Multi-influential interactions controls behaviour and cognition through a limited number of pathways in Down syndrome mouse models (Affymetrix reanalysis)

Down syndrome (DS) is the most common genetic form of intellectual disability with additional clinical features, caused by the presence of an additional copy of human chromosome 21 (Hsa21). A few number of patients with DS features, carry partial duplication of Hsa21 and their study provided novel insights into genotype–phenotype correlations. Despite the progress of genome analysis, the rareness of patients with partial duplication, and the human genetic heterogeneity, makes difficult to achieve a more detailed phenotypic map at present. As a complementary approach, we screened the in vivo DS mouse library with highly standardized behavioural tests, magnetic resonance imaging (MRI) and digged into hippocampal gene expression to go further in dissecting the genotype–phenotype correlations and in deciphering misregulated genes, functional pathways and biological cascades in DS models. Altogether this approach bring novel insights into the field. First, we unravelled the complexity of the genetic interactions between different regions of the chromosome 21 and how they play an important role in modulating the outcome of the behavioural and molecular phenotypes. Then, in depth analysis of misregulated expressed genes involved in synaptic dysfunction highlitghed six biological cascades centered around DYRK1A, GSK3beta, NPY, RHOA, NPAS4 and the SNARE complex. Finally, we provide a novel vision of the existing altered gene-gene crosstalk and molecular mechanisms that could be at play for both the DS clinical features and the rescue mechanisms by targeting specific hubs or well connected nodes that may be central to advance in our understanding of DS and therapies development. I downloaded and re-analysed the gene expression profile of the human derived dataset of 8 euploid and 7 Down syndrome adults published by Lockstone et al., (2007) with GEO accession number GSE5390. The re-analysis was performed using our bioinformatics pipeline, developed over the R environment the full description is on the supplementary material of our manuscript Duchon A., Muñiz MM et al., under revision. As a brief sumup, the pipeline consisted in seven major steps: Quality control (QC), pre-processing and normalization of the raw data, analysis of the structure and homogenicity of the data, differential expression analysis (DEA), differential functional analysis (DFA), Network connectivity analyses of pathways and genes and finally network topology and centrality analyses. Data quality control: The quality control (QC) workflow was implemented in two steps. The first step consisted in analysing the quality of the raw data, and the second in the analysis of the quality after the pre-processing and normalization steps. Data pre-processing and annotation: The raw intensity measurements of the arrays where normalized using limma RMA normalization algorithm producing the summaries at the probe/transcript level. The probes were gene and transcript annotated using the specific R package pd.hugene.1.0.st.v1. Data structure and homogeneity: We performed a study of i) the gene expression correlation across models and ii) the structure of the data and the separation between experimental groups using hierarchical clustering, PCA, tSNE and OPLS. Differential expression analysis (DEA): We carried on the differential expression analysis in R environment using FCROS (Dembele et al., 2014). FCROS is a fold change (FC) rank based method that works well with noisy datasets and gives strong reproducible results. FCROS computes for each pair of test/control samples (K pairs), a statistic associated with the k ranks of the FC values for each gene, and the obtained probability (f-value) is used to identify the differential expressed genes (DEGs) within an error level fixed by the analyst. Our fixed error level corresponds to a 5% False Discovery Rate (FDR), and is attained by setting the Ɑ parameter within this range 0.025<Ɑ>0.975.Differential functional analysis (DFA): To identify the altered functions due to the changes in gene expression and possibly get new insights into the mechanistic changes along the pathways between the two conditions trisomic vs. wild type on each model we used the generally applicable gene set enrichment for pathway analysis (GAGE, Luo et al., 2009) R package to carry on the functional expression analysis.The main aim of re-analysing this data was to get new insights into DS gene deregulation first in humans applying a more updated computational pipeline and secondly, by applying a comparative genomic approach with rodent Down syndrome datasets identify the conserved gene and functional deregulated profiles. I downloaded and re-analysed the gene expression profile of human foetal derived datasets published by Mao et al., (2003) with ArrayExpress accession number E-GEOD-1397. We re-analysed 2 samples coming from cerebrum-derived astrocytes, 3 samples of cerebellum, 4 samples of cerebrum and 2 samples of heart of Down syndrome individuals compared to 2 samples coming from cerebrum-derived astrocytes, 3 samples of cerebellum, 7 samples of cerebrum,2 samples of heart of euploid individuals .The re-analysis was performed using our bioinformatics pipeline, developed over the R environment the full description is on the supplementary material of our manuscript Duchon A. & Muñiz MM et al., under revision. As a brief sumup, the pipeline consisted in seven major steps: Quality control (QC), pre-processing and normalization of the raw data, analysis of the structure and homogenicity of the data, differential expression analysis (DEA), differential functional analysis (DFA), Network connectivity analyses of pathways and genes and finally network topology and centrality analyses. Data quality control: The quality control (QC) workflow was implemented in two steps. The first step consisted in analysing the quality of the raw data, and the second in the analysis of the quality after the pre-processing and normalization steps. Data pre-processing and annotation: The raw intensity measurements of the arrays where normalized using limma RMA normalization algorithm producing the summaries at the probe/transcript level. The probes were gene and transcript annotated using the specific R package pd.hugene.1.0.st.v1. Data structure and homogeneity: We performed a study of i) the gene expression correlation across models and ii) the structure of the data and the separation between experimental groups using hierarchical clustering, PCA, tSNE and OPLS. Differential expression analysis (DEA): We carried on the differential expression analysis in R environment using FCROS (Dembele et al., 2014). FCROS is a fold change (FC) rank based method that works well with noisy datasets and gives strong reproducible results. FCROS computes for each pair of test/control samples (K pairs), a statistic associated with the k ranks of the FC values for each gene, and the obtained probability (f-value) is used to identify the differential expressed genes (DEGs) within an error level fixed by the analyst. Our fixed error level corresponds to a 5% False Discovery Rate (FDR), and is attained by setting the Ɑ parameter within this range 0.025<Ɑ>0.975.Differential functional analysis (DFA): To identify the altered functions due to the changes in gene expression and possibly get new insights into the mechanistic changes along the pathways between the two conditions trisomic vs. wild type on each model we used the generally applicable gene set enrichment for pathway analysis (GAGE, Luo et al., 2009) R package to carry on the functional expression analysis.The main aim of re-analysing this data was to get new insights into DS gene deregulation first in humans applying a more updated computational pipeline and secondly, by applying a comparative genomic approach with rodent Down syndrome datasets identify the conserved gene and functional deregulated profiles.