Optimized photochemistry enables efficient analysis of dynamic RNA structuromes and interactomes in genetic and infectious diseases

Abstract

Direct determination of RNA structures and interactions in living cells is critical for understanding their functions in normal physiology and disease states. Here, we present PARIS2, a dramatically improved method for RNA duplex determination in vivo with >4000-fold higher efficiency than previous methods. PARIS2 captures ribosome binding sites on mRNAs, reporting translation status on a transcriptome scale. Applying PARIS2 to the U8 snoRNA mutated in the neurological disorder LCC, we discover a network of dynamic RNA structures and interactions which are destabilized by patient mutations. We report the first whole genome structure of enterovirus D68, an RNA virus that causes polio-like symptoms, revealing highly dynamic conformations altered by antiviral drugs and different pathogenic strains. We also discover a replication-associated asymmetry on the (+) and (-) strands of the viral genome. This study establishes a powerful technology for efficient interrogation of the RNA structurome and interactome in human diseases.

Fig. 1. PARIS2 outline and summary of improvements.

a Outline of the PARIS method and major improvements. Details of the improvements are presented in subsequent figures and tables. b Notes on the major advantages of the new method and fold improvement for each step.

Fig. 2. Systematic optimization of PARIS2.

a Structures of AMT and amotosalen. b Higher concentrations of crosslinkers increase crosslinking efficiency. Psoralen crosslinked HEK293 cells RNA was extracted using the standard AGPC (TRIzol) method. Small RNA: RNA in the range of 50–300 nt, including tRNA, snRNA, 5S and 5.8S rRNAs etc. c Total RNA yield (upper panel) and 18S + 28S percentage (lower panel) from panel (b). d Fraction of psoralen crosslinked RNA from DD2D gels. Data are mean ± s.d.; n = 2 biological replicates. e Charge and hydrogen bonding of RNA and DNA molecules in standard AGPC (TRIzol) purification. The ‘-’ indicates negative charges. The ‘..’ indicates exposed bases involved in hydrogen bonding. f Phase partition of DNA, RNA and protein in AGPC and PCI methods, and comparison to TNA. g Repartition of crosslinked RNA to the interphase in Trizol extraction. Purified total RNA was treated with or without 365 nm UV plus AMT, and then either directly precipitated using ethanol, or extracted from the 2 phases of TRIzol-chloroform. h Recovery of crosslinked RNA from the aqueous and inter+organic phases using the TNA method. Data are mean ± s.d.; n = 2 biological replicates. i TNA method outline. j Recovery of inact RNA from psoralen crosslinked HEK293 cells using TNA method, after DNase treatment. RNA integrity numbers (RIN) and two indicators of RNA quality, A260/A230 and A260/A280, were listed. k Quantity of RNA isolated from control and AMT/amotosalen crosslinked cells. Data are mean ± s.d.; n = 3 biological replicates. two-tailed, unpaired t-test. l Effects of UV damages and failed proximity ligations (non prox ligated) on the yield of gapped reads. m UVC excited RNA form various products, such as pyrimidine dimers, hydrates and strand breaks. Alternatively, the energy can be transferred to singlet quenchers like acridine orange, leaving RNA intact. S0 and S1: ground state and excited singlet state. n cDNA yield from RNA irradiated with 254 nm UV, with or without AO protection, normalized to non-photo-reversal sample. Top: Non-crosslinked. Bottom: AMT crosslinked. Box plots show center line as median, box limits as upper and lower quartiles, whiskers as minimum to maximum values. 10 min + AO: n = 4; others: n = 7, biological replicates. two-tailed, unpaired t-test. o cDNA yield obtained in RT-qPCR experiments for PUVA damaged 18S-rRNA, normalized to SSIII. MaRT: Marathon RT. Data are mean ± s.d.; SSIII/SSIV/TGIRT: n = 4; SSII/MaRT: n = 2, biological replicates. two-tailed, unpaired t-test. p cDNA yield for β-Actin using SSIV in different reaction buffers and different incubation time, normalized to a standard Mg2+ condition. Data are mean ± s.d.; n = 4 biological replicates. two-tailed, unpaired t-test.

Fig. 3. PARIS2 enables global profiling of SSU ribosome binding sites.

a The 40S ribosome small subunit (SSU), showing the beak side view, highlighting the 18S h18 helix exposed in the mRNA channel. rRNA in light gray, RPS in dark gray, except h18 left arm: 595–620, right arm: 621–641. b Rotated 40S, showing h26 and h44 in yellow. c The binding sites of mRNAs, snRNAs, and mitochondrial rRNAs (12S and 16S) on the 45S unit based on HEK293 mRNA PARIS2 data. d The binding sites of h18 and h26 on the meta mRNA. The tapering off signal on the 5′UTR is likely due to its limited length. The sharp drop off after the stop codon (20–30 nt short tail) is due to random RNA fragment length.

