Exploring the Anticancer Potential of Premna resinosa (Hochst.) Leaf Surface Extract: Discovering New Diterpenes as Heat Shock Protein 70 (Hsp70) Binding Agents

Premna, a genus consisting of approximately 200 species, predominantly thrives in tropical and subtropical areas. Many of these species have been utilized in ethnopharmacology for diverse medicinal applications. In Saudi Arabia, Premna resinosa (Hochst.) Schauer (Lamiaceae) grows wildly, and its slightly viscid leaves are attributed to the production of leaf accession. In this study, we aimed to extract the surface accession from fresh leaves using dichloromethane to evaluate the anticancer potential. The plant exudate yielded two previously unknown labdane diterpenes, Premnaresone A and B, in addition to three already described congeners and four known flavonoids. The isolation process was accomplished using a combination of silica gel column chromatography and semi-preparative HPLC, the structures of which were identified by NMR and HRESIMS analyses and a comparison with the literature data of associated compounds. Furthermore, we employed a density functional theory (DFT)/NMR approach to suggest the relative configuration of different compounds. Consequently, we investigated the possibility of developing new chaperone inhibitors by subjecting diterpenes 1–5 to a Surface Plasmon Resonance-screening, based on the knowledge that oridonin, a diterpene, interacts with Heat Shock Protein 70 (Hsp70) 1A in cancer cells. Additionally, we studied the anti-proliferative activity of compounds 1–5 on human Jurkat (human T-cell lymphoma) and HeLa (epithelial carcinoma) cell lines, where diterpene 3 exhibited activity in Jurkat cell lines after 48 h, with an IC50 of 15.21 ± 1.0 µM. Molecular docking and dynamic simulations revealed a robust interaction between compound 3 and Hsp70 key residues.


