To illustrate how Bn-AFF molecules flow from OFF to ON states in response to FKBP/cpFKBP domain folding (induced by rapamycin binding), we developed a thermodynamic model that describes the interconversion between all states of the switch. Assuming two-state character for all folding/unfolding reactions, the Bn domain of Bn-AFF can exist in three conformations (N-fold, CP-fold, or globally unfolded), and the FKBP and cpFKBP domains can each exist in folded or unfolded conformations. The minimal structural model of Bn-AFF switching therefore consists of 3 × 2 × 2 = 12 states. Ligand was allowed to bind to the states in which the FKBP and/or cpFKBP domains were folded, with the same K_d value of 1 nM. The free energies of FKBP and cpFKBP unfolding were set to −2 kcal/mol in all simulations, making 97% of those domains unfolded in the absence of rapamycin. Without ligand, most Bn-AFF molecules were kept in the OFF state by making the N-fold more stable than the CP-fold by variable amounts. The fraction of molecules in the ON state was calculated by the number of molecules in the CP-fold divided by the total number of molecules.

The main goals of the simulations were to quantify how MEF and LCE individually cause the molecules to flow from N-fold (OFF) to CP-fold (ON) states as a function of added ligand, and how combining MEF and LCE provides added benefits. Accordingly, we first simulated the MEF-only switch using a simplified model (6 states) generated from a construct with only FKBP in the N-frame and no cpFKBP in the CP-frame (Figure 1B; Supplementary Figure S1). We then simulated the LCE-only switch (6 states) using the complementary construct that contained only cpFKBP in the CP-frame (Figure 1C; Supplementary Figure S2). Finally, we combined the two to generate the simulation of the complete 12-state Bn-AFF switch (Figure 1D; Supplementary Figure S3). Full details of the models and simulations can be found in Supplementary Note S2.

The MEF condition was established as previously described (Cutler et al., 2009; Woloschuk et al., 2021) by disallowing any state in which the FKBP domain and the N-fold (of the Bn domain) were both folded. Simulated rapamycin dose-response curves were plotted as a function of the difference in unfolding free energies of the N-fold and CP-fold (ΔΔG_N/CP, where positive values indicate the N-fold is more stable than the CP-fold). ΔΔG_N/CP defines the extent to which the switch is OFF in the absence of rapamycin and thus determines the maximum signal change of the switch. For example, when ΔΔG_N/CP equals zero, the switch exists almost entirely as a 50/50 mix of N-fold and CP-fold (Figure 2A; Supplementary Figure S4A) and the maximum turn-on is a factor of two. The MEF-only simulation revealed that rapamycin addition converted the single-FKBP switch from a mostly OFF state to the fully ON state (Figure 2A; gray curves). The shapes of the response curves were that of the standard one-site binding isotherm. The apparent K_d values for activation became progressively higher than the canonical K_d (1 nM) as ΔΔG_N/CP increased, reflecting the energetic cost to push the balance from the N-fold to the increasingly destabilized CP-fold. This cost is paid by ligand binding energy. This relationship illustrates the trade-off between maximizing the signal change (high turn-on, enforced by large ΔΔG_N/CP values), and the resulting need for large concentrations of ligand to activate the switch.

The LCE condition was established by introducing an energetic penalty to folding of the CP-fold when the inserted cpFKBP domain is unfolded. The LCE penalty can be experimentally determined for a given protein pair, but technical considerations prevented us from doing so for the present system (vide infra). We estimated the LCE penalty to be −2 kcal/mol in the simulations. The simulated rapamycin response curves revealed that LCE alone activated switching (Figure 2A; green curves), but with three differences compared to MEF alone (Figure 2A; gray curves). First, the apparent K_d of rapamycin binding in the LCE model is lower than that in the MEF model at any given value of ΔΔG_N/CP, meaning LCE activates the switch more effectively than MEF at low ligand concentrations. Second, compared to MEF, LCE results in more molecules being OFF in the absence of ligand thus establishing a lower background signal. At high ligand concentrations, however, the LCE effect saturates and LCE may not fully drive conversion to the ON state. By contrast, MEF will always convert all molecules to the ON state, given high enough ligand concentration. In this respect, LCE and MEF are complementary.

