A guide to designing photocontrol in proteins: methods, strategies and applications

Drug development

The main goal of photopharmacology is to develop smart drugs that can be switched on and off by irradiation with light to tackle off-target toxicity. This area of application is quite comprehensive as it does include all stages of drug design meaning early precursor development towards a specific target protein, analysis of the mode of action in vitro and/or in vivo, and refinement of the drug. Moreover, light-responsive compounds are often used to study the molecular basis of a disease.

A prerequisite of photopharmacological drug design is to choose target proteins that are associated with a localized disease where the delivery of light and, hence, the spatiotemporal control by light is feasible. The target proteins must further be specific for the cells that are the cause of the disease, e.g., overproduced proteins in tumor cells or proteins that solely exist in bacteria. While examining the efficacy of the light-responsive drug it became apparent that the difference in activity prior to and after irradiation is more powerful in vivo than in vitro (Bieth et al. 1969; Wainberg and Erlanger 1971). I will use the term light regulation factor (LRF) to describe this difference in activity from now on. (Other publications use the term dynamic range for this factor.) For example, while an azobenzene-based drug shows only an LRF of 1.5-fold in vitro, it might cause a drastic phenotypic effect such as reduction of cell growth in vivo. This can be explained by the strong connectivity of proteins in a cell; small in vitro effects can be amplified especially when the target protein is part of a signaling cascade. Another important issue is that many synthetic light-responsive compounds are designed to permeate the cell-membrane (Mentel et al. 2011), which results in the local entrapment and increased concentration of the drug inside the cell. Thus, targeting of intracellular proteins is feasible, as well. Before a photopharmacological drug can be tested in a clinical study, however, its behavior in vivo requires critical analysis. The drug should neither be metabolized nor rapidly degraded (pharmacokinetics) and it should not demonstrate toxicity (pharmacodynamics).

One of the most pertinent diseases that is addressed by photopharmacological drugs is cancer. Solid tumors are highly localized and constitute an excellent target for the light-responsive compounds. Moreover, current therapies, first and foremost chemotherapy, lead to severe systemic side effects that have an enormous impact on the health and well-being of patients (Edwards and Aronson 2000). Hence, the field of photo-induced anticancer medication targeting, e.g., kinases or microtubule dynamics, is constantly growing (Dunkel and Ilaš 2021). A current example for kinase inhibitors are the azobenzene-based azoaxitinibs designed against the vascular endothelial growth factor receptor (VEGFR)-2 kinase, which show a more than 40-fold difference in their inhibitory strength upon irradiation (Heintze et al. 2020). Furthermore, a study recently showcased the attractiveness of the rarely used photoswitch styrylbenzothiazole for the photocontrol of the microtubule cytoskeleton as it is metabolically stable compared to previously designed azobenzene-based inhibitors (Gao et al. 2021). Another highly promising approach for photopharmacological cancer treatment is the photocontrol of proteolysis-targeting chimeras (PROTACs), which induce the degradation via the ubiquitin-proteasome pathway of pathogenic proteins by mediating binding of the respective protein of interest with an E3 ligase (Dunkel and Ilaš 2021; Zeng et al. 2021). A very successful example thereof is the development of two approaches, in which PROTACs were coupled with photocaging groups (Naro et al. 2020).

Next to cancer, local bacterial infections are a prominent target of photopharmacology. Here, photocontrol can help to tackle the build-up of antibiotic resistance in the environment and to prevent the attack of symbiotic bacteria in the human microbiome by developing drugs, which only exhibit an antibiotic activity in one of the two isomers. To this end, we recently developed a novel antibiotic precursor against Salmonella typhimurium tryptophan synthase, a key enzyme in the metabolism of bacteria. The compound comprises an azobenzene photoswitch, which inhibited tryptophan production reversibly and in an allosteric manner (Simeth et al. 2021). An example of irreversible photoactivation with caged antibiotics is the synthesis of vancomycin and cephalosporin derivatives, which were active against five pathogenic bacteria in a light-dependent manner (Shchelik et al. 2021). Notably, the two isomers not only cause different effects in bacteria, but they also evoke different resistance mechanisms than the initial light-unresponsive drug as could be shown with tetra-ortho-chloroazobenzene-trimethoprim conjugates and the antibiotic trimethoprim in the Escherichia coli strain CS1562 (Lauxen et al. 2021).

Other current examples include the study of neural networks and associated disorders by targeting GABA and nicotinic acetylcholine receptors (Bregestovski and Ponomareva 2021), glycine receptors (Maleeva et al. 2021) human cannabinoid receptors (Rodríguez-Soacha et al. 2021), G protein-coupled receptors (Donthamsetti et al. 2021) and muscarinic receptors (Barbero-Castillo et al. 2021). Moreover, azobenzene-based ligands were designed to analyze the distinct functions of the cryptochrome CRY1 in the circadian clock (Kolarski et al. 2021a).

Functional studies on proteins

Synthetic ligands have a long history as tools for the study of the function of proteins, particularly to investigate enzyme catalysis. Hence, it is not surprising that light-responsive ligands can also be used for fundamental research on proteins. Nevertheless, hitherto there are only rare examples, which primarily set out to investigate dynamics and allostery of proteins. It should be noted that I will use the term “dynamics” to summarize all kind of motions spanning from fast, sub-ms fluctuations to slow conformational changes in the ms and s time range. 20 years ago, one of the first examples demonstrated that protein conformation can be modulated with light by conjugating an azobenzene to an α-helix using two cysteine or homocysteine residues at positions i and i + 7 and two reactive moieties flanking the azobenzene core (Kumita et al. 2000, 2002). Through computational-based engineering of the ligand they achieved to control the stability of an α-helix in the target peptide; while the Z isomer sustained the α-helical structure, the E isomer led to its destabilization and degradation (Figure 5A). Although they did not apply this technique to a protein determining the effect on ligand binding or enzymatic turnover, they set the ground for further developments. Only recently, this strategy was revived to photocontrol the binding of a peptide ligand to the PDZ3 domain of the postsynaptic density-95 protein (Bozovic et al. 2020). The authors targeted an α-helix, which is localized outside the active site, but which has been shown to affect peptide binding nonetheless (Petit et al. 2009), and accomplished a temperature – as well as ligand-dependent LRF for peptide binding of 120-fold. They further proposed a thermodynamic cycle of ligand binding and photo-induced conformational change (Bozovic et al. 2020) as well as determined the speed of the signal transmittance from the α-helix to the active site (Bozovic et al. 2021) using extensive biophysical studies. Although an effect originating outside the active sites is nowadays often called allosteric, allostery is actually traditionally defined as the control of protein activity in the orthosteric site by binding of a ligand to an allosteric site (Liu and Nussinov 2016). Hence, our groups addressed the photocontrol of true allostery by designing a photoswitchable diarylethene-based inhibitor for imidazole glycerol phosphate synthase (Kneuttinger et al. 2018). In this model enzyme for allostery, ligand binding to one subunit activates the enzymatic activity of the second subunit over a distance of ∼25 Å (Douangamath et al. 2002). By mimicking the allosteric ligand using a diarylethene, we could not only competitively inhibit the catalytic turnover of the first subunit (HisF), but also control the activity of the second subunit (HisH) with the same efficiency by light (Figure 5B). Finally, bioconjugation of an azobenzene derivative to a lipase was used to explore the effect of photo-induced isomerization on protein dynamics and ultimately enzyme activity (Agarwal et al. 2012). By covalent attachment of the photoswitch, they cross-linked two surface loops of the lipase (Figure 5C), which they previously identified computationally to be essential for protein dynamics. While the cross-linking efficiency was unfavorably low, the catalytic activity of the enzyme was significantly increased compared to the wildtype enzyme and could be further enhanced by light. These examples demonstrate that small molecular light-responsive compounds are highly promising for functional studies of proteins. This trend is even more pronounced for the next photocontrol method.

Figure 5:

Exemplary photoswitch applications.

(A) Bioconjugation of an azobenzene inhibitor to two cysteine residues (blue) of an alpha helix result in destabilization of the secondary structure in the E configuration (left) while maintaining it in the Z configuration (right) (Kumita et al. 2000). (B) Photocontrol of allostery in the HisF:HisH bienzyme complex. While the open diarylethene (triangle) prohibits binding of the allosteric ligand (barbell), the closed diarylethene (rectangle) loses affinity towards the HisF active site and facilitates binding of the allosteric ligand. Binding of the allosteric ligand initiates the glutaminase reaction in HisH (Kneuttinger et al. 2018). (C) Bioconjugation of an azobenzene inhibitor to two surface loops via a cysteine (blue) and a lysine (green) residue mediates the photocontrol of lipase dynamics and as a result activity (Agarwal et al. 2012). Bold: Photoswitch cores as represented in Figure 4.

