Design and Synthesis of Photoswitchable Carbamazepine Analogs

The design of photoswitchable analog of carbamazepine followed a non-trivial azologization approach (dubbed crypto-azologization since the analogy is hidden in the ring structure) that proved to be effective in tricyclic drugs.²² Here, we expanded this method to include linearized azobenzenes beyond the originally reported ortho substitution (meta and para) as well as tricyclic diazocines (compounds 1–7 and 8, respectively, Figure 1).

We first examined the structure–activity relationship (SAR) of carbamazepine to identify the position of the pharmacophore and its interaction in the pore of voltage gated sodium channels (Na_V). Carbamazepine is part of a family of anticonvulsant drugs with very conserved structural elements that embrace one aryl ring (R), an electron donor atom (D), and a hydrogen bond acceptor/donor group (HAD), mutually separated by fixed distances.^{35, 36} In particular, the urea moiety in carbamazepine is involved in an amino-aromatic hydrogen bond in the binding pocket and establishes a crucial interaction for drug recognition.³⁷

Therefore, we retained these elements and designed the first cryptoazologs of carbamazepine by opening the seven-membered central ring and replacing it with the azo (−N=N−) bond. We named these linearized analogs “carbazopines”. The urea moiety was kept in ortho position as in the parent drug²² but, here, we also explored meta and para positions to test how flexible are the constraints of the pharmacophore (Figure 1B, Carbazopine-1, −2, and −3, respectively). These structures allow us to probe whether the conformational change upon isomerization localizes the urea functional group in the right cavity and how it depends on the urea substitution position.

Subsequently, we expanded the chemical library by introducing para and ortho substitutions in the unsubstituted aromatic ring, aiming to tune the photochromic properties (Figure 1C, Carbazopine-4 to −7). In particular, it has been reported that push-pull systems³⁰ and tetra-ortho substitutions with fluorine, chlorine, and methoxy groups³⁸ produce a red-shift in the absorption spectra of the azo compound that allows trans to cis conversion at wavelengths toward the red region of the spectrum, which are convenient because they display deeper tissue penetration than UV or blue light. Since the urea occupies the ortho position in one of the aromatic rings, we focused on bi-substitutions at the opposite ring.

Furthermore, we designed a photoswitchable analog of carbamazepine, keeping the middle ring closed. Bridged azobenzenes are receiving increasing attention in materials and biology.^{8, 39, 40} We based our approach on N-bridged diazocine, which adds an extra carbon atom to the central ring of carbamazepine, and we kept the urea at an equivalent position to that of the drug (Figure 1D, compound 8). We named it Carbadiazocine (8) in subsequent experiments.

Before synthesizing the compounds of Figure 1, we examined their drug-likeness and the predicted safety properties of Carbazopine-1 and Carbadiazocine (8) using ChemDraw and ProTox-II, see Supporting Information (SI). The results are shown in Table S1, S2 and Figures S1, S2, and S3, and predict that all compounds designed by cryptoazologization will retain or improve the drug-likeness and safety of carbamazepine.

Linear azobenzene derivatives Carbazopine-1 to −3 were prepared from the respective commercially available ortho/meta/para phenylenediamines (Scheme 1A). The urea moiety was introduced with a substitution reaction using as a reagent potassium isocyanate under acidic conditions. The obtained intermediates were subsequently coupled with nitrosobenzene under Mills conditions.

The library of para and ortho substituted derivatives (Carbazopine-4 to −7) was obtained via a divergent synthetic approach starting from the common intermediate (9, Scheme 1B). This intermediate was combined under Mills conditions with commercially available 1,3-dichloro-2-nitrosobenzene and the previously synthesized compound 12 to afford the ortho halogen di-substituted products Carbazopine-5 and −6. The synthesis of the nitroso compounds 12 and 13 was achieved by oxidation with oxone in a biphasic system of water (H₂O) and dichloromethane (DCM). The intermediate 13 was isolated after extraction of the organic phase without further purification and it was directly combined in acetic acid (AcOH) with 2,6-dimethoxyaniline and N,N-Dimethyl-1,4-phenylenediamine to provide the respective final products Carbazopine-4 and −7. The Mills reaction could only take place when the amino group was held in the most electron-rich aromatic ring, while the oxidation to the nitroso intermediate was favored in electron-poor aromatic rings. The azo coupling for all the substituted analogs (4–7) gave low yields (between 5–8 %), possibly due to steric hindrance of the ortho substituents. Indeed, a moderately better yield (26 %) was achieved for compound 6 bearing two fluorine atoms, comparable in size to hydrogen atoms.

