ABSTRACT: Work summary: Full-transcriptome methods have brought versatile power to protein biosynthesis research, but remain difficult to apply for the quantification of absolute protein synthesis rates. Here we propose and, using modified translation complex profiling, confirm co-localisation of ribosomes on messenger(m)RNA resulting from the ribosomal diffusional dynamics. We demonstrate that the stochastically co-localised ribosomes are linked with the translation initiation rate and provide a robust variable to model and quantify specific absolute protein output from mRNA. The new type of signal originating from the stochastic co-localisation of ribosomes on mRNA is evidenced by long disome-derived (potentially but rarely, multisome-derived) footprints outside the regions of non-random elongation stalls. Using the stochastic co-localisation component of the translation complex profiling data, we propose a new approach to assess absolute translation rates across mRNA, based on the calculation of Stochastic Translation Efficiency (STE) measure. STE employs a machine learning model trained with multiple TCP-seq-sourced variables. The variables, among others, include stochastic disome signal derived from the total disome footprint data and normalised to the ribosome (RS) footprint abundance across the respective open reading frames (ORFs). The variables also include SSU footprint occurrences over the respective start codons normalised to the RS signal averaged and normalised by the ORF length, to better distinguish between the cases of actively initiated or densely populated but slowly initiated mRNAs. Importantly, STE does not employ any variables resulting from normalisation of the signals of different nature, such as any normalisation of the footprint data by relative RNA abundance measured by RNA-seq. We propose STE as a more robust, sometimes more convenient, alternative to the classical TE (translation efficiency). The main distinguishing features of the STE are the inherent single-molecule-event nature of the main measure component (the long stochastic disome footprint signal), similar type of data used for the numerator(s) and denominator(s) of the calculated variables, resistance to the abundance changes and variations in the differential localisation of RNA and insensitivity to the library preparation techniques. We further hypothesise that similar high-throughput data-derived measures based on parameters dependent on random individual molecular co-incidences can be utilised to provide robust assessment of many molecular processes in dynamic biological systems. Methods summary: Wild-type (WT) yeast cell line BY4741 (MATa his3Δ1 leu2Δ0 met15Δ0 ura3Δ0; Dharmacon) (Non Starved, NS) was used for the glucose starvation (10 minutes; Starved, S10) studies. 2-3 replicates were used for each type of footprinting library and the NS and S10 biological points. The cells were subjected to snap-chilling followed by rapid formaldehyde fixation. The fixed material was separated into the ribosome-associated and non-associated fractions using ultracentrifugation. The ribosome-associated fraction was RNase-digested and separated into small ribosomal subunit (SSU), monosomal (ribosomes, RS), and disomal (disomes, DS) fractions of complexes using ultracentrifugation. The resultant translational complexes were de-crosslinked, and their RNA isolated to construct TCP-seq libraries, which were deep sequenced. Results summary: Preliminary theoretical investigations revealed monotonous dependence of the stochastic DS frequency on translation initiation rate in a broad range of initiation and elongation rates and variances, well-covering all biologically meaningful combination of parameters. Biomimetic in silico modelling utilising elongation and initiation rate distributions extrapolated from in vivo single-molecule measurements available in the published data confirmed the link of stochastic DS signal with the initiation frequency. The in silico-modelled stochastic DS frequency has shown resistance to local alterations of the elongation rate over ORFs and monotonous dependence on the initiation rate, even in the extreme cases of elongation rate discontinuity. DS footprint fractions derived from the TCP-seq material prominently demonstrated the presence of longer footprints, and revealed a complex pattern of footprint lengths. The TCP-seq DS footprint pattern allowed us to propose a generalised case of ribosomal co-localisation, whereby DS can form as a result of non-random co-localisations of serially occurring ribosomes, as well as co-localisations resulting from the spatial proximity and the geometry and spatial arrangement of the respective source poly(ribo)somes. Bioinformatically segregating all of the DS signal into its stochastic and non-random components, and combining the derived components with the other TCP-seq-sourced variables and known ORF lengths, we further implemented a machine learning approach that results in Stochastic Translation Efficiency (STE) measure with high correlation (Pearson’s r ~0.7 for train and ~0.6 for test data, respectively) to the absolute protein synthesis rates as directly measured in the published data using metabolic labelling and mass spectrometry. Applying STE to the NS and S10 cell stress scenarios, we identified mRNAs from 5,238 genes with distinct translational behaviour. Multiple classes of mRNAs demonstrated acute (~50) and moderate (~50) translational up-regulation in the starved cells, while others were moderately (~300) and severely (~280) down-regulated. We dissected TCP-seq variable contribution strength into the STE, revealing strong input and role of the scanning process and stochastic co-localisation of ribosomes in the definition of translation rate. We found that the stressed cells retained high median elongation rate capacity, while losing their ability to translate proteins with the maximum rate, thus highlighting that the safety translation shutdown is permissive of translation and is largely aimed at capping the highest-producing mRNAs. Using STE, we demonstrated that translational control in rapid stress response is vastly differential, with mRNAs showing highly specific magnitude and direction of the response. Conclusions: The stochastic disome signal, together with the other measurements sourced from TCP-seq and calculated as Stochastic Translation Efficiency (STE) measure, provide a link to the absolute translation initiation and protein biosynthesis rates. Using STE, it is possible to rank mRNAs by the absolute protein output and thus, characterise the ‘power’ of translation control elements across transcripts in a single setting or between different conditions. STE does not use bias-inducing normalisation to the RNA abundance or signals of different types and relies on self-normalised signal pairs. STE thus evades imprecisions arising from the different accessibility across mRNAs to translation machinery (such as in nucleus, phase-separated foci or condensates, etc.) and coincidental transcription- and stability-driven mRNA abundance alterations, and provides a robust measure suitable for investigating protein biosynthesis during rapidly changing transcriptome background, as occurs in cell stress response and reprogramming. With STE, we demonstrate that the prototypical example of translational control during yeast response to glucose starvation is more complex than previously thought and includes translational acceleration as well as re-adjustment of the RNA metabolism to suit new translational demands. More details can be uncovered in the future, as STE application is permissive to finer dynamics dissection and elucidation of very rapid cell responses at the level of RNA translation.
