**Atomistic Insight into the Role of Threonine 127 in the Functional Mechanism of Channelrhodopsin-2**

**David Ehrenberg 1,**†**, Nils Krause 2,**†**, Mattia Saita 1, Christian Bamann 3, Rajiv K. Kar 4, Kirsten Ho**ff**mann 2, Dorothea Heinrich 2, Igor Schapiro 4, Joachim Heberle 1,\* and Ramona Schlesinger 2,\***


Received: 28 September 2019; Accepted: 12 November 2019; Published: 15 November 2019 -

**Abstract:** Channelrhodopsins (ChRs) belong to the unique class of light-gated ion channels. The structure of channelrhodopsin-2 from *Chlamydomonas reinhardtii* (*Cr*ChR2) has been resolved, but the mechanistic link between light-induced isomerization of the chromophore retinal and channel gating remains elusive. Replacements of residues C128 and D156 (DC gate) resulted in drastic effects in channel closure. T127 is localized close to the retinal Schiff base and links the DC gate to the Schiff base. The homologous residue in bacteriorhodopsin (T89) has been shown to be crucial for the visible absorption maximum and dark–light adaptation, suggesting an interaction with the retinylidene chromophore, but the replacement had little effect on photocycle kinetics and proton pumping activity. Here, we show that the T127A and T127S variants of *Cr*ChR2 leave the visible absorption maximum unaffected. We inferred from hybrid quantum mechanics/molecular mechanics (QM/MM) calculations and resonance Raman spectroscopy that the hydroxylic side chain of T127 is hydrogen-bonded to E123 and the latter is hydrogen-bonded to the retinal Schiff base. The C=N–H vibration of the Schiff base in the T127A variant was 1674 cm<sup>−</sup>1, the highest among all rhodopsins reported to date. We also found heterogeneity in the Schiff base ground state vibrational properties due to different rotamer conformations of E123. The photoreaction of T127A is characterized by a long-lived P2 <sup>380</sup> state during which the Schiff base is deprotonated. The conservative replacement of T127S hardly affected the photocycle kinetics. Thus, we inferred that the hydroxyl group at position 127 is part of the proton transfer pathway from D156 to the Schiff base during rise of the P3 <sup>530</sup> intermediate. This finding provides molecular reasons for the evolutionary conservation of the chemically homologous residues threonine, serine, and cysteine at this position in all channelrhodopsins known so far.

**Keywords:** channelrhodopsin; resonance Raman; flash photolysis; hybrid QM/MM simulation; electrophysiology

#### **1. Introduction**

Channelrhodopsins (ChRs) are members of the group of microbial rhodopsins that are light-gated cation channels. They were originally found in the eyespot of the algae *Chlamydomonas reinhardtii* (*Cr*),

where they serve to identify optimal light conditions during phototactic movement [1,2]. Typically for rhodopsins, ChRs consist of seven α-helices spanning the membrane and a retinylidene chromophore, which is bound via a Schiff base to a lysine in the core of the protein.

Within the new research field of optogenetics, the coding sequences for light-responsive proteins together with regulating promoters can be introduced into complex organisms and expressed in a tissue-specific way, where cell processes can be triggered. Channelrhodopsin-2 (*Cr*ChR2) in neurons can depolarize the nerve cell when illuminated by light of the corresponding wavelength. Although this protein is already frequently used as optogenetic tool, aimed at medical approaches like restoring vision and hearing, the mechanism of action is still not completely understood. Comparison to related microbial rhodopsins where individual positions in the protein sequences are highly conserved could guide our approach towards mechanistically important amino acid residues which have demonstrated effects on photocycle kinetics, activity, and/or structural stability upon mutation.

Decades ago, the Khorana group investigated the role of threonines in the proton pump bacteriorhodopsin (bR) from *Halobacterium salinarum*. In that study, the T89V variant exhibited a blueshift of 146 nm of the visible absorption spectrum without impairing proton pumping [3]. It was concluded from FTIR and time-resolved absorption spectroscopy in the visible range that T89 exerts steric constraints during isomerization of the chromophore and is part of a hydrogen-bonded network including a water molecule and the residues Y185 and D212 [4]. Years later, the same group demonstrated that the hydroxyl group of T89 interacts with the protonated Schiff base in light-adapted bR and influences proton transfer to D85 during the photocycle [5]. However, from X-ray structures [6,7], it became apparent that the OH group of T89 forms a hydrogen bond to an oxygen of the side chain of D85 rather than to the Schiff base. It was concluded from polarized FTIR spectroscopy that the isomerization of the chromophore after light excitation leads to a shortening of the hydrogen bond between T89 and D85, indicating an interconnection to the Schiff base [8]. This tight complex persists from the K intermediate until the M intermediate [9].

The hydroxylic amino acid side chain of T89 of bR is conserved in many channelrhodopsins. The corresponding residue T127 of *Cr*ChR2 is located in immediate vicinity to the retinal Schiff base (Figure 1), with a distance of 3.7 Å of the threonine oxygen to the Schiff base nitrogen in the crystal structure. Other channelrhodopsins carry a serine [10] or a cysteine at this position. The latter appears in *Ca*ChR1 (from *Chlamydomonas augustae*) and *Cy*ChR1 (from *Chlamydomonas yellowstonensis)* [11]. Even anion-conducting channelrhodopsins, which do not conserve E123, have a threonine or a cysteine at this position [12].

**Figure 1.** X-ray crystallographic structure of *Chlamydomonas reinhardtii* channelrhodopsin-2 (*Cr*ChR2) (PDB: 6EID [10]). The protonated Schiff base of retinal and the side chains of key residues are shown as grey and yellow sticks, respectively. Water molecules are shown as red spheres. Dashed lines indicate hydrogen-bonding interaction.

In *Cr*ChR2, the residue T127 is connected via a hydrogen bond to E123, the homologous residue of D85 in bR [10]. The protonation state of E123 is not clear, but nearby ionized D253 serves as the acceptor of the Schiff base proton during the rise of the P2 <sup>390</sup> intermediate [13]. D156 has been shown to interact via hydrogen bonding with C128, denoted as the DC gate [14]. The X-ray structure as well as theoretical calculations confirmed that the terminal groups of both amino acids are interconnected via a water molecule [10,15]. D156 is protonated in the dark state and deprotonates during the P2 <sup>390</sup> to P3 <sup>520</sup> transition concomitantly to the reprotonation of the Schiff base. Thus, D156 is the proton donor of the Schiff base in *Cr*ChR2 [13]. The distance from the DC gate to the Schiff base is too large for direct proton transfer. It was shown in Reference [10] that the nearby residue T127 may facilitate the reprotonation process. To evaluate this hypothesis, we set out to investigate variants in which T127 was replaced by alanine or serine.

Channel activity of the T127 variants was examined by electrophysiology and exhibited reduced conductance in the T127 variants. Molecular spectroscopy (UV/Vis, FTIR, Raman) was applied to scrutinize the role of T127 in the molecular mechanism of *Cr*ChR2. We report here the exceptionally high frequency of the Schiff base vibration in the T127A variant, indicating strong interaction between the protonated Schiff base and the carboxylic side chain of E123 in this variant. QM/MM calculations supported and extended the spectroscopic results with detailed atomistic descriptions of the hydrogen-bonded network surrounding the retinal Schiff base. We believe that the characterization of the threonine variants at the molecular level is an important step towards the understanding of the link between the protein's spectroscopic properties and its function as a light-activated cation channel.

#### **2. Materials and Methods**

#### *2.1. Site-Directed Mutagenesis, Cloning, Expression and Purification of CrChR2*

The experiments were executed with recombinant *Cr*ChR2, which consisted only of the membranous part containing amino acid residues 1–307 of channelrhodopsin-2 of *Chlamydomonas reinhardtii* (UniProt: Q8RUT8). Due to insertion of the corresponding coding region with sequences for a C-terminal 10xHis tag and a linker (aa AS) behind the alpha factor signal sequence of the vector pPIC 9K into the *Eco*RI/*Not*I sites, the expressed protein led to a N-terminal extension of aa YVEFH and a C-terminal extension of aa ASHHHHHHHHHH. Based on this construct, which is referred in the following as *Cr*ChR2-WT or simply WT, T127 was substituted by serine (T127S) or alanine residues (T127A). The substitutions were introduced with oligonucleotide-directed mutagenesis using PCR and verified by sequencing of the *Cr*ChR2 coding sequence. The *Cr*ChR2 variants were expressed and purified as described previously [16].

#### *2.2. Molecular Spectroscopy*

Time-resolved UV/Vis experiments were performed with a commercial flash photolysis unit (LKS80, Applied Photophysics, Leatherhead, Surrey, UK), as described previously [17]. Briefly, a 10 ns laser pulse (Nd:YAG, Quanta-Ray, Spectra-Physics) tuned to 450 nm by an optical parametric oscillator (OPTA) was used to induce the photocycle. The energy density per pulse was set to 3 mJ/cm2. Five time traces were averaged at each wavelength with a repetition frequency of 0.33 Hz.

Samples used for FTIR spectroscopy were concentrated to ~4 mg/mL *Cr*ChR2 in an aqueous solution of 20 mM Hepes and 0.2% DM (n-decyl-β-D-maltopyranoside) at pH 7.4. For the FTIR experiments, approximately 8 μl of *Cr*ChR2 was dried on top of a BaF2 window. The protein film was rehydrated with the saturated vapor phase of a glycerol/water mixture (2/8 w/w) [18] and placed into the FTIR spectrometer (Vertex 80v, Bruker, Rheinstetten, Germany). Time-resolved rapid-scan FTIR spectroscopy was employed to resolve intermediates with a time resolution of about 10 ms.

Resonance Raman experiments were performed essentially as described in Reference [19]. Briefly, 5 μL of concentrated sample (5–10 mg/mL) was dried and rehydrated on a quartz crucible and subsequently cooled to 80 K using a N2 cryostat (Linkham). Rehydration occurred prior cooling via

vapor diffusion of 3 μl of either H2O/glycerol or D2O/glycerol mixtures (8/2 w/w) placed in the vicinity of the sample. The emission line at 457 nm of a diode-pumped solid-state (DPSS) laser (Changchun New Industries Optoelectronics Technology Co., Ltd., China) was used to induce Raman scattering.

#### *2.3. Electrophysiology*

Light-induced currents were recorded from oocytes in two-electrode voltage-clamp (TEVC) experiments after expression of the WT and the T127A or T127S variants for at least 3 days. To this end, in-vitro-synthesized RNA coding for a fusion protein of chop2 (residues 1 to 315) and egfp was injected into oocytes and incubated in oocyte Ringer (ORI, 90 mM NaCl, 2 mM KCl, 2 mM CaCl2, 1 mM MgCl2, 5 mM Hepes, pH 7.4) solution supplemented with 5 μM all-trans retinal. Oocytes were illuminated with light from a 75 W XBO lamp long-pass filtered at 420 nm in ORI. Currents of the T127 variants were normalized to the WT current amplitudes at −60 mV from oocytes recorded at the same day.

