Degradation of pharmaceuticals through sequential photon absorption and photoionization – the case of amiloride derivatives

Photodegradation of pharmaceutical and agrochemical compounds is an important concern for health and the environment. Amiloride derivatives are known to undergo clean photosubstitution in water. We have studied this apparent photo-S N Ar reaction in water and – for the first time – alcoholic solvents with a range of experimental and computational techniques. Available evidence points to a mechanism starting with photoexcitation followed by absorption of a second photon to eject an electron to give a radical cation intermediate. Subsequent substitution reaction with the protic solvent is assisted by a general base, possibly strengthened by the proximal solvated electron. Final recombination with the solvated electron generates the observed product. Quantum chemical computations reveal that excited state antiaromaticity is relieved when an electron is ejected from the photoexcited molecule by the second photon, leading to a weakly aromatic radical cation. The mechanism indicated here could have wide applicability to photoinduced degradation of similar heteroaromatic compounds in the environment as well as in protic solvents. There are also strong similarities to a class of increasingly popular synthetic photoredox methods.


Introduction
Pharmaceutical research is aimed at increasing quality of life globally by finding cures to debilitating or fatal diseases. The progress in recent years has been astounding, but the route to a new drug still faces many obstacles, in particular regarding safety, that impedes and increases the cost of pharmaceutical development. One important safety factor is drug product degradation, which can give impurities with potentially dangerous properties. The identification and investigation of all possible degradation products forms a significant part of the work that must precede the first dosing to human subjects. In some cases, degradation pathways are well understood and can be predicted using computational tools. 1 However, photochemical processes, leading to degradation and phototoxicity, are still less well understood than, for example, oxidative processes. Increased understanding of these processes could lead to better predictions of benefit to all drug development. The same processes are also important contributors to degradation of other substances in the environment, such as agrochemicals.
One of the most important photochemical degradation pathways of relevance to many drugs is photodehalogenation, which can occur either by photosubstitution, typically by the solvent, or photoreduction ( Figure 1a). 2 For example, chlorpromazine, an antipsychotic drug, reacts by photosubstitution in water, 3 ,4 while irradiation in methanol leads to both photoreduction and photosubstitution in a 1:1 ratio (Figure 1b). Reaction in more substituted alcohols favors the reduction product. 5 As with most photodegradations, the complete product profile is complicated and highly dependent on the reaction conditions, e.g., presence of oxygen. Irradiation under conditions simulating the aquatic environment leads to a total of 57 photoproducts, the major ones from photosubstitution and oxidation. 3 Another prominent drug undergoing photodehalogenation is diclofenac (Figure 1b). Upon irradiation, it undergoes ring-closure and loss of one of the chlorine substituents to form a carbazole. Further irradiation leads to both photoreduction (major) and photosubstitution (minor) in water or methanol. 6 Other drugs that undergo photodechlorination include frusemide, 7 chloroquine 8 and hydrochlorothiazide (Figure 1b). 9 In addition, similar photodegradations could occur in compounds used in the agrochemical sector. In the process of our development work at AstraZeneca, some of the authors became interested in the photodegradation of compounds related to the diuretic drug amiloride (1, Figure 2a). The photochemistry of 1 is dominated by the pyrazine moiety, and irradiation in water leads to photosubstitution of the chlorine substituent as well as further secondary photoreactions (vide infra). 2 The main pyrazine core in amiloride is central to its function and is found in many derivatives, such as benzamil, phenamil, DMA, EIPA and HMA. 10 To better understand the photodegradation mechanism of this family of compounds, we investigated the amiloride analogue 2 in water and other protic solvents. We find that the primary photoreaction of 2 is photosubstitution by the solvent. The mechanism likely involves photoionization to the radical cation, induced by sequential absorption of two photons, followed by a concerted SNAr-type reaction in which the chlorine is displaced by attack of solvent. This finding has implications not only for drug development, but also synthetic organic chemistry. In recent years, photochemistry has emerged as a new way to access thermodynamically unfeasible modes of reactivity, opening previously inaccessible routes to target compounds. 11 Of particular interest in this context is the work of Nicewicz, who showed that photoredox chemistry could be harnessed to effect nucleophilic aromatic substitution (SNAr) of nonactivated substrates. 12,13,14 The SNAr reaction typically requires activated substrates carrying electron-withdrawing groups (Figure 2b), but Nicewicz and coworkers showed that photooxidation of the substrate to the radical cation can increase its reactivity substantially. Here we show that similar reactivity can also be accessed without the use of a separate photoredox catalyst, by sequential photoabsorption of the substrate to access the radical cation intermediate. This sequential photoabsorption is especially notable in light of recent work of Wenger 15 and König 16 ,17 on harnessing the energy of two sequential photoexcitations to perform unprecedented transformations.  (1), experimentally studied derivative (2), computational model (2') and derivative lacking terminal amino group (2"). (b) SNAr reactions generally require an activated substrate with electron-withdrawing groups. Nicewicz showed that oxidation by a photoredox catalyst can activate the substrate. Here we show that photoionization by sequential absorption of two photons can also activate the substrate, without the need for a separate photoredox catalyst.

