The crystal data for compound 1 was collected on XtaLAB Synergy, Dualflex, Pilatus 300K diffractometer using graphite monochromated CuKα radiation (λ = 1.54158 Å) at a temperature 100 K, with data collection using Crysalis, and structure determination and refinement by Olex2 (Dolomanov et al., 2009). The structure was determined with the XT (Sheldrick, 2015) structure solution program using Intrinsic Phasing and refined with the XL (Sheldrick, 2008) refinement package using Least Squares minimization. The compound crystallizes in the orthorhombic space group, Pca2₁, with one molecule in the independent unit. The most important crystallographic data are summarized in Supplementary Table S1. Geometrical calculations were done by the PLATON program (Spek, 2009), while figures of crystal structures and analysis of H-bond graph sets by Mercury (Macrae et al., 2008).

CCDC no. of the title compound 1 is 2068871. File can be obtained free of charge from the Cambridge Crystallographic Data Center via www.ccdc.cam.as.uk/data_request/cif.

The geometries of neutral DFMO (2-(difluoromethyl)ornithine), its protonated cations as well as of corresponding hydrate and solvate (2-(difluoromethyl)ornithine hydrochloride hydrate) in the ground singlet state have been completely optimized using the Gaussian 09 program (Frisch et al., 2011). The stability of the optimized structures was tested by vibrational analysis (no imaginary vibrations). For calculations of stable conformers in both gas-phase and aqueous solution the density functional theory level of theory (Parr and Wang, 1994; Neumann et al., 1996; Bickelhaupt and Baerends, 2000) using the M06 hybrid functional (Zhao and Truhlar, 2008) and the polarized triple-ζ 6–311++G(d,p) basis sets (Hehre et al., 1986) for all atoms were used. The Conductor-like Polarizable Continuum Model (CPCM) (Klamt and Schűman, 1993; Barone and Cossi, 1998; Cossi et al., 2003) was exploited for the evaluation of the effect of hydration on the structure of the conformers studied. The gas-phase and solvated state molecular structures obtained by theoretical calculations were compared and discussed with the X-ray structure of these compounds in the crystalline state.

Hirshfeld surface (HS) analysis is a useful tool in providing information not only on nature of close non-covalent interactions but also on the topology of the ligand surface (molecular shape), electrostatic complementarity, interaction surfaces via visualization of the boundaries of a molecule (shape) and inter-contacts with adjacent molecules within the crystal (Spackman et al., 2016). Recently, HS methodology has been tested in the extended context – into the protein environment. It can be used to profile the shape of the protein pocket and generate the HS (with mapped electrostatic potential) around the ligand molecule (extracting the model data from the CSD) (Spackman et al., 2019).

Here, a thorough, either qualitative or quantitative, analysis of supramolecular interactions in 1 and 2 was carried out by HS approach (Spackman and Jayatilaka, 2009) using the CrystalExplorer 17.5 (Mackenzie et al., 2017) program, based on the X-ray crystallographic information data. 3-D HS maps and 2-D fingerprint plots (Spackman and McKinnon, 2002) were generated. The bond lengths of the hydrogen atoms were normalized to standard neutron diffraction values. The color scheme of the surface regions are related to the normalized contact distance, denoted as d _norm (a property reflecting the interactions of the molecule) (Spackman et al., 2016), defined in the context of the van der Waals (vdW) radii of the atoms and d _e , d _i , the distances from the HS to the nearest atom outside and inside the surface, respectively. Red areas characterize contacts shorter than the sum of vdW radii, white – equal to the sum of vdW radii, while blue — longer than the sum of vdW radii.

The electrostatic potential was calculated, and mapped on HS, using wave function STO-3-G basis sets at the Hartree-Fock theory level over the range of -0.025 au (Spackman et al., 2008).

The energies of intermolecular inter-contacts for the 3D energy framework construction were calculated in the same program by use of single-point molecular wave functions at B3LYP/6–31 G(d,p) level of theory, within a cluster of radius of 3.8 Ǻ (Mackenzie et al., 2017; Turner et al., 2017). The energy benchmark was calculated using Mackenzie’s method to scale energy frameworks. Quantitative analysis of interactions was performed on the base of the equation: E t o t   =   k e l e   E   ` e l e   +   k   p o l   E `   p o l   +   k   d i s   E `   d i s   +   k   r e p   E ` r e p where k characterizes scale factor values, E` <sub>ele</sub> – the electrostatic energy, E` _pol – the polarization energy, E` <sub>dis</sub> – the dispersion energy and E` _rep – the exchange repulsion energy (Mackenzie et al., 2017).

In the crystal packing, privileged and disfavored interactions formed by the pair of elements (X,Y) can be analyzed using the enrichment ratio (ER) (Jelsch et al., 2014). ER is a ratio between the interactions within the crystal and interactions derived from the theoretical calculations (HS analysis). The proportion S_X on the molecular surface, including the proportion of HS contacts (C_XY) can be described by the equation: S X   =   C XX   + 1 2 ∑ Y ≠ X C XY   ( ∑ X S X = 1 ) . (1)

Next, the ratio of interactions, and consequently ER, can be defined as: R XX = S X S X   and   R XY = 2 S X S Y   ( ∑ X R XY + ∑ Y ≠ X R XY =   1 ) . (2) E R XY   = C XY R XY (3) ER > 1 when the pairs of elements have high tendency to form contacts within the crystal, and ER < 1 when pairs avoid contacts with each other.

Absorption, distribution, metabolism, excretion, toxicity (ADMET) profiles for analyzed compounds and analogues of DFMO were evaluated using the pkCSM web-platform (Pires et al., 2015; Guan et al., 2019). The information was supplemented by other pharmacological parameters analyzed by SwissADME web-based interface, provided by the Molecular Modeling Group of the Swiss Institute of Bioinformatics (Daina et al., 2017).

