Isoform-selective NADPH oxidase inhibitor panel for pharmacological target validation

Unphysiological reactive oxygen species (ROS) formation is considered an important pathomechanism for several disease phenotypes with high unmet medical need. Therapeutically, antioxidants have failed multiple times. Instead, focusing on only disease-relevant, enzymatic sources of ROS appears to be a more promising and highly validated approach. Here the family of five NADPH oxidases (NOX) stands out as drug targets. Validation has been restricted, however, mainly to genetically modified rodents and is lacking in other species including human. It is thus unclear whether the different NOX isoforms are sufficiently distinct to allow selective pharmacological modulation. Here we show for five of the most advanced NOX inhibitors that indeed isoform selectivity can be achieved. NOX1 was most potently (IC50) targeted by ML171 (0.1 μM); NOX2, by VAS2870 (0.7 μM); NOX4, by M13 (0.01 μM) and NOX5, by ML090 (0.01 μM). Conditions need to be carefully controlled though as previously unrecognized non-specific antioxidant and assay artefacts may limit the interpretation of data and this included, surprisingly, one of the most advanced NOX inhibitors, GKT136901. As proof-of-principle that now also pharmacological and non-rodent target validation of different NOX isoforms is possible, we used a human blood-brain barrier model and NOX inhibitor panel at IC50 concentrations. The protective efficacy pattern of this panel confirmed the predominant role of NOX4 in stroke from previous genetic models. Our findings strongly encourage further lead optimization efforts for isoform-selective NOX inhibitors and clinical development and provide an experimental alternative when genetic validation of a NOX isoform is not an option. Graphical abstract


36
Reactive oxygen species (ROS) are considered an important pathomechanism for several 37 common diseases with high unmet medical need including cardiovascular and neurodegenerative as 38 well as cancer. Yet, clinical translation of this hypothesis into therapeutic application by use of 39 antioxidants that scavenge ROS has consistently failed [1], in some cases even increasing mortality 40 systemic manner, but rather by identifying for each disease condition the relevant enzymatic ROS 48 source and inhibit this source by selective pharmacological inhibitors [9]. 49 Several enzymatic sources of ROS exist [10,11]. NADPH oxidases (NOX), however, stand 50 out as they are the only known dedicated, evolutionary conserved ROS-forming enzyme family. All 51 other enzymes produce ROS as a byproduct or due to being damaged. Disease relevance has been 52 suggested for NOX1, contributing to diabetic atherosclerosis [ [24]. Target validation in these 56 cases, however, has been done primarily by gene knock-out (as reviewed in [25]) and, in the case of 57 NOX5 [21], by knock-in technology limiting most data sets to mice and, in the case of NOX4, in 58 addition to rat [15]. 59 ideally isoform-selective NOX inhibitors are desirable. It is unclear, however, whether this is or will 61 be achievable given the fact that all five relevant NOX isoforms (NOX1, NOX2, NOX3, NOX4 and 62 NOX5) are similarly structured transmembrane proteins containing highly conserved catalytic 63 density needed for the experiment (5 x 10 5 cells/well in 96 well plate). 104

RNA Extraction and cDNA Synthesis 105
RNA was extracted using RNeasy ® Micro Kit (Qiagen) according to the manufacturer's 106 protocol. cDNA was synthesized from 1 µg total RNA in 20 µl reactions using High Capacity 107 cDNA Reverse Transcription Kit (Thermo Fisher Scientific). After synthesis, the cDNA was stored 108 at −20°C.  To measure NOX2 activity, superoxide production in HL-60 cells was measured by 144 cytochrome C reduction as described previously [32]. Briefly, cytochrome C (100 µM) was added 145 to the DMSO-differentiated HL-60 cell suspension. After adding 5 x 10 5 cells to a 96-well plate and 146 addition of inhibitors, a basal reading was performed at 540 nm (isosbestic point of cytochrome C) 147 and 550 nm (SpectraMax 340; Molecular Devices, Sunnyvale, CA, USA). Subsequently, the 148 oxidative burst was initiated by the addition of 100 nM PMA. After incubation for 60 min. at 37°C 149 the absorbance was measured. Signals were normalized to the basal readings. In addition, 150 superoxide dismutase was added as control before PMA stimulation in selected wells. 151 Xanthine (X; final concentration: 50 µm) and cytochrome C (final concentration: 100 µm) 153 were dissolved in HBSS, and 100 µL aliquots of this solution were transferred to individual wells of 154 a 96-well plate. After addition of the oxidase inhibitors, the mixtures were allowed to equilibrate for 155 20 min. The reaction was started by the addition of 100 µL xanthine oxidase (XO) (final 156 concentration: 5 mU·mL-1), and absorbance at 540 and 550 nm was recorded 10 min. after the start 157 of the reaction. Superoxide production was calculated by normalization to the signals obtained at 158 540 nm. Similar experiments were performed with 0.4 mM luminol instead of cytochrome C. After 159 20 min. equilibration, the reaction was started with XOD (1 mU·mL-1), and chemiluminescence 160 was subsequently recorded for 20 min. in a Fluoroscan FL microplate reader. Signals were 161 calculated as AUC and normalized to the X/XOD-derived control signal. Accordingly, a counter 162 screen has been performed with cytochrome C or luminol without X/XO-generated superoxide. 163

