Tumor-Specificity, Neurotoxicity, and Possible Involvement of the Nuclear Receptor Response Pathway of 4,6,8-Trimethyl Azulene Amide Derivatives

Background: Very few papers covering the anticancer activity of azulenes have been reported, as compared with those of antibacterial and anti-inflammatory activity. This led us to investigate the antitumor potential of fifteen 4,6,8-trimethyl azulene amide derivatives against oral malignant cells. Methods: 4,6,8-Trimethyl azulene amide derivatives were newly synthesized. Anticancer activity was evaluated by tumor-specificity against four human oral squamous cell carcinoma (OSCC) cell lines over three normal oral cells. Neurotoxicity was evaluated by cytotoxicity against three neuronal cell lines over normal oral cells. Apoptosis induction was evaluated by Western blot and cell cycle analyses. Results: Among fifteen derivatives, compounds 7, 9, and 15 showed the highest anticancer activity, and relatively lower neurotoxicity than doxorubicin, 5-fluorouracil (5-FU), and melphalan. They induced the accumulation of a comparable amount of a subG1 population, but slightly lower extent of caspase activation, as compared with actinomycin D, used as an apoptosis inducer. The quantitative structure–activity relationship analysis suggests the significant correlation of tumor-specificity with a 3D shape of molecules, and possible involvement of inflammation and hormone receptor response pathways. Conclusions: Compounds 7 and 15 can be potential candidates of a lead compound for developing novel anticancer drugs.


Introduction
Azulene is a 10 π-electron non-benzenoid aromatic hydrocarbon with a fused structure of five-and seven-membered rings, showing a deep blue coloration. The resonance structure of azulene contains ionic cyclopentadienide and tropylium substructures, resulting in electrophilic substitution reactions at the 1-and 3-positions and nucleophilic addition reactions at the 2-, 4-, 6-, and 8-positions, along with the 2-position at the five-membered ring in some cases [1][2][3].
Azulene derivatives, including guaiazulene ( Figure S1A), are present in many plants and mushrooms, and applied as optoelectronic devices and ingredients used for hundreds of years in antiallergic, antibacterial, and anti-inflammatory therapies [4]. Several research studies have reported the applications of azulenes on oral diseases. For example, gargling with sodium azulene has been applied to maintain the compliance with afatinib treatment [5]. Azulene rinse has been applied to dry mouth and salivary gland dysfunction following radiotherapy, but with no convincing evidence [6]. Oral administration of marzulene (L-glutamine plus azulene) stimulates repair mechanisms of rat gastric mucosa after NaOH injury [7]. Guaizulene alleviated the paracetamol-induced glutathione (GSH) depletion and hepatic damage, possibly by its antioxidant activity [8]. Photodynamic activation of a lower concentration of guaiazulene suppressed inflammatory markers in peripheral blood mononuclear cells possibly by generating singlet oxygen [9]. On the other hand, very few reports of anticancer activity and quantitative structure-activity relationship (QSAR) analysis of azulenes have been reported [10][11][12].
In order to determine the tumor-specificity of azulene derivatives, we decided to use a set of four human oral squamous cell carcinoma cell lines and three human normal oral mesenchymal cells, rather than using human normal epithelial cells. This decision was based on our previous findings that (i) human normal epithelial cells (oral keratinocyte HOK, primary gingival epithelial cells HGEP) cannot be grown in regular culture medium (DMEM + 10%FBS) and, (ii) when HOK and HGEP were cultured in their specific growth factor-enriched medium, they began to rapidly grow like malignant cells, acquiring extremely higher sensitivity against anticancer drugs, resulting in the reduction of TS values below 1.0 [13].
In continuation of finding more tumor-selective compounds, a total of fifteen 4,6,8trimethyl azulene amide derivatives ( Figure 1) were investigated for their cytotoxicity against four OSCC cell lines and three normal human oral mesenchymal cells. Since many cancer drugs have been reported to show severe neurotoxicity [17][18][19], neurotoxicity of fifteen 4,6,8-trimethyl azulene amide derivatives was also compared with those of popular anticancer drugs.

