Development of a cancer-marker activated enzymatic switch from the herpes simplex virus thymidine kinase

Abstract

Discovery of new cancer biomarkers and advances in targeted gene delivery mechanisms have made gene-directed enzyme prodrug therapy (GDEPT) an attractive method for treating cancer. Recent focus has been placed on increasing target specificity of gene delivery systems and reducing toxicity in non-cancer cells in order to make GDEPT viable. To help address this challenge, we have developed an enzymatic switch that confers higher prodrug toxicity in the presence of a cancer marker. The enzymatic switch was derived from the herpes simplex virus thymidine kinase (HSV-TK) fused to the CH1 domain of the p300 protein. The CH1 domain binds to the C-terminal transactivation domain (C-TAD) of the cancer marker hypoxia inducible factor 1α. The switch was developed using a directed evolution approach that evaluated a large library of HSV-TK/CH1 fusions using a negative selection for azidothymidine (AZT) toxicity and a positive selection for dT phosphorylation. The identified switch, dubbed TICKLE (Trigger-Induced Cell-Killing Lethal-Enzyme), confers a 4-fold increase in AZT toxicity in the presence of C-TAD. The broad substrate specificity exhibited by HSV-TK makes TICKLE an appealing prospect for testing in medical imaging and cancer therapy, while establishing a foundation for further engineering of nucleoside kinase protein switches.

Introduction

The variety of approaches for treating cancer includes those that target different aspects of the disease. These approaches range from small molecule drugs that disrupt internal cancer cell replication to protein therapeutics, such as antibodies designed to specifically inhibit cancer-proliferating growth factor receptors (Niculescu-Duvaz et al., 1999; Jordheim et al., 2013; Primrose et al., 2014). While many of these therapeutic approaches show an ability to reduce tumor size and impede the spread of cancer, they also can subject the patient to harsh side effects. Among the most effective and recognized categories of cancer treatment is chemotherapy, which is well known for side effects that include hair loss, severe fatigue, nausea and loss of appetite (Chabner and Roberts, 2005). These side effects arise from cancer drugs killing normal, healthy cells along with cancer cells (Niculescu-Duvaz et al., 1999; Chabner and Roberts, 2005; Jordheim et al., 2013). The off-target activity of these drugs limits the dosage that can be administered to patients (Niculescu-Duvaz et al., 1999; Corrie, 2011).

A number of attempts have been made to increase specificity of cancer drugs toward cancer cells and reduce their toxicity toward healthy cells. This has led to the development of prodrugs, chemically inactivated drugs that depend on cellular metabolism to be reactivated and generate toxicity (Rautio et al., 2008; Huttunen et al., 2011). Such drugs aim to increase specificity toward cancer cells by relying on metabolic rate discrepancies between the cancer cells and regular cells; the typically high rate of metabolism within cancer cells generates more toxicity from prodrugs at lower doses compared with healthy cells (Sheng et al., 2009). Since this approach still allows prodrug activation in healthy cells, a modified form of the therapy called gene-directed enzyme prodrug therapy (GDEPT) was developed (Springer and Niculescu-Duvaz, 1996). GDEPT increases specificity toward cancer by delivering a non-native gene responsible for activating prodrugs into cancer cells (Springer and Niculescu-Duvaz, 2000). In theory, the targeted delivery allows only cancer cells to harbor the gene and its corresponding protein that is capable of activating the prodrug. However, most gene delivery systems lack the required target specificity to make GDEPT a viable approach for cancer treatment (Löbler et al., 2009; Sheng et al., 2009). Due to this lack of specificity in GDEPT systems, other target-specific methods such antibody-drug conjugates and nanomaterials have been investigated for drugging cancer. These methods can offer certain advantages over traditional GDEPT, including increased cancer specificity, but also possess their own limitations, such as tumor penetration issues and a greater risk of heptatoxicity (Löbler et al., 2009; Firer and Gellerman, 2012).

The lack of target-specific gene delivery systems for GDEPT could be addressed by engineering the non-native gene product to have its activity dependent on specific protein markers of cancer through the appropriate fusion of the enzyme to a protein that binds the cancer marker. Our laboratory has previously developed such a protein, dubbed HAPS59, by fusing the yeast cytosine deaminase (yCD) to a CH1 domain. The binding of the cancer marker, hypoxia inducible factor 1α (HIF1-α), to the CH1 domain increases the cellular accumulation of HAPS59, causing increased activation of the prodrug 5-fluorocytosine (5-FC) (Wright et al., 2011). This engineered ‘enzymatic switch’ can potentially provide the lacking cancer specificity by only activating the prodrug in the presence of specific cancer markers.

