The molecular aetiology of tRNA synthetase depletion: induction of a GCN4 amino acid starvation response despite homeostatic maintenance of charged tRNA levels

Abstract During protein synthesis, charged tRNAs deliver amino acids to translating ribosomes, and are then re-charged by tRNA synthetases (aaRS). In humans, mutant aaRS cause a diversity of neurological disorders, but their molecular aetiologies are incompletely characterised. To understand system responses to aaRS depletion, the yeast glutamine aaRS gene (GLN4) was transcriptionally regulated using doxycycline by tet-off control. Depletion of Gln4p inhibited growth, and induced a GCN4 amino acid starvation response, indicative of uncharged tRNA accumulation and Gcn2 kinase activation. Using a global model of translation that included aaRS recharging, Gln4p depletion was simulated, confirming slowed translation. Modelling also revealed that Gln4p depletion causes negative feedback that matches translational demand for Gln-tRNAGln to aaRS recharging capacity. This maintains normal charged tRNAGln levels despite Gln4p depletion, confirmed experimentally using tRNA Northern blotting. Model analysis resolves the paradox that Gln4p depletion triggers a GCN4 response, despite maintenance of tRNAGln charging levels, revealing that normally, the aaRS population can sequester free, uncharged tRNAs during aminoacylation. Gln4p depletion reduces this sequestration capacity, allowing uncharged tRNAGln to interact with Gcn2 kinase. The study sheds new light on mutant aaRS disease aetiologies, and explains how aaRS sequestration of uncharged tRNAs can prevent GCN4 activation under non-starvation conditions.

in our downscaled cell, we have on average 2500 ribosomes available for initiation in steady state.
Note that the simulation starts with empty mRNAs, i.e., initially, all ribosomes are free.
The initial initiation rates are listed in Table S2 for the downscaled cell, with the average initiation rate amounting to ⟨ ⟩ = 0.15⁄ .
The maximum charging rate for the tRNA-synthetases *01,, is estimated by dividing the total number of each amino acids incorporated into proteins during the cell lifetime by the abundance of the corresponding tRNA-synthetases (7) (8). Interestingly, the *01,, values highly correlate with the tRNA gene copy number. The values of the Michaelis-Menten constants H,, are not known for all synthetases, but the fact that the charging level of tRNAs under physiological conditions has been estimated to be 80% across different tRNA species (9)(10)(11)(12)(13) implies that the ratio = *01,, ⁄ H,, is similar for all different synthetases (see Table S4). We use the 80% charging level to estimate this ratio in the simulations (we use the charging capacity of = *01,, H,, ⁄ = 5⁄ ). Note that this ratio is proportional to the aminoacylation efficiency .0-,, H,, ⁄ .
Considering wobble base pairing via the wobble factor ( ) for each codon , we reduce the hopping rates of the codons which use the G-U wobble by 39% compared to their G-C wobble counterparts, and those of codons using the I-G wobble by 36% relative to their I-U counterparts, based on the results in (14). In addition to these choices, a 60% reduction has been introduced for the so-called 'missing tRNAs' which are non-perfect matches and which do not have a supplement (15). All wobble examples are listed together with the initial hopping rates in Table S3 for the downscaled cell. given system size, we can now retrieve the factor of proportionality , and for the considered downscaled system we obtain = 3.696 × 10 Y> . Note that although the average value does not depend on the system size and does not change throughout the simulation, the hopping rate in every single Gillespie step depends on the number of charged tRNAs of the corresponding type.
The average codon length of mRNA for each GO-Slim class was calculated. The ratio between total number of codons within this category and the number of mRNAs defines the average length of the single representative mRNA of this category. Then we determined the average number of each codon on the representative mRNA by dividing the number of codons of that type by the number of mRNAs within this category. Note that doing so, we neglected codons due to rounding effects, but the loss is less than 0.5%.
The code for the Global Translation Model is available through https://www.ebi.ac.uk/biomodels/ (23) with the following submission identifier following submission identifier: MODEL2001080004.
The code is written in C language and adapts some of the structures developed in (22) to access the codons of the mRNA lattices.