Molecular docking studies to establish favorable ligand binding geometries for both enantiomers of eflornithine (DFMO, α-difluoromethylornithine), namely (2S)-(–)2,5-diamino-2-(difluoromethyl)pentanoic acid (L-DFMO) and (2R)-(+)-2,5-diamino-2-(difluoromethyl)pentanoic acid (D-DFMO) as well as the developed DFMO-based derivative, namely (2S,5S)-5-amino-2-(difluoromethyl)-6-fluoro-2-[(fluoromethyl)amino]hexanoic acid (DFMO-analogue), were carried out on a four CPUs-based desktop PC computer equipped with AMD Phenom™ II X4 965 Processor 3.40 GHz and 32 GB of RAM on a Microsoft Windows 10 Professional 64-bit operating system using AutoDock Vina vs. 1.1.2 program for Windows (http://autodock.scripps.edu/) (Trott and Olson, 2010). The respective ligand molecules L-DFMO, D-DFMO and DFMO-analogue in non-ionizable form were prepared at first with ChemAxon MarvinSketch vs. 14.9.1.0 (http://www.chemaxon.com/marvin/) using general “Cleaning in 3D” option to assign with proper 3D orientation and then calculating conformers with MMFF94 force field parameters and saved as .pdb files. To obtain the minimum energy conformation for docking studies, the initial geometries of the afore-pretreated ligands were additionally optimized in Avogadro version 1.2.0. (http://avogadro.cc/), after adding all the hydrogens to the structure, and saved as .mol2 files. The energy of the ligand molecules was minimized using built-in feature of Avogadro including General Amber Force Field (GAFF) (Wang et al., 2004) with Steepest Descent Algorithm (100 steps). The minimum conformation energies obtained for each ligand were as follows: E _{calc.(L-DFMO)} = –209.069 kJ/mol, E _{calc.(D-DFMO)} = –209.452 kJ/mol, and DFMO-analogue E _{calc.(DFMO-analogue)} = –204.022 kJ/mol. The visualization of the optimized geometries were performed using molecular visualization software, POV-Ray for Windows v3.7.0. msvc10.win64 licensed under the terms of the GNU Affero General Public License (AGPL3) (see Supplementary Figure S2 in Supplementary Material). Next, the Gasteiger partial charges were calculated with AutoDock Tools vs. 1.5.6 (ADT, S3 http://mgltools.scripps.edu/) (Trott and Olson, 2010), whereas all torsion angles for each enantiomer of DFMO as well as for the respective DFMO-analogue were considered flexible, and all the possible rotatable bonds (8 out of 32 for DFMO enantiomers and 10 out of 32 for the DFMO-analogue), and non-polar hydrogen atoms were determined. The final ligands’ files were saved as PDBQT files (.pdbqt format), and were ready for docking procedure.

The crystallographic structure of human ornithine decarboxylase (h-ODC, PDB code: 2OO0) (Dufe et al., 2007) of the highest known resolution (1.90 Å) was downloaded from Brookhaven RCSB Protein Data Bank (PDB database, http://www.rcsb.org/pdb/). To avoid steric clashes within the model the crude target protein .pdb file was prepared by UCSF Chimera vs. 1.11.2 package (http://www.cgl.ucsf.edu/chimera/) (Pettersen et al., 2004) by removing all nonstandard molecules including conserved crystal water molecules (HOH), cofactor [pyridoxal-5′-phosphate (PLP)] and other small non-protein ligands [i.e., pentane-1,5-diamine (cadaverine, N2P), 3-aminooxy-1-aminopropane (APA; with ligand code: XAP), acetate ion (ACT)], respectively. For comparison of docking protocols, the protein with preserved PLP in active cavity was also prepared. Next, the polar hydrogen atoms were added and Gasteiger charges were calculated with AutoDock Tools 1.5.6 package using its standard utility scripts, and then the final protein file was saved as a PDBQT file. Next, a searching “grid box” was set by using AutoGrid function to perform docking in a (40 × 40 × 40 Å)-unit grid box (final size space dimensions x = 40 Å, y = 40 Å, z = 40 Å), centered on the catalytic cavity located in one of the subunits (namely “chain A”) of the homodimeric h-ODC as target coordinate (center_x = −33.462; center_y = 18.513; center_z = −23.794) with a grid spacing of 0.325 Å.