Multiple studies have shown that combining GDEPT systems can have synergistic effects in generating toxicity toward cancer (Willmon et al., 2006; Nawa et al., 2008). While many factors contribute to this observation, the bystander effect displayed in GDEPT systems can be tremendously useful in killing cancer cells. Combining enzyme–prodrug combinations may allow for proper attenuation of this effect without having to change dosage of each drug (Springer and Niculescu-Duvaz, 1996; Dachs et al., 2009). The enzyme–prodrug combination of cytosine deaminase (CD) and 5-FC is known for having a large bystander effect, whereas the combination of the herpes simplex virus thymidine kinase (HSV-TK) and ganciclovir (GCV) is known for its smaller bystander effect (Dachs et al., 2009). When the two enzyme–prodrug systems are used together, a large synergistic effect is seen (Willmon et al., 2006). For this reason, the prospect of creating an enzymatic switch out of HSV-TK that can be used alongside previously developed enzymatic switches from yCD is appealing (Wright et al., 2011, 2014). Having two switches with increased target specificity could allow for a synergistic approach to fighting cancer at even higher dosages. Furthermore, HSV-TK's broad substrate specificity lends itself to be used as a reporter gene with other nucleoside analogs besides GCV, and a switch derived from the enzyme could have important implications in medical imaging (Gambhir et al., 2000; Yaghoubi and Gambhir, 2006).

An established strategy for developing enzymatic switches involves fusing a signal-recognizing domain to an output domain (Ostermeier, 2005). Conceptually, the fusion of the two can couple signal recognition to a specific output. However, the difficulty in implementation of this strategy revolves around identifying the site of insertion of the signal-recognition domain that results in the switch property. A directed evolution approach involving the subjection of a library of variants with the signal-recognition domain in each insertion point on the output domain to selection pressures is useful (Guntas et al., 2005). The selection would identify specific switches without the need to predict or understand the mechanism by which the switch functions.

In this study, we used a directed evolution method to engineer the HSV-TK enzyme to have its cellular activity modulated by the presence or absence of the C-terminal transactivation domain (C-TAD) of the HIF1-α cancer marker. HSV-TK possesses a large active site and is promiscuous in phosphorylating a variety of different nucleoside analogs (Brown et al., 1995; Gardberg et al., 2003). HIF1-α is a transcription factor that is upregulated in hypoxic environments, such as tumors, but is almost non-existent in normoxic cells (Semenza, 2001). The transcription factor is involved in regulating cell proliferation, apoptosis and tumor angiogenesis in hypoxic conditions (Carmeliet et al., 1998). Sensitizing HSV-TK activity to respond to the presence of HIF1-α would provide a way to generate more cancer cell-specific toxicity. To create the switch, the CH1 domain of the p300 protein was inserted into the amino acid backbone of HSV-TK. In its native context, the C-TAD of HIF1-α interacts with the CH1 domain and the CH1 domain goes from a molten globule state to a structured state upon binding of the C-TAD (Freedman et al., 2002; Dial et al., 2003). Our aim was to exploit this interaction and have it influence the level of kinase activity in an HSV-TK/CH1 fusion protein. The directed evolution method we utilized involved a two-tiered selection scheme to sequentially select for an ‘On’ and an ‘Off’ state of a switch. This strategy led to the discovery of a protein switch we dubbed TICKLE (Trigger-Induced Cell-Killing Lethal-Enzyme) that confers C-TAD-dependent azidothymidine (AZT) toxicity to Escherichia coli cells. TICKLE establishes a way to modulate cellular activity of HSV-TK at the protein level and provides a platform to further engineer kinase-based protein switches tailored toward specific substrates.

Materials and methods

Unless otherwise specified, all molecular biology protocols were performed using New England Biolabs’ (NEB) High-Fidelity Phusion Master Mix for PCR, Invitrogen's Gel Purification Kit for gel extraction and Zymo's DNA Clean & Concentrator kit for DNA purification (5 µg loading capacity). All DNA phosphorylation reactions were conducted using NEB's T4 Polynucleotide Kinase with corresponding buffer.

Vectors and strains

The plasmid pMCC contains a wild-type hsv-tk gene and was a gift from Dr Margaret Black (University of Washington). The hsv-tk gene on the vector is under the control of the temperature sensitive lambda pR promoter system and confers ampicillin resistance (AmpR). The lambda pR promoter is constitutively active at 37°C and does not need to be induced. We created a previously described inactive variant of the hsv-tk gene as a control (Munir et al., 1993). The pGA-GSTHIF1α vector housing the C-TAD of HIF1-α attached to a GST-tag on its N-terminus was previously described, as was a corresponding control vector with GST only called pGA-GST (Wright et al., 2014). The AmpR gene was exchanged with a chloramphenicol resistance gene (CmR) for each pGA vector to make them compatible for switch assays. The vectors were dubbed pGAC-GST and pGAC-GSTHIF1α. Both GST and GSTHIF1α genes were also amplified using PCR and cloned into a spectinomycin resistance (SpecR) pSkunk2 vector (pSK2) under the tac promoter (Wright et al., 2014); these specific vectors were utilized during selections while pGAC-based vectors were used in toxicity assays due to their tightly controlled arabinose promoter. Unless otherwise indicated, all experiments were performed using the pMCC vector for expression of HSV-TK, the CH1/HSV-TK libraries and TICKLE. Vectors utilized in the creation of genomic deletions (pKD4, pKD46 and pRHAM) have been described previously (Russell and Ostermeier, 2014).

Escherichia coli  KY895 (-tdk) was acquired from Coli Genetic Stock Center at Yale University. Escherichia coli 5α was used for cloning and purchased from NEB. Escherichia coli NS01 (-tdk, -thyA) was a strain created using the Lambda Red Recombinase system and E. coli KY895 as a parent strain (Datsenko and Wanner, 2000). The strain has its chromosomal thyA gene replaced with a nptII gene to confer kanamycin resistance and a copy of thyA is provided on the pRHAM vector under the control of a rhamnose the promoter. Escherichia coli NS01 is conditional viable in the presence of rhamnose and non-viable in the presence of glucose and absence of rhamnose.

