Engineering aptamer dimers (apdimers) for optimization of synthetic riboswitches

Abstract

Riboswitches are compact RNA-based regulatory elements capable of modulating gene expression in response to small molecules, without the need for additional proteins. Various synthetic riboswitches have been engineered using in vitro-generated tetracycline and theophylline aptamers. However, many of these constructs exhibit suboptimal switching efficiency and background expression. Moreover, efforts to enhance their performance often involve time-consuming and costly screening processes. Here we report that artificial riboswitches can be efficiently optimized by engineering fusion aptamers that contain two binding pockets (apdimers). Following this rational approach, we generated cooperativity between both binding pockets, resulting in the improved performance of splicing-based and ribozyme-based synthetic riboswitches. We finally combined optimized tetracycline switches, yielding dynamic ranges exceeding 1000-fold with minimal background expression in the OFF state. In addition, we show that the optimized tetracycline riboswitches can be used to efficiently induce AAV-mediated transgene expression in mice. The presented strategy offers a straightforward and effective approach for the optimization of existing synthetic riboswitches and the design of novel riboswitches.

Graphical Abstract

Graphical Abstract.

Introduction

Gene expression regulation tools play an important role in the development and application of gene therapies [1]. Artificial riboswitches are modular RNA-based regulators that function by binding specific ligands via an aptamer, which triggers a conformational change to control gene expression. Since riboswitches can be inserted into a shuttle vector with minimal coding space requirements and do not rely on the expression of potentially immunogenic proteins, they are considered particularly promising for advancing gene therapies [1, 2]. Several synthetic riboswitches have been developed, based on different expression platforms that can be linked via communication modules to aptamers binding a desired ligand [3–11]. To advance riboswitches into clinical applications, especially for therapeutic purposes, highly efficient variants that respond to clinically relevant drugs like tetracycline with excellent switching performances are essential.

Aptamers that bind to specific ligands can be produced in vitro via the systematic evolution of ligands by exponential enrichment (SELEX) [12, 13]. The first engineered riboswitches designed for use in human cells included the theophylline-binding aptamer, developed by Jenison et al. in 1994, which has demonstrated functionality across various applications [14, 15]. Another very widely used in vitro-generated aptamer is the tetracycline aptamer developed by Berens et al. [16]. Tetracycline aptamers have been extensively studied and used to create several synthetic riboswitches [17]. In 2015, the Suess group published a tetracycline-inducible aptazyme design based on a type III Hammerhead ribozyme (HHR) [4, 18]. This construct was demonstrated to function in eukaryotic cells and organisms and was used for adeno-associated virus (AAV)-mediated transgene expression in mice [4, 19–21]. The Farzan group developed an OFF switch by attaching the tetracycline aptamer to a type I HHR and also used these constructs to control the expression of AAV-vectored transgenes in mice [5]. Both findings allowed for the application of tetracycline switches to regulate potential therapeutic genes, including VEGF-B for cardiovascular disease and prostaglandin F2α biosynthesis for glaucoma treatment [22, 23]. Tetracycline riboswitches that control the accessibility of splice sites were shown to efficiently regulate reporter gene expression in mammalian cells and C. elegans with the advantage of a comparably low background expression [7, 24]. These constructs regulate exon skipping by ligand-dependent aptamer folding, causing splice-site masking of an alternative exon with a premature stop codon. Switches based on alternative splicing were further used to study the function of CD20, a target for the treatment of B-cell-associated diseases [25].

The variety of approaches taken by researchers to design synthetic tetracycline-responsive switches has resulted in the development of a wide range of expression platforms, offering to interfere with transgene expression at different stages of messenger RNA (mRNA) maturation or focusing on different regulatory mechanisms [17]. However, most of these platforms are not optimized to induce maximum fold-change and display significant background levels of gene expression in the OFF state, which discourages the use of such systems for scientific or medical purposes. However, in principle tetracycline aptamer-based RNA switches can be further optimized, as exemplified by an advanced design in a recently published pA regulator system [8]. In this study, Yen et al. were able to design an exceptionally efficient switch by placing a polyA signal in a scaffold between three aptamers that additionally control a G-quadruplex-assisted splice site [8].

More general approaches to improve riboswitch performances often involve enhancing the interaction between the aptamer and the expression platform. This can be achieved through various approaches, including rational design [4, 5, 7] and NGS-based screening approaches [26]. Additionally, in some cases reducing background expression can be accomplished by arranging small constructs, such as aptazymes, in tandem [4, 5] or by combining different regulatory mechanisms within the designs [11, 24]. However, most of these attempts have not led to substantial improvements that would be desirable for the application of tetracycline riboswitches in therapeutic contexts [26, 27]. Furthermore, screening approaches can be labor-intensive and often require individual repetition for each riboswitch design. Currently, a simple and efficient optimization strategy is lacking.

Some RNA aptamers that have been identified in natural riboswitches bind more than one ligand cooperatively. The tetrahydrofolate aptamer contains two binding pockets [28]. Another example is the guanidine II riboswitch that consists of two ligand-binding stem-loops that dimerize and thereby show cooperativity with regard to ligand binding [29, 30]. The close proximity of the two binding pockets allowed for the use of alternative ligands that contain two guanidine groups connected by suitable linkers, thereby bridging both binding pockets and binding the riboswitch with increased affinity compared to the natural ligand [31, 32]. In 2001, Jose et al. fused two aptamers with responses to different ligands in a synthetic aptazyme construct and observed enhanced ribozyme activity upon addition of both ligands in vitro, suggesting the possibility to achieve cooperative ligand binding via the incorporation of two different aptamers in an artificial riboswitch design [33]. In a recent study, Kelvin et al. fused tetracycline- and different aminoglycoside-binding aptamers in order to generate efficient logic gates based on roadblock mechanisms in yeast [34].

