[1] 23w Why Have Clinical Trials of Antioxidants to Prevent Neurodegeneration Failed? -A Cellular Investigation of Novel Phenothiazine-Type Antioxidants Reveals Competing Objectives for Pharmaceutical Neuroprotection

[1] 192w Purpose Only a fraction of the currently established lowmolecular weight antioxidants exhibit cytoprotective activity in living cells, which is considered a prerequisite for their potential clinical usefulness in Parkinson's disease or stroke. Post hoc structure-activity relationship analyses have predicted that increased lipophilicity and enhanced radical stabilization could contribute to such cytoprotective activity. Methods We have synthesized a series of novel phenothiazinetype antioxidants exhibiting systematic variation in their lipophilicity and radical stabilization. Phenothiazine was chosen as lead structure for its superior activity at baseline. The novel compounds were evaluated for their neuroprotective potency in cell culture, and for their primary molecular targets. Results Lipophilicity was associated with enhanced cytoprotective activity, but only to a certain threshold (logP ≈ 7). Benzannulation likewise produced improved cytoprotectants that exhibited very low EC 50 values of ~8 nM in cultivated neuronal cells. Inhibition of global protein oxidation was the best molecular predictor of cytoprotective activity, followed by the inhibition of membrane protein autolysis. In contrast, the inhibition of lipid peroxidation in isolated brain lipids and the suppression of intracellular oxidant accumulation were poor predictors of cytoprotective activity, primarily as they misjudged the cellular advantage of high lipophilicity.

[1] 165w Oxidative stress is widely assumed to constitute a common and pathologically relevant event in acute and chronic n eurodegenerative conditions such as Parkinson's disease, Alzheimer's disease, or stroke (1,2). Accordingly, antioxidant treatment options have been extensively investigated in animals and in vitro, especially with respect to stroke, and have resulted in a large number of apparently effective treatment strategies (3,4). However, none of the strategies that have been tested clinically as disease-modifying therapy for stroke has demonstrated any substantial benefit in patients. The most recent, illustrative failure was the spin trap NXY-059 (5), which in the beginning had appeared to be a quite promising compound, as judged from at least some of the preclinical data (6). Similarly disappointing clinical outcomes of initially deemed efficacious compounds have been recorded for the lazaroid tirilazad (6,7), for the selenocompound ebselen (6,8), and for edaravone (9), even if the latter compound has been approved for clinical use in Japan on the basis of a single, borderline significant study (6,10).

[2] 183w In essence, these failures might be attributable to three major causes. First, oxidative stress may not be causally relevant for disease progression in the investigated cases ("mechanistic failure"). And even if oxidative stress had a causal role at disease onset or early during the course of a disease, it may not be causally involved in the pathological cascade at the time of intervention. This is especially important in stroke, in which clinical treatment was usually initiated after approximately 4 h (5,7) whereas the majority of oxidative damage seems to arise much earlier (11,12). What complicates the situation further is the general biological discrepancy between the studied animal cohorts (mostly young, healthy, genetically homogeneous rodents) and the ultimately targeted patient cohorts (usually elderly individuals with diverse backgrounds and comorbid conditions). Clearly, the mechanisms and manifestations of oxidative stress may also differ between those cohorts (13). This problem has been addressed in the 2009 update of the STAIR ("Stroke Therapy Academic Industry Roundtable") criteria (14); still, there has been only moderate progress in their implementation, probably due to the multiplied expenses that would be required.

[3] 158w Second, there has been insufficient knowledge about the biological equilibria that naturally maintain a certain level of oxidative activity for control and signaling purposes ("homeostatic failure"). Hence, in response to antioxidant supplementation, endogenous antioxidant defenses tend to retire, potentially resulting in an unaltered or even partially increased level of equilibrium damage (15). For example, switching from a marginally selenium-deficient diet to a selenium-adequate diet has been shown to result in a significantly lower expression of various antioxidant and detoxifying enzymes (16). While the overall consequences of compensatory downregulation of endogenous antioxidant defenses in response to antioxidant supplementation are unknown, it appears plausible that this mechanism is largely responsible for the observation that fruits, i.e. mixtures of antioxidants with prooxidants and electrophiles (17,18), seem to provide better protection against certain degenerative conditions than pure antioxidants alone (17). Endogenous antioxidant defense idleness may be particularly relevant for the outcome of any long-term antioxidant treatment of chronic conditions like Alzheimer's disease.

[4] 174w Third, a surprisingly large fraction of the clinical failures have been attributed to "chemical failure", i.e. the choice of unsuitable and inefficient compounds for reasons beyond scientific rationale. For instance, there is no indication that those drugs have selectively entered clinical stages that were the most potent in preclinical testing (4). In the case of NXY-059, there is clear evidence that inappropriate testing led to an overestimation of its potential benefit (19,20). In addition, it is difficult to understand how a negatively charged, protein and membrane repellent, blood-brain barrier negative (21) spin trap like NXY-059 has ever entered clinical testing as neuroprotective antioxidant when spin traps are notoriously weak antioxidants (22)(23)(24)(25). Chemical failure due to poor blood-brain barrier permeability might also explain the limited success of the otherwise quite potent, highly lipophilic compounds tocopherol (vitamin E) and ubiquinone (coenzyme Q) in Parkinson's disease (2,26,27) and Alzheimer's disease (2,28,29). Both molecules hardly enter the brain when supplemented at high doses (30,31), and rigorous control mechanisms of brain entry and metabolism seem to be operable (32,33).

[5] 150w In an effort to investigate more closely in how far chemical failure of new candidate drugs might have been avoided by in vitro testing, we have screened the cytoprotective potential of numerous biochemical antioxidants representing diverse structural classes and scavenging mechanisms (selected results were published in (24,25,34)). During that screening, the compound phenothiazine has repeatedly emerged as a uniquely potent neuroprotective substance, exhibiting an EC 50 value of ~30 nM in different experimental settings (24,25). To explore phenothiazine's potential for further refinement, we have designed and evaluated a series of novel phenothiazine derivatives exhibiting systematic variation in their lipophilicity/solubility and radical stabilization. Investigating their antioxidant and cytoprotective activities, we find that relevant increases in potency are achievable by using this drawing board strategy. However, at the attained level of efficacy with EC 50 values ≤ 10 nM, biophysical limits appear to impose restrictions to the design of ever better cytoprotectants.