Linearized HSV-TK insertion library

We utilized inverse PCR to develop a linearized DNA library with the hsv-tk gene opened up after 273 separate codons as described (Kanwar et al., 2013). Separate PCRs were performed using primer pairs targeting 273 different insertion sites in a 96-well format to get a linearized library of pMCC-hsv-tk. The library was named ‘linear pNYS’. Samples from every PCR were electrophoresed on 1.2% Tris-acetate-EDTA (TAE) gels to confirm the correct band size. Pooled PCRs were electrophoresed on a TAE gel and the desired band extracted from the gel. The resulting elution was purified further using Zymo Research's DNA Clean & Concentrator (25 µg loading capacity).

Degenerate and specific CH1 linkers

To create a degenerate CH1 linker library, we ordered oligonucleotides consisting of 0, 1, 2 and 3 degenerate NNK codons from Integrated DNA Technologies. The oligonucleotides were designed to bind to the 5′- and 3′- termini in the CH1 gene on the plasmid pRW0017 with the degenerate codons on the flanks (Wright et al., 2014). We then created the DNA library by PCR of CH1 using these degenerate primers. The PCR product was extracted from the gel and purified with Zymo Research's DNA Clean & Concentrator (5 μg loading capacity). The degenerate PCR library was dubbed ‘x(XXX)-CH1-x(XXX)’. We also used PCR to create CH1 constructs with predefined flanks composed of every permutation of GGS, (GGS)₂ and (GGS)₃ on the N-terminus and every permutation of STT, (STT)₂ and (STT)₃ on the C-terminus. We named this PCR library x(GGS)-CH1-x(STT). PCR was also used to amplify CH1 without added linkers for the purpose of a no-linker library.

HSV-TK–CH1 fusion and transformation

To create library pNYS01, we phosphorylated the amplified CH1 without any linkers and ligated it with 1 μg of linear pNYS in a 3:1 molecular ratio (CH1 to pNYS01). The pNYS02 and pNYS03 libraries were created analogously using x(GGS)-CH1-x(STT) and x(XXX)-CH1-x(XXX) as the inserts, respectively. The ligation reactions were conducted overnight at 16°C and purified after. The purified ligation was then transformed into NEB's 5α electrocompetent cells. Transformed cells were grown in liquid media at 30°C for 1 hour and then plated on large tryptone agar plates. Colonies were grown overnight at 37°C and then collected in liquid tryptone media with 15% glycerol. Collected stocks were frozen at −80°C for future use in selections while a portion was extracted for plasmid DNA of pNS01 using a miniprep kit.

One hundred nanograms of extracted, circular pNYS01, pNYS02 or pNYS03 were transformed into E. coli KY895 electrocompetent cells for use in selections. Transformed cells were incubated for 1 hour and plated at 37°C overnight on a large tryptone agar plate. Twenty or more colonies from transformants of each library were sequenced to ensure HSV-TK/CH1 fusions were properly made. Libraries displaying a CH1 insertion in at least 75% of the sequenced transformants were utilized in subsequent selections to identify protein switches.

Two-tiered selection and screening

We first subjected the libraries to a negative selection to select for the ‘Off’ state of the enzymatic switch. Bacteria containing switch candidates and pSK2-GST were plated on solid tryptone media with 10 μg/ml AZT, 0.3 mM IPTG, 25 μg/ml Amp and 25 μg/ml spectinomycin (Spec). The plates were incubated at 37°C overnight and the surviving colonies were collected in liquid media. After colony collection, the plasmid DNA was then extracted from the collected stock and 100 ng of the DNA was transformed into E. coli KY895 strain with pSK2-GSTHIF1α. These transformants were used in a positive selection to select for the ‘On’ state of the enzymatic switch. This specific selection protocol was adopted from a previous study selecting for higher activity HSV-TK variants (Black et al., 1996). In this selection, bacteria were plated on tryptone media with 10 μg/ml 5-Fluorouridine (5-FdU), 2.5 μg/ml uridine (dU), 10 μg/ml thymidine (dT), 25 μg/ml Amp, 0.3 mM IPTG and 25 μg/ml Spec. The plates were incubated at 37°C for 28 hours. After the selection, we picked the surviving colonies and grew them overnight at 37°C in tryptone media with 0.2% glucose (Glu) so that they could be screened for switching activity.

We used a liquid culture spotting assay to screen for switching activity. The overnight cultures from the positive selection were normalized to an optical density (OD) of 0.6 and then diluted 10 000-fold. In total, 1.5 µl of the diluted cultures was then spotted on two different types of solid tryptone media screening plates: the first screening plate tested for ‘On’ activity and had 10 μg/ml AZT with 0.3 mM IPTG, while the second plate tested for ‘Off’ activity and had 10 μg/ml AZT with 2% Glu. Plates were incubated at 37°C overnight for 18 hours and then compared to qualitatively assess cell density of corresponding spots between the ‘On’ and ‘Off’ plate. We selected candidates that displayed a stark difference in cell density and prepared overnight cultures at 37°C in liquid media. The cultures were extracted for plasmid DNA, transformed into a fresh cellular background (E. coli KY895 with pGAC-GSTHIF1α) and analyzed again in the same liquid culture spotting assay to confirm the switching phenotype. In this selection, 0.01% Ara was used instead of IPTG as the GST and C-TAD-GST genes were under control of the arabinose promoter. A switch to the more tightly controlled arabinose promoter from the tac promoter was done to minimize the effects of leaky expression that would dampen the switching phenotype.

