Rational evolution for altering the ligand preference of estrogen receptor alpha

Abstract

Estrogen receptor α is commonly used in synthetic biology to control the activity of genome editing tools. The activating ligands, estrogens, however, interfere with various cellular processes, thereby limiting the applicability of this receptor. Altering its ligand preference to chemicals of choice solves this hurdle but requires adaptation of unspecified ligand‐interacting residues. Here, we provide a solution by combining rational protein design with multi‐site‐directed mutagenesis and directed evolution of stably integrated variants in Saccharomyces cerevisiae. This method yielded an estrogen receptor variant, named TERRA, that lost its estrogen responsiveness and became activated by tamoxifen, an anti‐estrogenic drug used for breast cancer treatment. This tamoxifen preference of TERRA was maintained in mammalian cells and mice, even when fused to Cre recombinase, expanding the mammalian synthetic biology toolbox. Not only is our platform transferable to engineer ligand preference of any steroid receptor, it can also profile drug‐resistance landscapes for steroid receptor‐targeted therapies.

1. INTRODUCTION

Nuclear receptors (NRs) form a large protein family that modulate the development, homeostasis and metabolism of metazoans. The estrogen receptor alpha (ERα) is a member of the steroid receptor sub‐family and it is well‐known for its role in the female reproductive system, bone, and other tissues (Chen et al., 2022). By responding to estrogens, the ERα is involved in the complex regulation of mammary gland formation, the menstrual cycle and pregnancy. It is also one of the main therapeutic targets in breast cancer where its activity is blocked by the use of ligands, such as tamoxifen. Upon estrogen binding, the ERα undergoes a conformational change that enables the receptor to recognize estrogen response elements (ERE) in the regulatory regions of the ERα‐regulated network of genes (Mangelsdorf et al., 1995).

Because of their diversity, specificity, and modular design, NRs have been exploited as regulators in multiple tools for synthetic biology, including biosensors for environmental toxins (Hettwer et al., 2018; Ragavan et al., ²⁰¹³), controllable gene expression switches (Kuo et al., 2012; Liang et al., ²⁰¹³) or functional regulators (Hochrein et al., 2018; Liu et al., ²⁰¹⁶; Matsuda & Cepko, ²⁰⁰⁷). Major disadvantages of these applications of NRs are their cognate ligands that interfere with the native NR functions or are toxic to the host organism (Adeel et al., 2017; Ciriaco et al., ²⁰¹³; Tang & Newton, ²⁰⁰⁴). The development of novel synthetic NRs (XRs) that respond to specific inducers (X) that do not interfere with the host cell's normal physiology is therefore key to their broader application in synthetic biology.

Previously, random mutagenesis of the ERα resulted in a variant (ERT2) responsive to tamoxifen, a specific ERα modulator used in the treatment of hormone receptor‐positive breast cancer (Feil et al., 1997; Jordan, ¹⁹⁹⁷; Metzger et al., ¹⁹⁹⁵). The ligand‐binding domain (LBD) of this ERT2 variant has become the universal regulatory domain of genomic tools in animal models, including Cre recombinases and Cas9 endonucleases (Donocoff et al., 2020). However, in successive saturation mutagenesis, structural as well as dose optimization strategies (Donocoff et al., 2020; Matsuda & Cepko, ²⁰⁰⁷), further engineered ERT2 variants remain responsive to circulating levels of estrogens (Kristianto et al., 2017). Moreover, the high tamoxifen levels that are required for activation can interfere with host endocrinology (Kristianto et al., 2017; Xie et al., ²⁰²¹).

Engineering specific ligand recognition and transcriptional regulation is challenging. The key residues that need to be altered to accommodate de novo ligands are unknown. Moreover, these residues are distributed throughout the LBD. Since multiple residues need to be changed simultaneously, a near‐endless list of possible combinations of mutations needs to be generated (Glasgow et al., 2019; Katzenellenbogen et al., ²⁰¹⁸). Importantly, while ligand binding needs to be changed, the interactions with coregulators that are required for a proper functionality of the XR need to be preserved (Lu et al., 2009). Lastly, for applicability in metazoans it is important that the new XR no longer responds to the original ligand to prevent activation by endogenous hormones (Lu et al., 2009).

Protein engineering strategies for altering the ligand preference of ERα have important bottlenecks to overcome. First, the structural information highlighting key residues is limited and computational models are unable to predict the correct amino acid sequence to fold into a structure with the desired properties (Glasgow et al., 2019; Marshall et al., ²⁰⁰³; Pan & Kortemme, ²⁰²¹). Second, although numerous mutagenesis strategies are available to modify single or multiple target residues which use degenerate primers or randomly introduce mutations across the gene of interest, high‐throughput metazoan profiling platforms are lacking, thus limiting the size of mutant libraries that can be studied (Orencia et al., 2001; Packer & Liu, ²⁰¹⁵).

In order to alter the ligand preference of ERα toward tamoxifen, our experimental approach combines rational protein design to identify the residues of interest, with a mutagenesis protocol that targets multiple non‐contiguous amino acids at the same time. This results in a library of receptor variants that is integrated in yeast using CRISPR‐Cas9, thereby ensuring that every transformant expresses only a single receptor variant. This unique single variant integration scheme overcomes the limitations associated with plasmid‐based libraries, including plasmid copy number variations and multiple variants being expressed within a single transformant (Ilhan et al., 2018; Jahn et al., ²⁰¹⁶).

To identify the receptor variants with desired phenotypes, multiple selectable traits that convey growth and fluorescence are put under control of the receptor and introduced in S. cerevisiae that naturally lacks NRs. This creates a platform with the possibility to profile more than 10 ¹⁰ stably‐integrated receptor variants in parallel via in vivo compartmentalization per μg DNA (Wang et al., 2021). The practical bottleneck thus becomes the transformation efficiency of the host (Packer & Liu, 2015; Xiao et al., ²⁰¹⁵).

With this unique rational evolution approach, we have developed a human Tamoxifen Evolved estRogen Receptor α variant (TERRA), that is fully inducible by tamoxifen but not by estrogens. Moreover, we show that this variant can replace ERT2 as the regulatory domain of genome editing tools, in both cell culture and animal models. Taken together, our rational evolution strategy resulted in the TERRA receptor that is an ERα variant with optimal ligand preference for tamoxifen. This TERRA receptor can be further converted into various synthetic biology tools that are applicable across eukaryotes.

2. RESULTS

2.1. Establishing a Saccharomyces cerevisiae platform for evolving estrogen receptor alpha

To identify ERα variants with altered ligand preferences in S. cerevisiae, we have developed selectable traits that facilitate both growth and fluorescence in response to activated ERα. By replacing the native URA3 and HIS3 promoters with universal steroid receptor control elements (SRi) (Eerlings et al., 2021) and synthetic core promoters (Redden & Alper, 2015), yeast growth became dependent on ERα activity (Figure S1). The combination of two growth selective pressures promotes evolutionary stability when characterizing ERα variants (Sleight et al., 2010). Moreover, an ERα‐inducible yemCherry reporter was developed that permits fluorescence‐based profiling of ERα activity (Voth et al., 2001). Furthermore, a constitutively active yeCitrine expression cassette was designed that allows for normalization of ERα activity per cell. With CRISPR‐Cas9, these ERα‐dependent URA3, HIS3, and yemCherry reporter cassettes as well as the constitutively active yeCitrine expression cassette were introduced in S. cerevisiae in respectively the URA3, HIS3, HO, and CAN1 loci, establishing the pSRI1‐HIS3 pSRI2‐URA3 pSRI‐yemCherry pPGI1‐y eCitrine strain that served as a screening platform for ERα variants, the HURRY platform (Figure 1a).

FIGURE 1.