[1] 177w For the investigation of the influence of solubility-modulating side chains on cytoprotective activity, compounds 1-7 were drafted, synthesized (3-7) and analyzed by computational chemistry (Fig. 1). The compounds in this panel were designed (i) to possess an identical hydrogen-donating head group with basically identical radical stabilization, (ii) to comprise a series of members with gradually increasing lipophilicity, conferred by a hydrolysis-resistant substituent (4-7), and (iii) to include a charged member (3) with enhanced solubility in aqueous medium. A cationic amine group was chosen as charge carrier since comparably sized and hydrophobic drugs with amine groups are generally membrane-and blood-brain barrierpermeable (39). Examining the results of the computational characterization of the listed compounds (Fig. 1), it appears that the above criteria were largely met, with the exception that compound 2 may differ in its radical stabilization due to mesomeric interference of the cyano group (reflected in its unusually low LUMOr energy). Moreover, the employed method of logP estimation neglects the effects of ionic charges, such that the listed logP value of compound 3 might underestimate its true hydrophilicity.

[2] 381w The ability of the novel compounds to react directly with preformed radicals (TEAC assay) and to act as chainbreaking antioxidants in native rat brain membranes (TBARS assay) were investigated (Fig. 2). The results of the TEAC assay in Table I indicate that all of the tested compounds possessed a relatively similar reactivity towards the electrophilic ABTS +• radical. In aqueous medium, the acyl chain substituted and larger compounds 4-7 were disfavoured; however, there was no correlation with the predicted lipophilicity of the compounds. This was also the case in ethanolic medium, in which especially compounds 1-3 and 6 were more potent than trolox. Potentially, the low molecular weight of 1-3 and their expectedly higher diffusibility may have influenced the latter result. The relative similarity of all tested compounds 1-7 sharing the same reactive NH group is in accordance with observations made with phenolic antioxidants, in which TEAC activity seems to be primarily related to the number of phenolic hydroxyl groups (40). Regarding the prevention of iron/ascorbate-induced lipid peroxidation in vitro, the moderately lipophilic compounds 2-5 were identically potent, reaching almost complete protection at 100 nM concentration in a total cortical brain lipid preparation (Fig. 2a and b). At first sight paradoxically, high lipophilicity was associated with poorer efficacy (6 and 7), whereas the smallest compound, the intermediately lipophilic molecule phenothiazine itself, was the most potent lipid peroxidation inhibitor. Since compound 7 did not prevent lipid peroxidation even at high concentrations of 1000 nM (though it gained activity when increasing the incubation time from 30 min to 5-20 h; data not shown), it is concluded that the inefficacy of 7 is primarily attributable to its lipophilicity rather than to its higher molecular weight. This idea is supported by the relatively smaller disadvantage of high lipophilicity in a purified, more lipophilic lipid preparation depleted of vasculature (Fig. 2c and d). The resulting micelles containing more very long-chain fatty acids, both polyunsaturated (of neuronal origin) and monounsaturated (from myelin), may offer a more efficient incorporation of very lipophilic drugs, or aid in the solubilization of the latter. In general, though, extreme lipophilicity and the ensuing rapid precipitation or aggregation in the aqueous assay medium after administration seem to result in a slower penetration of adjacent lipid phases, and thus poorer efficacy in acute assays.

[3] 111w Regarding potential clinical applications of chain-breaking antioxidants, cytoprotection is arguably the most widely demanded property, for instance, to prevent degenerative conditions associated with severe cell damage or loss (2). Hence, the cytoprotective potency of the newly designed compounds was tested in murine hippocampal HT22 cells challenged with 5 mM glutamate (Fig. 3). The resulting "oxidative glutamate toxicity" follows a multi-step hierarchical pathway, of which at least two steps are oxidative in nature (41), and has been cited as a relevant cell culture model for neurodegenerative phenomena occurring in vivo (42). Contrasting the findings in rat brain lipids, all compounds including 7 were highly protective against oxidative cell death at 100-200 nM

[4] 166w S H N S H N CN S H N NH 2 S H N N H O S H N N H O S H N N H O S H N N H O 1 2 3 4 5 6 7 M W [Da] logP ΔH f [kcal/mol] E(LUMOr) [eV] 1 199 3.97 26.53 -0.456 2 224 3.38 26.78 -0.951 3 228 2.39 26.18 -0.483 4 298 3.39 35.41 -0.495 5 354 5.06 35.34 -0.494 6 410 6.73 34.95 -0.478 7 466 8.40 34.55 -0.480 Fig. 1 Structures and calculated chemical properties of the phenothiazines employed in this study. The depicted compounds are: 1, phenothiazine; 2, 2cyanophenothiazine; 3, 2-aminomethyl phenothiazine; 4, 2butyrylamidomethyl phenothiazine; 5, 2-octanoylamidomethyl phenothiazine; 6, 2-lauroylamidomethyl phenothiazine; 7, 2palmitoylamidomethyl phenothiazine. The listed properties are: M W , molecular weight; logP, octanol/water partition coefficient; ΔH f , difference in the heat of formation of a substance and its hydrogen-abstracted radical form; E(LUMOr), energy of the lowest unoccupied molecular orbital of the radical form.

[5] 32w concentration. This result was confirmed in a second assay system, propidium iodide staining (data not shown), which measures plasma membrane integrity rather than metabolic viability, as does the otherwise employed MTT test.