Figure 2B superimposes the responses of the MEF-only and LCE-only switches with that of the combined MEF-LCE switch (containing FKBP in the N-fold and cpFKBP in the CP-fold) in the case where ΔΔG_N/CP = 1 kcal/mol. The MEF-LCE switch combines the low background signal of LCE with the high ceiling of the MEF response. The net result is an increase in turn-on of the dual switch at all ligand concentrations, where turn-on is defined as the fraction of molecules in the CP-fold at a given [rapamycin] divided by the fraction of molecules in the CP-fold with zero rapamycin (Figure 2C). The combined switch affords as much as 15.1-fold and 6.7-fold improvements over MEF alone and LCE alone, respectively, in the conditions of Figure 2C. In summary, the simulations demonstrate that introducing two receptor domains can significantly enhance the AFF-mediated switching response, through the combined action of MEF and LCE.

We purified the isolated N-frame analog (Bn with FKBP inserted between residues 66–67) and CP-frame analog (cpBn with cpFKBP inserted between residues 110* and 1) to ascertain the effects of FKBP/cpFKBP insertion on Bn/cpBn stability, vis-à-vis the MEF and LCE mechanisms, respectively. The N-frame analog is expected to consist of folded Bn with an extended, disordered surface loop composed of the cpFKBP sequence (Figure 3A, top schematic). Upon addition of rapamycin or FK506, MEF predicts that the FKBP domain will fold and destabilize the Bn domain, potentially unfolding the latter and splitting it into two fragments connected by FKBP. As a test, we mixed the N-frame analog (24.6 kDa) with rapamycin, FK506, or DMSO vehicle and analyzed the proteins by size exclusion chromatography (SEC). The rapamycin and FK506 samples eluted at the same volume (apparent MW of 48 kDa) and earlier than the vehicle control (apparent MW of 33 kDa; Figure 3A, left). The observation that the apparent MW increased by less than a factor of two suggested that FKBP folding caused Bn to at least partially unfold, but not go on to form a domain-swapped dimer in which all domains are folded, a phenomenon that we reported earlier in other MEF systems (Ha et al., 2015; Ha et al., 2012; Karchin et al., 2017).

A unique signature of MEF is the seemingly paradoxical effect of the host protein becoming destabilized in the presence of ligand. Thus, as a further test we performed urea denaturation experiments on the N-frame analog in the presence and absence of FK506 using Trp fluorescence to monitor unfolding (Figure 3A, right). Since three of the four Trp residues are in the Bn domain, we expected that fluorescence would report mostly on Bn conformation. As expected, the N-frame analog appeared to be much less stable with FK506 (ΔG_unfold = 3.2 ± 0.1 kcal/mol) than without the drug (ΔG_unfold = 7.6 ± 0.7 kcal/mol) (Supplementary Figure S5A; Supplementary Table S1). Related to the drop in ΔG_unfold, another indication of MEF is the broadening of the Bn unfolding transition (demonstrated by a decrease in the cooperativity parameter m) in the presence of ligand. A decrease in m is normally interpreted as the presence of additional, poorly-resolved unfolding transition(s) caused by accumulation of one or more unfolding intermediates. In the case of MEF, however, m decreases due to the simultaneous unfolding of the host domain and folding of the inserted domain even as the overall transition remains two-state with no populated intermediate (Cutler et al., 2009). The observed m-value of Bn reduces to the difference in m-values of the Bn and FKBP domains under ideal experimental conditions, i.e., when Bn cannot fold when FKBP is folded, and when FKBP remains stable at denaturant concentrations that unfold Bn. These conditions were not rigorously established in the present case. Nevertheless, the decrease in m and ΔG_unfold provides evidence that FK506 binding destabilizes the N-fold as predicted by MEF.

