In vivo-like nearest neighbor parameters improve prediction of fractional RNA base-pairing in cells

Abstract

We conducted a thermodynamic analysis of RNA stability in Eco80 artificial cytoplasm, which mimics in vivo conditions, and compared it to transcriptome-wide probing of mRNA. Eco80 contains 80% of Escherichia coli metabolites, with biological concentrations of metal ions, including 2 mM free Mg2+ and 29 mM metabolite-chelated Mg2+. Fluorescence-detected binding isotherms (FDBI) were used to conduct a thermodynamic analysis of 24 RNA helices and found that these helices, which have an average stability of -12.3 kcal/mol, are less stable by ΔΔGo37 ∼1 kcal/mol. The FDBI data was used to determine a set of Watson-Crick free energy nearest neighbor parameters (NNPs), which revealed that Eco80 reduces the stability of three NNPs. This information was used to adjust the NN model using the RNAstructure package. The in vivo-like adjustments have minimal effects on the prediction of RNA secondary structures determined in vitro and in silico, but markedly improve prediction of fractional RNA base pairing in E. coli, as benchmarked with our in vivo DMS and EDC RNA chemical probing data. In summary, our thermodynamic and chemical probing analyses of RNA helices indicate that RNA secondary structures are less stable in cells than in artificially stable in vitro buffer conditions.

Graphical Abstract

Figure 1.

Thermodynamic characterization of RNA helices using fluorescence-detected binding isotherms (FDBI) in 1 M NaCl and Eco80. (A) Representative Van’t Hoff plots showing the relationship between the K_D and temperature, used to extract thermodynamic parameters (MeltR method 1). Points and error bars represent K_D values and standard error in the K_D values, respectively. Data here were collected on the FAM-CAGCCG-3′/5′-CGGCUG-BHQ1 helix in 1 M NaCl and Eco80. (B) Representative FDBI collected on the FAM-CAGCCG-3′/5′-CGGCUG-BHQ1 helix. Points represent the emission of a constant concentration of the FAM-labeled RNA in the presence of increasing concentrations of BHQ1-labeled RNA. Lines represent a global fit to determine thermodynamic parameters (MeltR method 2). (C) ΔG°₃₇ in Eco80 versus in 1 M NaCl for 24 RNA helices. Points and error bars represent the ΔG°₃₇ and standard error, respectively, determined using the global fitting method in MeltR. The R² represents the coefficient of determination of the linear model presented by the figure and represented graphically with a purple line. The grey line represents y = x. (D) The ΔΔG°₃₇ between Eco80 and 1 M NaCl versus the AU content of the helix. Features match those described in (C). (E) The ΔΔG°₃₇ between Eco80 and 1 M NaCl versus the length of the helix. Features match those described in (C). (F) ΔG°₃₇ NNPs in Eco80 versus NNPs in 1 M NaCl determined with FDBI. Points and error bars represent the ΔG°₃₇ and standard error, respectively, propagated from the fit to the NN model. The grey line represents y = x. (G) ΔG°₃₇ NNPs in 1 M NaCl and Eco80 determined in this work using FDBI compared to UV 1 M NaCl NNPs.

Figure 2.