Synthetase Sequestration Model
The synthetase sequestration model includes four main transitions for each amino acid type i: the binding rate , , the charging rate ^, , the usage rate ^, , and the synthetase-tRNA-complex dissociation rate _ . We start at time = 0 with 20 amino acids, with a starting configuration where all tRNAs are empty, i.e., the number of empty tRNAs is ,`= , and the number of bound and charged tRNAs is zero. To ensure steady state results, we allow the transient time -3045 = 5 × 10 a to pass before we start to record the number of charged, bound and empty tRNAs of each amino acid type.
The average charging level ⟨ , ⟩ is equal to the number of charged tRNAs of type i, weighted by the time the individual tRNAs of type i stay in the charged state, divided by the total number of tRNAs of this type and the total run time. Likewise, we obtain the bound level ⟨ , ⟩ and the empty level ⟨ , ⟩.
The synthetase sequestration simulation is carried out for the whole cell, i.e., we use the values from Table S4 for the number _,, of synthetase molecules and the catalytic rate .0-,, . The number of synthetase molecules reflects the total amount of synthetase in the cell, i.e, the free synthetase molecules and the ones captured within the synthetase-tRNA-complexes. However, the binding rates depend on the number of free synthetase molecules, i.e, , =^_ ,, − , f . Note that it is the total amount which is affected by doxycycline for the glutamine case: ^_ ,gh4 = _,gh4 with the Gln4 protein ratio .
The code for the Synthetase Sequestration Model is available through the BioModels database https://www.ebi.ac.uk/biomodels/ (23) with the following submission identifier: MODEL2001080005

Specification of parameters; Synthetase Sequestration Model
The usage rate constant > = 0.1⁄ is estimated to represent the average translation rate. The charging rate constants j,, = .0-,, are fitted to comply an average charging level of 80%, which leads to the factor of proportionality of = 0.2. The .0-,, and the enzymatic concentrations of the synthetases _,, are chosen analogously to the global translation model (Table S4). Note that as a matter of consistency, we adjust the number of glutamine synthetase molecules to the construct, _,gh4 ≈ 10 ; .
The rate constants _ = 43⁄ and n,gh4 = 0.007⁄ are taken from Uter (20).  (20). Note that in the stochastic simulation we reach a steady state in the bound level, since we never run out of substrate, as the tRNAs are replenished throughout the cycle. Additionally, the dissociation rate constant _ ≫ j,, is much larger than the charging rate constants, and therefore, the conditions are fulfilled for the approximation to be appropriate.

General expression for the usage rate
For each tRNA the = ∑ v v,, v is proportional to the ribosomal current v along the mRNA of type , the number v of copies of mRNA of type , the number v,, of codons on the mRNA of type decoded by tRNA of type .
Every time a ribosome hops with rate O from one codon to the next, a charged tRNA is used. Together with the ribosomal density O , the usage rate can be expressed by ,v of the -| codon which is decoded by tRNA on mRNA of type .

Autogenous Feedback
For the Global Translation Model (GTM) autogenous feedback we analysed different effects that reduced growth rate has on the initiation rate. In the GTM, the initiation rate is given by = _~3``, where ~3`` denotes the number of free ribosomes in the cytoplasm (not engaged in translation), and _ is a proportionality constant that comprises other effects on the initiation, such as the availability of initiation factors or secondary structures on the 5' UTR. In principle, in response to stress we expect that both _ and ~3`` are reduced ( _ decreases because the number of initiation factors can be assumed to decrease with the growth rate, and ~3`` decreases because ribosomes are coupled to the growth rate). Given the experimental data on how the number of ribosomes decreases with Gln4p ( Fig. 6), we tested what are the effects of the reduction of _ and ~3`` on the simulation results.
Using the data in Figure 6 for calibration of the ribosomal reduction, we obtain the simulation results shown in Fig. S5: when comparing these results to the ones obtained with no feedback (blue solid line), we see that the effect of ribosome reduction is very minor. However, the reduction in the initiation factors availability and therefore _ (assumed to decrease proportionally to the growth rate) has a marked effect (shown as a solid red line in Fig. S5), especially on the charging level of the Glutamine codons. Therefore, for the sake of simplicity, we have now only considered the effect of the reduction in _ .
Moreover, we analysed the effects of tRNA reduction with decreasing growth rate, since in principle not only ribosomes but also tRNAs are downregulated under stress, e.g., through the TOR pathway (21). In order to assess the effect of downregulation of tRNA on our results, we have tested the effects of a tRNA feedback in the GTM, by reducing the amount of tRNA proportionally to the experimentally obtained reduction in ribosomal content. The results are shown in Fig. S5. The green dashed line shows the results of the model incorporating the tRNA feedback. By comparing these results with the ones obtained with the GTM with no feedback (blue solid line), we see that the reduction of the number of tRNAs under stress has only a very minor effect. Therefore, we have neglected it for the sake of simplicity and do not consider tRNA reduction in the rest of the model simulations.

