Pervasive, Coordinated Protein-Level Changes Driven by Transcript Isoform Switching during Meiosis

Abstract

To better understand the gene regulatory mechanisms that program developmental processes, we carried out simultaneous genome-wide measurements of mRNA, translation, and protein through meiotic differentiation in budding yeast. Surprisingly, we observed that the levels of several hundred mRNAs are anti-correlated with their corresponding protein products. We show that rather than arising from canonical forms of gene regulatory control, the regulation of at least 380 such cases, or over 8% of all measured genes, involves temporally regulated switching between production of a canonical, translatable transcript and a 5' extended isoform that is not efficiently translated into protein. By this pervasive mechanism for the modulation of protein levels through a natural developmental program, a single transcription factor can coordinately activate and repress protein synthesis for distinct sets of genes. The distinction is not based on whether or not an mRNA is induced but rather on the type of transcript produced.

Keywords: LUTI; coordination; differentiation; gene expression; isoform; meiosis; ribosome profiling; transcription factor; translation; uORF.

Figure 1

Gene expression through meiotic differentiation, from mRNA to protein. A) Matched extract was used for mRNA-seq, ribosome profiling, and quantitative mass spectrometry. B) Hierarchical clustering of protein measurements for all quantified annotated genes (n=4464, columns) over all timepoints (rows) is shown. Total signal is normalized per column to allow comparison of patterns. See also Fig. S1, Tables S1, S3, and S6.

Figure 2

Many genes show a poor correlation between mRNA and protein in meiotic differentiation that is associated with alternate transcripts. A) A histogram of the Pearson correlation coefficients between mRNA and protein abundance measurements over all timepoints for all genes is shown. Note the general skew towards a positive correlation and a subdistribution (centered at ~−0.2) with a poor correlation. B) Mean levels of mRNA (blue) and protein (black) for each of the genes shown in the distribution above is shown. C) Protein abundances for Ndt80 targets. Columns are genes, rows are timepoints. Shaded boxes below denote discrete clusters representing patterns of protein abundance. The gray box denotes the set of genes that show protein abundance patterns that fit well with mRNA abundances in D. D) mRNA abundances matched to the columns in C. For E), F) and G), pink bars denote the timing of production of long transcripts and blue bars denote the timing of production of short, canonical transcripts. mRNA, ribosome footprints, protein, and TE are plotted for E) canonical Ndt80 target CDC5, F) ORC1 and G) NDC80. See also Fig. S2.

Figure 3

Annotation of genes regulated by transcript toggling. A) Top, transcript models are shown, along with the position of the MSE. Middle, mRNA reads (top) are plotted for CDC5, NDC80, and ORC1. Bottom, TE is plotted for these same genes. The timing of Ndt80 action corresponds with an increase in translation and TE for NDC80 and a decrease for ORC1. B) Enrichment is shown for predicted alternate meiotic transcripts [as defined in (Brar et al., 2012)] for genes in the Ndt80 regulon with poor mRNA:protein correlation (<0.4, Pearson) in Fig. S2A. ** p<0.01 by Fisher’s exact test. C) A model for Ndt80 action on three different types of targets for which it induces an abundant transcript. Canonical targets like CDC5 promote meiotic progression past pachytene. At these loci, a translatable transcript is made, resulting in rapid protein accumulation. Ndt80 induction induces an abundant and longer, untranslatable transcript isoform (LUTI) of ORC1, which results in decreased in protein levels. NDC80 is a target that was previously kept silent by a LUTI transcript. Ndt80 induces a short transcript that overcomes the silent transcript in the population and is well translated and allows protein accumulation, albeit at a slower rate than most canonical Ndt80 targets. D) The description of our pipeline for LUTI-type identification is shown and expanded in Fig. S3A. E) Enrichment of the newly annotated LUTIs is seen in the ORC1, CDC5 and NDC80 clusters in Fig. 2C. Fisher’s exact test was used for statistical significance. See also Fig. S3, Table S2.

Figure 4

Validation of LUTI cases predicted by systematic annotation. A) mRNA-seq and ribosome profiling data is shown over all timepoints for the MED7 locus. Existence of a long transcript that has translated uORFs and is poorly translated for MED7 is clear at mid-meiotic timepoints. B) Comparison of levels and timing between Northern blots for the MED7 ORF and the mRNA-seq, translation, protein, and TE measured from matched samples shows evidence for a poorly translated long transcript isoform. C) Z-score clustering to compare mRNA (top) and protein (bottom) patterns for each of the 380 predicted LUTIs discovered by the approach outlined in Fig. 3D. Below, the positions of the genes that are investigated in more detail in Fig. 4 and S4 are shown. D–J) Comparison of levels and timing between Northern blots and the TE in matched samples shows evidence for a poorly translated long transcript for D) RRD2, E) POP7, F) POP4, H) RAD16, J) SHS1. Pink bars represent the presence of the long transcript and blue bars represent the presence of the short transcript. G) W. blotting of a GFP reporter driven by pPOP4, I) pRAD16, and K) pSHS1. In a paired reporter deleted for the canonical SHS1 promoter, long transcript remains high and is increased relative to wild-type but protein production is low. Blots were run, transferred, and blotted together. L) Quantification of the Western blots in K. See also Fig. S4.

