Analysing GCN4 translational control in yeast by stochastic chemical kinetics modelling and simulation
© You et al; licensee BioMed Central Ltd. 2011
Received: 25 March 2011
Accepted: 18 August 2011
Published: 18 August 2011
The yeast Saccharomyces cerevisiae responds to amino acid starvation by inducing the transcription factor Gcn4. This is mainly mediated via a translational control mechanism dependent upon the translation initiation eIF2·GTP·Met-tRNAiMet ternary complex, and the four short upstream open reading frames (uORFs) in its 5' mRNA leader. These uORFs act to attenuate GCN4 mRNA translation under normal conditions. During amino acid starvation, levels of ternary complex are reduced. This overcomes the GCN4 translation attenuation effect via a scanning/reinitiation control mechanism dependent upon uORF spacing.
Using published experimental data, we have developed and validated a probabilistic formulation of GCN4 translation using the Chemical Master Equation (Model 1). Model 1 explains GCN4 translation's nonlinear dependency upon uORF placements, and predicts that an as yet unidentified factor, which was proposed to regulate GCN4 translation under some conditions, only has pronounced effects upon GCN4 translation when intercistronic distances are unnaturally short. A simpler Model 2 that does not include this unidentified factor could well represent the regulation of a natural GCN4 mRNA. Using parameter values optimised for this algebraic Model 2, we performed stochastic simulations by Gillespie algorithm to investigate the distribution of ribosomes in different sections of GCN4 mRNA under distinct conditions. Our simulations demonstrated that ribosomal loading in the 5'-untranslated region is mainly determined by the ratio between the rates of 5'-initiation and ribosome scanning, but was not significantly affected by rate of ternary complex binding. Importantly, the translation rate for codons starved of cognate tRNAs is predicted to be the most significant contributor to the changes in ribosomal loading in the coding region under repressing and derepressing conditions.
Our integrated probabilistic Models 1 and 2 explained GCN4 translation and helped to elucidate the role of a yet unidentified factor. The ensuing stochastic simulations evaluated different factors that may impact on the translation of GCN4 mRNA, and integrated translation status with ribosomal density.
KeywordsmRNA translation GCN4 Gillespie algorithm stochastic model
Reprogramming gene expression is an important means for cells to adapt to environmental changes. In eukaryotes, gene expression is regulated at multiple levels, including transcription, RNA splicing and translation. Translational control mechanisms, particularly acting at the level of translation initiation, can be a primary point of regulation for certain genes. The yeast GCN4 gene is one such example. It encodes a transcription factor that regulates expression of genes encoding amino acid biosynthetic (and other) enzymes. As such, it plays a central role in the amino acid starvation or GCN response [1, 2].
At the beginning of GCN4 mRNA translation, a 43S ribosomal subunit, incorporating an eIF2·GTP·Met-tRNAiMet ternary complex (TC), scans from the 5' end of the mRNA to initiate translation at uORF1. Following uORF1 translation termination, about half of the 40S subunits remain on the mRNA and resume scanning. When amino acids are abundant, the concentration of ternary complex is relatively high, these scanning 40S ribosomal subunits efficiently re-acquire ternary complex after uORF1 translation, forming active 48S preinitiation complexes. These 48S complexes reinitiate (i.e. recognise and subsequently translate) at downstream uORFs 3 and 4, which have 3' sequence contexts that promote ribosome release. This restricts the supply of ribosomes to the main GCN4 ORF and attenuates its translation. Hence, Gcn4 protein production is low under amino acid replete conditions (Figure 1B).
When yeast cells are starved of amino acids, phosphorylation of eIF2 by the Gcn2 kinase causes a reduced abundance of eIF2· GTP , and a consequential reduction in the concentration of ternary complex. 40S subunits scanning downstream of uORF1 have a reduced chance of re-acquiring ternary complex. Instead, 40S subunits frequently re-associate with eIF2·GTP·Met-tRNAiMet, only when scanning has progressed past uORFs 3 and 4, but before the main GCN4 AUG codon. The translation of the main GCN4 ORF under starvation conditions elevates Gcn4 synthesis by about 34-fold, which leads to the activation of amino acid biosynthetic genes .
uORF spacing, and its effect on ternary complex re-acquisition, is thus central to the GCN4 translational control mechanism. Intuitively, given a constant scanning speed, the time it takes for a 40S subunit to reach uORF4 from uORF1 should scale linearly with the corresponding intercistronic distance between uORF1 and uORF4. Naively, one would assume that the proportion of ribosomes that translate uORF4 is linearly dependent upon this distance, since the 40S subunits have more time to bind the ternary complex. However, it was found that this proportion depends nonlinearly upon the intercistronic distance . This unexplained observation motivated our use in this study of mathematical modelling of this stochastic process as an important tool to analyse the GCN4 control. Naturally, we chose to use a stochastic theoretical framework to address these issues.
Previous work from the Hinnebusch laboratory on how unnaturally short intercistronic distances between the upstream and main ORFs affect the rate of reinitiation at the main GCN4 ORF has implicated an additional unidentified factor (Factor X) in the recognition of the GCN4 start site (Figure 1B) . This factor is not required for uORF4 start codon selection. In contrast to ternary complex, its levels are low when amino acids are replete, and its levels are high under amino acid starvation conditions. This factor can help explain the difference in translational behaviours of uORF4 and the main GCN4 ORF under repressing and derepressing conditions. Factor X could be an unknown protein, or an identified initiation factor that is involved in start codon selection (see Discussion for more details). In this study, probabilistic modelling was used to evaluate how Factor X affects translation of GCN4 mRNA with different intercistronic distances.
