Tackling the Moving Mutant Target: Taxifolin Derivatives as Novel Putative E571K Exportin-1 Inhibitors for KRAS-mutant Lung Adenocarcinoma Therapy

Aim: To establish, through molecular modelling, safe and clinically acceptable putative antagonists of E571K-mutated exportin-1 among the bioactive compounds in various parts of Juglans mandshurica. Methods: The bioactive compounds were subjected to compendium of druglikeness and lead-likeness filter workflows prior to docking of the resultant compounds into E571K exportin-1 active site using PyRx AutoDock vina to establish their binding affinity and interaction profile. The evolutionary algorithm of Osiris property explorer DataWarrior software as well as lead-likeness filter were employed for generation of novel non-promiscuous analogues of the lead compound with better putative selectivity and clinical acceptability as E571K Exportin-1 antagonists.Results: The findings of this study present taxifolin as the putatively effective and lead-like E571K Exportin-1 inhibitor with high potential of qualifying for clinical evaluation but is associated with high promiscuity tendency in high throughput screening. The evolutionary derivation of novel analogues of the compound, however, results in the generation of putatively non-promiscuous, non-toxic, and lead-like E571K Exportin-1 antagonists with high synthetic accessibility and clinical developability for evaluation in the strategy for treatment of drug-resistant KRAS-mutant lung adenocarcinoma condition.


Introduction
The positive association between the recurring poor prognostic drug resistance and their varied genomic mutations widely emphasizes its clinical relevance in the various chemotherapeutic strategies for reducing cancer-associated morbidities and mortalities 66,67 . This include the missense single nucleotide gain-of-function E571K mutation of exportin-1 gene (XPO1) that thereby codes for positively charged lysine instead of negatively charged glutamate residues in exportin-1 and abnormally elevates its affinity for nuclear export signals (NES) 68,17 .
The deregulation of exportin-1 (also known as chromosomal region maintenance 1), as a major RanGTP-driven exporter of more than 1050 human NES-bearing proteins (including tumour suppressors such as p53, BRCA1 etc. 69 and protooncogenes) as well as several RNA species 70,71 across the cell nucleo-cytoplasmic partition, is an active driver of several oncogenic conditions including the highly untreatable KRAS-mutant lung adenocarcinoma 72,73 .
Although the recurrence of exportin-1 E571K mutation is mostly reported in some haematological malignancies 74,67 , while Taylor et al. (2019) opined its rarity in solid tumours; there are currently reported 0.13% of XPO1 E571K mutation cases in a solid tumour such as KRAS-mutant lung adenocarcinoma 53,75 , and developing effective selective inhibitor of nuclear export signals (SINEs) for tackling the isolated E571K-mutated exportin-1 is a plausible strategy with potential significance in improving the prognosis of this highly untreatable condition. Indeed, several currently developed selective inhibitors of nuclear export signals (SINEs), including the clinically evaluated selinexor (KPT-330), KPT-185 etc., has been suggested as effective inhibitors of both the mutated and wild type XPO1 protein but remain limited in potency as single agent and/or safety in single or combination therapy or both 51,18 .
We intend to find novel selective inhibitor (s) of nuclear export signals (SINEs), through molecular modelling, in this study that can effectively inhibit E571K-mutated exportin-1 with better potency and safety profile among the compounds reportedly identified in several studies 32,41,21,35,61,60,33,62,45,57,28,55 from various parts of Juglans mandshuricaa scientifically proven anticancer plant in several tumour cell lines 36,27 . This concept is validated by the fact that natural products are constituted of complex chemotypes with favourable safety profile and promising potential for discovery of novel nuclear export inhibitors 76 .