Tool development for in vivo fundamental studies

Just as for the other two photocontrol methods, the main focus of photoxenoprotein engineering lies currently within the application in vivo. For this purpose, many proof-of-principle studies have light-regulated key players of the cell to use as tools in future fundamental studies. While photoreceptors in optogenetics can be encoded relatively straightforward into the genome of organisms, the introduction of the synthetic machinery required for genetic code expansion has faced two major challenges (Chen et al. 2017a). ncAAs are no native substrate and, hence, the delivery to the target tissue in the model organism can be difficult to accomplish. The best strategies so far are direct injection into the tissue (Han et al. 2017; Kang et al. 2013) or supplementation in the drinking water (Chen et al. 2017b; Ernst et al. 2016; Liu et al. 2017a). The second challenge is the transfer of the DNA encoding the aaRS/tRNA pair and the mutated protein of interest into the genetic material of the organism. To this end, various vector systems have been introduced in utero by e.g., electroporation, such as plasmids (Kang et al. 2013) and vectors of viral origin, e.g., lentiviral (Chen et al. 2017b; Han et al. 2017). As a result, many model organisms could be modified to express ncAA-bearing proteins including mice (Chen et al. 2017b; Han et al. 2017), zebrafish (Chen et al. 2017b; Liu et al. 2017a), and fruit fly (Bianco et al. 2012).

Over the years, multiple studies developed and incorporated particularly photocaged ncAAs into the genome of eukaryotic cells or even organisms. In contrast, only one proof-of-principle example exists for the use of photoswitchable ncAAs in eukaryotic cells (Luo et al. 2018). The general success of these studies has been outlined comprehensively in various review articles (Ankenbruck et al. 2018; Baker and Deiters 2014; Bardhan and Deiters 2019; Courtney and Deiters 2018; Deiters 2018). While some studies use standard model proteins to test the ncAAs such as luciferase (Luo et al. 2017a, 2018; Uprety et al. 2014) or tobacco etch virus (TEV) protease (Luo et al. 2017a; Nguyen et al. 2014), multiple studies targeted various types of key proteins starting with enzymes for gene editing. The first example thereof is the spatiotemporal control of DNA recombination in mammalian cells by Cre recombinase, which has been rendered light-sensitive by incorporation of a photocaged tyrosine into the active site (Edwards et al. 2009). A later study exchanged a critical lysine residue with a coumarin-protected lysine that can be decaged with visible light and tested the photocontrolled recombination in the developing zebrafish embryo (Brown et al. 2018). An important success in this direction is the photocontrol of the CRIPR/Cas9 gene editing tool. Incorporation of a caged lysine into key positions for guide RNA and DNA binding rendered the Cas9 enzyme catalytically inactive and allowed for the photocontrolled as well as site-specific gene deletion of a transmembrane receptor (Hemphill et al. 2015). Another example of photocaged DNA/RNA processing enzymes is T7 RNA polymerase, in which NBY (Chou et al. 2010) and mNPK (Hemphill et al. 2013) were incorporated at positions essential for discrimination of ribose and deoxyribose substrates as well as binding of nucleoside triphosphates. Similarly, Taq DNA polymerase was photocontrolled with a photocaged tyrosine (Chou et al. 2009). Moreover, photocontrol of zinc-finger nuclease was achieved by inserting a photocaged tyrosine in the DNA binding site to induce a DNA double-strand break and subsequent homologous recombination upon irradiation (Chou and Deiters 2011). To ease the study of diseases, which are caused by a defect in DNA helicases, another study applied coumarin-caged lysine in the ATP binding site of the DNA helicase UvrD (Luo et al. 2017b). Besides gene editing, photocontrol of proteins involved in cell signaling networks is highly desired. As such, kinases are frequent targets. A prominent example is MAP kinase kinase 1 (MEK1), which has been put under photocontrol by incorporating caged lysine variants into a position essential for ATP binding. Light-dependent activation of the MEK/ERK pathway has then been attained in human embryonic kidney (HEK)293ET cells (Gautier et al. 2011) as well as in zebrafish embryos (Liu et al. 2017a), by which the effects of ERK phosphorylation during different developmental stages could be observed. A recent study has refined this strategy for gain-of-function studies on kinases in cells (Wang et al. 2019a). They introduced an orthogonal MEK1, which they photocontrolled by incorporating NBY in proximity to the active site, and at the same time turned off the endogenous MEK1 counterparts using kinase inhibitors. Since the orthogonal MEK1 was rendered resistant to those inhibitors through the mutation to tyrosine after light-induced decaging, they generated an experimental setup to distinctively switch MEK1 function on and off. Besides MEK1, photocaging has also been successfully applied to Janus kinase (Arbely et al. 2012) and adenylate kinase (Zhou et al. 2020). Further cellular processes have also been made accessible for photocontrol through photocaged ncAAs including nuclear translocation (Gautier et al. 2010), the GTPase activity of KRAS (Wang et al. 2019a) and apoptosis through regulation of caspase-3 activity (Wang et al. 2019a).

The here presented examples and many more compile a large toolbox of especially photocaged key proteins in vivo, which is powerful for spatiotemporally resolved studies in various biological fields. Compared to hybrid protein optogenetics and photopharmacology, the method of light-responsive ncAAs appears to be more versatile, as this technique is highly efficient not only for in vivo studies but also for diverse other applications, which I will elaborate in the following chapters.

Functional studies on proteins

Over the last decades, mutational analysis of proteins has become routine in biochemistry, in particular, for the analysis of protein function. For this, specific amino acid residues, which are considered to play a certain role, are substituted with one of the other 19 canonical amino acids in site-directed mutagenesis. Moreover, using ncAAs instead of canonical amino acids has widened the scope of functionalities that can be introduced into proteins (Agostini et al. 2017). It is therefore not surprising that site-directed incorporation of light-responsive ncAAs is used to analyze protein function, as their key benefit is the temporal resolution of the introduced effect.

Already in 1998, the photocaged tyrosine NBY was applied in a nicotinic acetylcholine receptor to investigate the influence of multiple highly conserved tyrosine residues on pore opening upon binding of acetylcholine (Miller et al. 1998). Through time-resolved kinetic analyses they found that each residue affected the electrochemical potential differently, which ultimately resulted in two activation phases of ion channel opening. Another study, almost a decade later, used photocaged serine to analyze the function of the transcription factor Pho4 (Lemke et al. 2007), which is exported from the nucleus in response to increased concentrations of inorganic phosphate in yeast (Komeili and O’Shea 1999). By exchanging the two phosphorylation sites S2 and S3 separately with photocaged serine, they could show that phosphorylation of S2 has slower export kinetics after irradiation than phosphorylation of S3. A good example of how photocrosslinking ncAAs can be utilized to investigate protein:ligand interactions was given by the investigation of serotonin transporters (Rannversson et al. 2016). By incorporating AzF at positions surrounding two distinct binding sites, S1 and S2, they substantiated that inhibitor binding solely occurs at the S1 site. Moreover, AzF has been employed for the investigation of the relationship between dynamics and activity in an n-methyl-d-aspartate (NMDA) receptor (Zhu et al. 2014). The gating mechanism of these receptors relies on conformational transitions at the interface between the n-terminal domains of the GluN1 and GluN2 subunits (Mony et al. 2011; Zhu et al. 2013), which have not been clearly understood at the time of the publication. The authors introduced AzF at this interface into a position, which has been known to influence the sensitivity of the receptor to allosteric effectors. Upon irradiation, the formation of a crosslink between the ncAA and a nearby loop put the receptor into a low-activity mode. This inactivation unambiguously showed that the flexibility of the GluN1 loop is decisive for the gating mechanism. The same group later achieved to use the same strategy to distinguish the mechanism of allosteric regulation by binding of Zn²⁺ and the small molecule ifenprodil (Tian and Ye 2016). They positioned AzF strategically at the interface of GluN1 and GluN2 close to the binding site of ifenprodil. Light-induced crosslinking of the two subunits led to the inactivation of ifenprodil inhibition. However, not ifenprodil binding but instead the transmittance of the allosteric signal from the binding site to the gate was interrupted and caused this inactivation. Since the inhibition by Zn²⁺ was not affected, they concluded that Zn²⁺ and ifenprodil follow differential structural pathways for inhibition. In another follow-up study they chose FC-AzoF to induce conformational changes in each structural domain of the NMDA receptor subunits by light-induced switching of the azobenzene moiety without making use of the click function (Klippenstein et al. 2017). Irradiation and further analysis of receptor function revealed that photoswitching especially influenced the agonist-binding domain leading to decreased gating efficiency and agonist affinity as well as the transmembrane domain hampering the permeation properties of the channel. This study nicely demonstrated that photoswitchable ncAAs can reversibly control protein activity through a light-induced conformational change. Our subsequent studies also follow this strategy aiming to light-regulate allostery in the prominent model enzyme complex imidazole glycerol phosphate synthase, which undergoes large-scale conformational rearrangements upon allosteric ligand binding (Wurm et al. 2021). To this end, AzoF as well as the photocaged ncAAs NPY and mNPK were incorporated at positions close to sites involved in these conformational rearrangements and thus important for allostery (Kneuttinger et al. 2019a). As a result, it was possible to photocontrol the allosteric activation of one subunit by the other. Furthermore, molecular dynamics (MD) simulations demonstrated that the interface configuration as well as the conformational arrangement of a catalytic residue were hampered when the transmittance of the allosteric signal was interrupted by the ncAA. In a second study, we deliberately exchanged a key residue in the subunit interface with AzoF to answer the long-standing question whether a closing of the interface occurs during the allosteric activation process (Kneuttinger et al. 2020). In fact, biochemical and X-ray crystallographic studies showed that AzoF incorporation facilitated the photocontrol of the interface configuration and that a closing increases enzyme activity implying that this conformational change is part of the allosteric signal.

