ORIGINAL RESEARCH published: 10 December 2020 doi: 10.3389/fphar.2020.602369 Evaluation of Serum/Urine Genomic and Metabolomic Profiles to Improve the Adherence to Sildenafil Therapy in Patients with Erectile Dysfunction Maria Santa Rocca 1†, Alessia Vignoli 2†, Leonardo Tenori 3, Marco Ghezzi 1, Maurizio De Rocco Ponce 4, Giannis Vatsellas 5, Dimitris Thanos 5, Roberto Padrini 6, Carlo Foresta 1* and Luca De Toni 1 1 Unit of Andrology and Reproduction Medicine—Department of Medicine, University of Padova, Padova, Italy, 2Consorzio Interuniversitario Risonanze Magnetiche di Metallo Proteine (CIRMMP), Sesto Fiorentino, Italy, 3Department of Chemistry “Ugo Schiff”, University of Florence, Sesto Fiorentino, Italy, 4Autònoma de Barcelona, Instituto de Investigaciones Biomédicas Sant Pau, Barcelona, Spain, 5Biomedical Research Foundation Academy of Athens (BRFAA), Athens, Greece, 6Department of Medicine, University of Padova, Padova, Italy Edited by: Lei Xi, Virginia Commonwealth University, Type V-phosphodiesterase-inhibitors (PDE5i) are the first choice drugs in the treatment of United States erectile dysfunction (ED), being effective in 60–70% of patients. However, approximately Reviewed by: Sai S. Koka, 50% of patients per year discontinue the treatment with PDE5i after reporting poor drug University of Houston, United States efficacy or major adverse drug reactions (ADR). To identify early markers of efficacy/safety Yuan James Rao, for the treatment of ED with PDE5i, the basal clinical characteristics of patients, integrated Washington University in St. Louis, United States with metabolomics analysis of serum and urine and genomic data, were here correlated *Correspondence: with the PDE5i efficacy and the occurrence of ADR upon administration. Thirty-six males Carlo Foresta with new diagnosis of ED were consecutively recruited and characterized at baseline for carlo.foresta@unipd.it † anthropometrics, blood pressure, blood glucose, lipid profile, serum levels of thyroid/sex These authors have contributed equally to this work hormones and erectile function evaluated by IIEF-15 questionnaire. Targeted Next Generation Sequencing (NGS) was applied to genes involved in PDE5i Specialty section: pharmacodynamics and pharmacokinetics. Fasting metabolic profiles of serum and This article was submitted to urine were assessed by nuclear magnetic resonance (NMR)-based metabolomics Translational Pharmacology, a section of the journal analysis. Patients were prescribed on-demand therapy with Sildenafil oro-dispersible Frontiers in Frontiers in Pharmacology film and followed-up after 3 months from recruitment. Baseline data were compared Received: 03 September 2020 with IIEF-15 score at follow-up and with the occurrence of ADR recorded by a dedicated Accepted: 04 November 2020 Published: 10 December 2020 questionnaire. Twenty-eight patients were finally included in the analysis. Serum LDL- Citation: cholesterol levels were increased in those reporting ADR (143.3 ± 13.2 mg/dl ADR vs. Rocca MS, Vignoli A, Tenori L, 133.1 ± 12.4 mg/dl No ADR; p  0.046). NGS data showed that specific variants Ghezzi M, De Rocco Ponce M, of PDE11A and CYP2D7 genes were more represented in drug responders (both Vatsellas G, Thanos D, Padrini R, Foresta C and De Toni L (2020) relative risk  2.7 [0.9–5.1]; p  0.04). NMR-based metabolomics showed the highest Evaluation of Serum/Urine Genomic association between serum LDL-cholesterol metabolites and the occurrence of ADR and Metabolomic Profiles to Improve the Adherence to Sildenafil Therapy in (Hazard ratio  17.5; p  0.019). The association between lipid profile and the ADR pattern Patients with Erectile Dysfunction. suggests major cues in the tailoring of ED therapy with PDE5i. Front. Pharmacol. 