Community control in cellular protein production: consequences for amino acid starvation
aa r X i v : . [ q - b i o . S C ] N ov Community control in cellular protein production:consequences for amino acid starvation.
Frank S Heldt , Chris A Brackley , Celso Grebogi and Marco Thiel , ∗
1. Otto-von-Guericke University Magdeburg, Universitaetsplatz 2, 39106 Magdeburg, Germany.2. SUPA, School of Physics and Astronomy, University of Edinburgh, Mayfield Road, Edinburgh, EH9 3JZ, UK.3. Institute for Complex Systems and Mathematical Biology, SUPA, King’s College, University of Aberdeen, Aberdeen, AB24 3UE,United Kingdom. * corresponding author. Email [email protected]
Abstract
Deprivation of essential nutrients can have stark consequences for many processes in a cell. We consideramino acid starvation, which can result in bottlenecks in mRNA translation when ribosomes stall due to lack ofresources, i.e. tRNAs charged with the missing amino acid. Recent experiments also show less obvious effectssuch as increased charging of other (non-starved) tRNA species and selective charging of isoaccepting tRNAs.We present a mechanism which accounts for these observations, and shows that production of some proteins canactually increase under starvation. One might assume that such responses could only be a result of sophisticatedcontrol pathways, but here we show that these effects can occur naturally due to changes in the supply anddemand for different resources, and that control can be accomplished through selective use of rare codons. Wedevelop a model for translation which includes the dynamics of the charging and use of aa-tRNAs, explicitlytaking into account the effect of specific codon sequences. This constitutes a new control mechanism in generegulation which emerges at the community level, i.e., via resources used by all ribosomes.
Subject Areas: biophysics, biomathematics, mathematical modelling
Keywords: mRNA translation, amino acid starvation, gene regulation
A revised version of this manuscript has been published as follows:
Frank S. Heldt, Chris A. Brackley, Celso Grebogi, Marco Thiel (2015) Community control in cellular protein pro-duction: consequences for amino acid starvation
Phil. Trans. R. Soc. A
Proteins form the basis of all biological processes within a cell, and one of the key characteristics of life is the ability ofan organism to synthesize its own proteins. Understanding how a cell regulates this synthesis remains one of the keyquestions of modern science. In this paper we propose a new mechanism for protein regulation which originates notfrom a complex network of gene control pathways, but instead emerges naturally from the interplay between differentprocesses via the supply and demand of different resources. This represents a new paradigm for gene regulation: thatof community based control of a complex system.The amino acids which make up protein molecules are either taken up directly, or formed from precursors found inthe cell’s environment [1]. When the environment lacks the required nutrients this will have a severe impact on theability to produce certain proteins; if the cell is to survive, it must react to stresses such as amino acid starvation. Asone would expect, if a cell is placed in an environment starved of a particular amino acid, this leads to bottlenecks inthe protein production process due to a reduction of the number of amino acid carrying complexes within the cell. Inthis paper we use mathematical modelling to investigate this scenario. Besides the expected effects, we also uncoversome counter-intuitive behaviour which provides insight to the origins of some recent experimental findings [2, 3],some of which have previously been unexplained. The process whereby amino acids are assembled into proteinsis called mRNA translation, and we consider this in the model organism
Saccharomyces cerevisiae (baker’s yeast).While previous studies have examined how starvation for amino acids activates feedback mechanisms in the initiationstage of translation [4], here we consider for the first time the effect on elongation and particularly ribosome traffic.The sequence of amino acids which gives a protein its unique structure and function is transcribed from thecell’s DNA into mRNA molecules. These are single strands of nucleotides which, in general, encode a single proteinsequence; every three nucleotides - one codon - represents one amino acid. Molecular machines called ribosomestranslate this information sequentially, moving from codon to codon along the mRNA, adding amino acids to agrowing chain. The process begins when a ribosome assembles at the 5’ end of an mRNA and scans for the initiationsignal (usually an AUG codon) [5, 6]. Ribosomes decode each codon via the binding of an aminoacylated-transferRNA (aa-tRNA) [7]. On encountering the required aa-tRNA the delivered amino acid is incorporated into the chain,and the ribosome moves one codon forward; a tRNA which is no longer bound to an amino acid is released. Thisprocess is known as elongation of the amino acid chain. Different codon species are decoded by different tRNAspecies, each of which can only carry a specific amino acid. We refer to the aa-tRNA complex as a “charged” tRNA.Uncharged tRNAs are loaded with their corresponding amino acid by aa-tRNA synthetases [8, 9]. Upon reaching astop codon, ribosomes release the complete protein and dissociate from the mRNA.1 minoacylationaa-tRNA tRNA aaR RRR RR RPInitiation Elongation Termination5' 3'mRNAk α k tr k tr k tr Figure 1: Schematic representation of mRNA translation. An mRNA consists of a sequence of codons, representedhere by boxes. Ribosomes ( R ) initiate translation from the 5’ end with rate k α . By using aa-tRNAs, they elongatewith rate k tr , provided the next codon is vacant. When a ribosome reaches the 3’ end, it dissociates from the mRNAand releases a protein ( P ). Used tRNAs are recharged by aminoacylation which requires amino acids ( aa ).There are two factors which determine the rate of protein translation: regulation at the initiation level [4], andthe rate at which ribosomes move during elongation, i.e., the translation rate of each codon [10]. The latter dependson the time that a ribosome has to wait before finding the correct charged tRNA, and therefore is proportional to theabundance of the aa-tRNAs in the cell [11]. Different tRNA species vary substantially in their abundances [12, 13],which gives rise to slow and fast codons. Also, since in yeast there are only 20 amino acids but 41 tRNA species,there is more than one tRNA for some amino acids (isoacceptors). In some cases a slow codon is used when a fastcodon for the same amino acid is available. One could reason that the biased use of such codons could regulate thedynamics of translation [14].There has been much recent experimental work investigating the effect of amino acid starvation on translation.For instance Dittmar et al. [2] examined the charging of different isoacceptors of the amino acid subject to starvation.Also, Zaborske et al. [3] showed that starvation for histidine, leucine or tryptophan causes a decrease in the charginglevel of not only the tRNA species carrying these amino acids, but also in tRNAs which deliver other amino acids(whose intracellular levels remained unchanged). Intriguingly, some tRNAs even show an increased charging levelwhen cells encounter starvation; as one of the main results of this paper we provide a mechanism for this counter-intuitive experimental finding .Most previous models of translation have assumed that the abundances of the aa-tRNAs are constant [15, 16, 17].A recent study [18, 19] included the finite rate of recharging of tRNAs in a probabilistic model of translation, andallowed aa-tRNA abundances to vary dynamically. It has also been shown that the balance between the supply andthe demand for tRNA resources is a crucial factor in determining protein production rates [20]. Here we examine theeffect of starvation of amino acids on tRNA recharging, and ultimately protein production rates.One might assume that the interesting effects described above could only occur as a result of complicated feedbackmechanisms of control. In this paper we show that actually such behaviour can arise naturally through careful selectionof codon sequence. We first use artificial mRNA sequences to show how behaviour such as an increased chargingof tRNAs as a response to starvation can be realized, before showing that the same effects occur when we considermRNA sequences taken from the S. cerevisiae genome. We see that regulation of protein production can be achievedthrough changes in the supply and demand of resources across all of the mRNAs in the system , i.e., at the communitylevel. It is possible to control how the protein production from a specific mRNA species responds to changes in thecommunity via a careful selection of coding sequence. We thus propose that there may exist a direct and rapidlyresponding mechanism of protein regulation at the translational level.
Figure 1 illustrates the two main features of our translation model, namely (i) ribosome traffic along mRNAs and(ii) the cycle of aminoacylation and usage of tRNAs. We also include explicitly the fact that ribosomes extend overseveral codons and cannot overlap. We now describe each of the aspects of the model in turn.
We derive a system of reaction rate equations for ribosome traffic on mRNAs by adapting a model of translation inbacteria [21]. We consider explicitly that mRNAs are precise sequences of different codons. Ribosomes bind to the2RNA with the rate k α , provided the initiation site and the subsequent codons are vacant R + m X h =1 X ◦ h k α −−→ X • + m X h =2 X ◦ h , (1)where X ◦ j (or X • j ) represents the j th codon of an mRNA. A filled circle denotes a codon occupied by a ribosome’srecognition (A) site (which we assume is the leftmost codon covered by that ribosome - note that this choice doesnot influence the results); an open circle indicates that either the codon is unoccupied, or that it is occupied by aregion of the ribosome other than the A site. R represents a free ribosome and m is the number of codons occludedby a translating ribosome. The rate k α encompasses all of the steps of translation initiation, including the bindingof several initiation factors (some of which are themselves regulated under cell stress conditions). Thus X • denotesa ribosome decoding the first codon downstream of the initial AUG. Ribosomes advance in a stepwise manner withrate k tr from codon j to j + 1 , provided the downstream sites are not occupied. This process requires a chargedtRNA that matches the current codon ( T ⋆C j , where C j indexes the species of codon j , and the star denotes thatthe tRNA is charged) and results in the release of its uncharged counterpart ( T C j ) X • j + T ⋆C j + j + m X h = j +1 X ◦ h k tr −−→ X ◦ j + T C j + X • j +1 + j + m X h = j +2 X ◦ h , (2)with j = 2 , . . . , n , where X • j is a ribosome decoding codon j , and n is the mRNA’s length in codons. We assume thatribosomes reaching the ( n − m )th codon continue translation without any hindrance from downstream ribosomes,i.e., we truncate the sums at h = n . The translation rate k tr is a constant representing all intermediate reactionsteps that the ribosome undergoes [17, 22]. We approximate this value from the average codon translation rate[23] and the abundance of tRNAs in yeast estimated from the gene copy number [24]. Note that unlike in othermodels [25, 21], the overall translation rate of a given codon depends on the concentration of the respective aa-tRNAand hence can vary with time. We make the approximation that the bare tRNA is released instantly after the chargedtRNA is captured; in actuality there are several binding sites on the ribosome, and the tRNA remains bound forseveral further elongation steps. If we assume that it is the availability