Saccharomyces cerevisiae platform for evolving nuclear receptors toward designer chemicals. (a). The pSRI1‐HIS3 pSRI2‐URA3 pSRI‐yemCheRRy pPGI1‐YeCitrine (HURRY) strain has the dual survival and fluorescent reporter cassettes under ERα control. Reporter cassettes are composed of the SRi enhancer region and synthetic core promoters 4, 9, or minimal CYC1 promoter resulting in pSRI1, pSRI2, or pSRI. (b) Evaluation of the estrogen responsive growth of the S. cerevisiae HURRY strain versus its parental strains that do not contain one or more reporter cassettes. These strains all express the S. cerevisiae codon optimized ERα (yERα) under control of the constitutive TEF1 promoter from the ARS208 locus, with the exception of the pTEF1‐yERα strain that contains this cassette in the CAN1 locus. The vehicle condition serves as negative control. The curves are fitted using non‐linear regression (Prism version 9.3.1). The error bars indicate the standard error of the mean of three biological replicates. (c) Characterization of the estradiol responses of yemCherry fluorescence of the S. cerevisiae HURRY strain versus its parental clones described in the panel. yemCherry fluorescence is corrected by the yeCitrine fluorescence and expressed relative to the vehicle condition. The curves are obtained as described for panel a.

To validate ERα‐dependent growth and fluorescence of the HURRY platform, we integrated the S. cerevisiae codon‐optimized wild‐type ERα (yERα) near the ARS208a locus. This genomic location promotes recombinant protein expression as well as genomic integration efficiency (Reider Apel et al., 2017). The estradiol response profile for wild‐type yERα in the HURRY platform was similar to that of the human ERα (hERα) in human cells with a half maximal effective concentration (EC50) of approximately 0.1–1 nM (Figure 1b,c and Table 1). Taken together, the yERα simultaneously induced expression of two genes allowing growth under selective conditions as well as expression of the yemCherry reporter gene in the S. cerevisiae HURRY platform in an estradiol‐dependent mode.

TABLE 1.

Sequence alignment and characterization of the estradiol‐ and tamoxifen responsiveness of the novel estrogen receptor α (ERα) variants discovered in this study in mammalian cells.

	Amino acid					Estradiol		Tamoxifen
	391	400	539	543	544	EC50 (nM)	Confidence interval (nM)	EC50 (nM)	Confidence interval (nM)
Wild‐type ERα	L	G	L	M	L	0.5	0.02 to 6.9	Non‐detectable	Non‐detectable
ERT2	L	V	L	A	A	34	6.6 to 165	2615	1072 to 11,790
PreTERRA	L	G	F	E	L	2.1	0.09 to 400	56	16 to 167
TERRA	T	G	F	E	L	83.3	26 to 264	122	39 to 382

For the development of novel ERα variants (XRs) that respond to a ligand of choice, we established a strategy to selectively mutate multiple non‐adjacent amino acid positions in ERα (see below). After CRISPR‐based integration of the receptor expression cassette, the ligand‐induced activity of the resulting ERα variants was validated with our high‐throughput profiling platform using survival and fluorescence as dual read‐out.

2.2. Evolving tamoxifen‐selective receptors through rational evolution

In this study, we decided to engineer ERα into a highly specific tamoxifen‐responsive receptor that outperforms the previously developed ERT2 variant (PDB ID:1QKU; Gangloff et al., ²⁰⁰¹). Our experimental pipeline for the adaptation of ERα to tamoxifen was divided in three sections (Figure 2a): (1) in silico identification of amino acid (AA) locations that are involved in the binding of tamoxifen, (2) simultaneous diversification of these AA positions in vitro, creating a ERα variant library via Darwin Assembly (Cozens & Pinheiro, 2018), followed by stable library integration near the ARS208a locus of the HURRY strain using CRISPR‐Cas9, and (3) characterization of these transformants using receptor‐dependent survival in growth‐selective media together with inducible fluorescence measurements.

FIGURE 2.

Methodology for engineering receptor proteins through rational evolution. (a) Schematic representation of the three sections of the rational evolution pipeline: (1) in silico identification of the amino acid positions in the human ERα‐LBD to be mutated, (2) simultaneous diversification of defined amino acid locations, and (3) characterization of NR variants for designer phenotypes. Rational evolution starts from double stranded plasmid DNA encoding the gene expression cassette of interest, including promoter (green arrow), CDS (orange region), terminator (red T), and 120 bp homology arms to the target site for genomic integration (red), near ARS208a. The plasmid DNA is singularized by a nicking endonuclease (at the purple dot) and exonuclease III. Next, the single stranded DNA plasmid is bound by a 5′‐biotinylated‐ (indicated by a red B) and a 3′‐3′dT boundary oligonucleotide and multiple 5′‐phosphorylated mutagenic primers with degenerate codons (red X) at the in silico determined putative sites. Non‐complementary overhangs extending the homology arms are indicated in red and tilted. Next, primers are extended and ligated in an isothermal assembly reaction. The assembled 5′‐biotinylated DNA strands are isolated by paramagnetic streptavidin‐coated beads (beige beads) and purified by alkali washing prior to PCR using outnested priming sites. This library of dsDNA mutant gene expression cassettes is stably integrated near the ARS208a locus of the S. cerevisiae HURRY strain by addition of a Cas9 expression vector that also transcribes the accommodating sgRNA during transformation. Every transformant stably expresses one unique receptor variant. To isolate the active XR fraction, the transformant pool is plated on medium lacking uracil and histidine but containing the chemical of choice. Transformants expressing a receptor variant that is activated by the chemical of choice are able to express URA3 and HIS3 and survive. If more than 1000 unique colonies are obtained, further selection by fluorescence‐based cell sorting using yemCherry as a proxy for receptor activity can be applied to isolate transformants expressing the highest yemCherry fluorescence signal in presence of the chemical of interest. From the sorted pool or immediately after growth selection, a maximum of 1000 colonies are subjected to clonal screening. Screening by overnight incubation with the native ligand or the chemical of interest, followed by detection of yemCherry and yeCitrine fluorescence intensities, can identify XRs that lost their responsiveness to the native ligand and improved the response to the chemical of choice (indicated in teal). After genotyping, the desired isolate can serve as a new starting point for subsequent rational evolution rounds. (b) Identification of hot spot amino acid positions for computed binding by the chemical of interest, tamoxifen in this study. Positions are chosen based on the individual conservation (right hand side y‐axis) and the energetic contribution to the ligand interface (left hand side y‐axis) of the human estrogen receptor α with estradiol or tamoxifen. Residues with high conservation scores and minimal contribution to tamoxifen binding are selected for the first rational evolution round to promote tamoxifen responsiveness (orange). Positions with conservation scores between 5 and 8 that show high estradiol binding energies are evolved to impair estradiol response (light blue).

To subsequently reduce estrogen‐induced activity of these transformants, another diversification and characterization round was performed aimed at estradiol‐interacting residues. This second adaptation is essential as it allows the use of the mutated LBD as regulatory domain of chimeric genome editing tools in organisms that naturally synthesize estrogens, including metazoans. Mutated LBD variants that remain susceptible to physiological estrogen levels will result in undesired off‐target effects. The following sections elaborate on each step of these optimization rounds.

