High-throughput identification of synthetic riboswitches by barcode-free amplicon-sequencing in human cells

Abstract

Synthetic riboswitches mediating ligand-dependent RNA cleavage or splicing-modulation represent elegant tools to control gene expression in various applications, including next-generation gene therapy. However, due to the limited understanding of context-dependent structure-function relationships, the identification of functional riboswitches requires large-scale-screening of aptamer-effector-domain designs, which is hampered by the lack of suitable cellular high-throughput methods. Here we describe a fast and broadly applicable method to functionally screen complex riboswitch libraries (~1.8 × 10⁴ constructs) by cDNA-amplicon-sequencing in transiently transfected and stimulated human cells. The self-barcoding nature of each construct enables quantification of differential mRNA levels without additional pre-selection or cDNA-manipulation steps. We apply this method to engineer tetracycline- and guanine-responsive ON- and OFF-switches based on hammerhead, hepatitis-delta-virus and Twister ribozymes as well as U1-snRNP polyadenylation-dependent RNA devices. In summary, our method enables fast and efficient high-throughput riboswitch identification, thereby overcoming a major hurdle in the development cascade for therapeutically applicable gene switches.

Fig. 1. Principle and schematic workflow for the sequencing-based identification of functional riboswitches in human cells.

a HEK-293 cells are transfected with a plasmid library harboring PCR-binding site (red arrowheads)-flanked aptazyme variants with a randomized motif (here: tetracycline (Tet)-hammerhead design). b Functional ON-switches are characterized by aptazyme auto-cleavage and mRNA-degradation in the absence of Tet, whereas upon Tet addition, cleavage is inhibited, resulting in stable mRNA. c Following cell lysis, RNA purification and reverse transcription into cDNA, riboswitch sequences are PCR-amplified and applied to amplicon-seq analysis. Functional sequences are identified from differential expression analyses (stimulated vs. unstimulated). ITR, inverted terminal repeat; CMV, cytomegalovirus; GFP, green fluorescent protein; pA, polyadenylation signal.

Fig. 2. Identification of Tet-responsive hammerhead ribozymes by amplicon-seq.

a Secondary structure of the Tet-hammerhead ribozyme library design. Blue nucleotides indicate the randomized motif. The arrowhead indicates the cleavage site. A/G mutation at the indicated position renders the ribozyme inactive. b Tet riboswitch construct frequency analysis of the original library and c conservation of complexity following transfection in HEK-293 cells. CV, coefficient of variation. d Volcano plots displaying the significance (FDR) of changes in riboswitch abundance as a function of the change in expression (log₂ fold change). e Correlation of expression induction upon Tet for constructs and associated fold changes reported by Beilstein et al. (X-axis) with fold changes measured in our screen (Y-axis). f Selected hits from the Tet riboswitch library screen, randomized sequence, mean counts per million (CPM), fold change (FC) and false-discovery rate (FDR) at 25 and 50 µM Tet, respectively. The bar plots show the raw counts under untreated, 25 µM and 50 µM Tet-stimulated conditions (three groups from left to right, n = 8 replicates each). g Functional hit validation in HEK-293 cells (n = 3 experiments, mean ± s.d.); Hh-act/inact: constitutively active/inactive ribozyme control. h Maximal fold changes and p-values measured upon stimulation vs. 0 µM Tet for all tested constructs. *p < 0.05, **p < 0.01, ***p < 0.001 (unpaired T-Test, two-tailed). Source data are provided as a Source Data file.

Fig. 3. Identification of Gua-responsive HDV ribozymes by amplicon-seq.

a Secondary structure of the Gua-HDV ribozyme library design. Blue nucleotides indicate the randomized motif. The arrowhead indicates the cleavage site. C/U mutation at the indicated position renders the ribozyme inactive. b Guanine riboswitch construct frequency analysis of the original library. c Volcano plots displaying the significance (FDR) of changes in riboswitch abundance as a function of the change in expression (log₂ fold change). d Selected hits from the Gua-HDV riboswitch library screen, randomized sequence, mean counts per million (CPM), fold change (FC) and false-discovery rate (FDR) at 30, 100, and 300 µM guanine, respectively. The bar plots show the raw counts under untreated, 30, 100, and 300 µM µM Gua-stimulated conditions (four groups from left to right, n = 8 replicates each). e Functional hit validation in HEK-293 cells at increasing guanine doses [µM] (n = 3 replicates, mean ± s.d.); HDV-act/inact: constitutively active/inactive ribozyme control. f Maximal fold changes and p-values measured upon stimulation vs. 0 µM guanine for all tested constructs. *p < 0.05, **p < 0.01, ***p < 0.001 (unpaired T-Test, two-tailed). Source data are provided as a Source Data file.