AZT toxicity assay

We conducted a liquid cell viability assay to assess the degree of switching between ‘On’ and ‘Off’ states of the HSV-TK switches. One million colony forming units (CFUs) from overnight cultures of E. coli KY895 containing the switch candidate and pGAC-GST or pGAC-GSTHIF1α were added to liquid LB media containing 0.2% Glu, 25 μg/ml of Amp, 25 μg/ml chloramphenicol (Cm) and different concentrations of AZT. Arabinose (0.01%) was added to wells containing cells with pGAC-C-TAD-GST and pGAC-GST. A positive control of cells with wild-type HSV-TK and a negative control of cells with knocked out HSV-TK (HSV-TKΔ) were also grown in identical fashion and used in the cell toxicity assay in the same liquid media conditions, excluding the addition of arabinose and Amp. After 6 hours of incubation at 37°C, we measured the OD of the cultures at 600 nm (OD₆₀₀).

Activity toward dT

Switch candidates were transformed into this E. coli NS01 housing pGAC-GST or pGAC-GSTHIF1α for a liquid cell viability assay. In 96 well culture plates, 500 000 CFUs of each strain type were inoculated into tryptone media containing 0.2% Glu, 5 μg/ml chloramphenicol (Cm), 0.01% arabinose, 25 μg/ml Amp, 25 μg/ml kanamycin (Kan) and different concentrations of dT. After 8 hours of incubation at 37°C we measured the OD₆₀₀.

Western blot analysis

We expressed TICKLE with and without the C-TAD in E. coli overnight and normalized the OD between the two cultures the next day. Additional cultures included cells expressing HSV-TK and cells expressing neither HSV-TK nor TICKLE. The cultures with an equal number of cells were then lysed using Novagen's BugBuster reagent as per the supplied protocol. The soluble fractions were collected after chemical lysis of the bacterial cultures and then electophoresed on a NuPage 4–12% Bis-Tris protein gel (ThermoFisher Scientific) using SDS-PAGE. The proteins from the gel were then transferred on to a polyvinylidene fluoride membrane using Biorad's Trans-Blot SD Semi-dry Transfer Cell for 15 min at 15 V. After the transfer, the primary antibody against HSV-TK (Santa Cruz Biotechnologies, Inc.) was incubated with the membrane in Tris-buffered saline-Tween (TBST) buffer at a 3:500 antibody:buffer ratio for 1 hour at 4°C. The anti-goat secondary antibody (Biorad Laboratories, Inc.) was loaded on to the western blot in TBST buffer at a 3:500 antibody:buffer ratio. No wash steps or blocking steps were performed. We visualized the western by adding horseradish peroxidase substrate using Biorad's Clarity Western ECL Substrate as per the protocol provided. The chemiluminescence was observed using Biorad's Universal Hood II and Quantity One Software.

Results

Library construction

To sensitize the HSV-TK enzyme to the C-TAD of HIF1-α, the CH1 domain needed to be incorporated into the kinase. We generated DNA libraries of HSV-TK/CH1 fusions in which the CH1 domain was incorporated at different insertion sites in HSV-TK using different types of linkers. These libraries were created by blunt-end ligation of the CH1-encoding DNA in between codons of the hsv-tk gene (Fig. 1). We developed three libraries. The first library was dubbed pNYS01 and had the CH1 domain directly incorporated at 273 different insertion points in HSV-TK. We only chose 273 of the 376 possible sites for domain insertion after scrutinizing the protein structure for solvent accessible regions and avoiding regions deeply embedded inside the protein. A second library, pNYS02, was created with every permutation of (GGS), (GGS)₂ and (GGS)₃ linkers on the N-terminus of the CH1 and every permutation of (STT), (STT)₂ and (STT)₃ linkers on the C-terminus. We chose flexible linkers since they have been successfully used in creating switches with CH1 (Wright et al., 2014) and introduction of flexibility in a linker can convert a non-switch into a switch for domain insertion proteins (Choi et al., 2015). The CH1 domain, with these linkers, was incorporated into the same 273 insertion points of HSV-TK. We also constructed a third library, pNYS03, which also used the same 273 insertion points as the other two libraries but had the CH1 domain connected to HSV-TK through all permutations of (XXX), (XXX)₂ and (XXX)₃ on both the N- and C- termini of the CH1 domain. X designates any amino acid and was encoded by NNK degenerate codons.

Fig. 1.

Construction of libraries. (A) Libraries with the CH1 domain inserted into HSV-TK were created using inverse PCR in which the plasmid containing hsv-tk was linearized at specific codons within the gene. The linearized plasmid was ligated with CH1 DNA to create the circularized plasmid library with the CH1 inserted into HSV-TK at different sites. Various linkers of different composition and length were added to the CH1 DNA by PCR before the ligation step to generate the libraries: (B) pNYS01, (C) pNYS02 and (D) pNYS03. Linkers shown are illustrative of the diversity.