At first, to validate the molecular docking protocols re-docking of the co-crystallized small molecule inhibitor, that is 1-amino-oxy-3-aminopropane (APA), into h-ODC (PDB access code: 2OO0) protein was performed. The docked complexes were superimposed to the original crystal structure to calculate the values of the root mean square deviation (rmsd). The re-docking of 2OO0 target protein structure and APA ligand reproduces the original pose with 1.48 and 2.73 Å rmsd, respectively. The average rmsd-value reaching 2.11 Å indicates that our docking methodology is adequate and can be utilized to search small molecule inhibitors of h-ODC. The overlaps between the docked pose of APA in h-ODC with retained PLP cofactor and the X-ray structure for h-ODC-APA complex are shown in Supplementary Figure S3 (Supplementary Information). Docking was performed into a rigid protein as well as using advanced protein flexibility by specifying flexible side chains. Each docking was performed with an exhaustiveness level of 48 concerning global search. For each ligand molecule 100 independent runs were performed using the Lamarckian Genetic Algorithm (GA) with at most 106 energy evaluations and a maximum number of generations of >27,000 Å³ (the search space volume). The rest of the docking parameters including the remaining Lamarckian GA parameters were set as default using the standard values for genetic Vina algorithms (the posed dockings were below 5.00 Å rmsd). The docking modes of each ligand, that is: L-DFMO, D-DFMO, and DFMO-analogue, were clustered and ranked based on a mutual ligand–protein affinity expressed as absolute free binding energies [ΔG _calc (kcal/mol)] as well as the rmsd-values in both modes regarding rmsd lower bound (l.b.), and rmsd upper bound (u.b.), respectively. The rmsd were computed referring to the input structure submitted to docking simulations. For h-ODC without PLP-cofactor the used random seed amounted to: (i) –324942292 for L-DFMO, (ii) –595761544 for D-DFMO, and (iii) –2089770112 for the DFMO-analogue; whereas for h-ODC with PLP-cofactor the used random seed amounted to: (i) +1194314488 for L-DFMO, (ii) –844931584 for D-DFMO, and (iii) –1490556468 for the DFMO-analogue. The best nine poses (modes) were selected according to AutoDock Vina scoring functions mainly based on binding energies and show mutual ligand–protein affinity (kcal/mol). For the protein with the cofactor in its structure, except considering binding energies, also the proximity of amine moiety of the halogenated ligands to electrophilic carbon atom of the formyl group of PLP was examined. Each binding mode was manually inspected to select only those conformations of the ligand molecule, which accommodated in the h-OCD catalytic cavity in the highest possible proximity to the L-ornithine substrate-binding site according to the crystal structure of h-ODC co-crystallized with both the PLP-cofactor and the APA inhibitor deposited as PDB: 2OO0. The results of docking scoring of the respective ligands to h-OCD without or with the PLP-cofactor are collected in Supplementary Tables S2–S8 (Supplementary Material), respectively. Additional data including the most important amino acid residues of h-ODC (PDB: 2OO0) involved in a hydrogen bond to the respective ligands as well as the results of the measurements of dihedral angels of L-DFMO and its analogue (9) docked to h-ODC (PDB: 2OO0) are available in Supplementary Tables S9–S13. The results generated by AutoDock Vina including optimized binding poses of all ligands in hypothetical complexes with the h-ODC protein as well as critical polar contacts between the respective atoms of those ligands and the receptor molecule (h-ODC, 2OO0) were visualized using the PyMOL Molecular Graphics System software, version 1.3, Schrödinger, LLC (https://www.pymol.org/). Visualization of ligands interaction diagrams were performed using Maestro Version 12.6.144, MMshare Version 5.2.144, Release 2020-4, Platform Windows-x64 for academics of the Schrödinger suite (Supplementary Figure S4). For comparison the protein-ligand interactions of the docked model were also analyzed by using freeware for academia BIOVIA Discovery Studio Visualizer 20.1.0.19295 software (Dassault Systèmes Biovia Corp.; https://www.3ds.com) (Supplementary Figure S5).

The title compound 1 crystallizes in the orthorhombic system, in the noncentrosymmetric Pca2₁ space group with unit cell parameters a = 10.41500 (10), b = 8.96290 (10), c = 10.70340 (10) Å. Crystal data as well as refinement parameters are summarized in Supplementary Table S1. The perspective view of the molecular structure of 1 with the non-hydrogen atom labeling scheme is presented in Figure 2. As depicted in the diagram, the asymmetric unit of 1 consists of one independent DFMO cation, a chloride ion and one water molecule. Bond length and angles exhibit normal values for ornithine derivatives (Cambridge Structural Database, Version 5.41) (Allen, 2002; Groom et al., 2016). Selected bond lengths and angles are listed in Supplementary Table S14, while geometrical parameters of H-bonds in Supplementary Table S15. The structure is stabilized by intra- and intermolecular either classical or non-classical O-H^…O(Cl), N-H^…O(Cl/F), C-H^…O hydrogen bonds between each of DFMO, Cl⁻ and water moieties, in the range of 2.6775 – 3.4213 Å. All H-atoms of both NH₃ ⁺ groups and O atoms of COO⁻ group are involved in the H-bonding as donors and acceptors, respectively.

The geometry optimization calculations were carried out using the hybrid meta exchange-correlation functional M06, which describes the molecular structure of organic compounds with high accuracy (Jacquemin et al., 2010; Zhao and Truhlar, 2011). Harmonic vibrational frequencies calculations of the optimized conformers computed in both surroundings confirmed that all of them correspond to minima of their potential energy surfaces. The initial geometries for the DFT geometry optimizations were obtained from the X-ray structure of the complex of this compound bound to human arginase I (Ilies et al., 2011). Because X-ray crystal structures of human arginase I complexed with the DFMO reveal the exclusive binding of the L-stereoisomer, this stereoisomer was considered in our calculations only. DFMO, when bound on the enzyme, is present in the zwitterionic form with basic N-amino group protonated (Ilies et al., 2011) (Supplementary Figure S6). The calculations were carried out for both, neutral DFMO molecules and its protonated cation at the terminal basic nitrogen atom N2. The important structural parameters are shown in Supplementary Table S16.

The M06 optimized structure of DFMO molecule in the both, gas-phase and hydrated state corresponds to the structure with the linear arrangement of the flexible C-C-C-C-C-C-N chain. The all-trans configuration is manifested by dihedral angles α, β and γ which are close to 180°degrees. The solvent affects the equilibrium structure only slightly (Supplementary Table S16).

The structure of neutral DFMO conformer resulted from the DFT calculation fits well with the experimental species of DFMO when bound at the human arginase I receptor. The DFMO molecule at this receptor is slightly bent, dihedral angle β[C(1)-C(4)-C(5)-C(6)] is of 146° and the C-C-C-C-N chain possess anticlinal conformation. The molecular superposition of the X-ray structure of biologically active conformation of DFMO (3gn0.pdb) and the corresponding thermodynamically stable structure isolated molecule is displayed in Supplementary Figure S7. Thus, the biological environment changes the molecular structure of this drug only slightly.

In clinical practice, eflornithine hydrochloride is used. We prepared suitable crystals of DFMO in the form of the monohydrate of eflornithine hydrochloride, 1. In the crystal state DFMO and HCl form a 1:1:1 complex with water. The hydrochloride acid forms an ion pair of the Cl⁻ ··· ⁺H₃N type with the basic terminal amino group of DFMO. The chloride anion in the crystal structure is coordinated with the basic center of DFMO by means of an amino group of another molecule. Crystal water contributes to the stabilization of the molecular crystal via the system of intermolecular hydrogen bonds.