Our aim in this paper has been to develop a quantitative understanding of GCN4 mRNA translational control, taking into account stochastic effects, and to use that model to understand some hitherto unexplained experimental observations. Previously we reported a simple probabilistic model of GCN4 mRNA translation . Here, we constructed a comprehensive probabilistic model that encompasses more mechanistic details based on Chemical Master Equation (Model 1). This approach gives the model a rigorous theoretical basis. This model was simplified to form a probabilistic Model 2. We used Model 2 to estimate ternary complex levels under repressing (replete) and derepressing (starvation) conditions. Based on these values, we developed a stochastic model (Model 3) to include the effects of steric hindrance caused by scanning ribosomes. Using the Gillespie algorithm, we performed stochastic simulations to investigate how translation of GCN4 mRNA is affected by different parameters, including 5'-loading of ribosomes and the scanning rate.
Translation of uORF4 and GCN4 protein coding ORF
is a relative percentage, and it is sensitive to ternary complex levels.
where and are the activities of the constructs with the wild type uORF1-GCN4 distance.
Grant et al.  showed that . and depend on (n1+n2) in markedly different fashions (Figure 3). is sensitive to the amino acid availabilities (Figure 3A). In contrast, exhibits similar (n1+n2) dependency under both repressing and derepressing conditions, resulting in the two curves almost coinciding with each other. After ruling out the possibility that mRNA secondary structures lower scanning rates in the uORF1-GCN4 constructs, the existence of a Factor X was hypothesized to explain this phenomenon . To reinitiate at the main GCN4 ORF, a 40S subunit needs to bind an extra Factor X in addition to ternary complex during its scanning. This factor is not required for uORF4 reinitiation. Factor X has low activities under amino acid replete conditions, and assumes high activities in response to amino acid starvation. In what follows, we use stochastic modelling to examine the contributions of both ternary complex and the hypothetical Factor X on GCN4 control.
Model 1 hypotheses
We applied stochastic chemical kinetics to model GCN4 mRNA translation. First we developed a probabilistic model of the likelihood that a ribosome either translates the inhibitory uORF4 or the main GCN4 ORF under repressing and derepressing conditions. Several simplifications were made to construct Model 1.
Recent studies suggest that reverse scanning is negligible [6, 7]. It is therefore not considered in the model. Ribosome scanning is a highly efficient process [6, 8]. Hence, we assume that ribosomes do not abort scanning. We further assume that ribosomes scan at a constant speed. In principle, the ternary complex can dissociate from the 40S ribosomal subunit. However, it was recently reported that this is a slow process in vitro . Therefore, this rare event is not considered. Finally, we assume that there is no "leaky scanning" (i.e. a 48S complex recognises and translates the immediately next ORF it encounters), and that all ribosomes dissociate from the mRNA after translating uORF4. It is worth noting that all reactions, including scanning, are expressed using a common dimension nt/s in this work. For more detailed discussions, please see Supporting Information section S0 (Additional file 1).
Derivation of a simple model from Chemical Master Equation
First we developed a probabilistic model of the likelihood that a ribosome either translates the inhibitory uORF4 or the main GCN4 ORF under repressing and derepressing conditions.
The regulation of the GCN4 is crucially dependent upon the distances n 1 and n 2 . Previous detailed analysis of the dependencies of GCN4 ORF translation on n 1 and n 2 revealed evidence that a post-uORF1 scanning ribosome has to acquire not only ternary complex, but also an unidentified Factor X to become competent to initiate at the GCN4 AUG . In what follows, we use stochastic modelling to examine the contribution of the hypothetical Factor X on GCN4 control, and the interplay between Factor X and ternary complex.
where a S and a D are propensity functions, which are equivalent to reaction rates in the conventional deterministic kinetics.
Alternatively, equation 7 can be directly obtained from Gillespie algorithm . In a system in which a ribosome moves at rate a S and disappears at rate a D , the probability that the scanning reaction happens is equal to propensity function of scanning (a S ) divided by the summation of propensity function of all possible reactions (a S + a D ). Consequently, the probability of a ribosome scanning n nucleotides without binding a ternary complex is the probability of consecutively selecting n times the scanning reaction, the same as defined in equation 7. This is an important intermediate conclusion that will be used in model developments that follow. Next, we derive a mechanistic Model 1 that considers both ternary complex and Factor X.
Probabilistic Model 1 formulation
According to equation 7, the probability of scanning i times before the ribosome binds any factor is P S1 = [a S /(a S + a TC + a X )] i .
The probability for a 40S subunit to bind a ternary complex before Factor X is P TC1 = a TC /(a S + a TC + a X ). After acquiring ternary complex, there are only two possible reactions in the system (i.e. reactions 10 and 12). The probability of scanning changes to P S 2 = a S /(a S + a X ), and the probability for the ribosome to traverse the remaining distance without associating with Factor X would be . Hence, gives the probability of assembling Factor X before reaching the GCN4 ORF.
where P X 1 = a X /(a S + a TC + a X ), P S 3 = a S /(a S + a TC ).
