A hallmark of Alzheimer’s disease (AD) pathology is the accumulation of brain amyloid β-peptide (Aβ), generated by γ-secretase-mediated cleavage of the amyloid precursor protein (APP). Therefore, γ-secretase inhibitors (GSIs) may lower brain Aβ and offer a potential new approach to treat AD. As γ-secretase also cleaves Notch proteins, GSIs can have undesirable effects due to interference with Notch signaling. Avagacestat (BMS-708163) is a GSI developed for selective inhibition of APP over Notch cleavage. Avagacestat inhibition of APP and Notch cleavage was evaluated in cell culture by measuring levels of Aβ and human Notch proteins. In rats, dogs, and humans, selectivity was evaluated by measuring plasma blood concentrations in relation to effects on cerebrospinal fluid (CSF) Aβ levels and Notch-related toxicities. Measurements of Notch-related toxicity included goblet cell metaplasia in the gut, marginal-zone depletion in the spleen, reductions in B cells, and changes in expression of the Notch-regulated hairy and enhancer of split homolog-1 from blood cells. In rats and dogs, acute administration of avagacestat robustly reduced CSF Aβ40 and Aβ42 levels similarly. Chronic administration in rats and dogs, and 28-day, single- and multiple-ascending–dose administration in healthy human subjects caused similar exposure-dependent reductions in CSF Aβ40. Consistent with the 137-fold selectivity measured in cell culture, we identified doses of avagacestat that reduce CSF Aβ levels without causing Notch-related toxicities. Our results demonstrate the selectivity of avagacestat for APP over Notch cleavage, supporting further evaluation of avagacestat for AD therapy.
Alzheimer’s disease (AD) is a progressive, neurodegenerative disorder characterized by memory loss and deficits in executive function—ultimately culminating in an impaired ability to perform activities of daily living, dementia, and death (Desai et al., 2004; Selkoe 2007). Although the etiology of AD remains unknown, the amyloid hypothesis contends that amyloid β-peptide (Aβ) is involved in the initiation and progression of AD symptoms (Hardy and Selkoe, 2002). Aβ peptides are formed by the sequential cleavage of amyloid precursor protein (APP) by β-secretase (also known as β-site APP-cleaving enzyme) and γ-secretase, which catalyzes intramembrane cleavage of a C-terminal fragment generated by β-site APP-cleaving enzyme (Steiner et al., 2008). Amyloidogenic processing of APP generates Aβ species of 36 to 43 amino acids due to different C-terminal γ-secretase cleavages (Steiner et al., 2008; Querfurth and LaFerla, 2010). Aβ40 is the most abundant species (80–90%); however, Aβ42 is considered largely responsible for the pathophysiologic events that occur in the AD brain (Walsh and Selkoe, 2004; Bergmans and De Strooper, 2010). Aβ42 has a propensity for aggregation and is the principal component of brain amyloid plaques that form in AD (Walsh and Selkoe, 2004; Bergmans and De Strooper, 2010). Thus, inhibiting γ-secretase activity by lowering brain levels of Aβ, particularly Aβ42, is a potential AD treatment strategy (Bergmans and De Strooper, 2010).
A challenge in developing γ-secretase inhibitors (GSIs) for AD therapy is to avoid effects on γ-secretase substrates other than APP (Barten et al., 2006; Bergmans and De Strooper, 2010). Various type I transmembrane proteins have been identified as γ-secretase substrates, including the Notch family of receptors (Notch1–Notch4) (Pollack and Lewis, 2005). Notch proteins mediate cell–cell communication, regulating cell-fate decisions throughout development and in adult tissues (Kopan and Ilagan, 2009). Cleavage of γ-secretase releases the Notch intracellular domain, which translocates to the nucleus and up-regulates gene expression (Kopan and Ilagan, 2009). As Notch proteins play a fundamental role in a wide range of tissues, GSI interference with Notch signaling can have adverse effects in several adult organ systems.