The X-ray structure of the 2-(difluoromethyl) ornithine hydrochloride hydrate, 1, was used as a starting geometry in our DFT calculations. However, in the absence of intermolecular crystal forces, the complex coordination of chloride anions and water molecules upon geometry optimization of a single 2-(difluoromethyl) ornithine hydrochloride hydrate dramatically changes its geometry. Thus, the geometry optimizations resulted in three different structures denoted as A, B and C, respectively (Figure 3). The complex A mimics the geometry of DFMO in the solid state. The amino (–NH₂) and carboxyl (–COOH) functional groups of this complex form a zwitterion and the Cl- ion is coordinated to both cationic (–NH₃ ⁺) groups. Water forms the hydrogen bond with the anionic carboxyl moiety.

Different starting geometries of 2-(difluoromethyl)ornithine hydrochloride hydrate resulted in two additional stable complexes B and C, Figure 3. The amino (–NH₂) and carboxyl (–COOH) functional groups in these systems are neutral and stabilized via the intramolecular hydrogen bond of the N-H···O=C type. The water molecule takes part on the hydrogen bonds acting both as proton donor and proton acceptor. In the complex B the chloride anion is coordinated by the carboxyl group of the parent molecule by means of the water molecule.

The relative stability of A, B and C structures of 2-(difluoromethyl)ornithine hydrochloride hydrate 1 is presented in Supplementary Table S17. Based on the relative Gibbs energies the structure B is the most stable species (Figure 3) both in the gas-phase and in the aqueous environment. However, the differences in relative stability of these species are low, and in the gas-phase all three complexes may coexist with almost equal probability. The aqueous environment clearly prefers the structure B. In the crystal structure of 2-(difluoromethyl)ornithine hydrochloride hydrate the entire molecule is nearly planar and close to the conformation of the thermodynamically stable structure B, Supplementary Table S17.

The enrichment ratios (ER) of the intermolecular interactions in the analyzed crystal structures were calculated on the HS methodology (Jelsch et al., 2014). Results are collected in Supplementary Table S18. Privileged close inter-contacts are as follows:

In 1: O^…H/H^…O, C^…H/H^…C, Cl^…H/H^…Cl, F^…H/H^…F

In 2: O^…H/H^…O, C^…H/H^…C, Cl^…H/H^…Cl

The highest values of ER are evaluated for the interactions: O^…H/H^…O (∼1.5) and Cl^…H/H^…Cl (1.5) in 1, O^…H/H^…O, C^…H/H^…C and Cl^…H/H^…Cl at the level of 1.4 in 2. The H^…H contacts are moderately enriched (0.85) despite their highest contribution into the HS in 1 and 2. The contacts F^…O/O^…F in 1 are impoverished, which is consistent with the HS analysis. Furthermore, the C^…H/H^…C and F^…H/H^…F ones have values of 1.35 in 1. However, C^…H/H^…C are an insignificant contributor in HS of 1.

The energies of intermolecular interactions were investigated (Supplementary Table S19) and the 3D topology of the crystal packing was visualized graphically via energy frameworks (Figure 6) to better understand the packing of the crystal structure and the supramolecular rearrangement in all space directions. The width of the cylinders (a size of 100) shows the relative strength of the energy between the molecules. The cylinders with the largest radius demonstrate the highest energy values. The red, green and dark blue tubes denote the electrostatic, dispersive and total energy, respectively. The molecular interactions in relation to the central (reference) molecule in a cluster of radius 3.8 Ǻ are shown in Supplementary Figure S10 (Supplementary Table S20). The molecular pairs are uniquely color-coded. The energies of the molecular pair-wise interactions were used to evaluate the net interaction energies. Generally, the intermolecular energy analysis revealed a significant contribution of the electrostatic energy term, related to the strong classical interactions, in the stabilization of the crystal lattice of both investigated systems. Nevertheless, in 1, additional weak interactions are observed, which should not be overlooked. More specifically, the total energies include the electrostatic term, namely E` <sub>ele</sub> = -134/-359.1 kJ/mol, the polarization term E` _pol = -69.5/-134.2 kJ/mol, the dispersion term E` <sub>dis</sub> = -124.7/-106.5 kJ/mol, the repulsion term E' <sub>rep</sub> = 161.9/370.5 kJ/mol and the total energy E` _tot = -201.8/-342.8 kJ/mol, for 1 and 2, respectively. In the light of these findings, in 1, the difference between the electrostatic and dispersion terms (responsible for weak interactions) is insignificant. In consequence, the topology of the electrostatic and dispersive energy terms in the systems 1 and 2 are quite different.

Supramolecular chemistry focuses on non-covalent molecular recognition resulting in supermolecules (Lehn, 1995). The latter are defined as the association of two or more chemical moieties into well-organized entities of higher complexity (Dunitz, 1991). Non-covalent interactions are common in a protein environment. Hydrogen bonding interactions (apart from hydrophobic interactions), and N-H^…O H-bonds as more frequent than O-H^…O, are the most popular in protein–ligand complexes (Ferreira de Freitas and Schapira, 2017). Therefore, we can say that bio-complexes are supramolecular bio-systems. The structure of the supermolecule can be defined by supramolecular synthons – structural units formed by synthetic operations requiring non-covalent interactions (Desiraju, 1995). In other words, synthon is a recognition unit based on chemical functionality. Homosynthon exists between the same functional groups, while heterosynthon – between two different functionalities (Vishweshwar et al., 2003; Walsh et al., 2003). The Etter’s H-bond rules (based on electrostatic potentials) were useful to determine donor-acceptor interactions leading to synthons (Etter et al., 1990). It is important to point out that the supramolecular synthon concept is promising in the future design of idealized ligands with effective binding inside the protein pockets via matching synthonic functionalities (from corresponding libraries) to the model ligands (Spackman et al., 2019). In this context, the potential transferability of supramolecular synthons from small-molecules to macromolecular bio-systems should be kept in mind (Bissantz et al., 2010; Groom and Cole, 2017; Bojarska et al., 2018a). Notably, supramolecular aspects of amino acids, essential components of proteins, are of prime importance in terms of the drug design and development.