Time-resolved single turnover currents of the T127 variants were also recorded from HEK293 cells expressing the same fusion protein as in the oocyte recordings. Cells were illuminated with a single flash (20 ns) from an Excimer-laser pumped dye laser (coumarin 2, 450 nm). Traces were sampled at 50 kHz and filtered at 10 kHz. The displayed traces are averages of eight signals.

#### *2.4. QM*/*MM Calculations*

The structural model for hybrid quantum mechanics/molecular mechanics (QM/MM) calculations was based on the crystal structure of *Cr*ChR2 (PDB entry: 6EID [10]) with an all-trans conformation of the retinal protonated Schiff base (RSBH+). Titratable residues were modeled corresponding to the experimental pH and considering their local environment. Based on spectroscopic studies, the residue E90 was modeled in the neutral form and the counterions of RSBH<sup>+</sup> (E123 and D253) were taken as negatively charged. The variants of *Cr*ChR2, T127A and T127S, were prepared with PyMol [20].

In order to account for the effect of T127 and its variants, we used a large QM region that comprised the RSBH<sup>+</sup> side chains of E123, T127, A127, S127, C128, D253, K257, and two water molecules in the close vicinity of RSBH+. The hydrogen link atom was placed at the QM/MM boundary between the Cδ and Cε atoms of K257, and between the Cα and side chain atom of remaining residues. The QM region was treated at the B3LYP/cc-pVDZ level of theory [21]. Corrections for the dispersion effect were included with D3/B-J damping variant [22]. All the remaining atoms were described at the MM level using the AMBER ff14SB force field [23]. The vibrational frequencies were calculated at the same level of theory. A scaling factor of 0.97 was used [24]. However, the QM region in frequency calculations did not include the water molecules. Furthermore, the QM/MM model was used to calculate the excitation energies using the simplified TD-DFT approach developed by Grimme and coworkers [25]. All QM/MM calculations were carried out using ChemShell interfaced with the quantum chemistry program Orca with DL\_POLY module for the force field [26].

#### **3. Results**

In the X-ray crystal structure of *Cr*ChR2, the distance between heavy atom pairs N(RSBH+) ... O(E123) and N(RSBH+) ... O(D253) is 2.81 and 3.24 Å, respectively. The QM/MM geometry optimization leads to slight adjustments in the retinal binding pocket, which modifies the corresponding distances to 2.82 and 3.48 Å. In WT, T127 was found to be hydrogen-bonded with E123 with an OH(T127) ... O(E123) distance of 1.70 Å (Figure 2A). Upon T127S exchange, the hydrogen-bonding network was retained with a slightly reduced distance of 1.67 Å between OH(T127S) ... O(E123) and an increased distance of 3.53 Å between N(RSBH+) ... O(D253) (Figure 2B). The difference between serine and threonine residues is the absence of the methyl group at the C<sup>β</sup> position in the former. Lack of this bulky group in serine might cause a reduced steric effect responsible for its approach to E123. In contrast, the T127A variant was missing a proton-donating group (Figure 2C); therefore, E123 had only one hydrogen-bonding partner, namely the Schiff base. The distance between N(RSBH+) ... O(E123) remained the same (2.81 Å), but the N–H bond length

was elongated by 0.016 Å (Table 1) because the proton is pulled more strongly towards E123 as a consequence of its altered hydrogen bonding.

**Figure 2.** Optimized geometry of (**A**) *Cr*ChR2-WT, (**B**) T127S, and (**C**) T127A variants. The retinal Schiff base and side chains of the counterion complex are shown as grey and yellow sticks, respectively. Dotted lines correspond to the hydrogen-bonding network.

**Table 1.** Bond length in selected RSBH<sup>+</sup> atom pairs, based on the hybrid quantum mechanics/molecular mechanics (QM/MM)-optimized geometries.


In *Cr*ChR2-WT, T127 was also found to be involved in a triad between O(E123), OH(T127), and S(C128). The corresponding distances in the optimized geometry were found to be 4.91 Å for O(E123) ... S(C128) and 4.39 Å for OH(T127) ... S(C128). In the T127S variant, the distance between O(E123) ... S(C128) was slightly reduced to 4.88 Å, whereas the distance between OH(T127S) ... S(C128) was increased to 4.47 Å. This indicated that the T127S variant led to a tighter network with the DC gate and the SB counterions. However, the T127A exchange perturbed the triad and decoupled the DC gate from the counterions, as the distance between O(E123) ... S(C128) was increased to 4.95 Å.

#### *3.1. Spectroscopic Properties of the Dark State of the T127 Variants of CrChR2*

As T127 is part of the retinal binding pocket, we first examined the potential influence of the amino acid exchange on the electronic properties of the retinal. It was evident from comparison of the UV/Vis spectra (Figure 3, red and blue spectra) that both T–A and T–S amino acid exchanges shifted the visible absorption spectrum only slightly to the blue as compared to WT (black spectrum in Figure 3). The second derivative of the absorption spectra (bottom panel) resolved a blue shift of about 2 nm in the T127A variant as compared to the WT (similar to T127S, not shown). The vibronic fine structure of the electronic absorption spectrum was retained in the T127A and the T127S variants. The latter observation was different to the C128T variant, where a 20 nm redshift of the retinal absorption was accompanied by a loss in vibronic fine structure [27].

We also computed the excitation energies for the WT protein and the two variants. The results displayed in Table 2 showed virtually no alteration of the excitation energies, which agreed well with the experimental results (Figure 3). These results present an additional validation of our QM/MM model.

**Table 2.** Computed excitation energies at the sTD-DFT CAM-B3LYP/cc-pVDZ level of theory.


**Figure 3.** UV/Vis absorption spectra for the *Cr*ChR2-WT, T127A, and T127S variants. The second derivatives of the absorption spectra are displayed in the bottom panel; the second derivative of the T127S spectrum has been omitted for clarity as it overlaps with the T127A trace.

As the wavelength of the electronic absorption in the visible correlated with the C=C stretching vibration [28], the strongest band in the resonance Raman spectrum of the T127A variant (Figure 4, green trace) was 1557 cm<sup>−</sup>1, slightly blue-shifted by 6 cm−<sup>1</sup> with respect to the WT (Figure 5A, [29]). All of the other Raman bands were essentially the same as in the WT, except for the C=N–H vibration of the retinal Schiff base (Figure 5B). The frequency of the latter was observed at 1674 cm−<sup>1</sup> for T127A, which is, to our knowledge, the highest frequency so far observed for any rhodopsin. It even outperformed the frequency of the Schiff base vibration in rhodopsin (1660 cm−1, [30]). The C=N–H vibrational band was downshifted by 42 cm−<sup>1</sup> upon H/D exchange, which was the largest isotope effect reported (Figure 4). In comparison, the C=N–H stretch in WT resonated with a frequency of 1657 cm−<sup>1</sup> and a ΔυH/<sup>D</sup> = 28 cm−<sup>1</sup> was reported [29]. As the hydrogen bond between T127 and E123 was removed in the T127A variant (Figure 2C), the carboxylic side chain of E123 was able to form a stronger salt bridge with the retinal Schiff base, resulting in a 0.016 Å elongated N–H bond (Table 1).

**Figure 4.** Resonance Raman spectra of the T127A variant in H2O (green trace) and in D2O (red trace). The indicated vibrational bands are discussed in the text.

**Figure 5.** (**A**) Resonance Raman spectra of *Cr*ChR2-WT and the variants E123D, T127S, and T127A in H2O. (**B**) Zoom-in of the frequency range of the Schiff base vibration after a three-point moving average smooth. Blue circles are the data points. Dashed grey lines are Gaussians fitted to the data points and continuous lines are sums of the fitted Gaussians.

Inspection of the vibrational band at 1674 cm−<sup>1</sup> revealed a strong asymmetry in the shape, with a pronounced shoulder at lower wavenumbers (green trace in Figure 5B). Consequently, the band was fitted by two Gaussians with frequency maxima at 1674 cm−<sup>1</sup> and 1661 cm<sup>−</sup>1. Such heterogeneity in the retinal binding pocket has been suggested from molecular dynamics simulations on WT [31] to be a result of two rotamers of the side chain of E123. In one rotamer configuration, E123 was in direct hydrogen-bonding interaction with the protonated Schiff base, competing with a water molecule and D253. The other rotamer configuration positioned the carboxylic group of E123 in a remote position from the Schiff base. A water molecule or the carboxylic group of D253 can then form a hydrogen bond with the Schiff base proton. The resulting effect on the strength of the N–H bond of the Schiff base was acutely determined by Raman spectroscopy.

The vibrational band of the Schiff base in WT has so far been treated as a single vibration [32]. In light of our results for the T127A variant, we revisited the Raman spectrum of the WT (Figure 5, black trace). The slight asymmetry in band shape was again considered by fitting two Gaussians, resulting in maxima at 1665 cm−<sup>1</sup> and 1661 cm<sup>−</sup>1.

From these results, we assigned the lower frequency of the C=N–H vibration at 1661 cm−<sup>1</sup> to the Schiff base being hydrogen-bonded to a water molecule and/or directly to D253. In this scenario, the substitution of T127 by A did not affect the frequency of the Schiff base vibration, as this residue did not interact directly with D253. This explained the low-frequency shoulder of the C=N–H band at 1661 cm−<sup>1</sup> which did not shift after amino acid exchange. The high-frequency band at 1674 cm−<sup>1</sup> in T127A corresponded to the population in which E123 was hydrogen bonded to the Schiff base and, with respect to the WT, lacked the hydrogen bond to the threonine (see Figure 2, right panel).

We tested this hypothesis by recording resonance Raman spectra for two other *Cr*ChR2 variants: T127S, where the hydroxyl group of the threonine side chain is conserved; and E123D, where the carboxylate side chain is shorter. In agreement with the WT and the T127A variant, all spectral features of the retinal chromophore were retained in the spectra of the T127S (blue traces in Figure 5) and E123D variants (red traces in Figure 5) with only minor differences. Analysis of the Schiff base vibration (Figure 5b) showed that the C=N–H band in E123D could be fitted sufficiently well with one Gaussian, whereas for T127S, two Gaussians were needed to fit the band. The band with a maximum at 1661 cm−<sup>1</sup> was present in both variants, supporting the hypothesis that this Schiff base vibration did not involve interaction with E123.

