We analyzed the ability of PBMCs preincubated with urolithin A, AhR agonist or antagonist to kill K562 (human chronic myelogenous leukemia, ATCC) target cells. K562 cells are widely used as a target cell to assess the cytotoxicity of PBMCs because these cells lack major histocompatibility complex (MHC) molecules resulting in the induction of NK activity (Tremblay-McLean et al., 2019). Conformity of the used cell line to the ATCC reference database was confirmed by the short tandem repeat (STR) analysis using the COrDIS Plus test system (GORDIZ, Russia).

Peripheral blood samples (8–10 mL of whole blood, heparin tubes) were obtained from patients with prostate adenocarcinomas (n = 20). Their mean age was 66 years (range, 55–77 years). Controls (n = 5) included peripheral blood of healthy, tumor-free, male individuals. Their mean age was 59 years (range, 54–76 years). All subjects gave informed consent to the study. The study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the Ethics Committee of Pirogov Russian National Research Medical University. Patients with human immunodeficiency virus (HIV)/hepatitis C virus/hepatitis B virus infection, diabetes, chronic inflammatory conditions, treated with chemotherapy or radiotherapy, or subjected to immunosuppressive therapies, were excluded from the study. Clinical features of the patients with prostate cancer are shown in Table 1.

The NK activity of PBMCs (we use cytotoxicity of PBMCs as the synonym of NK activity of PBMCs) was studied by flow cytometry by killing K562 target cells labeled with CFSE (5-, 6-carboxyfluorescein diacetate-succinimidyl ester) according to the method described in detail in (Murugin et al., 2011). PBMCs were isolated from venous blood by using Ficoll density gradient centrifugation, washed three times, and resuspended in complete culture medium RPMI containing glutamine (2 mmol/L, Paneco, Russia), and 10% fetal calf serum (Biosera, South Africa).

Freshly isolated PBMCs (10⁶/1 mL) were mixed with studied substances (urolithin A (6762, Tocris Bioscience, United Kingdom), CH223191 (antagonist of AhR, C8124, Sigma-Aldrich, United States) and β-naphthoflavone (agonist of AhR, A18543, Alfa Aesar, United States) in complete culture medium at various concentrations in 15-mL polypropylene tubes and cultured for 24 h (37°C; 5% CO2). PBMCs in a complete culture medium were used as a negative control; PBMCs in a complete culture medium and IFN-α (Invitrogen, United Kingdom) at a concentration of 1,000 IU/mL were used as a positive control for stimulation of the NK activity [as described in the article of Karsonova and coauthors (Karsonova et al., 2013)]. After the end of incubation, some supernatants were collected for multiplex plasma cytokine analysis, then PBMCs were washed once in RPMI and resuspended in a complete culture medium to test for NK activity. Some of the cells were used to make lysates for further PCR.

The NK activity of the PBMCs was measured by using a flow cytometry-based assay. Briefly, K562 target cells were labeled with CFSE (Invitrogen, Paisley, United Kingdom) and plated in 96-well U-bottom plates at 8 × 10⁴/well. Effector PBMCs were added at 4 × 10⁵/well such that the E:T ratio was 50:1 (in duplicates). In previous work different E:T ratios were tested ranging from 3.125:1 to 50:1 (Murugin et al., 2011). Significant and reliable differences between the groups (donors and patients) were observed at E:T ratio of 50:1 (Murugin et al., 2011), therefore, this ratio was used in the current work. To control for spontaneous target cell death, an additional two wells containing only target cells were prepared. The cells were incubated for 4 h (37°C; 5% CO2) and then resuspended and stained for 10 min with propidium iodide (PI, Sigma-Aldrich, United States) at 1 μg/mL. Duplicates were pooled, and the cells were analyzed by using a flow cytometer (FacsCalibur with CellQuest software, Becton Dickinson). In each sample, the percentage of dead targets was determined as the percentage of PI + events among CFSE + events. Specific killing was calculated as follows: % specific killing = [(% experimental target death − % spontaneous target death)/(100 − the % spontaneous target death)] × 100.