Introduction
The Premna genus, belonging to the Lamiaceae family, encompasses approximately 200 species that are primarily distributed across tropical and subtropical regions of Asia, The structure of known compounds was characterized through NMR and mass spectrometry experiments and their comparison with literature data.
1 was observed as a white, amorphous solid. The molecular formula of C 20 16), which suggested that 1 is a labdane diterpenoid [29]. COSY correlations were monitored between the signals at δ H 5.43 (H-7) and δ H 2.05 and 1.96 (H 2 -6). The doublet multiplicity of the signal at δ H 3.30 (H 2 -14) and the absence of COSY cross peaks showed that the corresponding protons were placed near non-protonated carbons. The other connectivity of the diterpenoid skeleton and the other functional groups was established from its 1 H-1 H COSY, HSQC, and HMBC spectra (Table 1 and Figure 2).   The relative configuration of 1 was determined by studying 1 H NMR coupling constants and NOESY correlations ( Figure 3). The NOESY correlations between H-5 and H-9 and between Me-18 and Me-20 made it possible to hypothesize the relative stereochemistry of 1. Thus, the structure of 1 could be proposed as 12-oxo-7,13(16)-labdandien-15-oic acid, and this compound has been accorded the trivial name, premnone D. The relative configuration of 1 was determined by studying 1 H NMR coupling constants and NOESY correlations ( Figure 3). The NOESY correlations between H-5 and H-9 and between Me-18 and Me-20 made it possible to hypothesize the relative stereochemistry of 1. Thus, the structure of 1 could be proposed as 12-oxo-7,13(16)-labdandien-15-oic acid, and this compound has been accorded the trivial name, premnone D. The determination of the relative configuration of C-9 of 1 was performed through the approach based on the combination of QM/NMR [19,30,31]. Specifically, this methodology, developed and optimized by us [19,20], relies on the prediction of NMR parameters (e.g., 13 C and 1 H NMR chemical shift) through calculations based on the density functional theory (DFT). In more detail, the comparison of both predicted and experimental NMR parameters, through a statistical analysis, is the crucial point for suggesting the stereo- The determination of the relative configuration of C-9 of 1 was performed through the approach based on the combination of QM/NMR [19,30,31]. Specifically, this methodol- ogy, developed and optimized by us [19,20], relies on the prediction of NMR parameters (e.g., 13 C and 1 H NMR chemical shift) through calculations based on the density functional theory (DFT). In more detail, the comparison of both predicted and experimental NMR parameters, through a statistical analysis, is the crucial point for suggesting the stereochemical assignment of organic and natural compounds [19,20,30]. The first step involved a wide-ranging conformational search of all the possible investigated stereoisomers of 1 by applying empirical methods, i.e., Monte Carlo Molecular Mechanics (MCMM), Low-Mode Conformational Sampling (LMCS), and Molecular Dynamics (MD) simulations (see computational details, Experimental Section). Specifically, the relative configuration of C-5 and C-10 has been assigned as S* and S*, respectively, since, in the literature, molecules with this core are characterized by trans arrangements considering the H-5 and Me-10. For this reason, two stereoisomers (1a-1b, Figure S14, Supporting Information) were considered for the prediction of chemical shifts. Afterwards, the MM conformer ensembles represented the input for the subsequent steps, involving quantomechanical geometry and energy optimization at the MPW1PW91/6-31g(d) level of theory. Eventually, the resulting optimized conformers were analyzed by visual inspection to eliminate further redundant conformers. Then, considering the Boltzmann distribution of the selected conformers for each investigated stereoisomer, 13 C and 1 H NMR chemical shifts were computed at the MPW1PW91/6-31g(d,p) level for 1a-b. For all the DFT calculations, the simulation of the solvent methanol was performed using the integral equation formalism model (IEFPCM) [20,30,32].
Subsequently, to compare the calculated and experimental value, the statistical parameter mean absolute error (MAE) was used (see computational details, Experimental Section). Finally, 1b featured the lowest 13 C and 1 H MAE values, i.e., 2.4 and 0.11 ppm, respectively (Tables S1 and S2), suggesting 5S*, 9S*, and 10S* as relative configurations for 1. Moreover, to corroborate the obtained results, we also employed the DP4+ method [33], a robust tool for performing the correct stereochemical assignment of organic compounds. Accordingly, 1b showed the highest DP4+ probabilities (100.0%), confirming the relative configuration 5S*, 9S*, and 10S*.
The HRESIMS spectrum of 2 showed the [M+Na] + ion at m/z 355.2242, and 13 C NMR showed 21 carbon signals. The multiplicities of carbons determined by 13 C NMR data led to the attribution of four methyl groups, one methoxy group, seven methylenes, one of which sp 2 hybridized, three methines, and two quaternary carbons, allowing for the proposal of the molecular formula C 21 H 32 O 3 and a labdane diterpene. The NMR spectra (Table 1) of 2 were very similar to those of 1, except for the presence of an ester functionality at C-17. Therefore, 2 was characterized as 12-oxo-7,13(16)-labdandien-15-oic acid methyl ester.
Concerning 4, four stereoisomers (4a-4d, Tables S3 and S4, Supporting Information) were considered for the prediction of chemical shifts. In this case, 4a showed the lowest 13 C and 1 H MAE values (2.2 and 0.13 ppm, respectively). Accordingly, the DP4+ method was employed, highlighting 5S*, 9S*, 10S*, and 12R* as the relative configuration for 4 (DP4+ probabilities = 97.83%). It is worth noting that, through the application of the DFT/NMR approach for compounds 1 and 4, only the relative configurations of the investigated compounds were suggested.

Surface Plasmon Resonance (SPR)
Our previous results suggest that diterpenes could modulate the Hsp70 activity. 1-5 were studied by Surface Plasmon Resonance (SPR) experiments to investigate their interaction with this chaperone; oridonin, a well-known Hsp70 inhibitor, was used as a positive control. The interaction between each of the labdanes (1-5) and Hsp70 was investigated in this study by a surface plasmon resonance (SPR)-based binding assay [24,[37][38][39].
Only 3 interacted efficiently with the immobilized protein. As a result of fitting the relative sensorgrams to a single-site bimolecular interaction model, the thermodynamic parameters for the resulting complex formation were determined. This approach allowed for the measurement of 90.8 ± 9.5 nM K D for the Hsp70/3 complex. Interestingly, 3 showed a similar affinity towards the chaperone compared with that determined for oridonin (Table 2, Figure 4). Instead, 4 showed a lower affinity than 3 (in order of micromolar), with a value of 1000 ± 25 nM.   (Table 3).