Photochemistry of amiloride
The primary photoreaction of 1 in water is substitution of Cl to give the hydroxylated product 1-OH ( Figure 3). 18,2 This substitution reactivity contrasts with similar compounds, e.g., frusemide and hydrochlorothiazide, which undergo both photosubstitution and photoreduction. One explanation put forth in the literature is that frusemide and hydrochlorothiazide form an ion-pair complex [Ar-Cl] •+ + [Ar-Cl] •− by electron transfer from one molecule in the excited state to another molecule in the ground state. 18 The resulting [Ar-Cl] •+ radical cation reacts with solvent to form Ar−OH, while the [Ar-Cl] •− radical anion cleaves off a chloride anion and abstracts a hydrogen atom from the solvent to form Ar−H. As 1 only gives the Ar-OH product, Moore and co-workers argued that the mechanism is different, and suggested that substitution occurs by attack of the water solvent on the radical cation of 1, formed by photoionization. 19 The hypothesis was supported by the observation of photoionization of 1 in aqueous solution accompanied by the formation of solvated electrons at both 265 nm and 353 nm, as shown by Hamoudi et al.. 20 The quantum yield of photoionization is 0.011 at pH 7 and increases with pH. The pH dependence of photoionization is similar to that of the quantum yield of amiloride photodegradation which increases from 0.009 at low pH (4.0) to 0.023 at high pH (10.4), with an inflection point at ca pH 8. 19 The location of the inflection point compares well with the pKa of 1 at 8.7. It is expected that the neutral base form of amiloride would give away an electron more readily than the pronated, cationic form. Interestingly, Hamoudi did not observe the formation of solvated electrons in i-PrOH. 20 Instead, the triplet amiloride was postulated to abstract hydrogen from i-PrOH. While the primary photoreaction of 1 is photosubstitution, the compound degrades further upon continued irradiation as its photoproduct 1-OH absorbs in the same region. Moore studied photodegradation under oxygen-free conditions using a medium pressure mercury lamp (Pyrex glass filter, >300 nm) and found that 1-OH forms with apparent first-order kinetics over the studied pH range of 4-11. 18 The 1-OH product exists predominately as its keto tautomer. After ca 50% conversion, unidentified secondary photoproducts started to appear. Using similar conditions, Calza and co-workers suggested a dihydroxy-substituted product which degraded upon continued irradiation ( Figure 3). De Luca and co-workers studied the pH dependence at 300-800 nm under aerobic conditions and suggested three secondary photoproducts based on LC-MS analysis ( Figure  3). 21 An explanation for the discrepancy between these studies may be that Calza and co-workers used anaerobic conditions, promoting reactivity from the triplet state of 1 and preventing singlet oxygen-induced degradation pathways.

Results and discussion
To clarify the mechanism of the primary photoreactivity of the amiloride family, we chose to study compound 2 in detail. Compound 2 was synthesized in a single step from commercially available building blocks according to a literature procedure (Scheme 1). 22 The absorption spectrum in methanol shows three main peaks in the UV region, at 210 nm, 273 nm and 360 nm ( Figure 4). The spectrum of 2 is very similar to the spectrum of 1 in water, which has main peaks at 212 nm, 285 nm and 362 nm. 23   Initial photochemical experiments at 350 nm in argon-saturated H2O gave the hydroxy-substituted product 2-OH, showing that 2 reacts by photosubstitution in the same way as 1 (Table 1). Unfortunately, we were not able to determine the NMR yield due to the overlap between the reactant and the product peaks. In methanol, the methoxy-substituted 2-OMe was obtained with 68% NMR yield after 2 h of irradiation with traces of an unknown by-product. Irradiation in the more substituted alcohol solvents EtOH, i-PrOH and t-BuOH for the same time also resulted in photosubstitution but with lower NMR yields (60%, 46% and 20% respectively, entry 3-5). As far as we know, this is the first report of photoreactivity in alcoholic solvents for amiloride-like compounds. To test for the influence of the terminal amino group, we synthesized derivate 2" (Figure 2), where the alkyl amine is replaced with a methyl group. Irradiation of 2" under the standard conditions led to 57% NMR yield of the methoxy-substituted product 2"-OMe, showing that the terminal alkyl amine is not essential for the observed photoreactivity. Varying the light intensity using 2-16 of the lamps in the Rayonet reactor showed consistent reactivity with 2-OMe as the photoproduct ( Figure S29). Reaction condition: A 0.002 M solution of 2 was irradiated at 350 nm for 2 hrs using a Rayonet reactor. a) Not able to determine the NMR yield as signals from the product and the starting material overlap. b) Only starting material was observed. * NMR yield.