Protein Preparation
The 3D coordinate files of crystal structure of an unliganded human (homo sapiens) chromosomal region maintenance 1 (E571K)-Ran-RanBP1 complex (PDB ID: 6X2O) was retrieved from the Research Collaboratory for Structural Bioinformatics (RCSB) protein data bank (http://www. rcsb.org), visualized using the Schrodinger molecular graphics program PyMol® and prepared using Schrodinger maestro protein preparation wizard. Briefly, the Ran-RanBP1 complex (chain A and B) were deleted while the remainder (chain C) was prepared under OPLS2005 force field at pH of 7.0±2.0. All steric clashes were corrected, hydrogen bond order fixed, while the missing side chains and loops were corrected using prime. Disulfide bonds were created, water molecules beyond 5.00Å were removed from the ionized (Het) groups, waters with less than 3 hydrogen bonds to non-waters were deleted while the bond orders were assigned.

Visualization of Docking Results
The ligand interactions at the target binding pocket and their three-dimension (3D) interaction with the target residues were visualized using Schrodinger Python Molecular graphics interface (PyMol®) version 2.2.0 software while the two-dimension (2D) ligandresidue interactions were obtained using BIOVIA Discovery Suite version 17.2.0.16349. The local minimum energy conformation binding mode of the compounds are regarded as the optimal binding pose at the target binding site in this study 56,46 .

2.5
Validation of Docking Study

Validation of Docking Protocol
The docking protocol of this study was validated by redocking of the same standard compound into the same target binding site and examining the relative binding pose of the ligands using the Python Molecular graphics interface (PyMol®) software.

Validation of Docking Algorithm
The binding energy calculation algorithm applied in this study was validated by comparative analysis of docking scores of the hits and the standard compound (KPT-185) in other different docking algorithms including Molegro virtual docker (PLANTS scoring function), LeDock, and iGEMDock.

Molegro Virtual Docker (PLANTS Scoring Function)
We employ the iterated simplex search algorithm (based on Nelder-Mead algorithm) with ant colony optimization component 4 of Molegro Virtual Docker for simulation of interaction between the E571K Exportin-1 active site and the hit compounds in this study, while the protein-ligand ANTS (PLANTS) scoring function with GRID 23 was used to calculate the potential energy values on a cubic grid prior to docking.

LeDock
We docked each of the three hits and the standard (KPT-185) into the target active site and calculated the docking scores through combination of simulated annealing and genetic algorithm for each ligand pose, orientation, and rotatable bond optimization in LeDock 63 , while the scoring function is on the basis of transferable and accurate energy function consisting of van der Waals interaction energy and a unique empirical-based hydrogen bonding energy 64 .

iGEMDock
The various possible conformations of each of the hits were also obtained through the genetic evolutionary algorithm of iGEMDock 59 that also include the ligand intramolecular energies for total energy calculation in this study.

Pharmacokinetics and Toxicity Profiling
We employed a compendium of workflow involving the web-based predictive tool SwissADME (http://www.swissadme.ch/index.php), and DataWarrior to profile both the hits and the lead compound derivatives in this study for their absorption, distribution, metabolism, excretion, and toxicity properties.

De novo Generation of Lead Compound Derivatives
An evolutionary library of derivatives of the lead compound was built in this study by creating compounds similar to current standard-of-care drugs for 50 generations using the Osiris property explorer DataWarrior software in which the substructures of the parent compound at the turn of each generation were randomly left untouched. The SkelSpheres descriptors were used in the generation of 128 compounds per generation while the maximum of 8 compounds were allowed to survive a generation.

Compounds Obtained
The total of 123 compounds reported by some studies 32,41,21,35,61,60,33,62,45,57,28,55 was retrieved in this study while a total of five (5) predictably drug-like and non-toxic compounds (Table 1) The druglikeness score of a compound, calculated from the compound toxicity descriptors and the degree at which it contains predominantly the fragments frequently present in commercially available drugs, helps validate its potential as a drug candidate 78 . The prefiltered five compounds from Juglans mandshurica in this study (Table 1) (Table 2), the potential of central nervous effects and toxicities via their blood-brain barrier impenetrability.
The pivotal balance between a compound molecular weight and other descriptive features of its bioavailability essentially defines its potency and safety profile. This is because increased molecular weight positively correlates with elevated bioactivity but inversely associates with gastrointestinal absorption 78  A definable ligand-target specificity is a function of the compound spatial orientation and geometry within the target binding site 19,84 . The molecular shape index range (0.385-0.667) of the retrieved compounds in this study (Table 2)   The molecular flexibility scores of the retrieved compounds in this study (Table 2), a function of their complexity, changeable dihedral bond angles, and geometry weighting factors 86,87 , are in the range of 0.136-0.751 on the scale of 0.0 (flexible) to 1.0 (completely rigid). This plays crucial role in defining the potential relative binding capacities of the compounds in the target active site as a ligand conformational rigidity (due to fewer rotatable bonds) positively correlates with its target binding affinity potential 29,15 .

