Discussion
We present a new mouse model for AD that was designed to test the consequences of inhibiting Aβ production after the onset of amyloid pathology. New lines of transgenic mice were developed for this study that express high levels of APPswe/ind under the control of a tetracycline-responsive promoter. We demonstrate that treatment with dox suppresses steady-state levels of both APPswe/ind and its C-terminal fragments, indicating that the mutant proteins have a relatively short half-life in vivo. Transgenic expression of APPswe/ind and consequent overproduction of Aβ42 cause early-onset amyloid deposition in untreated mice, in which deposits appear as early as 2 mo of age. Amyloid burden worsens significantly with age, and by 9 mo, the hippocampus and cortex of untreated mice are largely filled with aggregated peptide. We find that suppression of transgenic APP by more than 95% abruptly halts the progression of amyloid pathology. Importantly, this outcome occurs in animals already harboring considerable amyloid pathology, a situation similar to what might be expected in patients to be treated with secretase inhibitors. Somewhat unexpectedly, we observe no appreciable clearance of deposited amyloid even following periods of transgene suppression equal to the time taken for plaques to form. This latter finding indicates that compared to other disease-associated protein aggregates such as mutant huntingtin, which clear in less than 3 mo [32], the disaggregation of extracellular amyloid is relatively slow.
Notably, pharmaceutical γ-secretase inhibitors published to date show less strict regulation of Aβ production following chronic administration than we have attained here. Two independent compounds tested in Tg2576 [33] and TgCRND8 [34] transgenic mice show no more than 85% suppression of Aβ40 levels, and where measured, even less suppression of Aβ42 (approximately 60%) [35,36]. Thus, even after accounting for the much higher APP expression levels in our mice than in the Tg2576 and TgCRND8 lines, we have achieved better absolute suppression of Aβ production with the tet-off system than is currently possible with published γ-secretase inhibitors. Since even the most advanced future pharmaceutical agents are unlikely to attain more complete control of Aβ production than achieved here, this system provides a salient test of therapeutic intervention with Aβ-lowering compounds.
Although the progression of amyloid deposition was sharply arrested by this approach, we found that a substantial amyloid burden remained even after long periods of transgene suppression. We examined a small number of animals after 12 mo of dox treatment (beginning at 6 mo of age), and found that amyloid deposits were still relatively abundant. Longer-term treatments are now in progress. At the latest treatment interval analyzed by ELISA, animals administered dox for 6 mo showed elevated levels of PBS-soluble Aβ (see Figure 5) that could be interpreted as an indication that the plaques are slowly releasing peptide (or oligomeric Aβ) into the soluble pool and might eventually dissolve. Whether inhibiting Aβ production longer than 6 (or 12) mo may ultimately result in clearance of amyloid is under investigation; unfortunately, the life span of the model eventually limits this experiment. Therapeutics used in humans will have considerably more time to act than is possible within the life span of rodent models. Long-term treatments would certainly be possible and could be a key to effective therapy. Overall, however, we interpret our findings as evidence that AD therapies that significantly lower the production of Aβ (by either inhibiting secretase activity or inhibiting APP expression) may not quickly reverse preexisting pathology, but should effectively halt further deposition of amyloid.
In interpreting our study, it should be remembered that the earliest plaques to appear in these mice, like other APP transgenics harboring the Swedish mutation [27], are predominantly fibrillar deposits, which may be less tractable than the diffuse aggregates thought to come first in the course of the human disease. However, our data suggest that once diffuse deposits are formed in these mice, they are no more easily cleared in our system than cored plaques (see Figure S4). An additional consideration we recognize is that a small amount of transgene expression continues in the presence of dox and that endogenous mouse Aβ continues to be produced. It is possible that the combined low levels of endogenous mouse Aβ and nonsuppressed human peptide are sufficient to maintain existing deposits. However, these low levels of peptide are not sufficient to induce new amyloid formation, as CaMKIIα-tTA × tetAPPswe/ind mice raised on dox for up to a year do not develop amyloid lesions (data not shown). It is also clear that in this genetic system, we have raised the production of Aβ to levels not found in humans to accelerate pathology into an experimentally feasible time frame. This system allowed us to create an approximately 20-fold differential between APP/Aβ synthetic rates before and after treatment, yet the in vivo equilibrium between aggregated and disaggregated states of Aβ still favored the maintenance of amyloid deposits. In our opinion, it seems unlikely that amyloid deposits in human brain would be inherently any less stable than those formed in mouse brain. However, the human brain may harbor clearance mechanisms not shared by mice that would allow more efficient removal of preexisting amyloid.
One potential mechanism by which amyloid may be more efficiently removed in the human disease than in the mouse models is through microglial phagocytosis. Resident microglia in transgenic mouse models localize to tissue surrounding plaques but show little evidence of amyloid engulfment [37–40]. In contrast, microglia surrounding amyloid plaques in human brain show a much higher state of activation with greater expression of complement receptor [40]. Thus, the role of microglia in amyloid metabolism is minor in transgenic models compared to the human condition. Somewhat paradoxically, several studies further demonstrate that treatment with anti-inflammatory drugs to reduce microglial activation actually lowers amyloid load in APP transgenic mice, suggesting a role for mouse microglia in the formation and maintenance of amyloid aggregates [41–43]. However, this outcome may be alternatively explained by direct effects of many anti-inflammatory drugs on γ-secretase cleavage [44–47]. Nonetheless, the role of microglia in both the human condition and the mouse models is poorly understood, and differences in microglial reactivity between the two could lead to significantly faster amyloid clearance in the brains of patients with AD than we observe in the tet-off APP mice.