11:602369. doi: 10.3389/fphar.2020.602369 Keywords: drug safety, adherence, adverse drug reactions, translational research, nuclear magnetic resonance Frontiers in Pharmacology | www.frontiersin.org 1 December 2020 | Volume 11 | Article 602369 Rocca et al. Patient Profiling in PDE5i Treatment INTRODUCTION In this pilot study we applied unbiased metabolomics analysis of serum and urine to integrate genomic data in order to identify Therapy discontinuation represents a major public health markers of unfavorable efficacy/safety profile of the therapy with problem. Between 25 and 50% of patients worldwide do not PDE5i. To this aim, the genetic profile of a group of patients follow prescribed therapies, resulting in significant health and affected by ED was obtained by gene panel-sequencing including economic issues (van Dulmen et al., 2007). In the USA, genes involved in PDE5i pharmacodynamics and drug suboptimal adherence has been associated with 125,000 deaths, metabolism. Finally, genetic data and the metabolic profile of 10% of hospitalizations, and costs up to US$ 289 billion annually serum and urine at baseline were combined in order to correlate (Osterberg and Blaschke, 2005). However, no shared or high all findings with the subsequent response to PDE5i impact solutions have been identified so far (Nieuwlaat et al., administration in terms of drug efficacy and adverse drug 2014). A feasible strategy relies on precision medicine, such as the reactions. identification of tailored solutions addressing a patient’s specific adherence barrier and the subsequent scale-up to the population level (Zullig et al., 2018). METHODS A suitable application of these concepts is represented by the model of type V-phosphodiesterase (PDE5) inhibitors (PDE5i), Patients representing the first choice therapy for the erectile dysfunction This single-center, prospective, pathophysiological-genetic study (ED). ED, namely the inability to achieve and/or maintain an was conducted in agreement with the declaration of Helsinki adequate degree of erection for a complete sex intercourse, depends (Issue Information-Declaration of Helsinki, 2018) and was on the impaired production of nitric oxide (NO) by the approved by the Ethic Committee for Clinical Trials of the endothelium of cavernous vessels because of an underlying University-Hospital of Padova (Italy) with protocol number atherosclerotic process. This pattern can be pharmacologically 3982/AO/16 and subsequent amendments. overcome by the use of selective inhibitors of cGMP-dependent The study involved the recruitment of male patients with new PDE5, which are highly expressed in cavernous endothelium, in diagnosis of ED, eligible for treatment with PDE5i, attending the order to prolong the cGMP half-life and to enhance the residual Unit of Andrology and Reproductive Medicine of the University- NO-mediated vasodilating function (Hawksworth and Burnett, Hospital of Padova (Italy). Eligibility criteria for the study were: 2015). The estimated prevalence of ED worldwide varies from 3 age between 25 and 60 years old, diagnosis of ED with low or to 76.5% according to age range and, despite it is not considered a intermediate cardiovascular risk level assessed according to the life threatening condition, it is associated with important health cardiovascular risk charts of the III Princeton Consensus issues, such as reduced self-esteem and reduced quality of life (Yafi Conference (Nehra et al., 2012), body mass index (BMI) et al., 2016). The use “on-demand” of PDE5i is demonstrated to be between 19 and 40 kg/m2, and the signing of the informed effective in 60–70% of patients (Yafi et al., 2016). However, high consent to participate in the study. Exclusion criteria were rates of therapy discontinuation have been reported by several chronic diseases due to intestinal malabsorption, major studies, accounting for approximately 50% of patients/year. The endocrine disorders, renal or hepatic failure, severe neoplasms, reasons for the discontinuation are mainly related to the lack of or the presence of psychiatric conditions