of amino acids and not bare tRNAs whichwill limit elongation, then this is unlikely to have any substantial effect on our results.Eventually, ribosomes reach the stop codon and release a mature amino acid chain ( P ) X • n + T ⋆C n k tr −−→ R + P + T C n + X ◦ n . (3)The binding of release factors is known to be a fast process [26, 27], so we assume that translation of the last codondominates the rate of termination. Since we do not consider post translational modification of the chain the rate oftermination is equivalent to the rate of protein production, which we denote λ . We describe this system of reactionsusing a system of ODEs, as detailed in Sec. 2.3 below. The process of charging a tRNA with its cognate amino acid is facilitated by a family of aminoacyl tRNA synthetases.The precise mechanism is still controversial [8, 9] and different synthetases may work via different mechanisms. Thuswe do not model charging in detail, but rather use a simple description T k + aa k v k −→ T ⋆k , (4)where aa k represents an amino acid of type k . We choose the rate v k such that the recharging exhibits the kinetics ofa two substrate enzyme-catalysed reaction — an assumption which has been used by several previous authors [28, 29].Hence, the reaction rate is given by v k = v max ,k [ aa k ][ T k ] K m,aa k K m,T k (cid:16) [ aa k ] K m,aak (cid:17) (cid:16) [ T k ] K m,Tk (cid:17) , (5)where K m and v max ,k denote the Michaelis–Menten constants and maximum reaction velocity, respectively; theseare specific to each aminoacyl tRNA synthetase. Since each elongation step requires a single tRNA, we replace theconcentration of tRNA molecules [ T k ] in Eq. (5) with the number of molecules T k and express K m,T k in numbersof tRNAs. However we still consider the amino acids in terms of their concentration. We also neglect the change in [ aa k ] due to the ongoing recharging, and assume that it is constant in time.3 .3 Simulation Method We use reactions (1)-(3) to derive a set of ODEs for ribosome traffic. We consider a population of N identicalmRNAs described by a single set of differential equations. Defining y j as the proportion of mRNAs where there is aribosome translating codon j for ≤ j ≤ n , and y j = 0 for j > n , yields the equations dy dt = k α [ Q mh =1 (1 − y h )] − k tr T ⋆C y hQ m +1 h =2 (1 − y h ) i , (6) dy j dt = k tr T ⋆C j − y j − hQ j + m − h = j (1 − y h ) i − k tr T ⋆C j y j hQ j + mh = j +1 (1 − y h ) i for j = 2 , . . . , n. (7)Terms of the form (1 − y j ) represent hindrance of a ribosome’s movement due to the occupation of downstreamcodons: a ribosome stalls on codon j if any of the codons j + 1 to j + m are occupied, or it moves with rate k tr T C j if j + 1 to j + m are vacant. Given that translation is fast compared to most regulatory mechanisms, we assumethat the number of available ribosomes does not change significantly during translation, and that this does not limitthe initiation process; i.e., the ribosome availability is incorporated into the initiation rate k α . For simulations withmultiple mRNA species, we will consider multiple sets of ODEs.For the aminoacylation process described by Eq. (4), the number of charged tRNA molecules of species k isdescribed by dT ⋆k dt = v k − k tr T ⋆k X C k " y C k C k + m Y h = C k +1 (1 − y h ) , (8)where C k indexes codons that require the tRNA species k for decoding.Equations (6)-(8) form a model of protein translation that explicitly accounts for tRNA charging and its roleduring elongation. We solve the equations numerically using the SBPD package for the Systems Biology Toolbox2 [30] in Matlab (version 7.5.0 R2007b). All of the results which we present represent the steady state, i.e., wedisregard data from the first part of each simulation to account for transient effects due to the initial conditions. In the remainder of this paper, we consider ribosomes that cover m = 9 codons and choose an initiation rate k α =0 . s − (under normal conditions ribosomes are spaced on average 50 codons apart and translate 10 codons s − [27,23]. Although initiation itself is highly regulated under starvation conditions (for example via the GCN pathway[31, 4] or eIF3 initiation factor [32]), our focus here is on downstream regulation, so we keep k α constant – seeSec. 4 for further discussion. We examine the effect of changing the concentration of one amino acid species on theprotein production rate and the charging levels of different tRNAs. We define the latter as the ratio of the numberof charged tRNAs of a particular type to total number of tRNAs of that type, and denote it G k . In order to comparedifferent mRNA and tRNA species we normalize the protein production rate λ by dividing by the rate found for nonstarvation conditions. We also consider the probability y j of finding a ribosome translating the j th codon, referringto this as the ribosome density.A complete genome wide investigation using the above framework would present a substantial computationalchallenge; moreover the data such an undertaking would create would be difficult to interpret. For this reason werestrict our simulations to small subsets of the transcriptome, mostly considering only one mRNA sequence at a time.Before analysing simulations using real mRNA sequences, we first illustrate that the use of slow codons can controltranslation rates using simple artificial mRNA sequences which contain only two codon species, each coding for adifferent amino acid: a common species which makes up most of the mRNA, and a rare species of which there isonly a single codon. We refer to the latter as “species A” codons, decoded by tRNAs of type A carrying amino acidsof species A. We choose the number of tRNAs of each species such that their ratio is equal to that of the gene copynumbers of the most and least abundant tRNAs in S. cerevisiae [24]. Hence, translation of codon species A is slowcompared to the translation of the abundant codons. We investigate