The objectives of the studies reported here were several: firstly to characterize the pharmacology of BIA 10‐2474 using PF‐04457845 and JNJ‐42165279 as comparators, two fatty acid amide hydrolase inhibitors previously tested in humans and considered safe, and then to explore the selectivity of BIA 10‐2474 to identify potential off‐targets, both in vitro and in vivo. Our study concentrated on BIA 10‐2474 itself, as it represents the major species, >90% by AUC, found in plasma following its oral administration (Loureiro, Lopes, Bonifácio, Moser, & Soares‐da‐Silva, 2017) and secondly because the TSSC in its final report (TSSC, 2016) considered that the metabolites of BIA 10‐2474 were unlikely to play a role in the observed toxicity, because of their low circulating concentrations. This is reinforced by the present demonstration that brain levels of metabolites were substantially below those of the parent molecule. Furthermore, that the metabolite levels did not change between areas nor between acute and repeated administration. Finally, a study was designed to explore the potential effects of BIA 10‐2474 in an animal model known to be prone to have a high incidence of cerebral vascular stroke (Yang, Kimura‐Ohba, Thompson, & Rosenberg, 2016).

The present data characterizes BIA 10‐2474 as a time‐dependent mechanism‐based inhibitor of FAAH. In rat brain membranes, BIA 10‐2474 potency increased with pre‐incubation time, as expected for a time‐dependent inhibitor, similarly to that of PF‐04457845 and JNJ‐42165279. The IC₅₀ values determined at each pre‐incubation time (from 360 to 1134 nM) were about two orders of magnitude higher than those of PF‐04457845 and 2–4‐fold higher than those for JNJ‐42165279. However, when using an intact cell preparation expressing human fatty acid amide hydrolase, the IC₅₀ values determined for BIA 10‐2474 varied between 18 and 74 nM. This large variation in fatty acid amide hydrolase potency between membrane and whole‐cell preparations was not observed with either PF‐04457845 or JNJ‐42165279. Additionally, after oral administration the potency of BIA 10‐2474 to inhibit mouse brain or liver fatty acid amide hydrolase was similar to that of PF‐04457845. This higher potency in vivo has also been observed by other authors (Tong et al., 2017). The simplest explanation, as suggested by other authors (van Esbroeck et al., 2017), is that BIA 10‐2474 accumulates inside cells (Mateus, Matsson, & Artursson, 2013), but none of these compounds showed any evidence of such accumulation. Another possibility is that the effect was due to an active early metabolite instead of the parent BIA 10‐2474. However, metabolic studies performed in vitro with hepatocytes and in vivo in the rat with the radiolabelled BIA 10‐2474 (Loureiro, Fernandes‐Lopes, Moser, & Soares‐da‐Silva, 2018) showed no evidence for an early metabolite. Additionally, the only active metabolite identified had a similar in vitro potency to that of BIA 10‐2474. A differential interaction with the enzyme in its natural environment (in the membranes in vivo) or in the in vitro preparations could also be put forward. Although we do not have an explanation for this difference, the potency of BIA 10‐2474 determined in situ is more aligned with the concentrations that are active in vivo. Thus, at 0.1 mg·kg⁻¹ p.o., a dose‐producing near‐maximal inhibition of fatty acid amide hydrolase, we obtained maximum levels in plasma and brain of 33.0 ng·ml⁻¹ and 5.3 ng·g⁻¹ (equivalent to 109 and 18 nM), respectively.

In a broad screen of 221 targets, BIA 10‐2474 had no significant activity at up to 50 μM. Additional testing at a further eight targets at 10 μM similarly revealed no significant activity. In contrast, PF‐04457845 showed significant activity at several receptors, transporters and ion channels, similar to data published by Johnson et al. (2011). In addition, the present study also shows that BIA 10‐2474 has relatively little impact on CYPs and transporters up to concentrations of at least 30 μM, apart from inhibition of MATE2‐K with an IC₅₀ of 3.4 μM.

