Limits...
Rich RNA Structure Landscapes Revealed by Mutate-and-Map Analysis.

Cordero P, Das R - PLoS Comput. Biol. (2015)

Bottom Line: In experimental benchmarks, M2-REEFFIT recovers the structure landscapes of a 35-nt MedLoop hairpin, a 110-nt 16S rRNA four-way junction with an excited state, a 25-nt bistable hairpin, and a 112-nt three-state adenine riboswitch with its expression platform, molecules whose characterization previously required expert mutational analysis and specialized NMR or chemical mapping experiments.An artificial flavin mononucleotide riboswitch and a randomly generated RNA sequence are found to interconvert between three or more states, including structures for which there was no design, but that could be stabilized through mutations.These results highlight the likely pervasiveness of rich landscapes with multiple secondary structures in both natural and artificial RNAs and demonstrate an automated chemical/computational route for their empirical characterization.

View Article: PubMed Central - PubMed

Affiliation: Biomedical Informatics Program, Stanford University, Stanford, California, United States of America.

ABSTRACT
Landscapes exhibiting multiple secondary structures arise in natural RNA molecules that modulate gene expression, protein synthesis, and viral infection [corrected]. We report herein that high-throughput chemical experiments can isolate an RNA's multiple alternative secondary structures as they are stabilized by systematic mutagenesis (mutate-and-map, M2) and that a computational algorithm, REEFFIT, enables unbiased reconstruction of these states' structures and populations. In an in silico benchmark on non-coding RNAs with complex landscapes, M2-REEFFIT recovers 95% of RNA helices present with at least 25% population while maintaining a low false discovery rate (10%) and conservative error estimates. In experimental benchmarks, M2-REEFFIT recovers the structure landscapes of a 35-nt MedLoop hairpin, a 110-nt 16S rRNA four-way junction with an excited state, a 25-nt bistable hairpin, and a 112-nt three-state adenine riboswitch with its expression platform, molecules whose characterization previously required expert mutational analysis and specialized NMR or chemical mapping experiments. With this validation, M2-REEFFIT enabled tests of whether artificial RNA sequences might exhibit complex landscapes in the absence of explicit design. An artificial flavin mononucleotide riboswitch and a randomly generated RNA sequence are found to interconvert between three or more states, including structures for which there was no design, but that could be stabilized through mutations. These results highlight the likely pervasiveness of rich landscapes with multiple secondary structures in both natural and artificial RNAs and demonstrate an automated chemical/computational route for their empirical characterization.

Show MeSH

Related in: MedlinePlus

Comparison between reactivities predicted by REEFFIT and mutants designed to stabilize the (a) TBWN-A, (b) TBWN-B, (c) TBWN-C structures are given below each state (arrows in the structures mark mutations); each plot shows the measured reactivates for each pair of mutants (orange and green) and predicted REEFFIT reactivates for the corresponding state (blue with error bars). (d) Determining FMN dissociation constants for stabilizing mutants of the TBWN-A, TBWN-B, and TBWN-C structures of the Tebowned FMN switch using the LIFFT HiTRACE toolkit (showing data for residue 15, see Supporting Fig I in S1 Text) are shown below the mutants.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4643908&req=5

pcbi.1004473.g005: Comparison between reactivities predicted by REEFFIT and mutants designed to stabilize the (a) TBWN-A, (b) TBWN-B, (c) TBWN-C structures are given below each state (arrows in the structures mark mutations); each plot shows the measured reactivates for each pair of mutants (orange and green) and predicted REEFFIT reactivates for the corresponding state (blue with error bars). (d) Determining FMN dissociation constants for stabilizing mutants of the TBWN-A, TBWN-B, and TBWN-C structures of the Tebowned FMN switch using the LIFFT HiTRACE toolkit (showing data for residue 15, see Supporting Fig I in S1 Text) are shown below the mutants.

Mentions: After validation of M2-REEFFIT on diverse computational and experimental test cases, we used the method to estimate whether complex landscapes might arise in artificial RNA sequences without explicit design or selection. First, we analyzed the folding landscape of an imperfect RNA switch, ‘Tebowned’, that was designed to convert between two states upon flavin mononucleotide (FMN) binding (Figs 4 and 5) in early rounds of a riboswitch design puzzle in the Eterna massive open laboratory [33]. While the chemical mapping pattern of this RNA changed upon binding FMN, the measurements for the unbound state did not match the desired unbound structure, particularly near nucleotide A30 (red arrows in Fig 4a); this region should have been paired but instead was measured to be reactive. A priori, we could not distinguish whether this discrepancy was due to an incorrect balance of the two target bound and unbound structures or if there were other unexpected structures involved. To elucidate the discrepancy, we acquired M2 data for the Tebowned RNA. We first used REEFFIT to fit the M2 data using only the two desired structures, but this fit did not capture several features, such as the exposure at A30 (Fig G in S1 Text). In contrast, a global ensemble fit adequately captured these features (Fig 4a–4b, and Fig Gb in S1 Text). REEFFIT automatically clustered the modeled ensemble into three states: TBWN-A (56 ±16%) and TBWN-B (27 ± 12%), matching the desired switch structures, and a third structure TBWN-C present at 17 ± 11% fraction (red arrows in Fig 4a and Fig Ha-Hc in S1 Text). The unexpected state TBWN-C exhibits an apical loop around nucleotide A30, explaining the observed discrepancies in this region for the wild type RNA, and harbors a purine-rich symmetric loop that may be significantly stabilized compared to the energy assumed in current nearest-neighbor models [34]. Other helices were discovered to be populated at non-negligible fractions in the analysis, but were deemed uncertain (signal-to-noise ratios less than 1) from bootstrapping analysis.