the effect of starvation of amino acid speciesA, denoting the concentration [ aa A ] . Our results show that the mRNA sequence affects the rate of protein production, and more importantly its responseto amino acid starvation. Previous work [14] has shown that the response of the protein production rate of differentmRNAs to variation of the initiation rate depends very much on the position of the slowest codons. This allows fora classification of mRNAs based upon features of the coding sequence; a correlation between the different types ofmRNA and the function of the resulting proteins was also found. Inspired by this result, we start by showing that a4 type I y j D type IIcodon j y j [ aa A ] (mmol / L) n o r m a li ze d λ B type Itype II0 0.01 0.02 0.0300.20.40.60.81 [ aa A ] (mmol / L) G A C type Itype II A Figure 2: Steady state results from separate treatments of mRNA I and mRNA II: (a) Schematic representation ofmRNA types I and II. The position of the codon of species A (corresponding to the starved amino acid) is indicated.(b) Protein production rate per mRNA λ (normalized by dividing by the value found for large [ aa A ] ) as a functionof the amino acid concentration. (c) Charging level G A of tRNAs of type A. (d) Codon occupation density at high(brown line) and low (blue line) amino acid concentrations. The oscillatory features in these plots are due to thewidth of the ribosome, which is indicated by the length of the bar above the line in the upper plot.similar classification arises when we examine the response to starvation. Two different mRNA types emerge in thiscontext: type I mRNAs contain a codon of species A some distance downstream of the 5’ region, where a stallingribosome does not directly impair initiation; type II mRNAs have a codon of species A close to the 5’ end, where itis directly covered by initiating ribosomes.We illustrate the different behaviour of each mRNA type by treating them in isolation. In each case we considermRNAs of length n = 200 ; the configurations are shown in figure 2(a). For simplicity we assume that the rechargingparameters for each tRNA species are the same and use experimentally measured values for a common aminoacyltRNA synthetase: the Leucine-tRNA ligase (EC 6.1.1.4) [33, 34] . Since the model only considers a subset of thetranscriptome, parameters and tRNA numbers are scaled in order to match the demand–supply ratio for tRNAs in awhole cell.In figure 2(b), we plot the normalized protein production rate λ as a function of [ aa A ] . We assume that this isconstant in time; i.e., we solve for different, constant values of [ aa A ] . We see a different response for each mRNA.For type I mRNAs, at large amino acid concentrations, λ remains constant as [ aa A ] is reduced. Then when acritical concentration is reached there is a sharp change to a regime where λ reduces as [ aa A ] is reduced (a first–orderphase transition). In a model without amino acid starvation a first–order phase transition has been reported to resultdue to queue formation when the initiation rate is increased through a critical value [14]. We show here that a similareffect is seen due to starvation. The lack of amino acid molecules affects protein production by reducing the tRNAcharging level [figure 2(c)] which reduces the translation rate of species A codons. As [ aa A ] is reduced translationthrough this bottleneck becomes the limiting process: a ribosome queue forms. Queue formation upstream of codonA therefore characterizes the translation limited regime [figure 2(d)]. It is only in this queuing regime (small [ aa A ] )that λ is a function of [ aa A ] ; otherwise the protein production rate is proportional to k α . For type II mRNAs, theprotein production rate varies smoothly with [ aa A ] . A decreased charging level again impairs translation of speciesA codons which leads to an increased occupation density near the 5’ end of the mRNA [figure 2(d)]. The transitionis smooth since the slow codon is at the initiation site, so there can be no sharp onset of queuing.In summary, for type I mRNAs there is a first order phase transition as [ aa A ] is reduced: we move from a regimewhere protein production is limited by the initiation rate k α to a regime where it is limited by the slow codon. Fortype II mRNAs the protein production rate is limited by both initiation and the slow codon; i.e., due to the fact thatthe first codon is slow, initiation of new ribosomes is hindered. Figure 2 demonstrates that amino acid starvation impairs the ribosome traffic. In this subsection we describe howthis decreased traffic can affect the charging levels of a tRNA species delivering an amino acid that is not subjectto starvation. In figure 3(a) we introduce two further idealized mRNAs which contain three codon species, and wedenote these mRNA X and mRNA Y. We again starve for amino acids of species A, but this time these tRNAs are5 [ aa A ] (mmol / L) G A ( − ) , G B ( −− ) mRNA X B [ aa A ] (mmol / L) G A ( − ) , G B ( −− ) mRNA Y [ aa A ] (mmol / L) n o r m a li ze d λ C mRNA XmRNA Y mRNA Xcodon j y j D mRNA Ycodon j y j A Figure 3: Steady state results from separate treatments of mRNA X and Y: (a) Schematic representation of bothmRNAs. The positions of the codons of species A (abundant tRNA but starved amino acid) and the codons ofspecies B (rare tRNA) are indicated. (b) Charging levels G A and G B . (c) Normalized protein production rate λ as afunction of [ aa A ] . (d) Codon occupation density at high (brown line) and low (blue line) amino acid concentrationsfor each mRNA. Although difficult to see in the plot, in the high [ aa A ] case for mRNA X there is a small peak inthe density at the B-type codon ( j = 100 ).in high abundance. Codons of species B require a rare tRNA which carries an amino acid species that is not subjectto starvation. The third codon species makes up the rest of the mRNA and corresponds to an abundant tRNA withconstant charging