Subchronic dosing of GSIs can cause adverse effects in the gastrointestinal (GI) tract, thymus, spleen, skin, and hair in animals, likely resulting from reduced Notch signaling (Milano et al., 2004; Wong et al., 2004; Prasad et al., 2007; Kumano et al., 2008). In the GI tract, GSIs can cause decreased enterocyte differentiation, goblet cell metaplasia, and potential injury to the intestinal epithelium (Milano et al., 2004; Wong et al., 2004). These GSI-induced abnormalities in the gut resemble the Notch1- and Notch2-deficient murine phenotype (Riccio et al., 2008). In the thymus, GSIs alter T cell maturation and cause thymic atrophy (Milano et al., 2004), mirroring loss of Notch1 (Radtke et al., 2004). In the spleen, GSIs alter B cell maturation and cause atrophy of the marginal zone (Milano et al., 2004; Wong et al., 2004), mimicking loss of Notch2 (Saito et al., 2003). In the skin, GSI treatment of mice can cause epithelial cell hyperplasia reminiscent of the precancerous phase of squamous cell carcinoma (Li et al., 2007a) seen when γ-secretase activity is reduced by genetic manipulations (Xia et al., 2001; Zhang et al., 2007). Encouragingly, manipulating components of the γ-secretase complex to achieve partial inactivation suggests that Notch-related toxicities are threshold-driven and are not observed when γ-secretase activity is reduced by ≤30% (Li et al., 2007b). Studies with GSIs in mutant APP transgenic mice suggest that separation of pharmacologic activity and toxicity is possible (Barten et al., 2005; Hyde et al., 2006; Best et al., 2007). Furthermore, a recently identified mutation in APP protected against AD while causing a 40% decrease in Aβ formation (Jonsson et al., 2012).
Using preclinical models, we previously developed and screened a series of carboxamide-substituted sulfonamides as potential GSIs (Gillman et al., 2010). The compound designated avagacestat (BMS-708163) was selected for further evaluation based on potency, selectivity for effects on APP relative to Notch, oral bioavailability, and brain penetrance. In cultured H4-8Sw cells, avagacestat potently inhibited the formation of Aβ40 and Aβ42 with IC50 values of 0.30 and 0.27 nM, respectively (Gillman et al., 2010). Avagacestat exhibited a much weaker potency for inhibiting mouse Notch1 processing, with an IC50 of 58 nM. In vivo evaluation of avagacestat in dogs showed a plasma half-life that supports daily oral dosing, good brain penetration, and a correlation between reductions of Aβ40 levels in the brain and cerebrospinal fluid (CSF) (Gillman et al., 2010). Here, we evaluated the pharmacokinetic and pharmacodynamic properties of avagacestat, providing evidence for its selective effects on brain Aβ levels relative to effects on Notch signaling in rats, dogs, and healthy human subjects.
Materials and Methods
Avagacestat was supplied by Bristol-Myers Squibb Research and Development (Wallingford, CT) and was prepared following the general procedures of Gillman et al. (2010).
Cellular Assay for Notch Signaling.
Inhibition of Aβ40 and Aβ42 formation and mouse Notch1 signaling in cultured cells were performed as previously described (Gillman et al., 2010). Human Notch constructs (Notch1–Notch4) were generated by polymerase chain reaction and verified by sequencing as described for mouse Notch1 (Gillman et al., 2010). In contrast to mNotch1 ΔE, all four human Notch constructs contained intact cytoplasmic domains. Human Notch signaling assays were conducted as described for mouse Notch1 (Gillman et al., 2010).
Avagacestat Dosing Studies in Rats and Dogs.
All animal experiments were approved by the Institutional Animal Care and Use Committee of Bristol-Myers Squibb, Inc. Details of acute and chronic dosing studies in rats and dogs, and the collection of tissues for analyses, are detailed in Supplemental Materials and Methods. Paraffin sections of gut and spleen were stained using the periodic acid–Schiff base and hematoxylin and eosin methods, respectively. Histopathology was evaluated by light microscopy.
For human studies, all participants gave informed consent to participate. The local institutional review boards approved the studies, which were conducted in accordance with the Helsinki Declaration. All subjects received compensation for participation. Details of the human subjects who participated in single- and multiple-ascending–dose studies, avagacestat dosing of these subjects, and CSF sampling are detailed in Supplemental Materials and Methods.
Measurement of CSF and Brain Aβ Levels.
Collection of CSF, processing of brain tissue, and the measurement of Aβ levels are described in Supplemental Materials and Methods.
Plasma Avagacestat Analyses.
The analyses of plasma samples for avagacestat were conducted by liquid chromatography-tandem mass spectrometry as described previously (Gu et al., 2010).
Measurement of Notch-Regulated Hairy and Enhancer of Split Homolog-1 Levels.
Hairy and enhancer of split homolog-1 (HES-1) mRNA levels in blood were measured by quantitative reverse transcriptase polymerase chain reaction by standard techniques using an Applied Biosystems 7700 system (Life Technologies, Carlsbad, CA). Blood from rats and dogs was collected in Paxgene Blood RNA Tubes (Becton Dickinson, Franklin Lakes, NJ), and total RNA was isolated using Paxgene blood RNA kit (Qiagen, Valencia, CA). Glyceraldehyde-3-phosphate dehydrogenase mRNA levels were used as a normalization control.