In T127S, where the exchange retained the hydroxyl group that is hydrogen-bonded to E123 (see Figure 2), the high-frequency shoulder of the peak was only slightly blue-shifted by 5 cm<sup>−</sup>1, showing strong similarities to the C=N–H vibrational band of the WT. In E123D, however, the high frequency band was not present at all as a consequence of the shorter amino acid side chain length that prevented formation of a hydrogen bond with the Schiff base.

We focused our QM/MM calculations on the structure with the rotamer configuration of E123 in the upward orientation to be able to accept a hydrogen bond from the Schiff base proton. This configuration was in fact the one in which the C=N–H vibration exhibited the strongest effect in our T127 variants. Although the spectroscopic data showed that the hydrogen bond between the Schiff base and E123 was strongly strengthened upon replacement of T127, the distance between the Schiff base nitrogen and E123 was unchanged in the optimized QM/MM structure of the T127 variants. Thus, we went on to analyze the effect of the amino acid exchange on the length of the C=N and the N–H bonds of the Schiff base in our simulations (Table 1).

The length of the C=N bond was not significantly affected in T127S and T127A as compared to the WT. However, the hydrogen-bonding network in the vicinity of RSBH<sup>+</sup> was found to have a notable effect, with the presence or absence of hydroxylic group in the sidechain at position 127. The difference in N–H bond length in WT and T127S was negligible (~0.004 Å), which was attributed to the hydrogen-bonding network facilitated by the hydroxylic side chain. In contrary, the absence of such a network in T127A increased the N–H bond length by 0.016 Å. The longer N–H bond length in T127A was in line with our experimental findings that this variant accelerated deprotonation of RSBH<sup>+</sup> (see Section 3.3) and it also supported the high frequency of the C=N–H vibration of 1674 cm<sup>−</sup>1. QM/MM simulations yielded a frequency difference of 19 cm−<sup>1</sup> between WT and the T127A variant (Table 3), which reproduced the frequency upshift upon removal of the hydroxylic group of threonine but overestimated the experimental difference of 9 cm−<sup>1</sup> (Figure 5B). The exceptionally high frequency of the Schiff base C=N–H vibration in the T127A variant was, therefore, be assigned to the missing hydrogen-bond between A127 and E123 and the increase in distance between N(RSBH+) ... O(E123).


**Table 3.** Calculated vibrational frequencies of the C=N–H vibration using the QM/MM method.

#### *3.2. Channel Conductance of the T127 Variants*

The role of T127 in the functional mechanism of the ion channel was scrutinized via electrophysiological experiments on WT and its T127 variants expressed in *Xenopus* oocytes. The T127A and the T127S variants showed a strong decrease in the current amplitude (for T127A: Figure 6A, red traces). On average, the variants exhibited only 9 ± 1% (n = 29 for T127A and n = 18 for T127S) of the WT currents. From the residual current, we determined the kinetics of the channel closing upon switching off the light (Figure 6B). It was slower than the WT (τ = 12 ms), with τ = 22 ± 3 ms (n = 9) for the T127A variant and with τ = 23 ± 3 ms (n = 9) for the T127S variant.

**Figure 6.** Two-electrode voltage-clamp (TEVC) recordings of *Cr*ChR2-WT, T127A, and T127S variants expressed in *Xenopus* oocytes. (**A**) Light-induced currents at −60 mV and +60 mV holding potential. The traces for the T127A variant (red) were scaled by a factor of 10 to compare to the WT (black). (**B**) Histogram of the current amplitudes and closing kinetics. Currents were normalized to the WT at −60 mV. Bars represent the mean and the s.e.m. (n = 14 for WT, n = 29 for T127A variant, and n = 18 for T127S variant). (**C**) Time-resolved currents from T127A in HEK293 cells. Raw data of an average of eight traces at −80 mV to +40 mV in steps of 40 mV. The red curve shows a single exponential fit of the current decay with a time constant of 10 ms.

In time-resolved experiments on the T127A variant expressed in HEK293 cells, only very small current amplitudes were recorded. Signals from eight kinetic traces were averaged (Figure 6C) to be able to determine the channel closing kinetics. Similarly to the E123T variant [33] we observed a very fast outward current independent of the holding potential, which monitored a vectorial charge transport within the first 200 μs. The recovery kinetics after pulsed excitation exhibited single exponential behavior with a time constant of 10 ms, which was similar to the WT and consistent with our results from the TEVC recordings.

#### *3.3. Influence of T127 on the Photocycle Kinetics*

As the channel conductance was drastically reduced by the T127 replacement but the absorption spectrum of dark-state *Cr*ChR2 was hardly affected, we performed time-resolved UV/Vis spectroscopy (Figure 7). Excitation by a nanosecond laser pulse led to the depletion of the dark state of *Cr*ChR2, which was reflected by a loss of absorption at 470 nm (blue traces in Figure 7). The P1 <sup>500</sup> state rose at times beyond the resolution of our experiment, but the decay was resolved at 530 nm (green traces). The rise of the blue-shifted P2 <sup>380</sup> state with a deprotonated retinal Schiff base was recorded at 380 nm (red traces). Rise and decay of the succeeding P3 <sup>530</sup> intermediate were observed at 530 nm and the latter correlated with channel closure in *Cr*ChR2 [34]. The kinetics of the desensitized state P4 <sup>480</sup> were also observed at 530 nm.

The P2 <sup>380</sup> state was formed faster in the T127A variant (Figure 7, top panel) and decayed slower than in the WT (Figure 7, bottom panel), i.e., the Schiff base deprotonated at an earlier stage and was reprotonated later. The kinetics of P2 <sup>380</sup> rise in T127A had a half-life of t1/<sup>2</sup> = 0.7 μs, one order of magnitude faster than the rise of the P2 <sup>380</sup> state in WT. The decay of this blue-shifted intermediate in T127A, with a t1/<sup>2</sup> = 9 ms, was more than four times slower than in WT. The decay kinetics of the P2 <sup>380</sup> state were altered at the expense of the P3 <sup>530</sup> intermediate, which rose later and was significantly reduced in transient amplitude. This observation tallied the reduced channel conductance based on the correlation of the lifetimes of the open state and the P3 <sup>530</sup> state. The decay of the non-conductive P4 <sup>480</sup> state was not influenced by the T127A exchange. The less invasive replacement of threonine by serine in the T127S variant left the photocycle kinetics of *Cr*ChR2 unchanged (Figure 7, middle panel). Thus, we inferred that the hydroxyl group at position 127 was essential to a WT-like photoreaction, with a high accumulation of the late red-shifted intermediate P3 530.

**Figure 7.** Kinetics in the UV/Vis range recorded after pulsed laser excitation (450 nm). Traces at 380 nm (red lines), 470 nm (blue), and 530 nm (green) are shown for each variant. While the T127A variant (top panel) showed a prolonged lifetime of the P2 <sup>380</sup> state and reduced P3 <sup>530</sup> formation, the T127S variant (middle panel) had very similar kinetics to the WT (bottom panel).

#### *3.4. FTIR Di*ff*erence Spectroscopy on the T127 Variants*

FTIR spectroscopy was applied to gather information on the structural changes of the T127 variants (Figure 8). For comparison, difference spectra of *Cr*ChR2-WT were chosen at time points that represented mainly the intermediates P2 <sup>380</sup> (300 μs) and P3 <sup>530</sup> (6.7 ms), and were compared to the difference spectra of the T127A and T127S variants recorded at 8.4 ms after pulsed excitation. At this time, the P2 <sup>380</sup> intermediate was predominant in the variant with minor contributions from P3 <sup>530</sup> and P4 <sup>480</sup> intermediates.

The infrared difference spectra basically confirmed the observations made in the visible spectral range (Figure 6). The low accumulation of the P3 <sup>530</sup> state in the T127A variant was observed by comparing the difference spectrum at 8.4 ms with the one at 6.7 ms of the WT. The band at 1737 cm−<sup>1</sup> assigned to the deprotonation of the proton donor to the Schiff base, D156 [16], was reduced in intensity, which indicated low accumulation of the deprotonated D156 species. This observation was expected as soon as the lifetime of the P2 <sup>380</sup> state (with deprotonated Schiff base) was extended, as here in the T127 variant. The negative band at 1556 cm−<sup>1</sup> in the difference spectra of WT was due to the ethylenic stretching vibration of the retinal in ground-state ChR2 [32]. This band was hardly seen in T127A due

to spectral overlap by a positive contribution that is absent in the WT. Instead, a broad negative band at 1533 cm−<sup>1</sup> was registered, which was a marker for the contribution of the P4 <sup>480</sup> photoreaction [35].

**Figure 8.** Time-resolved FTIR difference spectra of ChR2-WT (black spectra, taken from Reference [13]), T127A (red spectrum), and T127S variants (blue spectrum). The ChR2-WT spectrum at 300 μs and 6.7 ms was dominated by the P2 <sup>380</sup> and P3 <sup>530</sup> states, respectively. The spectra of the T127 variants were recorded at 8.4 ms, at which the P2 <sup>380</sup> with smaller contributions of P3 <sup>530</sup> was predominant.

#### **4. Discussion**

T127 of *Cr*ChR2 is localized in close vicinity to the retinal Schiff base and is hydrogen-bonded to E123 which, in turn, is the hydrogen bond acceptor of the Schiff base. The position of the threonine is also strategic for Schiff base reprotonation, as its hydroxylic group is supposed to be part of the pathway between the proton donor D156 and the Schiff base. There is no direct hydrogen-bonding network in the ground state that connects the RSBH<sup>+</sup> and D156, but it is only with the rise of the photocycle intermediate P3 <sup>530</sup> that this connection is formed [13].

T127A substitution removed the hydroxylic group of the threonine, which affected the photocycle kinetics. The P2 <sup>380</sup> intermediate, during the lifetime of which the Schiff base is deprotonated, had a ~10 times faster risetime, meaning that the deprotonation of the Schiff base was facilitated in this variant. The acceleration of the rise of the P2 <sup>380</sup> intermediate may have been due to the stronger hydrogen-bonding interaction between the Schiff base and E123 that we reported in the present work.

The lifetime of the P2 <sup>380</sup> state was longer in the T127A variant and, as a consequence, the accumulation of the subsequent intermediate P3 <sup>530</sup> was very low. A longer lifetime of the state with a deprotonated Schiff base is a sign that the reprotonation pathway was blocked in T127A. The T127S variant conserved the hydroxyl group of the side chain, and we showed here that the photocycle kinetics were almost indistinguishable from the WT. We concluded, therefore, that the hydrogen-bonded network between D156 and the Schiff base, which is necessary for the reprotonation of the latter, involves the hydroxylic group of T127. The formation of this chain of hydrogen bonds is necessary for proton translocation and marks the transition between the P2 <sup>380</sup> and P3 <sup>530</sup> states.