level.Under non-starvation conditions species B codons are the slowest, but the initiation rate is not high enough forqueuing to occur. The X mRNAs (having their slow B type codon in the bulk) are of type I, and the Y mRNAs(having their slow B codon near the initiation site) are of type II. To clarify our definitions, in non-starvation conditionsmRNAs are classified as type I or II depending on the position of the slowest codons.Upon amino acid starvation, the charging of the A type tRNAs is reduced; for small enough [ aa A ] , type A codonswill become the bottleneck. As in the previous subsection we again treat each mRNA sequence separately, i.e., oursystem consists of many copies of the same mRNA sequence. From plots of the tRNA charging levels as a function of [ aa A ] we observe two distinct phases [figure 3(b)]. At high amino acid concentrations, variation of [ aa A ] only affectsthe abundance of aa-tRNAs of species A. In this regime, ribosome traffic and protein production remain constant[figure 3(c)] as the availability of tRNAs corresponding to the slowest codons (species B) is unchanged. When thestarved amino acids become very rare the system enters a second regime where codons of species A become theslowest; i.e., due to starvation charged tRNAs of type A have become depleted, and this causes ribosomes to queue[figure 3(d)]. Since for both mRNAs the species A codon is in the bulk, in this regime both mRNAs are effectivelyof type I. That is to say, although under non-starvation conditions mRNA Y is of type II, under species A amino acidstarvation it shows a type I response - the codon which becomes slowest is far from the initiation site.Intriguingly, for both mRNAs in the low [ aa A ] regime the charging level of (the non-starved) type B tRNAincreases with decreasing amino acid concentration. Increased tRNA charging in response to amino acid starvationhas been observed experimentally [3], and our model now provides a theoretical explanation for this phenomenon(see below). Decreasing [ aa A ] causes queues to form upstream of species A codons, reducing the overall elongationrate. This reduces the rate at which species B codons are used by the ribosomes, resulting in an increased charginglevel of B type tRNAs.The difference between mRNA X and Y is that when a slow site is near the initiation site it effectively reduces theinitiation rate. Since queuing depends on the ratio of the initiation rate and the translation rate of the slowest codon(under starvation this is species A), for mRNA Y [ aa A ] must reach a lower value before we see queuing [figure 3(c)]. The above results, along with those of previous studies [35, 10, 14], indicate that specific codon usage plays asignificant role in determining the translational dynamics of an mRNA. However, the transcriptome is comprised ofthousands of different mRNAs [36], and ribosomes translating these mRNAs all use a common pool of aa-tRNAresources. In particular, when resources are sparse different codon usage might govern the effectiveness with whichribosomes translating different mRNA species can utilize these resources.6 [ aa A ] (mmol / L) G A ( − ) , G B ( −− ) A [ aa A ] (mmol / L) n o r m a li ze d λ B mRNA XmRNA Y Figure 4: The two mRNA configurations depicted in figure 3(a) are treated simultaneously and have to compete fora common pool of charged tRNAs. (a) Charging levels G A and G B , and (b) normalized protein production rate λ as a function of [ aa A ] .We use the two mRNA configurations introduced in figure 3(a) to explore this scenario in more detail. In contrastto the previous subsection, here both sequences are included in the same simulation, using the same aa-tRNA pool.The abundance of aa-tRNAs [figure 4(a)] follows similar dynamics as presented for separate translation: a decreasein the species A amino acid concentration results in a decreased charging level of A type tRNAs and an increasedcharging of B type tRNAs. However the response of the protein production rates from each mRNA species [figure 4(b)]stands in sharp contrast to that of the previous subsection [figure 3(c)]. There is now a regime where there is anincreased protein production rate from Y type mRNAs as a result of a decrease in the amino acid concentration.
Toour knowledge this is the first time that such an increase has been shown in a model of protein translation, and thisis a key result in the present work. In this regime, codons of species A are flow-limiting for the X mRNA; however forthe Y mRNA, due to its proximity to the initiation site it is still the codon of species B which is the bottleneck. Sincethe flow on the X mRNA is reduced, the rate at which ribosomes on this mRNA use type B tRNAs is reduced. Thusribosomes on Y mRNAs are able to use these type B tRNAs, and the protein production rate increases. As [ aa A ] is further reduced the system moves into the familiar regime where protein production from both mRNA speciesreduces; yet even here Y mRNAs maintain a higher (normalized) λ compared with their X counterparts.In summary we have shown that even with very simple designer mRNAs the codon configuration can significantlyaffect the protein production performance. Real mRNA sequences consist of codons representing up to 41 tRNAtypes: they are inherently more complicated. This means that there is even more opportunity for control of proteinproduction given different starvation conditions. For example it might be possible to choose a sequence such thattranslation of a particular transcript is enhanced in response to starvation of one amino acid species, but is reducedin response to starvation of another. In order to investigate whether the qualitative behaviour obtained for the mRNA types (I and II) which we identifiedabove also emerge for realistic codon configurations in the context of their response to starvation, in this sectionwe consider the translation of two yeast mRNAs (YLR382C and YHR208W) during leucine