The pNYS01 library was constructed to test and identify variants that could function as a switch without the added flexibility of linkers. Since the library had the CH1 inserted into 273 sites of HSV-TK, the theoretical library size was only 2 × 273 = 546 (accounting for forward and backward insertions). The pNYS02 library was the first library to possess variants with linkers connecting the CH1 to HSV-TK. The Gly-Gly-Ser linker used for the N-terminus of CH1 is commonly utilized to generate flexibility between domains. Using Ser-Thr-Thr linker for the C-terminal linker of CH1 was a mistake that resulted from incorrect primer design for oligonucleotides used in PCRs to amplify the CH1-encoding DNA. We decided to move forward with the library despite the mistake, which turned out to be fortuitous, as will be shown. The pNYS02 library had a theoretical size of 4914 members since the CH1 construct could be incorporated into each of the 273 insertion point with nine possible permutations of linker combinations. The third library, pNYS03, was a very large library that was constructed to test 1, 2 and 3 amino acid linkers of variable compositions connecting the CH1 to HSV-TK. The theoretical library size was ~141.6 million members.

The pNYS01 and pNYS02 libraries comprised ~200 000 transformants each. Considering that their theoretical library size was 546 and 4914 variants, respectively, the probability that each of these libraries is complete is essentially 100% (Bosley and Ostermeier, 2005). Transformants from pNYS01 displayed successful HSV-TK/CH1 fusions in 21 out of 25 of the transformants sequenced and 4 duplicates were found in the 21 transformants. Transformants from pNYS02 had insertions in 24 out of 30 members sequenced and showed no identical sequences within the collection. The pNYS03 library consisted of 11 million transformants for testing, which is only a fraction of the possible number of different sequences. Sequencing of these transformants indicated that 27 out of 35 transformants possessed a CH1 insertion into HSV-TK and that none of sequences in the collection were identical.

Selections and screens identify a switch

We subjected all libraries to a two-tiered selection designed to identify HSV-TK/CH1 fusion proteins that had their cellular kinase activity increase upon the expression of the C-TAD of HIF1α. (Fig. 2) (Freedman et al., 2002). An initial negative selection in E. coli KY895 cells identified variants that lacked AZT phosphorylation activity in the absence of C-TAD expression. Escherichia coli KY895 cells lack endogenous thymidine kinase (Tdk) activity and cannot metabolize AZT into its toxic form. However, library variants capable of phosphorylating AZT into its monophosphate form, AZT-MP, would be able to kill their host cells from AZT-MP toxicity. We subjected surviving members of this negative selection to the subsequent positive selection to identify variants that could phosphorylate deoxythymidine into dTMP in the presence of the C-TAD of HIF1α. All endogenous means of dTMP production were inhibited through the addition of 5-FdU. There are two enzymes in E. coli capable of producing dTMP, thymidine kinase (Tdk) and thymidylate synthase (ThyA) (Ahmad et al., 1998). As mentioned above, E. coli KY895 does not possess Tdk and 5-FdU directly inhibits ThyA. Thus, growth depends on the ability of the HSV-TK/CH1 fusions to produce dTMP.

Fig. 2.

Genetic selections. (A) The negative selection in E. coli KY895 identifies variants that lack kinase activity in the absence of the C-TAD of HIF1-α. Variants that possess activity in this condition produce toxic AZT-MP, while variants lacking kinase activity do not and thus survive. (B) The positive selection, also done in E. coli KY895, selects for variants with the ability to phosphorylate dT to dTMP in the presence of the C-TAD. Cells that are unable metabolize the necessary dTMP are killed.

Approximately 10 000 CFUs of cells harboring pNYS01 and pNYS02 and 10⁶ CFUs of cells harboring pNYS03 were plated on negative selection plates. Our negative selections displayed a cell survival rate of ~70%. The high survival rate in the negative selection is not surprising, since we expected that most insertions of CH1 into HSV-TK would compromise enzyme activity. In the subsequent positive selection in the presence of coexpressed C-TAD, 5000 CFUs/plate of the pNYS01 and pNYS02 libraries were plated, whereas 500 000 CFUs/plate were spread on positive selection plates for the pNYS03 library. These selections resulted in 64 colonies from the pNYS01 library (on one plate), ~300 colonies for the pNYS02 library (on two plates) and ~3600 colonies from the pNYS03 library (on four plates). Over 2400 hits from the positive selections of pNYS01, pNYS02 and pNYS03 were screened for a switching phenotype. We screened all hits from the pNYS01 library and ~60% of the hits from the pNYS02 and pNYS03 libraries by a liquid culture spotting assay where cells from the positive selection expressing the variant were spotted on negative selection plates with and without C-TAD being expressed (C-TAD expression was induced by the addition of arabinose to the plate). Plasmids from hits in this screen were retransformed into fresh KY895 cells and retested by the spotting assay.