2.2.1. In silico identification of target residues

Structural information on the binding of estradiol and tamoxifen in the ERα‐LBD (PDB ID:1QKU; Gangloff et al., ²⁰⁰¹) and PDB ID:3ERT; Shiau et al., ¹⁹⁹⁸) was used to select specific target positions for improved accommodation of tamoxifen in the ERα‐LBD. Based on the conservation scores of the amino acid within the estrogen receptor lineage and their respective binding energy with estradiol and tamoxifen (Figure 2b; Figure S2), positions that are prone to promote tamoxifen binding when varied, were chosen. The conservation score was included in this selection step since the biological importance of a residue often correlates with its level of evolutionary conservation within the protein family (Landau et al., 2005). Residues G521, M522, H524, and L525 in helix 11 and L539, L540, L541, M543, and L544 in helix 12 were initially considered for rational evolution because tamoxifen sterically influences their positioning, thereby driving the change between agonistic and antagonistic conformation (Shiau et al., 1998). To keep the structure of helix 11 intact, H524 with a conservation score of 9 was excluded. For helix 12, we expected a tamoxifen‐specific change in conformation given the bulkier volume of this compound compared to the planar estrogens. Therefore, we chose residues L539, L540, M543, and L544 because they have a high conservation score (in this case, 9) and less negative binding energies with estrogen compared to tamoxifen meaning that they are more involved in tamoxifen binding. From helix 11, we added position L525 because it contributes to both estradiol and tamoxifen binding (binding energy of −1.89 kcal/mol for estradiol and −2.12 kcal/mol for tamoxifen) so varying it might result in a residue with clear preference for tamoxifen.

2.2.2. Simultaneous diversification of selected ERα residues and library integration

Subsequently, residues L525, L539, L540, M543, and L544 were simultaneously changed into all possible amino acid combinations by introducing the NNK‐degeneracy (Owen et al., 2016) in the open reading frame of ERα (Figure 3a; Figure S3). In agreement with the threefold oversampling rule (Bosley & Ostermeier, 2005; Reetz et al., ²⁰⁰⁸; Wu et al., ²⁰¹⁹), we ensured a minimal 95% coverage of the 3.36 × 10 ⁷ variant library by performing a fourteenfold oversampling. To establish this fourteenfold oversampling, 60 variant pools were independently introduced via CRISPR library integration near the ARS208a locus of the HURRY strain, leading to approximately 5.03 × 10 ⁸ independent yeast transformants, each expressing only one XR variant.

FIGURE 3.

Rational evolution of the estrogen receptor α (ERα) toward tamoxifen as an agonist. (a) Indication of hot spot residues for simultaneous randomization to facilitate tamoxifen binding (PDB ID:1QKU for representation) (b) Screening of the 913 rationally evolved NR variants in the HURRY strain for their activities in presence of 10 nM estradiol and 10 μM tamoxifen through flow cytometry. Fluorescence intensities are normalized to the 10 nM estradiol‐stimulated wild‐type yERα, which is set to 100%. Each data point represents the average reading of 10,000 single cells of that unique strain. (c) Evaluation of the activity of rationally evolved top isolates in mammalian HEK 293 T cells in the absence of a ligand (vehicle shown in black), in the presence of estradiol (pink), tamoxifen (blue), or combined ligands (green). For the combined ligands, 10 nM estradiol is selected as it represents the highest observed physiological concentration in metazoans. Physiological estrogen concentration ranges are indicated by gray shading. Luciferase values in presence of vehicle are set at 1. The error bars indicate the standard error of the mean from three biological replicates. The curves are fitted using non‐linear regression (Prism version 9.1.3). (d) Structural determination of putative amino acid locations to impair estradiol binding starting from the most tamoxifen‐selective identified variant of the first rational evolution round (preTERRA) (PDB ID:1QKU for representation) (e) Profiling of the 649 PreTERRA‐derived NR variants, accommodating mutations aimed at impairing estradiol binding, in the HURRY strain. Fluorescence intensities are normalized to the 100 nM estradiol‐stimulated wild‐type yERα, which is set to 100%.

2.2.3. In vivo characterization of ERα variants using dual read‐out

To isolate the fraction of transformants that were responsive to tamoxifen through expression of a certain XR variant, the transformant pool was plated on medium that lacks uracil and histidine but contained 10 μM tamoxifen. This yielded a total of 909 unique colonies that were individually tested for their yemCherry levels in presence of respective estradiol (10 nM, near saturation for yERα WT) and tamoxifen (10 μM, minimum activating concentration for yERα WT) via flow cytometry (Figure 3b). For every transformant, the yemCherry fluorescence intensities in these conditions were corrected for their respective yeCitrine signals followed by normalization against the wild‐type yERα intensity in presence of estradiol (Table S1). Subsequently, their tamoxifen selectivity was determined by taking the ratio of the normalized yemCherry fluorescence intensity in presence of tamoxifen over the normalized yemCherry fluorescent signal in presence of estradiol (Table S1). Based on those results, the coding DNA sequences (CDS) of the 20 clones that demonstrated the highest response in presence of 10 μM tamoxifen (1), the highest response in presence of 10 nM estradiol (2), and the highest tamoxifen selectivity (3) respectively, were determined by Sanger sequencing (Table S2–S4). Surprisingly, many highly active clones contain InDels or nonsense mutations that resulted in truncated receptor variants. These truncated receptors lack helix 12 or part of helix 11, that sterically obstructed the bulky sidechain of tamoxifen (Chakraborty & Biswas, 2014), thus explaining their tamoxifen preference.

To validate the improved tamoxifen sensitivity of the nonsynonymous receptor variants in metazoans, the top nine unique tamoxifen‐selective variants were codon‐optimized for evaluation in mammalian HEK 293T cells (Figure S4). With the exception of the H524Y mutant, all receptor variants demonstrated tamoxifen responsive reporter gene induction when co‐transfected in HEK 293T cells, confirming the agonistic character of tamoxifen for these variants.

However, the ERα variant activities in mammalian cells differed from the activities observed in yeast. The Y537N_L539W_L540I variant demonstrated the highest tamoxifen‐response in yeast, while the L525P_L539W_L540F_M543R variant reached the highest tamoxifen‐induced activity in HEK 293 T cells. The discrepancy in tamoxifen responsiveness between S. cerevisiae and mammalian cells could result from the absence or divergence of the NR coregulators in yeast. In mammalian cells, the L539F_M543E mutant (preTERRA) showed the highest tamoxifen sensitivity, with almost maximal activity at 100 nM tamoxifen (Figure S4). Compared to the current tamoxifen inducible system (ERT2), preTERRA showed a 46‐fold higher tamoxifen sensitivity, a two‐fold increase in maximal activity while lacking basal transactivation (Figure 3c; Figure S4). However, like ERT2, the preTERRA mutant was still responsive to physiological estradiol levels (1–10 nM), limiting its possible use as a metazoan synthetic biological tool. The next optimization round was therefore aimed at preventing estradiol responsiveness.

2.2.4. Preventing estrogen activation of the preTERRA variant

To impair estradiol binding to preTERRA, we focused on residues in helix 5 as these have higher negative binding energies for estradiol than tamoxifen (Figure 2b). Those residues with the strongest interactions with estradiol and close proximity toward tamoxifen were chosen as targets (Figure 3d). This included L384, L387, M388, and L391. The fifth residue for this second rational evolution round, G400, was chosen because of its involvement in improving agonistic response to tamoxifen in prior studies (Katzenellenbogen et al., 2018).

NNK‐degeneracy was used to simultaneously randomize the five denoted positions to all possible AA combinations, establishing a 3.36 × 10⁷ preTERRA variant pool (Figure S5). Again, up to 60 variant pools were independently introduced via CRISPR library integration in the HURRY strain aiming at 5.03 × 10⁸ separate transformants in total, ensuring a fourteenfold oversampling. To select for variants with reduced estradiol responsiveness, we used uracil‐ and histidine‐depleted medium containing 10 μM tamoxifen instead of 5‐fluoroorotic acid‐supplemented medium spiked with estrogens. Although this latter counterselection screening set‐up will remove estradiol‐induced variants, it predominantly yields transformants lacking a receptor or expressing a non‐functional receptor variant.