[6] 269w Compound 1, the parental phenothiazine, displayed a halfmaximal effective concentration (EC 50 value) of ~30 nM as reported (36,43). Derivatization with a nitrile group (2) or an aminomethyl group (3) at position 2 of the ring system led to a decreased protective capacity of the compounds (EC 50 values of ~50 nM and ~80 nM, respectively) (Fig. 3). The attachment of fatty acyl chains via an amide linkage resulted in an increasing efficacy with rising lipophilicity, from the C 4substituted compound (EC 50 value ~80 nM) to the C 12substituted compound (EC 50 value ~8 nM). However, the long-chain, C 16 -substituted compound did not profit from its additional lipophilicity, as it rather elicited the same, lowerthan-phenothiazine efficacy of ~80 nM like the C 4 -substituted compound. Moreover, the additional protective effect of the C 12 -substituted compound 6 at very low concentrations was not as definitive as the baseline protective effect of all phenothiazines at higher concentrations. In other words, maximum protection was only achieved at 200 nM concentration, almost irrespectively of chain length. These findings may be explained by the following working model: in general, lipophilicity and thus accumulation in membranes is associated with increased efficacy, since (at least) one of (at least) two sites of cytoprotective action of the phenothiazines is within the membrane. However, when a certain, assay-specific threshold is exceeded (i.e. with C 12 -substitution in total cortical lipids, or with C 16 -substitution in cell culture), activity drops again, due to decreasing membrane and cell penetration arising from high-affinity protein binding or aggregate formation. This working model was further tested in the following.

[7] 186w In order to gain additional insight into the potential role of oxidative protein modifications for neurotoxicity, we aimed at detecting proteins that were particularly affected in terms of their selective degradation and thus disappearance from the cellular proteome. Returning to the potentially more physiological and temporally protracted glutamate toxicity, it was investigated by gel electrophoresis which proteins would be particularly depleted in membrane and cytosolic fractions of HT22 cells collected at a relatively late stage in the glutamate toxicity cascade (after 18 h on 5 mM glutamate). At this point, treated cells were still visually indistinguishable from untreated cells and contained the same amount of proteins and peptides as assessed by the copper reduction-bicinchoninic acid (BCA) method. However, a pronounced proteolytic degradation of the majority of membrane proteins was observed (Fig. 6a). In the cytosolic fractions, though, only a very limited number of proteins were affected by proteolytic degradation. To particularize this observation, the specific behavior of a widely expressed transmembrane protein, Na/K-ATPase, was investigated. This protein has been shown to be specifically targeted by ubiquitin-dependent degradation in response to mitochondrial oxidants produced during hypoxia (46).

[8] 137w a b Na/K nsb Membrane fractions 10 nM 100 nM 1 3 6 10 nM 100 nM 1000 nM 10 nM 100 nM 1000 nM 1000 nM untr Cytosolic fractions M C M C M C M C glu M C 6 1000 nM 6 100 nM 6 10 nM 3 1000 nM M C M C M C M C M C 3 100 nM 3 10 nM 1 1000 nM 1 100 nM 1 10 nM M C untr glu untr glu 10 nM 100 nM 1 3 6 10 nM 100 nM 1000 nM 10 nM 100 nM 1000 nM 1000 nM untr glu untr glu 170 130 100 70 55 45 35 25 15 kDa 130 100 70 55 45 35 25 kDa 170 130 100 70 55 45 35 25 15 kDa

[9] 112w Fig. 6 Membrane protein autolysis and loss of Na/K-ATPase. (a) Cultivated HT22 cells were treated with 5 mM glutamate and different concentrations of compounds 1, 3, and 6. After 18 h, the cells were harvested and separated into membrane and soluble protein fractions before protein determination by the BCA method, gel electrophoresis, blotting and Ponceau S staining. The membrane samples exhibited pronounced proteolysis that was prevented by nanomolar concentrations of the antioxidants 1, 3, and 6. In the cytosolic fractions, only selective proteolysis was observed. (b) Anti-Na/K-ATPase α chain Western blotting of the same samples as in (a). nsb, non-specific binding of the antibody to an unknown component in the cytosolic fractions.

[10] 167w Western blot analysis demonstrated that the full-length α chain Na/K-ATPase effectively disappeared in glutamatetreated cells, a phenomenon that could be fully prevented through the administration of the indicated antioxidants 1, 3 and 6 at 100 nM concentration (Fig. 6b). The occasional appearance of fragment bands (e.g. with 10 nM compound 3) supports the idea that during oxidative glutamate toxicity, targeted proteolysis of membrane proteins occurs, most likely initiated by reactive oxygen species. Mechanistically though, it is unclear whether phenothiazines act directly to protect this and other proteins from oxidation, or whether they rather act at an earlier stage and prevent the oxidant-induced oxidant release from the respiratory chain that is characteristic of oxidative glutamate toxicity (41). Regarding the individual efficacies of the investigated compounds, the partial advantage of compound 6 over the parental structure 1 was not confirmed in this assay. Hence, the capacity to preserve cell viability and to prevent carbonyl formation does not run in perfect parallel with the capacity to prevent membrane protein degradation.

[11] 146w To explore the potential for antioxidant refinement that might be hidden in the aromatic core of phenothiazine, the three annulated benzophenothiazines 8, 9 and 10 were investigated (Fig. 7). Approving chemical expectation, these compounds were calculated to possess very low radicalization enthalpies, together with intermediate lipophilicity and a still relatively small molecular weight, making them approximately comparable to compound 4 regarding the latter two parameters. All benzophenothiazines were exceptionally neuroprotective in the MTT cell viability experiment, exhibiting EC 50 values of ~8 nM (Fig. 8). This finding was confirmed in propidium iodide (PI) stainings, in which the linearly annulated substance 9 demonstrated a significant protective effect at a concentration as low as 1 nM (p < 0.01). Compared with the parent molecule 1, benzannulation thus led to an about threefold increase in efficacy in terms of the protection of glutamatetreated HT22 cells from oxidative cell death.

[1] 81w Arguably, there is no other medical specialty in which the discrepancy between clinical demand and pharmaceutical portfolio is wider than in the field of neuroprotection. If ever achieved, neuroprotection has an immense potential to reduce morbidity and mortality, especially in aging populations. Still, after a series of clinically dissatisfying outcomes in the last two decades, it appears that even the very basic strategies of neuroprotection may be in need of reinvestigation (4,53). Indeed, there are preliminary examples that strategic alterations, e.g.

[2] 147w the focusing on narrower patient populations, may result in a higher probability of clinical success (54,55). As part of this endeavor, it seems advisable to go back to basic neuropharmacology again, and to confirm or reject the efficacy of any lead structure in relevant basic research models. In the current work, we have chosen oxytosis as primary endpoint, which is a glutamate-driven toxic cascade leading to oxidative cell death in clonal hippocampal cells (HT22 cells) (56). Cytoprotective activity in living cells can plausibly be considered as a prerequisite for any antioxidant to be used in the clinic, as it is very much unlikely that a noncytoprotectant in vitro will ever be a cytoprotectant in vivo. Oxytosis involves glutathione depletion, 12-lipoxygenase activation, calcium influx, and massive mitochondrial ROS production (41,57), and it has been shown to be part of the excitotoxicity cascade in NMDA receptor-negative juvenile neurons (42).

