Fig. 1. Phenotype search strategies. Metabolic networks are complex and it is unclear how to a priori obtain the optimal phenotype. Combinatorial tools for gene modification may be employed in a sequential, or gene-by-gene optimization, but may result in a local maximum. A simultaneous, multiple-gene modulation may be required to traverse more complex metabolic surfaces which could arise due to complex interactions and regulation.

Fig. 2. E. coli genome scan for single gene knockout target identification. The phenotype of every possible single gene knockout was simulated using FBA with MOMA as an additional constraint. The above genotype–phenotype plot illustrates the effect of single gene deletions on lycopene yield as measured by the fraction of the stoichiometric maximum yield (approximately 0.1 g Lycopene/g Glucose). A single knockout scan predicted eight genes whose deletion yielded enhanced product synthesis while satisfying a minimum growth requirement. The gene glyA appears twice since its function can be classified as both amino acid biosynthesis and vitamin/cofactor metabolism.

Fig. 3. In silico multiple gene knockout search strategy. The process of maximal sequential phenotype increase is illustrated. The production yields for each genetic background are simulated similarly to the method followed in Fig. 2, but in a different genetic background for the starting strain. A triple knockout construct based on the double mutant gdhA/gpmB was excluded as it violated the minimum growth rate requirement. On the other hand, predicted triple constructs in gdhA/aceE background continue to show an increase in lycopene yield. The path of maximal phenotype increase is given by the solid line. However, since a gdhA/gpmB knockout has been excluded for triple knockouts consideration due to growth rate, the next highest optimal path is followed. These results indicate that novel gene targets arise as the genotype is altered as result of gene knockouts. This is especially evident in the case of talB. Although talB increases the production level in a gdhA/aceE knockout background, it is detrimental in a gdhA only knockout background.

Fig. 4. Experimental results of single and multiple gene knockouts. Lycopene production is shown in ppm along with percent increases compared with the corresponding levels obtained in the parental strain. Total lycopene content increases with multiple knockouts obtained along the path of highest production. This data follows the trends of the simulations presented in Fig. 3. Error bars represent the standard deviations among replicate culture experiments.

Fig. 5. FDA results using 45 genes. (a)–(d) An FDA was applied using (a) top 45 discriminatory genes, (b) 2000–2045th genes, (c) 4000–4045th genes, (d) 6000–6045th genes, when 7000 genes were sorted with F statistic values. The results indirectly indicate that Wilks’ lambda scores successfully identify most discriminatory genes, because the group separation gradually gets worse and within-group variability increases, when the less discriminatory genes were used for a FDA classification. In each case, the projected values were calculated using linear combination of individual gene expressions, y1=v1g1+v2g2++v45g45. (e) The coefficients, v=[v1v2…v45] in FDA classification using the top 45 discriminatory genes, provides a transcriptional fingerprint of the selected discriminatory genes that can be used for discerning the difference between normal and cancerous tissues.