Relative binding rates for ternary complex and Factor X compared with a fixed scanning rate (30 nt/s) under the two conditions
Relative TC binding rate
(a TC /a S , %)
Relative Factor X binding rate
(a X /a S , %)
GCN4 translation is affected by intercistronic distances
Percentage of reinitiation at GCN4 as observed in experiments and predicted by Model 1
DR(%) measureda (predicted)
R(%) measureda (predicted)
Secondly, elongating n2 (while fixing n 1 ) would be predicted to increase the time it takes for a ribosome to scan to the GCN4 ORF after bypassing uORF4. This would be expected to favour binding of ternary complex and Factor X, both of which are required for GCN4 start codon recognition (Figure 4A, B). For instance, when n1 is reduced to only 32 nucleotides, it was found that doubling the distance n2 (from the natural distance of 151 nucleotides to 295 nucleotides) increases GCN4 translation activity by about 40% under derepressing conditions . This experimental observation is consistent with model prediction (Table 2: compare mutant 1 with mutant 2 under derepressing conditions).
In addition, Model 1 also predicts that about half of the ribosomes that bypass uORF4 will bind both the ternary complex and Factor X after scanning 100 nucleotides downstream of uORF4. This is in accordance with the experimental observation that GCN4 translation is lowered by around 50% when an additional artificial uORF is inserted 100 nucleotides downstream of uORF4 (Table 2: compare mutant 3 with mutant 4 under derepressing conditions) .
Interestingly, the model also predicts that the natural length for n2 does not guarantee that all ribosomes that bypass uORF4 will reinitiate at GCN4. For example, under derepressing conditions, lengthening n2 from 150 nucleotides to 600 nucleotides is expected to lead to an increase in GCN4 translation of up to 70%, irrespective of the n1 value (Figure 4A: compare the line marked by diamonds with the line by circles). This is an interesting prediction that could be tested experimentally.
GCN4 translation is regulated via modulation of ternary complex levels
Similarly, we have also investigated how changes in the level of Factor X would affect GCN4 translation under derepressing conditions. As shown in Figure 5B, Model 1 predicts that a five-fold reduction in Factor X would have no detectable effect upon optimum GCN4 reinitiation, while even a ten-fold decrease in the abundance of Factor X would only lead to a reduction of roughly 20% in the optimum reinitiation frequency at the GCN4 ORF. The reason why GCN4 reinitiation appears to be relatively insensitive to Factor X concentration is that the natural uORF1-GCN4 distance appears to be long enough to ensure efficient Factor X binding, even when Factor X levels are significantly reduced. For a construct that contains only uORF1 and GCN4 (as shown in Figure 1D), all ribosomes bind Factor X and translate GCN4 under both repressing and derepressing conditions, as shown in Figure 3B. We conjecture that the natural GCN4 mRNA has probably been evolutionarily selected to minimise any undesirable effects of Factor X on GCN4 translational regulation. Consequently, Factor X might not have a significant impact upon the dependence of GCN4 translation upon ternary complex levels when uORF1 and GCN4 are separated by the natural distance of 350 nucleotides or more. Any distance shorter than this, like in the experiments carried out by Grant et al, would make the effect of Factor X more obvious (Figure 3). In summary, Model 1 reveals that the intercistronic distances n1 and n2 are critical for the regulation of GCN4 mRNA translation, and that the natural distances may minimise the complicated effects of Factor X.
Model 2: The translation of naturally occurring GCN4 mRNA can be modelled without considering Factor X
Since we are interested in the translation of GCN4 mRNA with natural intercistronic distances, from now on, we neglect the existence of Factor X, and generate a simpler probabilistic model (Model 2, available in MATLAB format as Additional file 7. Model_2.m) to address how ternary complex binding rate controls GCN4 translation. Of course, this model would not be able to explain the data concerning GCN4 mutations with unnaturally short n 2 distance (Figure 3B). But our imperative is to study the translational control for GCN4 mRNA with natural uORF1 and uORF4 placements.
The binding rate for ternary complex relative to the scanning rate was estimated by fitting the experimental data in Figure 3A to Model 2. To parameterise Model 2, copy the following files (Additional file 4. isres.m; Additional file 5. srsort.c; Additional file 8. Model_2_parameterisation.m) into the same folder. Mex srsort.c in MATLAB and run Model_2_parameterisation.m. The resulting values (a TC relative to a constant scanning rate: repressing 4.5%; derepressing: 0.66%) were similar to those for Model 1 (compare these values with Table 1). In addition, we used Model 2 to predict the effects of ternary complex levels upon the translation of a GCN4 mRNA with natural intercistronic distances (Figure 1C). These predictions were then compared with those generated by Model 1. The ternary complex dependency curves for the two models were nearly superimposable, with similar optimal values for ternary complex binding rates (Figure 5B). This confirms that GCN4 translational regulation is essentially captured by the simplified Model 2, and that Factor X is not required to explain the translational behaviour of wild type GCN4 mRNA.
Model 3: Stochastic simulation of ribosome distribution on the GCN4 mRNA
The first and second probabilistic models analysed the translational control of GCN4 mRNAs under repressing and derepressing conditions, exploiting data generated using GCN4-lacZ fusions. We then extended this work by constructing a third stochastic model that exploits data about ribosome loading on GCN4 mRNA. These data provide another important reflection of the in vivo translational status of this mRNA. Arava et al.  surveyed the polysome size of different sections of the GCN4 mRNA: a 5'-section comprising the 5'-untranslated region, and a 3'-section representing the coding and 3'-untranslated regions. They found that the 5'-section carries about one ribosome under repressing conditions, and about two ribosomes under derepressing conditions, whereas the 3'-section of the GCN4 mRNA has no ribosomes under repressing conditions, and about four ribosomes under derepressing conditions . It is worth noting that 3-aminotriazole (3-AT) was used to induce amino acid starvation in these experiments, whereas in the above mentioned experiments that assayed GCN4-lacZ activity, gcd mutants were used to mimic derepressing conditions . In S. cerevisiae gcd mutants, the levels of ternary complex are believed to be lowered to a similar degree to the 3-AT condition, although charged histidyl-tRNA levels are not affected in these gcd cells . The presence of 3-AT also lowers the levels of charged histidyl-tRNA, inhibiting histidine biosynthesis . Hence, the differential polysome sizes under the two conditions have allowed us to explore the effects of reducing the levels of charged histidyl-tRNA.