T127 is important for the opening of the ion channel after light activation, since we have shown here that the T127A, as well as the more conservative T127S variants, showed less than 10% of the WT current in electrophysiology experiments. We note, though, that the reduced conductance may also have been related to lower expression yields of the variants in the different hosts (*Xenopus oocytes* and HEK cells). It has previously been shown that the channel closes with the decay of the P3 530 intermediate [36,37], but the channel opening is an optically silent transition in the visible range [38]. We confirmed from the present work that channel opening in *Cr*ChR2 is not related to changes in the photocycle kinetics as recorded by time-resolved UV/Vis spectroscopy. The T127S variant, in fact, showed WT-like photocyle kinetics, whereas a long-lived P2 <sup>380</sup> intermediate occurred in the T127A variant with both variants exhibiting low channel conductance.

A threonine residue at this position was conserved for most of the cation- and anion-conducting channelrhodopsins. It is also present in BR [12,39], where exchanges to valine or alanine resulted in blue shifts in the absorption maxima by 146 nm and 28 nm, respectively, with minor effects on pumping activity [3]. By contrast, the exchange of threonine to serine and alanine in *Cr*ChR2 had only negligible effects on the electronic properties of the retinal chromophore, as registered by its visible absorption. This result is particularly interesting in light of the exceptionally high vibrational frequency recorded for the C=N–H vibration of the retinal Schiff base in T127A, as the hydrogen-bonding of the retinal Schiff base has been proposed to be a molecular determinant of the opsin shift [40].

QM/MM calculations showed nearly identical excitation energies for WT, T127S, and T127A variants. The same trend was also observed in the experimental absorption maxima of the retinal chromophore of *Cr*ChR2. However, a large shift of 19 cm−<sup>1</sup> was calculated for the C=N–H vibration in the alanine variant, but no shift for the serine variant, again confirming the resonance Raman spectroscopic results. Since our QM/MM models were able to reproduce the relative trends in the UV/Vis and vibrational spectra of the variants, we considered these reliable for our analysis of the molecular changes. In the T127A variant, the hydrogen bond between the threonine and E123 was missing, affecting the hydrogen-bonding network around the Schiff base. To compensate for the missing hydrogen bond, the interaction between E123 and the Schiff base (N(RSBH+) ... O(E123)) increased, resulting in a longer N–H bond which manifested in the upshift of the C=N–H vibrational frequency. The elongation of the Schiff base N–H bond was 0.016 Å, and possibly accelerates the deprotonation of the Schiff base in this variant. Hence, the stronger interaction between the Schiff base and E123 provided an explanation for the faster rise time of the P2 <sup>380</sup> intermediate in the T127A variant, in a similar way to the D85E variant of bacteriorhodopsin [41].

Detailed analysis of the Schiff base vibration (coupled mode of the C=N stretching and the N–H bending vibration) by resonance Raman spectroscopy revealed the presence of two overlapping bands. The frequency shifts of these vibrational bands in the different variants were compared to the WT and their presence can be rationalized on the basis of the molecular model proposed in Reference [31]. In this model, the E123 side chain had one rotamer pointing towards the Schiff base and another rotamer pointing to a different hydrogen-bonding network that gave rise to two different vibrational bands of the C=N–H mode. This interpretation was supported by the band shape of the Schiff base vibration in the E123D variant, which was missing the high-frequency component. We can therefore lend support to the model of two rotamers of E123 in *Cr*ChR2, where a direct hydrogen bond with the Schiff base was formed in only one of the two configurations. The presence of two rotamer configurations may be essential for a voltage-sensing mechanism involving E123 [42]. While this seems to be a plausible scenario, further experimental evidence needs to be collected to support this model.

**Author Contributions:** Conceptualization, R.S. and J.H.; Funding acquisition, R.S. and J.H.; Investigation, D.E., N.K., M.S., C.B., R.K.K., K.H., D.H. and I.S.; Writing—original draft, R.S. and N.K.; Writing—review & editing, J.H. and M.S.

**Funding:** The research was funded by the German Research Foundation via the SFB-1078 projects B3 (to J.H.) and B4 (to R.S.). I.S. thanks the SFB 1078 for support within the Mercator program. R.K.K. acknowledges support from the Lady Davis Trust for Shunbrun postdoctoral fellowship. I.S. gratefully acknowledges funding by the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation program (Grant No. 678169 "PhotoMutant").

**Acknowledgments:** The publication of this article was funded by Freie Universität Berlin.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Light Stimulation Parameters Determine Neuron Dynamic Characteristics**

**Alexander Erofeev 1,\*, Evgenii Gerasimov 1, Anastasia Lavrova 1,2,3, Anastasia Bolshakova 1, Eugene Postnikov 4, Ilya Bezprozvanny 1,5 and Olga L. Vlasova 1,\***


Received: 8 August 2019; Accepted: 30 August 2019; Published: 5 September 2019

**Featured Application: This report highlights the importance of light stimulation parameters (frequency, duration, intensity) on the activity of neurons expressing channelrhodopsin-2. These results will allow neuroscientists to stably activate neurons during a repeated light pulse train in optogenetic experiments with ChR2.**

**Abstract:** Optogenetics is a recently developed technique that is widely used to study neuronal function. In optogenetic experiments, neurons encode opsins (channelrhodopsins, halorhodopsins or their derivatives) by means of viruses, plasmids or genetic modification (transgenic lines). Channelrhodopsin are light activated ion channels. Their expression in neurons allows light-dependent control of neuronal activity. The duration and frequency of light stimulation in optogenetic experiments is critical for stable, robust and reproducible experiments. In this study, we performed systematic analyses of these parameters using primary cultures of hippocampal neurons transfected with channelrhodopsin-2 (ChR2). The main goal of this work was to identify the optimal parameters of light stimulation that would result in stable neuronal activity during a repeated light pulse train. We demonstrated that the dependency of the photocurrent on the light pulse duration is described by a right-skewed bell-shaped curve, while the dependence on the stimulus intensity is close to linear. We established that a duration between 10–30 ms of stimulation was the minimal time necessary to achieve a full response. Obtained results will be useful in planning and interpretation of optogenetic experiments.

**Keywords:** membrane current; hippocampal neurons; optogenetics; light stimulation; channelrhodopsin-2

#### **1. Introduction**

Large scale neural activity patterns formed due to an interplay between dynamics of neurons and their network interconnections are principal fingerprints of brain functionality. At the same time, age and genetic disruption of neuronal connections can lead to neurodegenerative diseases such as Alzheimer's, Parkinson's, Huntington's and others. The number of incidence cases is increasing due to population ageing associated with new advances in medicine and technology. Therefore, the key problem is the difficult search for mechanisms of neurodegenerative disorders and new approaches to their treatment.

The most widespread conventional method for studying mechanisms of neuronal activity uses electrical stimulation of electrodes placed in the extracellular space. Electrophysiology allows for easy control of temporal resolution; nevertheless, in most cases it activates a lot of neurons simultaneously but not individual neurons. This is a significant disadvantage since specific types of neurons have an intrinsic activity pattern [1,2]. There is also an intracellular stimulation method which provides the necessary spatial and temporal resolutions, but its application is restricted to neuronal culture or brain slices [3].

One of the methods that allows for the study of the activity of certain neurons in vivo is the optical stimulation of neurons (optogenetics). In comparison with the methods mentioned above, it offers several advantages, such as a high spatio-temporal resolution with a parallel stimulation of certain brain areas [4,5]. The main principle of this method is based on the expression of light-sensitive ion channels called opsins in neuronal membranes. One of them, channelrhodopsin-2 (ChR2) isolated from the green alga *Chlamydomonas reinhardtii* [6,7], and its modifications is used practically for neuron excitation. ChR2 is activated with blue light (470 nm), which in turn induces a photoreceptor current due to non-selective cation flux into the cytoplasm of cells [6–8].

Since the method was developed, optogenetics has been widely used in neuroscience [4,8–11]. Recently, optogenetics has also been applied to investigate neurodegenerative diseases (an area of scientific interest of the authors). For example, the authors [12] used step-function opsin (SFO) for the long-term excitation of the hippocampal perforant pathway in amyloid precursor protein (APP) transgenic mice. As a result, a prolonged light excitation led to an increase in the level of amyloid deposits of peptide of 42 amino acid residues (Aβ42) that allows for the determination of a certain functional pathology in specific neuronal circuits in Alzheimer's disease. The review [13] contains a number of studies in which optogenetics is successfully used as a tool for activating or inhibiting specific regions or certain neurons involved in the development of neurodegenerative and neurological diseases such as Parkinson's disease [14], Huntington's disease [15], and epilepsy [16,17].

Despite the fact that optogenetics is one of the most developing areas of neurobiology and a widely used technique, there are some open questions related to the parameters of light stimulation for neurons expressing opsins. In this paper, we would like to emphasize the importance of light parameters during repeated light stimulation. Thus, our main goal is to determine the relationship between the light stimulation parameters (frequency, duration, intensity) and neuron activity during repeated light stimulation and define the optimal parameters for the stable activity of neurons.

#### **2. Materials and Methods**

#### *2.1. Animals*

The breeding colony of wild type mice of the same strain (C57BL/6J background, #000664) obtained from the Jackson Laboratory was established and maintained in a vivarium with 4–5 mice per cage and a 12 h light/dark cycle in the animal facility. All procedures were approved by principles of the European convention (Strasburg, 1986) and the Declaration of International Medical Association regarding the humane treatment of animals (Helsinki, 1996).

#### *2.2. Hippocampal Primary Culture*

The hippocampal cultures of mice were established from postnatal day 0–1 pups and maintained in culture as described earlier [18]. Briefly, after dissection and dissociation, neurons were plated on coverslips (pre-coated with poly-D-lysine, 0.1 mg/mL, #27964, Sigma, St. Louis, MI, USA) and cultured in neurobasal A medium with an addition of 1% fetal bovine serum (FBS) and 2% optimized neuronal cell culture serum-free supplement B27. On the third day of in vitro culture (DIV3), cytosine arabinoside (Ara-C) (40 M, Sigma, #C1768) was added to prevent the growth of glial cells. At DIV7 and DIV14, 50% of the medium was exchanged with fresh neurobasal A medium containing 2% B27 without FBS. At DIV7, neurons were transfected using the calcium phosphate method.