Fig. 4. Discovery of a snoRNA–rRNA interaction network in rRNA processing and LCC.

a Intermolecular interactions among U8, U13, 18S, and 28S based on PARIS2 data from human HEK and mouse brain. Two intramolecular U8 alternative conformations are labeled in red and blue colors. Each row illustrates interactions between one pair of RNAs. Each column illustrates regions in one RNA that participate in interactions. b Mapping LCC mutations onto the U8 interaction network. Stemloops (SL) are defined for all U8 conformations and their interactions. U8d U8 dimer. All mutations from LCC patients were summarized in the track, where the coverage [0–7] indicates counts of mutations. c MFE changes for all U8 related duplexes due to LCC-causing mutations, calculated using PARIS2-derived models and RNAcofold. Red: that mutations increased the MFE and lowered stability of the duplex.

Fig. 5. PARIS2 reveals dynamic structuromes of EV-D68 viral RNAs in cells.

a Schematic diagram of the experimental strategy. Virus-infected HeLa cells were crosslinked. RNA was extracted using the TNA method and the viral genome RNA was enriched using biotinylated oligos. VPg viral genome-linked protein. b Two strains of EV-D68 were chosen: VR1197 (isolated in 1963) and US47 (US/MO/14-18947, isolated in 2014). c Experimental conditions. HeLa and SH-SY5Y cells were infected with the two strains and treated with two inhibitors, AG and GA. d The enterovirus life cycle and inhibitors mechanisms of action. e Analysis strategies. Gapped alignments were first assembled into duplex groups using CRSSANT. DGs in two samples are compared to identify common and different ones. DGs within the same sample are compared to each other to identify alternative/dynamic conformations. f US47 genome structure is highly reproducible in HeLa and SH cell lines based on shuffling test (1000 times). cov relative coverage, f fraction overlap, mean overlap average number of overlaps from the 1000 shuffles. g Local structures are conserved between US47 and VR1197 strains in HeLa cells. Data in the US47 strain were lifted to the VR1197 coordinates. h Fractions of overlaps calculated at different relative coverage (cov) and fraction overlaps of the two arms (f). i, j PARIS2 detects extensive alternative conformations along the entire length of US47 genome. Relative coverage at least 0.01 (cov ≥ 0.01) was set as cutoff for duplex groups. Arcs on the top and bottom represent DGs that are involved in alternative conformations with each other. The alternative DGs track indicates number of alternative conformations each region is involved in, which is then summarized in a histogram (j). k Diagram showing the recovery of crosslinked dsRNA intermediates during replication, using antisense probes targeting the plus (+) strand. l Diagram showing the theoretical structures on the (+) and (−) strands, which are mirror images of each other (with the exception of G–U pairs, where the counterpart A–C pairing is less stable). m Comparison of duplex span on the two strands in all experimental conditions. Duplex span for RNAfold-predicted structures (default parameters) is calculated as the linear distance between base pairs. Duplex span for PARIS2-derived structures is the linear distance between the middle points of the two arms. All primary gap1 alignments were used for calculation. Distances were log-transformed before plotting the violin and box plots. In the box plots, wiskers represent the max and min. The top and bottom of the box represent the first and third quartiles. The red bar is the median. Here are the numbers of samples for each of the 22 violin + box plots. For RNAfold: n = 2223, 2248, 2287, 2290. For US in HeLa: 235,976, 3958, 150,676, 5538, 238,185, 2925. For VR in HeLa, 159,270, 2874, 89,940, 1687, 124,774, 1935. For US in SH: n = 127,123, 1744, 6167, 46, 74,388, 352. P values were calculated using the Mann–Whitney U test. n, o Comparison of DGs on the (+) and (−) strands in US47 (n) and VR1197 (o) strains. p Comparison of structure density on the (+) and (−) strands in the two strains. The three samples for each strain were colored red or blue, for the (+) and (−) strains, respectively. The gray shadowed area highlights the biggest difference in the 5′CL and IRES structures between the two strands.

Fig. 6. PARIS2 reveals dynamic and long-range structures in the 5′UTR of two EV-D68 strains.

a PARIS2 validates predicted structures (black arcs) and reveals new alternative conformations (colored arcs) in the 699 nt 5′UTR of the US47 strain of EV-D68. 5′CL cloverleaf structure, IRES internal ribosome entry site, including domains II–VI, a1–a5 five alternative conformations. Numbers in parentheses are the counts of gapped alignments in each DG. b Long-range structures connecting the 5′UTR to the rest of the genome in two EV-D68 strains. All DGs have cov ≥ 0.01 for these two samples. The two stars indicate the right-side anchors for the strongest long-range strucgtures in the two strains, near the CRE and the region near the stop codon at the end of peptide 3D (3Dstop). c Long-range structures in the 5′UTR are primarily anchored in three linker regions L1–L3 (e.g., L1 is after domain I). d Diagram showing the three strongest long-range structures L1-CRE (connecting L1 and near CRE, in US47), L2L3-start (start codon, in US47), and L1L2-3Dstop (in VR1197). e–g DGs supporting the three long-range structures. In L1-CRE (e), the first ~5000 nt of the genome is shown. In the L2L3-start (f), the first ~800 nt of the genome is shown. In L1L2-3Dstop (g), the entire genome is shown. h Quantifying the strength of DGs for the three structures in all conditions. H HeLa cells, S SH-SY5Y cells, U US47, V VR1197, A AG, G GA. i Stability of the three structures, measured in minimal free energy (MFE), in the two strains.