We next asked whether the CP-frame analog was stabilized by binding-induced folding of the cpFKBP domain as predicted by LCE (Figure 3B, top schematic). In agreement, adding rapamycin increased the fitted ΔG_unfold from 3.3 ± 0.1 kcal/mol to 5.7 ± 0.7 kcal/mol and C_m from 1.76 ± 0.01 M to 1.99 ± 0.06 M (Supplementary Figure S5B; Supplementary Table S1). The difference in ΔG_unfold (2.4 kcal/mol) theoretically equals the LCE penalty. Because cpFKBP unfolding may have contributed to the observed fluorescence change (due to the Trp residue in the cpFKBP domain), this equivalency was not proven. However, the observation that both ΔG_unfold and C_m increased with rapamycin addition strongly implies that the Bn domain was stabilized by rapamycin according to the LCE model.

A key step in constructing an AFF-based switch is tuning the thermodynamic balance between the N- and CP-frames. This process generates the ΔΔG_N/CP values depicted in the simulations of Figure 2A and determines the extent to which the switch is off in the absence of drug and how much ligand is needed to turn it on. The desired balance will depend on the application, which in our case is to destroy endogenous RNA in cells treated with rapamycin. We therefore tuned Bn-AFF in HEK293T cells using mRNA degradation as the reporter. Cells were transfected with plasmids expressing Bn-AFF with Ile (the WT residue), Val, Ala, and Gly at position 96* in the CP-frame (Wood et al., 2018). I96 is buried in the hydrophobic core of Bn and replacing Ile with smaller side chains was predicted to progressively destabilize the CP-fold, causing the switch to adopt the catalytically inactive N-fold and reducing background activity. A reporter plasmid expressing mCherry, tagged with a PEST sequence to reduce protein half-life, was co-expressed to provide fluorescence readout via mCherry mRNA degradation. The control was Bn-AFF made fully inactive by introducing the H102*A catalytic mutation into the CP-frame.

Transfecting cells with the H102*A control resulted in many brightly fluorescent cells, as expected (Supplementary Figure S6A). Transfection with WT or I96*V Bn-AFF (Supplementary Figures S6B, C) yielded very few fluorescent cells, suggesting those switches possessed significant RNase activity and therefore were mostly folded in the CP-frame. Consistent with this view, we observed many cells with dim fluorescence upon transfection with the I96*A mutant (Supplementary Figure S6D), and still more, brighter cells with the I96*G mutant (Supplementary Figure S6E). Cells expressing I96*G and WT constructs were comparably bright, implying that the I96*G mutation reduced RNase activity of Bn-AFF to baseline levels. Taken together, the results demonstrate that the Bn-AFF switch could be tuned in cells by rational mutagenesis of a buried residue in one of the folds.

We next evaluated ligand-induced switching of I96*A and I96*G Bn-AFF mutants in HEK293T cells (WT and I96*V variants were dropped from further study due to their high background activities). Treating the I96*A and I96*G cells with rapamycin reduced the fluorescence significantly, suggesting activation of RNase activity (Figure 4A). Turn-on was quantified by selecting fluorescent cells and calculating the average intensity of each cell in the mCherry channel before and after rapamycin treatment (Figure 4B). I96*A and I96*G switches yielded (3.1 ± 0.3)-fold and (2.9 ± 0.7-fold) reductions in mCherry fluorescence, respectively (three biological repeats). No significant change in activity for the H102*A control was noted [(0.94 ± 0.03)-fold reduction; three biological repeats]. We therefore selected the I96*A and I96*G variants of Bn-AFF for in vitro characterization and cell viability studies.