Here, we report that synthetic tetracycline and theophylline riboswitches with functionality in mammalian cells can be optimized by fusing two identical aptamers, generating more efficient sensor domains that contain two binding pockets. We will refer to this concept as apdimers in the following. This strategy opens the possibility of integrating apdimer versions into different riboswitch platforms, such as aptazymes or splicing-based switches, which results in cooperative ligand binding and improved switching performance. Combination of an apdimer-tetracycline-responsive HHR riboswitch and a splicing-based apdimer-tetracycline riboswitch resulted in a greatly optimized riboswitch system with superior switching performance of up to 1000-fold dynamic range. We further show that such optimized tetracycline riboswitches can be used to induce AAV-mediated transgene expression in mice with high efficiency and repeatedly while ensuring minimal background expression in the absence of tetracycline.

Materials and methods

Plasmid construction

Each riboswitch was introduced via PCR in a psiCHECK-2 vector, encoding Renilla luciferase (hRluc) and firefly luciferase (hluc+). Splicing-based riboswitches were inserted in a splicing cassette as previously described [7] to modulate hluc+ expression, while hRluc expression served for transfection control. Aptazymes were inserted into the 3′-UTR of hRluc, and hluc+ served as internal control. For combined constructs, the splicing cassette and aptazyme were both regulating hluc+ expression. For the IL-2 reporter, hluc+ was replaced by mouse IL-2 cDNA (Sinobiological). All oligonucleotides were purchased from Sigma–Aldrich. Plasmids were purified using a Zyppy Plasmid Miniprep kit (Zymo Research). The sequences of the constructs described in this study are listed in Supplementary Table S1. Selected hluc+ gene cassettes with catalytically active or inactive aptazymes were synthesized in an AAV cis plasmid backbone carrying the chicken β-actin/CMV immediate-early enhancer hybrid (CAG) promoter, and an SV40 polyadenylation (pA) signal. The expression cassettes were flanked by AAV2 inverted terminal repeats (ITRs).

Cell culture experiments

HeLa cells were grown in Dulbecco’s modified Eagle’s medium (Gibco DMEM, Fisher Scientific) supplemented with 10% (v/v) fetal calf serum (FCS) and 1% (v/v) penicillin/streptomycin (Fisher Scientific) at 37°C in 5% CO₂ humidified atmosphere. Twenty-four hours prior to plasmid transfection, 8000 cells were seeded in 96-well plates. Lipofectamine 3000 (Invitrogen) was used for transient transfection of HeLa cells. After incubation of 4 h at 37°C, the medium was replaced by Dulbecco’s modified Eagle’s medium (Gibco DMEM, Fisher Scientific) with tetracycline (Sigma–Aldrich) dissolved in DMSO or theophylline (Sigma–Aldrich) dissolved in 1 M NaOH and the same volume of 1 M HCl to restore the pH. Controls for theophylline were treated with Dulbecco’s modified Eagle’s medium, supplemented with 1 M NaOH and 1 M HCl. Luciferase expression was measured 24 h after transfection using the Dual-Luciferase Reporter Assay System (Promega). The luminescence was quantified with a Spark multimode microplate reader (Tecan), applying a settle time of 0 ms and an integration time of 2000 ms.

HEK293H cells were cultured in DMEM + GlutaMAX + 10% FCS at 37°C. 25 000 HEK293 cells per 96-well were seeded 24 h prior to transfection using the Lipofectamine 3000 kit (Invitrogen). Tetracycline (Tet-HCl, Sigma–Aldrich) was added to the cells 1–2 h after transfection. Luciferase activity was measured 48 h after transfection using the Dual-Glo Kit (Promega). Luciferase activity was background–corrected by subtracting the luminescence measured in non–transfected control cells.

ELISA

Secreted IL-2 was quantified using a Mouse IL-2 ELISA Kit (Invitrogen) according to the manufacturer’s protocol. Cells were seeded, transfected, and treated as described earlier. The supernatant was collected 24 h after transfection and secreted IL-2 was quantified. Briefly, a 96-microwell plate was coated with biotin-conjugated anti-mouse IL-2 antibody. The supernatant of transfected cells was diluted in PBS (uninduced samples and controls 1:1, induced samples 1:50) and was incubated on one plate with two IL-2 standard dilution series. The next day, plates were treated with Streptavidin-HRP and Substrate Solution (tetramethyl-benzidine), and the color reaction was stopped with 1 M H₂SO₄. The absorbance was determined at a Spark multimode microplate reader (Tecan), applying 450 nm as primary wavelength.

AAV production

AAV packaging and quantification have been described previously [35]. Briefly, AAVs were generated using a three-plasmid transfection [one plasmid encoding the AAV cap gene; the AAV cis-plasmid containing the expression cassette flanked by ITRs; and a pHelper (AAV Helper-free system, Agilent) plasmid] of adherent HEK293 cells. The vectors were purified by PEG precipitation, iodixanol gradients, and ultrafiltration. Genomic titers were determined by ddPCR.

Animals and treatments

Five-week-old male BALB/cAnNCrl mice used in this study were purchased from Charles River Laboratories. AAVs were diluted to the desired concentration in PBS and administered into the tail vein in a volume of 100 µl per mouse under isoflurane anesthesia (3.5–5 vol%). Tetracycline was formulated and administered as described previously [20, 21]. All animal experiments in this study were approved by the local German authorities and conducted in compliance with the German and European Animal Welfare Acts.