Computational Analyses
Molecular docking calculation and molecular dynamics were applied to investigate the binding of Hsp70 and 3. The Protein Data Bank [42] entry 5AQX was used to represent the structure of the enzyme and to explore the binding of possible inhibitors at the ATP binding site of the protein [43].
Glide [44] was used for docking simulation, focusing on the ATP binding site of the protein; the analysis of binding poses reported a hydrogen bond network between Hsp70 and 3. Specifically, the ligand was bound with Thr13, Thr14, and Gly339 amino acid residues in the binding cavity ( Figure S15, Supporting Information). The selected ligand reported a predicted binding affinity value of −6.258 (kcal/mol; Glide XP), with Hsp70 [45]. Then, 100 ns of the molecular dynamic simulation were carried out to evaluate the stability of the complex during the whole simulation time, and the results can be visualized in the protein-ligand RMSD plot ( Figure S16A, Supporting Information). The in silico investigation highlighted the stability of the binding during the whole simulation. The "Lig fit Prot" plot shows that the RMSD of the ligand is lower than the RMSD of the protein, underlining that the ligand is stable in its initial binding site. The protein-ligand contacts over time provided interesting information about the binding behavior of the complex. Interactions of 3 with the key residues Thr13 (H-bonds and water bridges) and Ser240, Gly202, and Lys271 were detected. Gly202 has been reported to interact with the α-helix at the back of the ATP binding site [43]. Thr13 is close to Thr15, a critical residue known to stabilize the protein's close conformation [24,43]. The residues around Arg275 have been indicated as selective determinants in the initial ATP-binding. 3 showed several hydrogen-bonded protein-ligand interactions mediated by a water molecule, including Arg72, Glu175, Gly202, and Gly339 ( Figure S16B, Supporting Information).

Discussion
The literature data reported an increasing number of Hsp70 inhibitors, acting with different mechanisms analyzed using chemical and biological approaches (for example, surface plasmon resonance, chemical proteomics, computational details, interaction on HSP70 cochaperones that modulate the chaperone activity of the protein through their binding to functional domains of Hsp70, Hsp70 ATPase activity test, regulation of Hsp70 client proteins, citrate synthase aggregation assay, co-immunoprecipitation experiments) [26,[46][47][48][49][50][51][52][53][54]. The investigation of the protein-ligand complex structure and behavior is fundamental in the drug discovery process. In a previous study, the treatment of leukemia-derived Jurkat cells with oridonin inhibited Hsp70 [55]. 3 stably interacted with Hsp70 efficiently compared with oridonin, as confirmed by the SPR and molecular modeling studies. The SPR-based binding assay results in this study showed the efficient interaction of 3 with the Hsp70 protein at 91 nM K D , which was similar to the chaperone compared with that determined for oridonin (positive control). This result was supported by an in silico investigation, which showed that 3 was accurately placed into the Hsp70 ATP binding site. In addition, 100 ns of molecular dynamic simulation reported several crucial interactions involved in the stabilization of the studied complex. Most recently, the kaurane diterpene ent-7β-acetoxy-18-hydroxy-15α,16α-epossikaurane (epoxysiderol) has been reported as a modulator of Hsp70 1A activity [24]. Particularly, 3 reported a similar hydrogen bonds pattern of epoxysiderol, involving different key residues for developing Hsp70 inhibitors. Moreover, epoxysiderol reported a similar binding affinity value of 3, interacting in the same ATP binding site [24]. All these findings suggest the molecular explanation of the experimentally detected binding affinity, suggesting 3 as a new possible Hsp70 inhibitor.
Molecular docking simulations were also performed for 1, 2, 4, and 5, which cannot bind the active site of Hsp70, in accordance with the biophysical data for SPR experiments. Further studies will be performed to identify natural compounds targeting Hsp70 1A in cancer cells.
Plants represent a remarkable reservoir of biomolecules exhibiting different biological activities. However, one major challenge in phytochemical studies lies in the intricate process of isolating and characterizing the active constituents. Often, these phytoconstituents are obtained in minute quantities, which hinders further investigations. Nonetheless, these limitations can be overcome through interdisciplinary collaborations that focus on producing active constituents via plant tissue and cell culture techniques [56]. Alternatively, chemical synthesis offers a viable and accessible approach for synthesizing biomolecules with exceptional bioactivity [57]. Moreover, the utilization of sophisticated nano-formulations holds promise in achieving the enhanced targeting of tumoral masses [58,59].

