Pharmacological Agents Targeting γ-Secretase Increase Risk of Cancer and Cognitive Decline in Alzheimer’s Disease Patients: A Systematic Review and Meta-Analysis
INTRODUCTION
Alzheimer’s disease (AD) represents a significant social and economic burden, and it has been forecast that its prevalence will quadruple by 2050, with 43% of cases requiring nursing home or equivalent high- level care [1]. The development of therapeutics that could reduce or eliminate disability would be of great benefit to society, and indeed is imperative to control- ling ballooning costs of healthcare in many Western countries with aging populations.
γ-secretase is a complex that completes the pro- cessing of amyloid-β precursor protein (AβPP), generating amyloid-β (Aβ) peptides of variable length [2]. Aβ1-40 and Aβ1-42 are both found in plaques, but Aβ1-42 has a greater tendency to oligomerize and form extracellular amyloid plaques in the brain, and is therefore considered more pathogenic. The goal of γ-secretase inhibition or modulation is to reduce the concentration of Aβ [3]. This approach stems from the amyloid cascade hypothesis, in which the accumulation of extracellu- lar amyloid plaques is the central early event in AD pathogenesis [4, 5].
However, the results of large- scale clinical trials of drugs targeting γ-secretase have been disappointing. Semagacestat (LY450139) is a γ-secretase inhibitor that was shown in early development to transiently reduce plasma Aβ40 after a single dose [6–8]. Phase II trials failed to show improvement in cognitive measures, while also being associated with increased gastrointestinal and dermatological adverse events in the treatment groups [9, 10].
A further Phase III trial found that high doses of semagacestat caused a clinical worsening in functional ability, and that the drug was associated with increased risk of adverse events including skin cancers and infections [11, 12].
γ-secretase is necessary for the enzymatic cleav- age of a number of other proteins, including the transmembrane receptor Notch, which is part of a crit- ical cell signaling pathway. Inhibition of Notch was believed to be responsible for gastrointestinal and der- matological toxicity in trials of semagacestat. It was posited by the authors that the adverse events may have been due to Notch inhibition [11]. Therefore, it was hoped that Notch-sparing γ-secretase inhibitors may avoid these adverse events.
Avagacestat (BMS-708169) is a Notch-sparing arylsulfonamide γ-secretase inhibitor [13]. Phase I trials involving the administration of single doses to healthy volunteers found the drug to be tolerable and to reduce Aβ1-40 concentrations [14, 15]; however, higher doses of 100 mg and 125 mg were poorly tolerated and caused cognitive deterioration in a sub- sequent Phase II study [16].
In a recently published Phase II trial, avagacestat use was associated with higher rates of serious adverse events, non-melanoma skin cancers, and cognitive deterioration [17]. Tarenflurbil (MPC-7869; R-flurbiprofen; Fluri- zanTM) is a non-steroidal anti-inflammatory drug (NSAID) R-enantiomer that modulates γ-secretase and selectively reduces Aβ42 without significantly affecting Aβ40 [18, 19]. While Aβ40 is more abun- dant, Aβ42 is more prone to oligomerization and hence is the principal component of amyloid plaques in the brain in AD [20].
Despite promising data in Phase I and II studies [21, 22], in a large Phase III trial tarenflurbil was unable to demonstrate efficacy in pri- mary cognitive outcomes, and significantly worsened Clinical Dementia Rating – Sum of Boxes (CDR-sb) scores [23].
Although previous large-scale clinical trials have demonstrated adverse events and a lack of efficacy associated with these drugs individually, it remains unclear whether these effects are generalizable to all drugs targeting γ-secretase. The purpose of this meta-analysis is to evaluate the safety and efficacy of γ-secretase inhibitors and modulators in AD from the results of randomized clinical trials (RCTs), and to perform a subgroup analysis including the γ- secretase inhibitors avagacestat and semagacestat, and excluding the modulator tarenflurbil.
