Inhibition of insulin-regulated aminopeptidase (IRAP, EC 3.4.11.3), a zinc metalloprotease
belonging to the oxytocinase subfamily of the M1 family of metalloproteases,[1 ] has attracted significant attention over the last two decades as an approach to
develop molecules with potential clinical applications as cognitive enhancers for
the treatment of, e.g., Alzheimer’s disease. Hence, considerable efforts have been
devoted to identifying inhibitors suitable for further optimizations into clinical
candidates.[2 ]
[3 ]
[4 ]
According to one hypothesis, the cognitive enhancement observed after administration
of IRAP inhibitors is due to a suppression of the enzyme-mediated degradation of the
substrates vasopressin[5 ] and oxytocin,[6 ] endogenous cyclic peptides known to improve parameters linked to cognition.[7 ]
[8 ]
[9 ]
[10 ] Newer findings suggest that IRAP inhibitors may also facilitate memory by increasing
the hippocampal dendritic spine density occurring via a GLUT4-mediated mechanism.[11 ]
The first druglike small-molecule IRAP inhibitors were reported by Albiston et al. in 2008,[12 ] built around a benzopyran scaffold. Subsequently, as a result of rational design
from substrate structures and by utilizing various screening approaches, numerous
druglike IRAP inhibitors encompassing very different scaffolds have been developed.[2 ] Examples include aryl sulfonamides,[13 ]
[14 ] diaminobenzoic acid derivatives,[15 ] pseudophosphinic peptide transition-state analogues,[16 ] spiro-oxindole dihydroquinazolinones,[17 ] amino acid derivatives of bestatin,[18 ] and imidazopyridines.[19 ]
We recently reported a high throughput screening of 400,000 small molecules towards
IRAP inhibition.[20 ] From this investigation, a qualified hit list of potent IRAP inhibitors was constructed.
The most potent of these compounds, QHL1 (Figure [1 ]), was selected for further investigation, and in a subsequent publication we reported
several analogues and evaluated their IRAP inhibition, resulting in the chloro-substituted
1 , with a 10-fold improvement in potency.[21 ]
Figure 1 Structure of compound QHL1 , 1 , and the general structure of compounds 2 –20 explored in this paper
We herein report various alterations of the sulfonamide function based on compound
1 . The synthesis and inhibitory capacity of 2 –20 as well as proposed binding modes, as deduced from molecular docking and free energy
perturbation (FEP) of selected IRAP inhibitors, are presented (Figure [2 ] and 3).
Figure 2 Synthesized analogues with inhibitory concentrations; IC50 value of 1 from previous publication[21 ]
The three analogues of compound QHL1 , 2 –4 , were synthesized via Fischer esterification of 2-(4-bromophenyl)acetic acid (1a ) to the methyl ester 1b (Scheme S1). This was followed by regioselective chlorosulfonylation using chlorosulfonic
acid. Due to its high reactivity, the sulfonyl chloride intermediate was carefully
extracted and directly reacted with amines A2 –A4 to achieve the sulfonamides 2.1 –4.1 . The sulfonamides were then debrominated using palladium catalysis (with simultaneous
solvent-mediated transesterification to the butyl ester when n -butanol was used) resulting in 2.2 –4.2 . This was followed by installation of hydroxamic acid by treatment with an excess
of hydroxylamine and KOH. Compounds 5 –20 were synthesized using the same strategy, starting from 2-(4-chlorophenyl)acetic
acid (2a ), but with the omission of the dehalogenation step.
The X-ray IRAP structure without inhibitor in the open-like conformation was retrieved
from the PDB (code: 4PJ6)[22 ] and prepared for computational simulations with the Protein Preparation Wizard pipeline
in Maestro (Schrödinger, version 2024-2). Preparation steps included proton addition
and assignment of the most likely protonation state for titratable residues and rotameric
assignment of planar residues (i.e., Asn, Gln, His). The docking pose obtained for
compound 1 in our previous work[21 ] (Figure [3 ]A) was used as a basis to build an analogous docking pose for each compound considered
(5 –8 , 11 –13 , and 15 –20 ) with Maestro, followed by energy minimization of the ligand implemented with the
OPLS4 force field[23 ] (Figure [3 ]B and supplementary Figure S1). The PDB files generated for each ligand, together
with the prepared protein file, were the starting point for molecular dynamics (MD)
simulations associated to relative binding free energy (RBFE) calculations using the
dual-topology QligFEP automated workflow.[24 ] QligFEP allows the user to easily define each vertex comparing compound pairs and
subsequently generate all free energy perturbation (FEP) input files and perform the
ulterior analysis. For each RBFE calculation, the associated MD simulations in both
protein and solvent environments were conducted on the MD software Q[25 ] under spherical boundary conditions, as described in the Supporting Information.