General
An Atago AP-300 digital polarimeter with a 1 dm microcell and a sodium lamp (589 nm) was used to obtain optical rotations. A Bruker DRX-600 NMR spectrometer (Bruker BioSpin GmBH, Rheinstetten, Germany) equipped with a Bruker 5 mm TCI at 300 K was employed to run NMR data. Data processing was carried out with Topspin 3.2 software. All 2D NMR spectra were acquired in methanol-d 4 (99.95%, Sigma-Aldrich, Milano, Italy), and standard pulse sequences and phase cycling were used for COSY, HSQC, HMBC, 1D-TOCSY, and ROESY spectra. HRESIMS data were obtained in the positive ion mode on a Q Exactive Plus mass spectrometer, an Orbitrap-based FT-MS system, equipped with an ESI source (Thermo Fischer Scientific Inc., Bremen, Germany). Column chromatography was performed over silica gel (70-220 mesh, Merck). A Shimadzu LC-8A series pumping system equipped with a Shimadzu RID-10A refractive index detector and Shimadzu injector (Shimadzu Corporation, Kyoto, Japan) on a Waters XTerra Semiprep MS C 18 column (300 mm × 7.8 mm i.d.) and a mobile phase consisting of a MeOH-H 2 O mixture at a flow rate of 2.0 mL/min were employed to purify the molecules. TLC was performed on silica gel 60 F 254 (0.20 mm thickness) plates (Merck, Darmstadt, Germany) as a spray reagent, and Ce(SO 4 ) 2 /H 2 SO 4 (Sigma-Aldrich, Milano, Italy) was used [60].

Plant Material
Premna resinosa (Hochst.) Schauer (Lamiaceae) leaves were collected at Al-Kurr, Makkah Province, Saudi Arabia, in June 2018 and identified by one of the authors, Prof. A. Bader. A voucher specimen was deposited at the Herbarium of Pharmacognosy Lab at the Faculty of Pharmacy, Umm Al-Qura University (n. UQU-SA-109).

Extraction and Isolation
The exudate of P. resinosa (2.1 g) was obtained by dipping 200 g of fresh leaves into 2.1 L of dichloromethane for less than 30 s and was dried at 40 • C. The dichloromethane extract was dissolved in CHCl 3 and subjected to silica gel CC eluting with CHCl 3 , followed Plants 2023, 12, 2421 9 of 14 by increasing concentrations of CH 3 OH in CHCl 3 (between 1% and 100%) and collecting seven major fractions (A-G).

Surface Plasmon Resonance Analyses (SPR)
To investigate the interaction between 3 and Hsp70, the surface plasmon resonance (SPR) analyses were performed using a Biacore 3000 optical biosensor equipped with research-grade CM5 sensor chips (GE Healthcare, Chicago, IL, USA), according to a previously detailed method [38]. Recombinant Hsp70 surfaces, a BSA surface, and an unmodified reference surface were prepared. Proteins (100 µg/mL in 10 mM sodium acetate, pH 5.0) were immobilized on individual sensor chip surfaces at a flow rate of 5 µL/min to produce densities of 8−12 kRU. 3, 4, and oridonin (positive control) were dissolved in DMSO. The six-point concentration series (0.025−4.000 µM) was prepared. Bioevaluation software (GE Healthcare) was used to elaborate simple interactions, which were adequately fit to a single-site bimolecular interaction model (A + B = AB), yielding a single K D sensorgram [37,61].

Cell Culture and Treatment
HeLa (cervical carcinoma) and Jurkat (T-cell lymphoma) cell lines were purchased from the American Type Cell Culture (ATCC) (Rockville, MD, USA). The cells were cultured in DMEM (HeLa) or RPMI 1640 (Jurkat), supplemented with 10% FBS, 100 mg/L streptomycin, and 100 IU/mL penicillin, at 37 • C in a humidified atmosphere of 5% CO 2 . Stock solutions of compounds were prepared in DMSO (50 mM) and stored in the dark at 4 • C. In all experiments, the final concentration of DMSO did not exceed 0.15% (v/v) [62].