After selection of the library on uracil‐ and histidine‐depleted medium spiked with 10 μM tamoxifen, the 649 surviving transformants were clonally evaluated via fluorescence‐based flow cytometry for their activity in the presence of respective estradiol (100 nM) and tamoxifen (10 μM) (Figure 3e; Table S5). The elevated estradiol concentration (100 nM vs. 10 nM in previous experiments) in this clonal evaluation was essential to pinpoint variants with further reduced estradiol responsivity compared to preTERRA. As expected, most variants had a lower estradiol response compared to the preTERRA variant, while two clones demonstrated a slightly higher tamoxifen response. Surprisingly, genotyping of the 30 most tamoxifen‐selective variants retrieved the preTERRA mutant or a synonymous variant in 16 cases (Table S6). The 14 remaining transformants contained at least one additional non‐synonymous mutation. Of the two above‐mentioned clones with slightly higher tamoxifen responsiveness, the most tamoxifen selective transformant contained the M384L mutation while the other variant expressed preTERRA.

The tamoxifen‐ and estradiol‐induced response of mammalian codon optimized versions of the top 10 unique tamoxifen‐selective variants were determined in mammalian HEK 293T cells (Figure 3c; Figure S6). All preTERRA derivatives retained their tamoxifen sensitivity and the majority (8 out of 10) displayed a reduced estradiol sensitivity. The tamoxifen‐response of the L391T_L539F_M543E mutant (TERRA) was similar to that of the preTERRA (L539F_M543E) but its responsiveness to estradiol was reduced 40‐fold in mammalian cells. In contrast to ERT2 and preTERRA, the TERRA variant was thus no longer inducible by physiological estradiol levels (1–10 nM) (Figure 3c), thereby overcoming an important limitation observed for the widespread ERT2‐based synthetic tools in metazoans (Donocoff et al., 2020; Jardi et al., ²⁰¹⁷).

2.3. Integration of tamoxifen engineered receptors into genomic tools

To compare the usability of the TERRA LBD with the ERT2 LBD as a regulatory domain for Cre recombinase activity, we transfected a TERRA‐Cre‐TERRA expression plasmid in a loxP‐GFP‐loxP‐RFP HEK 293 cell line (Figure 4a). The Cre recombinase will catalyze site specific recombination between two loxP recognition sites resulting in gene excision (Nagy, 2000). Agonist binding to the ERα LBD assists in the nuclear translocation of the Cre recombinase that can only execute its function when present in the nucleus (Feil et al., 1997; Metzger et al., ¹⁹⁹⁵). A duplication of the LBD was known to prevent ligand independent translocation for the ERT2 LBD‐based tools (Liu et al., 2016; Matsuda & Cepko, ²⁰⁰⁷) and hence was also applied to evaluate our TERRA LBD. The RFP‐fluorescence intensity within every well was counted after a 24 h stimulation with tamoxifen or estradiol. The TERRA‐Cre‐TERRA fusion induced recombination of the loxP‐GFP‐loxP‐RFP reporter with an EC50 of 96 nM for tamoxifen, while the ERT2‐Cre‐ERT2 fusion demonstrated tamoxifen‐mediated translocation events with an EC50 of 1420 nM (p = 0.0053). The estradiol responses in these systems were not significantly different (p > 0.05) (Figure 4b).

FIGURE 4.

Harnessing the TERRA variant into a viable genomic tool. (a) Dual tamoxifen‐inducible Cre recombinase expression cassette composed of the CAG promoter, the Cre recombinase CDS fused both up‐ and downstream by either the TERRA‐LBD or the ERT2‐LBD and the poly A signal. The HEK293‐loxP‐GFP‐loxP‐RFP cells maintain the loxP‐GFP‐loxP‐RFP reporter cassette composed of the CMV promotor, the GFP CDS with poly A signal flanked both up‐ and downstream by loxP sites followed by the RFP CDS and poly A signal. (b) Dose–response curve of the TERRA‐Cre‐TERRA and ERT2‐Cre‐ERT2 fusion proteins for estradiol and tamoxifen after stimulation for 24 h in mammalian HEK293‐loxP‐GFP‐loxP‐RFP cells. Physiological estrogen concentration ranges are indicated by gray shading. Error bars represent the standard error of the mean from three biological replicates. (c) Scheme for in vivo test of Cre‐TERRA. At the age of 8 weeks, mT/mG mice received an intraperitoneal injection of recombinant AAV9 encoding Cre‐TERRA or Cre‐ERT2. At 10 weeks of age, the mice were given either vehicle or tamoxifen via oral gavage. Image was created with Biorender.com. (d) One week thereafter, GFP and dTom fluorescence in the heart of the mice were quantified via microscopic analysis and were used to calculate the % of GFP‐positive area within the heart of the mouse. At least 2 mice were tested per condition. The results are shown as scatter plot with the mean ± SD as line and whiskers; statistical analysis was performed by an Ordinary Two‐way ANOVA with Sidak's multiple comparisons test. (e) Representative microscopic images of heart tissue of Cre‐TERRA or Cre‐ERT2 transduced mT/mG mice. The dTom and GFP fluorescent signals in these overlays are shown in red and green, respectively.

The functionality of the Cre‐TERRA system was further validated in the mT/mG mouse model (Figure 4c). The expression cassette for Cre‐TERRA or CRE‐ERT2 was packaged into recombinant adeno‐associated virus 9 particles (rAAV9) and delivered to mT/mG mice via intraperitoneal injection (Inagaki et al., 2006). Genomic recombination of the mT/mG reporter was induced by tamoxifen administration and resulted in the expression of membrane targeted GFP. Different doses of tamoxifen (190 and 0.19 mg/kg) were used in a previously optimized 2‐dose per 2‐day tamoxifen induction scheme (Jardi et al., 2017). Quantification of the GFP signal in heart tissue of the mT/mG mice demonstrated that Cre‐TERRA was a functional tamoxifen‐inducible gene switch in vivo that is operational at the lowest 0.19 mg/kg dose, while Cre‐ERT2 required the high tamoxifen dose of 190 mg/kg to become activated (Figure 4d,e). The absence of green fluorescence in the heart in the vehicle condition illustrated the absence of basal or leaky Cre‐TERRA activity.

2.4. Structural stability of preTERRA and TERRA after tamoxifen binding

To confirm the tamoxifen selectivity of preTERRA and TERRA, we compared the tamoxifen‐induced stabilization of their LBDs with the stability of tamoxifen‐bound ERα WT LBD. Unliganded wild‐type ERα‐, preTERRA‐ and TERRA‐LBDs were prepared through prokaryotic expression and subsequent purification. No soluble fraction of ERT2‐LBD was obtained because of its aggregation in inclusion bodies.

All purified LBDs demonstrated an increase in melting temperature in the presence of tamoxifen, indicative of a stabilizing effect of the ligand (Figure S7). At 1 μM tamoxifen, 45% stabilization of the TERRA LBD was achieved while 3% or less stabilization was observed for the wild‐type ERα and preTERRA LBDs, highlighting the improved tamoxifen sensitivity of the TERRA‐LBD. The apparent EC50 values of the recombinant proteins for tamoxifen were in the μM range (Figure S7) and were considerably higher compared to the values observed in mammalian cells (Table 1). Suboptimal protein folding due to the absence of cellular components including co‐regulators and heat‐shock proteins (Jeffreys et al., 2020) could explain this phenomenon. Although all unliganded LBDs had similar melting temperatures (46–48°C), the tamoxifen‐bound TERRA‐LBD melted at lower temperatures (52.6°C) compared to ERα‐LBD and preTERRA‐LBD that both melted at 56.7°C. Despite successful purification of the tamoxifen‐bound preTERRA‐ and TERRA‐LBDs in vitro, no diffracting crystals were obtained.

