DMS-MaPseq for genome-wide or targeted RNA structure probing in vivo

Abstract

Coupling of structure-specific in vivo chemical modification to next-generation sequencing is transforming RNA secondary structure studies in living cells. The dominant strategy for detecting in vivo chemical modifications uses reverse transcriptase truncation products, which introduce biases and necessitate population-average assessments of RNA structure. Here we present dimethyl sulfate (DMS) mutational profiling with sequencing (DMS-MaPseq), which encodes DMS modifications as mismatches using a thermostable group II intron reverse transcriptase. DMS-MaPseq yields a high signal-to-noise ratio, can report multiple structural features per molecule, and allows both genome-wide studies and focused in vivo investigations of even low-abundance RNAs. We apply DMS-MaPseq for the first analysis of RNA structure within an animal tissue and to identify a functional structure involved in noncanonical translation initiation. Additionally, we use DMS-MaPseq to compare the in vivo structure of pre-mRNAs with their mature isoforms. These applications illustrate DMS-MaPseq's capacity to dramatically expand in vivo analysis of RNA structure.

Figure 1. Sequencing library generation for RNA structure probing techniques

Schematic of library preparation strategies for cDNA truncation approaches (a) and for DMS-MaPseq (b).

Figure 2. TGIRT enzyme delivers higher signal and lower background for DMS-MaPseq

a, Distribution of mutation type generated by SSII/Mn²⁺ or TGIRT reverse transcription from in vivo DMS-treated yeast mRNA. b, Endogenous m¹A modifications in yeast 25S rRNA transcript reveal superior modification detection with TGIRT. Average percent modification (bar) detected at the position across two biological DMS-treated replicates (circles) with error bars representing standard deviation from the average. c, Nucleotide composition of mismatches from TGIRT or SSII/Mn²⁺ approaches. d, Yeast RPS28B mRNA positive control structure with nucleotides colored by DMS reactivity in vivo. Black boxes outline G/U bases with high background signal. DMS reactivity was calculated as the average ratiometric DMS signal per position across two biological replicates normalized to the highest number of reads in displayed region, which is set to 1.0. e, Genome-wide DMS-MaPseq replicates compared by Pearson’s r value and Gini index for yeast mRNA regions (requiring 15x coverage, resulting in 733 and 272 regions displayed for TGIRT and SSII/Mn²⁺, respectively).

Figure 3. Global analysis of in vivo DMS-MaPseq data

a, Signal decay observed after endogenous m¹A modification at position 642 in the yeast 25S rRNA in DMS-seq, but not in DMS-MaPseq. b, Histogram of ratiometric reactivity for negative control bases in the yeast 18S rRNA. The total number of negative control bases is 338, characterized as bases known to be base-paired. c, Scatterplot of GC content versus Gini Index in 50nt windows of deeply sequenced genes. Non-coding RNA regions include UTRs and all classes of mammalian non-coding RNAs. The total number of evaluated windows is 182. Pearson’s correlation = 0.32, p-value = 7.3e-6.

Figure 4. DMS-MaPseq enables in vivo RNA structure probing for specific RNA targets

a, Cumulative histogram of Pearson’s r values between yeast mRNA regions in DMS-MaPseq replicates at varied depths of average mismatch coverage. b, Fraction of genes exceeding the minimum average mismatch coverage of 20x in genome-wide human HEK 293T DMS-MaPseq data with varied sequencing depths. 0.006, 0.009, and 0.03 are the fraction of genes passing this threshold at 50, 100, and 200 million uniquely mapped reads, respectively. c, Schematic for targeted RNA structure probing via target-specific RT-PCR and NexteraXT tagmentation. d, ROC curve for DMS signal on yeast 18S rRNA using ratiometric data from target-specific tagmentation approach and from genome-wide DMS-MaPseq. e, f, Yeast HAC1 (e) and RPS28B (f) 3′ UTR mRNA positive control structures from target-specific priming with nucleotides colored by DMS reactivity in vivo. DMS reactivity calculated as the ratiometric DMS signal per position normalized to the highest number of reads in displayed region, which is set to 1.0.

Figure 5. Novel experimental applications for in vivo RNA structure probing

a, oskar 3′ UTR mRNA positive control structure from target-specific priming with nucleotides colored by in vivo DMS reactivity in D. melanogaster ovaries. DMS reactivity calculated as the ratiometric DMS signal per position normalized to the highest number of reads in displayed region, which is set to 1.0. b, oskar positive control region from (a) shown with average normalized DMS-MaPseq values from two biological replicates, one at 5 min DMS treatment and one at 10 min. Error bars represent one standard deviation. c, Ratiometric DMS-MaPseq from targeted amplification of the human FXR2 5′ UTR and exon1 sequence. Nucleotides accessible to DMS are noted with a value >0.03, which is the threshold representing the best agreement with our model. Position 1 corresponds to chromosome XVII:7614897.

Figure 6. Investigating RNA structure heterogeneity with DMS-MaPseq

a, Regions of heterogeneous structure exhibit indistinguishable structure signals when combined but can be distinguished by DMS-MaPseq, illustrated by normalized DMS-MaPseq data derived from the human MRPS21 ribosnitch A/C alleles. Allele-specific data represented as the mean of three technical replicates. Error bars represent one standard deviation. b, Targeted DMS-MaPseq data specific for the yeast RPL14A pre-mRNA and spliced mRNA isoforms reveal minimal structure difference in the common exon1 sequence (r = 0.88). Ratiometric in vivo DMS-MaPseq data is plotted with isoform-specific RT primer locations noted with arrows.