Finally, the synthesis of the N-bridged diazocine Carbadiazocine (8) followed a reported synthetic Scheme with minor modifications (SI, Scheme S3)⁴⁰ and the addition of the final N-substitution with the urea moiety (Scheme 1C). For instance, the yield for the Fmoc protection of the amino group was consistently improved (from 47 % to 94 %) using anhydrous 1,4-dioxane as solvent instead of the reported dry dimethylformamide (DMF), (SI, Scheme S3). The key step for ring closure and formation of the N-bridged diazocine ring gave a poor yield (13 %) compared to the reported one (40 %), mainly due to the formation of the side product 19. Since this side product results from excessive reduction of the nitro group, we tried to limit the exposure time to zinc/NH₄Cl from 10 minutes to 5 minutes. However, the final yield did not change. To overcome this limitation, an additional step was introduced to fully convert compound 17 into the di-amine product 19: the same conditions were retained, but the reaction was stirred for 1 hour, thus allowing the formation of the desired product 19 in 68 % yield. The intramolecular azo coupling for the ring closure of 19 was carried out in glacial acetic acid with either mCPBA (32 % yield) or oxone, which provided the diazocine 18 in 59 % yield.

The final carboxamination of the NH-diazocine 20 was not feasible using isocyanate in acidic conditions due to the reported tautomerization to the imine bridged hydrazine.⁴⁰ Therefore, we used triphosgene in the presence of triethylamine (Et₃N) to form the acyl chloride intermediate, followed by the addition of aqueous ammonia (NH₃) 30 %⁴¹ to provide the urea-substituted final product Carbadiazocine (8) in 72 % yield. The chemical characterization of every intermediate is reported in the Supporting Information file.

Photochemical Characterization of Carbamazepine Azobenzene and Diazocine Analogs

The photochemical properties of the synthesized compounds were characterized using UV/Vis spectrophotometry. We measured the ability of the molecules to efficiently isomerize and change their absorption spectra upon illumination and during thermal relaxation in the dark. All the linear compounds (Carbazopine-1 to −7) showed the typical absorption bands of azobenzene (photochemical behavior described in SI, see Figures S12–S19). The % of trans isomers at the photostationary states (PSS) was quantified by HLPC analysis at the isosbestic point (SI, Figures S20–S25). The amount of Carbazopine-1 trans isomer changed from 98 % in the dark-adapted state to a minimum of 22 % upon irradiation with 380 nm light (Figure 2ABC and Supporting Information Figures S12, S20 for absorption and HPLC measurements, respectively). Reverse switching from cis to trans can be obtained by irradiation at 500 nm and yields a maximum of 47 % trans isomer due to the overlap of the nπ* absorption band between the two isomers. Complete back-isomerization of Carbazopine-1 can be achieved by spontaneous thermal relaxation in dark conditions with a half-life of t_1/2=3.6 h (SI, Figure S12C).^{18, 42, 43}

In contrast to the linear azobenzenes, Carbadiazocine (8) is thermodynamically stable in its cis form as other diazocines.^{8, 40, 44} The absorption maximum of dark-adapted Carbadiazocine can be excited at 400–420 nm to achieve a mixture of 49 % cis and 51 % trans (Figure 2DEF and Supporting Information Figures S19 and S25). Interestingly, the distance between the cis and trans nπ* absorption bands allows for selective excitation of the trans isomer using wavelengths longer than 500 nm and achieving complete back-isomerization to the cis form. Quantitively recovering the dark-adapted configuration is a desirable feature as it allows to have fully reversible control over the activation and deactivation of the drug. Wavelengths in the 500–600 nm range also allow reducing toxicity and increasing tissue penetration compared to UV light. Hence, we irradiated 50 μM trans Carbadiazocine (8) using a 590 nm LED (120 mW, see SI) and observed quantitative conversion to the cis isomer in 45 min (Figure 2EF). Full conversion can be conveniently achieved in 1 min using a common halogen lamp (fiber optic gooseneck illuminator for surgery or microscopy, Supporting Information Figure S19F). The thermal relaxation half-life in the dark of trans-enriched Carbadiazocine (8) is t_1/2=2.2 h (SI Figure S19C), which offers a time window to test separately cis-relaxed and trans-enriched solutions in biological assays.

The photoisomerization of all the synthesized molecules is reversible and repeatable for several cycles of illumination without apparent loss in absorption or any evidence of degradation. The results of the photochemical characterization are outlined in Supporting Information and summarized in Table S3–S4 for the entire compound library and for Carbazopine-1 and Carbadiazocine (8).