Although the number of publications that use light-responsive ncAAs for functional studies on proteins are currently modest, the great potential of these time-resolved mutational tools promises that more studies might follow.

Therapeutic approaches

Similar to photopharmacology, photoxenoprotein engineering has the potential to broaden the efficacy of therapeutic approaches. To this end, light-responsive ncAAs can be used to add spatiotemporal control to a protein-based drug such as monoclonal antibodies (Raquel et al. 2021) or therapeutic enzymes (Yari et al. 2017). Similar to the fundamental idea of photopharmacology, photocontrol of these so-called biotherapeutics allows to activate them only at the target site and prevents harmful side effects.

Two recent studies have set out to apply this principle to antibodies. The therapeutic effect of antibodies relies on the blockage of cell surface receptors or the delivery of cytotoxic drugs to target cells. Since often the targeted receptors e.g., the epidermal growth factor receptor (EGFR) in cancer cells (Normanno et al. 2006), are also present on the surface of healthy cells, the therapy with antibodies can cause severe side effects (Hansel et al. 2010). Thus, the first study redesigned a so-called nanobody, which is a class of single domain antibody fragments (Ingram et al. 2018), to photocontrol its binding to EGFR (Bridge et al. 2019). To this end, they incorporated NBY into the antigen binding region of the 7D12 nanobody. This resulted in a largely reduced binding strength of the nanobody-EGFR interaction, which could be restored to its usually high affinity upon irradiation. The second study further promoted the success of this approach. They established efficient photocontrol in an anti-GFP nanobody, an EgA1 nanobody targeting EGFR, and a 2Rs15d nanobody targeting the human epidermal growth factor receptor 2 (HER2) with photocaged tyrosines (Jedlitzke et al. 2020). Another recent study tested the potential of photocontrolled cytotoxic proteins for the treatment of cancer. They applied NBY to inactivate lethal factor LF, which can enter cells with the help of the protein-protective antigen PA. Both proteins are components of anthrax toxin. In high dosage, LF is toxic to cancer and healthy cells as it breaks down MAP kinases. However, photocaging of LF allowed to activate LF in a controlled manner by irradiation resulting in significant reduction of tumor growth and tumor size (Wang et al. 2019a).

Production of challenging peptides and proteins

Protein biochemists face various difficulties in the production of proteins through heterologous gene expression. The most prominent example is that the protein or peptide in question is toxic to the host cell. Since the production using bacterial or yeast strains are still often preferred as standard technique over e.g., cell free systems, these problems need to be circumvented. While it often helps to use specialized strains or optimized expression conditions (Rosano et al. 2019), the production of some proteins remains challenging. Isolated cases have successfully applied photocaged ncAAs to tackle exactly this issue. To this effect the activity of the proapoptotic cysteine protease caspase 3 was switched off by incorporation of a photocaged cysteine in the active site to allow for the expression in yeast host cells (Wu et al. 2004). Similarly, the expression of Staphylococcus aureus toxins YoeBSa1/2 in E. coli host cells turned out to be extremely difficult with traditional approaches e.g., coexpression of the antitoxin YefM (Larson and Hergenrother 2014). However, when the RNase activity was reduced by incorporation of NBY into a tyrosine position crucial for toxicity, the toxins could be produced and reactivated by irradiation after cell lysis. The third example is the spatiotemporal control of inteins, which are protein domains able to autocatalytically excise themselves from the exteins of a precursor protein in a splicing process (Di Ventura and Mootz 2019). Inteins are valuable tools in synthetic biology because they allow for the reconstitution of split proteins, the post-translational attachment of tags or the generation of therapeutically significant cyclic peptides or proteins. Since the production of such cyclic peptides entails challenging purification steps, a recent publication rendered the M86 intein light-sensitive by incorporation of NBY into an essential position close to the active site (Böcker et al. 2015, 2019). Photocontrol of the splicing reaction thus promises to ease the purification and to facilitate high-throughput directed evolution screenings. Moreover, challenges in the production of mussel-derived adhesive proteins (MAPs) were addressed using photoxenoprotein engineering. The adhesive properties of MAPs are highly beneficial for various applications, however, the proteins are difficult to produce recombinantly, as they comprise DOPA as key amino acid. While neither post-translational hydroxylation (Hwang et al. 2007) nor other production strategies obtained sufficient yields, incorporation and post-production irradiation of NB-DOPA turned out to successfully obtain reasonable yields of MAPs (Budisa and Schneider 2019; Hauf et al. 2017).

From biotechnology to industry

The spatiotemporal control of protein function is not only highly desirable for fundamental or medicinal research studies. Many biotechnological and industrial processes would benefit strongly from an orthogonal and precise reaction initiation system. This is particularly true for the field of synthetic biology, which strives to equip, reengineer or rebuild biological modules with unnatural, artificial functions (Ausländer et al. 2017) for various applications. One example is the bottom-up reconstitution of a synthetic cell. To this end, spatiotemporal control over cell division is regarded as very important. A key player for this goal is the filamenting temperature-sensitive mutant Z (FtsZ), which polymerizes into a division ring, similar to tubulin in eukaryotes, for the initiation of bacterial cell division (Romberg and Levin 2003). To control FtsZ ring formation by light, ONBY was recently introduced into a key tyrosine position that is essential for FtsZ assembly (Sun et al. 2021). As a result, the GTPase activity of FtsZ, which is required for polymerization, and, hence, ring formation was regulated with an LRF of 22-fold. Another recent biotechnological study has demonstrated photocontrol of elastin-like polypeptides (ELPs) (Israeli et al. 2021), which constitute a family of artificial protein-based biomaterials. The most crucial feature of ELPs is their lower critical transition temperature, which triggers a reversible soluble-to-insoluble phase transition valuable for various applications such as drug delivery (Bellucci et al. 2013) and tissue engineering (Christensen et al. 2009). Incorporation of AzoF into these biomaterials facilitated reversible external control of the transition temperature and, hence, of the phase transition itself.

