We observed in a sample of Framingham Study participants a correlation of a cortical cataract score measured during adulthood with future development of AD, and with multiple measures of AD-related brain degeneration obtained from both MRI scan and cognitive testing nearly ten years later ( and Table S2
). Our unique GWAS identified genome-wide significant association of three intronic SNPs in CTNND2
with a bivariate outcome of cortical cataract and volume of the temporal horn, an MRI measurement that strongly correlates with AD progression 
. This association accounts for 19% of the heritable component of CC-THV. These SNPs were also significantly associated with the bivariate outcomes of cortical cataract and performance on several neuropsychological tests including the BNT, LMI, LMD and TMTB which are strongly associated with future AD risk 
evaluated at the same time as the MRI scan, suggesting that variants in CTNND2
may influence functional as well as structural brain changes. Our analyses also revealed a significant interaction of the top-ranked CTNND2
SNP (rs17183619) with APP
SNP rs2096488 on the degree of cortical cataract and on the bivariate outcome of cortical cataract and THV. Interestingly, the interaction between CTNND2
was strongest in the univariate cortical cataract model suggesting the possibility that this interaction may be stronger or is detectable earlier in the lens. In addition, we demonstrated that a rare missense mutation (G810R) located 249 bp from rs17183619 alters the distribution of δ-catenin and Aβ secretion in neuronal cells. Immunohistochemistry experiments using a δ-catenin antibody revealed punctate staining in the cortical and supranuclear region of the lens from autopsy-confirmed AD cases but not from subjects lacking AD-associated neuropathology.
The rarity of the G810R mutation makes it unlikely that this polymorphism explains the association with CTNND2, despite its proximity to rs17183619 and demonstrated effect on APP processing. It is more likely that common variants or multiple rare variants in tight linkage disequilibrium with rs17183619 alter the function or expression of CTNND2 perhaps by affecting splicing or transcription factor binding.
encodes an adhesive junction-associated protein of the β-catenin superfamily that has been implicated in brain and eye development 
. The δ-catenin protein is a component of the cadherin-catenin complex that recruits presenilin 1 to cadherins and inhibits Aβ production 
. SNPs rs17183619 and rs13170756 surround exon 14 that contains the G810R mutation and encodes part of the highly conserved set of 10 armadillo repeat domains in δ-catenin 
. Armadillo repeat domains 2–10 are necessary and sufficient to mediate the binding of δ-catenin to the hydrophilic loop of presenilin 1 in human brain 
. Presenilin 1-deficient mice show significantly reduced expression of δ-catenin 
and mice lacking normal δ-catenin display severe impairments in learning and memory tasks and in synaptic plasticity 
. Emerging evidence supports a critical role for δ-catenin in dendritic spine maturation and maintenance in the cerebral cortex 
. Recently, it has been suggested that δ-catenin tethers γ-secretase near synaptic membranes 
, supporting the previously proposed concept that γ-secretase complexes exist at the synapse where they are active and regulate synaptic function 
. Thus it is possible that variations in CTNND2
alter synaptic function via a mechanism involving γ-secretase.
The mechanism underling the change in Aβ secretion observed in cells transfected with the G810R mutation warrants further investigation. We have previously shown that δ-catenin interacts with the TM6-TM7 hydrophylic loop domain of presenilin 1 
, thus providing a plausible mechanism for direct modulation of γ-secretase activity. However, we also observed that the mutant δ-catenin has a markedly different subcellular distribution from the wild type protein in HEK293 cells transfected with the mutant δ-catenin construct, an observation consistent with the striking abnormal basolaminar distribution pattern noted in the epithelial layers of human AD lenses. Taken together, these observations suggest that δ-catenin may be aberrantly concentrated under the plasma membrane rather than in the nucleus in AD lens and brain. It is therefore conceivable that G810R alters intracellular trafficking of membrane proteins such as APP, or alters a signaling/transcriptional pathway that influences γ-secretase activity.
