C. Lawson, T.B. Ahmed Alta, G. Moschou et al.
European Journal of Medicinal Chemistry 225 (2021) 113751
BindingDB, a public, web-accessible database of measured
binding affinities [22], was used to provide suggested protein tar-
gets known to interact with structurally related diarylurea ligands.
In summary, the search with a simple diarylurea (PhNH(CO)NHPh)
employing the “substructure” setting resulted in 428 different
targets; no hits were obtained if the three methoxy groups were
incorporated. The range of targets identified included: multiple
kinases (including CDK2, VEGFR2 and Aurora A, >1200 diarylurea
hits in total against kinases); angiopoietin-1 receptor; carbonic
confirming their promise as potent lead compounds against (p53
mutant) medulloblastoma cell lines.
Interestingly, despite the promising in vitro cytotoxicity data,
in vitro analysis of the most promising compound (9c) against
CDK2, IGF-1R, VEGFR2 and Aurora A kinases [27], only returned
~30% inhibition against Aurora A (at 100 mM), with no notable in-
hibition against the other kinases, suggesting that these com-
pounds are not potent inhibitors against these kinases but that they
exert their effects, at least in part, through interaction with other
biological targets.
anhydrase 2; IDH2 (R140Q); integrin a4b1 (VLA-4); mast/stem cell
growth factor receptor Kit; mitogen-activated protein kinase 14;
platelet-derived growth factor receptors; and numerous purinergic
receptors, amongst others.
2.3. ADMET property predictions
To identify potential target kinases of interest (given the large
(>1200) number of diarylurea hits), all the kinases identified as hits
from the database were cross-referenced with kinases known to
play a role in brain tumours [23,24]. Based on this data, and
structurally similar, known kinase inhibitors, we considered
whether the active unmethylated diarylureas (series 4) had the
potential to act as kinase type II or type III inhibitors [25]. Therefore,
the number of target kinases were filtered even further, and only
kinases containing co-complex crystal structures of a bound type II
or III ligand were selected for docking.
The first-generation compounds were docked into the five
chosen kinases (CDK2; VEGFR2; Aurora A; IGF-1R and B-Raf), to
determine whether these compounds could potentially bind. In
general, the results indicated that the compounds could bind, and
specifically we observed that the trans, trans urea 4b was ranked
number one against 3 of the 5 kinases studied, and that the
unmethylated ureas and amides ranked better than their methyl-
ated counterparts.
Having confirmed naphthylamine derivative 4b as a lead com-
pound, a further 56 analogues were designed from commercially
available isocyanates containing polycyclic rings with various
differentially functionalised groups. Once created, the analogues
were prepared for docking against the same kinases utilised for the
generation 1 compounds, and eight candidates were selected for
synthesis (Table S3). Although we designed the compounds to
target kinase inhibition, the 3,4,5-trimethoxyphenyl ‘A’ ring moiety
still remained an integral feature for all generation 2 analogues as
some potent kinase inhibitors possess this structure, such as those
synthesised by Ganser et al., which exhibit nanomolar activity
Once synthesised, the compounds were evaluated for their
cytotoxicity, and Table 2 shows the biological activity against both
cell lines and the structures of the synthesised compounds and
compares the results of these second-generation compounds (9a-
9h) against the activity of 4b. Significantly, compounds 9a-9c are an
order of magnitude more active than 4b, and almost 60 times more
active in the case of naphthol analogue, 9c.
To further explore the translational potential of our most active
compounds (4b, 9a-d, 9h) from the cellular experiments, ADMET
property predictions were performed using the QikProp v5.8 pro-
gramme [28]. Unfavourable pharmacokinetic profiles are respon-
sible for the failure of many drug candidates [29], with CNS drug
development having additional requirements to consider, such as
the obstacle of blood-brain barrier (BBB) permeability [30]. Hence,
ADMET evaluation earlier in the drug design process is advisable,
with computational predictions providing a good starting point.
The results of the property predictions are presented in Table 3.
The predicted values should be considered approximate, with
some property predictions considerably dependent on conforma-
tion. Nevertheless, they provide useful information to be consid-
ered in future ligand optimisations. The first test of drug-likeness
we looked at was Lipinski's Rule of Five [31]; none of our active
compounds demonstrated any violations of these rules. Jorgensen's
Rule of Three results were promising with only one of the six
compounds, 9h, having a violation [32]. For 9h, a possible solubility
issue (log S) was flagged; an obvious consequence of its larger
anthracene aromatic substituent. The predicted Caco-2 perme-
abilities of the compounds are notable, with excellent permeabil-
ities (872e2963 nm sꢀ1) predicted for all compounds. Log BB
(logarithmic ratio between concentration of a compound in brain
and blood) and CNS values for the compounds are not ideal but
exhibit promise for the compounds in further optimisation efforts.
A log BB ¼ 0 implies an equal concentration of compound on both
sides of the blood brain barrier. The log BB threshold values for
blood brain barrier penetration vary in the literature as described in
Carpenter et al. [33], and include a log BB value of ꢀ1 [34], which
suggests that all the compounds pass, but a log BB ꢁ 0 is mainly
suggested. Additionally, while Veber et al. suggested that a PSA of
<140 Å2 is desirable for oral bioavailability [35], a lower threshold
of PSA < 90 Å2 (and MW < 450 Da) is proposed by van de Water-
beemd for CNS drugs [36]. Estimation of blood-brain barrier
crossing of drugs using molecular size and shape, and H-bonding
characteristics for drugs targeted to the CNS suggests all com-
pounds fulfil this [36]. The predicted values of logKhsa, binding to
human serum albumin, for all compounds are within the desired
range (ꢀ1.5-1.5) for 95% of known drugs. Finally, only compound 9h
again raised an initial concern regarding potential blockage of
HERG Kþ channels. In summary, the most active cellular com-
pounds are also promising CNS drug candidates for lead optimi-
sation considering the predicted pharmacokinetic profiles.
Of note is the improvement in activity of analogue 9c vs 4b by
simply incorporating a judiciously placed hydroxyl group; an in-
crease in activity not seen in isomer 9e where, in fact, all activity is
lost. Interestingly, any improvement in activity seen by the addition
of such an H-bonding group in 9c, is not required in the comparably
active indane analogue, 9a. A similar trend is also seen with indole
analogues 9b vs 9f where H-bonding and substitution presumably
play a large role in conferring activity. Furthermore, the anthracene
analogues 9g and 9h again highlight the importance of substitution
pattern here since 9h, substituted at position-1, similar to 4b, is
active and presumably can reside in the active site of its target,
whereas 9g, substituted at position-9, offers a more challenging
substrate for the target. On the whole, the most biologically active
generation-2 compounds are selective for the DAOY cell line,
3. Conclusion
To conclude, we have shown that potent anticancer activity has
been achieved with a series of diarylureas, primarily against the
DAOY medulloblastoma cell line, which represents a disease which
is highly aggressive and the most common high-grade childhood
embryonal tumour of the central nervous system, but that the
5
Ahmed Alta, Thowaiba Babikr