Notch/APP Selectivity of Avagacestat in Cell Culture
As in our previous studies (Gillman et al., 2010), we used an assay for Notch signaling in HeLa cells to evaluate the effects of avagacestat on the processing of each of the four human Notch proteins. These assays showed that avagacestat inhibited signaling of hNotch1, hNotch2, hNotch3, and hNotch4 with IC50 values of 41, 10, 35, and 38 nM, respectively. Using an IC50 value of 0.30 nM for Aβ40 (Gillman et al., 2010), the corresponding Notch/APP IC50 ratios were 137, 33, 117, and 127. The IC50 ratios demonstrate a much greater effect of avagacestat on APP cleavage relative to hNotch1, hNotch3, and hNotch4 processing in these cellular assays. Although the IC50 ratio for hNotch2 was less than for the other human Notch proteins, the results show a high degree of selectivity for APP relative to this isoform as well. These results extend our previous findings and show that avagacestat is a potent and selective inhibitor of APP processing relative to the signaling of human Notch proteins in cellular assays.
A range of Notch/Aβ IC50 ratios have been reported for avagacestat, from as low as 2.9 to as high as 193 (Martone et al., 2009; Gillman et al., 2010; Chávez-Guttiérez et al., 2012; Crump et al., 2012; Mitani et al., 2012) (Supplemental Table 1). This range is partly due to differences in methodology, such as cell-based formats or choice of enzymatic assays, and partly due to the increased potency of GSIs when substrate expression levels are higher (Burton et al., 2008). However, Notch/APP IC50 ratios also shift in a consistent fashion for other GSIs, such that the rank order of Notch/APP IC50 ratios remains unchanged. For example, avagacestat exhibits Notch/APP IC50 ratios that are consistently 15- to 43-fold greater than the corresponding ratios reported for semagacestat across multiple studies, as summarized in Supplemental Table 1 (Martone et al., 2009; Gillman et al., 2010; Chávez-Guttiérez et al., 2012; Mitani et al., 2012). This demonstrates that GSIs exhibit intrinsic and robust differences with respect to Notch/APP selectivity, and that avagacestat is one order of magnitude or more selective than semagacestat.
Effects of Avagacestat on Central Aβ Levels in Rats and Dogs
To investigate the effect of avagacestat on Aβ levels in vivo, the compound was administered to rats using a single oral dose of 2 or 10 mg/kg. In previous studies in rats and dogs, avagacestat administration produced plasma drug concentrations with half-lives that supported once-daily oral dosing (Gillman et al., 2010). In the present study, rat brain and plasma concentrations of avagacestat were similar, with a 1.1 brain/plasma ratio for mean area under the concentration-time curve (AUC). Brain Aβ40 and Aβ42 levels were significantly reduced at both doses relative to vehicle-treated animals in a dose-dependent manner in female rats. Female rats were used for the studies with Aβ measurements since avagacestat exposures in female rats were about 10-fold higher than in male rats. In particular, brain Aβ40 and Aβ42 levels were reduced by 40–50% at 2 mg/kg and by 75–85% at 10 mg/kg (Fig. 1A). Avagacestat also caused a dose-dependent decrease in CSF Aβ40 and Aβ42 levels similar to the effects in the brain (Fig. 1B). Levels of Aβ in both brain and CSF returned to pretreatment control levels within 50 hours after dosing with no significant increase above pretreatment control levels (Supplemental Table 2). Brain and CSF Aβ levels were highly correlated (Aβ40, r2 = 0.6445, P < 0.0001; Aβ42, r2 = 0.7410, P < 0.0001; Fig. 1C).
The relationship between plasma drug concentrations and the time course of Aβ response in both rat and dog brain and CSF was evaluated using an indirect response model (Dayneka et al., 1993; Sharma and Jusko, 1996; see Supplemental Materials and Methods). The model recapitulates the observed reductions in Aβ levels reasonably well. The goodness-of-fit was determined by visual inspection, Akaike information criterion, Schwartz criterion, examination of the residuals, and the coefficient of variation of the parameter estimates. The EC50 values relating total plasma drug concentrations and Aβ levels were estimated using this model (Table 1). These results show that the pharmacodynamic effects of avagacestat on Aβ40 and Aβ42 levels were similar in rat brain and CSF. Moreover, EC50 values for brain and CSF Aβ40 levels in dogs indicate similar cross-species acute effects of avagacestat on central Aβ40 levels (Table 1). The EC50 values from these models (114–206 ng/ml) were higher than the cellular IC50 values (0.14–3.5 ng/ml). The difference between the cellular IC50 and in vivo EC50 values likely results from the effect of substrate levels on absolute potency values and protein binding.