3. DISCUSSION

Here we reported a rational evolution strategy that enabled alteration of the ligand preference of ERα to tamoxifen. This strategy started with the identification of multiple sites in the ligand‐binding domain (LBD) that were randomized by simultaneous mutagenesis. The resulting ERα variant library was screened for ERα activity in response to tamoxifen in a S. cerevisiae strain with dual read‐out based on ligand‐induced survival or fluorescence. Both the Darwin assembly‐based mutagenesis method and the HURRY strain proved to be successful for the identification of active ERα variants. First, two subsequent rounds of structural optimization of the ERα LBD were sufficient to increase the sensitivity to tamoxifen yielding the preTERRA variant and, second, to decrease the response to estradiol resulting in the TERRA variant. During the first rational evolution round, we observed that many highly active clones contained InDels or nonsense mutations that resulted in truncated receptor variants. These truncated receptors lack helix 12 or part of helix 11, that sterically obstructed the bulky sidechain of tamoxifen (Chakraborty & Biswas, 2014), thus explaining their tamoxifen preference.

To enable the use of the final variant in mice, after each round of optimization in yeast, the top 10 tamoxifen sensitive clones were tested in mammalian cells. For selection of preTERRA and TERRA, priority was given to clones with high response in mammalian cells over clones with high response in the HURRY strain. This difference in activity in yeast compared to mammalian cells can be assigned to a (set of) coregulator (s) or heat shock proteins that co‐evolved in humans but is (are) absent or too diverged in S. cerevisiae. Uncovering the interaction networks in both organism via proximity‐dependent biotin identification (BioID) for instance, can elucidate the yeast‐ versus mammalian cell‐specific interactors.

Compared to the preTERRA variant and to ERα WT, the TERRA variant was indeed stabilized by tamoxifen as measured by an increase of the melting temperature of its LBD at 1 μM of tamoxifen. We further demonstrated that the TERRA‐LBD could function as a regulatory domain in the Cre‐lox system in HEK293‐loxP‐GFP‐loxP‐RFP cells and after viral transduction in mT/mG reporter mice. The TERRA‐LBD enabled temporal control of the Cre recombinase at much lower concentrations compared to the ERT2‐LBD. The sensitivity of the inducible Cre‐lox system in the HEK 293 cells increased almost 15‐fold when TERRA‐LBD was used as compared to ERT2‐LBD. Next, expression cassettes of these Cre‐TERRA or CRE‐ERT2 were packaged into recombinant adeno‐associated virus 9 particles (rAAV9) and delivered to mT/mG mice via intraperitoneal injection. Upon administration of the lowest tamoxifen dose (0.19 mg/kg), a clear increase of GFP‐fluorescence was observed in the heart of the Cre‐TERRA transduced mT/mG mice. At this dose, the level of GFP‐fluorescence in the heart tissue of Cre‐ERT2 transduced mice was similar to that of vehicle‐treated Cre‐ERT2transduced mice. The minimum effective concentration of tamoxifen (0.19 mg/kg) for activation of TERRA‐Cre observed in mice is one hundredfold lower than the lowest dose of tamoxifen (20 mg/kg) associated with trabecular and cortical bone formation, which labeled tamoxifen as a major confounder when studying bone phenotypes in mouse models (Xie et al., 2021). In‐depth physiological studies are required for illustrating the absence of or diminished influence of the reduced tamoxifen concentration on mice physiology.

Although the sensitivity of the TERRA‐Cre to tamoxifen was significantly increased in mice as mentioned above, the maximal Cre activity of the TERRA‐Cre seemed to be 2‐fold lower than for the ERT2‐Cre. This was illustrated by a 40% GFP‐positive area in the heart of AAV9‐CAG‐TERRA‐Cre transduced mT/mG mice, while 79% of the heart was GFP‐positive in the AAV—CAG‐ERT2‐Cre transduced mT/mG mice. A possible explanation for this could be the reduced overall stability of the tamoxifen‐bound TERRA‐LBD as evidenced by its melting temperature of 52.6 versus 56.7°C for wild‐type ERα‐LBD. This destabilized ligand‐bound conformation of tamoxifen‐bound TERRA‐LBD can be caused by the rearrangement of helix 12. Molecular dynamic simulations and crystallization studies will clarify the altered ligand binding interface of the preTERRA and TERRA variants. These insights facilitate algorithms such as the Protein Repair One‐Stop Shop (PROSS), that was successful in improving the thermal stability of the wild‐type ERα‐LBD (Kriegel et al., 2021), to highlight target residues for further engineering to improve protein stability.

3.1. Advantages of the HURRY platform

The here described rational evolution method has several advantages over directed evolution. Directed evolution starts from a wild‐type ligand binding domain and combines mutagenesis with natural selection to isolate variants with the phenotype of interest (Evans et al., 2011). A practical bottleneck here is the size of the variant libraries, which increases exponentially with the number of residues to be mutagenized and screened individually, making it near‐impossible to screen every mutation combination using traditional random mutagenesis strategies. Structural insights from rational protein design enable pinpointing hotspot residues for mutagenesis to alter ligand preference thereby creating smart libraries which limit the number of residues to vary and overcome this hurdle. Our strategy of combinatorial mutagenesis and efficient screening yielded de novo tamoxifen receptors that contain a minimum of two beneficial mutations to confer orthogonality (Karanicolas, 2012) which is less likely to be uncovered from traditional single‐site saturation or random mutagenesis.

Recently, continuous directed evolution methodologies have proven successful in the engineering of metabolic pathways (Crook et al., 2016; Molina et al., ²⁰²²). However, presence of InDels or nonsense mutations, as observed during the first rational evolution round, will impede these continuous evolution systems through accumulation of truncated receptor variants. These truncated receptors lack helix 12 or part of helix 11, which sterically obstructs the bulky sidechain of tamoxifen (Chakraborty & Biswas, 2014), thus explaining their tamoxifen preference. To avoid such truncated receptors, non‐continuous directed evolution approaches are necessary.

3.2. Alternative applications of the HURRY platform

Tamoxifen is not an inert ligand, it selectively modulates ERα and is therefore a well‐established therapeutic for breast cancer. Long term treatments inevitably induce resistance that can manifest due to the appearance of mutations in ERα (Toy et al., 2013). During the sequencing of our top performing ERα clones, we have picked up several well‐known tamoxifen‐resistant breast cancer variants of ERα (for example (Y537N)) (Ma et al., 2015), even when this site was not targeted by our mutagenesis approach. This indicates that the HURRY platform can also be adapted to predict drug‐resistance landscapes and mechanisms against new or currently used drugs that target NRs like ERα in breast cancer or the androgen receptor in prostate cancer. So far, such landscapes have only been derived from large patient groups and are unavailable for most NR‐related drugs (Ma et al., 2015). Especially in the field of personalized medicine, pre‐existing knowledge on resistance mutations can aid in providing the appropriate treatment for every patient.

Due to the use of universal steroid receptor responsive elements in the HURRY strain, our rational evolution approach can be adapted to any NR with minimal changes. Analogous systems with altered responsive elements can be engineered to evolve other ligand sensing elements such as riboswitches (Werstuck & Green, 1998) and allosteric transcription factors (Taylor et al., 2016). However, the specific plasticity and residue conservation of every NR or protein necessitates a case‐by‐case approach and may warrant iterative rounds of rational evolution. In an ideal case, each round of evolution is guided by structural information of the new XR‐ligand pair. While we only showed the application of the TERRA‐tamoxifen pair for temporal control of the Cre‐lox system, many other options, such as further development into a gene expression switch, Cas‐based genome editing tool or biosensor are possible.

4. MATERIALS AND METHODS

4.1. Resource availability

4.1.1. Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead Contacts, Christine Helsen (christine.helsen@kuleuven.be), Arnout voet (arnout.voet@kuleuven.be), or Kevin Verstrepen (kevin.verstrepen@kuleuven.be).

4.1.2. Materials availability

Experimental materials generated in this study are available upon request. Depending on the reagent MTA might be required.