Moreover, one of the currently most important goals that biological, biochemical and chemical research areas pursue is the development of more sustainable industrial processes. Already in 1998, a list of 12 principles of green chemistry was put together specifically for this endeavor (Anastas and Warner 1998). Within the plethora of research fields that strive to accomplish this goal, many use isolated proteins and especially enzymes for biocatalytic purposes. Owing to the rapid and vigorous development of bio-based strategies, more and more situations emerge, which require the spatial, temporal or spatiotemporal control of protein and especially enzyme function (Schmidt-Dannert and Lopez-Gallego 2016). A few first publications pave the way for the application of specifically photoxenoprotein engineering to fine-tune green chemistry approaches. One goal of biocatalysis is to achieve temporal control over the flux and dynamics of synthetic enzyme cascades (Schrittwieser et al. 2018), which resemble synthetic metabolic pathways. Thus, one of the first attempts to implement photocontrol in biocatalysis was the computationally guided creation of an artificial multienzyme complex from the enzymes of the synthetic pathway for the degradation of the toxic groundwater pollutant 1,2,3-trichloropropane that associates only upon phosphorylation and irradiation (Yang et al. 2017). For the light-induced complexation, the authors incorporated BzF into a redesigned Src homology 3 domain coupled to one enzyme subunit. Binding of this domain to its corresponding peptide fused with a second enzyme resulted in the conjugation of BzF with a methionine and, hence, covalent complex formation.

These and other publications (Claaßen et al. 2019; Schmermund et al. 2019) emphasize the potential of photocontrol in the fields of biotechnology, in particular synthetic biology, and industrial biocatalysis. Nevertheless, it remains to be seen whether photocontrol of proteins will prove to have high impact and which specific advantages photoxenoprotein engineering possesses for industrial applications.

In vivo fundamental studies

Optogenetics has been developed to regulate biological reactions in vivo. Specifically in the context of hybrid protein optogenetics, dynamic control of proteins of interest with light offers a convincing advantage: the effect of a modification can be observed in a spatially and timely resolved manner within the living system. This spatiotemporal control allows to better analyze and understand the effect of protein function and, hence, the dynamic character of cells, which comprises an intricate coordination of various input and output signals as well as actions and reactions. In comparison, static control approaches such as mutagenesis of a critical amino acid, overexpression or knockdown and knockout techniques provide comparatively little information on this dynamic environment. In this regard, the photoreceptor-based strategies offer many advantages. Primarily, the photosensor-tags are genetically encoded together with the target proteins, and they use light-sensitive cofactors, which are often naturally available within the cell, e.g., flavins for LOV domains. Moreover, some studies already accomplished the production of chromophores not inherent to the target system, e.g., PCB for phytochromes in animal cells, in vivo (Müller et al. 2013; Uda et al. 2017). Hence, hybrid protein optogenetics is widely developed and applied for the purpose of fundamental cellular and organismal studies. Even though engineering of hybrid proteins constitutes only a fraction of the massive publication activity in the field of optogenetics (>600 publications alone in 2021 according to Web of Science), the plethora of studies exceeds the limits of this general overview on photocontrol in proteins. Since other reviews comprehensively cover the many in vivo applications for fundamental studies (Kramer et al. 2021b; Losi et al. 2018; Oh et al. 2021; Repina et al. 2017; Tang et al. 2021), I herein concentrate on a brief overview on publications within the last year.

Hybrid protein optogenetic studies foremost concentrate on three cellular targets for the artificial control with light: signaling cascades, (translocation between) organelles and gene expression (Oh et al. 2021). Some of the most important players of cell signaling that constitute a target for photocontrol are phosphatases. To facilitate the regulation of tyrosine phosphatase 1B with light, three different strategies using LOV2 were compared in a very recent study, i.e., steric decaging, conformational change and heterotypic split protein association (cf. Figure 11B) (Hongdusit et al. 2022). The authors show that the conformational change strategy is particularly successful for this type of phosphatase allowing to analyze various tyrosine phosphatase related physiological processes in the cell. Another study focused on the intricately interconnected pathways of the two receptor tyrosine kinases ERK and AKT. To resolve the complexity of this signaling network they developed a two-way photocontrol system, which responds differently to blue and red/far-red light. While homo-typic clustering of cryptochrome 2 (CRY2) activated the AKT pathway, ERK signaling could be controlled with light through heterotypic association of phytochrome PhyB with its phytochrome interaction factor (PIF) (Kramer et al. 2021a). CRY2 was further used to assess the role of talin in the activation of integrins at the cell membrane. Talin was thereby recruited as a CRY2 hybrid protein to the plasma membrane through light-induced heterotypic interaction of CRY2 with a constitutively membrane-associated CIB protein (Liao et al. 2021). Their findings showed that binding of talin to the cell membrane additionally required the association of Ras-related protein 1 (Rap1)-GTP for the activation of integrines. Similarly, localization of the small GTPase RhoA, a key actuator in cytoskeletal function, was photocontrolled using the LOV domain from Botrytis cinerea (BcLOV4), which directly associates with lipids in the cell membrane (Berlew et al. 2021). Moreover, optogenetic systems have been established for the translocation of proteins to other organelles than the cell membrane. To shuttle proteins from the cytoplasm to the nucleus for example, LOV2 has been coupled to a nuclear export sequence, which is sterically decaged upon irradiation (Niopek et al. 2016). This so-called LEXY module was deployed in a recent study to investigate the function of the myocardin-related transcription factor A, which regulates actin-associated gene expression and promotes invasion and metastasis of tumors (Zhao and Grosse 2021). Besides such one-component systems, two-component systems can also be used for the analysis of nuclear import and export. To investigate the translocation rates of the transcriptional regulators YAP1 and TAZ, the authors of a recent publication tagged these proteins of interest with a fluorescent label as well as the synthetic peptide Zdk98 (Dowbaj et al. 2021). Zdk98 mediates light-dependent binding to LOV2, which was immobilized on the surface of mitochondria. Dissociation of the target proteins and, hence, nuclear import was then initiated by irradiation of this two-component system. Regarding the photocontrol of gene expression, the LOV domain from Rhodobacter sphaeroides was currently employed to inactivate the widely used Tet repressor by photoinduced dissociation of homodimers resulting in gene transcription (Dietler et al. 2021). Homotypic association/dissociation is a commonly used optogenetic strategy for the light-induced regulation of transcription. To facilitate this type of photocontrol with near-infrared light, bacterial phytochrome based photosensor modules were currently re-engineered to undergo homotypic oligomerization instead of heterotypic oligomerization. The resulting optogenetic transcription regulation system was called iLight (Kaberniuk et al. 2021).

Therapeutic approaches

Besides the fundamental research sector, optogenetics is also a promising therapeutic tool despite some limitations (Wichert et al. 2021). Ion channel based optogenetics pioneers this field; The most advanced example is the treatment of retinopathy by delivering a viral plasmid containing a microbial opsin directly to the eye in order to restore vision (Duebel et al. 2015). Nevertheless, optogenetic hybrid proteins are likewise developed for diverse therapeutic approaches. Current reviews give a comprehensive overview over present preclinical studies (Tang et al. 2021; Wichert et al. 2021) including the use of the cryptochrome system in Alzheimer’s disease (Lim et al. 2020), phytochrome as well as LOV based targeted bone regeneration (Hörner et al. 2019b; Wang et al. 2019b), and advancement of photoacoustic tomography using phytochromes for the improved identification of e.g., metastases (Li et al. 2018). These examples as well as two further topical approaches, described in the following, illustrate the wide range of medicinal applications for which photocontrol of proteins can be employed.

The first of these two approaches is the photocontrol of the smaller sized and disulfide-bond lacking single-domain counterparts of antibodies, nanobodies and monobodies. Using either the split protein (Yu et al. 2019) or the LOV2 based conformational change technique (Carrasco-López et al. 2020; Gil et al. 2020; He et al. 2021), nano- and monobodies targeting different fluorescent proteins, the Src Homology 2 domain of Abelson tyrosine kinase or actin, respectively, were engineered in the last years. The authors primarily show the advantage of their light-induced nano- and monobodies as so-called intrabodies for intracellular studies; however, they also point out the potential use for medicinal purposes, e.g., for anti-cancer treatment as outlined earlier for the light-sensitive ncAA-containing nanobodies in the photoxenoprotein engineering section.

Most remarkably, two optogenetic approaches have also been combined with smart electronic devices to facilitate a photocontrolled drug supply in mice (Mansouri et al. 2021; Shao et al. 2017). For this purpose, the production of human glucagon-like peptide-1 (hGLP-1), which is clinically licensed for the treatment of type-2 diabetes (Müller et al. 2019), was put under the control of a light-regulated transcription factor. While the first study uses the far-red light activated cyclic diguanylate monophosphate synthase BphS coupled to an endogenous signaling pathway (Shao et al. 2017), the second study has optimized a CarH (e.g., Figure 12) based two-component system that releases the transcription factor from the plasma membrane upon irradiation with green light (Mansouri et al. 2021). Moreover, while the first study requires a hydrogel implant containing the modified cells and a light-emitting diode (LED) array, the second study merely subcutaneously transplants the factory cells and then takes advantage of the green light functionality in smart wearable electronic devices, used to monitor health parameters, such as the Apple Watch (Raja et al. 2019). The authors of the combined hybrid protein optogenetics and smart watch study could further prove that their concept successfully treats diabetes and associated symptoms in mice (Mansouri et al. 2021).