Carbazopine-1 and Carbadiazocine (8) Reversibly Photocontrol Neuronal Firing in vitro

To identify active and photoswitchable carbamazepine analogs, we screened them using calcium imaging in cultured hippocampal neurons (see methods in SI). We looked for alterations in spontaneous calcium activity upon applying each compound in the dark, and under different illumination conditions to favor their cis and trans forms. Among the five linear compounds that are soluble in water with less than 1 % DMSO (Table S3), we found that only Carbazopine-1 inhibits neuronal activity in a light-dependent manner (Figure 3 and Supporting Information Figure S26), while the others do not show clear effects.

We further characterized the ability of Carbazopine-1 to photocontrol neuronal activity by directly measuring the frequency of action potential generation using patch clamp electrophysiology (Figure 3). At 30 μM, Carbazopine-1 in the dark or under 500 nm light robustly decreases the firing frequency from 3.6 Hz to 0.4 Hz, and 380 nm illumination readily reverts this effect and restores firing in a few seconds (Figure 3A). The effects of 380 nm and 500 nm illumination during Carbazopine-1 application were repeatable, as shown in the example recording and the firing frequency quantification below. On average, 30 μM of Carbazopine-1 under 500 nm light decreased the firing frequency to 8±5 % relative to the control (P<0.001, n=6), while 380 nm light induced a recovery up to 61±27 % (P<0.05, n=6). This control of neuronal firing with light is concentration-dependent: 10 μM Carbazopine-1 does not induce statistically significant changes in relative firing frequency (120±54 % under 500 nm light and 108±27 % under 380 nm light, n=5), whereas at 50 μM it blocks neuronal firing regardless of the illumination conditions (0 %, n=3 under 500 nm and 7±12 %, n=3 under 380 nm). Thus, 30 μM is the optimal concentration of Carbazopine-1 to photocontrol neuroinhibition in vitro. However, it is active in the dark and deactivated under violet light, which is a limitation to producing localized drug photoactivation.

We then tested the neuronal activity of Carbadiazocine (8) using the same assay. Figure 3C shows an example of current clamp recording and the corresponding quantification of the firing frequency. At 30 μM in the dark, Carbadiazocine (8) decreases the firing frequency to 72±5 % compared to control conditions (P>0.05; Figure 3CD), and under 400 nm light, it decreases to 28±13 % (P<0.01; n=4). This effect is reversible under 500 nm light and repeatable when illuminating again with 400 and 500 nm. The efficient and reversible modulation of action potential generation by hippocampal neurons indicates that Carbadiazocine (8) is a light-activated photoswitchable neuroinhibitor at a concentration of 30 μM in vitro. It is thus complementary to Carbazopine-1 and constitutes another good candidate to produce localized neuroinhibition with light.

Carbazopine-1 and Carbadiazocine (8) Reversibly Photocontrol Zebrafish Larvae Behavior

The photoswitchable derivatives of carbamazepine displaying activity in vitro (Figure 3) were further evaluated in vivo in zebrafish (Danio rerio) larvae at early developmental stage (7 days post fertilization, dpf) using a high-throughput behavioral assay.¹⁵ Compounds were diluted in zebrafish water and freely swimming larvae were incubated for 20 min to allow drug uptake^{15, 45} prior to analyzing their behavior under varying light conditions applied during the experiment (Figures 4 and 5 and SI). Carbazopine-1 at high concentrations (50–100 μM, red and green traces in Figure 4A top) transiently increased the swum distance during the initial relaxation period (RP) in the dark (0–20 min, shaded grey) in comparison to vehicle (black trace). Vehicle-treated larvae display a characteristic swimming response to 380 nm UV light (avoidance).^{46, 47} This behavior is blocked by Carbazopine-1 in a light-dependent manner (absence of swimming during UV periods and restoration under 500 nm green light). The statistical analysis for groups of 12—24 larvae is shown in Figure 4A bottom. Overall, Carbazopine-1 induced a dose-dependent, reversibly photoswitchable effect on the ability of larvae to swim. The compound appears to induce activity in trans compared to vehicle. Since the time course of the responses of Figure 4A top is masked by the kinetics of drug uptake and action, and by natural responses to light, we repeated the experiments with blinded larvae (treated with high intensity light, see Supporting Information methods and reference²³). Blinded larvae display reduced behavioral reactions to light changes but continue to react to mechanical (touch-evoked responses) and vibrational stimuli. We observed lower RP activity and photoresponses in vehicle-treated blind larvae while in 50 μM Carbazopine-1 the dark activity and photoresponses to 380/500 nm light were maintained (Figure 4B displays activity traces in the top panel and quantification in the bottom). These experiments are exemplified in the Supporting Information Movie M1. As a qualitative assessment of toxicity, we kept larvae in Carbazopine-1 for 24 h after the experiments, and we noticed that all fish larvae were dead at 100 μM dose.

