We have previously shown that PYR1 can be repurposed to create an agrochemical receptor that functions in planta to modulate stress tolerance¹¹; in that work, receptors were isolated from a library of variants created by single-site saturation mutagenesis. We reasoned that a structure-guided approach would facilitate the design of a larger collection of stable double mutants and ultimately enable the recognition of more ligands. There are 25 residues in PYR1’s ligand-binding pocket that make close contact to ABA and, therefore, 475 possible single mutants that can be combined to create 108,300 double mutants. We used our knowledge of the conserved receptor activation mechanism to remove six positions. To avoid combining unstable single mutants in the new library, we also used the Rosetta protein design software to predict the stability of the 361 single mutants at the remaining 19 positions. This analysis identified four positions that are particularly sensitive to mutation and were therefore restricted to a small number of amino acids in the final library; the remaining sites were allowed to mutate to any amino acid except cysteine or proline (Supporting File 1). Together, these steps condensed the library to a collection of 37,797 single and double nonsynonymous mutants (42,743 total mutants), which we constructed using site-directed mutagenesis. A total of 36,452 mutants were constructed using a pool of 301 single-mutant oligos in two sequential rounds of single-site nicking mutagenesis (NM)¹⁵; 6,291 of the mutants involved residues too close to one another (within eight residues) to be combined using single-mutant oligonucleotides and were instead constructed by NM using a pool of double-mutant oligonucleotides (Fig. 1b)¹². The two libraries were combined to yield the double-site mutagenesis (DSM) pocket library for subsequent screening. The DSM library was deep sequenced and determined to possess >99.8% of the desired double mutants (Supporting File 1).

With the improved library in hand, we set out to test its efficiency in a number of screens for biosensors. We first focused on developing cannabinoid sensors, in part to develop diagnostic reagents for synthetic cannabinoid mimics sold in products like ‘Spice,’ which have caused many hospitalizations¹⁶ and deaths^17,18. We screened for PYR1 mutants responsive to any one of a panel of 28 cannabinoids, screening for mutants responsive to 30 μM of each test chemical (Extended Data Fig. 1 shows chemical structures). Selections were accomplished in a Y2H strain in which expression of URA3 rescues uracil auxotrophy via PYR1 binding to HAB1. Before selection, mutations that produced ligand-independent URA3 activation were removed by a negative selection in the presence of 5-fluoroorotic acid. Our initial positive selections identified double mutants responsive to JWH-015 (Fig. 1c) and five other naphthoylindoles, as well as cannabicyclohexanol (CP 47,497), a different chemical scaffold and one of the active ingredients in ‘Spice’; this demonstrates that our library can yield sensors in a single screening step. Additional sensors were acquired by iteratively screening diversified cannabinoid-biased sublibraries that were created by shuffling hit receptors against both the parental DSM and previous single-site mutant (SSM) libraries. Ultimately, these efforts identified 12 PYR1-derived cannabinoid receptors that recognize 14 compounds, including sensors for CBDA, Δ⁹-THC and 4F-MDMB-BUTINACA (4F-MDMB), from a total of 28 cannabinoids screened (Fig. 2a,b and Extended Data Fig. 1). Overall, mutations in nine out of the 19 residue positions targeted in the parental library were obtained in the cannabinoid receptors (K59, V81, V83, L87, A89, Y120, F159, A160 and V164), along with two additional sites (H115 and M158) present in the SSM library used for DNA shuffling and affinity maturation. We note that in two cases, our selections converged on the identical sequences; receptors responsive to JWH-072 and JWH-015, closely related naphthoylindoles that differ by only a single methyl substituent, yielded nearly identical evolved sequences. Similarly, the sensors obtained for the closely related compounds 4F-MDMB and AB-PINACA contained identical mutations (Supporting File 2). In addition, wild-type PYR1 did not respond to any of the target ligands screened, nor did our evolved high-affinity sensors respond to ABA (Supplementary Fig. 1); collectively, these data show that PYR1’s ligand-binding pocket can be reprogrammed to recognize target molecules.

Most synthetic cannabinoids share a central indole or indazole scaffold. We anticipated that our evolved cannabinoid receptors might show cross-reactivity. To explore this, we tested several high-affinity sensors for cross-reactivity to cognate ligands (Supplementary Figs. 1 and 2). Although these tests indicate some cross-reactivity, particularly for PYR1^4F and to a lesser degree PYR1^JHW-072, in all cases, on-target sensitivity was at least tenfold higher than the off-target sensitivity. Thus, PYR1-derived sensors can provide sensitive and selective ligand detection, although this will vary by receptor and chemical.