Total RNA was extracted using HiPure Total RNA Kit (Magen Biotechnology, China) according to the manufacturer’s specifications. The synthesis of cDNA for qPCR was carried out using the RevertAid RT Reverse Transcription Kit (Thermo Scientific, United States). qPCR for CYP-1A and CYP-1B in PBMCs was carried out using the real-time PCR system (Roche, Lightcycler 96) with the reagent mix containing SYBR Green (Evrogen, Russia).

For the qPCR experiments, primers against the sequences of human CYP-1A1 (5′-GAT​TGA​GCA​CTG​TCA​GGA​GAA​GC-3′, 5′-CCA​AAG​AGG​TCC​AAG​ACG​ATG​TTA-3′), CYP-1B1 (5′-CTC​AAC​CGC​AAC​TTC​AGC​AAC​TTC-3′, 5′- AGA​GAG​GAT​AAA​GGC​GTC​CAT​CAT-3′), and the housekeeping gene glyceraldehyde 3-phosphate dehydrogenase (GAPDH) (5′-TGC​ACC​ACC​AAC​TGC​TTA​GC-3′, 5′-GGC​ATG​GAC​TGT​GGT​CAT​GAG-3′) (Vorontsova et al., 2023) were purchased from Syntol (Moscow, Russia). The cycling conditions were 95°C for 300 s followed by 40 cycles of denaturing at 95°C for 15 s and annealing/extension for 45 s at 60°C. Reactions were carried out in duplicate. A melt curve analysis was performed to confirm the specificity of qPCR products. Relative gene expression was calculated with the 2^−ΔΔCT method. Results were normalized to the expression levels of the untreated controls, represented as the black columns.

The level of various immunoregulatory factors in the cell culture of PBMCs was measured using a commercial kit (MILLIPLEX MAP Human Cytokine/Chemokine Magnetic Bead Panel—Premixed 38 Plex, HCYTMAG-60K-PX38, Millipore Sigma, United States) using the MAGPIX (Luminex Corp, United States). The detected cytokines included sCD40L, EGF, Eotaxin/CCL11, FGF-2, Flt-3 ligand, Fractalkine, G-CSF, GM-CSF, GRO, IFN-α2, IFN-γ, IL-1α, IL-1β, IL-1ra, IL-2, IL-3, IL-4, IL-5, IL-6, IL-7, IL-8, IL-9, IL-10, IL-12 (p40), IL-12 (p70), IL-13, IL-15, IL-17A, IP-10, MCP-1, MCP-3, MDC (CCL22), MIP-1α, MIP-1β, TGF-α, TNF-α, TNF-β, VEGF. The analysis was performed according to the manufacturer’s instructions. xPONENT 4.1 Software was used to obtain the raw data. The concentrations of 38 cytokines were determined using standard curves.

Statistical analysis of the data was performed using Prism 8 software. Shapiro-Wilk test and QQ plots were used to evaluate the distribution of the variables. In the analysis of NK activity and in the analysis of PCR data t-test was used for comparison with control group, and ANOVA followed by two-stage step-up method of Benjamini, Krieger and Yekutieli was used for multiple comparisons. Multiple t-test followed by two-stage step-up method of Benjamini, Krieger and Yekutieli was used for analysis of multiplex cytokine assay data.

We have used machine learning to perform computational prediction of AhR modulation by urolithin A. The source of data for the training set was the AICD database [an integrated “Anti-Inflammatory Compounds Database” (AICD) (Wang et al., 2019)], from which we have used data on half-maximal concentrations (EC50) required to activate the AhR. We have found molecular structure in SMILES format for 212 compounds. Python was used to create the model. After extracting the names of substances for which EC50 for the AhR was presented from the original database, the structures of the substances were obtained in MOL format using the RDKit library. Next, based on the MOL format, molecular fingerprints ECFP (Extended-Connectivity Fingerprints) were created for each substance, which makes it possible to present detailed data on the features of the molecular structure in a convenient form for machine learning (bit string). The radius for obtaining data for each atom was 4, and the size of the fingerprint line was 2048 bits.