We purified I96*A, I96*G, and H102*A Bn-AFF constructs from Escherichia coli and determined their enzymatic activities using the RNase Alert kit (Integrated DNA Technologies). As anticipated, in the absence of FK506, RNase activity of I96*A was greater than that of I96*G, with the latter value being close to background levels as judged by the H102*A control (Figure 5A). Addition of FK506 increased the RNA hydrolysis rates of I96*A and I96*G, demonstrating ligand-induced turn-on. No increase was observed for H102*A. Quantitating RNAse activity by initial velocity analysis revealed (1.9 ± 0.1)-fold and (5.3 ± 1.5)-fold turn-on values for I96*A and I96*G, respectively Figure 5B. These observations are consistent with the HEK293T cell data and support the switching mechanism in Figure 1D.

One aspect of AFF switching is that the fold shift can be slow (seconds to hours depending on the protein). This is because the switching rate is typically limited by unfolding of the N-fold (or CP-fold) which can be slow for stable proteins (Stratton and Loh, 2010; DeGrave et al., 2018; Bogetti et al., 2021). To determine the turn-on rate of Bn-AFF, we mixed the I96*G mutant with FK506, added RNA substrate at various times, and recorded the initial velocities. Interestingly, the fold shift reached completion within the dead time of the assay (∼5 min) Figure 5C. This fast rate was likely due to the modest thermodynamic stabilities of the N- and CP-frames, which lowered kinetic barriers for reaching equilibrium.

Barnase has been previously reported to kill eukaryotic cells via caspase-mediated apoptosis (Edelweiss et al., 2008; Balandin et al., 2011). We detected no evidence that the Bn-AFF variants shown in Figure 4 were cytotoxic to HEK293T cells, either by MTT viability assays or by noting membrane blebbing or cell rounding/detachment (not shown). We speculated that lack of toxicity was due to a negative feedback loop in which Bn-AFF kept its expression at sublethal levels by degrading its own message, as we observed with mCherry mRNA (Figure 4). The previous studies cited above introduced the Bn protein directly into cells, perhaps at levels sufficient to cause death.

To evaluate the effect of Bn-AFF switching on eukaryotic cell viability, we used temperature to perturb the conformations of Bn-AFF expressed in S. cerevisiae. The haploid FKBP-deletion (fpr1Δ) strain was used to make the yeast resistant to rapamycin (Xu et al., 2010) and isolate potential cytotoxic effects to Bn-AFF activity. We first determined the melting temperatures (T_m) of the CP-fold analogs containing the I96*A or I96*G mutations, in the presence of saturating FK506. These represent the active folds of the respective Bn-AFF mutants. At 37°C, the I96*G CP-fold analog (T_m = 23.2°C ± 0.8°C) was well beyond the unfolded baseline whereas the I96*A mutant (T_m = 30.5°C ± 0.2°C) was at the edge of the unfolding transition (Supplementary Figure S7). Reducing temperature to 30°C caused both I96G* and I96A* to substantially populate the folded state. We therefore reasoned that growing yeast at those two temperatures could provide a thermodynamic window for assessing rapamycin-induced switching and cell toxicity.

Cells were transfected with Bn-AFF plasmids and allowed to grow for 3 days at 30°C or 37°C, with either rapamycin or DMSO vehicle added to the plates. Yeast expressing I96*A exhibited a slow growth phenotype at both temperatures (Figure 6), whereas I96*G cells grew at the same rate as the H102*A and empty vector controls. Rapamycin did not further increase toxicity in I96*A cells, suggesting that the CP-fold was already significantly populated at the two temperatures. The above results are in broad agreement with expectations from the thermal melts (Supplementary Figure S7).