To understand the underlying molecular basis for cannabinoid recognition by our evolved receptors, we sought to obtain the structure of a receptor–cannabinoid–HAB1 complex and targeted two high-affinity sensors, PYR1^4F and PYR1^WIN. In our experience, PYL2 (a close relative of PYR1) forms crystals more readily than PYR1. PYL2 shares 88% pairwise sequence identity with PYR1 for the 25 ABA-proximal positions, and structures of the two proteins are globally alignable to 0.55 Å root mean squared deviation. We, therefore transposed mutations conferring ligand-selective responsiveness to PYL2, creating PYL2^4F and PYL2^WIN, which both retain nanomolar ligand responsiveness (Supplementary Fig. 2). In addition, we created a stabilized, catalytically inactive derivative of HAB1 less prone to oxidative inactivation by employing computational redesign, yielding ΔN-HAB1^T+ (derived from a HAB1 truncation that contains its PYR1-binding catalytic domain). This variant harbors 15 mutations, displays a ~7 °C increase in apparent T_m, and retains high-affinity ligand-dependent binding to PYR1, as measured by a yeast surface display assay that uses PYR1^M (H60P, N90S), a monomeric double mutant optimized for yeast surface display¹⁹ (Supplementary Fig. 3 and Supporting File 1). Using these engineered proteins in matrix screens, we obtained diffraction quality crystals for a ternary PYL2^WIN/WIN 55,212-2/ΔN-HAB1^T+ complex, whose structure was solved by molecular replacement (1.9 Å resolution; Supplementary Table 1). Diffraction quality PYL2^4F crystals were not obtained. Several rounds of structural refinement were carried out before modeling WIN 55,212-2 into the ligand-binding pocket’s unbiased electron density (Supplementary Fig. 4). A real-space correlation coefficient of 0.967 calculated between the unbiased electron density and (+)-WIN 55,212-2 indicates agreement between the model and observed electron density. We note that the evolved receptor recognizes the biologically active (+)-WIN 55,212-2 stereoisomer, although selections were conducted using a racemate (Supplementary Fig. 2).

A central feature of ABA recognition by native sensors is the formation of a closed-receptor conformer that is stabilized by a hydrogen-bond network between a structurally conserved water, ABA’s ring ketone, main-chain amides in the gate and latch loops and HAB1’s W385 Trp-‘lock’ residue^20–22. In our PYL2^WIN structure, WIN 55,212-2’s naphthoylindole ketone functions analogously to ABA’s ketone and is coordinated through water-mediated hydrogen bonds to backbone P92 in the gate, R120 in the latch and HAB1’s W385 lock residue (Fig. 2c).

Analysis of the PLY2-WIN structure also revealed that ligand binding is stabilized by an extensive network of hydrophobic contacts and a water-mediated contact to WIN’s morpholine oxygen (Supplementary Fig. 4). In comparison to PYR1, PYR1^WIN harbors three mutations (K59Q, F159A and A160I), and our structure illuminates their roles in allowing favorable binding. The most conspicuous effect is a relief of steric clash that would occur between F159 and WIN’s naphthalene ring in a wild-type receptor (Fig. 2d). The neighboring A160I mutation is positioned to enhance receptor-ligand surface complementarity by enabling the naphthoylindole to better pack in this position relative to the wild-type receptor. The K59Q mutation appears to reduce the electrostatic penalty of burying WIN’s positively charged morpholine ring but also organizes a water-mediated hydrogen-bond network at the base of the pocket (Fig. 2d and Supplementary Fig. 4). Thus, WIN’s binding to PYL2^WIN involves a combination of polar and hydrophobic contacts, which contrasts with its binding mode in the human cannabinoid receptor CB2, where binding involves exclusively hydrophobic contacts and a more extended ligand conformer²³ (Fig. 2e). In addition, the structure illustrates the success of our HAB1 redesign, showing that ΔN-HAB1^T+’s main chain is nearly superimposable with that of wild-type (0.85 Å root mean squared deviation) and that the key rotamers for residues involved in receptor interactions are maintained (Supplementary Fig. 3). Collectively, these data provide a mechanistic basis for the sensitive and selective cannabinoid recognition by our evolved receptor, illuminate the mutability of PYR1’s ligand-binding pocket and validate our computational HAB1 redesign.