Here, we focus on the survey of synthons driving the self-assembly, their topological properties in the supramolecular landscape in the title compound 1, in comparison with the ornithine supermolecule, 2. We report supramolecular relationships and synthon preferences resulting from the interplay of strong and weak interactions. Overall, ubiquitous linear and bifurcated H-bonds are formed in analyzed supramolecular systems. All types of interactions formed by amine and carboxyl groups, such as _(H2O)O-H^…O_(COO-), _(NH3+)N-H^…O_(COO-/H2O), _(CH2)C-H^…O_(COO-/H2O), but also those containing Cl⁻ and H₂O moieties, such as _(NH3+)N-H^…Cl⁻ _(CH2)C-H^…Cl⁻ and _(H2O)O-H^…Cl⁻, are presented in Figure 7. Notably, the amino-carboxy synthons are most important in the protein environment. An interesting additional issue may be the structural arrangement (motif) involving the NH₃ group, which interacts with the carbonyl group, H₂O and/or Cl⁻ entities.

In the title crystal 1, S(6) and R ² ₂(7) H-bonding motifs formed by oxygen atoms of COO⁻ group stabilize the structure (Supplementary Figure S11). The primary supramolecular H-bonding interactions are classical strong O(N)-H^…O=O and weak C-H^…O(F, Cl) and N-H^…Cl H-bonds. To be precise, at the first level of the supramolecular architecture, intramolecular cyclic and intermolecular linear H-bond patterns are sustained, namely: S(6) by _(CH2)C-H^…O=C_(COO), C(5) by _(NH3+)N-H^…O=C_(COO-), _(CH)C-H^…O=C_(COO-) or _(CH2)C-H^…F, C(7) by _(CH)C-H^…F and D(2) by _(NH3+)N-H^…O_(H2O), _(H2O)O-H^…O=O_(COO-), _(CH)C-H^…O_(H2O), _(NH3+)N-H^…Cl⁻ or_(CH2)C-H^…Cl⁻ (Supplementary Figure S12). H₂O molecules play supporting roles of bridging DFMO molecules and participate in the formation of supramolecular H-bonding patterns, for example they build the water-linear chain with the C ² ₂(9) H-bond descriptor, parallel to the b axis, (Figure 8). At the second level, diverse types of synthons are observed, including both cyclic and linear homo- and heterosynthons. The latter are dominant. The library of all H-bonding motifs found in the studied crystals is included in Supplementary Table S21. In particular, bifurcated H-bonds are observed (Figure 9). This type of interactions proposed by Albrecht and Corey (in glycine) (Albrecht and Corey, 1939) is represented by the H-bonding pattern, including more than one acceptor/donor (one H-bond donor is bound to two H-bond acceptors, and vice versa – two H-bond donors are bound to a single H-bond acceptor). Notably, bifurcated H-bonds are energetically favorable (Feldblum and Arkin, 2014). COO⁻, F, H₂O, Cl⁻ moieties participate in the formation of supramolecular motifs as a bifurcated acceptor, while NH₃ ⁺ CH(CH₂) as a bifurcated donor. Additionally, it should be mentioned that the water molecule plays an ambivalent role, both of a donor and an acceptor.

It is well known that supramolecular interactions play crucial roles in bio-systems, inter alia protein stabilization, molecular recognition, efficiency/specificity of enzymatic reactions (Frieden, 1975; Alberts et al., 1994). A bifurcation (and multi-furcation) is common in bio-systems. The H-bonds between the protein (target) and the ligand (drug) build unique H-bonding patterns, which can be described by synthon descriptors (Sarkhel and Desiraju, 2004; Panigrahi, 2008). In the context of this work, we observed that the bio-complex of arginase with DFMO displays a similar tendency to build the same synthons, such as the inter alia discrete chain D(2) created by the interactions between amino and carboxyl groups (N-H^…O-C) (Figure 10) (Ilies et al., 2011).

Due to the relevance of fluorine in bio-medicine, we also directed our attention to the identification of the structure-determining fluorine-based H-bonding patterns and careful topological analysis of the involvement of interactions containing fluorine in control of the supramolecular systems. The findings could be valuable in terms of the design of future innovative drugs as well as the advancement of the supramolecular chemistry of short peptides.

In 1, the contribution of C-H^…F interactions (distance below ∼2.6 Ǻ and angle above 120°) at the level of ca. 20% (based on the HS study) is noteworthy. Therefore, stabilization of the structure by defined fluorine-based supramolecular motifs cannot be neglected. Hirshfeld surfaces presenting C-H^…F patterns are depicted in Supplementary Figure S13 . We highlight the effect of numerous H-bonding motifs containing fluorine in relation to the role of diverse/subsequent interactions and their cooperativity in the construction of supramolecular architecture of 1. The fluorine atom participates through C-H^…F-C interactions in the formation of H-bonding patterns, both at the first and second levels of the supramolecular structure. In addition, auxiliary either strong or weak (N/C)O-H^…O and (C)N-H^…Cl interactions are involved in the formation of fluorine-based synthons at the second level (Supplementary Table S22). 5-membered and 7-membered chains formed by C-H^…F interactions only are the principal motifs in the studied system. The bifurcated F-atom participates in the building of only the R ¹ ₂(6) motif. The structural variety of homo- and heterosynthons, grouped into the rings, chains and discrete finite chains D-synthons, are presented in Figure 11. DFMO molecules are assembled into cyclic ring-shaped motifs through the dimer or catemer synthons. On the other hand, the dimer, trimer and pentamer of linear – chain synthons are observed. In the case of discrete chains – D-type motifs, we encountered only tetramer (catemer) units, but the co-operation of solvents such as water and Cl⁻ moieties in this kind of H-bonding motifs should be mentioned. In chain and ring types of H-bond patterns, homosynthons, apart from heterosynthons, are visible.