Biotechnology

As in the case of photoxenoprotein engineering, some rare examples for the application of optogenetic systems outside fundamental or medicinal research exist, which focus on the advancement of biotechnological methods. A widespread application is to control the properties of diverse biomaterials with light (Beyer et al. 2018; Hörner et al. 2019b). A very recent publication, for example, engineered a light-responsive silk-elastin-like protein biopolymer with the help of CarH (Narayan et al. 2021). CarH tetramerization in the dark thereby mediated hydrogel formation that is reversible through light-induced monomerization of the photosensory domain. Another current study developed an innovative pluripotent stem cell culture system, which can be maintained by blue light-dependent stimulation of the fibroblast growth factor (FGF) signaling pathway (Choi et al. 2021). This obviates the need to supplement stem cell cultures with expensive FGF 2. For this purpose, the authors used the LOV domain of Vaucheria fridida to artificially dimerize the FGF receptor upon irradiation. Other examples include the improvement of in vitro mass-spectroscopic workflows (Hörner et al. 2019a) and the development of optogenetic cell-sorting strategies (Yüz et al. 2018) as described elsewhere (Tang et al. 2021).

Expectations on light-triggered on/off switches

Controlling the biological activity of a protein with light is sometimes compared with a physical light switch, which turns light on and off. As a consequence, we might be misled to envision photocontrol processes to render a protein completely inactive in the dark state while fully active in the light state. However, just as the efficiency of turning light on and off with a physical switch depends on the initial darkness of the room and the power of the bulb, the effect of a photocontrol strategy on the biological activity depends on the photochemical properties of the chromophore and on the response of the protein to those molecular changes on the structural level. A detailed description of these properties is provided in the following to facilitate a better understanding of what can be expected of each photocontrol strategy.

On the molecular level of the chromophore

In general, the reaction of a chromophore in response to light depends on two essential factors. The first is the probability that a photon is absorbed at a specific wavelength λ given commonly by the extinction coefficient ε _λ [l mol⁻¹ cm⁻¹] or the absorption cross section σ _λ [cm²]. Both constants are interrelated and can be converted into each other by the formula ε _λ  = 2.62·10²⁰ σ _λ (Lenci et al. 2012). The second factor is the quantum efficiency Φ, which describes the probability that once the chromophore has been excited by the absorption of a photon it falls back to the ground state through e.g., decaging or isomerization. The product of these two factors Φ·ε _λ [l mol⁻¹ cm⁻¹] is proportional to the concentration of the product molecules formed by these light-induced reactions. Hence, the chromophores used as receiver should exhibit high extinction coefficients and high quantum yields to increase the absolute sensitivity to light in all photocontrol methods.

Light-induced decaging and photocrosslinking reactions exhibit several reaction steps after initial excitation of the chromophore. This is illustrated by the exemplary decaging reaction of NBY in Figure 13A (Il’ichev and Wirz 2000; Klán et al. 2013). The total rate k of this multi-step process is inversely proportional to the lifetime τ of the intermediate of the rate-determining step (k = 1/τ). Hence, (de)stabilization of the intermediates by solvent or adjacent acids and bases can significantly influence the total rate, even though the initial photoinduced step might be faster in comparison. In proteins such interferences appeared to result in the incomplete decaging of ncAAs incorporated at positions close to acidic or basic side chains (Kneuttinger et al. 2019a), although the exact reason of this effect is not entirely clear. In case of nitrobenzyl caging groups the decaging reaction can even last significantly longer than the microsecond time scale (Klán et al. 2013). In addition to disturbances in the reaction rate, the photoreactions can suffer from side reactions as outlined for the photocrosslinking ncAA AzF in Figure 9.

Figure 13:

Photochemical properties of synthetic chromophores.

(A) Decaging reaction of NBY (Il’ichev and Wirz 2000; Klán et al. 2013). Irradiation with light excites the nitro group and triggers the acid-base (X) catalyzed formation of a five-membered ring, which requires conformational freedom. The subsequent fragmentation to tyrosin and o-nitrosobenzylaldehyde is irreversible. (B) Representative absorption spectra of E and Z azobenzenes. (C) Isomerization of an azobenzene induced by light of λ ₁ shifts the equilibrium of E and Z until a photostationary state (PSS) is reached.

Reversible, light-induced switching reactions of chromophores occur in photosensory domains and synthetic photoswitches. The complex photochemical properties of these chromophores are best explained by means of the well-studied azobenzenes (Hoorens and Szymanski 2018). The foundation for the switching reaction is that the two configurations of the photoswitch exhibit different absorption characteristics in the UV/visible spectrum (Figure 13B). The planar structure with C_2h symmetry of the E-azobenzene (Brown 1966; Crecca and Roitberg 2006) prohibits a n → π* transition resulting in a weak absorption band in the visible region at ∼450 nm (Cusati et al. 2008). The allowed π → π* transition, however, is much stronger and causes the main absorption maximum in the UV range of ∼320 nm of the E isomer (Forber et al. 1985). On the other hand, the Z isomer adopts a non-planar conformation with C2 symmetry (Gagliardi et al. 2004; Mostad et al. 1971), which allows the n → π* transition causing a stronger absorption at ∼450 nm and weakens the π → π* transition leading to a red-shift and reduction in absorption strength (Forber et al. 1985). Isomerization in the directions E → Z and Z → E can be controlled by irradiation at either transition. Thus, overlapping of the characteristic absorption bands for each transition in the E and Z spectra makes quantitative switching from 100% E to 100% Z impossible. For most switchable chromophores this is the case. The exemplary scenario in Figure 13C illustrates the consequences of overlapping absorptions bands. Irradiation of the E isomer with light of the wavelength λ ₁ excites the azobenzene in the π → π* transition band. Relaxation to the ground state then results in the formation of the Z isomer dependent on Φ ^E→Z ·ε _λ1 ^E . The kinetics of this isomerization process occur in the picosecond timescale (Bandara and Burdette 2012; Chang et al. 2004; Satzger et al. 2003). While λ ₁ excites the E isomer, it also excites the newly formed Z isomer, which isomerizes back as a function of Φ ^Z→E ·ε _λ1 ^Z . This back-and-forth reaction continues until a thermodynamic equilibrium is reached, which is known as the photostationary state (PSS). The ratio between E and Z in this equilibrium is called the photostationary state distribution (PSD). Intentional, light-induced isomerization of the Z isomer back to the E isomer is achieved in the exemplary scenario with the wavelength λ ₂ and follows the same principle. For photoswitches with well separated absorption bands of each isomer, the π → π* and n → π* transitions can be addressed selectively obtaining quantitative PSDs of up to 99:1. In most cases, the absorption bands are reasonably well separated resulting in a sufficiently strong isomeric shift. However, for less well separated absorption bands the PSDs can be much lower than 85:15. Importantly, the time needed to reach the PSS can be regulated by adjusting the intensity of the applied light and the PSD can be controlled by changing the irradiation wavelength. In practice, this means that different light sources vary in their switching efficiency. Conventional light-bulbs emit light in an extremely broad wavelength range and additionally radiate a lot of heat during irradiation, which can negatively affect the biological reaction. Modern LEDs have narrower wavelength ranges and barely give off energy in form of heat. Hence, these light sources are much better suited for the photocontrol of proteins. Lasers are without a doubt the best but also the most expensive light source one can use for photocontrol experiments since the light beam is monochromatic and of high intensity.

Another factor that needs to be considered with photoswitches is the thermal stability of the two isomers. As we have seen in previous chapters most photosensory domains and most Z isomers thermally relax back to the thermodynamic more stable form after initial irradiation. These instable isomers exhibit half-lives of microseconds to minutes (Garcia-Amorós et al. 2016), whereas stable isomers last for hours to even years (Knie et al. 2014; Weston et al. 2014).

