The results of this study demonstrate that Aβ metabolism in the rhesus monkey is similar to healthy humans (R. J. Bateman et al., 2006
), which is expected because there are no significant amyloid plaques present in the rhesus monkey brain at this age (R. G. Struble et al., 1985
). In conjunction with in vivo
stable-isotope labeling, new generation of CNS Aβ was significantly reduced in response to γ-secretase inhibition. However, in contrast to the periphery, production of CNS Aβ did not rebound above baseline after cessation of inhibition. Defining the metabolic fate of APP in the CNS is critically important for the development of γ-secretase inhibitors to treat AD, as a substrate build-up of APP fragments could potentially lead to an overshoot in neurotoxic amyloid peptides. The lack of Aβ rebound in the CNS could be attributed to the shunting of APP (possibly beta C-terminal fragments, e.g. C99) to γ-secretase independent degradation. In support of this alternative, non-canonical processing, increased levels of APP fragments Aβ 1–14, 1–15, and 1–16 were observed in CSF samples of animals treated with the GSI, while Aβ 1–17 was decreased. The CNS elimination of APP through non-amyloidogenic pathways strengthens the case for the use of γ-secretase as a therapeutic target for AD and may also be relevant for physiological processing of APP.
Many treatments have demonstrated efficacy in preclinical AD murine models; however, to date, none have been successful in clinical trials (K. Duff and F. Suleman, 2004
). Failed translation from preclinical models to clinical studies may be a result of several reasons, including targeting of the wrong molecular mechanism of disease, failure to sufficiently modulate the target, or administration of the therapy too late in the course of the disease (R. J. Bateman and W. E. Klunk, 2008
). For Alzheimer’s disease, one limiting factor for translation from preclinical models to clinical trials is the lack of animal models that closely reflect human Aβ CNS physiology; such models would provide relevant information regarding the target, mechanism of action, and therapeutic effectiveness. Thus, there is a need to develop better models of therapeutic targets that would provide a bridge for the translation of murine models of amyloidosis to human trials in AD.
The CMP rhesus monkey is a nonhuman primate model that enables repeat sampling of CSF (D. B. Gilberto et al., 2003
;P. G. Nantermet et al., 2009
; S. Sankaranarayanan et al., 2009
). Pairing this model with in vivo
stable-isotope-labeling kinetics (SILK) enables direct measurement of newly generated CNS proteins and peptides such as Aβ. Results demonstrate that Aβ metabolism in the rhesus monkey is very similar to humans (R. J. Bateman et al., 2006
), which indicate that the CMP rhesus monkey model can be used for preclinical drug discovery and development studies. Repeated measures (pre- and post-treatment) studies with a crossover design are especially powerful when coupled with the low intra-subject variability. This combination allows for the use of fewer monkeys to clearly answer research questions ( and Supplemental Fig. 1
There are similarities and differences between the metabolism of Aβ in rhesus monkeys and humans. As expected, the average FSR and FCR are balanced in both the rhesus monkey (n=12) () and human (R. J. Bateman et al., 2006
). However, rhesus monkey CNS FSR and FCR are slightly faster than human CNS FSR and FCR (~10% per hour vs. ~8% per hour, respectively (, ( R. J. Bateman et al., 2006
)). Other than species differences, the observed disparity in Aβ metabolism may be due to CSF sampling location. Human CSF was sampled by lumbar intrathecal catheter, whereas monkey CSF was sampled at the cisterna magna at the base of the head (). The more proximal sampling location could also explain the slightly shorter delay in 13
incorporation observed in the monkeys versus that seen in humans (4 h vs. 5 h, respectively), which indicates that lumbar sampling only slightly delays the appearance of newly generated Aβ. Prior human studies demonstrated individual changes in CSF Aβ levels by 100–400% over several hours (R. J. Bateman et al., 2007a
). In this study of nonhuman primate CSF, there were individual CSF Aβ changes of 30–50%. The group changes in Aβ levels were similar between human lumbar CSF samples and nonhuman primate CMP samples when averaged (25–50% over a 24 hour period). However, CMP monkey samples did not demonstrate a progressive rise in Aβ levels. Possible reasons for differences in the intra-subject CSF Aβ variability and lack of CSF Aβ rise include species differences, behavioral activity, site of sampling, frequency or amount of sampling, and sleep/wake cycle differences.