starvation. Thesesequences code for proteins taking part in tRNA charging and amino acid metabolism, respectively: the leucine-tRNAligase (LeuRS, EC 6.1.1.4), which charges tRNAs with leucine; and the branched-chain-amino-acid transaminase(Transaminase, EC 2.6.1.42), which degrades leucine. Note that we do not consider the function of the proteins inour model but we can speculate on why it might be beneficial to the cell to have a different response for differentproteins.Figure 5(a) shows results for each of the two real mRNA sequences treated separately. We can clearly identifydifferent kinds of behaviour in each case during starvation for leucine. Protein production of the LeuRS mRNAshows a sharp phase transition indicating type I behaviour, whereas translation of the transaminase follows type IIcharacteristics. Hence in the model, a decrease in the leucine concentration directly affects the production rate ofthe leucine transaminase - we speculate that this is a desired effect, since this will reduce further leucine degrada-tion. Conversely, translation of the mRNA for the leucine recharging enzyme is robust to changes in the leucineconcentration down to a much lower level.In figure 5(b) we plot the charging levels of the four different leucine tRNAs for the case of LeuRS mRNAtranslation. Our results resemble previous experimental data [2] in that the four tRNA species are differently charged.Specifically the rare tRNA species Leu3 and Leu4 become rapidly discharged in response to starvation, whereas theabundant tRNAs (Leu1 and Leu2) maintain a higher charging level. This pattern is determined solely by theconcentration of the isoacceptors as we have not included any additional selective recharging in the model.If we consider ribosome traffic [figure 5(c)], we observe that the reduced availability of aa-tRNAs again leads toribosome queue formation. We show on the plot the positions of the two codon species which are decoded by theLeu3 tRNA, the charging level of which becomes most depleted. These are the CUC codon, which matches the Leu37 y j C codon j y j [ Leu ] (mmol / L) n o r m a li ze d λ A LeuRSTransaminase 0 0.05 0.1 0.15 0.200.20.40.60.81 [ Leu ] (mmol / L) G B Leu1Leu2Leu3Leu4
Figure 5: Steady state results from separate treatments of the mRNAs coding for the proteins YLR382C (LeuRS)and YHR208W (Transaminase): (a) Normalized protein production rate λ as a function of the leucine concentration.(b) Charging levels of the four leucine-tRNA isoacceptors (Leu1 (UUA), Leu2 (UUG), Leu3 (CUC, CUU) and Leu4(CUA, CUG)) for the case of LeuRS mRNA translation. (c) Codon occupation density of the LeuRS mRNA at low(upper panel) and high (lower panel) amino acid concentrations. Using dots along the top of the plots, we indicatethe positions of the two types of codon which are decoded by the aa-tRNA which becomes most depleted, i.e., Leu3(CUU in red and CUC in green).anticodon exactly, and the CUU codon which due to wobble base pairing is translated even more slowly [23]. As noted previously, recent experiments on amino acid starvation have uncovered unexpected behaviour which canbe explained by the above results. Dittmar et al. [2] find that different isoaccepting tRNAs for the same (starved)amino acid show different levels of charging. We have shown above that this can occur naturally due to differentabundances of the tRNAs and codons, i.e., the balance between supply and demand. This is consistent with theprevious prediction of [37].Zaborske et al. [3] investigated starvation for different amino acids in vivo using micro array techniques. Theyfound that the charging level for the staved amino acids decreases, as do charging levels of some other tRNAs(which they show is due to the general amino acid control pathway). Unexpectedly some tRNA species also showan increase in charging level; as we have shown here this could result from decreased rate of use of aa-tRNAs dueto bottlenecks forming on some mRNAs. We propose that a method for testing whether this behaviour does occurdue to community competition for resources would be an in vitro experiment where only the translational machineryis present, i.e., without other control pathways.
We have developed a model of protein translation which considers variable codon translation rates that depend onthe availability of aa-tRNAs. The framework explicitly accounts for aminoacylation and allows us to examine theinfluence of amino acid starvation on ribosome traffic. It leads to a new method of protein regulation where the useof slow codons determines how translation of an mRNA responds to changes in the availability of resources in thewider community.In the past, stochastic models have revealed the existence of two distinct mRNA types depending on the positionof the slowest codon, as well as a relationship between the mRNA type and the protein function [14]. Using simpleartificial mRNA sequences, we have found the same two types to yield different responses to amino acid starvation.An mRNA showing a type I response to starvation of a particular amino acid is, in terms of protein production,robust against fluctuations in amino acid concentration provided it remains above a critical level. In contrast, thetranslation of mRNAs showing a type II response is impaired by any change in amino acid concentration. The typeof response of a given mRNA may depend on which amino acid is subject to starvation.The two principal response types were also identified in realistic yeast mRNAs. We therefore propose thatproteins can be categorized with respect to starvation for a specific amino acid: those which are robust to changes inamino acid concentration, and those which are highly affected. We have presented a more natural example where theproduction of a protein involved in aminoacylation is robust, but the production of a protein responsible for degradingthat amino acid responds strongly. Although we cannot draw any firm conclusions from our limited simulations, onecould hypothesize that the production of proteins crucial to cell function could be prioritized over that of othersusing this mechanism. This would be consistent with the correlation between slow codon arrangement and proteinfunction reported in [14], as well as other recent studies which show that many mRNA sequences have slow codonspositioned close to the initiation site [38, 39]. 