Eco80-informed thermodynamic adjustments to the nearest neighbor (NN) model support reduced base pairing in vivo. (A) Composition of the E. coli structure set that was hand compiled from RFAM. (B) Composition of the ML training structure set that was downloaded from https://doi.org/10.5281/zenodo.4430150. (C) Effect of thermodynamic adjustments on the accuracy of structure prediction for the E. coli structure set and (D) ML set. Each point represents the accuracy of the prediction for a single RNA secondary structure. The center line of box plots represents the median (the 50th percentile), the hinges represent the first and third quartiles (the 25th and 75th percentile), and lower and upper whiskers represent the smallest or largest value that is no further than 1.5 times the interquartile range (the distance between the first and third quartiles) from the first or third quartiles. A Friedman test was applied to determine if changes in accuracy were statistically significant followed by a post hoc pairwise comparison of each NN model using a paired Wilcox test. The P value for the Friedman test is labeled and tested the null hypothesis that the data can be explained by random-non-parametric error in repeated measures of the same individual (RNA sequence). The remaining p values represent the post-hoc, paired Wilcox tests and tested the null hypothesis that measurements from two groups can be explained by random-non-parametric error in repeated measures of the same individual (RNA sequence). P values from the Wilcox tests were adjusted for multiple tests using a Bonferroni correction. (E) Effect of thermodynamic adjustments on the number of total structures in the predicted ensemble for each RNA in the ML structure set. Box plot features and statistics are described in (C). (F) Effect of thermodynamic adjustments on the percent of predicted paired nucleotides for each RNA in the ML set. Box plot features and statistics are described in (C). Outliers are shown explicitly as points. (G–I) Effect of thermodynamic adjustments on the percent of each type of base pair predicted for each RNA in the ML set. Box plot features and statistics are described in (C). Outliers are shown explicitly as points.

Figure 3.

Eco80-informed thermodynamic adjustments to the nearest neighbor model are consistent with fractionally folded RNA secondary structure in vivo. (A–C) Structure-seq2 provides information on the base-pairing state of RNA residues in E. coli via in vivo treatment with a chemical probe. The WC faces of As and Cs or Gs and Us were probed with DMS or EDC, respectively. Modifications were read out by reverse transcription stops and analyzed by NGS. (D) Moderate to high reactivity at a nucleotide indicates that nucleotides are in single-stranded or fractionally folded regions of RNA secondary structure. (E) Interpretation of low to no reactivity at a nucleotide is difficult because the WC face of nucleotides can be sequestered by many interactions other than WC base pairing. (F) Comparison of in vivo reactivity data colored red and the probability of a nucleotide adopting a single stranded conformation (P_ss) colored blue. The intensities of both red and blue colors are directly comparable. Left column: In vivo reactivity data (red) mapped to select regions of the E. coli 23S (rows 1 and 2) and 16S rRNA (row 3). The secondary structure networks are based on structures (PDB: 7OE1 and 6PJ6). Regions were selected for analysis based on high reactivity in regions that are base paired in the cryo-EM structure. Middle column: Predicted probability that a nucleotide is single stranded (P_ss) (blue) mapped to select regions of rRNA secondary structure. P_ss is based on a partition function analysis provided by the RNAstructure software using the UV 1 M NaCl NN model (left side) or the FDBI Eco80 NN model (right side). Right column: Reactivity versus the P_ss (RvP) plots for regions of the E. coli 23S and 16S rRNA. Orange circles represent a nucleotide in a partition function analysis using the UV 1 M NaCl NN model. Purple triangles represent a nucleotide in a partition function analysis using the FDBI Eco80 NN model. Arrow length represents the magnitude of the change in P_ss going from the 1 M NaCl NN model to the FDBI Eco80 NN model.

Figure 4.

Eco80-informed thermodynamic adjustments to the NN model and in vivo chemical probing are consistent with partially unfolded RNA secondary structure transcriptome-wide. (A) Reactivity versus the P_ss (RvP) plots using different adjustments to the NN model. Each point represents a nucleotide in the 91 mRNA transcripts with the best coverage in our Structure-seq2 dataset. The P_ss was calculated by folding mRNA in 50 nucleotide windows. Red lines represent the threshold T(a) function with a = 0.35 (Equation (9)). (B) Number of incorrectly predicted base pairs (false positives) versus the reactivity threshold parametric in adjustments to the NN model. (C) FDR versus the reactivity threshold parametric in adjustments to the NN model.

Figure 5.

Thermodynamic analysis of less stable RNA helices in Eco80. (A) Reaction diagram depicting the energetic contribution of various buffer additives to helix formation. (B) Thermodynamic contributions of Eco80 buffer components to RNA helix formation in Eco80. (C) Model for partial RNA helix unfolding in cells.