Figure 3 (A ) Proposed binding mode of for compound 1 . (B ) Initial docking poses of all derivatives here considered for FEP simulations, following
the same coloring scheme as in C. (C ) Results from FEP simulations. The experimental relative binding free energy shift
between each compound pair considered (gray) is compared with the FEP calculated value,
colored according to the nature of the substituent explored (see legend and text for
explanation), pentacyclic = 5-mebered heterocyclic. Error bars indicate the associated
s.e.m. of the calculation estimated from replica simulations (see the Supporting information).
Initial studies to explore the impact of different sulfonamide functions were performed
by modifying the original hit compound QHL1 from our previous screen.[20 ] In that work, we defined two possible binding modes for this scaffold, both showing
bidentate coordination of the catalytic zinc ion through the hydroxamic acid moiety
and a hydrogen bond with neighboring glutamic acid E431. A detailed exploration of
the SAR generated therein followed on the basis of systematic FEP simulations, performed
on each alternative binding mode.[20 ] The results of the FEP exploration allowed us to conclude that the dihydroisoquinoline
ring of 1 would remain in a solvent accessible cavity, lining up with Y961 and further stabilized
by hydrogen-bond interaction between the sulfonamide oxygen and Y549. We will herein
discuss the SAR of the present series on the basis of such a binding mode, using our
QligFEP approach for RBFE simulations (see Figure [3 ]).[24 ]
All three compounds lacking the chloro substituent in 1 (2 –4 ) demonstrated no inhibitory activity below 10 μM (Figure [2 ]). In all subsequent studies, the crucial chloro group[21 ] was maintained while modifications to the sulfonamide moiety were explored. We initially
explored the introduction of bromine in each of the four available positions, with
the goal of probing available space in the enzyme and to evaluate potential electronic
effects. Compound 5 with the electron-releasing halogen positioned para to the sulfonamide function, demonstrated an IC50 of 0.117 μM, the best value in the bromo series. The potency as IRAP inhibitors was
reduced twofold for compounds 6 and 7 (IC50 = 0.234 μM and 0.251 μM, respectively) and more than 10-fold for 8 (IC50 = 1.258 μM).
The RBFE exploration of the brominated derivatives partially captured these effects,
effectively showing a detrimental effect of moving the halogen from the optimal para position in 5 to the alternative locations in 6 , 7 , and to a minor extent on 8 (Figure [3 ]C, yellow bars, Table S1). While the closure of the full cycle (i.e., 5 → 6 → 7 → 8 → 5 ) yields a low-associated error of 0.22 kcal/mol, the mean unassigned error (MUE)
of the five possible pairwise transformations collected on Figure [3 ]C reached a considerable value of 0.73 kcal/mol, exceeding the average experimental
RBFE values of the five pairs involved <ΔΔGbind,exp > = 0.65 kcal/mol (brackets indicating an average value). This observation, together
with the relatively high average standard error of the mean (s.e.m.) obtained for
these calculations (<s.e.m.>) = 1.13 kcal/mol, error bars on Figure [3 ]C) weakens the statistical significance of the FEP results for the bromine analogues.
Opening of the aliphatic ring system of 1 to make the molecule more flexible resulted in a significant loss of activity: the
N -ethyl-substituted sulfonamide 9 demonstrated an IC50 value of 1.51 μM, while the corresponding value for the methyl sulfonamide 10 was 0.540 μM. These findings suggest that the locked aliphatic ring facilitates positioning
of the aromatic ring in a favorable position. The position and orientation of the
phenyl ring was further examined by alteration of the size of the nitrogen heterocycle.