In vivo bioluminescence imaging (BLI)

The distribution of hluc+ was monitored using an IVIS^® Lumina III bioluminescence imaging system (PerkinElmer) with a CCD camera. For this purpose, 150 mg/kg (7.5 ml/kg) D-luciferin in PBS was injected intraperitoneally into the mice 8 min before anesthesia. Animals were placed in supine position, and light emission was measured 10 min post-injection.

Ex vivo luciferase assay

On day 21, animals were weighed. Six hours after the second tetracycline administration, the mice were anesthetized by isoflurane, and the final in vivo luciferase imaging was performed. Thereafter, animals were anesthetized with isoflurane (3.5–5 vol%) and then euthanized via cervical dislocation. Livers were dissected, snap-frozen in liquid nitrogen, then immediately transferred into tubes with ex vivo luciferase buffer and stored at −80°C until analysis. For the ex vivo luciferase assay, the liver tissue was homogenized and assayed for luciferase activity using the Luciferase Assay System kit (Promega) according to manufacturer’s instructions. The luciferase activity was read with an Enspire (PerkinElmer). The tissue weights determined during necropsy were used to normalize luciferase activities.

Quantification of AAV vector genomes in tissue

Tissue samples were snap-frozen immediately after dissection. For DNA isolation, samples were homogenized in 900 µl RLT buffer (Qiagen) + β-mercapto-EtOH (1%) using a Precellys-24 homogenizer. Afterwards, samples were immediately placed on ice and subjected to phenol-chloroform-isoamyl alcohol (Sigma–Aldrich) extraction. DNA was purified using the KingFisher (ThermoFisher) with the MagMAX DNA Multi-Sample Ultra 2.0 Kit (Applied Biosystems™). AAV vector genomes were detected after DNA extraction via duplex dPCR (QIAcuity, Qiagen) in a 96-well Nanoplate (Qiagen) for simultaneous detection of vector genome and GAPDH DNA. A custom TaqMan assay (Thermo Scientific) was used to detect the target region in the CAG promoter. Oligonucleotide primers 5′-CCAAGTACGCCCCCTATTGAC-3′ (forward) and 5′-CTGCCAAGTAGGAAAGTCCCATAAG-3′ (reverse) and 5′-CCGCCTGGCATTATG-3′ (probe) were utilized for this assay. TaqMan assay Mm00186825_cn was used to detect mouse GAPDH.

Data analysis

Statistical data analysis was performed using GraphPad Prism 5 or 10 from GraphPad Software, Inc., San Diego, USA. All data were analyzed using a two-sample unpaired t-test, except for the mouse AAV data, which were analyzed using a one-way ANOVA with Dunnett’s post-test. Dose-response curves of in vivo tetracycline responses were fitted using the “log(agonist) versus response – variable slope (four parameters)” function in Prism 5, deriving the EC₅₀ and Hill coefficient (. This four–parameter logistic (variable slope) function (Hill fit) fits the following equation:

where X is the log₁₀(agonist concentration), Bottom and Top are the lower and upper plateaus, respectively, and is the Hill slope (Hill coefficient).

Results

Design of an optimized P1-connected tandem tetracycline apdimer for splicing-based riboswitches

To optimize riboswitches that are based on different expression platforms, we aimed for a method that focuses on the aptamer instead of the entire riboswitch construct. Recent results suggested that the fusion of two stem-loop aptamers improved the switching performance of a splicing-based xanthine/oxypurinol-responsive riboswitch [36]. We followed the hypothesis that such dimeric aptamers with two binding pockets bind their ligand in a cooperative manner. In addition, we speculated that cooperative binding could enhance the effect on the expression platform and thereby enhance riboswitch efficiencies. In order to assess whether this principle is transferable and broadly applicable, we set out to improve synthetic tetracycline riboswitches. Upon ligand binding, the tetracycline aptamer undergoes a conformational change that involves stabilization of stem P2 and P1 [5, 37, 38]. In a synthetic riboswitch construct, the aptamer is connected to the expression platform via one of these stems and transfers the stabilization upon ligand binding [4–8, 24, 39] Hence, we speculated that both stems could be opened and used to connect the aptamer to a second aptamer in order to generate cooperative ligand-binding (Fig. 1). In this study, we designed two different versions of these dimeric aptamer constructs: first, in a splicing-based construct, a tetracycline aptamer was attached to a P1-connected aptamer via its P1 stem, resulting in a design where both aptamers show the same orientation. In a second construct, we again attached a second aptamer via its P1 stem, but to an aptamer that is connected to the HHR ribozyme via the P2 stem. In this design, the aptamers have the opposite orientation.

Figure 1.

Schematic depiction of the design of dimeric tetracycline aptamer structures. In artificial riboswitch designs, the tetracycline aptamer is connected to an expression platform via stem P1 or stem P2. Fusion of two aptamers generates a dimeric structure (apdimer) with two binding pockets, proximal and distal to the expression platform, that could bind the ligand in a cooperative manner.