Ligand Docking
Potency and safety limitations abound around the potential effects of current SINE compounds as 463247 as having relatively higher binding affinities (-5.5 kcal/mol to -5.6 kcal/mol) when compared with that of KPT-185 as the standard compound, thereby suggesting relatively higher bioactivity efficacy potential of these hits as E571K exportin-1 inhibitors. It is also worthy of note that although taxifolin showed comparatively equivalent binding affinity to that of KPT-185 (-5 kcal/mol) in this study ( Figure 1); it displays relatively lower molecular flexibility score (Table 2) with the consequential tendency of comparatively advancing its efficacy due to its rigiditymediated lower entropy loss at the E571K exportin-1 active site. The superimposition of relative binding poses of the redocked KPT-185 at the E571K Exportin-1 active site (Figure 2) in this study shows the ligand docking procedure in this study as valid, thereby rendering the various docking scores as accurate.

Validation of Docking Algorithm
The results from other docking algorithms (   Exportin-1 active site.

Pharmacokinetics, Toxicity, and Lead-likeness Profile of Hit Compounds
The potency and safety profile of an orally bioavailable small molecule drug entity, a function of its pharmacokinetic profile, is partially defined by its lipophilicity-mediated gastrointestinal permeability and metabolism 5 . This study (Table 4) (Table 4) does not significantly inhibits CYP3A4; an enzyme responsible for degradation of most drug entities in the liver.
The blood brain barrier (BBB), dynamically interfacing the blood and brain, prevents toxic xenobiotics entry into the brain endothelium as well as supply it with requisite nutrients while the P-glycoprotein receptor-mediated transport system plays vital role in the outflux of unwanted xenobiotics from the BBB and several other organs 6,24 . The three hit compounds, unlike KPT-185, does not permeate the blood-brain barrier in this study (Table 4), thereby showing very low potential adverse clinical effects on the brain by the hit compounds in KRAS-mutant lung adenocarcinoma therapy. Although this study reveals KPT-185 as blood-brain barrier permeant; it is noteworthy to determine the comparative threshold in vivo at which it may induce neurotoxic side effects during the disease therapy. Also, this may alternatively associate with its ability to confer neuroprotection against xenobiotics-induced axonal damage as reported by Haines et al.
(2015) 20 . Only taxifolin, however, showed the properties of significantly less prone to Pglycoprotein receptor-mediated outflux (comparably similar to KPT-185) in this study (Table 4) and can play significant role in its high GIT absorbability-mediated biodistribution model; a feature made pronounced by its non-blood-brain barrier permeability.   The pan assay interference (PAINS) filter of the hit compounds in this study (Table 6) The synthetic accessibility score of this study hits (Table 6) (easy to synthesize) to 10 (very difficult to synthesize).