Cause of the relatively small sample size, we have used it for building the classification model–to find whether the studied substance might strongly activate the AhR or weakly activate the AhR. In our database, the median EC50 value was 1,748 nM. In this regard, we divided all substances into those that strongly activate the AhR, with an EC50 of less than 1,748 nM, and those that weakly activate this receptor, with an EC50 of more than 1,748 nM.

A Naive Bayes classifier was used to find the relationship between molecular structure and EC50. To train and test the algorithm, a division into a training and test set was used (test set of the data that was not used when training the model, the size of the test set was of 20%). The workflow of computational prediction is presented in Figure 1A. The chemical structures of urolithin A, AhR agonist (β-naphthoflavone) and antagonist (CH-223191) are shown in the Figure 1B.

We evaluated the NK activity of the PBMCs against target K562 cells in prostate cancer patients and healthy subjects (Figure 2A). We found that the addition of urolithin A increased the cytotoxic ability of PBMCs from patients with prostate cancer and healthy subjects (Figures 2B, C). For instance, urolithin A (10 μM) caused the mean increase of the NK activity of PBMCs by 23% (95% CI, 7%–38%). A lower concentration of urolithin A (5 μM) caused an average increase of the NK activity of PBMCs by 21% (95% CI, 5%–36%).

We have studied whether AhR is involved in the capability of urolithin A to induce NK activity of PBMCs (Figure 3). For this purpose, we have studied the influence of urolithin A on the NK activity of PBMCs in the presence of an agonist (β-naphtoflavone) or antagonist (CH-223191) of AhR (Figure 3A).

In our experiments, the antagonist of AhR, similar to urolithin A, increased the NK activity of PBMCs. In contrast, the agonist of AhR hasn’t shown such influence on the NK activity of PBMCs from patients with prostate cancer. Moreover, the agonist of AhR reduced to some extent urolithin A-mediated increase in PBMCs NK activity in patients with prostate cancer (p = 0.3).

Next, for further studying of the role of AhR in the urolithin A-dependent increase of PBMCs NK activity, we have examined the gene expression of target enzymes of AhR – CYP1A and CYP1B. Urolithin A suppressed the expression of CYP1B in both groups. Likewise, the antagonist of AhR suppressed the expression of CYP1B. Agonist of AhR increased CYP1B expression, mostly in PBMCs from patients with prostate cancer. And, finally, the agonist of AhR reduced the urolithin A-mediated decrease of CYP1B in both groups (Figure 3B).

We have made a structure-activity classification model, which allows us to predict the high activating potential (agonism) or low activating potential of the studied compounds to the AhR. When tested on the test set, the F1 Score was 0.727 (the harmonic mean between precision and recall). The accuracy of the model (AUC—Area Under the Curve) was 0.733.

Surprisingly, using this model, we have found that urolithin A belongs to the group of substances with relatively high potential to activate the AhR (EC50 less than 1,748 nM), as well as beta-naphthoflavone. Based on our model, the compound CH223191 (the antagonist of AhR), belongs to the group of substances with relatively low potential to activate the AhR, which suits to its role as an antagonist of AhR receptor. We have no dataset containing sufficient numbers of IC50 for AhR, that’s why only the EC50 dataset was used.

Since, apparently, antagonism of the AhR is not the only mechanism of increase of NK activity by urolithin A, we have preliminarily studied the influence of urolithin A (also in the absence and in the presence of AhR agonist or antagonist) on the production of panel of inflammatory cytokines by PBMCs of patients with prostate cancer and healthy subjects (Figure 4). We have used multiplex cytokine assay to preliminarily identify possible targets of urolithin A among cytokines for further studies. Some of studied cytokines can be involved in the increase of NK activity (e.g., IFN-α) (Karsonova et al., 2013), while some of them can be involved in the decrease of NK activity (e.g., IL-6 and IL-8) (Wu et al., 2019). According to our data, unstimulated PBMCs from prostate cancer patients produced significantly higher levels of a number of immunoregulatory factors (GRO, MCP-3, MIP-1a, IL-8) than PBMCs from healthy subjects. On the one hand, urolithin A caused a decrease in the production of immunoregulatory factors, which was most noticeable for fractalkine, IL-8, and MCP-3. On the other hand, urolithin A increased IFN-α levels to some extent, but more experiments are needed to draw more reliable conclusions.