We were interested in correlating the observed changes in GCN4 ribosome loading with changes in specific kinetic parameter values (e.g. rates of initiation, scanning, and elongation, etc) under derepressing and repressing conditions. The levels of ternary complex affect the abundance of charged 43S ribosomal subunits, and hence the rate at which these 43S ribosomal subunits load onto the 5'-end of the GCN4 mRNA. For simplicity, we treated the ternary complex binding rate and 5'-initiation rate as two independent factors in the subsequent discussion. To distinguish the most dominant kinetic parameter values in determining GCN4 ribosome loading, we took the relative ternary complex binding rates from Model 2. Using MATLAB, we placed these rates in a stochastic simulation framework that describes the behaviour of ribosomes on the GCN4 mRNA. As depicted in Figure 2, the GCN4 mRNA in Model 3 contains uORF1, uORF4 and the main GCN4 ORF. Also, Model 3 inherits all of the simplifications defined for Model 1 and Model 2 (above). Multiple ribosomes were allowed on a single mRNA simultaneously. We assume that each ribosome occupies 36 nucleotides of mRNA, irrespective of its position in the mRNA . If one scanning ribosome encountered a second ribosome, its progress would be sterically hindered. This issue was not addressed in either of the first two models. The biochemical equations that underpin Model 3 are available as additional files. To simulate Model 3, copy Additional file 9. GCN4_translation.m and Additional file 10. GCN4_codon_rate.txt into the same folder and run GCN4_translation.m in MATLAB.
Simulations were carried out using the Gillespie algorithm, starting each with an unloaded GCN4 mRNA. In these simulations we ensured that the system reached steady-state and that enough data were generated to generate statistically significant results. Simulations were carried out for 60 minutes, and data from 10-60 minutes were analysed to make sure that the system reached the steady state. Subsequently, we averaged the ribosome loading on each GCN4 mRNA over time to generate one data point for a specific condition. Fifty such generated data points were then averaged for each condition to calculate the ribosome loading. Because this system is ergodic, this method of averaging provides a view of the translation of multiple copies of GCN4 mRNA in many cells. Of course, we note that the half-life of the GCN4 mRNA is around 19 minutes under repressing conditions . Using simulations to understand the noise in GCN4 mRNA translation due to mRNA degradation is a different issue and is outside the scope of this paper.
5' polysome size is determined by the ratio between 5'-initiation rate and scanning rate
Using Model 3 we first investigated the impact of individual parameters upon ribosome loading on the 5'-section of the GCN4 mRNA (i.e. the 5'-leader region). These parameters included the rate of translational initiation at the 5'-end of the GCN4 mRNA a I , the 40S/48S scanning rate a S and the rate of ternary complex binding a TC . Besides, we also considered the rate of 60S association, the translational elongation rate, and the rate of translational termination (whether the ribosome remains associated or dissociates from the mRNA following termination). These events were not considered in Models 1 and 2 for the sake of simplication. The translational elongation rate for each codon and the termination rates for S. cerevisiae were taken from a previous study . Recent studies suggest that the rate of association of the 60S subunit is not rate limiting for translation . Neither is the rate of 60S association controlled under different nutritional conditions . Hence, the rate of this step was assumed to be constant and equal to the average translational elongation rate (30 nt/s). In addition, the GCN4 uORFs are short, containing only four codons, and consequently their translation was presumed not to be limiting. The rates of translational initiation a I ribosome scanning a S and ternary complex binding a TC were likely to be the most important parameters that influence the ribosome loading on the 5'-leader of the GCN4 mRNA. Therefore, we focused mainly on these parameters.
We first analysed the effects of 5'-translational initiation a I and ribosome scanning a S . The absolute rate of scanning in vivo is unknown, but it is expected to be at least comparable to the translational elongation rate. Therefore, we explored scanning rates within a physiologically relevant range (from 5 to 100 nt/s).
We further tested this idea. At scanning speed of 30 nt/s, we increased the absolute rates of 5'-initiation a I and scanning a S proportionately, and observed how 5' polysome size changed (Figure 7). Model 3 predicted that there will only be a slight increase in the ribosome loading on the GCN4 5'-leader (of about 5%) if these two rates are simultaneously elevated together over a five-fold range, under both repressing and derepressing conditions (Figure 7). If the ratio between the rates of 5'-initiation a I and scanning a S was changed by doubling the scanning rate from 30 to 60 nt/s, then this was predicted to lead to a two-fold decrease in the ribosome loading on the GCN4 5'-leader. These results reinforce the view that 5' polysome size is largely dependent on the ratio between the 5'-initiation a I and scanning rate a S . Figure 7 shows how this ratio changes 5' polysome size. These results can be explained by the theoretical results from the totally asymmetric exclusion progress [18, 19].