Carbadiazocine (8) also produced a dose- and light-dependent effect on swimming behavior (Figure 5). During the initial dark RP, no significant effect of the drug compared to the vehicle was observed but subsequent illumination produced robust and reversible photoresponses at 10–50 μM Carbadiazocine (8). In the absence of drugs and under constant visible wavelengths, such as 420 and 500 nm, intact zebrafish larvae do not show anxiety-related behaviors,^{48, 49} displaying scoot swimming and sporadic darting behaviors.⁵⁰ Changes in wavelength within this range elicit negligible responses (see Figure 5A, top panel, black vehicle trace between 20–55 min). However, switching the light off after a prolonged period of visible illumination triggers an enhanced photokinetic swimming behavior known as successive induction response^{51, 52} (Figure 5A top panel, black vehicle trace between minutes 55–60, 65–70, 75–80, and 85–90). In contrast to these native behaviors of untreated larvae, 10–50 μM Carbadiazocine (8) increases locomotion under 420 nm violet light and decreases it under 500 nm (green). These wavelengths isomerize the compound respectively to trans and back to cis (Figure 2). The effect is stronger but less reversible and repeatable at 50 μM (red traces in Figure 5A) probably due to exhaustion after the first cycles where treated larvae swum eight times more than in vehicle. Interestingly, light-off transitions (55–90 min) lead to smaller changes in locomotion (e.g. blue traces in Figure 5A) likely because the compound is thermally stable in the time frame of minutes after switching the light off (t_1/2=2.2 h, Figure 2). We also tested the effect on larval behavior of illumination with the halogen warm white lamp used for trans-cis isomerization of Carbadiazocine (8) (Figure 2). This is indicated by orange shaded bands in Figure 5A and it resulted in rapid and effective blockade of larval hyperactivity. The quantification and statistical analysis of these experiments in intact larvae are shown in the bottom panel of Figure 5A. Overall, Carbadiazocine (8) appears to be inactive in the dark compared to vehicle and elicits swimming behavior under violet illumination, which can be reversed with green or amber light. As in Figure 4, we aimed to isolate the drug effects by studying 10 μM Carbadiazocine (8) in blinded larvae (Figure 5B and SI). Vehicle treated animals show lower swimming activity and nearly absent photoresponses compared to intact ones. Behavioral photoresponses elicited by Carbadiazocine (8) are observed and statistically significant, with 420 nm light increasing locomotion and 500 nm light decreasing it. These experiments are exemplified in Supporting Information Movies M2 and M3. We also tested Carbadiazocine (8) toxicity qualitatively by keeping the larvae in their wells for 24 h after the assays, and we did not observe dead larvae at any concentration.

These results show that Carbazopine-1 and Carbadiazocine (8) enable robust and reversible photocontrol of wild-type zebrafish larvae behavior, and the latter does not produce apparent toxicity in the studied concentration range.

Carbadiazocine (8) Enables Non-Invasive Photocontrol of Neuropathic Pain in Rats

The favorable molecular properties of Figures 2 and 3 and the photocontrol in zebrafish larvae shown in Figures 4 and 5 prompted us to proceed with mammals. The neuroinhibition mediated by Carbazopine-1 and Carbadiazocine (8) offers opportunities to demonstrate a proof of concept of photocontrolled analgesia, so we tested them using nociception assays in a rat model of neuropathic pain (chronic-constriction injury, CCI, Figure 6 and Supporting Information Figures S27–S28). Briefly, animals were cannulated for intrathecal (i.t.) drug administration, and chronic pain was induced surgically by CCI. Sensory threshold changes were determined using two nociceptive tests (von Frey mechanical test and acetone thermal test, see Supporting Information Figure S27B) in lesioned and control paws. In preliminary experiments with photoswitchable ligands of α-adrenergic receptors (adrenoswitches),²⁴ we observed partial inhibition of pain behavior upon i.t. injection of the drugs, albeit at concentrations too high to allow photoswitching (SI Figure S27E–H). We then injected either vehicle (10 % DMSO), 65 μM carbamazepine, Carbamazopine-1 (60 μM in preliminary experiments, see below) or Carbadiazocine (8) (200, 600 μM) and performed nociceptive tests (Figure 6A and Supporting Information Figure S28). Carbamazepine induced significant analgesic effects (increase of the mechanical withdrawal threshold and decrease of the acetone score only in the lesioned paw, Figure 6BC), in agreement with reported results.⁵³ The corresponding time courses are shown in Supporting Information Figure S28CD.