MATERIALS AND METHODS
Study protocol
We followed the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines [24]. A systematic search of the databases MEDLINE (from 1950), PubMed (from 1946), EMBASE (from 1949), Google Scholar, the Cumu- lative Index to Nursing and Allied Health Literature (CINAHL), LILACS, ClinicalTrials.gov, and the Cochrane Central Register of Controlled Trials through to 15 July, 2016 was performed to identify relevant articles. The search used the terms ‘Alzheimer disease’ and ‘secretase inhibitor’, ‘semagacestat’, ‘LY450139’, ‘avagacestat’, ‘BMS- 708163’, ‘tarenflurbil’, ‘flurizan’, or ‘MPC-7869’, which were searched as text word and as exploded medical subject headings where possible. The ref- erence lists of relevant articles were also searched for appropriate studies. No language restrictions were used in either the search or study selection. A search for unpublished literature was not performed.
Study selection
We included studies that met the following inclu- sion criteria: (1) the drug used either directly inhibited or modulated γ-secretase; (2) the study reported adverse events as an odds ratio (OR), or the data was presented such that an OR could be calculated; (3) the 95% confidence interval (CI) was reported, or the data was presented such that the CI could be calculated; (4) the trial was prospective and placebo- controlled; and (5) the trial duration was at least two weeks. Additionally, for the comparison of cognitive measures, studies were only included if the individ- uals in the sample had probable AD. Studies with healthy volunteers were included for comparison of adverse effect outcomes only. We excluded studies that did not meet the inclusion criteria.
Data extraction
The data extraction was performed using a stan- dardized data extraction form, collecting information on the publication year, study design, number of cases, number of controls, total sample size, popu- lation type, country, mean age, number of adjusted variables, the risk estimates or data used to calcu- late the risk estimates, CIs or data used to calculate CIs, and the severity of dementia. Quality of the studies was assessed using the Jadad scale, though no studies were excluded on the basis of Jadad score [25].
Authors were not contacted for missing data. Adjusted ratios were extracted in preference to non-adjusted ratios, however, where ratios were not provided, unadjusted ORs and CIs were calculated. Where multiple dosage groups were included in a single trial, the results were pooled into a single treat- ment group. If necessary, the mean of the effect size between the dosage groups was taken for comparison to placebo.
Statistical analysis
Pooled ORs and 95% CIs were calculated for the effect of γ-secretase inhibitors and modulators on the risk of adverse events, serious adverse events, skin cancer, infections, and death using a random effects model [26]. A sub-group analysis was also performed, excluding drugs involving the γ-secretase modulator tarenflurbil, due to differences in the mechanism of action and Aβ products between γ-secretase inhibitors and modulators.
For cognitive assessments, a pooled mean change from baseline was calculated using a random effects model. We tested heterogeneity with Cochran’s Q statistic, with p < 0.10 indicating heterogeneity, and quantified the degree of heterogeneity using the I2 statistic, which represents the percentage of the total variability across studies which is due to heterogeneity. I2 values of 25, 50, and 75% corresponded to low, moder- ate, and high degrees of heterogeneity respectively [27]. We quantified publication bias using the Egger’s regression model [28], with the effect of bias assessed using the fail-safe number method. The fail-safe num- ber was the number of studies that we would need to have missed for our observed result to be nullified to statistical non-significance at the p < 0.05 level. Pub- lication bias is generally regarded as a concern if the fail-safe number is less than 5n+10, with n being the number of studies included in the meta-analysis [29]. All analyses were performed with Comprehensive Meta-analysis (version 2.0). RESULTS Study characteristics A broad search was performed, which identi- fied 8,301 citations for review. 