In principle, the PYR1-HAB1 CID mechanism enables rapid construction of multiple sense-response outputs, as has been demonstrated with other designed CID sensors¹. To explore the portability of the designed PYR1-HAB1 system, we selected two high-affinity receptors for evaluation of in vitro HAB1 inhibition, yeast transcriptional activation circuits and in vivo protein-fragment complementation with split luciferase. PYR1^WIN and PYR1^4F displayed nanomolar half-maximum effective concentration (EC₅₀) values using an inhibition assay that detects receptor activation by changes in HAB1 phosphatase activity using a fluorogenic substrate (EC₅₀ PYR1^WIN: 72 nM; PYR1^4F: 102 nM; Fig. 3 and Supplementary Fig. 5). Fusion of the transcriptional activator VP64 to ΔN-HAB1 and a zinc-finger DNA-binding domain to PYR1^WIN enabled inducible GFP expression from a synthetic cassette integrated into the genome of Saccharomyces cerevisiae with an EC₅₀ of 28 nM. The same transcriptional circuit built with PYR1^4F responded to 4F-MDMB with an EC₅₀ of 23 nM. Similar success was achieved with the NanoLuc split luciferase systems²⁴; NLuc^N-PYR1^WIN/NLuc^C-ΔN-HAB1 responded with an EC₅₀ of 56 nM, and NLuc^N-PYR1^4F/NLuc^C-ΔN-HAB1 responded with an EC₅₀ of 25 nM. Taken together, these results show that the PYR1/ΔN-HAB1 CID mechanism enables portability to both in vitro and in vivo formats and can be deployed in multiple outputs. Moreover, the luminescence and fluorescence reporting modes may be advantageous for ligand detection using low-cost instrumentation and/or in remote settings using engineered living cells.

Synthetic cannabinoids are frequently modified to evade detection by routine drug testing. For example, 4F-MDMB is a relatively new indazole that first appeared in 2018 and rapidly became one of the most prevalent synthetic cannabinoids used in the United States^25–27. Although mass spectrometry methods can sensitively detect this and most synthetic cannabinoids, lower-cost and easier-to-use immunoassays (e.g., ELISAs) dominate routine drug testing. Given this, we sought to convert our CID system into an ELISA-like system for microplate format measurements. To do so, we developed a hybrid sandwich-assay in which the PYR1 sensor is surface immobilized and then coincubated with biotinylated ΔN-HAB1^T+ and the ligand of interest. Detection of ternary complexes can then be quantified using streptavidin-linked horseradish peroxidase (HRP), as is commonly done with ELISAs. We developed an ABA-detection system using PYR1^M (LOD 2 nM, 0.7 ng ml⁻¹) and then adopted this optimized format for PYR1^4F-M and PYR1^WIN-M (Supplementary Fig. 5 and Supporting File 1). To evaluate the assay’s performance in forensically relevant samples, we tested it with spiked urine, blood and saliva samples and observed reliable detection of picomolar concentrations (LOD: 515 pM in urine, 1,105 pM in blood serum and 305 pM in saliva; Fig. 3d), which compares favorably with existing immunoassay kits that typically report low-nanomolar LODs²⁸. In addition, minimal cross-reactivity between the PYR1^WIN-M and PYR1^4F-M sensors and a panel of 14 cannabinoids was observed using this detection format (Fig. 3e and Supplementary Fig. 5). Combined with our other results, these data further demonstrate that PYR1’s native CID mechanism can be harnessed to develop multiple sense-response outputs and demonstrate that our optimized ELISA-like test enables selective and sensitive detection of target ligands using evolved PYR1-based sensors.

