The most efficient analysis was determined empirically and resulted in eight gene clusters of low diversity >0.75. The four largest clusters (<10%) were significantly CB-7598 enriched for proteins with function in specific cellular processes and cellular compartments (Fig. 4). Notably, the largest cluster was highly enriched for basic biochemical pathways, including carbohydrate metabolic process (GO:0005975), amino acid and derivative metabolic process (GO:0006519) and generation of precursor metabolites (GO:0006091); Bi-directional hierarchical clustering divided the populations into two main groups, with the three Canadian populations clustering away from the other five.

In concordance with the results from the hypergeometric test for similarity that showed up-regulation of mitochondrial proteins in Canadian populations, the cellular compartment ontology for mitochondrion (GO:0005739) was also significantly enriched in this cluster. Cluster 7 also provided additional confirmation of the mitochondria as the principle site for response to adaptive pressure: this grouping was highly enriched for a separate cluster of mitochondrial proteins, including components of Electron transport (GO:0006118), Generation of precursor metabolites and energy and Ion transport (GO:0006811). Finally, cluster 2 the second largest cluster, terms for Response to stress (GO:0006950), Protein folding and DNA Metabolic processes (GO:0006259) were enriched as was the Cellular component Nucleus with higher expression in Hawaiian, Chilean and Californian populations.

Overall the results of this cluster analysis are in agreement with the observations reported above in that the Canadian populations show higher levels of proteins involved in energy metabolism compared with the other populations. Figure 4 Results of clustering of all midgut proteins found differently expressed for any population. The clustering performed here was based on P-values from the Linear Mixed Effects analysis as they consider significance of the relative differences on protein levels. However, a close approximation to protein abundance in the Linear Mixed Effects model is the population effect value, which give a measure of relative change (log scale) of abundance for each protein compared to all the populations (see Methods); the average population effect values for each Drosophila KEGG [25] pathway is plotted in Fig.

5a,b. The dme01100 General Metabolism, dmel00480 Citrate cycle and dmel 00190 Oxidative phosphorylation pathways were used, along with two manually constructed composite pathways named Carbohydrate metabolism and Amino acid metabolism. Here functional specificity allows any noise associated with co-clustering proteins of different pathways to be eliminated. From the plot it is clear that populations Batimastat from Ontario, and Saskatchewan express higher levels of all 4 key metabolic pathways that emerged from the analysis of P-value clustering above.