Arava and colleagues found that when cells grown in rich media undergo amino acid starvation, the ribosome loading in the 5'-section of the mRNA changes from 1 to 2. The ratio a I ⁄a S is suggested to increase three-fold upon this change (Additional file 11. Figure S2). This might be due to a three-fold increase in a I , or a three-fold decrease in a S . It is also likely that both parameters may change simultaneously under derepressing conditions. This is discussed in detail in a later subsection.
5' polysome size is not significantly affected by the ternary complex binding rate a TC
We also investigated how ternary complex binding rate a TC affects ribosome loading in the 5'-section. Comparing Figure 6A (repressing) with Figure S2 A (derepressing, Additional file 11), it is apparent that this effect is negligible. Changes in a TC itself would only affect the probability for a 40S ribosomal subunit to bypass uORF4 and hence change the ribosome density in the section between uORF4 and GCN4 ORF. However, n 2 is only about 1/4 of the entire 5'-section. Consistent with Model 1, only about 25% of the 40S subunits downstream of uORF1 would bypass uORF4 under derepressing condition (Figure 3A). Taken these two factors together, ternary complex binding rate a TC does not significantly influence 5' polysome size. This is reinforced by the observation in Figure 7, where the 5' polysome size stays approximately the same under the two conditions.
Rates of 5' translation initiation a I and scanning a S affect other aspects of translation
Besides 5' polysome size, a I and a S affect other aspects of GCN4 mRNA translation. We found that when the rate of scanning is limiting, uORF1 translation becomes solely dependent on this parameter. In this case, a higher translation rate is not achievable through increases in the rate of translation initiation at the 5'-end of the mRNA. Hence, the rate of uORF1 translation tends to become saturated as 5'-initiation rates increase, and this trend is most pronounced at low scanning rates (Figure 6B: the curve for 5 nt/s). When scanning is not limiting, uORF1 translation rate is almost linearly proportional to the 5'-initiation rate (Figure 6B: the curve for 100 nt/s). In addition, at slow scanning rates, it takes a relatively long time for a ribosome to move from the first 36 nucleotides of the GCN4 mRNA to expose the 5'-end and allow another round of translational initiation. Hence, at slower scanning rates the 5'-end is less likely to be unoccupied (Figure 6C: please compare the curves for 5 and 15 nt/s). Similarly, at high translational initiation rates the 5'-end of the mRNA is more likely to be occupied.
Ternary complex binding rate affects ribosome loading on the 3'-section of the GCN4 mRNA
Next, we investigated how different parameters are predicted to change the ribosome loading in the 3'-section of the GCN4 mRNA (Figure 2), and related these to the experimental data on 3'-polysome size . Intuitively, the rates of 5'-initiation, ternary complex binding, and histidine codon translation in the GCN4 ORF would be expected to alter ribosome loading in the 3'-section of the GCN4 mRNA. Hence, it is not possible to adjust the 5'-initiation rate a I alone to satisfy the experimental data on ribosome loading in both the 3'-and 5'-sections of the mRNA simultaneously, because the values of a TC and histidine codon translation rates are unknown. As mentioned previously, the ratio a I /a S is suggested to increase from one to about three when cells change from repressing to derepressing conditions. We started by investigating two extreme cases that correspond to two ribosomes in the 5'-section of the GCN4 mRNA: firstly a three-fold decrease in scanning rate (to 10 nt/s, nominal value at 30 nt/s); and secondly a three-fold increase in 5'-initiation rate (to 0.26 s-1, nominal value at 0.087 s-1). This transformed the problem into whether values could be found for the rates of ternary complex binding and histidine codon translation that allow four ribosomes in the 3'-section of the mRNA under 3-AT conditions.
The rate of translational elongation on histidine codons influences the ratio of ribosome loading on the 3'-and 5'-sections of the GCN4 mRNA
To test this idea further, we then examined the effects of a three-fold increase in 5'-initiation activity. Figure 9C shows that when histidine codon translation rates are decreased around 0.002-fold, the 3'-ribosome loading was roughly four. In addition, the relative translational activity of the main GCN4 ORF is not impeded relative to the uORF1 translation rate (Figure 9D). This lends weight to the hypothesis that a decrease in the rate of translation of histidine codons causes a significant increase in ribosome loading on the GCN4 ORF during histidine starvation (treatment with 3-AT).
Ribosome loading on an mRNA may increase due to a limiting translational termination rate. This might also account for an increase in 3'-ribosome loading. Thus, we analysed the impact of translational termination upon ribosome loading on the GCN4 mRNA. Model 3 predicts that, under derepressing conditions, when translational termination becomes limiting, 5'-ribosome loading increases whilst 3'-ribosome loading decreases essentially to zero (data not shown). This is because, under these circumstances, ribosomes become restricted to the 5'-proximal region of the GCN4 mRNA before the uORF1 stop codon while they wait for translational termination. Hence, relatively few ribosomes are able to move beyond uORF1 to reach the GCN4 ORF. This in turn would lead to a significant decrease in 3'-ribosome loading. In conclusion, a limiting translational termination rate does not appear to account for the observed increase in 3'-ribosome loading under derepressing conditions.