4.1.3. Data and code availability

• All data reported in this paper will be shared by the lead contacts upon request.

• This paper does not report original code.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contacts upon request.

4.2. Experimental model details

4.2.1. Bacterial cultivation

All molecular cloning and plasmid propagation steps were performed in the chemically competent DH5α Escherichia coli strain (F ⁻ endA1 glnV44 thi‐1 recA1 relA1 gyrA96 deoR nupG purB20 φ80dlacZΔM15 Δ (lacZYA‐argF)U169, hsdR17 (rK–mK+), λ ⁻ ). The cultures were maintained at 37°C in Luria Bertani (LB) medium with the respective antibiotics.

Production of recombinant HIS‐tagged proteins was performed in the BL21 (DE3) RILP E. coli strain (F ⁻ ompT hsdS (rB ⁻ mB ⁻ ) dcm + Tetr galλ (DE3) endA Hte [argU proL Camr] [argU ileY leuW Strep/Specr]). Of every transformed the BL21 (DE3) RILP E. coli strain, 6 L cultures were grown in LB medium while shaking (200 rpm) to a density of OD₆₀₀ 0.6 at 37°C. Protein production was induced by supplementation of 0.5 mM isopropyl β‐d‐1‐thiogalactopyranoside (IPTG) and incubation shifted to 25°C.

4.2.2. Yeast strain development and growth condition

The BY4741 S. cerevisiae (MATa his3Δ1 leu2Δ0 met15Δ0 ura3Δ0) strain was sequentially modified via homologous recombination and CRISPR‐Cas9 (Mertens et al., 2019) to establish the HURRY strain (MATa his3Δ1::[pSRI1‐HIS3‐tADH1] leu2Δ0 met15Δ0 ura3Δ0::[pSRI2‐URA3‐tADH1] hoΔ::[p (SRI‐minCYC1)‐yemCherry‐tADH1] can1::pPGI1‐yeCitrine‐tADH1). S. cerevisiae codon optimized 3xFLAG‐ERα was subsequently integrated in a similar fashion near the ARS208a locus (Reider Apel et al., 2017) in the HURRY strain (MATa his3Δ1::[pSRI1‐HIS3‐tADH1] leu2Δ0 met15Δ0 ura3Δ0::[pSRI2‐URA3‐tADH1] hoΔ::[p (SRI‐minCYC1)‐yemCherry‐tADH1] can1::pPGI1‐yeCitrine‐tADH1 ars208a::pTEF1‐3xFLAG‐yERα‐tCYC1) for evaluating estrogen receptor activity. All integrations were verified by sequencing (Table S9). For standard cultivations, both liquid and solid media, yeasts were cultivated in synthetic complete medium (SC) supplemented with 2% (w/v) glucose at 30°C.

4.2.3. Cell lines and culturing

Human embryonal kidney 293 T (HEK 293 T, Gender: female) cells were obtained from the American Type Culture Collection (ATCC) while the human embryonal kidney 293 reporter strain (HEK293‐loxP‐GFP‐RFP, Gender: female) was purchased from GenTarget Inc. Both cell lines were cultivated in Dulbecco's Modified Eagle Medium without phenol red (DMEM; Thermo Fisher Scientific) supplemented with GlutaMAX (Thermo Fisher Scientific), 100 U/mL penicillin, 100 μg/mL streptomycin (Thermo Fisher Scientific), and 10% fetal calf serum (FCS; Thermo Fisher Scientific) or 5% charcoal‐stripped serum (CSS) for transfection experiments. Cells were incubated until 70%–80% confluency in a 5% CO₂ humidified 37°C incubator before sub culturing.

4.2.4. Mice housing

Homozygous B6.129 (Cg)‐Gt (ROSA)26Sor^{tm4 (ACTB‐tdTomato,‐EGFP)Luo}/J (mT/mG) mice were received from the Laboratory of Ion Channel Research of VIB‐KU Leuven and housed in the Animal Housing Facility of the KU Leuven under constant temperature, humidity and day/night cycle. Mouse had access to tap water and were fed ad libitum with Teklad Global 19% extruded protein Diet 2019X (Envigo) for improved optical fluorescence imaging. Litter mates of 8 weeks of age were randomly assigned to experimental groups.

Procedures were approved by the Animal Ethical Committee of the KU Leuven (P076/2021).

4.3. Method details

4.3.1. Molecular cloning

Expression vectors

The yeast expression vectors were derived from the p414‐TEF1p‐Cas9‐CYC1t (DiCarlo et al., 2013), a gift from George Church (Addgene plasmid # 43802; http://n2t.net/addgene:43802; RRID:Addgene_43,802). Via the In‐fusion HD Cloning Kit (TaKaRa), the Cas9 coding DNA sequence (CDS) was substituted for the S. cerevisiae codon optimized ERα coding DNA sequence (yERα) provided with the N‐terminal FLAG tag. Next, 120 bp homology arms for genomic integration near the ARS208a target site were introduced both upstream of the TEF1 promoter and downstream of the CYC1 terminator to yield the yERα template for rational evolution (Tables S7–S9).

Mammalian expression vectors of the rationally evolved ERα mutants were established via site‐directed mutagenesis (Tables S7 and S8) of the FLAG tagged human ERα CDS (hERα) in the modified pEGFP‐C1 plasmid (Eerlings et al., 2021). The pCMV‐β‐gal expression vector was acquired from Stratagene (La Jolla, CA).

Cre fusion constructs were generated in the pCAG‐ERT2CreERT2 vector, a gift from Connie Cepko (Matsuda & Cepko, 2007) (Addgene plasmid # 13777; http://n2t.net/addgene:13777; RRID:Addgene_13777), via restriction‐ligation of the PCR amplified human TERRA ligand‐binding domain (LBD).

Reporter constructs

Combining the universal steroid receptor enhancer region (SRi) with respectively the minimal synthetic core promoter (4 or 9) (Redden & Alper, 2015), the S. cerevisiae optimal Kozak sequence, the scHIS3 or scURA3 CDS and the ADH1 terminator, resulted in the SRi1‐URA3 and SRi2‐HIS3 reporters (Tables S7 and S8). The combination of the SRi enhancer region with the minimal CYC1 promoter, the yeast‐enhanced yemCherry CDS and the scDIT1 terminator yielded the SRi‐yemCherry reporter construct. The yemCherry CDS was isolated from the pGADT7‐ADH700‐yeCherry‐pTEF1 (S.C.)‐yeGFP‐DHFR construct, a kind gift from Thomas Wandless (Edwards & Wandless, 2010) (Addgene plasmid # 24584; http://n2t.net/addgene:24584; RRID:Addgene_24,584). The constitutively active yeast enhanced Citrine expression cassette was obtained via Gibson Assembly of PCR amplified PGI1 promoter, yeCitrine CDS and scCYC1 terminator (Table S7).

The universal steroid receptor luciferase reporter (SRi‐Luc) was used for profiling ERα activity in mammalian cells (Eerlings et al., 2021).

CRISPR vectors

The CRISPR vectors were derived from a modified pV1382 vector. First, the pV1382 (Vyas et al., 2018), a gift from Gerald Fink (Addgene plasmid # 111436; http://n2t.net/addgene:111436; RRID:Addgene_111,436), was digested with BglII (Thermo Fisher Scientific) and self‐ligated to remove the URA3 expression cassette. Next, the modified pV1382 vector was digested with BsmBI (Thermo Fisher Scientific) and dephosphorylated by shrimp alkaline phosphatase (Thermo Fisher Scientific). The linearized pV1382 was ligated with 5′‐phosphorylated hybridized DNA primers, spanning the sgRNA sequence of interest and the required overhangs for ligation, to yield the CRISPR vectors for genome editing (Tables S7 and S8). sgRNA sequences were identified via CRISPOR (Concordet & Haeussler, 2018).