#### *2.3. The Calcium Phosphate Method of Transfection*

At DIV7, neurons were transfected using the calcium phosphate method with a mammalian transfection kit (#631312, Clontech Laboratories Inc., Mountain View, CA, USA) according to the manufacturer's protocol with recommendations from [19]. Due to the fact that the excitation length (470 nm) of green fluorescent protein (GFP) coincides with the excitation length of chanalrhodopsin-2, we used the marker plasmid pCSCMV:tdTomato (#30530, Addgene, Watertown, MA, USA), which encodes a red fluorescent protein with an excitation wavelength of 530 nm. For multiple plasmid delivery, we used the calcium phosphate co-transfection method. Plasmids were mixed at a ratio of 3:1, with 3 parts containing the plasmid of interest.

#### *2.4. Whole Cell Patch Recordings in Hippocampal Cultures*

The recording of neuronal light-induced activity was performed with whole-cell voltage- and current-clamp techniques after 14 days of cultivation; at this time, neurons mature and form a stable network [20,21]. Whole-cell recordings were conducted in artificial cerebrospinal fluid (ACSF) external solution, which contains 124 mM NaCl, 26 mM NaHCO3, 10 mM glucose, 5 mM KCl, 2.5 mM CaCl2, 1.3 mM MgCl2 and 1 mM NaH2PO4 (pH 7.4, 290–310 milliosmole (mOsm)) [22]. An Olympus IX73 inverted microscope was used with the 40× objective and appropriate filters for fluorescence imaging. This microscope was equipped with a 4-channel light-emitting diode (LED) driver Thorlabs DC4104 to localize and target fluorescent ChR2-expressing neurons by expression of the tdTomato fluorescent protein.

Patch pipettes, pulled from borosilicate glass (1.5 mm outside diameter (OD), 0.86 mm inside diameter (ID), 3-5 MOm) were filled with a solution containing 140 mM K-Gluconate, 2 mM MgCl2, 2 mM NaCl, 2 mM ATP-Na2, 0.3 mM GTP-Mg, 10 mM 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES) (pH 7.35, 290–300 mOsm). Conventional whole-cell patch-clamp recordings were obtained via a MultiClamp 700 B double patch amplifier and a Digidata 1440A (Molecular Devices, Sunnyvale, CA, USA) coupled to the acquisition software pClamp 10.7. The holding potential in voltage-clamp mode was −70 mV, uncorrected for any liquid junction potential (in the range of −10 mV to −8 mV) between the internal and external solutions. After DIV14, light-evoked neuronal activity has been detected using the whole-cell patch clamp technique as shown in Figure 1. We have used 10 primary cultures for the measurement of membrane currents and potentials.

**Figure 1.** Primary hippocampal neuron transfected with FCK-ChR2-GFP and pCSCMV:tdTomato during the recording of photocurrents.

#### *2.5. Optogenetic Method*

During the recording of neuronal activity, a pulse train of blue light (470 nm) was performed with different parameters (frequency, duration, intensity). In our experiments, we used a frequency range of 1–30 Hz in increments of 5 Hz and a pulse duration range of 1–5 ms in increments of 1 ms, 10–50 ms in increments of 10 ms, and 100–500 ms in increments of 100 ms. Ten light stimuli were used per frequency or duration (Figure 2).

**Figure 2.** Photocurrents recorded using the whole-cell voltage-clamp technique at maximal intensity and light duration. t = 100 ms.

The maximum intensity of the blue LED (LED4D067, 470 nm, Thorlabs Inc., Newton, NJ, USA) was 35 mW mm−<sup>2</sup> with a maximum photo flux of 250 mW. The values of light intensities were normalized, taking the maximum value in each sequence as unity, i.e., all indicated values are shown as relative ones.

The LED source has a modulation voltage in the range of 0–10 V and a current in the range of 0–1000 mA. We measured the intensity at the maximum current, i.e., 1000 mA, which corresponds to a modulation voltage of 10 V. To change the intensity, we connected a pulse generator to the LED source and changed the modulation voltage from 0 V to 10 V; while we considered that at a voltage of 10 volts, we get the maximum intensity. The values of light intensities were normalized, taking the maximum value in each sequence as unity.

#### *2.6. Statistical Analysis*

The mathematical software OriginPro 9.0 and GraphPad Prism 7 were used to process the data and obtain standard errors.

#### **3. Results**

To define the relationship between the activity of the pyramidal neuron expressing ChR2 and the parameters of the light stimulus, we carried out experiments with a 10-light pulse train. We picked up the frequency at which the action potential (AP) was generated for each light stimulus (Figure 3). We considered the generation of action potentials because the membrane current changed at all frequencies used in this study.

These results demonstrated a reduced number of AP with increasing frequency under blue light pulsing. The frequency at which the AP was generated for each light stimulus was in the range of 1–5 Hz. Further measurements were carried out at a frequency of 1 Hz. These results are similar to the data published previously [23], but the range of optimal stimulation frequency was more narrow in our experiments. The reason for this discrepancy requires further study.

**Figure 3.** Action potential generation under blue light stimulation at different frequencies.

We have studied the effect of the light pulse duration in the range of 1–5 ms, 10–50 ms and 100–500 ms on the amplitude of photocurrents at various LED intensities; the value of maximum intensity has been normalized to unity. Each point in Figure 4 represents the mean value of the measured amplitudes of the response to a 10-pulse train with fixed duration and intensity (see Figure 2).

**Figure 4.** Dependence of the membrane current amplitude for pyramidal neurons on the duration of light stimulation at different light intensities (the value of intensity maximum is taken as 1), time–light pulse duration, *n* = 10. Plots of mean values with error bars indicate the standard error (SE). To obtain these mean values, fourth-order polynomial smoothing over neighboring points was performed in GraphPad Prism 7.

As shown in Figure 4, the dependency of the photocurrent amplitude on the light pulse is described by a right-skewed bell-shaped curve. The obtained dependency was identical for all light intensities used in the experiment. The obtained results agree with results published earlier [7,24].

It should be noted that in the range of 1–20 ms, the value of this amplitude increases, while at higher values of pulse duration (30–500 ms), the amplitude decreases. An explanation for this fact may be based on the kinetic model of the ChR2 photocycle, in which three channel states (open (O), desensitized (D) and closed (C)) were considered [7]. The three states model was the first model describing a photocycle of channelrhodopsin-2. However, after the spectral analysis of ChR2 [25], an existence of four kinetic intermediate states (*P*1, *P*2, *P*<sup>3</sup> and *P*4) was indicated. Based on this information, four [24,26] and six [27] state models have been developed. To explain our findings, we utilized the four-state model of ChR2 photocycle, which consists of two open states (*O*<sup>1</sup> → *O*2) and two closed states (*C*<sup>1</sup> → *C*2). Thus, the states defined by the spectral analysis can be interpreted as follows: *O*<sup>1</sup> → *O*<sup>2</sup> corresponds to the states *P*<sup>2</sup> and *P*<sup>3</sup> (open states with the time constants of 1 ms and 10 ms, respectively) and state *C*<sup>2</sup> (the time constant ~5s) corresponds to the state *P*<sup>4</sup> (desensitized). The state of *C* is the ground state of the channel and corresponds to *P*0. We suppose that the decrease of photocurrents after 30 ms is due to the transition in the *P*<sup>4</sup> state, i.e., inactivation of ChR-2.

We have also studied the time interval tau (τ), which is needed to reach the maximum photocurrent (Figure 5) for different pulse durations (t) atmaximum intensities (Imax).

**Figure 5.** Light-induced currents with a pulse duration of 100 ms.

We analyzed the relationship between τ and light pulse duration (t) (Figure 6). According to dependence in the range of 10–30 ms, τ corresponds to the light pulse duration.

**Figure 6.** Dependence of τ on the duration of light pulses.

After t = 50 ms, the time τ did not change. This suggests that light stimulation with a duration longer than 50 ms is impractical, because the response of neurons expressing ChR2 is not stable.

Figure 7 shows an example of ChR2 photocurrents of the primary hippocampal neuron for the first light stimuli at various pulse durations (1–5 ms, 10–50 ms, 100–500 ms).

**Figure 7.** Light-induced currents recorded from one hippocampal neuron at different light pulse durations in the voltage-clamp mode of the whole-cell configuration: 1–5 ms (**left**), 10–50 ms (**center**), 100–500 ms (**right**). Light intensity is at the maximum value.

As shown in Figure 7 (left panel), the maximum photocurrent is already achieved after the light stimulus itself in the case of short pulse durations. At higher pulse durations (Figure 7, middle panel), τ either coincides with the pulse duration, or achieves maximum value before the end of the light stimulus itself. At large values of pulse duration (Figure 7, right panel), the time interval τ is always shorter than the duration of the light stimulus.

In the next series of experiments, we studied the effect of light intensity on photocurrent amplitude at different light pulse durations.

The obtained curves (Figure 8) demonstrate a mostly linear dependency that is closed to the generation of sodium currents [28].

**Figure 8.** Dependence of the current amplitude (the average value for 10 light stimuli) for hippocampal neurons on the intensity of exposure at different light pulse durations: (**A**) 1–5 ms, (**B**) 10–50 ms, (**C**) 100–500 ms, *n* = 10. Plots of mean values with error bars indicate the standard error (SE). To obtain these mean values, fourth-order polynomial smoothing over neighboring points was performed in GraphPad Prism 7.

The obtained dependency of photocurrent amplitude on light intensity corresponds with previously published data [23].

#### **4. Discussion**

In this study, we analyzed a range of ChR2 stimulation conditions in experiments with primary hippocampal cultures.

We have determined an optimal frequency of light stimuli for generating APs in the current-clamp recording configuration. The optimal frequency range was between 1–5 Hz. Stimulation with frequencies less that 1 Hz was insufficient to generate APs, and stimulation with frequencies over 5 Hz results in the loss of fidelity of responses.

We have observed that the amplitude of ChR2 currents depends non-linearly (right-skewed bell-shaped curve) on the pulse duration. ChR2 photocurrent amplitudes were stable in the range of 10–30 ms, however, at low (1–5 ms) and large (100–500 ms) pulse duration values, the amplitudes changed in an almost stochastic manner (Figure 2).

Amplitude differences at low (1–5 ms) values of pulse duration are presumably explained due the fact that ChR2 only reaches the *O*<sup>1</sup> state and cannot change to *O*<sup>2</sup> with maximum bandwidth. At the same time, the observed amplitude differences at large pulse durations are explained due to the desensitization and degradation of ChR2 and the *P*<sup>4</sup> state [7]. Thus, we can conclude that the optimal interval of light stimuli is in the range of 10–30 ms.

The relationship between the photocurrent amplitude and the intensity of the light stimulus can be explained by the fact that increasing intensity leads to an increase of *O*<sup>2</sup> channel state duration. At intensities above 35 mW mm<sup>−</sup>2, the amplitude of the photocurrent of ChR2 may decrease; this assumption needs further research.