Given the success of the PYR1 scaffold as a platform for cannabinoid sensing, we sought to explore the possibility of rapidly generating sensors for a second class of compounds. To do so, we screened single and multisite mutational libraries against various different organophosphates, an important class of toxic, nonselective acetylcholinesterase inhibitors that were among the first insecticides broadly used in the 20th century. Due to their effects on nontarget organisms, most organophosphates have been banned in the United States, and they present an ongoing environmental monitoring challenge. We screened PYR1 libraries against a panel of 10 organophosphates (diazinon, pirimiphos, dimethoate, chlorfenvinphos, parathion, disulfoton, azinphos, bromophos, malathion and monocrotophos), none of which activated wild-type PYR1 (Supplementary Fig. 1; see Supporting File 1 for library details). These screens yielded receptors for diazinon, pirimiphos, dimethoate and chlorfenvinphos at concentrations between 5 and 100 μM (Supporting File 3). In addition, we screened the new DSM library against the same set of 10 compounds and obtained receptors for seven of these (Supplementary Fig. 6). To improve receptor affinity, we used recombination-based mutagenesis for four organophosphate receptors (diazinon, pirimiphos, chlorfenvinphos and dimethoate) by shuffling hits against parent libraries. This approach was repeated four times, reducing the ligand concentration at each step to ultimately yield improved sensors for these two compounds (Fig. 4a,b). The diazinon-responsive variant PYR1^DIAZI is a heptuple mutant (E8G, V81Y, L87M, F108Y, M158V, F159G and A160V), and the pirimiphos-responsive variant PYR1^PIRI is an octuple mutant (K59R, S92M, N119S, S122Q, E130G, F159T, A160T and V174A). These optimized sensors were also immediately portable to the split luciferase system with low-nanomolar sensitivity (Fig. 4c). Together, these data demonstrate that the PYR1 ligand-binding pocket can mutate to accommodate organophosphate ligands and provide a general system for developing organophosphate sensors.

To address the selectivity profiles of the evolved organophosphate receptors, we characterized their cross-reactivity to target ligands, given the close structural similarity of diazinon and pirimiphos. Both HAB1 inhibition and Y2H assays showed that the evolved receptors are highly selective to their on-target ligands (Fig. 4d). PYR1^PIRI was activated by low-nanomolar concentrations of pirimiphos but required high-nanomolar to low-micromolar diazinon concentrations for activation above background levels. Similar results were observed with the PYR1^DIAZI receptor, and neither receptor was activated by ABA. In a strict test of specificity, we used our yeast transcription circuit to characterize the off-target responses of these engineered receptors to a panel of six chemically similar organophosphates. The PYR1^DIAZI EC₅₀ to diazinon was greater than tenfold lower than all other molecules profiled (Fig. 4e). For example, PYR1^DIAZI responded with an EC₅₀ of 1.1 μM to azinphos but with an EC₅₀ of 43 nM to diazinon. Other off-target ligands responded with higher EC₅₀ values and with lower activation levels (Supporting File 1). Similarly, the PYR1^DIAZI receptor showed good discrimination between its on-target ligand diazinon and other organophosphates when tested at 20 μM in the ELISA-format mode (Supplementary Fig. 7). Lastly, we examined how receptor selectivity changed over the evolutionary trajectory of the PYR1^DIAZI sensor. Off-target ligand responses remained weak throughout the evolutionary process and decreased as affinity increased (Supplementary Fig. 7). Collectively, these data show that improvements in affinity were not obtained at the cost of increased promiscuity and that high-affinity and high-selectivity can be evolved using the PYR1 scaffold.

Using our designed PYR1-HAB1 system, we isolated sensors for 21 of the 38 compounds screened, which included a diverse set of ligands that fall into distinct chemotypes (Extended Data Fig. 2). Although structurally diverse, many of the ligands screened contain a carbonyl functional group that, as observed with WIN 55,212-2, can engage the Trp-lock to stabilize activated receptors (Fig. 2 and Extended Data Fig. 3). Prior work has shown that other H-bond acceptors (e.g., nitriles and others) can function in place of a carbonyl to activate both wild-type and engineered PYR1 receptors^11,29–31 (Extended Data Fig. 3). In addition, our evolved PYR1^CP sensor recognizes a ligand lacking a C=O (Fig. 2). Thus, the chemical scope of ligands compatible with our system should be quite broad. However, even if there is a bias towards carbonyl-containing ligands, approximately one-third of natural products and drugs contain a carbonyl³² and provide a large set of ligands. When coupled to the high hit rates obtained with our DSM library, it should be possible to evolve molecular switches controlled by a large number of drugs, natural products or metabolites. Although many technologies for chemically regulated dimerization have been developed, our system is unique, because it empowers the design systems controlled by user-specified ligands, which is particularly useful when specific properties (such as low cost or low toxicity) are required in downstream applications. Thus, the PYR1/HAB1 system provides an easily reprogrammable chemical-induced dimerization module that will enable new applications in biotechnology, synthetic biology and medicine.