Response to Subchronic Dosing.
In three studies, female rats were administered 3, 10, 30, 100, or 300 mg/kg per day for 4 days. Reductions in area between baseline and effect curve (ABEC) for brain Aβ40 were estimated to be 49–57% at 3 mg/kg per day and >85% at doses of 10 mg/kg per day or higher (Supplemental Table 3). When dogs were administered doses of 0.5, 2, and 10 mg/kg per day for 14 days, the brain Aβ40 values at 3–7 hours after the last dose were similar to those in previous experiments (Gillman et al., 2010), with significant Aβ40 reductions at all doses (28–80%) (Supplemental Table 4). Similarly, in 3-month and 6-month repeat-dose studies in dogs, significant decreases in brain Aβ40 (20–56% reductions) were seen 4–6 hours after that last dose at all doses tested (0.3, 1, and 3 mg/kg per day) (Supplemental Table 4).
Dose-Dependent Effects of Avagacestat on Notch-Related Toxicities in Rats and Dogs
At the end of each subchronic dosing study, GI toxicity was assessed by microscopic evaluation of goblet cell metaplasia in GI tissue. In the 4-day study, histologic evaluation of the duodenum from rats showed that none of the animals dosed at 3, 10, or 30 mg/kg per day had evidence of goblet cell metaplasia. In contrast, 5 of 7 rats dosed at 100 mg/kg per day and 3 of 4 rats dosed at 300 mg/kg per day had mild goblet cell metaplasia (Supplemental Table 3). In a 1-month repeat-dose study, male and female rats were given avagacestat at 3, 30, or 100 mg/kg per day. In female rats, no effects were observed at 3 mg/kg per day, but 5 of 10 and 7 of 10 animals at 30 and 100 mg/kg per day, respectively, had minimal goblet cell metaplasia (Supplemental Table 5). Goblet cell metaplasia was found at lower doses and drug levels after 1 month of dosing than after 4 days of dosing and will be discussed later. After a 1-month post-dose recovery period, 0 of 5 female rats at each dose had goblet cell metaplasia, indicating a reversal of the effect. Male rats at all doses did not show goblet cell metaplasia, consistent with much lower plasma drug concentrations in male versus female rats.
Subsequently, we evaluated the effects of avagacestat on goblet cell metaplasia in GI tissue in dogs. In dogs treated for 14 days, no GI abnormalities were found in the lower dose groups, whereas 3 of 5 dogs in the 10 mg/kg per day group had duodenal goblet cell metaplasia (Supplemental Table 4). Dosing over 3 and 6 months showed that 2 of 14 dogs at 3 mg/kg per day, but none of the dogs at 0.3 or 1 mg/kg per day, had goblet cell metaplasia (Supplemental Table 4). After a 2-month post-dose recovery period, 0 of 4 dogs had goblet cell metaplasia, consistent with the reversal effect seen in rats. In both rats and dogs, the goblet cell metaplasia observed was similar to that seen in other studies in which Notch activity was reduced or eliminated, either by GSIs or by genetic manipulations (Milano et al., 2004; Wong et al., 2004; Riccio et al., 2008). In these studies, goblet cell metaplasia preceded or was accompanied by the development of other changes in the GI tract, which are more variable and generally indicative of GI mucosal disruption and abnormal function.
Plasma drug exposure (as measured by AUC) was strongly related to goblet cell metaplasia in dogs and rats, with an apparent threshold drug level required to cause GI toxicity (Fig. 2A). To determine this threshold level, a discriminant analysis method was used to differentiate between animals affected by toxicity from those not affected. A discriminant analysis finds optimal cut points between groups that best differentiate the corresponding populations. In the case of the dogs, individual AUC values were used to predict group membership (toxicity affected/not affected). In the case of the rats, composite AUC values obtained from different sets of animals were used to predict group membership. Through the construction of a Mahalanobis distance statistic (Mahalanobis, 1936; Gnanadesikan and Kettenring, 1972), the distance of an observation from each group mean was calculated, taking into account the spread of the observations in each group. An optimal cut point was determined by assigning an observation to a group to which it had the smallest (Mahalanobis) distance. Prior to analysis, the AUC data were first converted into the log10 scale. After the analysis was completed, the results were transformed back to the original scale. This was done to meet assumptions of the statistical methods used. Using this analysis, we determined that the optimal AUC cut points separating the animals affected by GI toxicity from those unaffected were 31,623 ng•h/ml and 8710 ng•h/ml in dogs and rats, respectively (Fig. 3, A and C); cut point values correspond to the drug exposures that best define the threshold for toxicity. These threshold values were used to calculate therapeutic index values of 24 and 7 for a 25% brain Aβ40 ABEC reduction in rats and dogs, respectively, and values of 32 and 7 for a maximum reduction of 25% in brain Aβ40 levels in rats and dogs, respectively (Table 1). The plasma avagacestat AUC values targeting both 25% ABEC reduction and 25% maximum reduction were determined using the indirect response model for each species. The therapeutic index values of 7–32 were within the range of in vitro Notch/APP IC50 values (3–193).