I96*G cells grew normally at 37°C but slowly at 30°C, again consistent with its thermal stability (Supplementary Figure S6). Rapamycin did not reduce the growth rate of I96*G cells at 37°C, most likely because the CP-fold remained too unstable (at 14°C higher than its T_m) relative to the N-fold to retain sufficient activity for cell toxicity. When I96*G cells cultured at 30°C were shifted to 37°C for 1 d, however, we observed evidence for rapamycin-induced switching. After shifting to 37°C, the surviving I96*G cells were able to grow approximately as rapidly as the H102*A and empty vector controls in the absence of rapamycin but failed to proliferate in the presence of the drug. This result indicates that rapamycin had indeed enhanced the activity of I96*G Bn-AFF, either at 30°C or after the shift to 37°C despite rapamycin having no effect on I96*G cells that were cultured continuously at 37°C. The latter case may be explained by the I96*G protein staying in the active CP-fold long enough after the temperature shift to exert cytotoxic effects. The former scenario may have arisen from the protein degrading more RNA in cells cultured with rapamycin compared to those treated with vehicle at 30°C, thus preventing or delaying their growth when shifted to the permissive temperature.

For bacterial expression, C-terminal 6x-His tagged Bn-AFF constructs were cloned into the pETMT plasmid (Radley et al., 2003) downstream of the T7 promoter using NdeI/XhoI restriction sites. The pETMT vector contains barnase inhibitor barstar under a weak promoter to limit background activity of barnase. Proteins were expressed in E. coli BL21(DE3) and induced at OD₆₀₀ 0.5–0.6 with 0.5 mM isopropyl β-D-thiogalactopyranoside for ≥16 h at 18°C. Cell pellets were resuspended in 20 mM Tris (pH 7.5), 0.5 M NaCl, 10 mM imidazole, and 6 M guanidine hydrochloride and lysed by sonication. Lysates were loaded onto a nickel-nitrilotriacetic column, and the column was washed with at 10 column volumes of lysis buffer before elution with lysis buffer supplemented with 0.3 M imidazole. The proteins were then extensively dialyzed against ddH₂O and lyophilized.

For mammalian cell expression, Bn-AFF genes were cloned into the N1 vector (Clontech) downstream of the CMV promoter using AgeI/NotI sites. The proteins were tagged with a nuclear export signal to prevent barnase inhibition by DNA, and with yellow fluorescent protein to monitor transfection efficiency. For yeast expression, Bn-AFF genes were placed into the pBH787 vector (Gietz and Sugino, 1988) (gift from Brian K. Haarer) downstream of the TDH3 promoter using XbaI/SalI restriction sites. This vector contained the URA3 selection marker. All genes were fully sequenced and sequences are provided in Supplementary Note S1.

Simulations were performed in GNU Octave using models in Supplementary Figures S1–S3. Full details are provided in Supplementary Note S2.

HEK293T cells were cultured at 37°C in DMEM containing 10% FBS. Cells were split into a 12-well plate 1 day before transfection at a confluency of 60%–80%. Plasmids containing the Bn-AFF genes and the mCherry-PEST reporter were mixed in equal amounts, and 1 μg of DNA was transfected using calcium phosphate method. Briefly, CaCl₂ was added to the plasmid mix to 250 mM and equal amounts of 50 mM HEPES, 150 mM sodium phosphate, and 0.15 M NaCl (pH 7.0) were added dropwise while vortexing. The mixture was added directly to media after aging at room temperature for 20–30 min, and the media was replaced 15–20 h after transfection. After 36–48 h of expression, cells were washed with imaging media (DMEM, 50 mM HEPES pH 7.0, no phenol red), and 2.5 μM rapamycin or 0.025% DMSO vehicle was added to the imaging media.

After 12–15 h of rapamycin exposure, cells were mounted on imaging media and imaged with a Zeiss Axioimager Z1 upright microscope (10X/0.25 A-Plan Objective) equipped with a mercury lamp. mCherry fluorescence images were obtained using a 560 ± 40 nm excitation filter and a 630 ± 75 nm emission filter. YFP fluorescence was used to randomly identify regions with >50 cells and at least 3 fields were imaged. Cell images were quantified using Fiji (Schindelin et al., 2012). Background was subtracted using the built-in plugin using 50 pixels as the sliding ball radius. No further processing was applied, and the intensity of each cell was measured by manually drawing a region of interest around each cell and determining the average intensity using the Measure Particles plugin.