Factor X identity
Our modelling has provided insights into the identity of the cryptic Factor X, which was predicted to be one of the eukaryotic initiation factors involved in start codon selection, such as eIF1 or eIF5 . Factor X is needed for GCN4 start codon selection, but is dispensable for uORF4 reinitiation. Model 1 predicts that the rate of Factor X binding increases under derepressing conditions. This could be explained either by an increase in eIF levels or by an increase in their rates of association with the translation initiation complex under derepressing conditions. However, the absolute abundance of eIFs does not change under repressing and derepressing conditions . This prompted us to investigate the second possibility. A recent study suggests that eIF's bind the 40S ribosomal subunit cooperatively, such that the binding of one factor enhances the affinity of the initiation complex for other factors [9, 21, 22]. In addition, an intermediate eIF· eIF· eIF5 complex may be important for TC recruitment [23, 24]. Our model of general mRNA translation in yeast suggests that the level of eIF1· eIF3· eIF5 complex increases about 20-fold during histidine starvation (derepressing conditions) . It is likely that this complex binds the 40S subunit faster than the individual eIFs, thereby mimicking an increased eIF5 level that was proposed by Grant and coworkers . Hence, an increase in the level of the eIF1· eIF3· eIF5 complex formation, and the subsequent enhancement of eIF association with the 40S subunit, might explain the impact of Factor X upon translation.
For the sake of parameter identifiability, the binding of TC was assumed to be independent of Factor X in Model 1. Nevertheless, it would be interesting to investigate the cooperative effects using the model. In addition, if Factor X is indeed the eIF1· eIF3· eIF5 complex, this assumption will not hold due to the cooperation in factor binding. Including such cooperative effects will perhaps affect the quantitative predictions to certain degree but will not change the results qualitatively. On the other hand, a recent study from the Asano group suggests that Factor X might be an mRNA helicase such as Ded1 or Dhp1 . If this is the case, its binding to the ribosome (or mRNA ahead of it) can be considered as independent of TC binding.
GCN4 Regulation by ternary complex
Models 1 and 2 suggested that a decrease in ternary complex levels leads to a gradual increase in GCN4 mRNA translation (Figure 5). In other words, the dependence of GCN4 translation upon ternary complex levels reflects analogue-type behaviour rather than an on-off switch. The structure of equations 22 and 23 in Model 2 clearly demonstrates that this relationship is endowed by nature of the stochastic regulation and is independent of kinetic parameter values. The ternary complex binding rate under the derepressing conditions that we extracted from published experimental data  was estimated to be 0.168 nt/s (i.e. 0.56% of 30 nt/s; Table 1). This was close to the optimal ternary complex binding rate for GCN4 mRNA derepression (Figure 5). However, the 3-AT condition under which Grant and colleagues performed their experiments could be viewed as artificial in that it caused more severe amino acid starvation than natural starvation conditions, leading to lower ternary complex levels than for natural starvation. Hence, translation of GCN4 mRNA operates at much higher ternary complex levels in response to natural amino acid starvations, where the relationship between the two is more linear (e.g. the region between 0.5 to 0.7 s-1 in Figure 5A and 5B). Such dependence of GCN4 translation on ternary complex levels is perhaps advantageous. On one hand, GCN4 is a master transcription factor that remodels nearly a quarter of gene expression in yeast . Such a linear relationship at relatively high ternary complex levels allows incremental increases in GCN4 expression in response to natural starvation, without generating a disproportionate amount of such potent factor. On the other hand, it also allows the cell to mount a higher degree of GCN4 derepression in response to more severe conditions such as 3-AT treatment.
In vivo translational status of GCN4 mRNA
Our modelling has also provided insights into the observed increase in ribosome loading that occurs on the GCN4 mRNA following amino acid starvation. The existing experimental data are unable to distinguish whether this increase in ribosome loading is due to a higher 5'-initiation rate or to a decrease in ribosome scanning . However, these two conditions would have different outcomes in terms of absolute Gcn4 protein production rates (i.e. the higher 5'-initiation rate has roughly 3-fold higher effect than the lower ribosome scanning). In their study of the relationship between intercistronic distance and GCN4 translation, Grant et al.  inactivated all uORFs preceding the main GCN4 ORF by point mutation and measured the activity of the GCN4-lacZ constructs in both gcn and gcd mutants. Under derepressing conditions, the GCN4-lacZ activities were roughly the same in gcn cells (where GCN4 translation is constitutively repressed) and in gcd mutants (where GCN4 translation is constitutively derepressed). Their data indicate that the rates of 5'-initiation are comparable in gcn and gcd cells . However, the histidine analogue 3-AT elicits more severe amino acid starvation than is mimicked by gcd mutations. This is because, in addition to reducing ternary complex levels (like gcd mutations), 3-AT also reduces the levels of charged histidyl-tRNA by inhibiting histidine biosynthesis. Consistently, 3-AT is known to generate a strong protein synthesis defect, as reported in a recent study by Asano's group . Yet, without experimental evidence, we cannot rule out a possible change in 5'-initiation rates during amino acid starvation. To meet this challenge, we require systematic assays of ribosome density combined with measurements of GCN4 translation rates.
Genome-wide analyses of ribosome densities have become possible through the combination of deep RNA sequencing technologies and ribosome profiling . This powerful technology, which is capable of mapping ribosomes on mRNAs with single codon resolution, has provided direct confirmation of the translation of the uORFs in the GCN4 mRNA as well as the translational up-regulation of the main GCN4 ORF following amino acid starvation. Unexpectedly, increased translation of the GCN4 5'-leader region was also observed under these conditions  suggesting that additional aspects of GCN4 translational regulation remain to be elucidated. While our models do not reflect these as yet uncharacterised aspects of GCN4 translation, they have provided new insights into GCN4 translational regulation. Furthermore, while not all uORF-containing mRNAs are regulated using the same mechanisms as GCN4 [29–31], our models provide a useful platform for predictive studies on the translational regulation of other uORF-containing mRNAs.