[3] 183w Neuronal cell culture has proven very useful in the uncovering of false-positive results in antioxidant research, i.e. of seemingly promising biochemical antioxidants in vitro with still limited potential in vivo. For example, the relatively oxidative environment in cell culture has helped in the identification of pseudo-antioxidant compounds that under relevant circumstances, especially pathophysiological conditions, may rather exhibit an adverse, prooxidative redox chemistry (58). As a second example, rationally designed 6-amino-3pyridinols have been found to be unparalleled peroxyl radical scavengers in organic solutions (59). In cortical neurons and neuronal cells, though, they were not better than plain αtocopherol against the cytotoxicity of glutamate, 6hydroxydopamine, or MPP + (60), and thus at least an order of magnitude less effective than phenothiazine under comparable conditions (34,51). Notably, the oxytosis model employed in the current study predicts a very poor cellular antioxidant activity of all four flagship structures (NXY-059, (methyl) tirilazad, ebselen, and edaravone) that have advanced into clinical phases and have ultimately failed to provide any benefit (Fig. 9). Maybe some of these clinical failures could have been avoided by consulting the oxytosis model in advance.

[4] 187w The latter speculation is supported by the observation that efficacy against oxytosis has been found to be of high predictive value for efficacy in other, more complex disease models. For instance, phenothiazine has been shown to possess highly similar EC 50 values of neuroprotection (20-60 nM) in hydrogen peroxide-, MPP + -, and rotenone-treated neuroblastoma cells (34,36), in glutathione-depleted cerebellar neurons (36), in MPP + -treated midbrain neurons (34), and against MPP +and rotenone-induced dopaminergic neurodegeneration in vivo (in a C. elegans model of Parkinson's disease) (51). Beyond this, preliminary data indicate that phenothiazine has a long-lasting neuroprotective effect in a translational rat model of cerebral ischemia analyzed 10 days post intervention (61), which expands earlier data showing a protective effect of a substituted phenothiazine derivative in ischemic rats analyzed after 24-72 h (62). Since phenothiazine also seems to have a robust effect in the rotenone-based model of Parkinson's disease in rats (63) as well as in experimental traumatic brain injury in mice (64), phenothiazine and its derivatives clearly represent preferred lead structures for neuroprotective drug development by means of chemical refinement, as pursued in this study.

[5] 180w In the present work, we have investigated the potential of two strategies of antioxidant enhancement, namely increased lipophilicity, and improved chemical stabilization of the antioxidant radical that is transiently formed during the compounds' antioxidant action. Both parameters have been described to be of relevance in structure-activity relationship analyses performed in retrospect on phenothiazines (36,65), phenolic antioxidants (66)(67)(68)(69), or in a cross-comparison of phenothiazines with phenolic antioxidants (49). However, the extent to which antioxidant activity may still be positively influenced in living cells when starting with an already very potent compound like phenothiazine has never been investigated. Moreover, it is clear that both factors, increased lipophilicity and improved radical stabilization, almost inevitably come along with an increased molecular weight, which is generally associated with poorer antioxidant activity due decreased diffusibility (69). Besides, large aromatic and very hydrophobic structures are universally disfavored in pharmacology, be it for common reasons such as their increased cost of synthesis and their more complex administration (70), be it for specific reasons such as their much higher protein binding (71) and their poorer blood-brain barrier permeability (39).

[6] 224w We have found that both strategies yielded approximately the same benefit, namely EC 50 values decreased by a factor of three (Figs. 3, 8, and 9). This benefit materialized with a roughly similar degree of chemical effort (benzannulation and lauroylamidomethylation), resulted in still moderately sized molecules (249 and 410 Da, respectively), and may be viewed to be substantial, especially as the chosen starting material (plain phenothiazine) had been selected for its already high efficacy. For example, we have found an efficacy advantage of plain phenothiazine by a factor of ~1000 over trolox or edaravone in this study (Fig. 9), and by a factor of at least 30,000 over bilirubin, β-carotene, ergothioneine, epigallocatechin gallate, genistein, phenyl butyl nitrone (PBN), or uric acid in earlier studies (24). The aminic H-donating phenothiazines 1 (plain), 6 (C 12 -acylated) and 10 (benzo[c]-annulated) were compared to (b) four phenolic and enolic H-donors (α-tocopherol (TOC), butylated hydroxytoluene (BHT), trolox (TROX), and ascorbic acid (ASC)), and to (c) four structurally diverse, but clinically tested antioxidants (the selenocompound ebselen (EBS), the pyrazolone edaravone (EDA), the lazaroid U74389G, and the spin trap NXY-059 (NXY)). Neuronal HT22 cells were challenged with 5 mM glutamate and analyzed as in Fig. 3 (triplicate determinations). Phenothiazines were generally superior to phenols, which in turn were mostly superior to compounds that have been tested in clinical neuroprotection trials.

[7] 181w In both cases, though, the limits of each strategy became manifest rather rapidly. Regarding lipophilicity, there seems to be a clear optimum in living cells: while increasing lipophilicity was basically associated with increasing cytoprotective activity as described (36,49), an over-stretching of this parameter is possible (Fig. 3). Most likely, excessive lipophilicity prevents compounds from reaching their site of action, even if this site of beneficial action is the lipid bilayer. This conclusion may hold true in vivo, where excessive lipophilicity usually results in high-affinity protein binding (71) and the descent into lipoproteins (72), which should both negatively affect efficacy. In view of the existing transport systems for very lipophilic chemicals across the blood-brain barrier and within the brain, it remains to be determined whether the lipophilicity optimum in vivo might be somewhat higher than in the current cell culture study (logP ≈ 7). Still, the definitive exclusion of pharmacological doses of tocopherol (logP ≈ 12) from the brain sets an upper benchmark also for synthetic neuroprotective drugs (73), unless advanced delivery strategies will one day shift this benchmark further up (74).