Tumor-Specificity
Four human oral squamous cell carcinoma cell line (Ca9-22 originated from gingiva; HSC-2, HSC-3, HSC-4 from tongue) and three human mesenchymal normal oral cells (gingival fibroblast HGF, periodontal ligament fibroblast (HPLF) and pulp cells (HPC) were incubated for 48 h with various concentrations of compounds 1-15 and three reference compounds (doxorubicin (DOX), 5-fluorouracil (5-FU), melphalan (L-phenylalanine mustard, L-PAM) in DMEM supplemented with 10% fetal bovine serum (FBS) and antibiotics. Viable cell number was then determined by an MTT method in triplicate. Cytotoxicity of dimethyl sulfoxide (DMSO), used for dissolving these compounds, was subtracted. The cytotoxicity experiments were performed three times, and 50% cytotoxic concentration (CC 50 ) was determined from the dose-response curve done in triplicate ( Figures S2-S4), and listed in Table 1. We have noticed that Ca9-22 (derived from gingiva) is more sensitive than other oral squamous cell lines (HSC-2, HSC-3, HSC-4) (derived from tongue) to most of the azulene compounds, and oral squamous cells are more sensitive than normal mesenchymal cells. Since such trends, however, are reproducible in three independent experiments (as shown in supplementary data Figures S2-S4), we presented the mean value. When four cancer cells and three normal oral cells were used, compound 15 showed the highest tumor-specificity (TS = 17.9), followed by compound 7 (TS = 7.8) and compound 9 (TS = 5.7), higher than that of 5-FU (TS = 2.5) (D/B in Table 1). When two gingival tissuederived cells Ca9-22 and HGF were used, compound 15 again showed the highest tumorspecificity (TS = 20.1), followed by compound 9 (TS = 10.2) and compound 7 (TS = 9.6) (C/A in Table 1).
For clinical application of candidate compounds, the database of their tumor-specificity and cytotoxicity against tumor cells provides useful information. Thus, PSE (Potency-Selectivity Expression (the ratio of the tumor-specificity to the CC 50 against cancer cells) × 100) was calculated for all compounds. 100C/A 2 in Table 1). PSE values of all derivatives were two-orders lower than that of doxorubicin (PSE = 12,325; 6270). However, PSE values of compounds 7, 9, and 15 were higher than that of 5-FU (PSE = 0.6; 2.7), and PSE value of compound 15 was higher than that of L-PAM (PSE = 56.5; 16.6). Based on these data, compounds 7, 9, and 15 were chosen for further analysis.
Similarly, actinomycin D (1 µM), and higher concentration of compound 7 (160 µM), low and high concentrations of compound 9 (60 and 120 µM) increased the relative proportion of subG 1 population (that reflect DNA fragmentation) to a nearly comparable level attained by actinomycin D. On the other hand, compound 15 at higher concentration (80 µM) only slightly increased subG 1 population (p < 0.05) (Figure 4).  Western blot analysis demonstrated that compounds 7 and 9, but not compound 15, induced the cleavage of poly ADP-ribose polymerase (PARP) and procaspase 3, suggesting the induction of apoptosis ( Figure 5). Since the extent of apoptosis induction was much lower than that induced by actinomycin D, the possibility of induction of other types of cell death such as necrosis has been suggested.   Figure S5. N.D., not determined.

Possible Nuclear Receptor/Stress Response Pathways
Nuclear receptors and stress response pathways that are possibly involved in the inhibition of OSCC growth by 4,6,8-trimethyl azulene amide derivatives were predicted. The specific cytotoxicity against OSCC cells were correlated with nuclear factor-kappa B (NFκB) agonist, estrogen receptor alpha with stimulator antagonist, thyroid stimulating hormone receptor (TSHR) antagonist, and glucocorticoid receptor agonist (Figure 7). These data suggest that the tumor-specificity of 4,6,8-trimethyl azulene amide derivatives might be coupled to the signaling pathway of NFκB and estrogen, thyroid stimulating hormone and glucocorticoid receptors.
Dose-response curve demonstrated that compound 15 with the highest TS value inhibited the growth of Ca9-22 cells without complete killing of the cells ( Figure 2C). Compound 15 induced trace amounts of a subG 1 population (that reflects DNA fragmentation) (p = 0.029) (Figure 4) and caspase-3 activation (assessed by cleavage of PARP and procaspase 3) ( Figure 5).
On the other hand, both compounds 7 and 9 were rather cytotoxic, inducing cell shrinkage ( Figure 3C-F) and the accumulation of a subG 1 population above 80 and 60 µM, respectively (p < 0.027) (Figure 2B), to a similar extent that was attained by actinomycin D (Figure 4). However, their caspase 3 activating activity was much lower than that of actinomycin D ( Figure 5). This suggests the possibility that compounds 7 and 9 may induce different types of cell death from apoptosis.
Finally, a possible signaling pathway of 4,6,8-trimethyl azulene amide derivatives was estimated by Toxicity Predicators. Their tumor-specificity was matched at a higher probability with the onset of a signaling pathway of NFκB, and receptors for estrogen, thyroid stimulating hormone, and glucocorticoid ( Figure 7). This is consistent with the reported anti-inflammatory activity of guaiazulene [4,9,[20][21][22]. On the other hand, the involvement of other signaling pathways such as caspase and androgen receptor ( Figure S7), transforming growth factor β (TGFβ), peroxisome proliferator activated receptor δ (PPARδ), endoplasmic reticulum stress (ER stress) response, and retinoid X receptor-α(RXR) ( Figure S8) seems to be low. This further supports the non-apoptotic cell death induced by compounds 7, 9, and 15. It remains to identify the type of cell death induced by these compounds.

Materials
The following chemicals were obtained from the indicated companies: Dulbecco's modified Eagle's medium (DMEM) from Thermo Fisher Scientific (
Differentiated PC12 cells were prepared by the "overlay method", as described previously. In short, PC12 cells were cultured in the serum-free DMEM supplemented with 50 ng/mL NGF, and at Day 3 overlayed with fresh NGF solution. The Day 6 cells with ex-tended neurites were used for the experiment.