On the conformational level of the protein

The response of the protein to light-induced changes of the chromophore on the molecular level is best studied for photoreceptors (Möglich and Moffat 2010; Ziegler and Möglich 2015). The theory draws upon the fundamental functionality of allosteric proteins, which undergo a conformational transition resulting in (de)activation after binding of a ligand at a site distant from the orthosteric center. One way to visualize this is the energy landscape of the proteins (Hans et al. 1991). To this end, the conformational transition induced by ligand binding (allostery) or light (photoreceptors) and thus the change in biological activity can be described as a shift in conformational populations (Kar et al. 2010). Greatly simplified, the photoreceptor is assumed to exist in an equilibrium between a standby conformation (S) of low activity and a working conformation (W) of high activity. In the absence of light, the chromophore in the protein adopts the thermally equilibrated dark-adapted state, which is defined by the equilibrium constant K_dark (Figure 14A). For light-activated photoreceptors, the standby conformation S is energetically favored over the working mode W; for light-inactivated photoreceptors, W is more stable than S. Moreover, K_dark is proportional to the difference in free energy ΔG_dark of both conformations in the energy landscape of the protein (Figure 14B, slate blue). Isomerization of the chromophore upon irradiation introduces a free energy perturbation ΔΔG, which triggers a shift in the equilibrium to reach the lit state described by K_lit and ΔG_lit (Figure 14A). This results in the stabilization of W to W* and destabilization of S to S* in light-activated photoreceptors or vice versa in light-inactivated photoreceptors (Figure 14B, green). The rates of the forward (k ₁ ) and reverse (k ₋₁ ) reaction between the dark and lit state are highly simplified in this principle assuming a unimolecular, single reaction step. The rate constant k ₁ is very fast and depends on the product Φ·ε _λ controlled by the wavelength as well as formation of the PSS regulated by the intensity of light, as we have seen in the earlier chapter. The back reaction in this principle is defined by the thermal reversion of the chromophore to the dark-adapted state and k ₋₁ therefore depends on its half-live. This thermodynamic cycle of photoreceptors is called the “photocycle” (Mowery and Stoeckenius 1981; Polland et al. 1986).

Figure 14:

The response of the protein to light-induced changes of the chromophore.

(A) Photocycle of photoreceptors explaining the thermodynamic equilibria in the dark-adapted (slate blue) and lit state (green). The difference between activities results from a shift in S and W conformational populations as demonstrated in panel (B). The size of the activity difference, which correlates with the fraction of the working mode (f _W ), depends on the initial position of the equilibrium ΔG_dark and the size of the shift ΔΔG. While only small changes in f _W are achieved when ΔG_dark is >>0 (or <<0, respectively) as shown in panel (C), large differences can be obtained when ΔG_dark approximates zero as illustrated in panel (D).

The response of the protein to the different isomers of the chromophore is thus relatively clear in the case of hybrid protein optogenetics. In contrast, the effect of the photocaged or photoswitchable ligands in photopharmacology is based on the change in their affinity towards the binding site of a protein. This determines the strength of competition with the native ligand. Hence, conformational transitions induced by the light-sensitive ligands mostly do not play a role. This is different for photoswitchable bridges in photopharmacology, in which the photoswitchable ligand is bioconjugated at two sites in the protein (Figure 3), and for light-sensitive ncAAs in photoxenoprotein engineering (Figure 8). Although the photocycle for these two photocontrol techniques has hitherto not been investigated it can be assumed that the photocages, photocrosslinkers and photoswitches influence the protein conformation both in the dark-adapted as well as the lit state. The population shift principle as described for photoreceptors above could therefore also be relevant for these photocontrol strategies.

On the activity level

The fundamentals from the previous two chapters allow us to better understand the effects of the on/off switch on the activity of the protein. As was mentioned earlier, the activity difference in proteins controlled by light-sensitive ligands in photopharmacology is primarily determined by the difference in binding strength of the caged and decaged state or the two isomers of the ligand, respectively. In most cases these differences are not particularly high resulting in LRFs of ∼2-fold in in vitro experiments. However, the same ligands have been shown to mediate a much more impressive effect in cellulo or even in vivo attributed to various factors (Hüll et al. 2018). First, the ligands usually have a good cell permeability increasing their local concentration through compartmentalization (Velema et al. 2014). More importantly, light-sensitive drugs are often designed to target essential parts of a signaling network (Hüll et al. 2018); consequently, the signaling effect multiplies the initially low LRF making the drug highly effective. Finally, the two states of the ligand can provoke different pharmacodynamic and pharmacokinetic effects, which again might influence the LRF in vivo.

For the implementation of photocontrol using strategies from photopharmacology it is important to distinguish the photochemical properties of ligands from those of photosensory domains in hybrid protein optogenetics. Photocaged ligands do not follow a thermodynamic cycle as the total reaction is irreversible. Hence, irradiation is only required until the decaging reaction is complete. As the decaging efficiency correlates with the product Φ·ε _λ this process depends on the wavelength spectrum of the light source and takes microseconds to several minutes (Klán et al. 2013). Photocontrol using photoswitchable ligands with a thermally instable state requires constant irradiation to maintain the PSS. However, the goal of photopharmacology is to produce thermally stable photoswitches and as a result most synthetic ligands exhibit thermal half-lives of hours to years. In such cases, a brief irradiation to obtain the PSS is sufficient. Fine-tuning of the LRF is best achieved by adjusting the wavelength of irradiation as the PSS is strongly dependent on Φ·ε _λ of the two isomers. In contrast, changing the light intensity is not as efficient since it only affects the duration until equilibration, which usually takes sub-seconds to minutes depending on the wavelength of irradiation.

The principles, which determine the LRF in strategies of photoxenoprotein engineering, are hitherto not entirely clarified. The effects are not as simple as for light-sensitive ligands in photopharmacology; differences in binding strength do not matter since the photocages, photocrosslinks and photoswitches are introduced translationally into the protein. As indicated in the previous chapter, the population shift principle as described for photoreceptors may apply to the artificial proteins created in photoxenoprotein engineering, as well. Hence, the following underlying concept of activity tuning in photoreceptors is fundamental to further analyze the possible effects of irradiation on the activity level in ncAA-containing proteins.

In hybrid protein optogenetics, regulation of the activity strongly depends on the photocycle (Figure 14) and its diverse factors (Möglich and Moffat 2010; Ziegler and Möglich 2015). To simplify, we first assume that the standby mode is 0% and the working mode is 100% active, and that the photoreceptor is activated. In this case, the thermodynamic principle demands a large shift of the equilibrium between S and W and hence a large increase in the fraction of W (f _W ) to maximize the LRF. The shift depends on two factors, the initial position of the equilibrium in the dark-adapted state ΔG_dark and the size of the free energy perturbation ΔΔG induced by irradiation. One might assume that in order to obtain the largest activity difference the shift needs to start at ΔG_dark >> 0 approximating 0% W. However, large ΔΔGs would be required to achieve a significant difference in f _W with this starting point (Figure 14C). In comparison, a much bigger f _W is obtained with the same ΔΔG when starting from ΔG_dark∼0 at an equilibrium of ∼50% S and ∼50% W (Figure 14D). In addition to ΔG_dark and ΔΔG the activity of both conformations S and W is decisive. Contrary to what we assumed at the beginning the activity of proteins is rarely zero; instead, it is regulated down by factors ranging from below 10-fold to more than 1000-fold depending on the strength of the interference. The larger the difference in activity of S and W is, the higher the LRF. Remarkably, the interplay between ΔG_dark, ΔΔG, and the difference in activity of S and W has been optimally adjusted by nature in photoreceptors. Engineering of artificial photoreceptors can achieve high LRFs of >100-fold (Carrasco-López et al. 2020); however, often photoreceptor variants only exhibit a modest LRF of 2- to 10-fold (Gasser et al. 2014; Hongdusit et al. 2020; Ryu et al. 2014), which can be further improved to >10-fold by optimization screenings by which primarily ΔG_dark is shifted (Strickland et al. 2010), whereas ΔΔG is assumed to be challenging to alter (Ziegler and Möglich 2015). Nevertheless, it has been shown that even LRFs of ≤10-fold can cause significant phenotypical changes in vivo (Barends et al. 2009; Wu et al. 2009).