While no protein switches were found in pNYS01, one HSV-TK/CH1 fusion from each of the pNYS02 and pNYS03 libraries reproducibly showed increase AZT susceptibility with coexpression of C-TAD of HIF1α. Both constructs shared the same amino acid sequence (Fig. 3a) but differed at the nucleotide level. The degenerate NNK codon linkers utilized in pNYS03 could not have produced the DNA sequence found in pNYS02, thus ruling out cross-contamination between the libraries as an explanation. Based on DNA sequencing, the CH1 domain is inserted after the 150th amino acid of HSV-TK (Fig. 3a). This insertion site is part of an unstructured loop in HSV-TK (Fig. 3b). The N-terminus of the CH1 was connected to the HSV-TK enzyme through a Gly-Gly-Ser linker, while the C-terminus was connected using a Ser-Thr-Thr linker. We were struck and encouraged by the fact that both the pNYS02 and pNYS03 libraries resulted in selection of the same protein and that the selections performed on the third library (with randomized linkers) resulted in selection of the same STT C-terminal linker that was an ‘error’ in the design of the second library. We designated this switch TICKLE for ‘Trigger-Induced Cell-Killing Lethal-Enzyme’.

Fig. 3.

TICKLE switch. (A) Sequence of TICKLE. TICKLE comprised a CH1 domain insert in between the 150th and 151st residues of HSV-TK. The N-terminus of the CH1 is connected to HSV-TK by a Gly-Gly-Ser linker and the C-terminus is connected by a Ser-Thr-Thr linker. (B) Structure models of HSV-TK (PDB #1KIM (Champness et al., 1998)) adjacent to CH1 with the C-TAD of HIF1-α bound (PDB #1LE3 (Freedman et al., 2002)). Space-filled spheres indicate the sites of fusion (i.e. insertion) of CH1 into HSV-TK. The insertion site in HSV-TK (between residues 150 and 151) is within an unstructured loop (residues 148–153) in the crystal structure, thus residues 147 and 154 are indicated instead. Thymidine (dT) is shown in the active site of HSV-TK.

TICKLE switch renders cells AZT sensitive in a C-TAD-dependent manner

AZT toxicity assays were conducted in solid and liquid media to determine the extent to which expression of C-TAD of HIF1-α caused cells expressing TICKLE to be sensitive to AZT. As we expressed the C-TAD as a fusion with glutathione S-transferase (GST), we expressed GST alone as the negative control. In the spot assay, serial dilutions of cells were challenged to grow in the presence of AZT. For cells expressing TICKLE, coexpression of C-TAD-GST made the cells more susceptible to AZT compared with coexpression of GST (Fig. 4a). The liquid assay evaluated the relative growth of E. coli KY895 containing TICKLE by OD, in the presence of different levels of AZT. The presence of C-TAD-GST caused a 4-fold increase in AZT toxicity relative to GST (Fig. 4b). Regardless of whether C-TAD is expressed, TICKLE is inferior to HSV-TK at making the cells susceptible to AZT (Fig. 4b). Thus, the insertion of the CH1 domain as found in TICKLE, greatly compromises the ability of the HSV-TK to confer AZT sensitivity to cells, but this toxicity can be partially restored to higher levels by the presence of C-TAD. Presumably, this increased AZT toxicity results from the C-TAD binding to the CH1 domain of TICKLE and either increasing its specific enzyme activity or increasing its cellular abundance.

Fig. 4.

AZT toxicity assay. (A) C-TAD coexpression increases the AZT toxicity of E. coli KY895 cells expressing TICKLE. An equal number of cells expressing TICKLE + GST or TICKLE + C-TAD-GST were spotted on negative selection plates containing 10 μg/ml AZT and incubated 18 hours at 37°C. The different spots represent a series of 1:2 serial dilutions of the culture, with the highest cell concentration at the right. (B) Escherichia coli KY895 cells expressing the indicated proteins were challenged to grow in liquid media for 6 hours in the presence of different concentrations of AZT. The relative OD₆₀₀ normalizes the OD₆₀₀ to that in the absence of AZT. Circles, TICKLE + C-TAD-GST; squares, TICKLE + GST; triangles, HSV-TK (positive control) and diamonds, inactive HSV-TK (negative control). (C) Circles, TICKLE (T244S) + C-TAD-GST; squares, TICKLE (T244S) + GST; triangles, TICKLE (S243G, T244G, T245S) + GST; diamonds, TICKLE (S243G, T244G, T245S) + C-TAD-GST. Error bars represent the standard deviation (n = 3).

We also examined the importance of the STT sequence in the C-terminal linker of TICKLE by mutating it back to the intended GGS linker of the second library (dubbed TICKLE (S243G, T244G, T245S)) as well as mutating it to SST (dubbed TICKLE (T244S)). AZT toxicity assays showed that the STT linker was important if not vital to TICKLE's switch function. TICKLE (T244S) caused only a 2-fold switching effect, compared with the 4-fold effect seen in wild-type TICKLE (Fig. 4c). The T244S mutation decreased switching by increasing AZT toxicity in the absence of C-TAD. Replacing the STT linker with the intended GGS linker not only abolished switching but also eliminated TICKLE's ability to make the cells sensitive to AZT (Fig. 4c).

