Modeling alternative translation initiation sites in plants reveals evolutionarily conserved cis -regulatory codes in eukaryotes

Figure 1.. Identification of prediction models and the associated features that predict plant alternative translation initiation sites (TISs). (A) Illustration of the identified true-positive (TP; blue) and true-negative (TN; gray) TISs that were categorized based on the location of initiation codons (i.e., 5′ UTR, CDS, or 3′ UTR) and their sequences in a transcript. The “known” features include the Kozak motif and position weight matrices (PWMs) generated from short sequences centered on TISs (Joshi et al. 1997; Reuter et al. 2016). The “ORF” features consist of various nucleotide compositions/sizes of the alternative TIS-initiated open reading frames (ORFs) and the annotated ORFs and also the RNA structures/nucleotide compositions of their surrounding regions. The “contextual” features based on k-mer enrichment analyses (k = 1–3) consist of the nucleotide and amino acid sequence contexts in the 200-bp region centered on the TISs (see the Supplemental Methods). Gray line indicates mRNAs; black boxes, annotated AUG TIS-initiated ORFs; and blue boxes, alternative TIS-initiated ORFs. (B,C) Numbers of the identified TP and TN TISs, categorized as described in A, in the tomato (B) and…

Figure 2.. The features that are most informative for predicting plant 5′ UTR TISs. (A) Comparison of the importance scores derived from the model and the statistical significance of differences (−log10(FDR), determined by a Wilcoxon signed-rank test with Bonferroni correction) between tomato 5′ UTR–AUG TPs and TNs for the features used in the best model. Rho indicates Spearman's rank correlation coefficient. The black line indicates the fitted linear regression line, and the gray area indicates the 95% confidence level interval. (B) The means of the feature values in the tomato 5′ UTR–AUG TP and TN data sets (right) and the frequency of features identified in 10 randomly balanced data sets (left) for the feature elimination–determined top 10 features (ranked using their importance) (see Supplemental Fig. S3A). The rank and frequency indicate the importance of a given feature in the prediction model and their robustness using 10 randomly balanced data sets. The features with a frequency greater than seven within the top 10 are shown. Orange indicates the TIS group with the higher feature value. (C–H) As described in A,B, but for the tomato 5′ UTR–nonAUG TIS group (C,D), the Arabidopsis 5′…

Figure 3.. The C/U nucleotide compositions and the flanking sequences of 5′ UTR TISs. (A,B) Sequence logo plots showing the differential enrichment of A/U/C/G nucleotides between TPs and TNs in the regions 15-bp upstream of and 13-bp downstream from the TISs, represented as the log2 ratio of the site frequencies between TPs and TNs, for the 5′ UTR TISs and annotated AUG sites in tomato (A) and Arabidopsis (B). (C,D) Enrichment of sites with the indicated 3-mer sequences in the tomato (C) and Arabidopsis (D) TIS groups, represented as the log2 ratio of the site frequencies between TPs and TNs in the 180-bp region centered on tomato TISs with a 10-bp window. (E,F) As described in C,D, but for the A, U, C, and G mononucleotides. (G) As described in Figure 2A, but shown for the regression coefficients (x-axis) for the features used in the best linear regression model of predicting human TISs. (H) As described in Figure 2B, but for the top six features with highest regression coefficients in human TIS prediction model. (I) Summary of the ML-revealed features of the Kozak motif (Kozak), the mononucleotide C content (“C”), and CU-rich tracts for their importance across different TIS groups.

Figure 4.. Characterization of putative RNA cis-elements predictive of plant TISs. (A,B) Relationship between the number of C/Us (y-axis) and the enrichment of sequence occurrence (x-axis) for all 5-mers in the tomato (A) and Arabidopsis (B) 5′ UTR–AUG and 5′ UTR–nonAUG TIS groups. The enrichment is represented as the log2 ratio of median sequence occurrence between TPs and TNs in the 200-bp regions centered on TIS sites. Rho indicates Spearman's rank correlation coefficient. (C,D) As described in Figure 3C, but for the enrichment of putative RNA cis-elements in the tomato (C) and Arabidopsis (D) 5′ UTR TIS groups. The putative “UCUUC” and “UCUCU” elements with the highest importance score and with the highest enrichment, which are indicated by arrows and highlighted in blue and orange in C,D, are shown.