Mechanistic scenarios
In order to elucidate the reaction mechanism, we investigated several plausible scenarios ( Figure  5). After Moore's hypothesis of photoionization and subsequent nucleophilic attack by the solvent (Figure 5.3) was formulated, 19 another possible mechanism for photoinduced dehalogenation was reported by Fagnoni and co-workers for similar substrates. 24 This mechanism involves initial formation of an aryl cation by heterolytic cleavage of the C-Cl bond in the S1 or T1 state ( Figure  5.1). The resulting ground state singlet aryl cation then reacts with a nearby solvent molecule. We call this mechanism "photo-SN1", by analogy with its ground-state equivalent. We have further computationally investigated the possibility of a more traditional SNAr mechanism in S1 or T1 ( Figure 5.2). For completeness, we also list photoreduction ( Figure 5.4), which is not observed for neither 1, 2 nor 2 and which will not be considered further.

Photo-SN1 DFT calculations
As an initial plausibility check of heterolytic chloride dissociation and aryl cation formation in S1 and T1, we used TD-DFT calculations on the model compound 2' (Supporting Information, Section 2.1). The calculations showed a barrier for chloride dissociation of only 5.0 kcal/mol in the T1 state and 5.4 kcal/mol in the S1 state. These low barriers would make the reaction feasible at least in the T1 state, given its longer lifetime. However, the calculations also showed that the reaction was endergonic both in T1 (ΔG = 3.9 kcal/mol) and S1 (ΔG = 5.0 kcal/mol), raising the question whether it would be disfavored thermodynamically. Given that dissociation would happen, the resulting aryl cation would relax to a ground state of either singlet or triplet multiplicity. 25 Our DFT calculations indicate that the triplet is the ground state, 1.6 kcal/mol below the singlet, which is also confirmed by high-level CASPT2 calculations (2.3 kcal/mol). The calculations show that the low-lying singlet aryl cation would add solvent molecules (H2O, MeOH, i-PrOH or t-BuOH) without any barrier, while the ground state triplet cation is unreactive (Supporting Information, Section 2.1.3). However, with an energy gap of only ca 1-2 kcal/mol between singlet and triplet, inter-system crossing would be feasible and allow the triplet state to convert to the reactive singlet state. In summary, the photo-SN1 mechanism seems plausible from a computational point of view, but as we shall see below, it turns out that it is not consistent with experimental findings.