We identified an effect of BIA 10‐2474 against serine hydrolases both in vitro and ex vivo. The results obtained after chronic treatment in the ex vivo activity‐based protein profiling study confirmed most of the off‐targets previously identified by van Esbroeck et al. (2017), notably ABHD6 at exposure levels only slightly greater than those required for fatty acid amide hydrolase inhibition, and at higher exposures ABHD11 and PNPLA6. Although not detecting specifically activity against the rat orthologue of human CES2, we did identify activity against three members of the same https://www.uniprot.org/uniprot/?query=family:%22type-B+carboxylesterase%2Flipase+family%22 (CES 1C and two carboxyl ester hydrolases). We did not detect an effect against hormone‐sensitive lipase nor, as expected, fatty acid amide hydrolase2. However, we identified additional off‐target activity against the lipid metabolism‐related enzymes PLA2G15, lysosomal thioesterase PPT2, calcium‐independent PLA₂ (PLA2G6) and ABHD10. Significant inhibition was also seen on the androgen‐induced protein 1 and signal peptidase complex catalytic subunit SEC11A. Possible reasons for the differences in off‐target identification include the interactions being specific to rodent enzymes (present study as opposed to human cells, van Esbroeck et al., 2017), and/or due to a metabolite or requiring very long exposure as they were primarily seen at 100 mg·kg⁻¹ after 28 days. The differences between our in vitro and ex vivo activity‐based protein profiling data suggest that BIA 10‐2474 is less potent against all serine hydrolase enzymes in vitro, not just fatty acid amide hydrolase and ABHD6. Although we have not measured brain levels of BIA 10‐2474 at the doses used in our ex vivo study, data from GLP toxicology studies (Harris et al., 2019; Hayes et al., 2019) at comparable doses (10, 30, and 90 mg·kg⁻¹ p.o.) and for comparable durations (28 and 30 days of treatment) provide plasma C _max values of approximately 11, 47, and 101 μM at 10, 30 and 90 mg·kg⁻¹ p.o. (irrespective of treatment duration), suggesting maximum brain concentrations around 20 μM at the 100 mg·kg⁻¹ of dose used here (assuming the same ratio of 1:5 between brain and plasma as we see here for 0.1 and 0.3 mg·kg⁻¹ in mice).

Lipid metabolism was altered in rat brain tissues in animals dosed with 100 mg·kg⁻¹·day⁻¹ as compared with the control group. All identified PEs and PCs containing AA were significantly increased in the BIA 10‐2474‐treated group reflecting accumulation of lipid species that are precursors of AEA, which, although not quantified in the lipid profiling study, is increased as a result of fatty acid amide hydrolase inhibition (as shown in mouse studies). Interestingly, phospholipids not only with AA but also with very‐long‐chain fatty acids were increased, representing an impact on downstream products of these lipid species. Our data are in part aligned with that previously reported (van Esbroeck et al., 2017), with increased long‐chain PC species and DAG. Further comparisons cannot be drawn since in our study, small lipids, such as fatty acids, N‐acetyl amines and small lysophospholipids, were not quantified due to method limitations. Although several of the serine hydrolases identified as BIA 10‐2474 off‐targets are involved in lipid metabolism, they metabolize classes of lipids, and a direct correlation between the lipid alterations observed and the activities of those enzymes cannot be made at present.

Some of these enzymes are associated with neurological syndromes in humans. Loss‐of‐function mutations in PLA2G6 (Arber, Li, Houlden, & Wray, 2016; Illingworth et al., 2014) and PNPLA6 (Synofzik et al., 2014; Topaloglu et al., 2014) have been identified as causing neurodevelopmental disorders and PPT2 knockout mice display a neurodegenerative phenotype, although PPT2 deficiency has not been described in humans (Gupta et al., 2003). Could any of the off‐targets identified here be responsible for the toxicity seen in the clinic with BIA 10‐2474? At the maximum plasma exposures seen in the clinic (around 2 μM at 50 mg and 6 μM at 100 mg, Rocha et al., 2017), it is almost certain that ABHD6 and CES would be inhibited. As far as we can ascertain, there are no reports of neurological consequences of either gene knockout or pharmacological inhibition of these targets. The other off‐targets mentioned above will probably have lower levels of inhibition at the measured clinical exposure, but we cannot determine at this stage to what extent any inhibition might be physiologically meaningful. It should be noted however that inhibitors of PNPLA6 can induce neurotoxicity in rodents if inhibition is greater than 80% (Wu & Casida, 1996) and animal models of the different point mutations and/or knockout mice of these targets reproduce the associated clinical symptoms (Long & Cravatt, 2011; Winrow et al., 2003).

However, the available data make it unlikely that the identified off‐targets were significantly occupied at the clinical doses, since other elements besides exposure to the inhibitor are involved such as the relative amounts of enzymes present in relation to both the inhibitor and substrate concentrations and the rate of recovery of enzyme activity. Nevertheless, the exposure levels of BIA 10‐2474 used in regulatory toxicology studies would have been sufficient to engage all of the off‐targets we have identified. However, neither the neurological syndromes nor the histopathological hallmarks of these human syndromes were seen in these toxicology studies. There was some mild axonal swelling in the medulla oblongata in the rat toxicology study at 90 mg·kg⁻¹ after 13 weeks of treatment (in 3/10 male rats and 7/10 female rats) and vacuolation in the same region in about half the male rats at the same dose after 26 weeks of treatment. However, these signs were not observed in the 28‐day toxicology study (manuscript under editorial review), consistent with the absence of CNS histopathology in the present study.

