The mechanistic basis of diverse, environmentally dependent genetic interaction in a metabolic pathway

Metabolic imbalances underlie a large spectrum of diseases, spanning congenital and chronic conditions and cancer. Our ability to explain and predict such imbalances remains severely limited by the diversity of underlying mutation effects and their dependence on the genetic background and environment, but it is unclear whether these complicating factors can be reduced to simple quantitative rules. To characterise their interplay in determining cell physiology and fitness, we systematically quantified almost 4,000 interactions between expression variants of two genes of a well-known sugar-utilisation pathway containing a toxic metabolite in the model bacterium, Escherichia coli, in different environments. We detect a remarkable variety of types and trends of intergenic interaction in this linear pathway, which cannot be reliably predicted from the effects of each variant in isolation, along with a dependence of this epistasis on the environment. Despite this apparent complexity, the fitness consequences of interactions between alleles and environment are explained by a mechanistic model accounting for catabolic flux and toxic metabolite concentration. Our findings reveal how, contrary to a common assumption, the nature of fitness interactions is governed by more than just the topology of the molecular network underlying a selected trait. Our prospects of predicting disease and evolution will therefore improve by expanding our knowledge of the links among proteome, metabolome and physiology. The experimental system consists of 2 combinatorially mutagenised promoters (each driving expression of a metabolic gene) carried on a plasmid. Plasmids are barcoded with 20 randomised nucleotide tags, so that every gentoype is associated to multiple different barcodes. Sequencing thus consisted of two steps: first, MiSeq was used to uncover which barcodes were attached to which genotypes (one sample - "pTet-pLac-BC-linked"); second, the library was grown together and barcodes tracked over time with HiSeq in 4 independent experiments. A preliminary experiment consisted of 6 time-points (6 samples). The 3 others were performed in 3 different experimental environments, one of which replicated the preliminary experiment, and consisted of 4 time-points each (3x4=12 samples).