Adeno‐associated viral vector designs

The transfer plasmid for rAAV9 production was derived from a bespoke AAV‐CAG‐GFP (AAV serotype 9) construct. In brief, the AAV‐CAG‐GFP was enzymatically digested with MluI and SfbI (FastDigest, Thermo Fisher Scientific) to remove the GFP expression cassette together with its polyA sequence. Next, the Cre‐ERT2 (LBD), Cre‐ERT‐TERRA (LBD) and hGH polyA sequences were PCR amplified with primers that allow subsequent overlap extension PCR to fuse the Cre‐ERT2/−TERRA (LBD) with the PCR fragment of the hGH polyA. After overlap extension PCR, the fragment was cloned into the linearized AAV‐CAG‐GFP backbone using NEBuilder HiFi DNA Assembly (New England Biolabs Inc.) to obtain the AAV‐CAG‐Cre‐ERT2 and AAV‐CAG‐Cre‐TERRA vectors for rAAV9 production. The integrity of the inverted terminal repeats (ITRs) was checked by sequencing the vectors after SmaI and Eam1105I digestion of the ITRs.

4.3.2. Survival assay

From an overnight culture in synthetic complete medium (2% glucose), the cell number was determined by Cell Counter (Bio‐Rad) and seeded at a density of 50.000 cells/mL in uracil and histidine depleted minimal medium spiked with the chemical of interest. Cells were incubated for 36 h at 30°C on a shaking platform (750 rpm) before determining OD₆₀₀ by plate reader (Thermo Fisher).

4.3.3. Rational evolution

Identification

The ERα‐LBD (PDB ID:1QKU) (Gangloff et al., 2001) was taken as a reference for rational design. The residues surrounding the hormone binding pocket and on helices 11 and 12 were considered for diversification. Specific target positions were determined on the basis of their conservation scores and binding energy with estradiol (E8875, Sigma‐Aldrich) and tamoxifen (T5648, Sigma‐Aldrich). Conservation scores were obtained from ConSurf web‐server (Ashkenazy et al., 2016). The binding energies were calculated by using snapshots from molecular dynamic simulations (GROMACS) (Abraham et al., 2015) followed by the Generalized Bonn approximation.

Diversification

The diversification strategy was adapted from the in vitro combinatorial mutagenesis method, Darwin Assembly (Cozens & Pinheiro, 2018). The yERα for rational evolution (300 ng/μL) was made single‐stranded by co‐incubation with 30 U of ExoIII (New England Biolabs) and 40 U of Nt. BspQI (New England Biolabs) for 2 h on 37°C. Next, the 5′ phosphorylated spiked primers targeting the identified sites (44 pmol each) and the boundary primers (2.2 pmol each) were assembled onto the single‐stranded plasmid (0.22 pmol) by incubating for 1 h at 50°C in 2× DA buffer (0.05 U/μl Q5 High‐Fidelity DNA polymerase, 8 U/μl Taq DNA ligase, 2 mM NAD⁺, 0.4 mM of every dNTP, 10% (w/v) PEG 8000, 2 mM DTT, 1× CutSmart buffer). In parallel, 5 μL Streptavidin coated beads (Dynabeads™ MyOne™ Streptavidin T1; Thermo Fishier Scientific) were blocked in 2× BWBS‐T (20 mM Tris*HCl (pH 7.4), 2 M NaCl, 0.2% Tween‐20, 2 mM EDTA) for an hour on a spinning wheel at room temperature. After isothermal assembly, the biotinylated DNA fragments were immobilized on the pre‐blocked beads by incubation for 3 h at room temperature on a spinning wheel. To release the immobilized DNA products, the beads were washed twice with warm NaOH (30 mM at 37°C) and once with EB‐T (10 mM Tris*HCl (pH 8.8), 0.01% Tween‐20, 0.1 mM EDTA) before resuspending in 10 μL EB (10 mM Tris*HCl (pH 8.8)). The resuspended beads were immediately used for PCR with outnest primers (Table S8) followed by gel purification with the GeneJet Gell extraction kit (Thermo Fisher Scientific). The purified, double stranded, linear libraries were used as repair template in the CRISPR‐directed library transformation.

For the CRISPR‐directed library integration, 15 batches of the HURRY strain were independently cultivated in 50 mL YPAD (1% yeast extract, 2% peptone, 0,004% adenine, 2% glucose) on 30°C for 3–5 h in a shaking incubator (200 rpm) until the exponential phase was reached. Following a washing step with 5 mL lithium acetate (LiAc) (0.1 M), yeast cells were resuspended in 100 μL LiAc (0.1 M) and incubated for 10 min at room temperature before aliquoting in 30 μL. To every aliquot, 1 μg CRISPR plasmid targeting near the ARS208a locus (Table S7), 10 μg purified linear yERα mutant library, 40 μg single‐stranded carrier DNA and 750 μL PL solution (42% PEG 3350, 120 mM LiAc) were added. The cell suspensions were incubated for 30 min at 30°C before delivering the 14 min heat shock at 42°C. After centrifugation (3000 rcf 3 min), the yeast pellets were resuspended in CaCl₂.2H₂O (5 mM) and incubated for 5 min at room temperature. Next, cells were washed twice with YPD (2% glucose) before plating on YPD agar (2% glucose) for overnight incubation on 30°C. The following day, the transformants were replica plated on uracil and histidine depleted minimal medium supplemented with tamoxifen (10 μM) and grown for 48–72 h at 30°C.

Characterization

Per round of selection, transformants from 60 independent CRISPR‐directed library integrations were plated on uracil and histidine depleted minimal medium supplemented with tamoxifen (final concentration of 10 μM). Cells expressing yERα mutants that can induce the NR‐selective growth pressures are able to grow on this medium.

Each clone was incubated overnight in synthetic complete medium with 10 nM estradiol, 10 μL tamoxifen or without ligand. After 16–24 h, the respective yeCitrine and yemCherry levels were measured on a flow cytometer with high‐throughput sampler (Attune NxT Flow Cytometer). The clonal yemCherry fluorescence intensities were corrected by taking the ratio of the clonal yemCherry fluorescence and the corresponding yeCitrine signals. Next, the corrected yemCherry fluorescence in presence of tamoxifen or estradiol for every variant was normalized to the fluorescence intensity obtained for the wild‐type yERα in presence of estradiol. Subsequently, the tamoxifen selectivity of every mutant was calculated by determining the ratio of the normalized tamoxifen signal over the normalized estradiol signal for that particular variant.

4.3.4. Mammalian transactivation assays

Luciferase assay

HEK 293T cells were seeded 24 h prior to transfection in a 96‐well plate (Greiner) at a density of 15,000 cells per well in DMEM with 5% charcoal‐stripped serum. The seeded cells were transfected with a mixture of 100 ng SRi‐Luc, 10 ng of hERα, or human codon optimized mutant of interest and 5 ng of pCMV‐β‐gal using GeneJuice® Transfection Reagent (Novagen) conform the manufacturer's instructions. After overnight incubation, the medium was refreshed and spiked with vehicle or increasing concentrations of either tamoxifen or estradiol. After 24 h stimulation, cells were lysed in 25 μL Passive Lysis Buffer (Promega). Luciferase activity and β‐galactosidase activity were measured and processed as described (Eerlings et al., 2021). The observed luciferase activities were corrected by the corresponding β‐galactosidase activities.

Fluorescent assay for Cre activity

HEK293‐loxP‐GFP‐RFP cells were seeded 24 h prior to transfection in a 96‐well plate (TPP) at a density of 15,000 cells per well in DMEM with 5% charcoal‐stripped serum. The seeded cells were transfected with 1 ng of the Cre‐fusion protein of interest, 5 ng of pCMV‐β‐gal and 109 ng pGEM‐T using GeneJuice® Transfection Reagent as per manufacturer's instructions. After overnight incubation, the medium was refreshed and spiked with either tamoxifen, estradiol or vehicle. The total fluorescence intensity (GFP and RFP) on a per well basis was determined after 24 h stimulation with respectively estradiol and tamoxifen stimulation with the Incucyte ZOOM live‐cell imaging system (Essen Bioscience).