Thus, we found that the following parameters of light stimulation: frequency (F) = 1–5 Hz, t = 10–30 ms, I ≤ Imax are optimal for multiple light stimulation, i.e., the activity of neurons expressing ChR2 will be stable throughout the stimulation period. This is also shown in Figure 9.

Figure 9 demonstrates the cumulative effect of duration and intensity on the activity of neurons expressing ChR2. Thus, the photocurrent amplitude is almost constant for the pulse duration, which is equal to 10 ms in the full range of intensities, while there is a wide variation of the amplitude values at other durations (1 ms and 300 ms) and intensities.

**Figure 9.** Graph representing the relative difference between the first two current spike amplitudes as a function of the light impulse duration and their relative intensity. The maximal difference is normalized to unity; the relative light intensities are indicated by markers. This graph was designed as follows: the chosen pulse duration was assigned to a value of the difference between amplitudes of the first and second pulses.

#### **5. Conclusions**

In this study, we determined the relationship between the light stimulation parameters (frequency, duration, intensity) and neuron activity during repeated light stimulation and defined the optimal parameters for the stable activity of neurons. It was determined that the optimal frequency of light stimuli for generating APs is in the range of 1–5 Hz. We demonstrated that the dependency of the current amplitude on light pulse duration is described by a right-skewed bell-shaped curve, while the dependence on stimulus intensity is close to linear. We discovered that the light pulses between 10–30 ms in the full range of intensity are optimal for activation of ChR2 in cultured hippocampal neurons. We established that a 10 ms duration of stimulation was the minimal time necessary to achieve full response. The obtained results will be useful in the planning and interpretation of optogenetic experiments.

**Author Contributions:** Conceptualization, O.L.V. and A.E.; methodology, O.L.V.; validation A.L. and O.L.V.; formal analysis, A.E., E.G. and A.L.; investigation, A.E. and E.G.; resources, A.B. and I.B.; writing—original draft preparation, A.E., A.L. and O.L.V.; writing—review and editing, E.P. and A.B.; visualization, A.E.; supervision, O.L.V.; project administration, I.B.; funding acquisition, I.B. and E.P.

**Funding:** This research was funded by the state grant 17.991.2017/4.6 (IB) and by the Russian Science Foundation Grant No. 19-15-00184 (IB). The financial support was divided in the following way: experiments depicted in Figures 1–6 were supported by the state grant 17.991.2017/4.6, experiments depicted in Figures 6 and 7 were supported by the Russian Science Foundation Grant No. 19-15-00184, and experiments in Figures 8 and 9 were supported by the Russian Science Foundation, project 19-15-00201 (AL and EP).

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Review* **Two-Photon Excitation of Azobenzene Photoswitches for Synthetic Optogenetics**

#### **Shai Kellner and Shai Berlin \***

Department of Neuroscience, Ruth and Bruce Rappaport Faculty of Medicine, Technion-Israel Institute of Technology, Haifa 3525433, Israel; shaikellner@gmail.com

**\*** Correspondence: shai.berlin@technion.ac.il

Received: 1 December 2019; Accepted: 20 January 2020; Published: 23 January 2020

**Abstract:** Synthetic optogenetics is an emerging optical technique that enables users to photocontrol molecules, proteins, and cells in vitro and in vivo. This is achieved by use of synthetic chromophores—denoted photoswitches—that undergo light-dependent changes (e.g., isomerization), which are meticulously designed to interact with unique cellular targets, notably proteins. Following light illumination, the changes adopted by photoswitches are harnessed to affect the function of nearby proteins. In most instances, photoswitches absorb visible light, wavelengths of poor tissue penetration, and excessive scatter. These shortcomings impede their use in vivo. To overcome these challenges, photoswitches of red-shifted absorbance have been developed. Notably, this shift in absorbance also increases their compatibility with two-photon excitation (2PE) methods. Here, we provide an overview of recent efforts devoted towards optimizing azobenzene-based photoswitches for 2PE and their current applications.

**Keywords:** two-photon; azobenzene; optogenetics; photoswitch

#### **1. Introduction**

Photoreceptors are protein light-detectors that allow the cell/organism to respond to its environment [1]. Most, if not all, photoreceptors are a two-component system: a genetically-encoded protein coupled to a chromophore. Following photon-absorption, chromophores undergo unique alterations (for instance in geometry or charge [2,3]), resulting in the modulation of the protein partner to which each chromophore is bound to. In most instances, this process is completely reversible, whereby removal of light (i.e., return to darkness) facilitates the relaxation of the chromophore to its initial, thermally stable state. This also restores the function of the protein back to as it were prior illumination. These features, explicitly non-invasive reversible control over proteins, have made it possible to develop fantastic genetically-encoded optical tools—denoted 'optogenetic' tools—for the manipulation of defined subsets of cells [4]. Many tools today are based on naturally-occurring photoreceptors, for instance opsins [5–8]. However, the need to photocontrol other signaling mechanisms, proteins or molecules, have pushed forward the development of complementary strategies, explicitly Synthetic optogenetics [9,10].

Synthetic optogenetics relies on the use of non-natural, synthetic chromophores photoswitches—to relay photon absorption to modulation of structure and function of their target molecule (Figure 1a). Photoswitches do not require a unique chromophore binding pocket or protein domains as in the case of chromophores (e.g., [11]). Instead, photoswitches can be directly conjugated to various locations on their target molecule with use of different genetic or chemical handles (see below) [9,12]. However, unlike chromophores that can be synthesized by various cells (even if the cells are not intrinsically light-sensitive), photoswitches need to be supplemented to the preparation. This is both a blessing and a curse. Blessing—since the preparation remains naïve until the photoswitch

is added. Curse—it adds another layer of complexity to the experiment, especially in vivo, such as injection prior experimentation.

**Figure 1.** Synthetic optogenetics. (**a**) Cartoon depiction of a photoswitchable glutamate receptor (space filling based on PDB: 3KG2). For simplicity, only two of four subunits are presented. In addition, only one subunit is shown to tether the photoswitch (color-coded shapes). The photoswitch consists of a ligand (purple rectangle), an azobenzene-core (red hexagon) and a maleimide for attachment (green circle). In the dark, the photoswitch is in *-trans* and channel is closed (leftward cartoon, light blue). Photoisomerization of the photoswitch to *-cis* by red-shifted light (*h*ν) allows insertion of the glutamate headgroup (purple rectangle) into its binding pocket inducing channel opening (rightward cartoon, green receptor; dashed arrow depicts conductance of ions). (**b**) Chemical structure of an exemplary photoswitch; MAG (color coded as in (a); maleimide—green; azobenzene—red; glutamate—purple). Visible light photoisomerization (common to most photoswitches): 360–405 nm light photoisomerizes MAG from *-trans* to *-cis*; illumination at ~460–560 nm (or thermal relaxation; κ<sup>B</sup> T) returns MAG to *-trans*.

Photoswitches can be custom-tailored to photomodulate a large variety of effectors. For instance, some have been fashioned to mechanically restructure membrane lipids or DNA molecules [13–15], to act as light-activated forceps [16] or to include pharmacological agents (e.g., pore-blockers) to block, agonize, or antagonize receptors, ion channels, or enzymes (Figure 1b) (also see reviews [9,17,18]). It is noteworthy to mention that although this method dates back to the late 1960s (Figure 2) [19], it has particularly gained momentum in early 2000 owing to critical advances in chemistry, biology, and imaging methods. Here, we focus on the most recent efforts devoted to push it to the next level, namely towards multiphoton activation of photoswitches for in vivo applications in opaque preparations. Owing to the versatility of the technique, we foresee a bright future for 2PE-compatible photoswitches in biology and medicine.

**Figure 2.** Graphical illustration of 'synthetic optogenetics' emerging in the scientific literature. Cumulative plot of number of publications across years was performed by searching for specific terms (chemical-optogenetics, photopharmacology, optopharmacology, synthetic optogenetics, optogenetic pharmacology, photoswitch, azobenzene and photo), as well as by searching for publications from recognized researchers in the field (e.g., Erlanger BF, Poolman B, Feringa B, Lester HA, Trauner D, Kramer RH, Isacoff EY, Woolley GA, Gorostiza P, Paoletti P, Ellis-Davies GCR, etc.). Search was continuously performed in various search engines, notably PubMed Central, Google Scholar, and bioRxiv. Last query was performed on November 2019.

#### **2. Photoswitches**

At the heart of photoswitches lies a light-sensitive core [20]. Of the different kinds, the most prevalent moiety used in biological applications is the azobenzene (Figure 1b) [21]. Its widespread use arises from favorable features, such as high quantum yield, minimal photobleaching, and fast responsivity at biologically-relevant conditions, namely pH, temperature and solubility in water-based solvents [21–23]. However, its ability to undergo *trans*-to-*cis* photoisomerization is undoubtfully the main reason; giving rise to robust changes in geometry and end-to-end distance of the molecule (Figure 1b).

Azobenzene photoswitches commonly include additional chemical groups placed on either side of the light-sensitive core. These can be quite diverse, such as drugs, chemical- or biological tethers, dyes, as well as lipidic structures [9]. The nature of these groups will dictate the mode of action of the photoswitch and whether it is to remain diffusive or immobilized to its target. Diffusive photoswitches (photochromic ligands; PCL) are, in principle, reversible caged-compounds [24]. In contrast, immobilized photoswitches (photoswitchable tethered ligand; PTL [25]) include a unique chemical tether that can conjugate specific residues of the protein (e.g., maleimide to bind cysteines [26]; benzylguanine to a SNAP domain [27]). These restrict the binding (thereby the effect) of the photoswitch to a defined molecule. Importantly, specific amino acids (a.a.) or domains can be genetically-encoded, thereby giving users access to defined populations of cells. Today, most targets for PTLs require genetic modifications, with cysteines as the preferred choice. However, recent advances have made it possible to conjugate photoswitches to endogenous, non-modified receptors by use of strong electrophilic moiety in the PTL to couple with reactive amines and hydroxyl groups, naturally present in several a.a. side chains [28]. Another scheme to bypass protein modification is by attaching the photoswitch next to its intended target. This may include expression of membrane anchors (e.g., a transmembrane

domain with the tether facing the extracellular) to conjugate the photoswitch which can then modulate adjacent endogenous proteins [29].