impeding the efficacy and serious side effects (Corona et al., 2016). Thus the participation in the study. identification of early/predictive markers of poor drug efficacy/ A total of 36 patients were consecutively recruited during the safety profile for PDE5i may represent a key strategy to improve the baseline outpatient evaluation from April 2017 to June 2019. treatment of ED and related comorbidities. Eight of these patients were excluded from further analyses since The molecular bases underlying the unfavorable efficacy/ they missed the follow-up visit. Each patient was evaluated for the safety profile of PDE5i are currently under-investigated. clinical history, concomitant drug therapy, anthropometric Available studies are focused on the pharmacogenetic parameters (weight, height, BMI), blood pressure, blood implications in ED, identifying a panel of genes, related to NO glucose and lipid profile, serum levels of thyroid-stimulating and cGMP pathway, presenting genetic variants that are hormone (TSH), total testosterone (T), luteinizing hormone associated with poor response to PDE5i. However, clinical (LH), and prolactin. The presence of diabetes was diagnosed data on the prognostic value of these markers are not yet by fasting blood levels of glucose ≥126 mg/dl or the use of available (Lacchini and Tanus-Santos, 2014). In addition, the glycemic lowering agents. The presence of dyslipidemia was onset of side effects after PDE5i dosing has been classically diagnosed for blood LDL levels ≥ 130 mg/dl or the use of ascribed to the inhibitory side-effect on other PDE activities, lipidemia lowering agents. The presence of hypogonadism was such as PDE1, 2, 4, 6, and 11. However, this hypothesis is mainly diagnosed for serum T levels ≤ 10.4 nmol/L. The erectile function supported by in vitro studies (Bischoff, 2004; Taylor et al., 2009). was assessed by the administration of the International Index of In this context, metabolomics is considered as a novel and the Erectile Function 15 questions-validated questionnaire (IIEF- valuable tool to characterize the clinical traits associated with 15), and the cardiovascular risk was stratified according to the drug response. Performed in an unbiased manner, association of international guidelines (Nehra et al., 2012). The IIEF-15 metabolomics to systems biology has provided insight into gene comprises a series of question focused, respectively, on the networks, metabolic changes, and genotype/phenotype erectile function (questions 1–5 and 15; IIEF-15ED), the quality correlations (Widmann et al., 2013). of the orgasm (questions 9 and 10; IIEF-15ORGASM), on the sexual Frontiers in Pharmacology | www.frontiersin.org 2 December 2020 | Volume 11 | Article 602369 Rocca et al. Patient Profiling in PDE5i Treatment TABLE 1 | Basal demographic and clinical parameters of patients with erectile dysfunction included in the group, distinguished for response to type five phosphodiesterase inhibitors and related adverse drug reactions. Response to PDE5i Adverse drug reaction Parameter All patients Responders Non-responders ADR NO ADR (N = 28) (N = 24) (N = 4) (N = 12) (N = 16) Demographic/Clinical Age (years ± SD) 47.6 ± 13.9 48.8 ± 13.25 40.5 ± 17.9 48.1 ± 10.4 47.2 ± 16.12 Height (cm ± SD) 175.5 ± 8.9 175.5 ± 9.5 175.5 ± 4.5 176.2 ± 6.7 175.1 ± 10.1 Weight (kg ± SD) 81.7 ± 17.3 81.6 ± 17.6 82.8 ± 18.3 78.7 ± 16.4 83.5 ± 18.0 BMI (kg/m2 ± SD) 26.4 ± 4.7 26.3 ± 4.5 26.9 ± 5.9 25.2 ± 3.7 27.1 ± 5.2 SBP (mmHg ± SD) 133.2 ± 12.3 131.6 ± 11.4 133.9 ± 12.6 135.0 ± 12.9 132.1 ± 12.9 DBP (mmHg ± SD) 81.8 ± 6.0 80.9 ± 5.4 82.1 ± 5.9 82.5 ± 5.0 81.4 ± 6.9 Glucose (mg/dL ± SD) 102.3 ± 9.8 100.3 ± 8.4 103.1 ± 7.3 94.7 ± 6.8 106.2 ± 9.0 Total Chol (mg/dl ± SD) 185.4 ± 38.9 181.0 ± 38.1 187.9 ± 37.6 187.0 ± 30.2 184.4 ± 39.5 HDL Chol (mg/dl ± SD) 48.7 ± 13.9 49.5 ± 14.6 46.0 ± 12.5 46.0 ± 14.4 50 ± 14.8 LDL Chol (mg/dl ± SD) 143.7 ± 17.1 143.3 ± 25.7 149.9 ± 11.2 143.3 ± 13.2 133.1 ± 12.4* Triglycerides (mg/dl ± SD) 102.1 ± 47.2 98.4 ± 48.9 108.6 ± 34.4 133.0 ± 38.5 86.7 ± 45.9 Total T (nmol/l) 16.9 ± 8.5 17.5 ± 7.2 15.8 ± 9.5 13.1 ± 7.2 18.0 ± 9.8 LH (IU/ml ± SD) 5.2 ± 3.11 4.0 ± 2.6 5.2 ± 3.4 4.9 ± 3.8 5.0 ± 2.9 Prolactin (ng/ml ± SD) 10.7 ± 0.3 10.5 ± 4.6 12.1 ± 2.4 9.8 ± 3.5 11.3 ± 4.9 Organic ED (n/%) 12/42.9% 11/45.8% 1/25% 5/41.7% 7/43.8% Diabetes (n/%) 7/25.0% 7/14.8% 0/0% 2/16.7% 5/31.3% Dyslipidemia (n/%) 10/35.7% 8/33.3% 2/50% 3/25.0% 7/43.8% Hypogonadism (n/%) 4/14.3 2/8.3% 2/50% 1/8.3% 3/18.8% Thyroid diseases (n/%) 1/3.6% 1/4.2% 0/0% 0/0% 1/6.3% Concomitant Drug Therapy Glycemic lowering agents 3/10.7% 3/12.5% 0/0% 1/8.3% 2/12.5% Lipid lowering agents 2/7.1% 2/8.3% 0/0% 1/8.3% 1/6.3% Antihypertension agents 5/17.9% 5/20.8% 0/0% 0/0% 5/31.3% Antiplatelet/anticoagulants 3/10.7% 3/12.5% 0/0% 0/0% 3/18.75% Poly-therapy 3/10.7% 3/12.5% 0/0% 0/0% 3/18.75% Erectile Function IIEF15ED (Score ± SD) 16.5 ± 6.8 17.1 ± 6.4 13.0 ± 8.4 19.5 ± 6.4 14.5 ± 6.4 IIEF15ORGASM (Score ± SD) 7.9 ± 2.5 7.9 ± 2.2 8.0 ± 4.0 8.2 ± 2.5 7.7 ± 2.5 IIEF15DESIRE (Score ± SD) 6.8 ± 2.1 7.0 ± 1.9 5.5 ± 2.6 7.5 ± 2.4 6.4 ± 1.8 IIEF15SATISFACTION (Score ± SD) 9.3 ± 4.9 9.2 ± 3.8 9.5 ± 10.3 12.3 ± 4.2 7.3 ± 4.3 IIEF15QOL (Score ± SD) 6.2 ± 3.9 6.5 ± 3.7 4.5 ± 5.0 7.3 ± 4.1 5.4 ± 3.7 Abbreviations: PDE5i, type 5 phosphodiesterase inhibitors; ADR, adverse drug reaction; SD, standard deviation; BMI, body mass-index; Chol, cholesterol; HDL, High density-lipoprotein; LDL, Low density-lipoprotein; T, serum testosterone; ED, erectile dysfunction; Poly-therapy, the use of two or more different drug categories; IIEF15, 15-question International Index of Erectile Function; QOL, quality of life. Significance: *  p < 0.05 vs. ADR. desire (questions 11 and 12; IIEF-15DESIRE), on the sexual increased drug tolerability profile compared to the generator film satisfaction (questions 6–8; IIEF-15SATISFACTION), and on the coated tablet or standard oral dosing, representing a strategy to overall quality of life (questions 13 and 14; IIEF-15QOL). The ameliorate the patient’s compliance to the therapy and the presence of cavernous artery alterations was diagnosed by protocol (De Toni et al., 2018). dynamic Penile Color-Doppler Ultrasound analysis (P-CDU) At recruitment, and in any case before initiating the on- after intra-cavernous PGE1 injection as previously described demand therapy with PDE5i, each patient was requested to (Caretta et al., 2019). Patients with increased intima-media provide a sample of peripheral blood for the assessment of thickness of the cavernous artery wall (i.e., IMT > 0.3 mm) genetic screening, which was stored at −80°C until use. were then defined as affected by organic ED. Additional samples of blood serum and early morning urine at To avoid the dispersion of subjects on the basis of the available fasting conditions were obtained for metabolomics assessment. PDE5i drugs, all patients recruited in the study were prescribed to All samples were centrifuged at 1,000 × g for 10 min to remove use Sildenafil orodispersible film (ODF) as part of the standard of cell debris and then stored at −80°C until use. All specimens were care, whose dosage was determined by the physician according to processed within 1 h from sampling. of good clinical practice criteria in the management of ED For the evaluation of side effects associated with PDEi (Greenberg et al., 2019). Patients were also instructed to take administration, patients were asked to fill out a questionnaire the drug in starving conditions and to hold the ODF under the for the adverse drug reaction (ADR-questionnaire), adapted from tongue for 15 min without the assumption of water, followed by the form of the Italian Drug Agency (https://www.aifa.gov.it/ swallowing as previously described (De Toni et al., 2018). This moduli-segnalazione-reazioni-avverse). The ADR-questionnaire specific