Advertisement

LINKING GENE PATHWAYS AND BRAIN ATROPHY IN ALZHEIMER'S DISEASE

      Background

      Multivariate imaging genetics studies are likely to elucidate complex genotype-phenotype relationships in neurodegeneration and ultimately highlight novel links between brain physiology and molecular biological functions. In this study we aim to identify molecular genetic pathways underlying brain atrophy in Alzheimer’s disease (AD). We use a data-driven statistical model of the joint variation between cortical thickness, subcortical volumes, and genetic variation.

      Methods

      We applied a multivariate approach based on partial least squares to model the joint relationship between very large genetic arrays (1,167,126 single nucleotide polymorphisms) and cortical and subcortical measures from 639 individuals in the ADNI cohort. We enforced a sparse solution, i.e., only few relevant genetic loci were selected, and used a novel cross-validation strategy to ensure the stability of the modelled genetic component. Further, we used gene set enrichment analysis to identify significant biological regulatory functions associated to brain atrophy in AD. The generalization of the model was finally tested on an independent cohort of 553 patients affected by mild cognitive impairment (MCI).

      Results

      Brain areas showing significant cortical thinning were the temporal cortex, the parahippocampal cortex, posterior cingulum, posterior hippocampi, amygdalae (anterior portion), thalami and putamen at the subcortical level (Figure1). The corresponding genetic component was comprised of 27 significant molecular genetic pathways, mostly related to neuronal functions such as myelination and nervous system development (Figure2-3). At the cellular level, we identified pathways related to synaptic properties, and neuronal and cell projections. The obtained genes also identified the glutamatergic synapse at the cellular metabolic level, and several phospholipase C-mediated components. The model detected statistically significant association with pathological markers (MMSE: p = 3.8e-2, time to conversion: p=4.9e-3, and hippocampal volume: p=1.5e-4) when independently tested on 553 patients with mild cognitive impairment (Figure 4).

      Conclusions

      This study links AD-related brain measures to several biological regulatory functions mediated by common genetic variants. Most importantly, the identified metabolic and cellular pathways are linked to known biochemical mechanisms of AD, and highlight potential targets for developing novel therapeutic agents.
      Figure thumbnail fx1
      Figure 1Modelled phenotype component
      The colour indicates the probability of maximal correlation between cortical thiskness loci and genotype variation. At the cortical level, the involved areas are the temporal and the parahippocampal cortex, as well as the posterior cingulum. At the subcortical level, the involved areas are the posterior hippocampi, amygdalae (frontal part), thalami and putamen.
      Figure thumbnail fx2
      Figure 2Modelled genotype component
      The circular Manhattan plot shows the probability of a given genetic locus to be maximally correlated with the cortical thickness variation (Figure 1). The genes proximal to the significant loci are listed in the inner radial list, depending on their physical position.
      Figure thumbnail fx3
      Figure 3Pathways analysis on the set of 389 genes identified by the model
      The table shows the significant terms obtained from the pathway analysis (multiple comparison corrected). The used ontologies were: Gene Ontology term, KEGG pathway, Reactome, TRANSFAC motif term, and miRBase microRNAs. The analysis was performed with the g:Profiler software (http://biit.cs.ut.ee/gprofiler/).
      Figure thumbnail fx4
      Figure 4Testing the model on the independent MCI cohort
      When independently tested on the MCI cohort, the survival analysis shows that the individual projections of the modelled genetic component are significantly associated with pathological markers (time to conversion, MMSE, and hippocampal volume)