Rich RNA Structure Landscapes Revealed by Mutate-and-Map Analysis.

Cordero P, Das R - PLoS Comput. Biol. (2015)

Comparison between reactivities predicted by REEFFIT and mutants designed to stabilize the (a) TBWN-A, (b) TBWN-B, (c) TBWN-C structures are given below each state (arrows in the structures mark mutations); each plot shows the measured reactivates for each pair of mutants (orange and green) and predicted REEFFIT reactivates for the corresponding state (blue with error bars). (d) Determining FMN dissociation constants for stabilizing mutants of the TBWN-A, TBWN-B, and TBWN-C structures of the Tebowned FMN switch using the LIFFT HiTRACE toolkit (showing data for residue 15, see Supporting Fig I in S1 Text) are shown below the mutants.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4643908&req=5

pcbi.1004473.g005: Comparison between reactivities predicted by REEFFIT and mutants designed to stabilize the (a) TBWN-A, (b) TBWN-B, (c) TBWN-C structures are given below each state (arrows in the structures mark mutations); each plot shows the measured reactivates for each pair of mutants (orange and green) and predicted REEFFIT reactivates for the corresponding state (blue with error bars). (d) Determining FMN dissociation constants for stabilizing mutants of the TBWN-A, TBWN-B, and TBWN-C structures of the Tebowned FMN switch using the LIFFT HiTRACE toolkit (showing data for residue 15, see Supporting Fig I in S1 Text) are shown below the mutants.
Mentions: After validation of M2-REEFFIT on diverse computational and experimental test cases, we used the method to estimate whether complex landscapes might arise in artificial RNA sequences without explicit design or selection. First, we analyzed the folding landscape of an imperfect RNA switch, ‘Tebowned’, that was designed to convert between two states upon flavin mononucleotide (FMN) binding (Figs 4 and 5) in early rounds of a riboswitch design puzzle in the Eterna massive open laboratory [33]. While the chemical mapping pattern of this RNA changed upon binding FMN, the measurements for the unbound state did not match the desired unbound structure, particularly near nucleotide A30 (red arrows in Fig 4a); this region should have been paired but instead was measured to be reactive. A priori, we could not distinguish whether this discrepancy was due to an incorrect balance of the two target bound and unbound structures or if there were other unexpected structures involved. To elucidate the discrepancy, we acquired M2 data for the Tebowned RNA. We first used REEFFIT to fit the M2 data using only the two desired structures, but this fit did not capture several features, such as the exposure at A30 (Fig G in S1 Text). In contrast, a global ensemble fit adequately captured these features (Fig 4a–4b, and Fig Gb in S1 Text). REEFFIT automatically clustered the modeled ensemble into three states: TBWN-A (56 ±16%) and TBWN-B (27 ± 12%), matching the desired switch structures, and a third structure TBWN-C present at 17 ± 11% fraction (red arrows in Fig 4a and Fig Ha-Hc in S1 Text). The unexpected state TBWN-C exhibits an apical loop around nucleotide A30, explaining the observed discrepancies in this region for the wild type RNA, and harbors a purine-rich symmetric loop that may be significantly stabilized compared to the energy assumed in current nearest-neighbor models [34]. Other helices were discovered to be populated at non-negligible fractions in the analysis, but were deemed uncertain (signal-to-noise ratios less than 1) from bootstrapping analysis.

Bottom Line: In experimental benchmarks, M2-REEFFIT recovers the structure landscapes of a 35-nt MedLoop hairpin, a 110-nt 16S rRNA four-way junction with an excited state, a 25-nt bistable hairpin, and a 112-nt three-state adenine riboswitch with its expression platform, molecules whose characterization previously required expert mutational analysis and specialized NMR or chemical mapping experiments.An artificial flavin mononucleotide riboswitch and a randomly generated RNA sequence are found to interconvert between three or more states, including structures for which there was no design, but that could be stabilized through mutations.These results highlight the likely pervasiveness of rich landscapes with multiple secondary structures in both natural and artificial RNAs and demonstrate an automated chemical/computational route for their empirical characterization.

View Article: PubMed Central - PubMed

Affiliation: Biomedical Informatics Program, Stanford University, Stanford, California, United States of America.

ABSTRACT
Landscapes exhibiting multiple secondary structures arise in natural RNA molecules that modulate gene expression, protein synthesis, and viral infection [corrected]. We report herein that high-throughput chemical experiments can isolate an RNA's multiple alternative secondary structures as they are stabilized by systematic mutagenesis (mutate-and-map, M2) and that a computational algorithm, REEFFIT, enables unbiased reconstruction of these states' structures and populations. In an in silico benchmark on non-coding RNAs with complex landscapes, M2-REEFFIT recovers 95% of RNA helices present with at least 25% population while maintaining a low false discovery rate (10%) and conservative error estimates. In experimental benchmarks, M2-REEFFIT recovers the structure landscapes of a 35-nt MedLoop hairpin, a 110-nt 16S rRNA four-way junction with an excited state, a 25-nt bistable hairpin, and a 112-nt three-state adenine riboswitch with its expression platform, molecules whose characterization previously required expert mutational analysis and specialized NMR or chemical mapping experiments. With this validation, M2-REEFFIT enabled tests of whether artificial RNA sequences might exhibit complex landscapes in the absence of explicit design. An artificial flavin mononucleotide riboswitch and a randomly generated RNA sequence are found to interconvert between three or more states, including structures for which there was no design, but that could be stabilized through mutations. These results highlight the likely pervasiveness of rich landscapes with multiple secondary structures in both natural and artificial RNAs and demonstrate an automated chemical/computational route for their empirical characterization.

Show MeSH
Related in: MedlinePlus