dosage was chosen because of its association with an provided a direct yes/no question about the drug-response and a Frontiers in Pharmacology | www.frontiersin.org 3 December 2020 | Volume 11 | Article 602369 Rocca et al. Patient Profiling in PDE5i Treatment TABLE 2 | The gene panel evaluated in 28 male patients with erectile dysfunction An Agilent SureSelect XT custom library panel for 22 genes included in the study and the corresponding number of gene variants. (CYP3A4, CYP2C9, CYP2D6, CYP3A5, NOS1, NOS3, VEGFA, ID Gene N° variants out ACE, GNB3, PDE1A, PDE1B, PDE1C, PDE2A, PDE3A, PDE3B, of 2,629 total PDE4A, PDE5A, PDE6A, PDE7A, PDE9A, PDE10A, PDE11A) variants was designed with the Agilent SureDesign software. For a total CYP2C9 18 Region of interest (ROI) size of 66.183 kb, 2.747 probes were CYP2D6 504 designed, covering in total 83.156 kb. CYP2D7 113 Targeted NGS library preparation was carried out with the CYP3A4 33 Agilent SureSelect QXT library prep kit and the libraries were CYP3A5 14 GNB3 24 then loaded on a 500-cycle (2 × 250 paired ends) reagent cartridge LOC101929829 126 (Illumina, San Diego, CA, USA) and run on a MiSeq sequencer LOC105373764 50 (Illumina, San Diego, CA, USA). Raw sequencing data analysis NOS1 115 and variant calling was performed with the Agilent SureCall NOS3 142 software using GRCh37/hg19 as reference genome. For each run, PDE10A 91 PDE11A 204 average read depth within the ROI was 222X. Of the ROI, 96.5 PDE1A 87 and 87.2% was covered by at least 50 and 100 reads respectively. PDE1B 49 PolyPhen-2 (http://genetics.bwh.harvard.edu/pph2/) was used as PDE1C 123 an in silico tool to predict possible impact of non-synonymous PDE2A 232 variants on the structure and function of the protein. PDE3A 79 PDE3B 28 PDE4A 45 NMR Analysis PDE5A 94 All serum and urine samples were analyzed via Nuclear Magnetic PDE6A 36 Resonance (NMR) spectroscopy. One-dimensional 1H NMR PDE7A 28 spectra for all samples were acquired using a Bruker 600 MHz PDE9A 165 VEGFA 24 spectrometer (BrukerBioSpin) operating at 600.13 MHz proton ACE 205 Larmor frequency and equipped with a 5 mm PATXI 1H-13C-15N and 2H decoupling probe including a z axis gradient coil, an automatic tuning-matching (ATM) and an automatic and refrigerate sample changer (SampleJet). A BTO 2000 direct yes/no question about the occurrence of side effects thermocouple served for temperature stabilization at the level secondary to PDE5i ingestion. In case of a positive answer to of approximately 0.1 K at the sample. Before measurement, the latter question, the patient was asked to answer specific samples were kept for at least 5 min inside the NMR probe questions regarding the type of the side effect, the duration head for temperature equilibration (310 K serum, 300 K urine). and the intensity personally experienced, quantified by a score Samples were prepared and NMR spectra acquired following ranging from 0 (very weak) to 5 (strong). the procedures detailed by Vignoli et al. (Vignoli et al., 2019a). The standard of care of patients with ED includes a follow-up For each serum sample, three standard 1D 1H NMR spectra visit after approximately 3 months from basal evaluation. On that namely NOESY, CPMG (selective detection of low molecular occasion, patients enrolled in the study were re-evaluated for the weight metabolites), and Diffusion-edited (selective detection of erectile function by the IIEF-15 questionnaire, and the ADR- high molecular weight molecules) spectra were acquired. For each questionnaire was collected to obtain representative data of the urine sample a standard 1D 1H NOESY NMR spectrum was safety-efficacy profile of the therapy with PDE5i. We defined acquired. Before applying Fourier transform, raw NMR data were ΔIIEF-15ED as the difference between the score values of IIEF-15 multiplied by an exponential function equivalent to 0.3 Hz line- on questions focused on ED, at follow-up and at