Competition experiments
To test the photo-SN1 hypothesis experimentally, we turned to competition experiments. We first tested competition in solvent mixtures. The aryl cation intermediate is extremely reactive and would not discriminate between nucleophiles. In MeOH/H2O (1:1 v/v), 2-OH and 2-OMe were obtained in an equal ratio (Table 2, entry 1). As the molar ratio of MeOH to H2O is 1:2.2, there is a slight preference for MeOH over H2O as nucleophile. With MeOH/EtOH (1:1 v/v), a 3:1 mixture of 2-OMe and 2-OEt was obtained (Table 2, entry 2). As the molar ratio is only 1.4:1, MeOH is here preferred over EtOH. In MeOH/i-PrOH (1:1 v/v), eight times as much 2-OMe was obtained than 2-O-i-Pr, showing that MeOH is clearly superior to i-PrOH as nucleophile as the molar ratio is in its favor only by 1.9:1. Although the differences in reactivity between the different nucleophiles is strong evidence against the photo-SN1 mechanism, they could in principle be consistent with subtle differences in preferential solvation of the aryl cation. 26 Therefore, we carried out another round of experiments. Aryl cations are known to coordinate to acetonitrile, resulting in the formation of acetamides after attack by water ( Figure 6). This reaction, which is the photochemical equivalent of the Ritter reaction, has been observed previously for aryl cation intermediates of other compounds by Albini and co-workers. 27 Our DFT calculations (Supporting Information, Section 2.1.4) show that acetonitrile coordination to the singlet aryl cation of 2' is highly exergonic (ΔG = −53.0 kcal/mol). Subsequent addition of H2O occurs with a barrier of 17.4 kcal/mol, which should be feasible at room temperature. Therefore, acetamide product 3 would be expected to form upon irradiating 2 in MeCN or MeCN/H2O mixtures. However, irradiation in neat MeCN did not result in acetamide 3 or the alternative cyclized imidazopyrazine 4. Irradiation in solvent mixtures of MeCN with MeOH or H2O (9:1 v/v) only gave trace amounts of 2-OMe and 2-OH, respectively, and no evidence of 3 or 4. If the aryl cation is formed, it should react statistically as addition of MeOH/H2O and MeCN are all barrierless processes. Furthermore, irradiation of 2 at 365 nm in DMSO in the presence of KCN (10 equiv.) did not result in any cyano-substituted product. Also, no reaction occurred under irradiation in MeCN with tetrabutylammonium cyanide (TBACN, 10 equiv). Experiments with π-nucleophiles also make the photo-SN1 mechanism unlikely. The triplet aryl cation, which is predicted to be the ground state according to our calculations, should readily add π-nucleophiles to give the allylated or vinylated product, respectively. 24 We tested reaction with both allyltrimethylsilane and styrene in MeCN/MeOH (9:1 v/v) with irradiation for 2 hrs at 365 nm, but no addition product was found. Lack of reactivity with styrene and allyltrimethylsilane also indicate that radicals are not generated during the reaction.
In conclusion, an array of experiments are inconsistent with the photo-SN1 mechanism ( Figure 6). The non-statistical outcomes of the competition experiments in mixed solvents and the lack of reactivity with MeCN and CN − are not consistent with barrierless addition of the closest nucleophile to a very reactive singlet aryl cation intermediate. The lack of reactivity with πnucleophiles is not consistent with a triplet aryl cation intermediate. One possible explanation why the reaction does not occur despite the calculated small barrier is the endergonicity of the initial dissociation of Cl − in S1 and T1.

Concerted or stepwise SNAr
We further considered the possibility that the photoreaction proceeds via a SNAr mechanism in the excited state. However, DFT calculations showed a high barrier for attack by H2O in the T1 state (36.4 kcal/mol, using another H2O as general base). This barrier is much too high for the reaction to take place in this short-lived excited state. Interestingly, the reaction is calculated to be concerted, with displacement of Cl − occurring simultaneously with C-O bond formation. Considering OH − as the nucleophile (standard state of 10 −7 M) lowered the barrier to 18.9 kcal/mol, which we deem unlikely given the short lifetime of the excited state and that the reaction occurs even at low pH (vide infra). The barrier is of similar magnitude in the S1 ππ* state (22.1 kcal/mol). For more details, see Supporting Information, Section 2.2. Based on these calculations, the photo-SNAr mechanisms is highly implausible.

Detection of photocurrent and thermodynamics of photoionization
Having excluded the photo-SN1 and photo-SNAr mechanism, we then assessed photoionization followed by nucleophilic attack by the solvent (Figure 5.2). As shown recently by Nicewicz in the context of photoredox catalysis, arene radical cations are much more prone to undergo SNAr reactions than the neutral compounds. 12,13 We investigated photoionization of 2 by photoelectrochemistry in MeOH, EtOH and i-PrOH. A photocurrent corresponding to the release of solvated electrons was detected in MeOH and was found to correlate with light intensity and applied bias potential (see Supporting Information, Section 9). The formation of solvated electrons is in accordance with the flash photolysis experiments by Hamoudi. 20 Photocurrent measurements in a series of alcohols (MeOH, EtOH and i-PrOH) detected similar photoinduced currents within experimental error in all tested solvents, in line with the observed formation of substitution products. The detection of a photocurrent for 2 in i-PrOH is different compared to 1, for which Hamoudi detected solvated electrons in H2O but not in i-PrOH.
Hypothesizing that more electron-rich arenes should photoionize more efficiently, we compared the reactivity of 2 with its ester analogue S2 (see Supporting Information, Section 8 and 9). Cyclic voltammetry showed that S2 is more electron-poor than 2, as evidenced by its higher redox potential. Consistent with its electronic properties, we observed smaller generated photocurrent and lower reactivity by a factor of 6 as compared to 2. The monoamino-substituted pyrazine derivative S1 has an even higher redox potential and no detectable photocurrent. Below, we discuss the influence of the frontier orbital energies for the rationalization of photoionization.
While Hamoudi and co-workers argued that photoionization of 1 is a monophotonic process at 353 nm and a biphotonic process at 265 nm, 20 we find that this is not thermodynamically feasible. The calculated solution-phase vertical ionization potential (IPvert) of 2 is 6.08 eV (204 nm), very close to the 5.98 eV (207 nm) calculated for 1. Considering the possibility of adiabatic ionization, and that the electron is transferred to bulk water with a hydration enthalpy of −1.34 eV, 28 the bestscenario cost of photoionization is reduced to 4.57 eV (271 nm) for 2 and 4.43 (280 nm) for 1. Clearly, these values are still too high for a monophotonic process at 353 nm, even allowing for a large error in the calculated values. As noted by Grabner and El-Gogary, 29 a linear relationship between laser power and concentration of solvated electrons, as was observed by Hamoudi, is not sufficient evidence to conclude a monophotonic process. Measurements over a wide range of irradiation powers and correcting for non-linear behavior with pulse energy would be needed, which was not done by Hamoudi. 20 For a more extensive discussion, see Supporting Information, Section 2.3.2. We therefore conclude that the photoionization of 1 and 2, as shown by the observed photo-current, is a biphotonic process not only at 265 nm but also at 353 nm. This sequential photoionization could progress either through the singlet or triplet excited states, although the triplet would be more likely considering its longer lifetime.