8hen considering a codon species whose tRNA delivers an amino acid which is not subject to starvation, we findthat the charging level of this species can in fact increase in response to starvation. This agrees with previous studieswhich suggest that tRNA charging levels depend on the balance between their supply and demand [37, 40, 20], andthe phenomenon has also been observed experimentally [3]. Reduced protein production from some mRNAs can freeup resources, which can then be used by other mRNAs leading to increased protein production.In contrast to previous models we have explicitly taken the codon sequence into account. Our description is,however, deterministic and is essentially a mean field treatment. This latter point refers to the fact that since y i refers to an average over mRNAs (see Materials and Methods), we do not take into account any spatial correlationsin the density. Although correlations are likely to have a large impact on density profiles around bottleneck sites,by comparing with stochastic models which do take correlations into account [41, 19] we predict that the overallbehaviour is unlikely to change qualitatively. We have also neglected any active variation in the initiation rate whichmay result as the cell actively responds to starvation: for example, the GCN pathway [31, 4] gives rise to a globalreduction in initiation rate, while other pathways act to increase or reduce translation of specific mRNA species.Nevertheless, by keeping k α constant in this study we have been able to identify a translation control methodbased on tRNA and codon usage. Whilst taking into account variation in initiation rates (and also variation of theabundance of different mRNA species [42]) would no doubt lead to a more complicated response to starvation, onemight expect an even more important role for tRNA charging dynamics as regulation at different levels changes thepattern of supply and demand [20].In summary, we find that counter-intuitive effects emerge when considering amino acids as a constrained resource.Starvation can lead to different tRNA species having decreased or increased charging levels depending on the codonconfiguration of all transcribed mRNAs. This mechanism opens a new perspective on protein production controlwhich emerges naturally at a community level due to the complex nature of the system. Acknowledgements
The authors thank M. C. Romano, I. Stansfield, L. Ciandrini, A. Kort, and M. Rehberg for helpful discussions. Thiswork was funded by BBSRC grants BB/F00513/X1 and BB/G010722, and the Scottish Universities Life ScienceAlliance (SULSA).
References [1] Alberts, B., Johnson, A., Lewis, J., Raff, M., Roberts, K., and Walter, P. (2002) Molecular Biology of the Cell,Fourth Edition, Garland Science, 4th edition.[2] Dittmar, K. A., Sørensen, M. A., Elf, J., Ehrenberg, M., and Pan, T. (2005) Selective charging of tRNAisoacceptors induced by amino-acid starvation.
EMBO Rep., , 151–157.[3] Zaborske, J. M., Narasimhan, J., Jiang, L., Wek, S. A., Dittmar, K. A., Freimoser, F., Pan, T., and Wek,R. C. (2009) Genome-wide analysis of tRNA charging and activation of the eIF2 kinase Gcn2p. J. Biol. Chem., (37), 25254–25267.[4] Hinnebusch, A. G. (2005) Translational regulation of GCN4 and the general amino acid control of yeast.
Annu.Rev. Microbiol., , 407–450.[5] Sonenberg, N. and Dever, T. E. (2003) Eukaryotic translation initiation factors and regulators. Curr. Opin.Struct. Biol., , 56–63.[6] Nakamoto, T. (2009) Evolution and the universality of the mechanism of initiation of protein synthesis. Gene, , 1–6.[7] Ibba, M. and Söll, D. (2004) Aminoacyl-tRNAs: setting the limits of the genetic code.
Genes Dev., , 731–738.[8] Delarue, M. and Moras, D. (1993) The aminoacyl-tRNA synthetase family: modules at work. Bioessays, ,675–687.[9] Ibba, M. and Söll, D. (2000) Aminoacyl-tRNA synthesis. Annu. Rev. Biochem., , 617–650.[10] Sørensen, M. A., Kurland, C. G., and Pedersen, S. (1989) Codon usage determines translation rate in Escherichiacoli. J. Mol. Biol., , 365–77.[11] Varenne, S., Buc, J., Lloubes, R., and Lazdunski, C. (1984) Translation is a non-uniform process. Effect oftRNA availability on the rate of elongation of nascent polypeptide chains.
J. Mol. Biol., , 549–576.912] Ikemura, T. (1982) Correlation between the abundance of yeast transfer RNAs and the occurrence of therespective codons in protein genes. Differences in synonymous codon choice patterns of yeast and Escherichiacoli with reference to the abundance of isoaccepting transfer RNAs.
J. Mol. Biol., , 573–597.[13] Ikemura, T. (1985) Codon usage and tRNA content in unicellular and multicellular organisms.
Mol. Biol. Evol., , 13–34.[14] Romano, M. C., Thiel, M., Stansfield, I., and Grebogi, C. (2009) Queueing phase transition: Theory of trans-lation. Phys. Rev. Lett., , 198104.[15] MacDonald, C. T., Gibbs, J. H., and Pipkin, A. C. (1968) Kinetics of biopolymerization on nucleic acid templates.
Biopolymers, , 1–5.[16] Shaw, L. B., Zia, R. K. P., and Lee, K. H. (2003) Totally asymmetric exclusion process with extended objects:A model for protein synthesis. Phys. Rev. E, , 021910–1–17.[17] Basu, A. and Chowdhury, D. (2007) Traffic of interacting ribosomes: Effects of single-machine mechanochem-istry on protein synthesis. Phys. Rev. E, .[18] Brackley, C. A., Romano, M. C., Grebogi, C., and Thiel, M. (2010) Limited resources in a driven diffusionprocess. Phys. Rev. Lett., , 078102.[19] Brackley, C. A., Romano, M. C., and Thiel, M. (2010) Slow Sites in an Exclusion Process with Limited Resources.