baseline. Patients broadening factor. Phase and baseline distortions were reporting the ineffectiveness of the drug, on the basis of their automatically corrected, and then transformed spectra were personal experience, were considered as non-responders. calibrated (TSP singlet at 0.00 ppm for urine, and glucose doublet at 5.24 ppm for serum) using TopSpin 3.2 Targeted Next Generation Sequencing (BrukerBiospin). Genomic DNA was extracted from peripheral blood leucocytes of subjects using QIAamp DNA Blood Mini Kit, according to the Statistical Analysis manufacturer’s protocol (Qiagen Inc., Hilden, Germany). The Statistical analysis was conducted with SPSS 21.0 for Windows quality of the DNA was determined using a NanoDrop-1000 (SPSS, Chicago, IL). The patients were divided on the basis of two (Thermo Fisher Scientific Inc, Waltham, MA, USA) and Qubit categorical endpoints: drug responders/non responders and 2.0 fluorometer (Thermo Fisher Scientific Inc, Waltham, MA, patients with or without ADR. Fisher’s exact test was used to USA). A Qubit dsDNA BR (broad range, 2–1,000 ng) Assay Kit compare the occurrence of each identified variant between and Qubit dsDNA HS (high sensitivity, 0.2–100 ng) Assay Kit subgroups. Furthermore, differences in continuous variables were used with a Qubit fluorometer according to the between subgroups were analyzed by Student’s t test. p values manufacturer’s protocol. < 0.05 were considered as statistically significant. Frontiers in Pharmacology | www.frontiersin.org 4 December 2020 | Volume 11 | Article 602369 Rocca et al. Patient Profiling in PDE5i Treatment TABLE 3 | Differential clustering of gene variants detected in 28 male patients with erectile dysfunction included in the study, grouped according to drug response of presence of adverse drug reaction. ID target SNP Responders Non-responders ADR NO ADR Variant PDE2A rs1980091 p  0.04; RR  2.7 [0.9–5.0] Synonymous variant PDE2A rs392565 p  0.02; RR  3.2 [1.1–4.6] Synonymous variant PDE2A rs426907 p  0.02; RR  3.2 [1.1–4.6] Intron variant PDE3A rs7966459 p  0.04; RR  2.7 [0.9–5.0] Intron variant PDE11A rs10201180 p  0.04; RR  2.7 [0.9–5.1] Splice region variant CYP2C9 rs1799853 p  0.02.RR  0.0 [0–1.02] Missense variant CYP2C9 rs9332119 p  0.02.RR  0.0 [0–1.02] Intron variant CYP2D7 rs56127449 p  0.04; RR  2.7 [0.9–5.1] Missense variant Abbreviations: DR, drug responsive; NDR, not drug responsive; ADR, adverse drug reactions; NADR, not adverse drug reactions; RR, Relative Risk (95% confidence intervals are reported within squared brackets). FIGURE 1 | Score plots of the first two principal components of the PCA analysis on the three serum NMR spectra acquired: (A) CPMG; (B) NOESY; (C) Diffusion. All Diffusion spectra are reported in panel (D), the two outlier samples are colored in red (one responder patient and one non-responder patient). The correlation between a continuous outcome and a set of All data analyses were performed using the “R” statistical continuous variables was evaluated by stepwise regression environment (R Core Team, 2014). Multivariate data analyses analysis, while the correlation with a dichotomous outcome were performed on processed binned data. Principal was evaluated with a logistic regression analysis. Component analysis was used as first exploratory analysis to All 1D NMR spectra, in the range of 0.02–10 ppm, were identify possible outliers. Random Forest (RF) algorithm binned into chemical shift bins of 0.02-ppm. The (Breiman, 2001) was used for classification. RF is a corresponding spectral areas (obtained using AssureNMR 2.2 classification algorithm that uses an ensemble of unpruned software, Bruker BioSpin) were used as input variables for decision trees (forest), each of which is built on a bootstrap multivariate statistical analysis. The region containing residual sample of the training data using a randomly selected subset of water signals (4.5–6.0 ppm) was removed