First, we aimed for an optimized version of the tetracycline aptamer, which would improve the dynamic range of previously described splicing-based tetracycline switches (Fig. 2A) [7]. We opened stem P2 of the tetracycline aptamer and connected it to a second tetracycline aptamer. The fusion stem was initially designed to closely resemble the original closing stem in order to minimize disruption of the aptamer structure. In addition, we introduced a point mutation (A42G) in the aptamer that improves the performance of tetracycline switches in human cells by influencing the stability of the aptamer (Fig. 2B) [8, 40]. To probe the effect of this mutation independent from the effect of the apdimer design, we introduced it into the Tet9_ex (herein referred to as Tet_splice_1) construct, reported in Finke et al., which contains only a single aptamer [7]. This resulted in construct Tet_splice_2, and we compared the switching performance of both constructs in transfected HeLa cells (Fig. 2 and Supplementary Fig. S1). The A42G mutation indeed seems to improve the switching performance of the splicing-based construct Tet_splice_1, resulting in an increase in the dynamic range from 28-fold to 39-fold for Tet_splice_2 (Fig. 2C). Moreover, Tet_splice_2 displays higher sensitivity with an EC₅₀ of 21.8 µM, compared to the original construct Tet_splice_1 (EC₅₀ = 49.8 µM) (Supplementary Fig. S1). The Tet_splice_apdimer_1 construct containing the engineered apdimer switches gene expression to a similar level compared to Tet_splice_1, but with a notable decrease in expression in the OFF state. This background reduction causes a significantly higher dynamic range of up to over 200-fold induction (Fig. 2C). We also placed the apdimer in the Tet13_el construct that does not contain the alternative exon and uses the aptamer with two UAA stop codons in frame as described in Finke et al. [7]. This design achieved a slightly higher ON state than Tet_splice_apdimer_1 and a dynamic range of over 150-fold, while requiring significantly less coding space (Supplementary Fig. S2). Additionally, we introduced a second point mutation A53U into both aptamers of Tet_splice_apdimer_1 (Fig. 2B), resulting in Tet_splice_apdimer_2. This A53U mutation was reported to increase the regulatory property of tetracycline-aptamer-controlled translation in yeast by Hanson et al. [41]. For Tet_splice_apdimer_2 >300-fold induction of gene expression was observed (Fig. 2C).

Figure 2.

Apdimer construction and comparison of switching performances of two point mutations in the tetracycline aptamer in HeLa cells. (A) Schematic depiction of the splicing-based riboswitch constructs. The inclusion of an alternative exon (alt. exon) containing a stop codon (red) is regulated by a single aptamer (Tet_splice) or an apdimer (Tet_splice_apdimer). (B) Sequence of the Tet_splice and Tet_splice_apdimer constructs. The 5′ splice site masked in the P1 stem is highlighted in red, and introduced mutations are indicated in red boxes and labeled, respectively. (C) Dual luciferase assay showing the performance of tetracycline apdimer-containing splice switches in HeLa cells compared to constructs with the conventional tetracycline aptamer 24 h after transfection and induction with 50 µM tetracycline (light gray). Values show the luciferase activity normalized to a psiCheck control without riboswitch. Shown is the mean ± SD of three biological replicates. **P < .01, ***P < .001.

We speculated that the apdimer aptamers bind their ligands in a cooperative manner, thereby causing the improved switching performance. To investigate whether both aptamers contribute to the optimized performance, we created tetracycline-binding-deficient control constructs. We introduced a mutation that abrogates tetracycline binding (A9U, Fig. 3A) into each tetracycline binding pocket of Tet_splice_apdimer_1 [7]. While mutating the distal aptamer (Tet_control_1) reduced the switching effect to 36-fold (comparable to the single aptamer design), mutation of the proximal aptamer (Tet_control_2) abolished switching completely (Fig. 3B). We thus concluded that the improved switching efficiency is a result of both aptamers binding tetracycline.

Figure 3.

Comparison of Tet_splice_apdimer_1 and controls with individually mutated binding pockets. (A) Sequence of the Tet_splice_apdimer_1 and control constructs. An A to U mutation (A9U) rendering the aptamer binding-deficient was introduced in the distal (Tet_control_1) or proximal (Tet_control_2) aptamer as indicated in red boxes. (B) HeLa cells were transfected with psiCheck vectors containing the splicing cassette with active or control tetracycline switches in the firefly coding sequence. Shown is the mean of three biological replicates in the absence (dark gray) or presence (light gray) of 50 µM tetracycline. The gene expression was normalized to a psiCheck control without regulatory element. **P < .01. (C) Concentration-dependent induction fold of Tet_splice_2 and (D) Tet_splice_apdimer_1. The concentration of tetracycline for half-maximal effect on gene expression (EC₅₀) and the Hill coefficient (n_H) obtained from a Hill fit are reported; see the “Materials and methods” section. Data points represent the mean (± SD) of three independent experiments performed in triplicates.

Next, we compared the concentration-dependent response curves of Tet_splice_2 containing the monomeric aptamer and Tet_splice_apdimer_1 containing the dimeric aptamer (Fig. 3C and D). A sigmoidal curve fit allowing for a variable slope revealed a Hill coefficient of 1.9 for the dimeric construct, opposed to 1.0 for Tet_splice_2, demonstrating positively cooperative behavior. However, the sensitivity of Tet_splice_apdimer_1 (EC₅₀ = 23.1 µM) compared to Tet_splice_2 (EC₅₀ = 21.8 µM) was not increased (Fig. 3C and D).

