Proteomes evolve under many different constraints including the minimization of the energy required to produce them (1
), and the establishment of biochemical networks that optimize metabolic fluxes or increase fitness by other means (2
). Another constraint, which is not widely studied, is that viable proteomes must be producible with a limited gene expression machinery. Gene expression is in essence a two-step process, whereby protein templates are produced during transcription, and the proteins proper during translation. Although specific limitations apply at every level, translation is overall the more resource intense step. The main components of the translation machinery are tRNAs, mRNAs and ribosomes. Particularly the latter are very costly to produce for the cell and have been proposed to limit gene expression and cell growth as a whole (1
The optimality of a particular proteome is not only a function of its environment, but will also depend on its metabolic maintenance costs. It appears to be generally accepted knowledge that cell resources somehow
limit the achievable proteomes (1
), yet at present we do not have a detailed understanding of this limitation. Indeed, while the particular details of many biological mechanisms involved in gene expression are well understood now, we do not know how the individual processes interact with one another at a systems level.
To understand this we will focus on a particularly well-studied model organism for translation, baker’s yeast (Saccharomyces cerevisiae
). Its translational machinery consists of ~200 000 ribosomes, 4500 expressed mRNA species and 3 million tRNAs distributed over 42 tRNA species (3
). Estimates of the total number of mRNA within the cell at any one time range from 15 000 to 60 000 (3
). For the purpose of this contribution we will primarily focus on the lower end of this range, which is dynamically the most complex area. However, where applicable, we will discuss how conclusions are affected if the mRNA content is higher. The molecular detail of the translation-factor-mediated interactions between the individual components is now well understood (7
). With all these available data, we possess in principle the information to fully characterize yeast translation. However, due to the complexity of the system we are still unable to understand the implications of these data. Even less do we understand the ‘fitness landscape’ of translation, that is how changes in parameters affect the ability of the cell to translate efficiently.
We have recently developed a stochastic simulation model of eukaryotic translation (8
) that can be used to systematically probe the system and understand the implications of available knowledge about translation. For the present study, we parametrized this model using ribosome footprinting data by Ingolia (9
) and other parameter values from the literature (especially (3
)). We thus have a simulation model of S. cerevisiae
translation under a specific growth condition (fast growth in rich medium) that represents in detail tRNA concentrations, individual ribosomes and mRNAs. The model can also provide a system-wide picture of ribosomal traffic jams on particular mRNAs including modulations of initiation rates caused by traffic jams that block the initiation sequence.
Computational modelling of translation per se
is not new. The first model dates back to 1969 (10
) and there has been a steady stream of improved models ever since. In recent years activity in this field has increased substantially. However, the focus of current models tends to be rather narrow as they concentrate on isolated aspects of translation, such as codon usage (11–13
), ribosome–ribosome interactions (14
), initiation (16
) or elongation (18
This study is an attempt to explore the system-level properties of translation. Our interest is not primarily to reproduce a particular known (or conjectured) behaviour of the bio-system. Instead we wish to use the simulation model as a computational synthesis machine to generate a systems understanding of translation. This includes exploring the structure of the parameter space that defines translation. Remarkably, we find that the overall translation rate (proteins per second) can be achieved by a wide range of parameters. In contrast, if not only the overall translation rate, but also the translation rates of each individual ORF are taken into account, then the range of parameters that leave the system invariant narrows drastically. We also find that within the range of physiologically plausible parameters, ribosomes are limiting translation. Our simulations show that a ribosome limited regime has a number of characteristics that are beneficial to the cell: Firstly, the high metabolic cost of ribosomes warrants their economical use. Secondly, too many ribosomes lead to traffic jams and hence sub-optimal use of resources which, thirdly, also makes it difficult for the cell to maintain a specific proteome.