H 2 O 2 measurement in cell-free system 164
The fluorescence of Amplex Red was measured in presence and absence of H 2 O 2 (0.25 µM), 165 reflecting NOX4 and NOX5 output. After addition of inhibitors and 50 µl Amplex Red reaction 166 mixture containing 100 U·mL-1 HRP and 10 mM Amplex Red to each well, the plate was incubated 167 at 37°C for 10 min. Thereafter, the plate was read at excitation (530-560 nm) and emission ~ 590 168 nm at 37ºC and for 60 min. Data was calculated as the AUC over 60 min. and data were normalized 169 to H 2 O 2 output in absence of inhibitor. 170

Statistical analysis 184
Data analysis was performed using Prism 6.0g software package (GraphPad Software, San 185 Diego, USA). IC 50 values were calculated with a non-linear regression analysis using an algorithm 186 for sigmoidal dose-response with variable slopes. Results are expressed as means ± SEM. 187 Statistical differences between means were analyzed by one-way ANOVA followed by Bonferroni 188 correction for multiple comparisons. A value of P < 0.05 was considered statistically significant. 189

Gene expression of NOX isoforms 191
To check the expression of the different NOX isoforms, we analysed NOX1, 2, 4 and 5 192 mRNA expression in control vector-transfected HEK293 cells, HEK293 cells transfected with 193 NOX1, 2, 4 and 5 and HL-60 cells. NOX isoforms mRNA expression was measured by  and mRNA levels were normalized to the expression of β -actin. 195 As reported previously [26, 32], NOX1 was expressed in NOX1-transfected HEK293 cells 196

NOX pharmacological isoform selectivity 202
To characterize the NOX isoform selectivity of the current second-generation NOX 203 inhibitors we determined concentration-dependency and efficacy of GKT136901, Ml171, 204 VAS2870, M13 and ML090 on NOX1, 2, 4, and 5 ( Fig. 2A). Our preference was for cell (lines) 205 expressing a specific NOX isoform in a highly selective manner and ideally physiologically. The 206 latter is only known for NOX2 and HL-60 cells [33,34]. For NOX1, NOX4 and NOX5, there is, to 207 our knowledge, no such native human cell line. Therefore, for these three isoforms we used 208 transiently transfected HEK-293 cells, using previously validated methodology [27]. 209 Specifically, NOX2 expressing HL-60 cells or HEK293 cells transfected with NOX1, 210 NOX4 or NOX5 were incubated in presence of increasing concentrations of each compound. ROS 211 generation was subsequently induced by PMA to stimulate NOX1, NOX2 and NOX5, which was 212 assayed in presence of ionomycin. Thereafter, ROS production was assayed using a panel of 213 cellular assays with structurally unrelated probes; i.e. Amplex red, luminol or cytochrome C ( Figure  214 chemiluminescence and cytochrome C-reduction were used to quantify superoxide generation. As 216 controls, medium, vector-transfected HEK293 cells and non-stimulated HL-60 cells were used ( Fig comparable IC 50 values but enhanced E max for NOX5 inhibition. These data suggest that NOX 227 inhibitors indeed display differential isoform targeting, albeit the single compound selectivity is yet 228 not sufficient. Therefore, combined analysis of a NOX inhibitor panel with sufficient differential 229 potencies and selectivities is suggested. 230