There are three chemical entities in the crystal of 1: the DFMO cation, the chloride anion, and the water molecule. Both the cation and water molecule act as proton donors or acceptors while the anion is only the acceptor participating in charge-assisted hydrogen bonds. Taking into account the dimensionality of the synthons described at the first level of the graph theory, the smaller species participate only in 0D (definite, D) motifs. In turn, the cations interact via N-H^…O hydrogen bonding forming the first-level C(5) motif running along the [001] direction and joining molecules related by the c-glide plane. According to Energy Frameworks calculations (Figure 6), this is the most significant interaction in the crystal, and thus might be regarded as a Large Synthon Aufbau Module (LSAM), (Desiraju, 1995; Ganguly and Desiraju, 2010). The symmetry of this supramolecular chain can be described with the pc11 rod group symmetry (International Tables for Crystallography, 2006). In a crystal, each chain is surrounded by eight others, but at three different distances. The lack of close-to-hexagonal packing indicates the presence of other important interactions between the chains. Indeed, the plethora of weak interactions of C-H^…O and C-H^…F type are observed in the [100] direction, joining the cations related by the a-glide plane into a zigzag chain. As a result, wavy layers perpendicular to (010) planes are observed, and thus extending the cationic LSAM into two dimensions with the pba2 layer group symmetry. Interestingly, the shortest interactions of the cation with water molecules, resulting in the C ² ₂(7) chain motif along the 2 ₁ screw axis, might be treated as reinforcement of the 2D structure (Figure 12A,B). Between the layers, the anions are located, interacting via H-bonds with both the cation and solvent. The cationic layers are also joined into the 3D structure by the longest N-H^…O hydrogen bonding with water molecules (thin lines in Figure 12B).

In the case of crystal 2 each cation interacts via N-H^…O with four others. The most important supramolecular entity is a chain of rings propagating along the 2 ₁ screw axis consisting of two types of H-bonds. The chains related by translation in [001] direction are joined by remaining N-H^…O hydrogen bonds, and resulting in a cationic square-grid undulated layer on the plane (100) (Figure 12C). The layer exhibits the p12 ₁1 layer group symmetry. Taking into account the calculated energies for intermolecular contacts in the crystal (Figure 6; Supplementary Table S19), the chains of the p2 ₁ rod group symmetry might be treated as LSAMs. However, they do not pack hexagonally, which in this case might result from both the additional cation-cation hydrogen bonds as well as relatively short hydrogen bonds with chloride anions. The anions are located on both sides of the layers joining them into a 3D structure (Figure 12D). The enhancement of these N-H^.Cl interactions compared to the structure 1 might be due to the lack of solvent molecules in the crystal.

DFMO exerts its action within cells, where the cell machinery of a polyamine synthesis is located. The drug, however, is very hydrophilic, and unable to cross lipid cell membranes. This is also reflected in its inability to cross the blood-brain barrier, as pointed out by Levin (Levin and Ellingson, 2018), and this difficulty in reaching the primary site of action may well reduce the efficacy of the drug, and prevent it from accessing certain target tissues where diseases amenable to its actions are manifested. To address this problem, an investigation looking at the physical characteristics of a range of analogues was conducted, to determine whether any of these could alter the properties of the molecule sufficiently to enhance its membrane permeability. The requirement for the analogue to fit into the confined space of the ligand-binding pocket of ODC, combined with the need to increase the hydrophobicity of the molecule, led to the use of the fluorine atom to substitute for hydrogen atoms in DFMO; of all the common atoms forming covalent bonds with carbon, fluorine is the smallest and most hydrophobic. Further details on properties of fluorine in the parent molecule have already been provided in the introduction. The on-line tool SwissADME and pkcsm were employed to calculate a range of parameters for each analogue, evaluating their hydrophilicity/hydrophobicity balance, and predicting the potential for passive transport across the BBB, and across lipid membranes in general. Results from this analysis are shown in Supplementary Table S23.

As can be seen, according to the criterion of the BBB permeability – one of the parameters specifically singled out by SwissADME, neither ornithine nor DFMO are able to cross membrane barriers, either in the charge or the neutral form (lines 1 – 4). Introduction into DFMO of additional methyl groups and/or fluorine atoms (both hydrophobic moieties) resulted in an increase in lipophilicity and a decrease in hydrophilicity relative to the parent compound, and these molecules (5–9) were considered to be able to cross the BBB by SwissADME. However, examination of the interactions of DFMO with the human arginase binding site in the X-ray crystal structure 3gn0 (Ilies et al., 2011) showed that the carboxyl group in DFMO plays a key role in binding, and it was feared that conversion to the amide, as in analogues 6 and 7, may interfere with this binding interaction. Consequently, these analogues were discounted as potential candidates for effective inhibition, and the analogue 9 was chosen for further study since this showed the greatest reduction in hydrophilicity.

In order to rationalize pharmacological activity of DFMO, 1, and its novel analogue (9), as potent inhibitors of human ornithine decarboxylase (h-ODC), an in silico enzyme−substrate docking protocol using non-commercial AutoDock Vina software (v1.1.2) (Trott and Olson, 2010) was applied. In shortening, before docking procedure was performed, the respective crystal structure of h-ODC protein was taken from the Protein Data Bank (PDB) with the 2OO0 code (Dufe et al., 2007) and used as a receptor after initial processing of the crude .pdb file by removing all non-protein molecules (i.e., conserved water molecules, cofactor, unique ligands, and ions) and addition of polar hydrogen atoms to avoid steric clashes within the model (see Experimental section for details). Docking was performed using the centered grid-box coordinates only at the A subunit since both monomers of h-ODC share the same residues and the similar conformation. In turn, ligand molecules were optimized in terms of geometry and further prepared for the docking procedure as PDBQT files by using the Python script “prepare_ligand4.py” that comes with AutoDock Tools vs. 1.5.6 (ADT, S3 http://mgltools.scripps.edu/) (Trott and Olson, 2010), and which included the optimization of the rotatable bonds network among the respective ligand molecule. Since DFT analyses performed by us revealed that the biological environment changes the molecular structure of DFMO in a negligible extent; we decided to use neutral (non-ionizable) DFMO-based molecules in all docking calculations.