Injection of Carbazopine-1 induced notable side effects in rats, including reduced spontaneous motor activity, back-arching, twitching, closed eyes, and vocalizations. Due to these adverse reactions, further testing of Carbazopine-1 was not pursued in this study. In contrast, both isomers of Carbadiazocine (8) at a concentration of 600 μM were well tolerated, showing no apparent signs of anesthesia, sedation, or notable alterations in behavior, and were thus studied in detail. The mechanical and thermal pain were assessed after i.t. injection of 600 μM dark-relaxed and 400 nm pre-illuminated Carbadiazocine (8) (black and blue dots, respectively, in Figure 6DE). Dark-relaxed Carbadiazocine (8) induced significant analgesic effects, both in mechanical and thermal assays and only in the lesioned paw, not in the control paw. In contrast, the activity of preilluminated Carbadiazocine (8) was not significantly different from that of the vehicle in either test. The corresponding time courses are shown in Supporting Information Figure S28C–F.

These promising results suggested that it might be possible to photoinduce on-demand, spinal-mediated analgesia by injecting the inactive preilluminated Carbadiazocine (8) and using an external halogen lamp to non-invasively isomerize it, taking advantage of the ability of the compound to isomerize under 590 nm light (Figure 2 and 5). Strikingly, when we performed this experiment, the animals remained still under the lamp, a behavior that suggested readily perceptible pain relief (Figure 6F and Supporting Information Movie M4). Detailed quantification of nociceptive tests confirmed significant analgesic effects of illumination, only in the lesioned paw and in the presence of Carbadiazocine (8) (Figure 6GH). Further studies are warranted to confirm the absence of side effects associated with local administration of Carbadiazocine (8) in the spinal cord. These investigations would provide valuable insights into the safety and efficacy of this approach, further informing its potential application in clinical settings.

We discuss below the major results of the article and their advantages and limitations compared to previously reported results (more detailed discussion can be found in the SI).

Our integral drug design involved (1) selecting protein targets validated for pain and epilepsy (α-adrenoceptors, Na_V channels), (2) choosing a clinically prescribed drug as a parent compound to obtain drug-like photoswitchable analogs, and (3) retaining only those analogs that can be used in vivo and that allow future optimization of their properties (including safety, ADMET, and formulation). The absence of activity of meta and para analogs (2, 3) further validates the concept demonstrated with active ortho cryptoazologs (1) for tricyclic drugs.²² In addition, we reported the first instance of diazocine crypoazalog (8) retaining the original tricyclic structure. Using halogen lamps and violet LEDs, the compound allows in vivo photoswitching of analgesia for neuropathic pain in rats with non-invasive illumination. Future improvements may afford bi-directional quantitative switching and lead to even stronger differences in firing rate with light.

The surprising differences in photoswitch sign between in vitro and in vivo assays can be attributed to the contributions of both excitatory and inhibitory neurons and to the complex effects of carbamazepine derivatives on conductance and gating mechanisms (activation and inactivation).^{54, 55} Note also that the three-dimensional geometry of diazocines (whose side ring planes adopt an angle of ~80° in cis and ~180° in trans)⁵⁶ differs from that of carbamazepine (~130°).³⁷ The angle of linear cis azobenzene ranges between 60°–90° depending on the substitutions.

Both Carbazopine-1 and Carbadiazocine (8) retain the in vitro potency of carbamazepine and other AEDs (tens of μM)⁵⁷ and the latter does not show in vivo toxicity. Carbadiazocine (8) is less active in rats upon violet illumination (50 % trans-like) and converts quantitatively to 100 % cis with either green or orange light, the latter allowing deeper tissue penetration. Thus, doubling the concentration of the active isomer of Carbadiazocine (8) suffices to produce analgesic effects, probably because small changes in membrane potential near the firing threshold have large effects in neuronal activity. Carbadiazocine (8) is a small, drug-like, and water-soluble molecule that may retain the favorable characteristics of carbamazepine, including oral bioavailability and permeation through the blood–brain barrier. Taken together, our approach has overcome several hurdles to develop a photoswitchable AEDs with properties matching those of conventional drugs and allowing the use of non-invasive illumination in rats. Potential uses include the inhibition of pain signaling, seizures, and bipolar disorder.

The development of photoswitchable ligands of ion channels involved in neuronal excitability has largely focused on potassium channels,⁵⁸ with fewer photoswitches targeting sodium channels,⁵⁹ despite their unique role in action potential generation. Sustained efforts to develop photoswitchable AEDs and other drugs should lead to treatments offering highly localized and on-demand neuroinhibitory efficacy without systemic adverse effects.