To fine-tune the LRF in an optogenetic experiment it must be considered that most photosensory domains have only short thermal half-lives of tens to 10 thousands of seconds corresponding to merely a few hours (Möglich and Moffat 2010). This means the photosensory domain reverts back from the lit state to the dark-adapted state with the rate constant k ₋₁ . A brief light pulse to reach the PSS is therefore not sufficient to evoke a long-lasting effect. The samples thus need to be constantly irradiated. Depending on the rate constant k ₁ , which can be adjusted by the intensity of light, different equilibria between the lit and the dark-adapted state can be established facilitating a fine-tuning of the LRF.

Coming back to photoxenoprotein engineering, the activity of the dark-adapted state and the activity of the lit state are decisive for the LRF just as in hybrid protein optogenetics. This factor is exploited for the photocontrol with photocaged ncAAs by targeting essential residues in the orthosteric site for replacement. High LRFs of >100-fold can be achieved by this strategy since the photocage efficiently blocks the function of the residue (Jedlitzke et al. 2020). While this factor is just as important for the other strategies in Figure 8, an equilibrium between S and W conformations, and hence ΔG_dark and ΔΔG, will play an increasingly pronounced role as well. An easy example to illustrate this is the photocontrol of multimerization using light-sensitive ncAAs. In the dark-adapted state, complex formation of two subunits might be hampered by the presence of the ncAA; however, the monomers (equivalent to the standby mode S) and the complex (equivalent to the working mode W) will still exist in equilibrium described by the dissociation constant K _D . Light-induced changes of the ncAA might then facilitate multimerization, which translates into a shift in equilibrium and an improvement of K _D comparable to the cycle in Figure 14A. Moreover, it is highly likely that the third photoxenoprotein engineering strategy, the photocontrol of a conformational change, follows the same basic principles as the photocycle of photoreceptors. Notably, the term photocycle would only fit for photoswitchable ncAAs as the change in equilibrium is irreversible for photocaged and photocrosslinking ncAAs. Recent findings proved that introduction of the light-sensitive moiety into strategically chosen positions close to existing conformational transitions can result in high LRFs of >10- (Kneuttinger et al. 2019a, 2020) and >100-fold (Kneuttinger et al. 2019b). Even the photoswitchable bridge strategy of photopharmacology, for which these principles most likely apply as well, achieved LRFs of substrate binding of >100-fold (Bozovic et al. 2020, 2021).

Despite the analogy to hybrid protein optogenetics regarding the photocycle, the strategies of photoxenoprotein engineering differ in terms of implementation. For the same reasons as for the photocages and photoswitches in photopharmacology, no constant irradiation is required for light-sensitive ncAAs. Likewise, fine-tuning of the LRF is more efficient by changing the wavelength of irradiation than light intensity.

Photopharmacology

The design of photocontrol in photopharmacological strategies first and foremost requires chemical engineering of the light-sensitive ligand. In principle, the design needs to include moieties that mediate high affinity binding, which are often copied from native ligands, and photocaged or photoswitchable components with beneficial photochemical properties. Moreover, the ligands should be synthetically accessible and rapidly producible. This, however, means that new ligands have to be designed for each new target protein making chemical engineering quite challenging. Customizing the photochemical properties such as the thermal stability of the Z isomer (Kolarski et al. 2021b) or the shift to longer wavelengths (Konrad et al. 2016) can be realized by adding substituents to the initial scaffold, but is quite tedious since these substitutions might have adverse effects and make synthesis difficult. Nevertheless, many advances have been made towards for example red-shifted light-sensitive ligands in the last couple of years (Leistner et al. 2021; Morville et al. 2021).

While for the engineering of free ligands only the active site of the target protein needs to be considered, bioconjugated PTLs and PORTLs further require difficile design of the spacer between the protein anchor site and the active site. The design of photoswitchable bridges is even more complex since here the photoswitch is attached to two sites in the protein and must induce significant conformational changes to light-regulate protein activity. Both attachment sites and spacer size need to be precisely adjusted. Thus, this protein-based design component is often supported by computational approaches. Using photoswitchable bridges to control the α-helical content of proteins is a strategy, which reaches back two decades. One of the earliest publications (Kumita et al. 2000) clicked an azobenzene-based photoswitch to two cysteine residues positioned seven residues apart. Molecular modelling revealed that only the Z configuration facilitated the secondary structure, whereas the E configuration of the photoswitch was predicted to disturb the α-helix because it was too long. In a follow-up study, the authors again used computational methods to analyze whether steric interactions of residues in between the two cysteines influence the photoswitch, which does not appear to be the case (Kumita et al. 2002). Another approach identified dynamical surface regions in lipase B from Candida antarctica that promote catalysis by computational modelling (Agarwal et al. 2012). The authors conjugated an azobenzene photoswitch to these regions. The rationale behind this strategy is the anticipation that the conformational switch of the azobenzene upon irradiation with light might introduce energy into the dynamical regions and hence result in activation of catalysis. In fact, lipase B activity could be light-dependently increased by 24%; after taking the high percentage of non-crosslinked enzymes into account the LRF increased from ∼8- to 52-fold. While this approach reflects the power of inducing conformational changes in proteins by light, it also demonstrated that low crosslinking efficiencies are most critical for bioconjugation-based photocontrol strategies. Yet another study made use of a computational approach to calculate crosslinking sites for an azobenzene-based photoswitch in cytosine deaminase (Blacklock et al. 2018). The Rosetta-based simulations successfully predicted a site at which the photoswitchable bridge mediated photocontrol with an LRF of ∼3-fold and a site at which it did not have an effect. In the former design the azobenzene ligand crosslinked two subunits of the enzyme dimer, whereas in the latter design it was placed in an α-helix. Additional mutagenesis is rarely used to improve the LRF of the ligand-protein-pair in photopharmacology; however, one study proved that this kind of optimization is possible (Schierling et al. 2010). The authors installed a photoswitchable bridge in the homodimeric restriction enzyme PvuII and initially achieved an LRF of ∼7-fold. Additional amino acid exchanges could further increase the LRF to ∼16-fold.

Photoxenoprotein engineering

The design of ncAA-based photocontrol in proteins shares some processes with photopharmacology and some with hybrid protein optogenetics. In a first step, a suitable ncAA has to be chosen for the desired strategy and application. Similarly to hybrid protein optogenetics, a library is then created, in which the chosen ncAA(s) is/are incorporated at various sites in the target protein. The selection of positions is primarily based on prior knowledge of the target system including e.g., residues involved in catalysis, ligand binding or allostery, for which the three-dimensional structure of the protein may be beneficial. In a subsequent low-throughput screening variants are identified, which exhibit light-dependent behavior with activities of the lit state comparable to wildtype. Since the effect of the light-sensitive ncAAs is often difficult to anticipate, some computationally based protocols have been developed to ease the choice of positions for ncAA incorporation. Comparable to examples in photopharmacology, a computational approach was recently used to detect regions of high conformational flexibility in firefly luciferase, in which incorporation of AzoF could result in efficient photocontrol (Luo et al. 2018). As a result, the authors obtained two out of seven initial AzoF-variants with which the chemiluminescence of firefly luciferase could be repeatedly switched on and off in live mammalian cells. A more comprehensive study devised a Rosetta-based strategy to determine sites close to the active site, in which the caged tyrosine NBY achieves high LRFs (Wang et al. 2019a). The authors initially screened 60 sites in firefly luciferase feeding the results into the algorithm of Rosetta (Park et al. 2016) to create a powerful prediction tool. They further accomplished to photocontrol multiple cellular processes by photocontrolling various target proteins thereby showing the general applicability of their approach, which they dubbed “computationally aided and genetically encoded proximal decaging (CAGE-prox)”. Apart from these studies for improving the protein engineering component of photoxenoprotein engineering, plenty of new ncAAs are designed e.g., with optimized photochemical properties, comparable to the design of ligands in photopharmacology. Nevertheless, chemical engineering of ncAAs is limited by the incorporation capability of the heterologous aaRS/tRNA pairs. Evolved aaRSs may tolerate minimal changes of an ncAA; however, larger changes usually require a new evolution experiment.