Given the relatively minor role played by microglia in other mouse models of amyloidosis, we think it unlikely that these cells have influenced the rate of amyloid clearance in the tet-off APP mice. Even so, we considered the possibility that chronic dox treatment may have altered the activation state of microglia in our treated mice. Dox is structurally similar to minocycline, a reported anti-inflammatory drug and inhibitor of microglial activation [48]. However, if dox does have anti-inflammatory activity, then, based on previous studies with other anti-inflammatories, we would have expected to find less amyloid in the dox-treated animals. Clearly, that was not the case. While it is possible that dox acts in some other way to slow amyloid clearance, data from multiple studies demonstrate that microglial responses are normally weak in the mouse AD models [37–40], and thus it is doubtful that dox-mediated microglial inhibition affected the outcome of our study.
The persistence and stability of amyloid deposits in our system is unexpected given the speed with which Aβ aggregates are cleared in other mouse models of therapeutic intervention. Anti-Aβ antibodies injected directly into the brain have been shown to eliminate amyloid deposits in as little as 1 wk after treatment [49–51]. Peripheral antibody injection decreases amyloid load more broadly, and although it does not appear to act as quickly as local injection, can significantly reduce amyloid load within 2 mo of initial treatment [52,53]. More recently, an alternative approach has shown that lentiviral transfer of neprilysin can also reduce the number of aggregates in the area of the injection site [54]. Careful study of the mechanism behind several of the antibody-mediated therapies has suggested that activated microglia play an important role in the removal of fibrillar plaques after immunization [50,52,55]. However, it has been noted that deletion of the Fc receptor (the primary receptor for microglial opsinization of antibody–antigen complexes) in APP transgenic mouse models has no impact on the effectiveness of antibody-mediated therapy [56,57]. It is, nevertheless, possible that lack of microglia activation is the major difference between the slow clearance described here, where no perturbation of the immune system is expected, and the rapid clearance described in studies involving antibody or viral injection. In isolation, mild activation of microglia by injection damage or opsinization may not be adequate to induce substantial phagocytosis, but when combined with an Aβ-lowering agent, such as neprilysin or Aβ-targeted antibodies, the two may work in concert to clear peptide deposits. Consistent with this hypothesis, strong activation of microglia through transgenic expression of TGFβ [58] or central injection of lipopolysaccharide [59,60] can by itself substantially reduce plaque burden in APP transgenic mice. But in the case of acute antibody- and/or injury-mediated activation, once the inflammation has passed, and the antibody and bound peptide have been cleared and degraded, the remaining Aβ quickly reaggregates and amyloid pathology is reestablished [49]. This finding reinforces the notion that without continued stimulation, microglia in mouse models do not maintain the same level of sustained activation that may occur in humans.
SantaCruz et al. recently published a study of mice that express P301L human tau via a similar vector system [61]. As in our tet-off APP mice, SantaCruz et al. found that tau neurofibrillary tangles, like amyloid plaques, are not cleared efficiently following transgene suppression. The lack of clearance in both models of AD pathology comes as a stark contrast to the rapid removal of protein aggregates found in similar tet-off mouse models of Huntington [32] and prion disease [62]. In these cases, disrupting the input of new monomer to the system via dox-mediated transgene suppression led to relatively rapid clearance of protein aggregates. By contrast, our study and that of SantaCruz et al. suggest that protein aggregates in AD may be more tenacious than in other neurodegenerative disorders. Perhaps once aggregated, Aβ and tau are either inherently more stable than other protein aggregates or more resistant to intra- and extracellular clearance mechanisms.
One question we were not able to address in this study is whether abrogating synthesis of new Aβ halts the progression of cognitive decline. Studies from the tet-off tau mice suggest that protein clearance may in fact not be required for cognitive improvement following transgene suppression [61]. At present, because of unexpected noncognitive behavioral abnormalities, it is not clear whether the tet-off APP mice can be used to address the same question in the context of amyloid pathology. Both lines of tTA/APP mice we studied here display extreme hyperactivity visible as cage circling and quantified by activity monitoring (see Figure 7). Many of the double transgenic mice showed similar circular patterns of swimming near the edge of the tank when tested in the Morris water maze. Expression of this phenotype makes standard tests of learning and memory uninterpretable. Hyperactivity nonspecifically inhibits choice-driven changes in movement, the key element behind all cognitive behavioral paradigms. We are currently working to determine whether hyperactivity correlates with expression of the APPswe/ind holoprotein or its proteolytic derivatives. Preliminary studies suggest that hyperactivity does not appear quickly when dox-reared mice are shifted to nonmedicated diets (J. L. J., unpublished data). These data may indicate that the neuroactive culprit is not immediately present after transgenic APP synthesis is initiated, but requires additional time to develop. Alternatively, hyperactivity may be caused by neuronal alterations due to transgene expression during early postnatal development. Further experiments are needed to distinguish between these possibilities.
In summary, we demonstrate that abrogating Aβ production halts the progression of pathologic changes in a transgenic mouse model of Alzheimer-type amyloidosis. However, despite dramatic reductions in Aβ synthesis, neuritic plaques are stable structures in vivo that do not quickly disaggregate. It is possible that a combination of therapies to limit Aβ production, increase Aβ degradation, and enhance phagocytosis of deposited amyloid may be required to reverse damage associated with AD. However, if started early enough in the course of disease, secretase inhibitors alone could provide substantial benefit in slowing pathogenic processes linked to amyloid deposition. Even at later stages in the disease, the presence of substantial microglial activation in human AD [40] suggests that simply slowing the formation of new amyloid deposits may allow ongoing phagocytosis to diminish preexisting lesions. However, the development of safe and effective secretase inhibitors will ultimately be required to determine whether the human brain has the capacity to repair amyloid-associated damage of AD once the progression of pathology is arrested.