from all datasets. Urine variables (bins) (Verikas et al., 2011; Touw et al., 2013). The can present large variations due to dilution, thus they were percentage of trees in the forest that assign one sample to a normalized using probabilistic quotient normalization (PQN; specific class can be inferred as a probability of belonging to a Dieterle et al., 2006), whereas serum spectra were not normalized. given class (Vignoli et al., 2019b). In our case, each tree was used Frontiers in Pharmacology | www.frontiersin.org 5 December 2020 | Volume 11 | Article 602369 Rocca et al. Patient Profiling in PDE5i Treatment FIGURE 2 | Confusion matrices and score plots of the first two principal components of the Random Forest models discriminating patients with erectile disfunction induced by organic causes (green) and those for whom the disease was related by other causes (orange) calculated on the NMR spectra acquired: (A) serum NOESY; (B) serum CPMG; (C) serum DIFFUSION; (D) urine NOESY. to predict whether a sample represents a patient that RESULTS experienced or not adverse effects, or a patient with ED due or not due to organic causes. For all calculations, the R package Of the 36 patients enrolled in the study, eight missed the follow- “Random Forest”4 was used to grow a forest of 3,000 trees, using up visit without giving notice or reasons and were therefore the default settings. excluded from the analyses. Twenty-eight patients, all of Quantification of metabolites (in both serum and urine) and Caucasian ethnicity, were finally included in the analysis and lipid fractions (in serum) was performed using the B.I. (Bruker were prescribed to take Sildenafil ODF, in particular Rabestrom® IVDr 2.0.2.) platform (Jiménez et al., 2018). Wilcoxon rank-sum (IBSA Farmaceutici Italia Srl, Lodi, Milan, Italy), the only test (Wilcoxon, 1945) was used to infer differences between the marketed Sildenafil ODF formulation available on the market metabolite/lipid levels of groups on the biological assumption at the time of the study. The demographic clinical characteristics that metabolite concentrations are not normally distributed. A of patients at baseline are reported in Table 1. p value < 0.05 was deemed significant. In this pilot study, we We found that the use of PDE5i for the treatment of ED in the chose to consider and discuss results not adjusted for multiple whole study group was associated with a significant increase of comparisons in order to decrease the risk of missing promising the mean IIEF-15ED score (respectively: 16.5 ± 6.8 units at basal vs biomarkers. However, we are aware that this could increase the 21.6 ± 6.9 units at follow-up; p  0.007). No significant increase risk of a type I error. Furthermore, for each analyzed variable was observed for the specific questions focused on the orgasm, (metabolites and lipids) the area under the receiver operating sexual desire, sexual satisfaction, and quality of life. characteristics curve (AUROC) was obtained using the R package At the follow-up visit, four out of 28 patients reported no drug “caTools.” response to PDE5i on-demand and, therefore, they were The prognostic capacity of the lipid fractions was evaluated considered as non-responders. Importantly, the mean calculating logistic regressions, using the standard R function prescribed drug dosage between responders and non- “glm.” responders was not statistically different (71.9 ± 38.4 mg vs. Frontiers in Pharmacology | www.frontiersin.org 6 December 2020 | Volume 11 | Article 602369 Rocca et al. Patient Profiling in PDE5i Treatment FIGURE 3 | Bins that mainly contribute to the discrimination between patients with erectile dysfunction induced by organic causes and those for whom the disease was related by other causes calculated on the NMR spectra acquired: (A) serum NOESY; (B) serum CPMG; (C) serum DIFFUSION; (D) urine NOESY. The importance is measured by the Gini index. 