Comment by Sascha Weggen and Stefanie Hahn
To understand how presenilin mutations affect γ-secretase modulators (GSMs), it's worth elaborating on the differences in the studies by Hahn et al. (Hahn et al., 2011) and Kretner et al. (Kretner et al., 2011). Both studies used similar compounds. Both used E-2012 as an example of a potent non-acidic GSM without a carboxylic acid group. In addition, Kretner et al. used GSM-1 as a potent acidic GSM, whereas we used BB25, which is a close analogue of GSM-1 with similar potency. Kretner et al. concluded from their results that the attenuating effects of most presenilin mutations could be overcome by second-generation GSMs. In contrast, Hahn et al. concluded the opposite: that the efficacy of second-generation GSMs is reduced by presenilin mutations, in a similar fashion to what was previously reported by both of our groups for the less potent NSAID-type GSMs such as sulindac sulfide (Czirr et...  Read more

Comment by Sascha Weggen and Stefanie Hahn
To understand how presenilin mutations affect γ-secretase modulators (GSMs), it's worth elaborating on the differences in the studies by Hahn et al. (Hahn et al., 2011) and Kretner et al. (Kretner et al., 2011). Both studies used similar compounds. Both used E-2012 as an example of a potent non-acidic GSM without a carboxylic acid group. In addition, Kretner et al. used GSM-1 as a potent acidic GSM, whereas we used BB25, which is a close analogue of GSM-1 with similar potency. Kretner et al. concluded from their results that the attenuating effects of most presenilin mutations could be overcome by second-generation GSMs. In contrast, Hahn et al. concluded the opposite: that the efficacy of second-generation GSMs is reduced by presenilin mutations, in a similar fashion to what was previously reported by both of our groups for the less potent NSAID-type GSMs such as sulindac sulfide (Czirr et al., 2007; Page et al., 2008). The different outcomes of the two studies appear to be due to differences in the compound concentrations used and the statistical analysis.

Kretner et al. used higher GSM concentrations. We conducted our experiments in a dose-dependent fashion with a wide range of concentrations from 50-1,000 nM for E-2012 and BB25, whereas Kretner et al. used 1 μM in most experiments. The IC50 of these compounds in cell-based assays is around 50-200 nM. Our experience is that, if you go high enough with the compound concentration, one can break almost every presenilin mutation. However, we do not think that these concentrations are a realistic approximation of the in-vivo situation. A general problem with current GSM structures is that they are highly lipophilic. This means that high doses will be required to reach effective free drug concentrations in vivo. In this respect, concentrations five- to 20-fold above the in-vitro IC50 seem overly optimistic.

We are also not convinced that second-generation GSMs are principally different from the older NSAID-type GSMs. Kretner et al. suggested this because they found that many more mutants did not respond to sulindac sulfide as compared to GSM-1. However, they used sulindac sulfide at 50 μM, which is around its IC50 in cell-based assays, and GSM-1 at 1 μM, which is five times its IC50. Higher sulindac sulfide concentrations are very toxic on cells. Therefore, it is simply not possible to do the experiment, but if one could go high enough with the sulindac sulfide concentrations without inducing toxicity, we would predict sulindac sulfide would behave exactly the same as the new GSMs.

More important are the differences in the statistical analyses between our studies. Kretner et al. tested for significance of drug treatment versus vehicle control in each cell line separately. In many mutant cell lines, they observed a significant drop in Aβ42 levels after drug treatment and concluded that the mutants do respond to the drug. In contrast, we tested (at a given drug concentration) for a significant difference between the Aβ42 reductions in cell lines expressing mutant presenilin versus a control cell line expressing wild-type PS1. In this way, one can observe a significant drug effect with presenilin mutants, but the effect is still less in comparison to wild-type presenilin.

In conclusion, our studies seem to report contradictory results for similar or identical compounds, but these differences can be explained by the experimental design and the statistical analysis. So what is the truth? As an example, imagine treatment of a sporadic AD patient and a PS1 FAD patient with a GSM. Let's say the wild-type patient shows a 35 percent drop in CSF Aβ42 after GSM treatment. Most companies would likely be pleased with such an effect size. Now, in line with the data in both of our studies, the presenilin mutation in the FAD patient might attenuate the drug effect and cause only a 20 percent reduction in Aβ42. If you do a statistical analysis in the FAD patient of treatment versus non-treatment, a 20 percent drop is likely statistically significant. So is the 15 percent difference between treatment efficacy in the wild-type and the mutant patient. In this sense, the statistics in both studies are correct. However, although it may be possible to significantly lower Aβ42 in the FAD patient, the 15 percent difference in Aβ42 reduction could have profound implications for the clinical efficacy of the compound. Obviously, we do not know whether FAD patients with presenilin mutations will ever be treated with a GSM in the future, and this thought experiment might be a purely academic exercise. Nevertheless, these two papers nicely illustrate how differences in statistics can lead to completely different interpretations of the data. By just looking at the experimental raw data, there is very little difference between our studies.

View all comments by Sascha Weggen