Importantly, the reason why we decided to choose h-ODC deposited in PDB as 2OO0 and not the apo-enzyme (PDB: 1D7K; refined to 2.10 Å resolution) (Almrud et al., 2000) or its complex with small ligands (PDB: 2ON3; refined to 3.00 Å resolution) (Dufe et al., 2007) is two-fold: the crystal structure of 2OO0 protein is of the highest known resolution (1.90 Å) and it was additionally co-crystalized with 1-amino-oxy-3-aminopropane (APA; with ligand code: XAP), which is an isosteric analogue of putrescine and a significantly more potent inhibitor of h-ODC than DFMO with inhibition constants in the nM-range (K _i = 10−25 nM) (Khomutov et al., 1985; Stanek et al., 1992; Milovica et al., 2001; Seiler, 2003). The APA molecule complexed with h-ODC is bound in the substrate-binding pocket in close proximity to the h-ODC functional cofactor, namely pyridoxal-5′-phosphate (PLP). Thanks to a high-resolution X-ray structure of the complex between APA and h-ODC we could easily identify, which of the amino acids contribute to each active site of a dimeric enzyme as well as which of the residues play an important role for both PLP binding and interactions with APA. Moreover, the validation of the docking protocol was performed by re-docking of APA inhibitor into h-ODC receptor and by comparing the binding pose of the most energetically stable protein-ligand complex with the co-crystallized 2OO0 enzyme-APA inhibitor structure. As both the X-ray structure and docked structure overlapped within a similar space inside the receptor and at a root mean square deviation (rmsd) value of 2.11 Å our docking methodology was considered as reliable for binding simulations of the rest of the studied ligands. Only then, we set the docking calculations with the grid box centered on the substrate-binding site at the coordinates retained as close to those used for APA (see Experimental section for details). To maintain accurate simulations of the ligand−receptor interactions, docking was performed at significantly increased exhaustiveness of 48 with the search space volume >27,000 Å³.

As a result of docking calculations nine of the most energetically favourable binding modes being the relevant docking poses for the ligand−protein complexes for each substrate molecule (L-DFMO, D-DFMO, DFMO-analogue) and target 2OO0 protein were generated, and the results of their binding affinity energies expressed as ΔG _calc (kcal/mol) are given in Supplementary Tables S3–S8 (Supplementary Material). The visualization of the representative binding modes of L-DFMO, D-DFMO, and the DFMO-analogue to h-ODC with close contacts to residues in the active site is presented in Figure 13. The inspection of the productive pose of L-DFMO showed that the inhibitor possesses close polar contacts between both the amino groups and the side chains of Tyr323 (B chain) and Asp361 (B chain), while its carboxyl moiety interacted with Asp332, Arg277 and Ser200, respectively. In turn, D-DFMO is stabilized by strong 2.1–2.5 Å-long polar interactions between both amine moieties and the backbone atom of Asn385, Glu384 and Asp332 residues, whereas the carboxylic acid group forms H-bonds with Arg277 and/or Ala388. Complexes of h-ODC and the selected top-scoring pose of the DFMO-analogue with interacting amino acid residues revealed that primary amine group forms the hydrogen bonding with Ala388 and Asp332; secondary amine functionality interacts with Asn385, while carboxyl group is H-bounded with His335. For more details see also Supplementary Figures S14–S17 (Supplementary Material).

Validation of molecular docking of both enantiomers of the model DFMO inhibitor yielded minimum value of –4.7 kcal/mol for L-DFMO, with average ΔG _calc = –4.50 kcal/mol, while D-DFMO docking resulted in the minimum value of –4.8 kcal/mol, with average ΔG _calc = –4.54 kcal/mol. Of course, the lower the binding energy, the more suitable the binding of the inhibitor molecule to the receptor. Remarkably, L-DFMO underwent six hydrogen bonds which represents the highest count in our docking studies using all three inhibitors and h-ODC, respectively (Figures 13A,B). In turn, it is worth noting that the developed DFMO-analogue presents docking scores ranging from –4.8 to –5.1 kcal/mol with the average value for ΔG _calc reaching –4.91 kcal/mol (Supplementary Table S5, Supplementary Material). These results suggest as the DFMO-analogue might be slightly more potent than both DFMO enantiomers in terms of inhibition of h-ODC catalytic activity. However, when comparing only the values of the lowest energies of the formation of the ligand-receptor complexes without any critical inspection of all the generated poses and general inhibitory mechanism of h-ODC by derivatives of amino acids, one could conclude that the D-stereoisomer of DFMO is more efficient in inhibition of a catalytic activity of h-ODC than its L-DFMO counterpart. Of course, this is in strong contrary with inhibitory effects of DFMO antipodes on h-ODC established during in vitro tests. Considering that the inhibitor dissociation constant (K _D) values for the formation of h-ODC–DFMO complexes are at 1.3 FM0.3 μM conc. for L-DFMO and 28.3 DF3.4 μM conc. for D-DFMO (Qu et al., 2003), the inhibitory potency toward h-ODC is ca. 20 times greater for L-DFMO when compared with D-DFMO. Therefore, further evaluation of the binding poses was performed to select only productive conformations of the ligand molecules, which can form covalent coenzyme-substrate adducts that cease the catalytic activity of PLP-dependent ODC. In this regard, the second series of docking calculations were performed with h-ODC containing the PLP-cofactor in the catalytic cavity to include possibility of the formation of the phosphopyridoxyl-amino acids intermediate with studied ligands according to one of the postulated mechanisms of ODC-inhibition by amino acid-like substrates (Wu et al., 2007; Correa-Basurto et al., 2008; Preeti et al., 2013). Details of the contacts between the halogenated compounds and homodimeric form of human ornithine decarboxylase in the most energetically stable binding pose including the hydrogen-bonding network connecting the studied inhibitors and the h-ODC active site are depicted in Figure 14.