Computational Details
All the steps described in this paragraph were accomplished for all the stereoisomers investigated. For the generation of the starting 3D structure of 1 and 4, Maestro 12.7 was applied, and then, MacroModel 13.1 was employed for the optimization of the 3D structures, using the OPLS force field and the Polak-Ribier conjugate gradient algorithm (PRCG, maximum derivative less than 0.001 kcal/mol).
Subsequently, wide-ranging conformational search cycles at the empirical molecular mechanics (MM) level were carried out, performing Monte Carlo Multiple Minimum (MCMM) (50,000 steps) and Low Mode Conformational Search (LMCS) rounds (50,000 steps).
In addition, MD simulations of 10ns were performed, setting different temperatures (i.e., 450, 600, 700, and 750 K), a time step of 2 fs, an equilibration time of 0.1 ns, and a constant dielectric term of methanol to consider the presence of the solvent.
The minimization step (PRCG, maximum derivative less than 0.001 kcal/mol) of the sampled conformers was performed, and, using the "Redundant Conformer Elimination" MacroModel 13.1+, all the conformers differing by more than 21.0 kJ/mol (5.02 kcal/mol) from the most energetically favored and a 0.5 Å RMSD (root-mean-square deviation) were saved for the next steps. Indeed, the non-redundant conformers represented the input for QM calculations with Gaussian 09 software [32]. The MPW1PW91 functional and the 6-31G (d) basis set were employed for the conformer's geometry optimization step at the DFT level [63]. The integral equation formalism version of the polarizable continuum model (IEFPCM) related to MeOH was used for the solvent simulation [64]. To ensure the absence of further possible redundant conformers, the optimized geometries were also analyzed by visual inspection. The MPW1PW91 functional, the 6-31G (d,p) basis set, and methanol IEFPCM were used to compute the 13 C and 1 H NMR chemical shifts of each conformer of each of the accounted isomers of 1 and 4. Eventually, final 13 C and 1 H NMR datasets were obtained, considering the energy-based weight of each conformer on the total Boltzmann distribution.
The calibration of calculated chemical shifts was carried out, accounting for a multistandard approach (MSTD) [65] and according to this procedure: sp 2 13 C and 1 H NMR chemical shifts were computed, taking benzene as reference [65], while sp 3 13 C and 1 H chemical shifts were calibrated considering TMS (Tables S1-S4, Supporting Information).
In parallel, for the computation of the DP4+ probabilities, a further dataset was obtained, only considering TMS as reference (Tables S1, S2, S5 and S6, Supporting Information).
The final comparison of the experimental and predicted NMR data was accomplished through the ∆δ parameter (Tables S1-S4, Supporting Information): where δ exp (ppm) refers to 13 C/ 1 H experimental chemical shifts and δ calc (ppm) refers to calculated shifts.
Subsequently, the mean absolute error (MAE) values for the final statistical analysis were computed as follows: MAE = ∑(∆δ) n i.e., the summation (Σ) of all calculated absolute error values (∆δ), normalized considering the number of chemical shifts (n) (Tables S1-S4, Supporting Information).
The DP4+ probabilities for each investigated stereoisomer of 1 and 4 were obtained, taking into account the chemical shift data obtained by using only TMS as the reference compound (Tables S1-S4, Supporting Information) and then manually selecting the sp 2 atoms in the available DP4+ Toolbox (Excel file).

Molecular Docking
The 3D chemical structures of 1-5 were built using Maestro v. 11.1 (Schrödinger Suite 2020-4) [44]. The generated structures were prepared using LigPrep software (Schrodinger Suite) [44] and were then minimized using an OPLS3e force field. For each ligand, all possible tautomers and the protonation state at a pH of 7.0 ± 2.0 were enumerated.
The protein 3D structure was prepared using the Schrödinger Protein Preparation Wizard [44], starting from the Hsp70 X-ray structure co-complexed with the inhibitor KC7 (PDB code: 5AQX). Water molecules were removed, all hydrogen atoms were added, and bond orders were assigned. The grid box was placed on the co-complexed ligand using the Receptor Grid Generation tool [44]. Molecular docking experiments were performed using Glide software (also part of the Schrödinger Suite) [44] and the Standard Precision (SP) and Extra Precision (XP) scoring/sampling mode.

Molecular Dynamics
The XP docked pose of 3 bound to Hsp70 (PDB code: 5AQX) was submitted to molecular dynamics simulations using the Desmond Molecular Dynamic [44]. The starting complex was prepared by the System Builder in Desmond. A cubic box with a 10 Å buffer distance was set; the TIP3P water model for solvation and the OPLS3e force field were employed, and 16 Na + ions were added for obtaining the electroneutrality. The built systems were then minimized by the Minimization tool in Desmond. MD simulations of 100 ns at 310 K, using a recording interval of 100 ps and an ensemble class NPT (1.01 bar), were then performed.

Conclusions
Our study has provided valuable insights into the phytochemical constituents of P. resinosa-specifically, its exudated metabolites. These compounds hold great potential for further exploration, not only within the same genus but also in other plants that contain similar classes of compounds. Utilizing cheminformatics, we successfully identified the active pharmacophore and elucidated the mechanism of action of the isolated labdane diterpenes. Future studies should focus on developing sophisticated micro-and nanoformulations to optimize therapeutic dosing and enable targeted delivery to diseased organs. Additionally, plant tissue and cell culture approaches and chemical synthesis may overcome the limitations associated with the isolation and characterization of bioactive constituents present in trace amounts. By collaborating across different disciplines, we can harness the diverse biological activities of plant-derived biomolecules and accelerate the development of effective anticancer therapies.