TICKLE switch requires C-TAD coexpression for rescuing cells deficient in dTMP production

To further confirm TICKLE functioned as a switch, we tested its ability to rescue growth of cells deficient in dTMP production. dTMP is essential to cell survival. A conditionally lethal cell strain (NS01) was created from the tdk knock out strain KY895. NS01 has thyA deleted from the chromosome and a plasmid containing thyA under the control of the rhamnose-inducible promoter system. In the presence of glucose (and absence of rhamnose), the inability of NS01 to produce sufficient dTMP makes the strain non-viable, but growth can be rescued by the addition of rhamnose (Fig. 5a). In the absence of rhamnose, growth could be rescued by expressing HSV-TK (Fig. 5b).

Fig. 5.

dT growth assay. (A) Escherichia coli NS01 is a conditional lethal strain that requires rhamnose for growth due to a deficiency in dTMP production. One thousand CFU cells were plated on tryptone media plates with 25 μg/ml kanamycin and containing 0.2% glucose (Glu) (top; non-permissive condition) or 0.2% rhamnose (Rha) (bottom; permissive condition) and incubated 18 hours at 37°C. (B) Escherichia coli NS01 expressing the following proteins were grown in non-permissive liquid media for 8 hours in the presence of different concentrations of dT. Circles, TICKLE + C-TAD-GST; squares, TICKLE + GST; triangles, inactive HSV-TK (negative control) and diamonds, wild-type HSV-TK (positive control). The relative OD₆₀₀ was normalized to the OD₆₀₀ of the positive control at 50 μg/ml dT. Error bars represent the standard deviation (n = 3). Points that appear to be without error bars result from errors that are smaller than the size of the symbol.

We utilized this strain in a growth assay that tested TICKLE's ability to generate dTMP in a C-TAD-dependent manner. NS01 cells expressing TICKLE and C-TAD-GST or GST were challenged to grow under non-permissive conditions at different concentrations of dT. TICKLE was expressed from the pMCC plasmid while C-TAD-GST and GST were expressed from pSkunk2. TICKLE allowed the cells to grow, but only if C-TAD was expressed (Fig. 5b). Cells not expressing the C-TAD showed a consistently low OD regardless of dT concentration.

C-TAD increases TICKLE cellular abundance

Immunoblotting was used to better understand the switching mechanism behind TICKLE causing increased AZT toxicity in the presence of the C-TAD. One possibility is that C-TAD binding to TICKLE increases TICKLE's cellular abundance, thus increasing the total AZT kinase activity in the cell. Such a mechanism has been observed with previous switches built by domain insertion (Heins et al., 2011; Wright et al., 2011) and can result from the binding event increasing the thermodynamic or proteolytic stability of the switch (Choi et al., 2013). Cultures of E. coli KY895 expressing TICKLE with and without C-TAD coexpression were incubated overnight. We lysed an equal number of cells from the cultures (based on OD) and examined the soluble fraction of the lysate by western blot using anti-HSV-TK antibodies (Fig. 6). As in previous studies (Garapin et al., 1981; Waldman et al., 1983), HSV-TK expression in E. coli resulted in a multiple bands, the smaller of which are presumably proteolytic products. Expression of TICKLE resulted in higher weight bands as expected by the addition of the CH1 domain, but the intensity of the higher weight bands (especially the largest size band at ~53 kDa, the expected size of TICKLE) was much higher when C-TAD was coexpressed.

Fig. 6.

Relative accumulation of TICKLE as a function of C-TAD coexpression. A western blot using anti-HSV-TK antibodies was performed on an SDS-PAGE gel of the soluble fraction from an equal number of KY895 cells expressing the indicated proteins. Lane 1, HSV-TK; Lane 3, TICKLE + C-TAD-GST; Lane 4, TICKLE + GST; Lane 5, no heterologous protein expressed. Lane 2 is the molecular weight marker.

These results are consistent with the following model of TICKLE's mechanism. In the absence of C-TAD, TICKLE is an unstable protein and is rapidly degraded. In the presence of C-TAD, its binding to the CH1 domain of TICKLE decreases the rate of TICKLE's turnover in the cell, causing a higher level of cellular HSV-TK activity. We believe this explanation is the simplest and most likely one; however, we have not yet directly interrogated whether there is a binding event between TICKLE and C-TAD. The presence of bands smaller than TICKLE but larger than those of HSV-TK suggest a reason for why TICKLE is not a better switch than it is. These proteolytic products may lack the ability to bind CH1 but retain HSV-TK enzyme activity that would be unregulated by C-TAD.

Discussion

The development of TICKLE as an enzymatic switch establishes a method to modulate the cellular HSV-TK activity by the expression of the C-TAD of HIF1-α. Non-target toxicity during cancer therapy is a major obstacle in increasing dosage levels of drug. TICKLE's net level increases in the cell when in the presence of the C-TAD of a cancer marker, potentially providing a way to selectively activate prodrugs in cancer cells. The ‘Off’ state of the enzymatic switch diminished levels of the enzyme when compared with the ‘On’ state. The 4-fold increase in toxicity that is generated by the presence of the C-TAD is believed to be due to a stabilizing effect that occurs in the presence of the C-TAD that leads to greater accumulation of the enzyme. Studies have shown that the CH1 domain exists in a molten globule state and transitions into a structured state after binding to the C-TAD (Freedman et al., 2002). This molten globule state of CH1 may cause TICKLE to be cellularly unstable, leading to increased degradation by proteases and lower net cellular activity toward AZT in bacteria. The binding of the C-TAD would stabilize the CH1 domain and TICKLE, which would lead to higher accumulation of this protein in the cell. This mechanism is supported by western blot analysis of TICKLE.

