It is positive that Cruchaga et al. support the association between the poly-T repeat (“523”) and risk of late-onset Alzheimer’s disease (LOAD) that we first reported in 2009. Additionally, this group completed a phylogenetic analysis of the ApoE-Tomm40 region that effectively confirmed the evolutionary-relatedness of the ApoE-523 haplotypes and allele frequencies in Caucasians, again reported by us in 2009. The two findings that differ from our original report are 1) longer alleles of 523 are associated with later onset and a protective effect and 2) lack of association between the 523 polymorphism and age of LOAD onset. Both are explainable by data not defined in the manuscript.

1. We have reported (Hayden, K: AAIC meeting 2011) that there is a biphasic response associated with the 523-very long allele, and indeed have observed, for several clinical series, that very long (VL)/VL carriers can have a very late onset (>80 years of age). It is also observed that the very young onset subjects with late-onset Alzheimer's disease (  Read more

It is positive that Cruchaga et al. support the association between the poly-T repeat (“523”) and risk of late-onset Alzheimer’s disease (LOAD) that we first reported in 2009. Additionally, this group completed a phylogenetic analysis of the ApoE-Tomm40 region that effectively confirmed the evolutionary-relatedness of the ApoE-523 haplotypes and allele frequencies in Caucasians, again reported by us in 2009. The two findings that differ from our original report are 1) longer alleles of 523 are associated with later onset and a protective effect and 2) lack of association between the 523 polymorphism and age of LOAD onset. Both are explainable by data not defined in the manuscript.

1. We have reported (Hayden, K: AAIC meeting 2011) that there is a biphasic response associated with the 523-very long allele, and indeed have observed, for several clinical series, that very long (VL)/VL carriers can have a very late onset (>80 years of age). It is also observed that the very young onset subjects with late-onset Alzheimer's disease (<60 years of age) are VL/VL carriers, not long/long (L/L), as would be "expected" (our data and Bekris: AAIC 2011). This is a complex genetic phenomenon that we are investigating by doing appropriate sequencing and variant identification. We are not surprised that this group reported a different direction for the VL/VL genotype since we have observed this same phenomenon. We cannot explain why, for the ApoE3/4 group (supplementary figure S4), the short/long (S/L) carriers appear to develop LOAD earlier than the VL/L; this is in direct contrast to robust results that we have reported in the original paper and that have been reported by Caselli et al. (AAICD 2010). However, as the authors observe, there is no statistically significant difference between the curves, and, indeed, by inspection they are very close and not distinguishable. There are two major data points involved; the marker itself is accurate but the accuracy of age-of-onset ascertainment has a long history of differences between populations observed and followed prospectively and retrospectively. In our first paper we were careful to define the age of onset in the series. It is not described how Cruchaga et al. scored these very important data points in the series of AD patients they collected (see below).

2. Cruchaga reports that they did not find a relationship between “523” and age of onset. This stands in opposition to the robust results that we and others (Caselli) have reported from longitudinal studies of LOAD documenting earliest age of cognitive impairment. It is a shame that a longitudinal series of subjects defined by earliest neuropsychiatric data were not isolated from the dataset and scored for age of onset of cognitive impairment—especially since this center has such a long longitudinal experience. The authors attempted to re-create an analysis similar to a prospective analysis by including controls in the survival analysis (panel B of figure). However, there is a major difference between running a longitudinal study over time where you track events, record onset, and censor observations where events have not occurred, and retrospectively constructing a sample from cases and controls and placing them into a survival analysis. This may explain some of the discrepancy between the results, at least with respect to the survival analysis. However, the major difference is likely due to determination of the age of onset, especially in the ADNI cases (see below). Details are not provided in the paper as to how the age of onset was ascertained (self-reported, reported by physician as part of medical record, or reported by caregiver?), and age of onset of LOAD is recognized to be difficult to determine accurately.