Design of a P2-connected tetracycline apdimer for aptazyme optimization

Since fusing aptamers resulted in the most effective optimization of splicing-based tetracycline switches, we investigated whether we could use the same strategy to improve the fold change of a tetracycline-responsive aptazyme. Several studies have utilized tetracycline aptamers linked to HHR ribozymes as synthetic riboswitches [4, 5, 18, 42]. If connected to stem I or stem II of the ribozyme, the communication module between aptamer and ribozyme allows tetracycline-dependent inhibition of the ribozyme. Binding the ligand initiates structural changes and compacts the aptamer, which then stabilizes the P1 and P2 stems [5, 37, 38]. This, in turn, disrupts base pairing between the ribozyme loops necessary for its catalytically active conformation [4, 43–45] (Fig. 4C). Here, we connected the P2 stem of the tetracycline aptamer to stem I of the HHR type III ribozyme, resulting in the construct Tet_HHR, which combines design principles from previous studies [4, 5].

Figure 4.

Evaluation of tetracycline apdimers that are connected to the HHR via the P2 stem. (A) Schematic depiction of the aptazyme-based riboswitch constructs. The aptazyme constructs contain one (Tet_HHR) or more tetracycline aptamers (Tet_HHR_apdimer_1–3) connected to the HHR type III. In the tandem construct, two copies of the Tet_HHR_apdimer_2 aptazyme are inserted into the 3′-UTR. (B) HeLa cells were transfected with a psiCheck control vector and psiCheck vectors containing the tetracycline aptazymes constructed with a single tetracycline aptamer (Tet_HHR) or tetracycline apdimer versions (Tet_HHR_apdimer_1, Tet_HHR_apdimer_2, Tet_HHR_apdimer_3) or two copies of Tet_HHR_apdimer_2 (Tet_HHR_apdimer_2_tandem) in the Renilla luciferase 3′-UTR. (C) Sequence of the Tet_apdimer constructs and controls. Point mutations introduced in the construct Tet_HHR_apdimer_3 and the non-binding controls are indicated in red boxes. Construct Tet_HHR_apdimer_1 contains a shorter stem, which is indicated in blue brackets. The nucleotides involved in crucial interactions between the ribozyme loops are highlighted in bold. (D) Comparison of the optimized Tet_apdimer constructs with controls of Tet_apdimer constructs containing mutations in the binding pocket of the distal (Tet_control_1) or the proximal (“bridging”) aptamer (Tet_control_2). Bar plots show the mean of at least three biological replicates in the absence (dark gray) or presence (light gray) of 50 µM tetracycline. The gene expression was normalized to a psiCheck control without regulatory element. **P < .01, ***P < .001, ****P < .0001.

With the aim of designing a more efficient aptazyme with lower background expression, we chose to optimize the P2-connected construct again via apdimer incorporation. We opened the tetracycline aptamer at the P1 stem and attached a second aptamer via its P1 stem, resulting in an aptazyme design containing two aptamers with different orientations (Tet_HHR_apdimer_1, Fig. 4A and C). Here, we used the same fusion stem to attach the distal aptamer as in the apdimer design for the splicing switches, but retained the first AU base pair of the P1 stem of the proximal aptamer. In a second design, we added an AU base pair to stem P2 of the ribozyme-connected aptamer to stabilize the communication module (Tet_HHR_apdimer_2, Fig. 4C). We then again introduced the aptamer-stabilizing mutations A42G and A53U as described above (Tet_HHR_apdimer_3, Fig. 4C). To compare the apdimer versions with Tet_HHR, we transfected HeLa cells with all three constructs and measured the luminescence after induction with tetracycline. All apdimer constructs display a higher dynamic range compared to the original Tet_HHR switch (Fig. 4B). The highest fold change was achieved by Tet_HHR_apdimer_3, the mutated and stabilized version containing an additional AU base pair. However, this construct also shows the highest background gene expression with ∼8%. The non-mutated construct Tet_HHR_apdimer_2 exhibits reduced background expression of ~5% and a similar fold change. When inserting two Tet_HHR_apdimer_2 aptazymes in tandem, we observed further reduction of the background expression and only a small decrease of the fold change (Tet_HHR_apdimer_2_tandem, Fig. 4A and B).

To further examine the contribution of each aptamer to the switching performance in the ribozyme-based design, we again mutated both binding pockets of Tet_HHR_apdimer_2 individually. Mutation of the distal aptamer decreased the dynamic range to four-fold (Fig. 4D). In contrast, mutation of the proximal aptamer, directly connected to the type III HHR, resulted in almost complete depletion of the switching performance. This led us to the conclusion that the connection of a second, distal aptamer adds a ligand-dependent stabilization of the proximal aptamer, allowing for the generation of more efficient ribozyme-based riboswitch platforms.

Design of a P1-connected theophylline apdimer for splicing-based riboswitches

To investigate the transferability of the apdimer concept to other systems, we additionally designed a theophylline apdimer. In a theophylline-responsive splice ON switch (Theo_splice), we opened the P3 stem and connected a second aptamer via the P1 stem (Fig. 5A) [46]. Also in this context, the initial fusion stem was designed to preserve the original closing stem; however, stem elongation was necessary to obtain functional and more efficient constructs. By adjusting the length of the connecting stem, we could generate constructs with suitable stability (Fig. 5B and C). A P1/3 connection stem of four base pairs appeared to be too unstable, as it resulted in a non-switching construct (Theo_splice_apdimer_1). However, constructs with a connection stem of 5, 6, and 7 base pairs increased the switching performance of the Theo_splice riboswitch, while a stem of 8 base pairs increased the expression levels but lowered the fold-change (Fig. 5C). Addition of a GC base pair to the P1/3 connection stem resulted in the most efficient construct (Theo_splice_apdimer_2) reaching an induction ratio of >90-fold (Fig. 5C). In summary, an improved performance was achieved by replacing the single-aptamer version with an apdimer construct.