Figure 5

A single transcription factor coordinately induces long, translationally silent transcript isoforms and canonical transcripts, with opposing effects on protein production. A) Northern blotting reveals rapid increases in mRNA following Ndt80 induction for canonical Ndt80 target, CDC5, the canonical transcript isoform of CYC8, and long isoforms of POP7, MED7, and ORC3. Pink and blue bars at the right of blots indicate canonical and LUTI isoforms. B) These increases in mRNA result in decreased TE for POP7, MED7, and ORC3, while CDC5 and the short CYC8 isoform are better translated after Ndt80 induction. C) Northern blotting reveals Ndt80-dependence to induction of CDC5, the canonical transcript isoform of CYC8, and long isoforms of POP7, MED7, and ORC3, *denotes a background band. D) mRNA abundance increases for traditional and long, poorly translated Ndt80 transcript isoform targets occur with similar timing and are dependent on Ndt80. Dotted lines show mRNA abundances without addition of β-estradiol. Solid lines show measurements with. E) Outline of experiments in A) and C) and expected effects on gene expression. F–I) mRNA (blue) and protein (black) are shown with (solid line) and without (dotted line) Ndt80 induction for F) canonical targets CDC5 and G) HRR25, H) the canonical transcript isoform of CYC8, and I) POP7^LUTI. Note that induction of canonical mRNAs results in an Ndt80-dependent increase in mRNA and protein, while induction of the POP7 LUTI results in an Ndt80-dependent increase in mRNA but decrease in protein, relative to no Ndt80 induction. See also Fig. S5, Table S4, Table S5.

Figure 6

Newly identified LUTIs result in strong apparent translational control and show unusual spatio-temporal transcriptional coordination. A) A histogram of the ratio between the highest and lowest TE measured is shown for the genes that we predicted to be LUTI-regulated (Fig. 3D), revealing a higher difference for these genes compared to others. B) The incidence of directly adjacent and divergently oriented genes with their 5′ ends close is shown for our newly defined prospective LUTIs and a control set, chosen to include the 380 genes with the lowest protein:translation Pearson correlation in our dataset. C) We examined the full set of 380 predicted LUTI cases for evidence of neighboring, correlated transcripts, with 94 of these cases oriented divergently and with apparently co-regulated long transcript isoform cases (Fig. 3D). D) mRNA seq data for the RAD53/RRD2 locus is shown, demonstrating their divergent, neighboring orientation. RAD53 shows a regulated longer transcript of the predicted size that is poorly translated. RAD53 regulation looks similar to the regulation for RRD2 (Fig. 4D, 6D) but we had not previously annotated translated AUG-initiated uORFs for RAD53. E) Translation levels, mRNA, protein, and TE for RAD53 are shown above Northern blotting of matched samples. See also Fig. S6, S7.

Figure 7

A model for the control of protein levels by transcript isoform toggling in meiosis. A) A subset of genes are encoded by two isoforms, differing in their 5′ end. These isoforms result from two transcription start sites (TSSes) and the choice between these two TSSes may be controlled by the relative levels of the two transcription factors (TF2 and TF1) that can bind the proximal and distal TSS, respectively. If TF2 binds the proximal TSS, a canonical transcript is produced with a short 5′ leader that is well translated and results in protein accumulation. If TF1 accumulates, it binds the distal TSS and produces a longer transcript at this locus. This transcript includes the sequence for the encoded gene, but ribosomes do not decode this region into protein due to uORFs in the extended leader region. Analysis of one case shows that transcription of the LUTI can silence the proximal TSS in cis (Chia et al., 2017). The difference in translatability of the two transcripts is more important than the abundance of transcript at these loci. Further, by this model, TF2 ultimately activates gene expression and TF1 ultimately represses gene expression. B) In contrast, canonical transcriptional repression involves either loss of binding of an activating transcription factor or the additional presence of a repressor molecule. C) In a developmental process, the LUTI mechanism can enable coordinated activation of genes required for the next cellular state and repression of genes involved in the past cellular state. D) This mode of regulation allows a relay of sequential activation and repression to time protein levels to a window of action. E) The modular nature of LUTI regulation allows genes to be turned on an off in a coordinated manner for windows of different lengths of time. For example, gene A protein production would be turned on by TF1 and turned off by production of a LUTI by TF2. Gene C, would also be turned on by TF1, but would stay on longer, until shut down by TF3.