Identity of the nucleophile
Having shown that photoionization occurs in all protic solvents investigated, we then turned to the next step of the reaction: nucleophilic attack on the radical cation. The nucleophile could be either a neutral or deprotonated solvent molecule depending on the pH. DFT calculations showed that attack by H2O on the model 2' •+ occurs with a considerable barrier of 27.5 kcal/mol using the SMD implicit solvent model. However, the barrier was lowered considerably to 19.6 kcal/mol by adding another explicit water molecule acting as a general base. Barriers of similar height were obtained for 2 •+ and was not affected significantly by protonation of the terminal amine (see Supporting Information, Section 2.3.1). Considering the known limitations of DFT for reactions of charged species in solution, we believe that these barriers are consistent with reaction on the microsecond timescale as dictated by the expected lifetime of the radical cation. 30,31 A competing pathway in neutral or slightly basic solution is attack by hydroxide or alkoxides. We measured the pH of a 0.8 mM aqueous solution of 2 to 8.5-8.7, which would mean a hydroxide concentration of ca 4 μM (Supporting Information, Section 5). DFT calculations indicate that OH − and MeO − react with 2' •+ in a barrierless manner, while EtO − , i-PrO − and t-BuO − show manageable barriers of 5.9 kcal/mol, 5.4 kcal/mol and 11.1 kcal/mol, respectively (a standard state of 10 −7 M is used for the alkoxides). These computational results are consistent with the slower reactivity in EtOH, i-PrOH and t-BuOH as compared to MeOH and H2O (Table 1). However, there are also other plausible explanations for the observed difference in reactivity. One such explanation is the lower rates of autoprotolysis of the more substituted alcohols (Table 3). For example, the ambient concentration of OH − in H2O is 14 orders of magnitude larger than that of t-BuO − in t-BuOH. As 2 is a weak base, the alcohols' pKa is also relevant ( Table 3). The more substituted alcohols are harder to deprotonate, and we therefore expect lower concentrations of alkoxide nucleophile. The nucleophilicities of the alkoxide anions follow the same trend (Table 3). To test the hypothesis that OH − is the main active nucleophile, we performed competition experiments with AcO − in a 1:1 ratio (v/v) of AcOH and H2O. When both AcOH and H2O are present, the ratio AcO − /OH − should be determined by their relative pKa, ca. 10 12 . Detailed calculations show that AcO − should be present in a concentration of ca 0.02 M with a pH of 1.8, meaning that the ratio of AcO − to OH − is 3 ·10 10 (see Supporting Information, Section 4). Computationally, we find that attack by AcO − on 2' •+ has a barrier of only 8.0 kcal/mol (10 −2 M standard state). In case the reactive species is the anion, acetate should therefore outcompete hydroxide by a wide margin. From the experiment performed in a 1:1 ratio (v/v) of AcOH and H2O, we see trace amount of 2-OH by LC-MS (227 m/z) and no acetylated product 2-OAc.
Although OH − could well contribute at higher pH, we therefore find that H2O must play a role, and likely dominates the reactivity at neutral or low pH. As indicated by the calculations above, this reactivity would rely on general base catalysis to increase the nucleophilicity of H2O. To test this hypothesis, we carried out irradiation of 2 in D2O vs H2O, and found that the reaction in H2O is ca. 10% faster than in D2O. This is clearly a solvent isotope effect, as expected if a selectivitydetermining nucleophilic attack by a solvent molecule is assisted by a general base. The observed rate enhancement is a function of a competition between the rate of nucleophilic attack and the unknown rate of electron recombination (leading back to starting material). We therefore cannot determine an exact value for the KIE, but we can see that it is clearly larger than one. We therefore conclude that the nucleophile at neutral and low pH must be the neutral species, assisted by a general base in solution. Another piece of evidence against OH − as the active nucleophile comes from experiments by De Luca et al. They observed similar rates of formation of photoproduct at pH 3 and pH 7, even though the concentration of OH − is negligible at pH 3 (10 −11 M). 21 As the photochemical reaction is electroneutral, the radical cation must be reduced after the substitution step (Figure 7). This implies that the solvated electrons cannot quantitatively react with solvent to produce dihydrogen and hydroxide (alkoxide). Due to its negative charge, it may instead enhance the ability of the local water molecule to act as a general base towards the nucleophile in the substitution reaction, eventually recombining with the product radical cation to give the observed product. Considering that the solvent acts as nucleophile with a high effective concentration, we believe that this process is competitive with decay of the solvated electrons, which likely occurs on the μs timescale. 34 Figure 7. Suggested mechanism for photosubstitution of 2. Photoionization occurs by sequential absorption of two photons, with the second absorbed either from the S1 state (less likely) or T1 state (more likely). The resulting radical cation is attacked either by solvent or hydroxide/alkoxide with concerted loss of Cl − . Deprotonation and recombination with the electron leads to the substitution product 2-OH. The solvated electron plausibly acts as a general base to assist reactivity.