Figure 5.

Comparison of switching performance of Theo_splice and Theo_splice_apdimer versions in HeLa cells. (A) Schematic depiction of the theophylline apdimer design. The aptamers are fused between stem P3 and P1. (B) Sequence and the predicted secondary structure of the theophylline aptamer in the splicing-based expression platform (Theo_splice) and the dimeric structure (Theo_splice_apdimer_1) containing fusion regions of various lengths (apdimer_2–5). (C) Dual-luciferase assay showing the performance of theophylline apdimer-containing splice switches (Theo_splice_apdimer1–5) in HeLa cells compared to the construct with the conventional theophylline aptamer (Theo_splice) 24 h after transfection and induction with 2 mM theophylline (light gray). Values show the luciferase activity normalized to a psiCheck control without riboswitch. Shown is the mean ± SD of at least three biological replicates. **P < .01, ***P < .001, ****P < .0001.

Combination of splicing- and aptazyme-based tetracycline apdimer switches and 4-epitetracycline as effector

Finally, we combined both optimized tetracycline-responsive constructs by placing the Tet_HHR_apdimer_2 construct in the 3′-UTR of a firefly luciferase that contains the splicing construct Tet_splice_apdimer_1 (Tet_apdimer_combi_1) (Fig. 6A). Previous studies have shown that combining aptazymes and splicing-based switches can improve the overall switching performance [11, 24]. We evaluated the switching performance of the construct in HeLa cells. The dynamic range of this combined construct is increased to >1000-fold due to a reduction of the background gene expression in the OFF state that amounts to 0.01% expression compared to the parental psiCheck vector (Fig. 6B). When combining Tet_splice_apdimer_2, which carries the A42G, and A53U mutations and the aptazyme-based Tet_HHR_apdimer_3 (Tet_apdimer_combi_2), the background expression increases slightly but also higher induction levels are reached, leading to a similar induction ratio.

Figure 6.

Comparison of performance of the original Tet_splice switch and the combined construct containing the Tet_splice_apdimer switch and the Tet_HHR_apdimer switch in the 3′-UTR. (A) Schematic depiction of the combined apdimer-containing splicing-based and aptazyme-based riboswitch constructs (Tet_apdimer_combi constructs). (B) HeLa cells were transfected with psiCheck control vector and psiCheck vectors containing the tetracycline switches and incubated without (dark gray) or with (gray) 50 µM tetracycline after normalization to a psiCheck control without regulatory element. Shown is the mean of at least three biological replicates, each performed in triplicates. (C) Dose response curve of the Tet_apdimer_combi_1 switch. Shown is the mean fold change in luciferase expression of three biological replicates (mean induction ratio) after psiCheck normalization. In addition, the concentration of tetracycline for half-maximal effect on gene expression (EC₅₀) and the Hill coefficient (n_H) obtained from a Hill fit are reported. (D) Concentration-dependent response of the Tet_apdimer_combi_1 riboswitch to tetracycline (blue) and 4-epitetracycline (red). Shown is the mean fold change in luciferase expression of three biological replicates (mean induction ratio) after psiCheck normalization.**P < .01, ***P < .001, ****P < .0001.

For Tet_apdimer_combi_1, we further measured the concentration-dependent increase in gene expression upon addition of tetracycline and determined an EC₅₀ of 28.3 µM (Fig. 6C). With this switch combination, 200-fold induction is already achieved at a concentration of 12.5 µM tetracycline.

One of the most prevalent degradation products of tetracycline is 4-epitetracycline (4-ET). This less antibiotic degradation product is formed upon epimerization under mildly acidic aqueous conditions (pH 3–6) and makes up ∼5% of excreted tetracyclines in patients [47–49]. Since the tumor microenvironment is characterized by a lower pH compared to the physiological pH, formation of 4-ET might play a role for applications in cancer therapy [50, 51]. To investigate whether the optimized switches respond to 4-ET with the same efficiency, we additionally measured a concentration dependency using this ligand (Fig. 6D). In line with previous studies, we observed no deficit in the switching performance with 4-ET and even improved fold changes [52]. Thus, we conclude that the performance of these improved switches is not hampered by the formation of 4-ET under physiological conditions. Moreover, 4-ET might represent an alternative for clinical applications due to its reduced antibiotic activity and improved effects on tetracycline-aptamer-dependent riboswitches.

Controlling interleukin-2 expression with tetracycline splice–aptazyme combination switches

The herein developed fusion constructs achieve high dynamic ranges and display very low background expression, which is an advantage for potential therapeutic applications. Especially gene therapies that involve the expression of toxic genes or genes that trigger the immune system require tools for precise dosing that could be realized with such artificial riboswitches [21]. Hence, we aimed to investigate the transferability of the results from the luciferase reporter to a therapeutic gene. Interleukin-2 (IL-2) exhibits dose-dependent effects on the immune system, stimulating regulatory T-cells at low doses, while high doses activate a broader immune response. The latter is utilized clinically as an anti-tumor therapy (aldesleukin) [53–55]. Despite its therapeutic potential, IL-2’s short half-life necessitates frequent administration, which can lead to side effects. To address this, approaches employing IL-2 vectorization are under investigation [56]. Given its dual role in immune modulation, precise control over IL-2 expression could significantly enhance its therapeutic efficacy and safety in both autoimmune and cancer treatments. We therefore introduced the optimized Tet_apdimer_combi_1 switch into the sequence of a murine interleukin-2 gene (mIL-2) and quantified mIL-2 expression in HeLa cells in presence or absence of tetracycline (Fig. 7).