Extension of the ring system of 1 by incorporation of a methylene group to create 11 resulted in a more than 10-fold reduction of activity (IC50 = 0.794 μM). The sulfonamide 12 comprising the smaller five-membered ring system and with the nitrogen atom connected
to the phenyl ring resulted in a similar outcome (IC50 = 0.676 μM).
An enlargement of the ring system in 12 by addition of a methylene group was essentially deleterious for the activity and
13 demonstrated an IC50 value of 3.16 μM. A comparison of compound 1 and its isomer 13 demonstrates that a correct location and direction of the bicyclic system is crucial
for efficient inhibition as 13 is 50-times less efficient at IRAP inhibition than 1 . Thanks to the dual topology approach implemented in QligFEP, we could simulate such
effects on ring-size variation with high accuracy, with the transformations 11 → 12 → 13 → 11 showing qualitative and quantitative agreement with experimental data (Figure [3 ]C, light-green bars, Table S1). In other words, based on the binding mode proposed,
the FEP simulations can accurately explain the relative differences in affinity, with
a MUE = 0.13 kcal/mol, in this case significantly smaller than the average experimental
RBFE values of this compound series (<ΔΔGbind,exp > = 0.62 kcal/mol). Following these insights, the aliphatic ring was left intact,
while the phenyl ring of 1 was replaced with various heteroaromatic systems. The indole and pyrazole derivatives
14 and 15 , both possessing hydrogen-bond-donating capacities, were inefficient inhibitors (IC50 = 6.92 μM and 1.38 μM, respectively). Conversely, the thiophene and oxazole compounds
16 and 17 with hydrogen-bond-accepting abilities resulted in better outcomes, inhibiting IRAP
with IC50 values of 0.117 μM and 0.480 μM respectively. Within this series, we restricted the
FEP exploration to 5-membered heterocycles, including the transformations 15 → 16 → 17 → 15 (Figure [3 ]C, dark green bars, Table S1). As in the previous case, the accuracy of the calculations
was remarkable (MUE = 0.25 kcal/mol), showing enough sensitivity of the proposed binding
mode to quantitively assess the SAR of these compounds, which on average resulted
in experimental RBFE variations of <ΔΔGbind,exp > = 1 kcal/mol.
The exploration continued with other hydrogen-accepting heterocycles and replacement
of the phenyl ring with pyridine to provide 18 resulted in a compound with an IC50 value of 0.100 μM, while its regioisomer 19 with the nitrogen atom located at approximately the same position as the nitrogen
in the oxazole derivative 17 exhibited an IC50 value of 0.044 μM. Hence, 19 acts as a better IRAP inhibitor than the lead compound 1 . The pyridine derivative 19 also has a more than 10-fold better ligand-lipophilicity efficiency (LLE) than that
calculated for 1 , the latter devoid of the nitrogen atom (LLE = 6.42 vs 5.00, pIC50 – logD) and a better LLE than its pyridine regioisomer 18 (LLE ) 5.91 pIC50 – logD). The LLE is a parameter-linking potency and lipophilicity and is applied
in efforts to estimate druglikeness.[26 ] An incorporation of an additional nitrogen atom in the ring system to provide the
dihydropyridopyrimidine hydroxamic acid derivative 20 was productive and resulted in the best IRAP inhibitor in the series with an IC50 value of 0.034 μM.
These bioisosteric effects were again well captured by our structural model of IRAP
inhibitor binding, with the RBFE simulations 18 → 19 → 20 → 18 showing the gradual improvement in affinity described above (from 18 to 20 , Figure [3 ]C, blue bars, Table S1) with very high sensitivity (MUE = 0.19, significantly under
the average value of experimental RBFE values, <ΔΔGbind,exp > = 0.65 kcal/mol).
In summary a series of 16 new chloro-substituted benzylhydroxamic acids were synthesized
and their capacity to inhibit IRAP assessed. We report that replacement of the dihydroisoquinoline
moiety in 1 , with a dihydropyridopyrimidine, leads to a twofold improvement of the inhibition,
and 20 exhibited an IC50 value of 34 nM. The binding mode proposed was challenged with extensive FEP simulations,
strongly supporting the observed structure–activity relationship (SAR) in this series
of benzylhydroxamic acids. Further evaluations of this series of new IRAP inhibitors
are ongoing.