De novo Generation of Lead Compound Derivatives
The lead-likeness of taxifolin, among the three hit compounds, in this study (Table 5) presents it as a putative drug-like E571K Exportin-1 inhibitor with high likelihood of qualifying for clinical evaluation and development. Its potential efficacy, however, is limited by its promiscuity tendency in high throughput screening which is a precursor to off-target binding, as well as its violation of Brenk rule of lead-likeness selection due to its constituted catechol substructure. Therefore, exploring the possible chemical space that can be generated from random mutable combination of the compound fragments (based on similar flexophore and dissimilar skelsphere descriptors to that of taxifolin as the parent compound 48 may presents viable approach toward discovery of its better analogue (s) with the most promising property profile as lead-like E571K Exportin-1 inhibitors.
A nature-mimicking evolutionary approach of randomly mutating the taxifolin structure based on  (Table 7).  The resultant five taxifolin analogues with no Brenk and PAINS alert in this study shows high gastrointestinal permeability but possess property, unlike their parent compound (Table 4), which rendered them as substrate for P-glycoprotein receptor (Table 8). This, given the fact that they are non-blood brain barrier-permeable (Table 10), may contributes toward their reduced potential of inducing any side effects associated with the neurological system while aiding in preventing accumulation of the drug entities, via excretion, en route the action site. Also, the analogues, similar to the parent taxifolin compound (Table 4), display non-inhibition tendency toward the various studied cytochrome P450 isoforms, except compound 235 of generation 26 (Table 8).
The bioavailability score of the analogue compounds with no Brenk and PAINS alert (0.55), in similarity to that of the taxifolin parent compound, also present them as orally bioavailable compounds (Table 9) while obeying the various conventional oral bioavailability and druglikeness model rules.
Furthermore, the lead-likeness evaluation of these no-Brenk-and-PAINS-alert taxifolin analogues shows them as non-toxic compounds with high tendency of clinical developability (Table 10) while their range of synthetic accessibility score (3.48-3.59) presents them as easy to synthesize (comparably with that of their parent compound) for further preclinical and clinical evaluation.

Bioactivity Profile of Taxifolin Derivatives with No Brenk and PAINS Alert
Evaluating the bioactivity probability of the obtained lead-like taxifolin derivatives with no Brenk and PAINS alert in this study (Table 11) characteristically reveals the compounds as highly active ligand of the nuclear receptor such as E571K Exportin-1, when compared with that of KPT-185.
Both compound 235 and 250 of generation 26 and 28 respectively exhibits the relatively highest bioactivity against nuclear receptors (Table 11), when compared with KPT-185. It is also noteworthy that although these taxifolin analogues tend to also exhibit moderate activity in modulating ion channel, inhibiting kinase and protease enzymes, as well as actively inhibiting Gprotein coupled receptors and enzymes; their lack of PAINS alert, unlike the parent taxifolin, presents the plausibility of restrictively advancing their bioactivity specifically against the nuclear E571K Exportin-1.

Binding Affinities of Taxifolin Derivatives with No Brenk and PAINS Alert
Comparative evaluation of binding affinity potential of the lead-like non-promiscuous taxifolin analogues as putative E571K Exportin-1 inhibitors against the parent taxifolin compound and KPT-185 in this study ( Figure 5) reveals compound 235 of generation 26 as having relatively higher binding affinity (-5.3kcal/mol) than both taxifolin (-5kcal/mol) and KPT-185 (-5kcal/mol) while both compound 220 and compound 379 respectively of generation 24 and 43 shows comparatively equivalent binding affinity with the parent compound and KPT-185 ( Figure 5). It is also noteworthy that the characteristic display of highest binding affinity ( Figure 5) and highest bioactivity tendency against nuclear receptors (

(A) Interaction between Compound 220 and E571K Exportin-1 (B) Interaction between Compound 235 and E571K Exportin-1 (C) Interaction between Compound 250 and E571K Exportin-1 (D) Interaction between Compound 294 and E571K Exportin-1 (E) Interaction between Compound 379 and E571K Exportin-1
Examining the interaction between the non-promiscuous lead-like taxifolin analogues with the binding site of E571K Exportin-1 in this study reveals formation of electrostatic Pi-cation dipole moment ( Figure 6) with the active lysine residue 360; a mechanism earlier stated (section 3.4) as having plausibly significant contributory role in E571K Exportin-1 inhibition. This is in addition to compound 220 ( Figure 6A), 235 ( Figure 6B), and 250 ( Figure 6C) of generation 24, 26, and 28 respectively forming Pi-alkyl bond with Leu 363 of E571K Exportin-1 active site; a feature which is notably absent in their parent compound ( Figure 3A). Compound 235 of generation 26, however, formed additional Pi-alkyl interaction with the active histidine residue 367 ( Figure 6B) that may have contributed toward its relatively higher binding affinity ( Figure 5) and led credence to our