In summary, a diversity of modelling platforms was used in this study to probe the principles governing control of GCN4 at the translational level, and to probe the contributions made by different soluble translation factors to the control mechanism. The predictions of the models employed were validated by comparison with experimental data, and all reproduced the dependence of GCN4 translation on varying ternary complex levels, a crucial feature of GCN4 regulation.
Overall, the study revealed that the natural intercistronic distances in the GCN4 mRNA are sufficiently long to allow a scanning ribosome to acquire Factor X even when the levels of this factor are low. This suggested that Factor X is largely not a relevant factor in the translational regulation of GCN4 with natural intercistronic distances. Deployment of a stochastic model with awareness of steric interactions between ribosomes (queuing effects) revealed that changes in histidine codon translation rate, rather than alterations in ternary complex acquisition, was the key factor governing increases in ribosome loading on the GCN4 ORF under amino acid starvation conditions. Thus, via mathematical modelling and simulation, we have revealed novel features of GCN4 regulation, an important paradigm of eukaryotic translational control.
We thank Prof. Katsura Asano (Kansas State University) for his valuable discussions.
Funding. TY gratefully acknowledges the support of a 6h Century Studentship (University of Aberdeen) and an ORSAS studentship (Scottish Funding Council). TY, MCR, GMC and AB are supported by the CRISP project (Combinatorial Responses In Stress Pathways) funded by the BBSRC (BB/F00513X/1) under the Systems Approaches to Biological Research (SABR) Initiative. MCR acknowledges the financial support from SULSA and BBSRC (BB/F00513/X1, BB/G010722). This work was supported by BBSRC grants to IS (BB/F019084/1 and BB/G010722/1).
- Natarajan K, Meyer MR, Jackson BM, Slade D, Roberts C, et al.: Transcriptional profiling shows that Gcn4p is a master regulator of gene expression during amino acid starvation in yeast. Mol Cell Biol. 2001, 21 (13): 4347-4368. 10.1128/MCB.21.13.4347-4368.2001PubMed CentralView ArticlePubMedGoogle Scholar
- Hinnebusch AG: Translational regulation of GCN4 and the general amino acid control of yeast. Annu Rev Microbiol. 2005, 59: 407-450. 10.1146/annurev.micro.59.031805.133833View ArticlePubMedGoogle Scholar
- Szamecz B, Rutkai E, Cuchalová L, Munzarová V, Herrmannová A, et al.: eIF3a cooperates with sequences 5' of uORF1 to promote resumption of scanning by post-termination ribosomes for reinitiation on GCN4 mRNA. Genes Dev. 2008, 22: 2414-2425. 10.1101/gad.480508PubMed CentralView ArticlePubMedGoogle Scholar
- Grant CM, Miller PF, Hinnebusch AG: Requirements for intercistronic distance and level of eukaryotic initiation factor 2 activity in reinitiation on GCN4 mRNA vary with the downstream cistron. Mol Cell Biol. 1994, 14 (4): 2616-2628.PubMed CentralView ArticlePubMedGoogle Scholar
- You T, Coghill GM, Brown AJP: A quantitative model for the translational control of GCN4 in yeast. 2nd Foundations of Systems Biology in Engineering Conference Proceedings (Stuttgart). 2007, 121-126.Google Scholar
- Berthelot K, Muldoon M, Rajkowitsch L, Hughes J, McCarthy JE: Dynamics and processivity of 40s ribosome scanning on mrna in yeast. Mol Microbiol. 2005, 51 (4): 987-1002.View ArticleGoogle Scholar
- Kozak M: Regulation of translation via mRNA structure in prokaryotes and eukaryotes. Gene. 2005, 361: 13-37.View ArticlePubMedGoogle Scholar
- Kozak M: Pushing the limits of the scanning mechanism for initiation of translation. Gene. 2002, 299: 1-34. 10.1016/S0378-1119(02)01056-9View ArticlePubMedGoogle Scholar
- Passmore LA, Schmeing TM, Maag D, Applefield DJ, Acker MG, et al.: The eukaryotic translation initiation factors eIF1 and eIF1A induce an open conformation of the 40S ribosome. Mol Cell. 2007, 26 (1): 41-50. 10.1016/j.molcel.2007.03.018View ArticlePubMedGoogle Scholar
- Gillespie DT: A general method for numerically simulating the stochastic time evolution of coupled chemical reactions. J Comp Phys. 1976, 22 (4): 403-434. 10.1016/0021-9991(76)90041-3.View ArticleGoogle Scholar
- Runarsson TP, Yao X: Search biases in constrained evolutionary optimization. IEEE T Syst Man Cy C. 2005, 35 (2): 233-243.View ArticleGoogle Scholar
- Arava Y, Boas FE, Brown PO, Herschlag D: Dissecting eukaryotic translation and its control by ribosome density mapping. Nucleic Acids Res. 2005, 33 (8): 2421-2432. 10.1093/nar/gki331PubMed CentralView ArticlePubMedGoogle Scholar
- Bergmann JE, Lodish HF: A kinetic model of protein synthesis. Application to hemoglobin synthesis and translational control. J Biol Chem. 1979, 254 (23): 11927-11937.PubMedGoogle Scholar