[8] 118w It is interesting to note in this respect that the best molecular predictors of cytoprotective activity were the inhibition of protein oxidation (Fig. 5) and membrane protein autolysis (Fig. 6), which is consistent with a central role of membrane protein oxidation in neurodegenerative processes (75). Remarkably, the human brain appears to be particularly prone to high membrane protein oxidation, as opposed to high lipid peroxidation, for unknown reasons (13,76). In terms of predicting neuroprotective activity, the widely employed DCFA oxidation assay returned a misleading result (Fig. 4). It seems that the DCFA assay and the Fe 2+ -induced brain lipid peroxidation assay (Fig. 2), have their optimum lipophilicity at lower logP values than authentic cell culture neuroprotection experiments.

[9] 114w Regarding increased radical stabilization, a certain surprise of the current study is the relatively small efficacy gain comp a r i n g p l a i n p h e n o t h i a z i n e w i t h e a c h o f t h e benzophenothiazines (Fig. 7). Even substantially lower radicalization enthalpies only translated into approximately threefold lower EC 50 values of cytoprotection. Given the poor developability of compounds with aromatic ring counts ≥ 5, it is unclear whether further expansion of the aromatic system of benzophenothiazine might ever yield applicable drugs. Nevertheless, other radical stabilization strategies beyond simple benzannulation certainly deserve further study.

[10] 113w In summary, the current data demonstrate that increased lipophilicity and enhanced radical stabilization can both enhance antioxidant cytoprotective activity in an already nanomolar potency lead structure like phenothiazine. However, they also indicate that both strategies come along with downsides, as exaggerated lipophilicity might readily nullify cytoprotective activity due to solubility problems, whereas widely expanded carbo-aromatic systems have limited potential for drug development for different reasons. Comparing phenothiazine with established antioxidant drugs including those that have advanced to the clinic, phenothiazine appears to be clearly more attractive than the latter, which were astoundingly weak cytoprotectants throughout. Our findings support the idea that "chemical failure" was centrally involved in the clinical ineffectiveness of these drugs.

[1] 88w All reagents and solvents were purchased from Sigma-Aldrich (St. Louis, MO, USA). Biochemicals were obtained from the same manufacturer at the highest available purity, including 2,2′-azino-bis(3-ethylbenzothiazoline-6-sulfonic acid) (ABTS), 2′,7′-dichlorofluorescin diacetate (DCFA), 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyl tetrazolium bromide (MTT), 2,4dinitrophenylhydrazine (DNPH), glutamate, hydrogen peroxide, propidium iodide (PI), thiobarbituric acid (TBA), as well as the antioxidant chemicals ascorbic acid (vitamin C), butylated hydroxytoluene (BHT), ebselen, phenothiazine, αtocopherol (vitamin E, synthetic), and trolox. Edaravone was from abcam (Cambridge, UK), NXY-059 was from Tocris (Bristol, UK), and U74389G (methyl tirilazad) was from Biomol (Hamburg, Germany).

[2] 205w Compound 1phenothiazine was from Sigma-Aldrich. Compound 2 -2-cyanophenothiazine was from Bernhagen-Chemie (Bad Homburg, Germany). Compound 3 -2-aminomethyl phenothiazine was synthesized by reduction of 2-cyanophenothiazine 2. The educt was dissolved in methylene chloride (30 mM) and reduced with two volumes of 1 M lithium aluminum hydride (LiAlH 4 ). After 24 h at RT, 0.01 volumes of 1 M HCl and 0.02 volumes of water were added to quench the reaction. The resulting, light yellow suspension was thoroughly extracted with ether, washed with water, and dried. 1 H-NMR δ (400 MHz; (CD 3 ) 2 CO): 1.8-1.95 (2H, m), 4.25 (2H, s), 6.7-6.97 (7H, m), 7.73 (1H, s). MS (FD): 228.1. Compound 4 -2-butyrylamidomethyl phenothiazine was synthesized by acylation of 2-aminomethyl phenothiazine 3 with butyryl chloride. To this end, compound 3 (10 mM) was dissolved in dimethyl formamide containing 10 mM pyridine and cooled to 0°C. A stoichiometric amount of butyryl chloride was slowly added, before the reaction mixture was heated to 50°C and stirred for 2 h. After cooling to RT and stirring for an additional 24 h, the reaction was terminated with three volumes of water. The resulting mixture was cooled to 0°C, suction filtered, and the residue was recrystallized from methanol/water (9:1, v/v).

[3] 277w 1 H-NMR δ (400 MHz; (CD 3 ) 2 CO): 0.86 (3H, t), 1.55-1.64 (2H, m), 2.15 (2H, t), 4.19 (2H, d), 6.62-6.98 (7H, m), 7.31 (1H, s), 7.81 (1H, s). MS (FD): 298.2. Compounds 5, 6, 7 -2-octanoylamidomethyl phenothiazine, 2l a u r o y l a m i d o m e t h y l p h e n o t h i a z i n e , a n d 2palmitoylamidomethyl phenothiazine were synthesized as described above for compound 4, using corresponding stoichiometric amounts of octanoyl chloride, lauroyl chloride, and palmitoyl chloride, respectively. Compound 5: 1 H-NMR δ (400 MHz; (CD 3 ) 2 CO): 0.84 (3H, t), 1.24-1.27 (8H, m), 1.56-1.60 (2H, m), 2.16 (2H, t), 4.18 (2H, d), 6.63-6.98 (7H, m), 7.3 (1H, s), 7.8 (1H, s). MS (FD): 354.3. Compound 6: 1 H-NMR δ (400 MHz; (CD 3 ) 2 CO): 0.85 (3H, t), 1.25 (16H, s), 1.57 (2H, m), 2.16 (2H, t), 4.18 (2H, d), 6.62-6.97 (7H, m), 7.3 (1H, s), 7.8 (1H, s). MS (FD): 410.3. Compound 7: 1 H-NMR δ (400 MHz; (CD 3 ) 2 CO): 0.85 (3H, t), 1.36 (24H, s), 1.58 (2H, m), 2.14 (2H, t), 4.18 (2H, d), 6.62-6.97 (7H, m), 7.3 (1H, s), 7.79 (1H, s). MS (FD): 466.4. C o m p o u n d s 8 , 9 , 1 0b e n z o [ a ] p h e n o t h i a z i n e , benzo[b]phenothiazine, and benzo[c]phenothiazine were a kind gift from Dr. James V. Crivello (Rensselaer Polytechnic Institute, Troy, NY, USA). Their synthesis had been performed as described (35).