For urea denaturation experiments, lyophilized proteins were dissolved in 50 mM HEPES (pH 7.0), 100 mM NaCl, 6 M Urea, and 0.01% Tween-20. The solution was then rapidly diluted (∼200-fold, to a final protein concentration of 0.5–1.0 µM) into the same buffer and to the buffer that lacked urea. FK506 (20 µM) or the equivalent volume of DMSO was added to both solutions, and the solutions were mixed in various ratios using a Hamilton Microlab 540B dispenser. Urea concentration was measured by refractive index. After mixing, the samples were equilibrated at 4°C for 12–15 h and their tryptophan fluorescence spectra was recorded on a Horiba Fluoromax-4 fluorometer (280 nm excitation, 320–450 nm emission, 4°C). The maximum emission wavelength was determined by fitting the spectra to an exponentially modified gaussian function using Igor Pro (Wavemetrics Inc.). The maximum emission as a function of [urea] was then fit to the 2-state linear extrapolation equation (Santoro and Bolen, 1988) to obtain ΔG_unfold, C_m, and m-value using Kaleidagraph (Synergy Software).

Thermal denaturation experiments were performed by recording fluorescence spectra every 2°C from 10°C–50°C with ∼30 s equilibration between temperatures, using the same instrument settings described above. Solution conditions were 0.5 µM protein, 20 mM sodium phosphate (pH 7.0), 100 mM NaCl, and 20 µM FK506. T_m values were determined by plotting the derivative of the maximum emission wavelength as a function of temperature and fitting to a gaussian function using Kaleidagraph.

Lyophilized proteins were dissolved in 6 M GdnHCl and rapidly diluted to 20 nM into 50 mM HEPES (pH 7.0), 100 mM NaCl, 0.005% Tween-20. After equilibrating at room temperature for at least 20 m, 50 μM FK506 or an equivalent amount of DMSO vehicle was added and the solutions were kept overnight at room temperature. The solutions were then diluted to 0.1–1.0 nM and 20 nM of RNaseAlert substrate (IDT technologies) was added. Fluorescence changes were recorded using a SpectraMax i3x plate reader (Molecular Devices) with the following conditions: 96-well fully opaque plate, 150 μL per well, 1 mm read height, 485 nm excitation, 535 nm emission, and 30 s interval between measurements). Background signal was subtracted using a well containing only substrate in buffer. The first 7–10 m of the reactions were fit to a linear equation to determine initial rate.

The rapamycin-resistant haploid yeast strain (MATa ura3∆0 leu2∆0 his3∆1 fpr1∆::G418 ^R ) was transformed with the plasmid containing Bn-AFF using the lithium acetate/PEG method (Gietz and Schiestl, 2007) with sheared salmon sperm DNA as the carrier. Transformed cells were directly spotted onto plates containing 7 g/L yeast nitrogen base without amino acids, 0.25% casamino acids, 0.05 mg/mL tryptophan, 0.0025 mg/ml adenine, and either 1 μM rapamycin or 0.005% DMSO vehicle. The plates were supplemented with additional rapamycin or DMSO by spreading the solution once dry and drying them further for 3 h at 37°C. The cells were diluted by 10, 100, and 1000-fold, and 5 μL of each dilution was spotted onto the plates. The 100-fold dilutions were chosen for quantification by manual counting. Empty vector (plasmid containing no barnase gene) and no vector (only salmon sperm DNA) served as controls in all experiments.

All experiments described in this study were performed a minimum of three times (technical repeats). The mammalian cell and yeast experiments consisted of three biological repeats with new cells freshly split and transformed on different days. For activity assays, the I96*G mutant was purified two separate times. The turn-on ratios (determined from multiple technical repeats) were identical within error. Data are plotted as mean ± s.d. The statistical significance, where stated, was determined by a t-test with unequal variance using Kaleidagraph. All data and code generated in this paper is available on the Mendeley database (DOI 10.17632/j2rm8p4jr5.1). The plasmids will be made available upon request.