Excited state antiaromaticity as a potential driving force for photoionization
Having established that photoionization by sequential absorption of two photons is the most likely mechanism, we now turn to the question of driving force. One possible explanation for the facile photoionization of 2 is based on excited state antiaromaticity. Ground state aromatic compounds become antiaromatic in their lowest singlet and triplet ππ* states according to Baird's rule, the excited state equivalent of Hückel's rule. 35 Aromatic compounds are much more reactive when photoexcited, 36 , 37 and it has recently been shown that new photochemical reactions can be developed with relief of excited state antiaromaticity as the driving force. 38,39 We now hypothesize that excited state antiaromaticity can also be a driving force for photoionization from the excited state.
Aromaticity in the ground state is linked with high ionization potentials (IP), 40 while non-aromatic and anti-aromatic compounds display lower IPs (Figure 8a). The reason is that antiaromatic compounds have smaller frontier orbital splittings, and their HOMOs are consequently higher in energy than for aromatic compounds. Photoionization could occur either in a monophotonic process directly from the HOMO, or from the LUMO in a biphotonic process via initial excitation to the excited state (HOMO → LUMO). For the monophotonic process, ground state aromatic compounds like 2 should have a relatively high IP. For the biphotonic process, on the other hand, excitation occurs from a relatively high-lying LUMO and should be more facile. Within this context, removing an electron from the LUMO could be seen as a way for excited state antiaromatic compounds to relieve their antiaromaticity. This qualitative picture is corroborated by our calculations, showing that the IP decreases much more going from ground to excited state for ground state aromatic compounds than for antiaromatic ones (see Supporting Information Section 2.3.3). Indeed, intramolecular charge transfer of an electron from an excited state antiaromatic benzene ring was recently found to occur during photodissociation of a protecting group. 41,42 Excited state antiaromaticity is also involved in the fast excited-state deactivation of DNA base pairs, where it serves as a driving force for electron-driven proton transfer. 43 Now, how does the aromaticity of 2 change upon photoexcitation and subsequent ionization to the radical cation? We investigated this with calculations on the model compound 2'. Ring currents according to the anisotropy of the induced current density (ACID) method 44 reveal a moderate diatropic (aromatic) ring current in the S0 ground state of 2', while the relaxed T1 state is antiaromatic with an appreciable paratropic ring current (Figure 8a). Photoionization to the radical cation 2' •+ is accompanied by loss of antiaromaticity, leading to non-aromaticity. These findings are further corroborated by multicenter index (MCI) calculations and nucleus-independent chemical shift (NICS) scans (Figure 8b). 45,46,47 Furthermore, the similar MCI values for the relaxed T1 (0.0057) and S1 states (0.0089) indicate that the S1 state of 2' is also antiaromatic. The loss of aromaticity according to MCI is consistent with that for the parent pyrazine going from the ground state (0.0644) to the corresponding ππ* triplet (0.0019) and singlet (0.0061) excited states. Due to the longer T1 than S1 lifetime, it is likely that photoexcitation occurs from the T1 state. In summary, a contributing factor to the facile photoionization of 2 could therefore be antiaromaticity in the excited state, which is alleviated when going from the T1 state to the radical cation.