Figure 5.. Mutation of CU-rich sequences on plant cellular mRNAs attenuated the TIS activities. (A) Immunoblotting analyses of proteins with translation initiated from the TISs indicated in Supplemental Figure S8 (orange arrows) and with translation driven by the upstream 100-nt wild-type (WT) sequence or sequences with mutations of CU tracts (mCU) or UCUUC/UCUCU (mUCUUC/mUCUCU) sites. Proteins were expressed in Nicotiana benthamiana (tobacco) leaves using the Agrobacterium-mediated transient expression system. Vector indicates tobacco leaves infiltrated with agrobacteria containing the expression vector (i.e., the GFP or FLAG-containing plasmid without a target gene sequence). (Bottom) The protein abundance of the reporter genes relative to actin for three biological repeats (dots) and the corresponding means and standard errors are shown. (B) The known CU-rich sequences found on human transcripts and on plant and animal viral mRNAs promote translation efficiency as reported in the indicated literature. (C) Illustration of swapping the WT (orange box) and the CU-mutated (mCU; gray line) sequences for the TIS-upstream region of a 5′ UTR–AUG TIS of Solyc03g096920.3.1 examined in A. The…

Figure 6.. The features of misclassified TN TISs. (A,B) Prediction score distribution of the TN TISs for the 5′ UTR–AUG and 5′ UTR–nonAUG TIS groups in tomato (A) and Arabidopsis (B). The mean threshold (dashed lines) for the classification of TN → TN (the TN TISs predicted as TNs; gray) and TN → TP (the TN TISs predicted as TPs; light brown) and the mean percentage (%) of the TN TISs misclassified as TP (TN → TP group) derived from the best models in the 10 randomly balanced data sets (as indicated in Fig. 1) are shown. (C,D) Dot plots show the frequency (y-axis) of a given feature used for TIS prediction in 10 randomly balanced data sets and the feature enrichment (FDR; x-axis) between the TN → TN and TN → TP groups for the TIS groups indicated in A,B. The red line represents the threshold (frequency of seven or more) of important features as indicated in Figure 2. (E,F) Violin plots show the feature value distributions for the features of methionine counts or PWM–5′ UTR–TP and “Arg” or “Ser” that were most enriched in C,D for the TN → TN (gray) and TN → TP (light brown) groups, indicated in A and the TP → TP (the TP TISs predicted as TPs, dark brown) group. The red dot represents the…

Figure 7.. Prediction of monocot and dicot TISs and their characteristics. (A) Distribution of the TIS prediction scores generated by the tomato best models for the 5′ UTR–AUG and 5′ UTR–nonAUG TIS groups identified by the RiboTISH algorism and with RiboTISH-reported FDRs (FDR percentile in which the 0–25 category includes TISs with lowest FDR values) using ribosome profiling data sets generated from dicot plants including Arabidopsis (suspension cells), tomato (leaves) and Arabidopsis seedings. (B) As indicated in A, but for the data sets generated from monocot plants including maize and rice. (C) Pairwise Pearson correlation coefficient of the features among dicot and monocot plants for the 5′ UTR–nonAUG TISs identified by the RiboTISH algorism. Monocots: (At) Arabidopsis, (Sl) tomato, and (At-Se) Arabidopsis seedlings; monocots: (Zm) maize and (Os) rice. (D) GO biological process terms significantly enriched (FDR < 0.05) in genes that have RiboTISH-reported TISs with the top 10% highest prediction scores and with the top 25% highest CG contents in the upstream 100-bp regions.