#### **3. Photomodulating Cellular Activity**

One prototypical example of a synthetic optogenetic tool is SPARK (Synthetic Photoisomerizable Azobenzene-Regulated K+-channel) [30]. SPARK was designed to control neuronal excitability. This channel was designed to tether MAL-AZO-QA. MAL-AZO-QA includes an azobenzene (AZO) flanked by a cysteine-reactive maleimide (MAL) and a potassium channel blocker (quaternary ammonium; QA). For the maleimide to specifically tether SPARK, the channel had a cysteine residue inserted in one of its external loops (as water-accessible cysteines are not common in proteins [31]). Prior illumination, the photoswitch is found in its elongated *-trans* form; spanning 17 Å. The authors designed the length of the photoswitch to equal the distance between the cysteine residue and the pore so that the QA drug would easily reach the pore and block the channel. The functional outcome of blocking this potassium channel is robust neuronal excitability and action potential firing. This effect was completely reversed by near-UV illumination (~400 nm), pushing the photoswitch back to *-cis*, a much shorter form (end-to-end distance ~10 Å), thereby physically removing the blocker away from the pore leading to immediate silencing of the neuron [30]. This could be repeated many times, by quickly toggling the photoswitch back to *-trans* by ~500 nm light (rather than slow thermal relaxation in the dark). Similar schemes have been adopted to photocontrol a variety of channels and receptors with the, almost exclusive, use of visible light [9,12].

#### **4. Shifting from Visible to Near-Infrared**

Despite progresses made since the very first demonstrations of Synthetic optogenetics by the Erlanger group in the late 1960s [19,32], current illumination schemes remain as they were back then, namely rely on the use of visible light, typically near-UV (~400 nm) for *trans-to-cis* photoisomerization and blue-shifted wavelengths (~500) for *cis-to-trans* [9,12,15,17,18,24,33,34]. Aside the potential cytotoxicity of these wavelengths [35,36], visible light is not well-suited for deep tissue penetration due to strong absorption and scatter from endogenous components [37]. The use of longer, near infrared (NIR; >700 nm) wavelengths is advantageous as these are less absorbed by biological tissues and water [38], induce less photodamage [36], and are less prone to scatter.

Synthetic optogenetics is slow to adapt NIR illumination owing to the very poor absorbance of these low energy wavelengths by the azobenzene core, a common challenge with other optogenetic tools [39], not to mention the relatively little information extant for NIR absorption by photoswitches ([40,41] and see below). In fact, though the -*cis* azobenzene was first discovered almost a century ago (see [42]), the debate regarding the exact mechanism for photoisomerization, and how different factors affect it, remains lively to this day [43,44]. Following excitation, the azobenzene molecule proceeds from S0 to S1 and S2 states, with distinct absorption bands for each transition (n–π\* transition excite azobenzene compounds to S1 state and π–π\* transition leads to S2 state) [45,46]. However, many different factors can have dramatic influences on the these, for instance substitutions on the phenyl rings, solvent properties, temperature, to name the more common factors. However, it is much less known by which of the proposed mechanisms the isomerization of azobenzene proceeds, namely rotation, inversion, concerted inversion, and inversion assisted rotation (for more details see [44]). These uncertainties arise from different approaches used, or different experimental settings. For instance, whereas non-viscous polar solvents favor a rotation mechanism, viscous non-polar favor inversion [44]. It has also been shown that the use of different wavelengths can affect the isomerization mechanism. In fact, these differences are also seen with other compounds such as fluorescent proteins [47,48] and extant data of how wavelengths affect isomerization mostly pertains to one-photon excitation of azobenzene compounds [49]. This actuality makes it very difficult to infer the mechanisms for NIR excitation as the absorption properties between these different illumination schemes are substantially different [41]. For instance, use of 1-photon for isomerization shows that

the transition from *trans*-to-*cis* path develops triexponentially with times of 0.3, 3, and 16 ps (where the first two reflect population relaxation and the third corresponds to the final relaxation to ground state [49]) and these are not affected by solvent viscosity rather by wavelengths at early (t∼1 ps) and late (t∼100 ps) times but show similar photoisomerization behavior on a 10 ps time scale [50].

Some of these hurdles can be overcome by using, at least, two NIR photons; each with half the energy of needed, to be absorbed in a single event (within ~10−<sup>18</sup> s) in order to reach the molecule's excited state [51]. For instance, if a photoswitch/chromophore efficiently absorbs near-UV light at 400 nm, in principle it should be possible to excite it by two simultaneous photons at approximately 800 nm each [41]. This technique is denoted two-photon excitation (2PE) [52]. However, it is important to note that 2PE is a third-order nonlinear process that depends not only on the absorption cross-section (σ2) of the molecule, but also on the concentration of incident photons. Therefore, even small differences in laser conditions can lead to very different absorptions; making it difficult to compare compounds, or even the relative strength of transitions. Nevertheless, recent reports addressing these issues show that the fundamental excitation of the S0→S2 transition by 2P are similar and are due to resonance absorption [50].

2PE allows to excite molecules found deeper within the tissue (up to mm), at high three-dimensional, sub-micrometric spatial resolution [51,53]. This provides exquisite means to activate optogenetic tools at the level of single cells, or even subcellular regions, with minimal light escaping to undesired or nearby regions [54]. However, many optical tools display low 2PE absorption cross-section. This is worsened by low expression of the protein under control or its low conductance (in the case of an ion channel) [55,56]. Together, these limitations largely render 2PE unusable in these instances. Improving 2PE could be obtained by increasing the 2PE cross-section absorption of a photoswitch by chemical modification (s) (see below), by increasing the expression and of the protein or to simultaneously illuminate larger regions-of-interest (ROIs) such as entire somata. Expansion of the illumination area can be obtained by parallel excitation techniques that provide a high flux of photons to numerous ROIs simultaneously, in contrast to rastering methods (line scanning) that provide photons to single pixels sequentially [55,57]. A collection of these improvements have been employed for 2P-photolysis of caged-compounds as well as photoactivation of various opsins for quite some time now [56,58–62] and, only recently, in synthetic optogenetics (Figure 2).

#### **5. Two-Photon Compatible MAG-Based Photoswitches**

One of the most commonly employed tethered photoswitch is MAG [9] (Figures 1b and 2). MAG, similar to MAL-AZO-QA (see above), has an azobenzene-core (A), flanked by a cysteine-reactive maleimide (M), but instead of a channel blocker bears a glutamate molecule (G). This photoswitch is therefore intended for glutamate-binding proteins. The MAG photoswitch has been employed to activate glutamate receptors (e.g., LiGluR, LiGluN, mGluRs) in its -*cis* form, not to mention to antagonize a glycine-binding receptor (the GluN1a-subunit [26]). Akin to other azobenzene-based photoswitches, MAG undergoes very efficient -*trans* to -*cis* photoisomerization when irradiated with near-UV light (MAG0; λtrans-cis 340−400 nm), and reverts from -*cis* back to -*trans* by thermal relaxation or, significantly more rapidly, by blue-green wavelengths (MAG0; λcis-trans 440−580 nm). Nevertheless, with regards to 2PE, MAG exhibits very poor absorption in the red-to-NIR range (~700–1400 nm) [41].

In the case of first generation MAGs (MAG0 [26,63,64]), endowed with a symmetrically-substituted azobenzene (Figure 2 and Table 1), the 2P-absorption cross section is low (σ2 = 10 GM at 820 nm [41]), albeit on the order of magnitude of a common fluorescent protein such as eGFP (σ2 = 30 GM at 927 nm [47]). This, along the lower expression and density of channels at membrane of neurons, results in very little or no capacity to photocontrol cellular activity when 2PE is employed [41]. This prompted the design of second-generation MAGs with increased 2P-absorption cross sections. More precisely, this required lowering the energetic barrier of *trans-to-cis* isomerization by specifically adjusting and shifting MAG's spectral properties towards 'red'-er wavelengths.



*Appl. Sci.* **2020**, *10*, 805

One strategy employed involved breaking the symmetry around the azobenzene core (below in Figure 3 and Table 1) [74]. This was obtained by creating a push–pull system, where one benzene ring of the azo-core was decorated with an electron donating group, while the second azo-unit was supplemented with an electron withdrawing group. These led to an asymmetric azobenzene with a significantly red-shifted absorbance [21,68,74–77]. This strategy was employed in one of the very first 2P-compatible PTLs, where the MAG photoswitch was redesigned to include an asymmetric aminoazobenzene core (tertiary amine in the 4 -position) to act as a strong electron-donating group, denoted MAG2p (Figure 3) [65]. Indeed, this photoswitch exhibited a red-shifted 1P-absorption spectrum peaking at 420 nm; ~60 nm red-shifted compared to the parent MAG. Notably, this photoswitch could be activated by 2PE (Table 1). In a subsequent report, the authors estimated its 2P-absopriton cross section and suggest it to be ~five times higher than that of the parent photoswitch (MAG2p, σ2 = 56 GM [72] at 850 nm, Table 1). The authors go on to show that, when coupled to a glutamate receptor (LiGluR [63]) optimal 2P-responses are obtained at 900 nm. Another unique feature of this photoswitch is that it is no longer bi-stable, rather spontaneously reverts back to -*trans* in the dark. Notably, decrease in the thermal stability of the -*cis* isomer by red-shifting of absorbance of azobenzenes is a well-described phenomenon [74]. The functional outcome of the thermal relaxation of the photoswitch is a rapid cessation of channel activation, seen as a decrease in current once illumination stops (τoff = 150 ms [67]; ~10,000-fold times faster than thermal relaxation of MAG, τoff = 25 min [75]).

A push–pull scheme was similarly employed, very shortly after, by Kienzler et al. [75]. Here, the authors have modified MAG to include an asymmetric azobenzene-core with a tertiary amine at the 4'-position as the electron-donor, but an acetamido-group at the 4 -position, as the weak electron withdrawing group (Figure 3). These modifications resulted in a larger shift in peak 1P-absorption (~100 nm) of the -*trans* isomer (λpeak = 460 nm), therefore denoted MAG460 (Figure 3) [75]. Expectedly, this modification reduced the stability of the -*cis* isomer (MAG460, τoff = 0.71 s) [75]. Importantly, this shift led to an 8-fold increase in the photoswitch's 2P-absorption cross section (*trans*-MAG460, σ2 = 80 GM at 850 nm) [41]. We have later employed this photoswitch to successfully activate LiGluR [41]. Though the photocurrents obtained by 2PE were of smaller amplitude than when using 1P-illumination, they were sufficient to evoke action potential firing in cultured neurons by parallel excitation methods (digital-holography; Table 1) [41,78]. Lastly, we also employed a fluorescent reporter as readout for Ca2+-activity (R-GECO) and find it to provide reliable responses (Δ*F*/*F*), demonstrating that this method is also suitable for all-optical interrogation of cells [78].