Conclusions and outlook
We have studied the apparent photo-SNAr reaction of amiloride analogs by a combination of experimental and theoretical methods, to evaluate possible reaction mechanisms. The simplest possibility, a conventional SNAr reaction in the excited state, can be excluded by the high calculated barriers in both singlet and triplet states. It is therefore clear that the excited state needs to evolve to a more reactive state. This could occur by dissociation of the chloride in an SN1-type fashion, giving an aryl cation, or by ionization, giving an arene radical cation. Heterolytic chloride dissociation is found to be possible, albeit endergonic. However, the observed reactivity profile does not agree with either a singlet or triplet aryl cation, and we can therefore exclude a photo-SN1 reaction pathway. Due to the anti-aromaticity of the excited state, the energy of the excited electron is high, but not sufficiently high for spontaneous ionization. However, a second photon can eject the electron into the surrounding solvent. The solvated electrons are detected, giving support for the ionization hypothesis. The experimental selectivity profile combined with calculated barriers indicates that the radical cation reacts with a neutral solvent molecule, assisted by another solvent molecule acting as a general base. It is plausible that the basicity of the solvent is strengthened by the nearby solvated electron. We have observed that several protic solvents are competent nucleophiles, but neutral solvents such as acetonitrile are unreactive, giving further support to the hypothesis that a general base is necessary for reactivity.
From a pharmaceutical perspective, understanding the degradation mechanism is the first step in creating predictive tools able to alert developers of drugs to potential light sensitivity. The mechanism studied here could also be an important tool to understand the fate of agrochemicals, which generally are exposed to light, water, and air in the environment. In organic synthesis, the reaction studied here is closely related to recent advances allowing mild SNAr reactions of electronrich substrates. 12

Supporting information description
The supporting information contains a detailed description of the materials and methods, calculations of singlet-triplet gap of aryl cations, analysis of the nature of excited state of compound 2, an analysis of mono-or biphotonic photoionization of amiloride, pH measurements, experiments with deuterated solvent, a computational analysis of the effect of protonation on barrier of attack on radical cations, UV and fluorescence spectra, cyclic voltammetry and photocurrent measurements, NMR spectra and chromatograms. Cartesian coordinates for computed structures are given in a separate file.

Spectroscopy
Absorption spectra were obtained on a scanning UV-Vis spectrometer and a diode-array UV-Vis spectrometer with matched 1.0-cm quartz cells. Fluorescence spectra were recorded on a diodearray automated combined luminescence and UV-Vis spectrometer in 1.0 cm quartz fluorescence cuvettes at 23 ± 1 °C. The sample concentration was set to keep the absorbance below 0.5 at λirr and a correction for self-absorption was applied for each spectrum. Emission spectra were also corrected using standard correction files. Each sample was measured 3-5 times, and the spectra were averaged. Due to the photosensitivity of the compounds, fresh samples were used for each measurement.

Cyclic voltammetry
The electrochemistry was recorded in methanolic solutions of pyrazine derivative (c = 1 mM) using tetrabutylammonium hexafluorophosphate as a conducting salt (c = 0.1 M). The electrochemical measurements were accomplished using a glassy-carbon disc working electrode, a platinum counter electrode, and a silver pseudo-reference electrode. The cyclic voltammograms in the measurement range (+1.2 V to −0.7 V) were recorded three times.

Photochemical experiments
Standard experimental procedure The photochemical reactions were carried out for 2 hours in a Rayonet reactor with 16 UV lamps (300-460 nm, λmax = 350 nm). Solutions of concentration 2 mM in quartz test tubes were degassed by bubbling with argon for 10 min prior to irradiation. After 2 hours of irradiation, residual solvent was removed under reduced pressure and the crude product was analyzed using 1 H NMR and LC-MS.