In general, the design process of ncAA-based photocontrol largely depends on the incorporation efficiency of the amber suppression technique. Several concerns regarding this method exist. A major drawback is the low ratio of full-length versus truncated protein, which is caused by the competition of the aminoacylated heterologous tRNA with release factor 1 for binding to the amber codon. Another matter is the potentially low expression yields of the full-length protein; a common rule of thumb approximates the yields to 10% compared to wildtype. Usually, a wide yield range of 1–100 mg of protein per liter expression medium is common (Liu and Schultz 2010). Moreover, the fidelity of incorporation, meaning that canonical amino acids are incorporated in addition to the ncAA, is also a highly discussed issue. Besides these major concerns, practical limitations exist such as high wastage of ncAA due to poor influx rates of the ncAA into the host cells (Chin 2017), varying incorporation efficiencies in different positions of a protein (Young et al. 2010) and the need of optimizing expression conditions anew for each target protein. Nevertheless, various advancements have been made in recent years of which I describe some in the following.

The ratio of full-length versus truncated protein was maximized by the engineered E. coli strain C321.ΔA, in which release factor 1 is knocked out and all chromosomally encoded TAG codons are replaced by the other amber codons (Kuznetsov et al. 2017; Lajoie et al. 2013). A significant drawback of this strain is its slow growth rate, which makes the expression of proteins especially in large scale quite difficult reducing the overall yields. A recent study found that primarily the carbohydrate and energy metabolisms are downregulated in this strain (Yi et al. 2021). Hence, other studies have tried to improve the system. One example is the strain B-95.ΔA, which is based on the quite common expression host strain E. coli BL21(DE3) and which also lacks release factor 1 (Mukai et al. 2015). Since a large amount of TAG codons in the genome of E. coli does not only serve as a stop codon in one open reading frame, but also encode amino acids in an alternative open reading frame, the authors only deleted 95 of the 273 TAG codons present. As a result, the strain retained the good growth properties of the BL21(DE3). Besides elimination of truncated proteins, increased efficiency of multi-site incorporation is a major benefit of the C321.ΔA strain (Zheng et al. 2016). Further advancements in this direction are engineered strains (Robertson et al. 2021) or cell-free protein synthesis systems (Cui et al. 2020), which allow to incorporate different ncAAs in response to sense codons instead or in addition to the amber codon. A second challenge needed to overcome for multi-site incorporation are low affinities of the evolved aaRSs for the targeted ncAA (O’Donoghue et al. 2013; Umehara et al. 2012). These low affinities might partly be owed to the high copy number of plasmids encoding the aaRS in evolution experiments. Two studies circumvented this issue by developing an in vivo evolution approach, in which libraries of chromosomally integrated aaRSs were used (Amiram et al. 2015; Israeli et al. 2021). In fact, the engineered aaRSs showed increased aminoacylation efficiencies for diverse ncAAs including AzF and AzoF. In an alternative approach, the affinity of the aaRS for NBY was increased by evolving up to 17 residues—significantly more compared to the six residues in the traditional approach (Deiters et al. 2006)—in the computationally-guided redesign including some that are able to form specific interactions with the bound ncAA (Baumann et al. 2019). As a result, the yields for single-site incorporation were improved to 100% compared to wildtype and the yields for multi-site incorporation were markedly increased as well. In addition to the aaRS, expression yields highly depend on the tRNA and the expression conditions. Thus, an early publication developed an enhanced plasmid system, pEVOL, which encodes two copies of the evolved aaRS, coupled to a constitutive and an arabinose-induced promoter, respectively, and a suppressor tRNA that was optimized in directed evolution experiments (Guo et al. 2009; Young et al. 2010). This plasmid system has become the golden standard for ncAA incorporation experiments. Expression yields could also be significantly improved by choosing the Vibrio natriegens based Vmax X2 expression host (González et al. 2021). This strain exhibits rapid growth rates allowing to increase expression yields, is fully compatible with plasmids designed for expression in E. coli and additionally shows reduced misincorporation of endogenous amino acids into the amber codon. Finally, wastage of ncAA was reduced by an engineered leucine-binding protein that actively imports a broad repertoire of ncAAs into the bacterial cell and hence significantly reduces the ncAA concentration needed for the expression of the target protein (Ko et al. 2019).

Hybrid protein optogenetics

In optogenetics, it is not uncommon that a natural photoreceptor exists with the exact requirements needed for the desired application, e.g., certain ion channels, and which is then used in divergent tissues. However, most researchers interested in photocontrol of proteins aim to put a specific target protein under the control of light. For this purpose, hybrid proteins are designed in optogenetics by fusion of a photosensory domain with the protein of interest. Although this appears easily achievable, this approach requires cumbersome and elaborate engineering in multiple steps. A generally applicable protocol has hitherto not been established due to the fact that each protein is individual; a strategy, which facilitates a high LRF for one protein, might not be able to attain photocontrol of another protein at all.

In a first step, an appropriate photosensory module needs to be chosen. Attention especially needs to be paid to whether irradiation should occur with low-energy red or infra-red light; if this is the case, phytochromes are the photosensor of choice (Figure 12). In the next step, a library is created comprising various protein fusion constructs of photosensor and target protein (Möglich and Moffat 2010; Mathony and Niopek 2021). The different constructs may vary in the linker sequence and length or the position in the target protein to which the linker is attached. For the identification of initial light-regulatable constructs a small library and a low-throughput screening usually suffices. Nevertheless, some design rules should be kept in mind when starting to engineer an artificial photoreceptor. For non-associating strategies (Figure 11) to work, the target protein needs to be structurally related to the native effector protein (Mathony and Niopek 2021). Moreover, since the interaction between photosensor and effector is critical for these strategies, the properties of the linker, such as its length, sequence, and structure, are decisive for the design (Möglich et al. 2009; Ryu et al. 2014; Ziegler and Möglich 2015). Associating strategies (Figure 11), on the other hand, are not critically limited by structural homology of the target protein to the native effector or the linker identity. For both strategies, however, the fusion site is of crucial importance. The photosensor can either be attached at the termini or within the target protein, which opens plenty of possibilities for photocontrol but unfortunately also for failures. Many successful studies have therefore relied on prior knowledge of the target protein to choose appropriate sites; this knowledge might include allosteric sites (Gehrig et al. 2017) or other activation mechanisms (Smart et al. 2017) as well as previous protein engineering studies (Fukuda et al. 2014; Hattori et al. 2013).

Ideally, initial screening has identified one or more variants, which allow the photocontrol of the chosen target protein. In the next step, the variants are typically further optimized by introducing point mutations, deletions or insertions in directed evolution experiments (Mathony and Niopek 2021; Möglich and Moffat 2010). These aim at either optimizing the linker length (Bubeck et al. 2018; Dagliyan et al. 2019) or directly improve the photochemical properties of the photoreceptor by modulating the recovery time (Pudasaini et al. 2015; Ziegler and Möglich 2015) or increasing the LRF (Strickland et al. 2010). Most studies thereby shift the equilibrium between the S and W conformation (Strickland et al. 2010; Yao et al. 2008; Ziegler and Möglich 2015).

The most critical step in this outlined design process is probably the initial search for a light-regulatable photoreceptor, which is not very intuitive. Hence, multiple studies tried to ease this first step by computational design approaches (Dagliyan et al. 2019) or by a switch to high-throughput screenings. Coevolution studies using statistical coupling analysis identified allosteric regulation hotspots on the surface of E. coli dihydrofolate reductase. Fusion of these hotspots to the LOV2 domain from A. sativa resulted in a two-fold light-induced regulation of enzyme activity and the authors speculated that this approach might be used for the design of other photoreceptors (Lee et al. 2008; Reynolds et al. 2011). In a similar approach the LOV2 domain was inserted into non-conserved surface loops, which were identified to be allosterically coupled to the active site of the target protein by molecular dynamics simulations (Dagliyan et al. 2016). Various proteins could be photocontrolled by this strategy in living cells. Still, these computational approaches are far from generalizable. While the latter study suggested to place the photosensor module into surface loops, authors from another study could not accomplish photocontrol this way; instead, the photosensor needed to be inserted into a β-sheet (Hoffmann et al. 2021). Thus, high-throughput screening of fusion construct libraries designed by the Mu-transposon strategy (Edwards et al. 2008; Nadler et al. 2016) or gene library synthesis (Coyote-Maestas et al. 2020) constitutes an alternative approach. To this end, the photosensor module is inserted at random positions in the target protein. This approach is only feasible for proteins, which exhibit easily detectable activities for high-throughput in vivo selection or microtiter-plate-based screening processes. In the course of the experiment, the photoreceptors are sieved to eliminate constitutively active or inactive variants and to enrich constructs that show a significant LRF between activities in the dark and lit state.