Phys. Rev. E, , 051920.[20] Brackley, C. A., Romano, M. C., and Thiel, M. (2011) The Dynamics of Supply and Demand in mRNATranslation. PLoS Comput. Biol., , e1002203.[21] Kremling, A., Bettenbrock, K., and Gilles, E. D. (2007) Analysis of global control of Escherichia coli carbohydrateuptake. BMC Syst Biol, , 42.[22] Fluitt, A., Pienaar, E., and Viljoen, H. (2007) Ribosome kinetics and aa-tRNA competition determine rate andfidelity of peptide synthesis. Comput. Biol. Chem., , 335–346.[23] Gilchrist, M. A. and Wagner, A. (2006) A model of protein translation including codon bias, nonsense errors,and ribosome recycling. J. Theor. Biol., (4), 417–434.[24] Percudani, R., Pavesi, A., and Ottonello, S. (1997) Transfer RNA gene redundancy and translational selectionin Saccharomyces cerevisiae.
J. Mol. Biol., , 322–330.[25] Lakatos, G. and Chou, T. (2003) Totally asymmetric exclusion processes with particles of arbitrary size.
J. Phys.A: Math. Gen., , 2027–2041.[26] Heinrich, R. and Rapoport, T. (1980) Mathematical modelling of translation of mRNA in eucaryotes; steadystates, time-dependent processes and application to reticulocytest. J. Theor. Biol., , 279–313.[27] Arava, Y., Wang, Y., Storey, J. D., Liu, C. L., Brown, P. O., and Herschlag, D. (2003) Genome-wide analysisof mRNA translation profiles in Saccharomyces cerevisiae. Proc. Natl. Acad. Sci. USA, , 3889–3894.[28] Chassagnole, C., Fell, D. A., Raïs, B., Kudla, B., and Mazat, J. P. (2001) Control of the threonine-synthesispathway in Escherichia coli: a theoretical and experimental approach.
Biochem. J., , 433–444.[29] Elf, J., Berg, O. G., and Ehrenberg, M. (2001) Comparison of repressor and transcriptional attenuator systemsfor control of amino acid biosynthetic operons.
J. Mol. Biol., , 941–954.[30] Schmidt, H. and Jirstrand, M. (2006) Systems Biology Toolbox for MATLAB: a computational platform forresearch in systems biology.
Bioinformatics, , 514–515.[31] You, T., Stansfield, I., Romano, M. C., Brown, A. J., and Coghill, G. M. (2011) Analysing GCN4 translationalcontrol in yeast by stochastic chemical kinetics modelling and simulation. BMC Systems Biology, , 131.[32] Sha, Z., Brill, L. M., Cabrera, R., Kleifeld, O., Scheliga, J. S., Glickman, M. H., Chang, E. C., and Wolf,D. A. (2009) The eIF3 Interactome Reveals the Translasome, a Supercomplex Linking Protein Synthesis andDegradation Machineries. Mol. Cell, , 141–152.[33] Chirikjian, J. G., Kanagalingam, K., Lau, E., and Fresco, J. R. (1973) Purification and properties of leucyl-tRNAsynthetase from Bakers’ yeast. J. Biol. Chem., , 1074–1079.1034] Nawaz, M. H., Pang, Y. L. J., and Martinis, S. A. (2007) Molecular and functional dissection of a putativeRNA-binding region in yeast mitochondrial leucyl-tRNA synthetase.
J. Mol. Biol., , 384–394.[35] Robinson, M., Lilley, R., Little, S., Emtage, J. S., Yarranton, G., Stephens, P., Millican, A., Eaton, M., andHumphreys, G. (1984) Codon usage can affect efficiency of translation of genes in Escherichia coli.
NucleicAcids Res., , 6663–6671.[36] Hereford, L. M. and Rosbash, M. (1977) Number and distribution of polyadenylated RNA sequences in yeast. Cell, , 453–462.[37] Elf, J., Nilsson, D., Tenson, T., and Ehrenberg, M. (2003) Selective charging of tRNA isoacceptors explainspatterns of codon usage. Science, , 1718–1722.[38] Tuller, T., Carmi, A., Vestsigian, K., Navon, S., Dorfan, Y., Zaborske, J., Pan, T., Dahan, O., Furman, I., andPilpel, Y. (2010) An Evolutionarily Conserved Mechanism for Controlling the Efficiency of Protein Translation.
Cell, , 344.[39] G, C., NN, S., M, F., von Rohr P, MT, F., AC, R., P, G., G, G., and Y, B. (2010) A role for codon order intranslation dynamics.
Cell, , 355.[40] Elf, J. and Ehrenberg, M. (2005) Near-critical behavior of aminoacyl-tRNA pools in E. coli at rate-limitingsupply of amino acids.
Biophys. J., , 132–146.[41] Derrida, B., Domany, E., and Mukame, D. (1992) An Exact Solution of a One-Dimensional Asymmetric ExclusionModel with Open Boundaries. Journal of Statistical Physics, , 667–687.[42] Gingold, H. and Pilpel, Y. (2011) Determinants of translation efficiency and accuracy. Mol Syst Biol.,7