Correlation analysis of the current and the glutamine content
To address the question whether the mRNAs with a high glutamine codon density are more sensitive to increased doxycycline concentration, we did a correlation analysis of the ratio 0 (+ )/ 0 (− ) of the current with and without doxycycline of the GO-Slim category a with the glutamine codon content per mRNA within this GO-Slim category.
In the GTM with no feedback, there is a strong correlation between 0 (+ )/ 0 (− ) and the CAG codon content (Spearman's correlation coefficient of -0.88). Depletion of charged Gln tRNAs causes elongation arrests, and therefore, a reduction in the current.
With autogenous feedback, on the other hand, there is a strong reduction in the correlation with the glutamine content (Spearman's correlation coefficient of 0.22), which is not surprising, because here, the balance between supply and demand is restored; translation initiation rate is decreased so that the charged Gln tRNA available can keep up with the demand.
Note that there is a weaker correlation between reduction in current and the non-rare CAA codon content per mRNA (Spearman's correlation coefficients -0.76 without feedback and 0.18 with feedback).

Synthetase sequestration at different charging levels
As a matter of consistency with the literature, we used the average 80% tRNA charging level as data to parameterise our model. If we instead use 60% tRNA charging level, as suggested by the experimental results in Fig. 7, the results remain qualitatively the same as shown in Fig. S1A.
The glutamine charging level stays more or less constant whereas the empty level increases with increasing doxycycline concentration (decreasing Gln4 protein ratio) at the expense of the bound level.
Note that a smaller global charging level of the cell leads to a smaller global current. Therefore, the response on doxycycline evolves slightly differently. We use the GTM to inform the SSM how the threshold for the doxycycline sensitive response of the usage rate is shifted; for 80% charging the feedback for the SSM is apparent at Gln4 protein ratio 0.2 which corresponds to the threshold current -|3`5| = 0.03/ (see Fig. S1B). For 60% charging, however, this value of the current is already reached at the Gln4 protein ratio of 0.35, as shown by the red arrows in Fig. S1B. Figure S1 shows the profile of the ribosomal density along an mRNA from GO Slim category 66, 'regulation of transport' (bottom), which has a large abundance of slow codons (19). In contrast, category 66 has only two mRNAs in the downscaled cell, but with an average length of 920 codons per mRNA and 8.6 CAG codons on each mRNA, see Table S10, below. category 16 66 The position of the glutamine codons on the mRNA in (Fig. S4) is highlighted by the grey dashed bars.

Density profiles
The black line represents the situation without doxycycline. Both, the blue and the red lines represent the situation in the presence of doxycycline, corresponding to a Gln4 protein ratio of 5%, with (red line) and without (blue line) feedback loop.
Without feedback and in the presence of doxycycline, the ribosome density profile in (Fig. S4) is highly inhomogeneous (blue line); ribosomal queues build up behind glutamine codons. When the feedback mechanism is switched on, the profile is much smoother and the queues vanish (red dashed line).
In category 16, the ribosome density along the mRNA in the case with doxycycline is large compared to the case without doxycycline (see figure 6H), whereas in category 66, with doxycycline the ribosomal density is also large at the entrance of the lattice but decreases rapidly along the mRNA, due to the bottleneck created by the accumulation of rare codons.
In contrast, when the autogenous feedback mechanism is activated, ribosomes are equally distributed in a low density-like regime all over the mRNA, no queueing is visible for any of the two categories.
Note that with feedback, the average density in both categories decreases considerably in the presence of doxycycline, which is also reflected in the protein production rate, see (Fig. 6D) in the main text.
Note that the ribosomal particles in the simulation have a footprint of = 9, which leads to the plateau peaks obtained in the case without feedback. Note further that here, the translational site of the ribosome is on the left side, which produces the overhang peaks.     Figure S7; Hygromycin resistance of a wild-type strain and the Gln tRNA synthetase tet-off strain.
Panel A-C: the resistance to hygromycin of the GLN4 tet-off strain was compared to its progenitor wild-type strain BGY2 by measuring mid-log phase growth rates in cells growing in growing in YPD medium containing doxycycline at either 0, 0.04 and 0.08 µg/ml. Panel D:, this data was processed for each concentration of doxycycline to show the growth rates as a percentage of those obtained in 0 µg/ml hygromycin. Cells were grown at 30°C, 400 rpm, in 96-well plates in a LabTech International Omega plate reader. Expressing mid-log phase growth rates in the presence, or absence of hygromycin indicated that depletion of the Gln4p tRNA synthetase did not render the cells more sensitive to hygromycin relative to the effect of hygromycin on wild-type cells (panel D).         Table S9: Transcriptional response of translation initiation factor genes significantly repressed in the GLN4 tet-off strain in response to doxycycline treatment Notes (a); Significantly repressed (green shading, bold font) and induced genes (orange shading) are indicated.
(b); A false discovery rate (FDR) of 0.05 was indicated the significance or otherwise of an adjusted q value (Benjamini-Hochberg).