[4] 109w The purity of the newly synthesized compounds was assessed from their FD-MS and 1 H-NMR spectra. In general, FD-MS spectra did not contain any other peaks than the following: the molecular ion (M), its dimer (2 M), and their potassium adducts (M+K, 2M+2K). Educt peaks or conceivable side product peaks (e.g. doubly fatty acid modified phenothiazine) were absent in all cases; a small +16 peak (indicative of oxidation) of ~1% reconstructed ion current was seen with compounds 3 and 7. Similarly, all peaks in the 1 H-NMR spectra were assignable to the specified compound, residual water, or non-deuterated acetone (deuterated acetone was used as solvent for the NMR analyses).

[5] 105w Quantum chemical properties relevant for the antioxidant characterization (36) of the investigated compounds were calculated using the Chem3D software version 9.0 (CambridgeSoft Corporation, Cambridge, MA, USA). After geometrical optimization, the implemented MOPAC routine (AM1 basis set) was used to determine for each compound the difference (ΔH f ) between the heat of formation of the compound and its radical form, and the energy of the lowest unoccupied molecular orbital of this radical form (E(LUMOr)). Estimated logP values of the new phenothiazines were obtained by an incremental procedure based on literature values for molecular fragments. Here, the "best estimate" method implemented in ChemDraw 9.0 was employed.

[6] 101w The activity of the newly synthesized compounds to rapidly react with stable, preformed free radicals in vitro was investigated in terms of their "Trolox Equivalent Antioxidative Capacity" (TEAC), i.e. their capacity to quench the synthetic ABTS +• radical cation (2,2′-azino-bis(3-ethylbenzothiazoline-6-sulfonic acid). ABTS +• radical cations were generated as described (37). The decreased absorption of the ABTS +• solution at 734 nm was measured after 5 min incubation with the indicated compounds to be tested. After linear regression over four suitable concentrations, the ABTS +• -quenching activity of each compound was put into relation to the activity of the reference compound trolox.

[7] 74w Glutamate-sensitive clonal murine hippocampal neurons (HT22) were a kind gift from Dr. Pamela Maher (The Salk Institute, La Jolla, CA, USA) and were cultivated under 5% CO 2 in a humidified incubator at 37°C. Dulbecco's Modified Eagle Medium (DMEM) supplemented with 10% heatinactivated fetal calf serum (FCS), 1 mM pyruvate, 100 mg/l streptomycin and 100 U/ml penicillin was used as cultivation medium. All cell culture reagents and supplements were from Invitrogen (Carlsbad, CA, USA).

[8] 123w Cell viability was quantified in terms of the cells' capacity to reduce the chromogenic tetrazole MTT as described (36). HT22 cells were seeded onto 96-well plates at a density of 3 × 10 3 cells per well. After 24 h, the cells were incubated with various concentrations of the compounds to be tested, followed by the addition of 5 mM sodium glutamate. On the next day, MTT was added at a concentration of 0.5 mg/ml for 3-5 h. Thereafter, the cells were lysed by adding 0.1 ml/well of a solubilization solution (40% dimethylformamide (DMF), 10% sodium dodecyl sulfate (SDS), adjusted to pH 4.0 with acetic acid) overnight, and the amount of reduced MTT was analyzed at 560 nm with a multiwell plate reader.

[9] 82w Plasma membrane integrity towards a DNA-intercalating fluorescent dye, propidium iodide (PI), was used as a metabolism-independent marker of cell survival (36). To this end, HT22 cells were treated as in the MTT cell viability assay. After the glutamate challenge, the cells were loaded with 1 μg/ml PI for 10 min at 37°C. To quantify the effects of the investigated compounds, both the viable and the dead cells were manually counted by phase contrast and fluorescence microscopy (545 nm excitation/610 nm emission), respectively.

[10] 112w The intracellular oxidant tone in oxidatively challenged HT22 cells was assessed in terms of the oxidation of the fluorigenic substrate 2′,7′-dichlorofluorescin to 2′,7′dichlorofluorescein. HT22 cells were plated and cultured to confluency in 96-well plates and thereupon incubated with the indicated compounds in 200 μl DMEM without serum and phenol red for 3 h at 37°C. Subsequently, 5 μM 2′,7′dichlorofluorescin diacetate (DCFA) was added for an additional hour, before removing the surplus, extracellular DCFA by washing with fresh DMEM. The cells were then stressed with 500 μM hydrogen peroxide for 5 min. Changes in total fluorescence were measured at 37°C in a temperaturecontrolled multiwell plate reader at 485 nm excitation/ 535 nm emission.

[11] 87w Ongoing lipid peroxidation in cultivated neuronal cells was determined by quantifying their 8-isoprostane release. Confluent HT22 cells in 96-well plates were washed with fresh DMEM and incubated with the investigated compounds for 4 h at 37°C. Subsequently, 500 μM H 2 O 2 was added for 1 h, before 50 μl of the cell supernatant was sampled for 8isoprostane analysis by competitive enzyme immunoassay (EIA) (Cayman Chemicals, Ann Arbor, MI, USA), which was conducted as detailed in the manufacturer's instructions. The actual detection limit was ~5 pg/ml.

[12] 160w Protein carbonyl group accumulation was evaluated as a broad-spectrum index of protein oxidation. HT22 cells seeded in 6-well plates were preincubated with the indicated compounds for 3 h, before 500 μM H 2 O 2 was added for an additional 3 h. The cells were washed with PBS and harvested in lysis buffer (2% SDS, 10% sucrose, 62.5 mM Tris-HCl, pH 7.4), before determining the protein content of the generated samples with a copper/bicinchoninic acid (BCA) kit (Thermo Scientific, Waltham, MA, USA). The cell lysates were mixed with 0.5 volumes of derivatization solution (10 mM 2,4-dinitrophenylhydrazine (DNPH) in 2 M HCl), incubated for 20 min at RT, and quenched with the same volume of 3 M Tris base. Denaturing SDS-polyacrylamide gel electrophoresis (PAGE) using isocratic 10% gels and the subsequent Western blotting routine for the immunological detection of the introduced dinitrophenyl moieties were performed using standard protocols as described (34). The anti-DNP antibody (rabbit polyclonal, dilution 1:1000) was from Invitrogen.