Findings in the nonhuman primate model also have implications for human studies. For example, in this study, we determined that plasma is more accurate than CSF for leucine precursor measurements of CNS Aβ generation (). As the CSF normalized ratio exceeded the theoretical maximum of one, the CSF 13
-leucine is an underestimate of the precursor pool labeling. The labeled leucine precursor enrichment in monkey plasma is higher than CSF, similar to results observed in humans (R. J. Bateman et al., 2006
). This suggests that plasma labeled leucine measurements are more accurate for calculations of Aβ metabolism in human, as well as the nonhuman primate.
The utility of this CMP rhesus monkey translational model was demonstrated by quantifying Aβ metabolism before and after acute exposure to the GSI MK-0752. A prior human GSI SILK study indicates that a decrease in Aβ production can be measured to a sensitive degree (R. J. Bateman et al., 2009
). In this nonhuman primate study, GSI efficacy (blocking Aβ production) and duration was higher compared to that observed in the human GSI study. Possible reasons for these observed differences include the GSI compound used, the dose of GSI administered, and, possibly, species differences. The extended effect may be due to the longer half-life of MK-0752 (10 h) in rhesus monkey vs. LY450139 (2.5 h) in human (E. Siemers et al., 2005
Consistent with previous observations, (T. A. Lanz et al., 2004
; T. A. Lanz et al., 2006
; M. S.Michener et al., 2006
; L. B. Rosen et al., 2006
;E. R. Siemers et al., 2006
;C. R. Burton et al., 2008
) plasma Aβ levels did rebound after the GSI was cleared (). However, concentrations of Aβ in the CSF did not overshoot baseline or placebo Aβ levels (). Measurement of production rates directly confirms that this GSI reduced Aβ levels by decreasing the generation of new Aβ without a subsequent rebound (). These results suggest that the Aβ precursor is being degraded by another pathway during inhibition (E. Portelius et al., 2009
). In vitro
studies have suggested that in the presence of γ-secretase inhibition, β-secretase and α-secretase generate Aβ 1–14, 1–15, and 1–16, and that Aβ isoforms longer than 1–16 are reduced (E. Portelius et al., 2009
). Our results support this hypothesis by demonstrating for the first time in a nonhuman primate model, the rise in Aβ 1–14, 1–15, and 1–16 with a concurrent decrease of Aβ 1–17 (). Further, the time course of altered APP metabolites indicates that Aβ 1–17 is possibly a cleavage product of Aβ 1–40 and Aβ 1–42, as all are decreased at similar times (0–48 hours), while Aβ 1–14, 1–15, & 1–16 are increased during and after inhibition (0–144 hours). Potential candidates known to cleave at the Aβ 1–17 site include endothelin converting enzyme, insulin-degrading enzyme, and matrix metalloproteinase-9 (P. Yan et al., 2006
). These findings advance our understanding of the mechanisms of APP processing in the CNS, and provide novel information about the effects of GSI therapy on APP processing.
In conclusion, our study demonstrates that the effect of therapeutics developed to target Aβ production or clearance can be evaluated directly in a preclinical, nonhuman primate model. This will aid in the selection of clinical candidate compounds, and optimization of dose and timing of drug administration in a model which is similar to human Aβ physiology. In addition, molecular mechanisms of CNS protein processing can be explored by utilizing therapeutics, which modulate key CNS enzymes with physiologically relevant models. These methods and findings hold promise to improve the likelihood of successful clinical trials for the treatment of Alzheimer’s disease and other CNS disorders.