Our study showed only modest association of CTNND2
with cortical cataract and no evidence for association with measures of brain degeneration when the lens and brain traits were considered independently in univariate analysis. These observations suggest that CTNND2
accounts for a very small portion of the genetic component of late-onset AD and age-related cortical cataract captured by the phenotype classification system used in the FEOS. Alternatively, AD-linked cortical cataract may be a distinct disorder as suggested by the distinctive subequatorial supranuclear phenotype observed in late-onset AD and Down syndrome 
. Another explanation is that CTNND2
variation affects a very specific process or pathway that is best represented by the bivariate measures of degeneration in the lens and brain that are mechanistically related to altered binding of δ-catenin to presenilin 1. This hypothesis is supported by our data showing apparent increased accumulation and abnormal cellular distribution of δ-catenin in lenses from neuropathologically-confirmed AD patients. The significant interaction of CTNND2
SNPs on the univariate measure of cortical cataract highlights the importance of investigating physiological changes in the lens that may precede neuronal loss.
One of the innovative aspects of this study is the application of a bivariate framework to identify genes for co-heritable traits. This approach detected genome-wide significant association of CTNND2 variants with the bivariate outcomes of cortical cataract and temporal horn volume, whereas association with this gene was not statistically remarkable when these traits were considered separately. Family-based samples are particularly well suited for establishing co-heritability of traits before consideration as a bivariate outcome in genetic association studies. Thus, the Framingham Offspring Study provided a unique opportunity to test our hypotheses because of its family-based design, prospective follow-up of individuals unbiased by selection for ocular or neurological disease, detailed ophthalmological data acquired about ten years prior to the first brain MRI examination, longitudinal monitoring of cognitive status, and availability of GWAS data.
While our results are significant and mechanistically plausible, our study has several limitations. The association findings of CTNND2
SNPs with quantitative outcomes derived from measures of cortical lens opacification and AD-linked neurodegeneration should be replicated in an independent sample even though they met genome-wide significance criteria. However, to our knowledge there are no other large cohorts having eye measurement data obtained many years before acquisition of brain MRI scan data. For this reason, we validated these findings through genetic association analyses with measures of cognitive performance and by experimental approaches. Our association findings for CTNND2
with bivariate outcomes including cognitive function are consistent with results from a recent GWAS in a prospectively followed cohort showing near genome-wide significance for association of rate of cognitive decline with rs2973488 
, a CTNND2
SNP located 2 kb distal of rs13189742 which was strongly associated with the CC-THV bivariate outcome (Table S5
). Another limitation relates to the use of an ordinal variable that precludes inclusion of precataractous lens pathology. Nonetheless, we were able to demonstrate statistically significant association with multiple variables derived from this less precise measure of lens opacity. We also note that meaningful comparisons of co-heritability estimates for clinically ascertained AD with specific cataract and MRI phenotypes in the same sample could not be made as only seven incident AD cases currently exist among the 1249 subjects who participated in both the eye and MRI examination arms of the study. However, analysis of an enlarged sample including 139 AD cases who had the eye exam only revealed significant co-heritability of CC and PSC with AD (Table S2
). Thus, the co-heritability estimates for lens and MRI traits are especially remarkable since all of these subjects were dementia-free at the time of the ophthalmic examination and ten years following at the time of the initial MRI examination. Future studies of this cohort are likely to provide additional insight into the relationship of AD pathogenesis in the brain and lens.
Taken together, prior observations and the results from our study support the existence of a pathway leading to AD-linked pathology in the brain and lens, a hypothesis that supports a systemic rather than brain-limited focus for age-dependent AD pathogenesis. This hypothesis is indirectly supported by the epithelial origin of the lens and brain (surface ectoderm and neuroectoderm, respectively) and the long-lived nature of the terminal differentiated cell types affected by AD pathology in the lens and brain. Further investigation is needed to determine the specific molecular and cellular mechanisms underpinning presumptive linkage of AD pathology in these two anatomical compartments. The implication that a genetic variant can alter the function of a protein affecting cortical cataract and AD suggests that these two systemically distinct diseases may be related, thus suggesting possible convergent pathogenic mechanisms. Moreover, δ-catenin, and possibly other members of the cadherin-catenin complex, may provide new therapeutic targets for AD and cortical cataracts. Finally, detection of AD-linked lens pathology could serve as a peripherally accessible biomarker to facilitate discovery, development, evaluation, and implementation of emerging AD therapeutics.