The effect of avagacestat on the spleen was examined in subchronic dosing studies in rats and dogs (Supplemental Tables 4 and 5). In dogs, plasma drug concentrations were strongly related to splenic, marginal-zone lymphoid depletion, with an apparent threshold of 5012 ng•h/ml (Figs. 2B and 3B). The data from the 1-month dosing study in rats were consistent with a similar threshold, although data in the threshold region were too sparse for an accurate estimate. The composite AUC cut point separating affected rats from unaffected rats was ~3376 ng•h/ml (Fig. 3D). These threshold values yielded therapeutic index values of 3 and 4 for a 25% brain Aβ40 ABEC reduction and a maximum reduction of 25% in brain Aβ40 levels in rats and dogs, respectively (Table 1).
For both GI and spleen toxicity, the cut points determined for rats were based on composite AUCs obtained from groups of unequal sizes and are further limited by the small numbers of animals that experienced toxicity. Thus, cut points calculated for the 2 rat data sets may be less reliable than those for dogs.
Avagacestat caused a concentration-dependent decrease in circulating B cells in dogs (Fig. 2C). The plasma drug concentrations required to decrease B cells below normal levels, defined as 2 standard deviations from the mean normal level (543 cells/μl, S.D. = 272, range = 21–1192, N = 20), were 3000–4000 ng•h/ml (similar to the concentration that causes splenic pathology). To further evaluate the effects of avagacestat on Notch signaling, we determined changes in HES-1, a transcription factor whose expression is regulated by Notch intracellular domain (Ang and Tergaonkar, 2007). Avagacestat caused a concentration-dependent decrease in HES-1 mRNA levels in blood from rats and dogs (Fig. 2D). HES-1 reductions were related to the plasma drug concentrations of avagacestat, with EC50 values of 7000 and 40,000 ng•h/ml in dogs and rats, respectively. Other potential Notch-dependent effects were also monitored in subchronic dosing studies in rats and dogs. No changes in hair color, skin, or neuronal viability as assessed by histopathology were observed following avagacestat administration.
Effects of Avagacestat on CSF Aβ Levels and Notch Signaling in Healthy Human Subjects
Single- and multiple-ascending–dose studies in healthy human subjects showed that avagacestat has a good tolerability profile, with good pharmacokinetic properties (Dockens et al., 2012; Tong et al., 2012a). In particular, plasma levels of avagacestat exhibited linear pharmacokinetics up to a 200-mg single dose (and 150 mg given as multiple doses), with a Tmax ranging from 1 to 2 hours, a typical biphasic profile, and a terminal half-life of approximately 40 hours (Tong et al., 2012a). Based on these results, doses were chosen to study the effect of avagacestat on CSF Aβ levels.
In a continuous CSF study, placebo or a single oral dose of 50 mg or 200 mg avagacestat was administered to human subjects with indwelling CSF catheters to allow repeated (i.e., hourly) sampling (Tong et al., 2012b). Several predose CSF samples were obtained, and the Aβ values in these samples were used to normalize Aβ measurements for each subject. CSF Aβ values in the placebo group increased steadily during sample collection. The mechanism responsible for this placebo-induced increase is unknown, but similar results have been reported by others using similar CSF collection techniques (Bateman et al., 2009). To obtain a conservative estimate of Aβ reduction, only Aβ values below baseline were considered. Subjects dosed with 200 mg had median CSF Aβ40 concentrations at 63% of baseline at 12 hours post dose. The median time of maximal inhibition was 16 hours at 200 mg, and Aβ40 values remained below baseline 24 hours after dosing. For the 50-mg dose, the corresponding 12-hour CSF Aβ40 levels (percentage of baseline) varied considerably and were within a range similar to those of the placebo group. Median time of maximal inhibition was 11 hours at 50 mg.
