesis. In fact, because Aβ itself has been found to interact with wnt receptors (Chacon et al., 2008; Magdesian et al., 2008), components outlined in Figure 3 can be coopted by the amyloid hypothesis to link Aβ to tau hyperphosphorylation in a serial manner. However, in the dual pathway scenario, reducing Aβ alone, although predicted to have some effect, might be insufficient to completely block tau hyperphosphorylation and subsequent neurodegeneration. From a therapeutic perspective, if the dual pathway hypothesis is validated, diagnosing AD patients at the molecular level would become important, to determine whether the serial or dual pathway model underlies an individual’s pathophysiology. So, for example, patients with primary defects in APP processing (mutations, copy number, etc.) would be predicted to benefit most from Aβ reducing agents, while, according to the model proposed here, patients expressing the APOE4 genotype or having retromer deficiency would not.
An important therapeutic implication of the dual pathway model is that the findings upon which it is based suggest that it might be plausible to develop drugs that simultaneously target both Aβ and tau. Even among defenders of the amyloid hypothesis, the recent drug failures have led to the general acknowledgment that effective therapeutics should target both abnormalities. The prevailing view is that a mixture of different drugs will be needed: those that reduce Aβ and plaques and others that reduce tau phosphorylation and tangles. The observation, as reviewed here, that singular molecular defects may drive both abnormalities, provides a pharmacological rationale for searching out and developing single agents that can ameliorate both core defects of AD.
We believe that clinical findings from a growing number of Aβ-reducing drug trials in LOAD suggest that alternative models linking Aβ with tau are worth considering, and molecular findings related to LOAD provide evidence that an alternative model is biologically plausible. The specific molecular pathways outlined here are just exemplars, showing that Aβ and tau abnormalities can be linked through common upstream drivers. We believe that there are other molecular pathways pathogenic in LOAD that can act in a similar manner (Geschwind, 2003). Of course, currently, the dual pathway model is offered only as a working hypothesis. As is the case for any biologically sound hypothesis, it can be refuted, confirmed, or modified by future studies in human patients and animal models.
This work was supported by NIH grants AG008702, AG025161, and AG07232 (to S.A.S.) and AG017216 and NS048447 (to K.D.) and Alzheimer’s Association (S.A.S. and K.D.).