[13] 33w Data were expressed as mean ± standard deviation of the number of experiments indicated in the figure legends. Significant differences between treatment groups were analyzed using Student's t-test or, where indicated, two-way ANOVA.

[14] 73w To investigate the apparent discrepancy between the results in cultured cells and isolated brain lipids, the site of antioxidative activity in toxin-treated HT22 cells was probed. To this end, two indices were monitored: intracellular accumulation of oxidizing species as reflected by the aqueous DCFA assay, and the formation of 8-isoprostane from arachidonic acid to quantify lipid peroxidation (44). Hydrogen peroxide was used as oxidant in these assays to ensure that the readout would

[15] 321w concentration [nM] 0,1 1 10 100 1000 10000 ] % [ n o i t a d i x o r e p d i p i l 0 20 40 60 80 100 120 4 5 6 7 concentration [nM] 0,1 1 10 100 1000 10000 ] % [ n o i t a d i x o r e p d i p i l 0 20 40 60 80 100 120 1 2 3 a b concentration [nM] 0,1 1 10 100 1000 10000 ] % [ n o i t a d i x o r e p d i p i l 0 20 40 60 80 100 120 1 2 3 concentration [nM] 0,1 1 10 100 1000 10000 ] % [ n o i t a d i x o r e p d i p i l 0 20 40 60 80 100 120 4 5 6 7 c d Fig. 2 Lipid peroxidation in vitro. (a, b) The formation of TBARS in a crude rat brain lipid preparation, and (c, d) in a purified lipid preparation depleted of vascular material was monitored, following a prooxidative challenge (10 μM Fe 2+ /200 μM ascorbate, 30 min) with or without the addition of compounds 1-7 at different concentrations (1 h preincubation; duplicate determinations). Unsubstituted phenothiazine 1 and derivatives with short-chain substituents 2-5 tended to be more effective than the more lipophilic C 12 -and C 16 -substituted molecules 6 and 7. Table I TEAC (Trolox Equivalent Antioxidative Capacity) Values of Compounds 1-7 in Two Different Solvents Ethanolic Aqueous trolox 1.00 ± 0 1.00 ± 0 1 2.05 ± 0.05 1.10 ± 0.01 2 1.70 ± 0.30 1.07 ± 0 3 1.32 ± 0.12 0.83 ± 0.04 4 0.70 ± 0.08 0.35 ± 0 5 0.72 ± 0.03 0.38 ± 0.02 6 1.62 ± 0.09 0.39 ± 0.02 7 0.87 ± 0 0.35 ± 0.03

[16] 175w The TEAC assay reflects the direct reactivity of a compound with a preformed, stable free radical (ABTS radical cation); values > 1 represent higher reactivity than the reference compound trolox (data from n = 2, each determined from 4 concentrations) quantify only direct, redox-related effects of the tested compounds rather than any potential interference with signaling events involved in oxidative glutamate toxicity. The small, moderately lipophilic phenothiazines 1-4 were found to be potent suppressors of intracellular oxidant accumulation (Fig. 4). However, activity was lost with increasing lipophilicity (Fig. 4b), as expected for a method quantifying aqueous oxidant accumulation. In fact, the most potent cytoprotective agent 6 turned out to be basically devoid of any effect in this assay, ruling out the possibility that aqueous peroxide suppression was relevant for its cytoprotective effect. In contrast, compound 6 was approximately equally effective in the suppression of cellular lipid peroxidation as compound 1, unsubstituted phenothiazine (Fig. 4c), which did demonstrate activity in the aqueous DCFA assay (Fig. 4a superior activity in the brain lipid peroxidation assay (Fig. 2a).

[17] 36w These findings indicate that inhibition of cellular lipid peroxidation (or a closely related redox process running in parallel) is indeed the mechanism from which to explain the cytoprotective efficacy of the C 12 -substituted compound 6.

[18] 21w By analogy, it is plausible to assume that also the other phenothiazines' cytoprotective effect is finally attributable to the same mechanism.

[1] 24w Conclusions Lipophilicity, radical stabilization and molecular weight appear to form an uneasy triangle, in which a slightly faulty selection may readily abolish neuroprotective activity.

[2] 7w KEY WORDS neuroprotection . oxidative stress .

[3] 190w The formation of thiobarbituric acid-reactive substances (TBARS) in isolated, native lipid bilayers was analyzed to quantify the chain-breaking antioxidant activity of the novel compounds. Cerebrocortical brain membranes were prepared from adult male Sprague-Dawley rats and stored as described (36). For lipid peroxidation experiments, the membrane preparation was diluted with PBS to reach a concentration of 0.5 mg/ml protein. After 1 h preincubation with the compounds to be tested, peroxidation was initiated by the addition of 10 μM Fe 2+ SO 4 2-and 200 μM sodium ascorbate at RT. After 30 min, the reaction was stopped by adding 3 volumes of 5% trichloroacetic acid in 1 M HCl, and 1 volume of 0.5% thiobarbituric acid in 5 mM NaOH. The samples were heated to 95°C for 30 min, followed by cooling and centrifugation at 10,000g for 10 min to precipitate membrane proteins and other acid-insoluble material. The fluorimetric measurement of the TBA adduct-containing supernatant was performed at 531 nm excitation/572 nm emission in a multiwell plate reader. Under these conditions, more than 95% of the fluorescence was attributable to a single HPLC peak coeluting with an authentic malondialdehyde-thiobarbituric acid adduct peak.

[4] 148w In a second experiment, a purified cerebrocortical lipid preparation depleted of vascular material was investigated. To this end, a high-dextran centrifugation step (38) was inserted into the above protocol (36). Minced cortical tissue was homogenized on ice in 5 volumes of DMEM with a Potter-Elvehjem apparatus at 500 rpm for 1 min. After centrifugation at 3000g for 5 min, the pellet was homogenized in the same manner in 10 volumes of DMEM adjusted to 13% Dextran T70 (Roth, Karlsruhe, Germany) and centrifuged at 5400g for 15 min, which separated two lipid-rich fractions. The floating fraction was transferred to 10 volumes of nonreducing "lipid buffer" (20 mM Tris-HCl, pH 7.4, 1 mM MgCl 2 , 5 mM KCl) and sonicated, before continuing with the above protocol as detailed in (36). All fluids were thoroughly flushed with nitrogen before use; all preparative steps were performed on ice or at 4°C.

