Submitted to the poster abstracts track at ISMB 2005
Multiple Knockouts Analysis of Genetic Robustness in the Yeast Metabolic Network
David Deutscher, Isaac Meilijson, and Eytan Ruppin; Tel Aviv University, Israel
50-word short abstract
Genetic robustness, a constant phenotype in face of genetic perturbations, is widespread in biological systems. By analyzing the
results of multiple concurrent knockouts to the metabolic genes of S.cerevisiae, we provide the first large-scale study of metabolic
network robustness, portraying its architecture and shedding new light on its evolution.
1000-word abstract
Genetic robustness characterizes the constancy of the phenotype in face of heritable perturbations. In laboratory conditions only
19% of the genes in the yeast S. cerevisiae are essential, i.e., their null mutation is lethal to the organism. All other genes are
apparently dispensable. Previous studies had provided three explanations accounting for this observed dispensability: a gene function
might be buffered (a) by duplication or overlap at either the sequence or the molecular function levels; (b) by an alternative
biochemical pathway; or (c) a gene might be involved in processes that are required only under untested environmental conditions.
Here we go beyond gene essentiality and dispensability and chart the architecture of robustness against gene knockouts of the yeast
metabolic network, employing large scale deep multiple knockouts in an in-silico model.
Extending the common notion of essentiality we define a gene as contributing to the organism’s viability and growth if it is a
member of an essential gene-set. This denotes a set of genes whose combined knockout results in a mutant strain with very slow or no
growth, but where the growth rate of each mutant missing only a subgroup of these genes remains high. The functioning of any one
gene in an essential set hence buffers against the concomitant knockout of all other genes in the set, providing a basic functional
backup and indicating the existence of pair-wise backup interactions (also termed synthetic, aggravating or synergistic interactions).
We denote the system as k-robust to a specific gene knockout according to the size k of the smallest essential set that includes the
knocked-out gene. Thus, the system is only 1-robust to knockout of an essential gene, 2-robust to knockout of a gene involved in a
synthetic lethal pair, etc.
We study genetic robustness using Palsson and colleagues' previously reconstructed Flux Balance Analysis (FBA) model of the
metabolic network of the yeast, incorporating 708 genes, 1175 reactions, and 584 metabolites, and focus on the 483 genes with known
ORFs. The FBA analysis takes into consideration the structure, stoichiometry, and basic thermodynamics of the metabolic network,
applying mass-balance constraints to predict phenotypes with general prediction accuracy of 70-90%. Our analysis is performed in
two stages: In the first, we exhaustively search through the space of all possible combinations of concomitant knockouts of up to four
genes. We record the essential sets found, and list the contributing genes with their k-robustness levels. Since further exhaustive
search is currently computationally infeasible, the second stage employs a novel stochastic sampling method to identify genes
involved in essential sets which include up to 8-10 genes. Essential sets and backed up genes are marked as backed by functional
duplication, alternative pathways, or both.
Using exhaustive multiple knockouts search we found 48 essential genes, 16 essential pairs, 16 triplets, and 38 essential
quadruples, overall involving 157 contributing genes. The gene knockout sampling method identified an additional 174 contributing
genes with k-robustness levels greater than four. Overall, the contributing genes total 74% of the tested genes, compared with only
10% of the genes that are identified in-silico as essential using traditional single-knockouts (and 13% previously found in-vivo). This
indicates that a large majority of the genes are involved in processes already required in the standard laboratory rich environment
used, even though the individual genes are not essential. Backups more often arise from alternative pathways than from gene
functional duplication, the former being solely responsible for 45% of the backed-up genes and partially responsible for 33% more.
Alternative pathways are particularly dominant in genes with high k-robustness levels, suggesting that their role in genetic robustness
might be underestimated when the investigation is limited to shallow knockout depths. In qualitative agreement with previous
observations, we find that the number of essential sets per gene is usually small, averaging 6.4 sets or 22 pair-wise interactions per
gene, though few genes are involved in many interactions.
We used the GO-Slim biological process annotations from SGD to test if any biological process category is significantly enriched
or depleted with backed-up genes. Indeed, two main metabolic functions, amino acid and carbohydrate metabolism pathways, are
highly backed-up (p-values 10-8, 0.09), together with the genes catalyzing transport processes (p=0.005). In contrast, genes
functioning in lipid metabolism contain significantly more essential genes than expected by chance (p=2∙10-7). Looking at the backup
interactions between the functional categories using the full GO annotation, we find that, overall, inter-functional backup interactions
are abundant: only 14% of backup interactions are between genes annotated with similar GO terms, comparable with the 27 % found
previously found experimentally, and differing from previous observations in small scale systems.
A previous study of the mechanisms for dispensability found that 37-68% of the dispensable metabolic genes are conditionspecific (they are dispensable on rich medium but there is an alternative environment where there is a non-zero flux through their
reactions), concluding that environmental specificity is the dominant explanation behind dispensability. Our findings shed new light
on these results: First, we find that 91% of these condition-specific genes are contributing already in the standard rich environment.
Second, we find a significant correlation between the k-robustness of genes and their environmental specificity (R=0.50, pvalue=4∙10-18, N=268; data courtesy of Balázs Papp). That is, genes with many backups tend to catalyze reactions that are activated
only in few, specific, environments. Hence, while environmental specificity may be the source of the observed genetic robustness, the
converse may also hold: the availability of backups can allow high k-robust genes to flexibly adapt to specific environments during
evolution. To examine the extent to which the k-robustness of genes may actually confer them with a functional backup from an
evolutionary perspective, we compared the genes' k-robustness levels with their evolutionary retention (based on a gene loss
calculation). The resulting significant correlation (R=-0.32, p=2∙10-9, N=331) shows that high k-robust genes indeed tend to evolve
faster. In summary, this work provides the first large-scale study of metabolic network robustness, portraying its architecture and
shedding new light on its evolution.