Synthesis Assay for Cytotoxic Activity
Cells were inoculated at 2 × 10 3 cells/0.1 mL in a 96-microwell plate. After 48 h, the medium was replaced with 0.1 mL of fresh medium containing different concentrations of test compounds. Control cells were treated with the same amounts of DMSO present in each diluent solution. Cells were incubated for 48 h and the relative viable cell number was then determined by the MTT method, as described previously [15]. We have prepared the controls that contain DMSO (1, 0.5, 0.25, 0.125, 0.063, 0.031, 0.016, 0.008%). Cytotoxicity induced by DMSO alone was subtracted from each well of a 96-microwell plate. The CC 50 was determined from the dose-response curve of triplicate samples.

Western Blot Analysis
The cells were washed, lysed, and their protein extracts subjected to Western blot (WB) analysis, as described previously [30]. All protein samples of cell lysates (15 µg) were separated by SDS-PAGE using a Mini-Protean 3 Cell system (Bio-Rad Laboratories, Hercules, CA, USA). After electrophoresis, the separated proteins were transferred onto a PVDF filter using a Trans-Blot Turbo System (Bio-Rad Laboratories). The blots were blocked at room temperature for 50 min in skim milk (Morinaga-Nyugyo, Tokyo, Japan) and then probed for 120 min with a primary antibody cocktail (1:250) from Apoptosis Western Blot Cocktail kit (Abcam, Cambridge, UK). The blots were washed three times with Tris-buffered saline (pH 7.6) containing 0.05% Tween 20 and then probed for 90 min with a horseradish peroxidase-conjugated secondary antibody cocktail (1:100) from the kit. Immunoreactivities were detected using Amersham ECL Select (Cytiva, Tokyo, Japan). Images were acquired using ChemiDoc MP System (Bio-Rad Laboratories) and Image Lab 4.1 software (Bio-Rad Laboratories) [30].

Calculation of Chemical Descriptors
pCC50 (i.e., the −log CC 50 ) was used for the comparison of the cytotoxicity between the compounds, since the CC 50 values had a distribution pattern close to a logarithmic normal distribution. The mean pCC 50 values for normal cells and tumor cell lines were defined as N and T, respectively [31]. Thus, T represents -log (mean CC 50 against OSCC), N represents −log (mean CC 50 against normal oral cells, T-N represents −log (mean CC 50 for normal cells/mean CC 50 for OSCC). The chemical structures were cleaned and standardized (removing salts and adjusting the protonation state (neutralize)). Then, the 3D-structure of each compound was generated by CORINA Classic (Molecular Networks GmbH, Nürnberg, Germany) and determined optimal 3D-structure with force field calculations (amber-10: EHT) in Molecular Operating Environment (MOE) version 2020.09 (Chemical Computing Group, Quebec, Canada). Chemical 2D and 3D descriptors were calculated with MOE version 2020.9 and Mordred version 1.2.0 (Python library) [32] based on optimal 3D structures. Descriptors that were duplicated and contained missing values and outliers were excluded from this analysis. Note that we regarded results outside 1st quartile −3 × interquartile (IQR) to 3rd quartile +3 × IQR range as outliers. Then, the multicollinearity of descriptors was analyzed. A threshold of 1 of the absolute value of the correlation coefficient was adopted as multicollinearity. When multicollinearity was detected, only descriptors that showed the highest correlation with objective variables among multicollinear descriptors was adopted. Finally, 1520 descriptors were used for this analysis.

Calculation of Chemical Descriptors
The activities against 59 signaling pathways [33], agonist and antagonist activities of the nuclear receptor, and stress response pathway were calculated by the chemical structures. In other words, all azulene derivatives were classified as positive or negative based on the calculated probabilities in Tox21 activity scores of 1 or higher for each signaling pathway using the Toxicity Predictor [33].

Statistical Analysis
Each experimental value was expressed as the mean ± standard deviation (SD) of triplicate or quadruplicate measurements. One-way ANOVA and Dunnett's post-test were performed using IBM SPSS 27.0 statistics (IBM Co., Armonk, NY, USA). The correlation between chemical descriptors and cytotoxicity or tumor specificity was investigated using simple regression analyses by scikit-learn and SciPy with Python 3.8.5. Student's t-test was performed using JMP Pro 6 (SAS Institute, Cary, NC, USA). The significance level was set at p < 0.05.

Conclusions
The present study demonstrated that three compounds from fifteen 4,6,8-trimethyl azulene amide derivatives selectively inhibited the growth of human oral squamous cell carcinoma cell lines. Their actions are either cytotoxic or cytostatic, accumulating subG 1 . The 3D-structure may be the determinant of tumor-selectivity. There was possible involvement of inflammation and hormone receptor pathway rather than caspase pathway in the selective cytotoxicity against OSCC cells. At present, TS values of compounds 7, 9, and 15 are not so high compared with anticancer drugs (Table 1). Since compound 9 was cytotoxic, further chemical modification of this compound that enhances the TS value should be done. On the other hand, compound 7, compound 15, and 5-FU showed cytostatic growth inhibition, and their action may be potentiated through combination with other types of anticancer drugs [34].