Figure 7.

Controlling interleukin-2 expression using the tetracycline splice-aptazyme combination switch. (A) Top: Schematic representation of the dose-dependent immunosuppressive or -stimulatory effect of IL-2. Bottom: Design of the IL-2 construct under control of the Tet_apdimer_combi_1 riboswitch with the splice-switch inserted at different positions (splice sites 1 and 2), which result in IL-2_Tet_combi_1 and IL-2_Tet_combi_2. The sequences were inserted into a psiCheck vector, replacing the firefly sequence. (B) Relative IL-2 expression evaluated via ELISA. Shown is the mean of three biological replicates in the absence (dark gray), presence of 10 µM (medium gray) or 50 µM (light gray) tetracycline. Values below the detected background were set to 0. The gene expression was normalized to a psiCheck vector containing the IL-2 gene sequence without regulatory element. Bar plots show the mean (± SD) of three independent replicates, each measured in triplicate. *P < .05.

Upon addition of 50 µM tetracycline, we observed induction of ∼250–500-fold with the splice-aptazyme combination using two different insertion sites for the splicing-based construct (Fig. 7A and B). The experiment was additionally carried out at a concentration of 10 µM tetracycline, which reflects a concentration that is more likely to be achievable in the targeted tissues in mammals [20, 47]. Also at this concentration, induction of ∼50–70-fold could be reached when testing the splice-aptazyme combination. Importantly, we observed that inserting the splicing construct at a splice site closer to the 5′-end of the mRNA (IL-2_Tet_combi_1) resulted in overall reduced expression levels compared to the insertion of the same construct at a splice site further downstream (IL-2_Tet_combi_2, Fig. 7B). Consequently, the leakiness in the absence of tetracycline was further minimized and, in some measurements, became undetectable. Therefore, we consider the reported dynamic range in such experiments not very meaningful. However, we have included this value to indicate that stable induction is achieved with this system.

Regulation of AAV-mediated transgene expression by tetracycline apdimer containing riboswitches in mice

We have previously shown that the tetracycline-dependent K19 aptazyme (TetK19) has the potential to control adeno-associated virus (AAV)-mediated transgene expression in mice, albeit at relatively high background activity in the absence of its ligand [20, 21]. We therefore decided to adopt selected tetracycline-dependent, apdimer-based riboswitches for AAV-mediated gene delivery. In a first step, we cloned hluc+ with or without riboswitches in an AAV-plasmid backbone with the strong ubiquitous chicken-β-actin (CAG) promoter. This created plasmids pAAV control, pAAV-TetK19, pAAV-Tet_HHR_apdimer_2_tandem, and pAAV-Tet_apdimer_combi_1 active (Fig. 8A). In addition, catalytically inactive versions [19] were generated, resulting in constructs pAAV-TetK19 inactive and pAAV-Tet_HHR_apdimer_2_tandem inactive. Transfection of these plasmids in HEK293H cells confirmed a 9.8-fold range for pAAV-Tet_HHR_apdimer_2_tandem active and a 1959-fold range for pAAV-Tet_apdimer_combi_1 active after induction with 31 µM tetracycline (Fig. 8B). pAAV-TetK19 did show a 1.6-fold induction that did not reach statistical significance. Constructs with no or inactive switches showed no tetracycline response. The stronger expression observed for pAAV-based constructs compared to psiCheck reflected the difference in promoter strength between these constructs (CAG versus HSV-TK). In summary, these data suggest modularity of the apdimer switches, as their functionality was confirmed in an AAV plasmid backbone, with a strong promoter and in HEK293H cells.

Figure 8.

Regulating AAV-mediated transgene expression by tetracycline-apdimer riboswitches. (A) Schematic representation of AAV gene expression cassettes containing no (pAAV control) or the indicated catalytically active/inactive tetracycline riboswitches. (B) hluc+ activity in HEK293H cells 48 h after transient transfection of psiCheck or AAV plasmids containing gene expression cassettes shown in panel (A) and incubation in the absence (dark gray) or presence (light gray) of 31 µM tetracycline. Shown is the mean (± SD) of three biological replicates. (C) In vivo luciferase activity of mice injected i.v. with 5 × 10¹² vector genomes (vg) per kg body weight of AAV8-packaged constructs shown in (A) and treated with 30 mg/kg tetracycline on days 14 and 21 after AAV administration. N = 8 animals per group. Shown is the mean (± SD). Dashed lines indicate an undetermined relation. Color code also applies to (D) and (E). (D) Ex vivo luciferase activity in liver tissue isolated on day 21 from mice in panel (C). Shown is the mean (± SD). Statistical analysis was performed comparing it to AAV8 control group. (E) Quantification of vector genomes in liver tissue isolated on day 21 from mice in panel (C). Shown is the mean (± SD). *P < .05, **P < .01, ***P < .001, ****P < .0001.