Recombinant AAV production and purification

Recombinant AAV9 was produced in HEK293T (ATCC) cells using standard procedures (Tordo et al., 2018). Briefly, the transfer construct was co‐transfected with pDG9 (Kohlbrenner & Weber, 2017) and 72 h later cells & supernatant were harvested. Clarified cell lysate was generated by freeze‐thawing the resuspended cell pellet, treating with benzonase to remove cellular and non‐encapsidated DNA and pelleting the debris. The clarified cell lysate was combined with the filtered supernatant and purified by affinity (POROS™ GoPure™ AAV9, ThermoFisher Scientific) and anion exchange (POROS™ 50 HQ, ThermoFisher Scientific) chromatography. Purified rAAV vectors were dialyzed, filter sterilized, aliquoted and stored at −80°C. Viral genome titers were determined by ddPCR and capsid titers by AAV sandwich ELISA (Progen). Alkaline gel electrophoresis was performed as previously described to assess viral genome integrity and purity (Fagone et al., 2012). SDS‐PAGE was performed to assess the purity of the viral preparations (Kohlbrenner et al., 2012). Finally, endotoxin levels were determined by LAL assay (Charles River).

4.3.5. Biochemical analysis

Protein purification

The HIS‐tagged ERα‐LBD (WT, preTERRA and TERRA) were expressed from a pET15b in E. coli BL21 (DE3) RILP cells. 6 L cultures were grown with shaking to a density of OD₆₀₀ 0.6 at 37°C and induced for 4 h. with 0.5 mM when isopropyl β‐d‐1‐thiogalactopyranoside (IPTG) at 25°C. Proteins were purified according to previously reported procedure, however in absence of any ligand (Fanning et al., 2016). At the final size exclusion chromatography (SEC) step (Superdex75 16/60 column), the dimeric fraction was isolated and concentrated to 1 mg/mL and dialyzed against the assay buffer (50 mM HEPES pH 8, 500 mM NaCl and 0.5 mM TCEP).

Differential scanning fluorimetry

Differential scanning fluorimetry (DSF) was performed using a QuantStudio 3 Real‐Time PCR Systems (ThermoFisher) to determine the thermal stability of the proteins. His‐tagged protein samples (~1 μg) were incubated with tamoxifen 0.5–50 μM, 2.5 μL of 8× Protein Thermal Shift Dye (Applied Biosystems) in a total volume of 20 μL within MicroAmp™ Optical 96‐Well Reaction Plates. After 30 min of incubation, the temperature was raised from 25 to 95°C with a velocity of 0.05°C/sec. The fluorescence intensity was measured and the melting temperature (Tm) was determined by the derivatives of the melt curve (dFluorescence/dT) using Protein Thermal Shift Software version 1.3 (Applied Biosystems). Half maximal effective concentration (EC50) was determined and the 95% confidence interval was calculated by using PRISM 9.3.1.

4.3.6. Mice

Viral injection and detection of Cre activity

Via intraperitoneal injection, AAV9‐CAG‐Cre‐TERRA or AAV9‐CAG‐Cre‐ERT2 particles (1.8 × 10¹² vg/mouse) were administered to male mT/mG mice of 8 weeks old. Following the viral injection, mice were kept in IVC for 2 weeks prior to tamoxifen administration. Tamoxifen (Sigma‐Aldrich T5648‐5G) was dissolved in sunflower seed oil with 100% EtOH (9:1) at a concentration of 40 mg/mL. The tamoxifen administration protocol consists of 2 doses of 190 mg/kg p.o. on two consecutive days (Jardi et al., 2017), additionally a 1/1000 dilution of this dose was used (0.190 mg/kg p.o.) in the same regimen. One week after tamoxifen or vehicle administration, the heart was dissected from the mouse and frozen for tissue sections. For ex‐vivo analysis of the Cre‐TERRA activity, we quantified the dTom to GFP conversion in the dissected heart of the mouse using the tissue cytometer TissueFAXS iPLUS (TissueGnostics GmbH, Vienna, Austria). Frozen sections of the heart were imaged at 20× magnification in FITC channel (excitation 494 nm, emission 520 nm) and the Texas Red channel (excitation 595 nm, emission 620 nm). At least 25 fields of view of the heart were imaged to obtain an average value of the whole heart. The parameters “Total tissue area,” “GFP‐positive,” and “dTom‐positive” area in mm² were obtained via analysis of the images with StrataQuest (TissueGnostics GmbH, Vienna, Austria) using the “total area measurement” algorithm. Those parameters were further used to calculate the percentage of GFP‐positive area within the heart. A scatter plot with the mean and SD indicated as a line with whiskers is given for each condition. Minimal two animals per condition are used for statistical analysis. An Ordinary Two‐way ANOVA with Sidak's multiple comparisons test was performed to compare Cre‐TERRA with Cre‐ERT2 at different tamoxifen doses. p‐Values of less than 0.05 are considered significant.

4.4. Quantification and statistical analysis

Data represent the mean value ± standard deviation (SD) or mean values ± standard error of the mean (SEM) of the number of biological replicates (N) as specified in the figure legends. If not otherwise stated statistical analysis was done using GraphPad Prism 9. Statistical analysis is described in the “Method details” section or in the associated figure legends. Levels of significance were p < 0.05 (*); p < 0.01 (**); p < 0.001 (***).

AUTHOR CONTRIBUTIONS

Els Henckaerts: Resources; writing – review and editing; supervision; methodology; project administration. Vitor B. Pinheiro: Supervision; writing – review and editing; validation; methodology. Karin Voordeckers: Resources; supervision; conceptualization; writing – review and editing; validation. Duy Tien Ta: Writing – review and editing; investigation; visualization. Anton Gorkovskiy: Writing – review and editing. Frank Claessens: Supervision; resources; conceptualization; writing – review and editing; validation; data curation; project administration. Kevin Verstrepen: Resources; supervision; conceptualization; writing – review and editing; validation; project administration. Irina Thiry: Writing – review and editing; investigation; visualization. Tien Nguyen: Writing – review and editing; software; investigation; visualization; methodology. Sarah El Kharraz: Writing – review and editing; investigation. Xiao Yin Lee: Writing – review and editing; investigation; software; visualization; writing – original draft; methodology. Roy Eerlings: Writing – review and editing; writing – original draft; methodology; conceptualization; investigation; formal analysis; visualization; data curation; project administration. Purvi Gupta: Writing – review and editing; software; data curation; investigation; visualization; methodology; writing – original draft. Bram Vandewinkel: Writing – review and editing; investigation; visualization; methodology; writing – original draft. Lisa Moris: Writing – review and editing. Wout Devlies: Writing – review and editing. Florian Handle: Writing – review and editing; validation. Elien Smeets: Writing – review and editing. Christine Helsen: Resources; funding acquisition; supervision; project administration; conceptualization; writing – review and editing; writing – original draft; formal analysis; validation; visualization; data curation; methodology. Arnout Voet: Resources; funding acquisition; supervision; conceptualization; writing – review and editing; writing – original draft; software; formal analysis; validation; data curation; methodology; project administration.

CONFLICT OF INTEREST STATEMENT

The authors declare no competing financial interests.

Supporting information

DATA S1 Supporting Information.

DATA S2 Supporting Information.

Eerlings R, Gupta P, Lee XY, Nguyen T, El Kharraz S, Handle F, et al. Rational evolution for altering the ligand preference of estrogen receptor alpha. Protein Science. 2024;33(4):e4940. 10.1002/pro.4940