Non-specific anti-oxidant and assay artefacts 231
To screen for possible assay interference between GKT136901, ML171, VAS2870, M13 or 232 ML090 and the NOX1 assay, a cell-free system was performed in which each inhibitor was 233 screened in presence of luminol, 1mU/ml xanthine oxidase (XO) and 1mg xanthine (X) generating 234 superoxide. At concentrations 1µM GKT136901 did not show interference with either the 235 molecular probe or with the X/XO system (Fig. 4A). Similar findings were obtained with 0.1 µM 236 ML171, 10 µM VAS2870 and 30 nM ML090 (Fig. 4A). In contrast, 1 µM M13 enhanced 237 chemiluminescence (Fig. 4A). To identify direct interactions between assay components and 238 GKT136901, ML171, VAS2870, M13 or ML090 a cell-free counter screen was performed without 239 suggesting direct interference with luminol-based chemiluminescence (Fig. 4B). In contrast, in 241 presence of ML090 the signal was enhanced (Fig. 4B). 242 To study non-specific antioxidant effects of VAS2870, the effect of 10 µM VAS2870 was 243 studied in a cell-free system in presence of cytochrome C, and X/XO-derived ROS. In these assays, 244 VAS2870 showed significant antioxidant effects (Fig. 4C). 245 To study whether the effects of GKT136901, ML171, VAS2870, M13 and ML090 on 246 NOX4 and NOX5 are specific, a cell-free assay was performed with NOX4 or NOX5 effective 247 concentrations of each inhibitor in presence of 0.25µM H 2 O 2 in an Amplex Red assay. ML171, 248 VAS2870, M13 and ML090 did not directly affect Amplex Red-based detection of H 2 O 2 (Fig. 4D). 249 However, GKT136901 reduced Amplex Red-based H 2 O 2 signals suggesting assay interference. 250 Hence, the assay was repeated without H 2 O 2 to study potential direct assay component interference. 251 Indeed, GKT136901, ML171, M13 and ML090 reduced the signal as compared to the control (Fig.  252 4E) whereas in presence of VAS2870 the signal was enhanced (Fig. 4E). 253

Target validation of NOX in ischemia-induced hyperpermeability model 254
As predicted from the inhibitor screen, M13, GKT136901 or ML171 protect against 255 ischemia-induced hyperpermeability in HBMEC. Our data suggested that a NOX inhibitor panel 256 may be used for target validation of selective NOX isoforms. We thus tested our NOX inhibitor 257 panel in an in vitro human model of ischemia-induced hyperpermeability (Fig. 5A) where NOX4 is 258 involved in subacute hypoxia-induced increases in cell permeability whereas NOX1, NOX2 do not 259 [31] and NOX5 only acutely [20]. Primary HBMEC cultures were subjected to 6h of hypoxia 260 followed by 24h of re-oxygenation (Fig. 5B)  VAS2870 or ML090 treatment showed no effect (Fig. 5C). These data provide a proof-of-concept 267 for pharmacological target validation using NOX inhibitor panels. 268  Fig. 4C); the same concentration, however, resulted in a 91.5% decrease of the 286 NOX2-dependent signal (Fig. 3C). 287 M13, a previously unreported compound from Glucox Biotech, was identified as a first-in-288 class relative NOX4 selective inhibitor as it was 200-times more potent in inhibiting NOX4 (IC 50 ~ 289 0.01 µM) versus NOX1 (IC 50 ~ 0.2 µM) and showed negligible NOX2 inhibition and almost no 290 effect on NOX5. Although M13 interfered with the Amplex Red assay by 9% (Fig. 4E), this was 291 considerably less than the 40% reduction of the NOX4-dependent signal (Fig. 3D). 292 ML090 was previously described as NOX1 selective [43]. In our hands, IC 50 values for 293 NOX1, 4 and 5 were quite similar suggesting ML090 to be a rather pan NOX inhibitor. Two 294 interferences were observed. The 36.8% inhibition of the NOX1-dependent signal (Fig. 3D) may 295 represent a slight underestimation because ML090 per se surprisingly increased the luminol signal 296 by 13.4% (Fig. 4B). ML090 also interfered with the Amplex Red signal, yet lowering it by 11.9% 297 (Fig. 4E); the NOX5 dependent signal, however, was reduced by 46.8 % (Fig. 3E). Therapeutically, 298 ML090 protects from vascular dysfunction in a rabbit model, an effect that has been attributed to 299 NOX1 inhibition [44]. According to our here presented data, however, an effect involving NOX4 300 and, in particular, NOX5 cannot be excluded as rabbits, unlike mice and rats, do express this 301

isoform. 302
Compared to all compounds, GKT136901, a chemical analogue and pharmacological sibling 303 of the clinically most advanced NOX inhibitor, GKT831, and claimed to be NOX1 and NOX4 304 specific, stood out in a problematic manner. GKT136901interfered both with the Amplex Red 305 assay, for NOX4 and 5, and the luminol assay, for NOX1, and, on top of this, directly scavenged 306 ROS. This latter antioxidant effect has been observed also by others using an alternative H 2 O 2 assay 307 [45] or by scavenging peroxynitrite [28]. These characteristics thus appear to limit interpretation of 308 results using GKT136901 as a NOX4/5 inhibitor and may cast doubt on the specificity with respect 309 In summary, all tested NOX inhibitor compounds displayed different isoform selectivity 324 profiles suggesting differential chemical targeting and a proof-of-principle that NOX isoform 325 selective small molecule inhibition will become possible. This should stimulate focused library 326 synthesis and structure-activity programs for further lead optimization to obtain truly isoform