Response to Subchronic Dosing.
In a 28-day, multiple-ascending–dose study, Ctrough CSF Aβ levels in young healthy men, taken immediately before the last daily dose, were compared with the CSF Aβ levels prior to dosing (Dockens et al., 2012). This analysis showed that avagacestat caused dose-dependent reductions in Ctrough CSF Aβ levels (Fig. 4A). In particular, the 100- and 150-mg doses caused approximately 30 and 60% reductions in CSF Aβ levels, respectively. Importantly, CSF Aβ40 and Aβ42 responses were similar. The mean concentrations of CSF Aβ40 and Aβ42 in the 15-mg and 50-mg dose groups were similar to the baseline levels, consistent with reduced drug levels in these dose groups at the Ctrough sampling time. These results were consistent with acute response observed in the single-ascending–dose study in which CSF Aβ40 concentrations returned to baseline prior to the end of the dosing period.
We further investigated the relationship between plasma drug concentrations at trough (i.e., 24 hours after dosing, day 28) and CSF Aβ40 levels from the same time point in humans. As shown in Fig. 4B, there was a close relationship between plasma drug concentrations and CSF Aβ40. Data collected 24 hours after 1 or multiple doses to rats and dogs were overlaid with the human data (Fig. 4C), showing that the pharmacodynamic effects of avagacestat on CSF Aβ levels are similar between rats, dogs, and humans. In contrast to findings in preclinical species, we did not detect significant decreases in the levels of B cells or HES-1 mRNA in humans after 28 days of dosing at exposures of up to 10,000 ng•h/ml (Fig. 5). GI adverse events (AEs), which will be described in detail elsewhere, occurred at low frequency in young, healthy subjects. Most of the AEs, with the exception of mouth ulcer and flatulence, occurred in one subject.
Several potent, small-molecule GSIs have been developed (Olson and Albright, 2008; Wu and Zhang, 2009). Structure-function studies have shown that GSIs selected for in vivo use, the azepines and sulfonamides, bind a common allosteric inhibitory site on presenilin (PS), rather than the site targeted by GSIs in the isostere class and are believed to interact with the active site of the enzyme (Olson and Albright, 2008; Steiner et al., 2008). Developing an effective and tolerable AD therapy that avoids negatively affecting γ-secretase substrates (other than APP) remains a challenge for GSIs. Evidence suggests that Notch is one of the most important among these off-target substrates because of its critical role in self-renewing cell differentiation in adult tissues (Saito et al., 2003; Milano et al., 2004; Radtke et al., 2004; Wong et al., 2004; Pollack and Lewis, 2005; Li et al., 2007a; Prasad et al., 2007; Kumano et al., 2008; Riccio et al., 2008; Kopan and Ilagan, 2009;). The binding of azepine and sulfonamide GSIs with an allosteric site on PS, rather than the catalytic site, raises the possibility of differential inhibitory interactions as PS processes different substrates, such as APP and Notch. Consistent with this hypothesis, in vitro studies have shown varying substrate differentiation for GSIs in the allosteric class, but not for GSIs in the isostere class (Lewis et al., 2003; Kreft et al., 2008; Olson and Albright, 2008; Martone et al., 2009; Basi et al., 2010; Chávez-Guttiérez et al., 2012; Mitani et al., 2012). One GSI, LY-411575, was shown to reduce Aβ40 and Aβ42 levels in plasma and in the brains of CRND8 transgenic mice while causing intestinal goblet metaplasia and thymic atrophy (Wong et al., 2004). The most clinically advanced of the GSIs was semagacestat (LY-450139) (Henley et al., 2009). In a phase II study of patients with AD, semagacestat 30 mg per day for 1 week followed by 40 mg per day for 5 weeks significantly reduced plasma Aβ levels but not CSF Aβ levels (Siemers et al., 2006). Higher doses were then evaluated in AD patients (100 and 140 mg), but AEs affecting skin and hair occurred and CSF Aβ levels were not significantly affected (Fleisher et al., 2008); the drug was advanced to phase III trials. However, two 21-month phase III trials of semagacestat in patients with mild-to-moderate AD were terminated early (Imbimbo et al., 2011) due to its detrimental effects on clinical measures of cognition and activities of daily living. Semagacestat was also associated with increased rates of skin cancer, presumably related to “mechanism-based toxicities” (Imbimbo et al., 2011). Some investigators have proposed that both the toxicities and effects on cognition were related to interference with Notch signaling (Ross and Imbimbo, 2010).