In this case, docking results revealed completely different binding affinity for both DFMO enantiomers in favor of L-DFMO, which obtained the lowest energies of the respective ligand-receptor complex formation for h-ODC [ranging from –4.7 to –5.4 kcal/mol with average values of ΔG _calc = –4.96 kcal/mol (Supplementary Table S6 Supplementary Material)]. These results are also impressive in terms of binding affinity when compared to nanomolar APA inhibitor, which interactions inside the substrate-binding cleft of h-ODC complexed with PLP revealed binding energy of only −3.8 kcal/mol for the best mode inspected (Supplementary Table S2, Supplementary Material). What is even more important, hypothetical formation of a Schiff base of PLP with the α-amino group of the fluorinated substrates might be achievable only in the case of L-DFMO located significantly closer to the cofactor (6.2 Å) when compared with two other studied inhibitors (>14.3 Å). Moreover, L-DFMO was stabilized by multiply strong H-bonds of 2.6–3.1 Å distance formed between the carboxyl moiety of the ligand and the hydrogen atom of the –OH group of the Ser200 residue from one side and the guanidine group of Arg277 as well as –COOH of Asp322 from the other side. However, the interactions between Asp322 and L-DFMO are rather unfavorable since acceptor-acceptor clashes are observed in this case (Figure 15). In turn, the α-amino group of L-DFMO forms a 2.7 Å-long hydrogen bond with the carboxylic acid group of the Asp361 (B) side chain and/or Tyr331, whereas the terminal δ-NH₂ group interacts with the oxygen atom of the hydroxyl moiety of Tyr323 (B) in the close distance of 2.6 Å. In addition, single π–σ contact between methylene group of the alkyl chain and Tyr389 was detected. Interestingly, only in the case of L-DFMO two extra interfacial residues, namely Asp361 (B), Tyr323 (B), and Cys360 (B), that belong to the B subunit of the dimeric enzyme, participate in the stabilization of the ligand molecule in the complex with h-ODC (Figures 14A,B). In the case of L-DFMO additional halogen bonding interactions between oxygen atom of carboxylic group belonging to Asp361 (B) residue and fluorine atoms of α-difluoromethyl group was detected (Figure 15).

In sharp contrast, although D-DFMO essentially occupies the position of the substrate in the active site interacting with the crucial Arg277 residue, which is fundamental for the h-ODC activity and overall catalysis, it is positioned far away from the cofactor location and the formation of an aldimine intermediate with PLP is obviously not possible (Figures 14C,D). Other potential polar interactions between the less favored D-stereoisomer and h-ODC within a radius of 2.4–2.5 Å involve Ser241 and Ala388. In the case of D-DFMO halogen bonding was detected between α-difluoromethyl group of the ligand and Asp332 residue (Figure 15). More specifically, the optimized binding mode of the bulky DFMO-analogue revealed that the ligand is also pulled far away from the PLP location, thus accommodating in the larger hydrophilic cavity surrounded by the subsequent aliphatic amino acids: Gly240, Ser241, Lys337, Val244, Ala281, Ala334, and Asn385 (Figures 14E,F). Furthermore, docking investigations revealed that additional binding interactions via fluoromethyl group present in DFMO-analogue was observed as halogen bonds between F atom and Asp243 was observed in this case (Figure 15). It turned out that such functionalization of the α-amine group in DFMO-analogue has significantly changed the way in which this ligand is accommodated in the h-ODC active site. In contrary to L-DFMO molecule its analogue lost 2.7 Å-long H-bonding with Asp361(B) and thus is pulled far away from the PLP functional cofactor. Instead of this DFMO-analogue gained a closer (and thus stronger) 2.0 Å-long hydrogen bond interactions with hydroxyl moiety of the Ser241 residue located in a different region of the binding pocket and few other halogen bonds with Ala388 and/or Asp332 (Figure 15). The main lesson to be drawn from the in silico data presented in this paper is that closely related compounds sharing DFMO scaffold can bind to the active site of h-ODC in substantially different manners as a consequence of subtle structural alterations in terms of fluorine atom redistribution. In this case the binding potency cannot be correlated in a straightforward manner just to the number and type of polar interactions observed, but other interactions including, i.e., van der Waals forces or steric clashes in protein structure, should also be included. Probably it is also the restricted size of the pocket nearby the PLP cofactor that forces the DFMO-analogue to adopt different conformation which is unable to enter deeply into this site. Therefore, the only reliable tool for rationalization of the predictions made by computational modeling is the experimental approach including enzymatic activity assay. Notable, the DFMO-analogue furnished promising results in terms of consistency of the formed complexes with h-ODC by spending vast majority of poses within the selected binding cavity, while both DFMO enantiomers occupied the enzyme active site only with maximum 2–4 poses from the best mode according to the calculated rmsd-values, respectively (Supplementary Tables S6–S8, Supplementary Material). Since rmsd is very important parameter that imparts the information of the overall stability of the protein-ligand complex in terms of calculating standard deviation from the initial structure, it is worth to note that for DFMO-analogue bound to target protein, the rmsd-value for the best pose was found to be 1.58 Å with the average rmsd of 3.10 Å suggesting its stability. Of course, to estimate which of the inhibitor forms the most stable complexes, the best arrangements of the ligands along with spatially optimized binding cavities of the enzyme molecule should be further subjected to a molecular dynamics simulation (Bitencourt-Ferreira and Filgueira de Azevedo, 2019; Bitencourt-Ferreira et al., 2020; Bitencourt-Ferreira et al., 2021). Moreover, to confirm if binding of the DFMO-analogue has potential physiological relevance, in vitro inhibitory evaluations with isolated enzyme must be performed. For more details on the formation of potential h-ODC-DFMO-analogue complexes see Supplementary Figures S15–S19 appended in Supplementary Material.