For subsequent experiments in mice, we focused on the liver as tetracycline reaches high exposure in this organ [20]. Therefore, we packaged the AAV cassettes in AAV8, a serotype with strong hepatotropism. In the mouse liver, AAV8-mediated transgene expression is known to peak at 14 days post-delivery and remains stable afterwards. The study design included injecting male BALB/c mice with six different AAV vectors at 5 × 10¹² vector genomes (vg)/kg. Luciferase activity was monitored by bioluminescence imaging on day 13 (one day before the first tetracycline administration), on day 14 (6 h after the first tetracycline administration), on day 15 (24 h after the first tetracycline administration), and on day 21 (6 h after the second tetracycline administration). The BLI signal was restricted to the upper abdominal cavity, suggesting transgene expression was limited to the liver (Supplementary Fig. S3). Treatment with 30 mg/kg tetracycline induced luciferase expression 3.8-fold, 12-fold, and 85-fold in the groups AAV8-TetK19 active, AAV8-Tet_HHR_apdimer_2_tandem active, and AAV8-Tet_apdimer_combi_1 active, respectively (Fig. 8C). In line with our previous data [21], luciferase levels had returned to baseline one day after the initial tetracycline exposure (day 15). The tetracycline re-challenge (day 21) produced a fold change that was comparable to the respective initial induction level for all three groups with active switches. As expected, the AAV control group and the groups receiving inactive switches showed both high and comparable luciferase activity at baseline (day 13). A slight increase in transgene expression was observed in these control groups at the first tetracycline exposure (day 14) and upon re-challenge (day 21), ranging from 1.1- to 1.8-fold. The luciferase enzyme activity in protein lysates from liver tissues obtained at necropsy (day 21, 6h) corroborated the BLI data (Fig. 8D). Notably, while expression levels induced by tetracycline reached 71% and 81% of AAV8 control for AAV8-TetK19 active and AAV8-Tet_HHR_apdimer_2_tandem active, respectively, they accounted for only 2% for AAV8-Tet_apdimer_combi_1 active. An even pattern of liver transduction across all treatment groups confirmed homogenous dosing (Fig. 8E).

Discussion

RNA-based induction of gene expression is a promising tool for the development of spatiotemporally inducible gene therapies. In the past years, many synthetic riboswitches with responses to tetracycline and theophylline were developed based on the SELEX-derived aptamers from Jenison et al. and Berens et al. [14, 16]. Here, we show that tetracycline and theophylline riboswitches can be optimized by fusing two aptamer structures to generate aptamers that bind their ligands in a cooperative manner. In addition, we used this approach to improve splicing-based tetracycline riboswitches and aptazymes, showing that it can be applied to optimize riboswitches based on different riboswitch mechanisms. Combination of the improved constructs resulted in an even higher fold change and lower background gene expression. We observed an improved switch performance in response to the C4 epimer of tetracycline, 4-epitetracycline (4-ET). Thus, this tetracycline derivative with reduced antibiotic activity [49] could serve as an alternative ligand in clinical applications. We further show that the optimized tetracycline switches are suitable to control gene expression of a luciferase reporter in different cell lines and vectors as well as the expression of IL-2, which is used for therapeutic applications. In addition, the functionality of the riboswitches and improved performance of the apdimer constructs enabling regulation of AAV-mediated transgene expression in mouse liver tissue could be confirmed. The best-performing apdimer construct exhibited over a 2000-fold reduction in background gene expression compared to the benchmark TetK19, thereby offsetting the reduced maximal expression. These properties would be particularly advantageous for transgenes such as interleukin-12 (IL-12), which exert beneficial effects at low effective doses but have a relatively narrow therapeutic window [21].

To date, riboswitch optimization has mostly focused on changing length and composition of communication modules. Lately, further optimization strategies have been introduced mainly for splicing-based riboswitches that include the introduction of anti-stems for balancing ON- and OFF-states as well as the inclusion of splicing-modulating G-quadruplex structures [8, 36, 57]. In addition, multiple fused aptamer domains have been used in the so-called pA regulator system, which was optimized by adding additional expression platform elements [8].

The apdimer design presented in this study represents a simple yet effective optimization approach that focuses on the aptamer domain. However, our findings suggest that the fusion stem linking the aptamers may require adaptation depending on the specific aptamer context. Based on our results, starting with linker sequences that closely resemble the original closing stem may provide a useful strategy for construct design, while systematic variation of stem length and composition could further optimize construct performance in future studies.

The versatility of the tetracycline aptamer, demonstrated by its applicability in various contexts such as aptazymes, U1-snRNP-based or splicing-based switches, as well as in more complex systems like the pA regulator, highlights its suitability for integration into diverse environments [4–9, 17, 39, 58]. This also underlines the robustness of the tetracycline-induced structural changes even when positioned at different locations within the mRNA and surrounded by different sequences. However, we want to point out that some applications could necessitate specific features of the riboswitch platform mechanism, for example independence of transcriptional processes upon application in RNA viruses [59, 60]. To leverage the various designs effectively, an easily implementable optimization strategy could catalyze the emergence of numerous efficient riboswitches based on the preferred mechanism.

The tetracycline aptamer is a rare example of an in vitro-generated aptamer that is highly functional in many different riboswitch designs. Often SELEX-derived aptamers do not transfer a sufficient structural rearrangement upon ligand-binding [61]. However, the recently developed approach of capture-SELEX holds promise to facilitate the selection of ligand-specific aptamers that undergo structural rearrangement while bound to the ligand [62, 63]. With the development of riboswitch-suitable aptamers capable of binding additional therapeutically applicable ligands with favorable pharmacological properties, riboswitches will become more attractive for therapeutic applications. Since we have applied the presented apdimer strategy successfully to optimize tetracycline-, theophylline-, and oxypurinol-triggered switches [36], we are optimistic that the approach should be easily applicable to the development of more efficient riboswitches containing further aptamers.

Supplementary data

Supplementary data is available at NAR online.

Conflict of interest

V.H. and J.S.H. have filed a patent application that includes parts of the apdimer concept (PCT/EP2025/055 596).

Funding

Open access publication costs were provided by the University of Konstanz.

Data availability

The data underlying this article are available in the article and in its online supplementary material.