Fig. 4. Identification of Gua-responsive hammerhead ribozymes by amplicon-seq.

a Secondary structure of the guanine hammerhead ribozyme library design. Blue nucleotides indicate the randomized motif. The arrowhead indicates the cleavage site. A/G mutation at the indicated position renders the ribozyme inactive. b Guanine riboswitch construct frequency analysis of the original library. c Volcano plots displaying the significance (FDR) of changes in riboswitch abundance as a function of the change in expression (log₂ fold change). d Selected hits from the Gua-HHR riboswitch library screen, randomized sequence, mean counts per million (CPM), fold change (FC) and false-discovery rate (FDR) at 30, 100, and 300 µM guanine, respectively. The bar plots show the raw counts under untreated, 30, 100, and 300 µM µM Gua-stimulated conditions (four groups from left to right, n = 8 replicates each). e Functional hit validation in HEK-293 cells at increasing guanine doses [µM] (n = 3 replicates, mean ± s.d.); Hh-act/inact: constitutively active/inactive ribozyme control. f Maximal fold changes and p-values measured upon stimulation vs. 0 µM guanine for all tested constructs. *p < 0.05, **p < 0.01, ***p < 0.001 (unpaired T-Test, two-tailed). Source data are provided as a Source Data file.

Fig. 5. Identification of Tet-responsive Twister ribozymes by amplicon-seq.

a Secondary structure of the Tet-Twister library design. Blue boxes indicate the randomized motifs of the nine sub-libraries, connecting the Twister ribozyme and the Tet aptamer. The arrowhead indicates the cleavage site. GU/UG mutation at the indicated position renders the ribozyme inactive. Inset: Library construct frequency analysis by DNA sequencing. b Volcano plots displaying the significance (FDR) of changes in riboswitch abundance as a function of the change in expression (log₂ fold change) for the nine Tet-Twister sub-libraries. Selected hits are labeled with their sequence motif. c Selected hits from the Tet-Twister library screen, randomized sequence, mean counts per million (CPM), fold change (FC) and false-discovery rate (FDR) at 12.5, 25, and 50 µM Tet, respectively. The bar plots show the raw counts under untreated, 12.5, 25, and 50 µM Tet-stimulated conditions (four groups from left to right, n = 8 replicates each). d Functional hit validation in HeLa cells at increasing tetracycline doses [µM] (n = 3 replicates, mean ± s.d.); psi-Check: ribozyme-free control construct; Tw-act/inact: constitutively active/inactive ribozyme control. e Maximal fold changes and p-values measured upon stimulation vs. 0 µM Tet for all tested constructs. f Network analysis of motif similarity for the CG_3N3N library at 25 µM Tet vs. untreated and an FDR < 1%. Each line segment represents a single nucleotide change. The “T9” construct is circled in red. *p < 0.05, **p < 0.01, ***p < 0.001 (unpaired T-Test, two-tailed). Source data are provided as a Source Data file.

Fig. 6. Identification of Gua-responsive riboswitches based on a U1-snRNP-dependent mode of action.

a Secondary structure of the Gua-U1-snRNP riboswitch library design. The U1-snRNP binding site (CAGGUAAGU) is included into a prolonged stem P1 of the guanine aptamer. Blue nucleotides indicate the randomized motif. b Mode of action of U1-snRNP-dependent riboswitches. In the ligand-unbound conformation, the U1-snRNP binding site is available, allowing for U1-snRNP binding and interference with the polyadenylation process. Upon ligand binding, the aptamer stem is stabilized and the U1-snRNP binding site is masked, enabling polyadenylation and subsequent gene expression. c Guanine riboswitch construct frequency analysis of the original library. d Volcano plots showing the significance (FDR) of changes in riboswitch abundance as a function of the change in expression (log₂ fold change). e Selected hits from the Gua-U1-snRNP riboswitch library screen, randomized sequence, mean counts per million (CPM), fold change (FC) and false-discovery rate (FDR) at 30, 100, and 300 µM guanine, respectively. The bar plots show the raw counts under untreated, 30, 100, and 300 µM µM Gua-stimulated conditions (four groups from left to right, n = 8 replicates each). f Functional hit validation in HEK-293 cells at increasing guanine doses [µM] (n = 3 replicates, mean ± s.d.); U1-bs: construct harboring the freely accessible U1-snRNP binding site. g Maximal fold changes and p-values measured upon stimulation vs. 0 µM guanine for all tested constructs. *p < 0.05, **p < 0.01, ***p < 0.001 (unpaired T-Test, two-tailed). Source data are provided as a Source Data file.