[5] 168w Late-stage autolysis of cellular proteins was evaluated as marker for preceding oxidative protein damage. HT22 neurons at 70% confluency were treated with the indicated compounds and 5 mM glutamate for 18 h. After washing with PBS, the cells were harvested in 200 μl homogenization buffer (5 mM Tris-HCl, pH 7.4, 1 mM EDTA, 1 mM DTT, 10 μM phenothiazine) containing 1% protease inhibitor-cocktail (P2714, Sigma-Aldrich). After sonication (3 × 5 s), the lysates were centrifuged at 100,000g for 90 min at 4°C. The membrane pellets were resuspended in 50 μl homogenization buffer, followed by protein determination by the BCA method, SDS-PAGE, and transfer to nitrocellulose membranes (34). For global analysis of protein autolysis, the blots were stained with Ponceau S (0.2% in 5% acetic acid). For immunological detection of specific proteins, the blots were washed with TBST (Tris-buffered saline containing 0.05% TWEEN 20) and processed as described (34). The primary antibody against the α 1 subunit of Na/K-ATPase (mouse monoclonal, dilution 1:3000) was from Acris Antibodies (Herford, Germany).

[6] 196w From the above considerations, it cannot be inferred that an inhibition of the formation of 8-isoprostane was the direct mediator of the cytoprotective effect of compound 6. After all, it is well possible that 8-isoprostane merely constitutes a proxy for some other, parallel redox process that would finally explain the efficacy and maybe even the partial efficacy advantage of 6. Hence, we have investigated a few candidate targets that might be particularly well-protected by this compound, notably proteins. Cells that had been stressed with 500 μM H 2 O 2 for 3 h as in the preceding experiment were harvested and analyzed for protein carbonyl content, a global marker of protein oxidation (45). Quantitative analysis of four independent Western blotting experiments yielded an approximately twofold induction of protein carbonyls due to peroxide treatment (Fig. 5). Of the three investigated compounds 1, 3 and 6 with their different solubility characteristics, only compound 6 was capable to significantly suppress carbonyl formation at 100 nM concentration, potentially explaining its relative cytoprotective superiority with respect to plain phenothiazine (1). At 1000 nM concentration, though, all compounds were roughly the same effective and approximately halved the peroxide-induced increase in protein carbonyls.

[7] 252w ] % [ s l y n o b r a c n i e t o r p 0 50 100 150 200 250 a b 1 3 6 # # 100 nM 1000 nM 100 nM 1000 nM 100 nM 1000 nM untr Anti-DNP Ponceau 1 3 6 untr untr H 2 O 2 H 2 O 2 100 nM 1000 nM 100 nM 1000 nM 100 nM 1000 nM H 2 O 2 170 130 100 70 55 45 35 25 kDa 170 130 100 70 55 45 35 25 kDa Acute intracellular reactive oxygen species accumulation in response to hydrogen peroxide application (500 μM, 5 min) was measured in DCFAloaded cells (5 μM, 1 h) that had been pretreated with compounds 1-7 for 3 h (quadruplicate determinations). The smaller and more hydrophilic molecules 1-4 were more efficacious in the quenching of cytosolic peroxide reactivity than the long-chain substituted compounds 6 and 7. (c) Measurement of 8-isoprostane formation under the same conditions as above (triplicate determinations). Phenothiazine 1 and the C 12 -substituted phenothiazine 6 were about equally effective and had a dose-dependent effect on 8-isoprostane formation (**p < 0.001). In view of the diversity of chemical reaction mechanisms that can lead to protein carbonylation under oxidative stress (45), we would deem that even an incomplete attenuation of carbonylation (by half) constitutes a biochemically significant effect. After all, it is hard to imagine that a single antioxidant may ever inhibit all of those chemical reaction mechanisms simultaneously and fully.

[8] 301w To obtain a critical, quantitative assessment of the improvement realized by the novel substances, we compared the potency of compounds 6 and 10 with the effects of ascorbate, which is of pivotal importance for redox regulation in the brain (47,48), and with different phenolic antioxidants, namely α-tocopherol (49), trolox (49), and butylated hydroxytoluene (BHT), which is frequently used as preservative in food or cosmetics (50). Included into the comparison were the four widely cited structures already mentioned in the introduction (NXY-059, (methyl) tirilazad, ebselen, and edaravone); the latter compounds have all been promoted into clinical testing in humans, but have ultimately failed to elicit significant neuroprotective effects (5)(6)(7)(8)(9). The results indicate that phenothiazines, as a group, indeed appear to possess superior neuroprotective properties, supporting their choice as lead structures in the current study. Among phenothiazine's competitors, the natural chain-breaking antioxidant α-tocopherol came closest in potency, exhibiting an only fivefold lower cytoprotective activity than phenothiazine (EC 50 value ~150 nM versus ~30 nM for phenothiazine). However, this potency gap many be larger in vivo, due to permeability handicaps of extremely lipophilic compounds like tocopherol (49,51). The neuroprotective capacity of the small, antiatherogenic phenol BHT (52) was about an order of magnitude lower than that of tocopherol (EC 50 value ~2 μM), followed by the relatively inefficient substance ascorbate, with an EC 50 value of ~15 μM. Compared to its natural concentration in the brain of ~1 mM (48), this may yet be viewed to be a rather low dose to produce neuroprotection. Surprisingly, none of the clinically investigated drugs achieved a reasonably low, i.e. nanomolar EC 50 value, and the spin trap NXY-059, which has earned a great deal of investment (5,6), was the only substance that did not show any neuroprotective effect up to the highest tested concentration of 1 mM.

[9] 36w S H N S H N S H N 8 9 10 M W [Da] logP ∆H f [kcal/mol] E(LUMOr) [eV] 9 249 4.34 24.10 -0.529 10 249 4.34 23.39 -0.735 8 249 4.34 12.11 -0.538