- Wang Y, Liu CL, Storey JD, Tibshirani RJ, Herschlag D, et al.: Precision and functional specificity in mRNA decay. Proc Natl Acad Sci USA. 2002, 99 (9): 5860-5865. 10.1073/pnas.092538799PubMed CentralView ArticlePubMedGoogle Scholar
- Gilchrist MA, Wagner A: A model of protein translation including codon bias, nonsense errors, and ribosome recycling. J Theor Biol. 2006, 239 (4): 417-434. 10.1016/j.jtbi.2005.08.007View ArticlePubMedGoogle Scholar
- Algire MA, Maag D, Savio P, Acker MG, Tarun SZ, et al.: Development and characterization of a reconstituted yeast translation initiation system. RNA. 2002, 8 (3): 382-397. 10.1017/S1355838202029527PubMed CentralView ArticlePubMedGoogle Scholar
- Zaman S, Lippman SI, Zhao X, Broach JR: How Saccharomyces Responds to Nutrients. Annu Rev Genet. 2008, 2.1-2.55. 42, Google Scholar
- Derrida B, Domany E, Mukamel D: An exact solution of a one-dimensional asymmetric exclusion model with open boundaries. J Stat Phys. 1992, 69 (3/4): 667-687.View ArticleGoogle Scholar
- Romano MC, Thiel M, Stansfield I, Grebogi C: Queueing phase transition: theory of translation. Phys Rev Lett. 2009, 102 (19): 198104-PubMed CentralView ArticlePubMedGoogle Scholar
- Hoyle NP, Castelli LM, Campbell SG, Holmes LE, Ashe MP: Stressdependent relocalization of translationally primed mRNPs to cytoplasmic granules that are kinetically and spatially distinct from P-bodies. J Cell Biol. 2007, 179 (1): 65-74. 10.1083/jcb.200707010PubMed CentralView ArticlePubMedGoogle Scholar
- Dever TE, Yang W, Aström S, Byström AS, Hinnebusch AG: Modulation of tRNA(iMet), eIF-2, and eIF-2B expression shows that GCN4 translation is inversely coupled to the level of eIF-2.GTP.Met-tRNA(iMet) ternary complexes. Mol Cell Biol. 1995, 15 (11): 6351-6363.PubMed CentralView ArticlePubMedGoogle Scholar
- Olsen DS, Savner EM, Mathew A, Zhang F, Krishnamoorthy T, et al.: Domains of eIF1A that mediate binding to eIF2, eIF3 and eIF5B and promote ternary complex recruitment in vivo. EMBO J. 2003, 22: 193-204. 10.1093/emboj/cdg030PubMed CentralView ArticlePubMedGoogle Scholar
- Asano K, Clayton J, Shalev A, Hinnebusch AG: A multifactor complex of eukaryotic initiation factors, eIF1, eIF2, eIF3, eIF5, and initiator tRNA(Met) is an important translation initiation intermediate in vivo. Genes Dev. 2000, 14 (19): 2534-2546. 10.1101/gad.831800PubMed CentralView ArticlePubMedGoogle Scholar
- Singh CR, Udagawa T, Lee B, Wassink S, He H, et al.: Change in nutritional status modulates the abundance of critical pre-initiation intermediate complexes during translation initiation in vivo. J Mol Biol. 2007, 370 (2): 315-330. 10.1016/j.jmb.2007.04.034PubMed CentralView ArticlePubMedGoogle Scholar
- You T, Coghill GM, Brown AJP: A quantitative model for mRNA translation in Saccharomyces cerevisiae. Yeast. 2010, 27: 785-800. 10.1002/yea.1770View ArticlePubMedGoogle Scholar
- Watanabe R, Murai MJ, Singh CR, Fox S, Ii M, et al.: The eukaryotic initiation factor (eIF) 4G HEAT domain promotes translation re-initiation in yeast both dependent on and independent of eIF4A mRNA helicase. J Biol Chem. 2010, 285 (29): 21922-21933. 10.1074/jbc.M110.132027PubMed CentralView ArticlePubMedGoogle Scholar
- Udagawa T, Nemoto N, Wilkinson CR, Narashimhan J, Jiang L, Watt S, Zook A, Jones N, Wek RC, Bähler J, Asano K: Int6/eIF3e promotes general translation and Atf1 abundance to modulate Sty1 MAPK-dependent stress response in fission yeast. J Biol Chem. 2008, 283 (32): 22063-22075. 10.1074/jbc.M710017200PubMed CentralView ArticlePubMedGoogle Scholar
- Ingolia NT, Ghaemmaghami S, Newman JR, Weissman JS: Genome-wide analysis in vivo of translation with nucleotide resolution using ribosome profiling. Science. 2009, 324 (5924): 218-223. 10.1126/science.1168978PubMed CentralView ArticlePubMedGoogle Scholar
- Vilela C, Linz B, Rodrigues-Pousada C, McCarthy JEG: The yeast transcription factor genes YAP1 and YAP2 are subject to differential control at the levels of both translation and mRNA stability. Nucleic Acids Research. 1998, 26 (5): 1150-1159. 10.1093/nar/26.5.1150PubMed CentralView ArticlePubMedGoogle Scholar
- Gaba A, Wang Z, Krishnamoorthy T, Hinnebusch AG, Sachs MS: Physical evidence for distinct mechanisms of translational control by upstream open reading frames. EMBO Journal. 2001, 20 (22): 6453-6463. 10.1093/emboj/20.22.6453PubMed CentralView ArticlePubMedGoogle Scholar
- Palam LR, Baird TD, Wek RC: Phosphorylation of eIF2 facilitates ribosomal bypass of an inhibitory upstream ORF to enhance CHOP translation. J Biol Chem. 2011, 286 (13): 10939-10949. 10.1074/jbc.M110.216093PubMed CentralView ArticlePubMedGoogle Scholar