8,257 trials were discarded, with many being narrative reviews, dupli- cates, animal studies, opinion pieces, editorials, or otherwise irrelevant. Overall, 44 abstracts were selected for full-text review. Of these, 34 were dis- carded, many being single-dose studies, case studies, conference abstracts without sufficient data for anal- ysis, results reported in multiple articles, or animal studies. Overall, 10 trials (n = 5,227) were selected for inclusion in the meta-analysis. All 10 trials were randomized controlled trials (Fig. 1). Of these studies, two were Phase I, five were Phase II, and three were Phase III. The study characteristics are summarized in Table 1. Note that the IDENTITY-2 trial results remain unpublished; however the data was published on ClinicalTrials.gov (NCT00762411) and hence this data was included in the pooled analysis. All tri- als reported at least basic adverse events data. Eight of ten studies provided detailed data on serious adverse events. Six trials provided sufficient data on the Alzheimer’s Disease Assessment Scale, cogni- tive component (ADAS-cog), Alzheimer’s Disease Cooperative Study – Activities of Daily Living Inven- tory (ADCS-ADL), and Clinical Dementia Rating, sum of boxes (CDR-sb) changes for pooled analy- sis. Four trials provided data for Mini-Mental State Examination (MMSE) changes that could be pooled. The methodological quality of RCTs was assessed using the Jadad scale, which gives a rating out of a total of 5 based on randomization, blinding, and reporting of withdrawals [25]. Six of the trials achieved the highest possible score of 5. Three trials [16, 17] scored 4, as while the trials were described as randomized, the method of randomized alloca- tion was not provided in detail. One of these trials was the unpublished IDENTITY-2, for which lim- ited methodological information was available. If IDENTITY-2 was carried out with a similar proto- col to the IDENTITY trial, then the score for that trial would be 5. One trial [31] was not described as double-blinded, and did not describe the method of randomization, and hence scored 2. No trial was excluded on the basis of the Jadad score. Quantitative data synthesis Pooling of data from all three drugs (Table 2) revealed significantly increased risk of adverse events (OR 1.38; p = 0.01), serious adverse events (Fig. 2B; OR 1.71; p < 0.001), and skin cancers (Fig. 2C; OR 4.77; p < 0.001). There was no significantly increased risk of infections or mortality. Upon pre-planned exclusion of studies (Table 3) involving tarenflur- bil, the pooled odds ratios for adverse events (OR 1.57; p < 0.001) and serious adverse events (OR 1.71; p < 0.001) increased. The risk of developing infections was found to be significantly increased (OR 1.36; p < 0.001), where this result had pre- viously not been significant (p = 0.05). While the pooled odds ratio for mortality remained not sta- tistically significant, there was a trend suggesting increased risk in the treatment group which was accentuated upon exclusion of tarenflurbil (OR 1.93; p = 0.14). Heterogeneity was not significant for seri- ous adverse events, infections, skin cancers, or mortality (p > 0.05), but was significant for adverse events both before and after exclusion of tarenflurbil (p < 0.001 and p = 0.008, respectively). With regards to cognitive outcomes, prior to exclusion of tarenflurbil studies significant cognitive worsening was present in ADAS-cog (Fig. 2A; dif- ference in means 1.33, 95% CI 0.58–2.08; p < 0.001) and MMSE (difference in means –0.66, 95% CI –0.96 to 0.35; p < 0.001) but not ADCS-ADL (difference in means –0.23, 95% CI –1.17 to 0.71; p = 0.64) or CDR- sb (difference in means –0.03, 95% CI –0.54 to 0.48; p = 0.91). In the subset of studies that excluded of tarenflurbil (Table 4), statistically significant worsen- ing of ADAS-cog (difference in means 0.85, 95% CI 0.41–1.29; p < 0.001), MMSE (difference in means –0.70, 95% CI –1.04 to –0.36; p < 0.001), and ADCS-ADL (difference in means –1.12, 95% CI –2.04 to –0.20; p = 0.02) was noted, with no change in CDR-sb (p = 0.91). Substantial heterogeneity was present in the pooled analysis of all cognitive out- comes, both before and after exclusion of tarenflurbil (p < 0.05).