Another interesting strategy that has been employed to sensitize the MAG photoswitch towards NIR wavelengths is by adding a light-harvesting molecule (i.e., antenna). Two such photoswitches were designed under the name of maleimide–azobenzene–glutamate antenna, or MAGA. The first—MAGA2p [65]—consisted of the same asymmetric core as us found in MAG2p along an added light-harvesting naphthalene-derivative (with a high 2P-absorption cross section; σ2 ≈ 200 GM at 780 nm [79]) (Figure 3). The antenna was incorporated to sensitize the *trans*-azobenzene by resonant electronic energy transfer (RET) (Figure 3). This photoswitch could therefore be photoactivated by two distinct mechanisms: direct excitation of the push–pull azo-core by 2PE and RET by the photosensitizing antenna. This photoswitch displayed low thermal stability, with rapid relaxation of the currents (τoff = 265 ms [65]) that, when excited using a laser-scanning method, would not allow efficient activation of the channel or robust cellular responses. In a subsequent improved version, MAGA ligand-2 (Figure 3) [66], the authors maintained an antenna for harvesting light, albeit swapped it by a pyrene molecule (σ<sup>2</sup> = 55 GM), but reverted back to a symmetrical azo-core with the intention to counteract the pyrene's lower 2P-absorption cross section (∼4-fold lower than naphthalene) by increasing the thermal stability of the photoswitch. Indeed, the resulting photoswitch exhibited a highly stable *-cis* isomer (MAGA ligand-2, τoff = 2.0 h in DMSO). Thus, the increased thermal stability of the -*cis* state should make this photoswitch more suitable to accumulate activated photoswitches by scanning imaging methods. However, it is worthy to mention that the photosensitizing antenna

employed drastically reduced the solubility of the photoswitch and, likely, its usability in biological experiments. Indeed, so far MAGA ligand-2 was not employed with cells (Table 1).

**Figure 3.** Chemical structure and absorption of exemplary 2P-compatible photoswitches. Left, names of photoswitches; right, one photon (1P) peak absorption (λmax), also depicted by middle color gradient.

The most recent development in MAG photochemistry includes a compromise between the -*cis*' thermal stability and 2P-absorptivity by addition of a strong inductive electron withdrawing group in the *ortho* position, in conjunction with an asymmetric azo-core (Figure 3). This new design has been based on previous *ortho* substitutions made onto azobenzene molecules that can affect the steric or electronic barrier for isomerization, not to mention to robustly slow down its thermal relaxation rates [68,74]. Examples of this include the symmetrical azo-core photoswitch denoted toCl-MAG1 (Figure 3); highly decorated with tetra-*ortho*-chloro substitutions, exhibiting a very slow thermal relaxation (τcis = 3.5 h at 37 ◦C in PBS, pH = 7.0) [68], not to mention red-shifting its peak 1P absorption to 470 nm (the most rightward shift seen in PTLs) (Figure 3). When computationally modeled onto the azo-core of MAG0 and MAG2p, these substitutions yielded new photoswitches, designated *MAGslow* 2*p*\_*F* (Figure 3) [67]. Surprisingly, these substitutions did not lead to a major shift in the 1P absorption of the photoswitch (λpeak = ~360 nm), however they significantly increased the thermal stability of the -*cis* variant when compared to previous MAG2p (MAG2p, τcis = 118 ms in 80% PBS:20% DMSO [65]; *MAGslow* <sup>2</sup>*p*\_*F*, τcis = 10 min in 99% PBS:1% DMSO [67]). Modeling also showed that the photoswitch should exhibit a slightly increased 2P-absoprtion cross section (MAG2p, σ2 = 56 GM; *MAGslow* <sup>2</sup>*p*\_*F*, σ2 = 69 GM), therefore potentially resolving the problem of insufficient accumulation of opened channels during illumination. Although the authors do not show comparison of 1P vs. 2P-mediated current size (which should differ), they do however extensively characterize the responses obtained by a fluorescent calcium activity reporter (RCaMP2) and show that equivalent responses are obtained when using optimal 1P-(405 nm) or 2P-excitation (780 nm) (Table 1). These results are in-line with our own, showing the compatibility of this photoswitch, and technique, for all-optical interrogation of cells.

#### **6. Novel Photoswitches with 2PE-Potential**

Other azobenzene-based photoswitches have also been rendered 2p-compatible by use of other strategies. Of these, the methoxy-substitutions at the *meta* positions and C2-bridged azobenzene are particularly interesting [74]. Methoxy *meta-*substitutions are estimated to strongly shift the absorbance of the azo-core to much longer wavelengths than most reported photoswitches (i.e., >100 nm), specifically into the far-red and infra-red regimes [80] and these should likely have a high 2P-absoprion cross section. However, these modifications will most likely lead to very rapid thermal relaxations. One such example is compound 28 with near-IR absorbance peaking at 620 nm (but also sufficiently activated by 730 nm), but with an ultra-rapid thermal back reaction (τcis = 10 μs) (Figure 3) [74].

The second approach, the C2-bridged azobenzene, also show a significant red-shifted absorption spectrum of the -*trans* isomer (λmax > 540 nm) [72,74,81], implying an increased 2P-absorption cross section [41], but with a much more favorable (i.e., slower) thermal relaxation rate (on the order of minutes at room temperature in aqueous solution) than methoxy *meta*-substitutions. The drawback of the molecule is that the -*cis* state is the more thermodynamically stable form. This means that if the photoswitch is active in *-cis*, it would be active when applied onto the preparation. Thus, it is preferable should the photoswitch be active only in *-trans*. An additional drawback is that the stable -*cis* isomer poorly absorbs red light and isomerization to -*trans* requires UV to near UV illumination. Recently, two *trans*-active diffusible photoswitches were designed based on a C2-bridged azobenzene, denoted locked-azobenzene (LAB) [82] or Glu\_brAzo [72]. Both photoswitches bear a glutamate moiety as ligand, thus intended for activation of glutamate receptors, and exhibit highly similar structural and spectral properties (Figure 3). Of note, despite their similarities, LAB is shown to selectively activate NMDARs, whereas Glu\_brAzo efficiently activates both Kainate and AMPA receptors. Regardless the receptor type, Glu\_brAzo shows more practical features for use in vivo, particularly increased solubility in aqueous media owing to the addition of a bulky ionic group. This is noteworthy as solubility issues are a very common limitation of most photoswitches, in particular azobenzene-based ones [9,21]. This additional group is also suggested to improve the performance of the photoswitch by increasing the steric congestion around the glutamate moiety with the intention to lower the interaction of the glutamate moiety with the receptor when the photoswitch is found in its stable -*cis* isomeric

form. To design Glu\_brAzo, the authors searched for *trans*-active photoswitches that target glutamate receptors. This was not very challenging as most diffusible GluR-photoswitches are, surprisingly, active in the extended thermally stable -*trans* form (e.g., [83,84]). They honed-in on a *trans*-active photoswitch denoted GluAzo [85]. For photoactivation, *cis*-Glu\_brAzo (i.e., inactive) requires λpeak = 395 nm to shift to -*trans*. As noted above, this is quite a limitation for two-photon activation. However, this photoswitch exhibits very desirable bi-stability, with very slow spontaneous back-isomerization from *trans*-to*-cis* (t1/2∼4 h, at RT in aqueous media). This step could be accelerated by green light illumination (λpeak = 480 nm), suggesting that it may also be compatible with 2PE. However, the 2P-absorption cross section was not assessed for Glu\_brAzo (or LAB) and is very hard to predict based solely on 1P absorbance (seen above and e.g., [47]). An additional limitation of Glu\_brAzo is its overlapping absorption spectra of both *-cis* and *-trans* isomers, so much so that even the use of optimal 1P wavelengths result in a mixture of photostationary states, with merely ~60% of the active -*trans* isomer.

#### **7. Future Prospects**

There is growing interest in synthetic optogenetic tools in fields such as in neuroscience [8,12,23], heart physiology [86–88] and, intuitively, vision restoration [89–92]. Thus, we expect these to motivate further developments of the strategy in the upcoming years, in particular towards progression of the technique towards enhancing multiphoton absorption. Synthetic optogenetic tools can be combined with electrophysiological recordings and optical tools (e.g., Ca2+-probes [78]), making it highly tailored for in vivo use in different animal models in an all-optical manner. We also suggest that several photoswitches might even make it to the clinic [93], in particular photoswitches that target native proteins (e.g., [28,29]) or those applicable in the blood [74]. In addition, synthetic optogenetic tools have shown promise for studying (and maybe treating) Parkinson's disease-related receptors [94], inhibiting cellular division in cancer [95], not to mention controlling cellular excitability, highly relevant for brain diseases such as epilepsy [96].

Here, we have briefly summarized several of the newest developments in the field of synthetic optogenetics, more precisely efforts devoted towards rendering photoswitches absorbent of longer, red-shifted wavelengths and their ability to undergo 2PE. These properties are highly desired for making the method more operational in vivo. Despite this motivation, it remains that most laboratories focus on designing unique photoswitches for 1P-applications. We believe that this stems from limited information on nonlinear multiphoton properties of photoswitches and the difficulties in determining, *a priori*, their multiphoton absorption properties [47,67], relaxation rates and whether optimization (derivatization) of 1P-functional photoswitches towards 2PE will not render them inoperative.

It appears that bistable photoswitches provide an optimal starting point for designing next-generation 2PE-compatible photoswitches. The slow relaxing photoswitches with light-harvesting antennae and symmetrical azo-cores display high 2P-absorption cross section that should allow for robust photoactivation of optical actuators in vivo. However, they will require further modifications to increase their solubility in aqueous solution prior use with cells. The bistable *MAGslow* <sup>2</sup>*p*\_*<sup>F</sup>* exhibits slightly better features. Though of lower 2P cross-section (<70 GM) and almost no shift in 1P absorption maxima, its lower thermal relaxation allows for larger responses. The effect of this, and other, photoswitches could be further improved by additional means such as use of higher 2P-laser intensities (keeping in mind that higher intensities may be harmful to cells [52]), increase in the expression of the optical actuators [97,98] and, importantly, use of parallel excitation methods for simultaneous activation of larger regions of interest in all axes [55,57,78,99].

In summary, we see current hurdles as wonderful incentives and opportunities for designing better photoswitches and protein actuators. Indeed, many labs around the world are working intensely to address these issues. Consequently, we anticipate that the potential of synthetic optogenetics towards in vivo use and the clinic will be realized within the next few years.

**Author Contributions:** Conceptualization, S.B. and S.K.; methodology, S.B.; data curation, S.B. and S.K.; writing—review and editing, S.B. and S.K.; visualization, S.B.; supervision, S.B.; funding acquisition, S.B. The research submitted is in partial fulfilment for a Doctoral degree for S.K. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the Israel Science Foundation (ISF), grant number 1096/17 (S.B.).

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*Applied Sciences* Editorial Office E-mail: applsci@mdpi.com www.mdpi.com/journal/applsci

MDPI St. Alban-Anlage 66 4052 Basel Switzerland

Tel: +41 61 683 77 34 Fax: +41 61 302 89 18