Experiments with π nucleophiles
Allyltrimethylsilane and styrene (30 equiv.) were added to two independent quartz test tubes containing 2 in MeCN/MeOH (9:1, v/v). The test tubes were sealed with rubber septa, degassed using argon for 10 mins and irradiated for 2 hours at 350 nm. Excess solvent was removed under reduced pressure and the crude products were analyzed using 1 H NMR. In both the cases, only traces of 2-OMe were observed and no addition product could be seen. Most of the starting material remained unreactive. Allyltrimethylsilane was completely consumed after irradiation, as shown by 1 H NMR (see Supporting Information, Figure S31). The allyltrimethylsilane was probably consumed by protodesilylation under the slightly acidic conditions to give volatile propene. 48 Importantly, neither styrene nor allyltrimethylsilane show any absorbance above 300 nm and were therefore not photoexcited.

Solvent isotopic experiments
Two separate 2 mM solutions of compound 2 were prepared in H2O and D2O. The solutions were then degassed for ca 20 min by bubbling with argon and irradiated simultaneously in the Rayonet reactor with 18 350 nm UV-lamps. After every 30 mins of irradiation, samples were carefully withdrawn from the solutions while maintaining the inert atmosphere in the reaction flask using argon. The residual solvent was removed under reduced pressure, and the crude 1 H NMR spectra were recorded and analyzed for NMR yield. The solutions were irradiated for a total period of 2 hours.

Photocurrent experiments
The photocurrent measurements were accomplished using a spectroelectrochemical cell irradiated with a UV-light source. The light was gradually turned on and off and the induced photocurrent was measured by a chronoamperometric method applying a bias potential to accelerate the ionized electrons towards the electrode. A detailed description of experimental procedure is shown in the Supporting Information.

Quantum-chemical calculations
All quantum-chemical calculations, unless otherwise noted, were carried out with Gaussian 16, revisions A.03 and B.01. 49 Geometry optimizations were done at the B3LYP-D3(BJ)/6-31+G(d)/SMD level, i.e., with the B3LYP functional 50 , the 6-31+G(d) basis set 51 , the SMD solvation model 52 and the D3-BJ dispersion model. 53 Stationary points on the potential energy surface were confirmed with frequency calculations and transition states corroborated by quick reaction coordinate calculations. 54 Final energies were obtained from M06-2X-D3/6-311+G(d,p)/SMD single point calculations at the B3LYP geometries, i.e., using the M06-2X functional 55 , the 6-311+G(d,p) basis set 56 , the SMD solvent model and the D3 dispersion model, 57 together with thermal contributions from B3LYP-D3(BJ)/6-31+G(d)/SMD. Standard state corrections were applied to give a standard state of 1 M for reactants and 55.5 M for H2O, unless otherwise noted. Triplet state calculations used the same levels of theory but with unrestricted DFT.
Optimization and frequency calculations in the excited S1 state used TD-DFT with B3LYP-D3(BJ)/6-31+G(d)/SMD based on a ground-state restricted reference. Final energies were obtained by combining the thermal corrections at the optimization level of theory with single-point energies at the M06-2X/6-311+G(d,p)/SMD level, enforcing equilibrium solvation treatment with the keyword "IOp(9/73=2)". MS-CASPT2 58,59 calculations of the aryl cations were done with OpenMolcas 18.0 60 with an active space of 6 electrons in 7 orbitals and the ANO-RCC-VDZP basis set. 61 An imaginary shift of 0.2 a.u. and an IPEA shift of 0.0 a.u. 62 were used. Further details on the reference wave function and active space orbitals can be found in the Supporting Information, Section 2.1.2.
Ionization potentials and electron affinities were calculated with the ΔSCF approach, using the energy of the neutral molecule and its radical anion/cation. Vertical values were obtained by single-point calculations of the anion/cation at the geometry of the parent molecule, while adiabatic values were obtained by allowing the anion/cation to relax. For vertical values, we used the electronic energy difference and for adiabatic values we used the free energy difference.
NICS scans 45,46 were performed with Aroma 1.0, 63 and ACID plots 44 with the AICD 2.0.0 package, both using Gaussian 09, revision E.01. 64 MCI values 47 were obtained using the ESI-3D program, 65 based on the Quantum Theory of Atoms in Molecules atomic partition and the integration scheme as provided by the AIMAll package. 66 MCI values for pyrazine in the lowest ππ* triplet state and its radical cation were calculated by optimizing the structures in D2h symmetry with the orbital occupation altered (keyword "guess=alter") and employing symmetry in the SCF calculation (keyword "scf=symm"). For the ππ* singlet excited state, we used TD-DFT, optimizing the third root. For the singlet and triplet states, the D2h-symmetric geometry was not a minimum.