The results of the semagacestat trials underscore the importance of APP selectivity for a GSI ultimately used in AD patients. Our previous studies demonstrated that avagacestat is a highly potent inhibitor of Aβ40 and Aβ42 production in cell culture, with IC50 values of ~0.3 nM (Gillman et al., 2010). In the present study, we show that avagacestat has IC50 values of 41, 10, 33, and 38 nM for signaling inhibition of the human Notch1–Notch4 proteins, respectively. The corresponding Notch/APP selectivity ratios were 137, 33, 117, and 127. Although other groups have reported lower Notch/APP ratios for avagacestat, relative differences in the ratio between GSIs were maintained. For example, avagacestat exhibits Notch/APP IC50 ratios that are consistently 15- to 43-fold greater than the corresponding ratios reported for semagacestat across multiple studies (Supplemental Table 1) (Martone et al., 2009; Gillman et al., 2010; Chávez-Guttiérez et al., 2012; Mitani et al., 2012). Thus, GSIs can exhibit intrinsic and significant differences with respect to Notch/APP selectivity. Furthermore, avagacestat is one order of magnitude or more selective than semagacestat.
Avagacestat selectivity for APP cleavage and its favorable pharmacokinetic properties support its use as an in vivo Aβ-lowering compound. In our previous studies, we showed that intravenous dosing of avagacestat in dogs resulted in low total body clearance (4.20 ml/min/kg), high volume of distribution at steady state (3.87 l/kg), and an oral bioavailability of 42% (Gillman et al., 2010). Furthermore, dosing achieved a high brain-to-plasma concentration ratio (>1), demonstrating good brain penetrance; the calculated half-life of 11.4 hours and good oral bioavailability supported daily dosing of the drug (Gillman et al., 2010). Here, we showed that acute oral dosing of avagacestat in rats caused a rapid and robust decrease in brain and CSF Aβ levels, which were highly correlated and dose dependent (Fig. 1; Table 1). Chronic dosing studies with daily oral administration of avagacestat in rats and dogs also showed dose-dependent effects, and plasma concentrations were correlated with reductions in central Aβ levels (Fig. 4C). In rats and dogs, the estimated EC50 values relating plasma drug concentration and central Aβ responses ranged from 105 to 206 ng/ml.
Development of GSIs as AD therapies has been limited by deleterious Notch-related effects on the GI tract, spleen, immune system, skin, and cognition (Milano et al., 2004; Wong et al., 2004; Prasad et al., 2007; Kumano et al., 2008; Ross and Imbimbo, 2010; Imbimbo et al., 2011). However, studies in transgenic mouse models show that modest decreases in Aβ production by GSIs (15–30%) can reverse mutant APP-induced cognitive and synaptic deficits (Comery et al., 2005; Martone et al., 2009). Likewise, a recently identified mutation in APP was shown to have a protective effect against AD and cognitive impairment while causing a 40% decrease in Aβ formation (Jonsson et al., 2012). By comparison, a 30% reduction of γ-secretase activity in mice ameliorated amyloid burden but had no adverse effects on cognition (Li et al., 2007b). Studies of partially inactivated γ-secretase complexes suggest that in mice, the threshold for GI and spleen toxicity and skin tumors corresponds to a 30–50% reduction in γ-secretase activity (Li et al., 2007b). These findings suggest that Notch-related toxicities may be threshold-driven, and that GSIs with better selectivity for APP over Notch cleavage may be able to lower central Aβ levels enough to provide cognitive benefits without causing toxicities related to Notch-signaling interference.
Consistent with the murine Notch-toxicity thresholds, our studies in rats and dogs indicated that avagacestat caused threshold-dependent goblet cell metaplasia in GI tissue, marginal-zone lymphoid depletion in the spleen, and reductions in B cells. A reduction in the threshold for goblet cell metaplasia was observed when dosing was extended from 4 days to 1 month in rats, likely resulting from the system not being at steady state after 4 days with respect to both changes in the GI tissue and drug levels. However, thresholds for Notch toxicity in dogs were indistinguishable in the 14-day, 3-month, and 6-month dosing studies. The differences in the threshold for Notch toxicity between rats and dogs and between GI tissue and spleen are not understood.