3. When we analyzed the ADNI data made available to us, we came to a similar conclusion about the relationship between “523” and age of onset—effectively, we also did not find a statistically significant relationship between these variables. However, we also did not find a statistically significant difference in age of onset between ApoE3/3 and ApoE3/4 individuals in this dataset either, which stands in contrast to over 18 years of experience with ApoE (after the first report by Roses and Saunders) and LOAD. Looking over how these cases were ascertained, and by what criteria age of onset were scored, we felt that we could not compare or report the ADNI data. Our collaborators confirmed the non-uniformity of the age of onset (of what, it is not clear...cognitive impairment, diagnosis, etc.?) in this series collected from more than 50 centers for a totally different purpose.

4. The most accurate series for determining the age of onset of cognitive impairment of the Alzheimer's type is by prospective, serial testing of cognitively normal individuals. In prospective series followed less than eight to 10 years, like that reported by Caselli et al., it is unclear what proportion of cognitively impaired patients will go on to defined AD (as collected by Cruchaga et al.). The rate of progression to AD and the proportion affected over time is currently being monitored in several large prospective collections. Retrospective cohort series are also being investigated, but we judge these analyses to be less robust, as accurate neuropsychiatric data at a given age is often not the first characteristic scored in these series.

I suspect these same observations will apply to other AD case-control series that have variable ascertainment and are retrospective. To be clear, we have never claimed that the 523 data, or the ApoE4 data, should be used for the diagnosis of AD. We have published these data to support the use of the data for pharmacogenetic enrichment of a high-risk population selected from a geographically defined area who are neuropsychiatric tested at entry and found to be normal by currently accepted standards. This is not an "AD study," but a study of prognostic prediction that will be qualified by the clinical trial population. I believe there are valid reasons why there are differences between the results of Cruchaga et al. and ours. Since the delay of the onset of cognitive impairment study that will be done, with regulatory oversight already in progress, claims of non-confirmation will be viewed with the same degree of interest as that of the lack of confirmation of ApoE in the first years after we published almost 20 years ago. It is an interesting note that several of the overt ApoE skeptics of 1993-1995 were also the same negative voices heard at ICAD 2011. As Yogi Berra, legendary catcher, and later manager of the New York Yankees, commented: "It's like déjà vu all over again!"

View all comments by Allen Roses

Allen Roses referred to my work, so I'd like to clarify that in a longitudinal study of more than 600 cognitively normal individuals, we found a difference between the VL/VL and S/S variants among ApoE3/3 individuals prior to age 60 (presented at the Alzheimer's Association International Conference last month; paper under review). The VL/VL group showed little evidence of a test-retest effect (on the long-term memory measure/trial 7 of the auditory verbal learning test) in our longitudinal study, and that is not normal. On the other hand, we did not find any evidence for accelerated decline after age 60. The results are therefore complex, and I am all in favor of additional clinical data with age of onset defined as clearly as possible. If there are cohorts of younger individuals that might be in a position to replicate (or refute) our findings, that too would be of obvious importance and (together with the new perspective of preclinical stage AD) underscores the need for inclusion of younger individuals in AD research. For example, might Tomm40 contribute to a cognitive...  Read more

Allen Roses referred to my work, so I'd like to clarify that in a longitudinal study of more than 600 cognitively normal individuals, we found a difference between the VL/VL and S/S variants among ApoE3/3 individuals prior to age 60 (presented at the Alzheimer's Association International Conference last month; paper under review). The VL/VL group showed little evidence of a test-retest effect (on the long-term memory measure/trial 7 of the auditory verbal learning test) in our longitudinal study, and that is not normal. On the other hand, we did not find any evidence for accelerated decline after age 60. The results are therefore complex, and I am all in favor of additional clinical data with age of onset defined as clearly as possible. If there are cohorts of younger individuals that might be in a position to replicate (or refute) our findings, that too would be of obvious importance and (together with the new perspective of preclinical stage AD) underscores the need for inclusion of younger individuals in AD research. For example, might Tomm40 contribute to a cognitive phenotype that fosters earlier clinical recognition even in the absence of actual Tomm40-mediated accelerated decline? This is an example of the kind of new model we may need to consider, though certainly not the only possibility.

View all comments by Richard Caselli