The protein accumulation mechanism has been seen with protein switches designed in the past using the CH1 as the signal-recognition domain (Wright et al., 2011, 2014). These previous studies concerned the development of protein switches using yCD as the output domain and showed effector-enabled toxicity in both bacterial and mammalian systems. The yCD switch was developed using the same general protocol of subjecting bacterial libraries containing yCD/CH1 fusions to selection pressures to identify ‘on’ and ‘off’ states of a potential switch. The switches found through this process had switching properties in bacterial and mammalian systems. CD, in combination with 5-FC, is prominently used in GDEPT studies and its synergy with HSV-TK and GCV in producing additional toxicity together is well documented (Rogulski et al., 1997; Aghi et al., 1998; Willmon et al., 2006). While the efficacy of TICKLE in mammalian systems needs to be tested, an HSV-TK-based protein switch could be used in combination with a yCD-based protein switch to produce a synergistic effect. HSV-TK's broad substrate specificity is key to this synergy; it allows faster metabolism of substrates activated by yCD and may, in fact, result in higher toxicity levels produced directly from 5-FC (Longley et al., 2003). This broad specificity may also exist in TICKLE.

Translating TICKLE's switching phenotype in cancer cells may be a challenge due to differences in proteolytic mechanisms and switch degradation rates between mammalian and bacterial cells. The TICKLE gene will have to be optimized for expression in human cells and expressed such that it is in the same cellular compartment as HIF-1a. However, none of these issues presented a barrier in previous work utilizing the yCD/CH1 fusion switches (Wright et al., 2014). This work showed that genes encoding accumulation-based, hypoxia sensing switches can be appropriately altered and transferred to human cells where they function as switches by the accumulation-based mechanism.

Beyond implications in cancer therapy, HSV-TK has also been investigated for its potential as a reporter gene in animal models and clinical studies (Heyman et al., 1989; Gambhir et al., 2000; Yaghoubi and Gambhir, 2006). The kinase, in combination with 2′-[18F]fluoro-5-ethyl-1-beta-D-arabinofuranosyluracil (18F-FEAU), has been utilized in positron imaging tomography. The difference in activity levels based on HIF1α availability in TICKLE would also make the switch an attractive candidate to be tested for the purpose of medical imaging and may be able to identify hypoxic environments in animal models more efficiently.

Mutational analysis of the linker regions of TICKLE indicated that the linker composition plays an important role in the protein's ability to behave as a switch. As seen with TICKLE (S243G, T244G, T245S), mutations to the linker abolished almost all activity regardless of the presence of C-TAD. A construct consisting of single point mutation in TICKLE (T245S) at the linker region resulted in reduced activity in the ‘on’ state and decreased switching efficiency by 2-fold. These results underscore the influence of the linkers in sensitizing TICKLE to an effector molecule. The sensitivity of the linker to mutation suggests that linker optimization might lead to a second generation TICKLE switch that has superior switching properties. A combinatorial linker library combined with our two-tiered selection could be used for such optimization.

The addition of the thymidine analog, 5-FdU, to inhibit ThyA for positive selection used to identify TICKLE provides added stress to cells through its genotoxic effects after 5-FdU phosphorylation. Our E. coli NS01 strain has potential for use in an alternative positive selection method that avoids the use of 5-FdU to select for TICKLE variants with higher ‘on’ states. This method would be a more direct means of causing dTMP deficiency within cells without the use of toxic nucleoside analogs. This effectively removes an activity ‘cap’ placed on TICKLE variants during the original positive selection and may allow identification of switches with greater activity in their ‘on’ states. Switches with better ‘off’ can be selected by increasing the concentration of AZT.

In addition, the binding promiscuity of CH1 must also be addressed. The CH1 domain has binding affinity toward other transcription factors besides HIF1-α (Newton et al., 2000; Wright et al., 2014). Other binding partners might also activate TICKLE. We have previously shown that a CH1-yCD protein switches created in our laboratory can be activated by CITED2 in E. coli, and we suspect that activation of this switch in Flp-In 293 human cells in the absence of HIF1-α accumulation might result from binding of some other protein besides HIF1-α (Wright et al., 2014). However, while the binding of transcription factors other than HIF1-α may result in the activation of TICKLE, the main binding partners of CH1 are upregulated in cancerous environments and may still induce increased toxicity toward cancer cells (Bai and Merchant, 2007; Yu et al., 2009; Wright et al., 2014).

Conclusion

We have established a kinase-based proteins switch that is capable of generating increased prodrug toxicity toward E. coli cells in the presence of the C-TAD of HIF1-α. The selective toxicity based on the presence of hypoxia-induced elements makes TICKLE an appealing candidate for future investigation in mammalian cancer systems. While the enzymatic switch may not yet have sufficient differences between its ‘on’ and ‘off’ states and may suffer from effector specificity issues, further rounds of directed evolution may solve these issues. With the broad substrate specificity provided by the HSV-TK domain, the enzymatic switch may be capable of activating an array of substrates in a switch-like fashion and could be utilized for a variety of applications.

Funding

National Institute of General Medicine at the National Institutes of Health [grant numbers R01 GM066972, T32 GM080189].