Importantly, our studies suggest that Aβ levels can be significantly reduced without measurable toxicity. Chronic dosing studies in dogs identified doses of avagacestat that can achieve a ≤30% reduction in central Aβ levels without putative Notch-related toxicities in the GI tract or spleen (Supplemental Table 4). Avagacestat exhibited a therapeutic margin between Notch-dependent toxicities and Aβ reductions in rats and dogs (Table 1). In rats and dogs, avagacestat has a 24- and 7-fold margin, respectively, between a 25% reduction in brain Aβ ABEC and threshold levels for GI toxicity. We further observed a 3- and 4-fold margin, respectively, between a 25% reduction in brain Aβ ABEC and the threshold levels for spleen and B cell changes.
Notably, avagacestat may also have a therapeutic margin in humans. Evaluation of the pharmacokinetic–pharmacodynamic properties of avagacestat was extended to single- and multiple-ascending dose studies in healthy human subjects. We showed that administration of avagacestat to human subjects decreased CSF Aβ levels at Ctrough in a dose-dependent manner (more than 50% at higher doses). In contrast to the biphasic response of plasma Aβ to avagacestat in humans (Dockens et al., 2012; Tong et al., 2012a) and animals (Gillman et al., 2010), human CSF Aβ had a typical monophasic concentration–response relationship. It is likely that the monophasic response of CSF Aβ and brain Aβ to avagacestat reflects the higher substrate levels in the brain compared with the peripheral tissues that contribute to plasma Aβ given the effect of substrate levels on Aβ response (Burton et al., 2008). The AEs observed in healthy subjects treated with avagacestat included low-frequency GI AEs (Dockens et al., 2012). While no changes in lymphocytes were observed following 28 days of dosing in humans at exposures that caused spleen changes in rats and dogs, it is possible that 28 days were not enough to observe such changes in humans. Unlike Notch-dependent GI toxicity, the effects of Notch-dependent hematopoietic toxicity on immune function are unknown and will need to be carefully monitored if such abnormalities are observed in humans. Subsequent studies in patients with mild-to-moderate AD showed that the 100- and 125-mg doses were not well tolerated due to GI and dermatologic AEs and trends for worsening of cognition (Coric et al., 2012). In contrast, the 25- and 50-mg doses had dose-dependent effects on CSF Aβ species in an exploratory analysis plus acceptable safety and tolerability to support future studies.
The authors thank Wendy Clarke, Jason Corsa, Harley Ferguson, Tracey Fiedler, Margi Goldstein, Nina Hoque, Larry Iben, Cathy Kieras, Carol Krause, Alan Lin, Maria Peirdomenico, Ding Ren Shen, Sarah J. Taylor, Mark Thompson, Sam Varma, Allen Wong, and Victoria Wong.
Participated in research design: Albright, Barten, Berman, Bronson, Burton, Castaneda, Coric, Denton, Gillman, Guss, Houston, Huang, Lentz, Macor, Meredith, Olson, Pilcher, Polson, Rhyne, Sankaranarayanan, Slemmon, Starrett, Tong, Zaczek.
Conducted experiments: Barten, Coric, Denton, Dockens, Gillman, Guss, Lentz, Olson, Pilcher, Polson, Rhyne, Slemmon, Starrett, Tong, Toyn, Wang.
Contributed new reagents or analytic tools: Denton, Gillman, Olson, Pilcher, Rhyne, Slemmon.
Performed data analysis: Denton, Dockens, Feldman, Gillman, Guss, Huang, Lentz, Meredith, Olson, Pilcher, Polson, Raybon, Rhyne, Sanderson, Sankaranarayanan, Simutis, Starrett, Sverdlov, Tong, Wang, White.
Wrote or contributed to writing of the manuscript: Albright, Barten, Berman, Coric, Denton, Dockens, Feldman, Gillman, Guss, Houston, Huang, Lentz, Meredith, Olson, Pilcher, Polson, Raybon, Rhyne, Sanderson, Simutis, Sverdlov, Tong, Wang, White.
This work was funded by Bristol-Myers Squibb. Editorial and writing assistance was provided by StemScientific and Oxford PharmaGenesis and was funded by Bristol-Myers Squibb. All of the authors were employees of Bristol-Myers Squibb at the time when this work was done; many of the authors own in excess of $10,000 of company stock. Additionally, some authors are inventors on patents related to the subject matter.
- area between baseline and effect curve
- amyloid β-peptide
- Alzheimer’s disease
- adverse events
- amyloid precursor protein
- area under the concentration-time curve
- cerebrospinal fluid
- γ-secretase inhibitors
- hairy and enhancer of split homolog-1
- Received September 20, 2012.
- Accepted December 27, 2012.
- Copyright © 2013 by The American Society for Pharmacology and Experimental Therapeutics