There have been suggestions of other potential off‐targets such histone deacetylases, calcitoninhttps://www.guidetopharmacology.org/GRAC/ObjectDisplayForward?objectId=48 and https://www.guidetopharmacology.org/GRAC/ObjectDisplayForward?objectId=2363 and https://www.guidetopharmacology.org/GRAC/ObjectDisplayForward?objectId=2359 and https://www.guidetopharmacology.org/GRAC/ObjectDisplayForward?objectId=590r (Molinski et al., 2017). However, the present data show that BIA 10‐2474 does not have affinity for any of these targets. Recently, Huang et al. (2019), using alkynylated analogues of BIA 10‐2474 and metabolites as activity‐based protein profiling probes showed that the two desmethyl metabolites were able to covalently bind catalytic cysteines of aldehyde dehydrogenases in human cells, namely, https://www.guidetopharmacology.org/GRAC/ObjectDisplayForward?objectId=2595, ALDH1B1 and https://www.guidetopharmacology.org/GRAC/FamilyDisplayForward?familyId=764#2467. These enzymes were not identified in the present study, although the data are not comparable, as the activity‐based protein profiling probes used in this study were designed to target specifically serine hydrolases.

Kerbrat et al. (2016) described several types of cerebral lesion in the affected subjects taking part in the clinical trial with BIA 10‐2474 as detected by MRI scans. They included lesions, microhaemorrhage and oedema, which were severe in the subject that died and mild to severe in two affected survivors. The SHRSP were established from a substrain of the spontaneously hypertensive rat (SHR) and shows a high incidence of cerebral stroke. We consider the SHRSP model that we have used here particularly relevant to the abnormalities described as the use of relatively young rats (60 days old) limits the role for hypertension but still results in cerebral haemorrhage in conjunction with high‐salt intake and low‐protein diet (Yang et al., 2016). Daily treatment with BIA 10‐2474 at the high dose of 30 mg·kg⁻¹ p.o. did not affect mortality rate in this model and, in surviving animals, did not increase the incidence of haemorrhage or oedema. At this dose, the primary off‐targets identified by activity‐based protein profiling ex vivo, such as ABHD6 and CES, would have been inhibited as well as SEC11/11a, lysosomal thiophosphatase and androgen‐induced protein 1 (AIG1; see Table 6). However, the phospholipases identified by van Esbroek (van Esbroeck et al., 2017) and by us that have been associated clinically with neurological symptoms (i.e. PNPLA6 and PLA2g6) were probably unaffected at this dose.

It is important to point out some of the limitations of the present study in relation to identifying off‐target interactions that could be responsible for the clinical signs seen with BIA 10‐2474. The most obvious one is that we have used the rat for most of our studies. However, there is no other way to reproduce the duration of drug exposure that occurred in the clinical trial. It is nevertheless important that we have identified a very similar range of off‐targets to van Esbroeck et al. (2017) with what appears to be a similar relative potency compared with fatty acid amide hydrolase. However, relative potency of irreversible inhibitors such as BIA 10‐2474 and PF‐04457845 at serine hydrolase targets in vivo depends on several factors such as the inactivation constants, the quantity of enzyme present and the duration of exposure. None of the studies carried out so far with BIA 10‐2474 have done that. It would be important to characterize the interaction of the compounds to the identified off‐targets and to evaluate their in vivo off‐target occupancy at the different exposures to properly extrapolate our data to the clinical situation. It should also be noted that we, like others, have mostly considered CNS targets as being responsible for the clinical trial accident and serine hydrolases in particular. Although our study in SHRSP was carried out to explore possible mechanisms outside the CNS (albeit related to CNS vasculature), further studies outside the CNS and looking more broadly at protein targets using proteome‐wide screening techniques, such as surface plasmon resonance or MS imaging (Yang et al., 2016; Zhou, Liu, Rothschild, & Lim, 2016) could be of relevance. One other aspect that needs to be considered is that in this study and most of the non‐clinical studies with BIA 10‐2474, the rats were relatively young, although at the end of the 26‐week toxicology study, they would correspond to 25‐ to 30‐year‐old humans according to Sengupta (2013) and no pathology of the type seen in the clinic was observed. The subjects affected by toxicity of BIA 10‐2474 in the clinical trial were aged between 27 and 49 (Kerbrat et al., 2016) and so only slightly older than the rat‐equivalent age. Nevertheless, it will be useful in future studies to explore the effect of age on the sensitivity of the identified off‐targets to inhibition by mechanism‐based inhibitors such as BIA 10‐2474 as well as their rate of recovery.

The present study provides detailed information on the interaction of BIA 10‐2474 with fatty acid amide hydrolase and has explored potential off‐targets using traditional in vitro binding methods for non‐serine hydrolase targets and both in vitro and ex vivo activity‐based protein profiling for serine hydrolases. Although BIA 10‐2474 showed good selectivity in binding studies, the activity‐based protein profiling studies mostly confirmed the off‐targets already identified by van Esbroeck et al. (2017) and also identified several additional serine hydrolase off‐targets. Our data suggest that most of these interactions would have occurred at the doses used in preclinical toxicology studies. However, without further clinical information, such as the autopsy results and measured CNS exposure, it is not possible to determine if these off‐targets can be associated with the toxicity observed in the clinical trial with BIA 10‐2474. At present, the available data do not allow conclusions to be made about the specific mechanisms that could have caused the clinical trial accident.