51.2 ± 43.9 mg, respectively; p  0.428), ruling out the posology as responders and non-responder patients. In addition, the a major bias in the outcome. Major concomitant drug therapies response to PDE5i as a dichotomous parameter was not were the use of glycemic lowering agents, lipid lowering agents, associated with any of the basal clinical parameters in the anti-hypertension agents and antiplatelet/anticoagulants logistic regression analysis. Similarly, ΔIIEF-15ED as a (respectively 3, 2, 5, and 2 out of 28 patients). In addition, 3 continuous variable was not significantly correlated with any out 28 patients reported a poly-therapy with two or more of the basal clinical parameter in the stepwise regression different drug categories. However, non-significant clustering analysis. of patients reporting concomitant drug therapies was detected The collection of the ADR questionnaire at follow-up showed between responders and non-responders. that 12 out of 28 patients reported side effects. In particular, 10 As expected, the IIEF-15ED score of non-responder patients patients had side effects directly related to vasodilation, such as at follow-up showed no significant increase, compared to headache, flushing, and tachycardia (respectively 7, 2, and 1 out of baseline (respectively: 13.0 ± 8.4 units at baseline vs. 23.3 ± 12), whereas two patients reported altered vision. Furthermore, in 7.9 units at follow-up; p  0.125). Differently, responders three out of 12 patients, the side effect was of such an extent to showed a significant increase of the mean value of IIEF-15ED discourage the further drug intake despite reporting an effective score (respectively: 17.1 ± 6.4 units at baseline vs. 21.3 ± 6.9 response to the drug. Accordingly, patients disclosing the units at follow-up; p  0.045). The comparison of basal clinical discontinuation of the therapy had a higher score of the side characteristics showed non obvious differences between effect intensity compared to those patients that continued the Frontiers in Pharmacology | www.frontiersin.org 7 December 2020 | Volume 11 | Article 602369 Rocca et al. Patient Profiling in PDE5i Treatment FIGURE 4 | Confusion matrices and score plots of the first two principal components of the Random Forest models discriminating patients that experienced (red dots) and not experienced (blued dots) adverse effects to the administration of PDE5 inhibitors calculated on the NMR spectra acquired: (A) serum NOESY; (B) serum CPMG; (C) serum DIFFUSION; (D) urine NOESY. adherence to the therapy (respectively: 3.7 ± 0.6 units vs. 0.7 ± 1.1 Additionally, 1,251 mapped within the coding regions. The units; p < 0.001). Considering the majority of vascular side effects, number of detected variants for each gene included in the patients reporting ADRs were considered as a whole group in the panel is reported in Table 2. subsequent analyses. Interestingly, a higher number of variants was observed in The comparison of clinical characteristics at baseline showed that PDE5i responders compared to non-responders (p < 0.001). patients reporting ADR had higher serum LDL levels (respectively: In addition, a higher number of variants was observed in 143.3 ± 13.2 mg/dl ADR vs. 133.1 ± 12.4 mg/dl No ADR; p  0.046). patients that did not show dyslipidemia at baseline, compared The presence of ADR as a dichotomous parameter was significantly with those patients showing dyslipidemia (p  0.04). associated with serum LDL levels at logistic regression analysis Finally, Table 3 shows the 8 single nucleotide variants that (Score  4.634; p  0.0319). However, this association was not showed a significant and specific clustering into one of the 4 maintained at stepwise regression analysis where ADR intensity was groups of patients showing a defined outcome: responders, non- considered as a continuous variable. responders, reporting ADR, reporting no ADR. In particular, Importantly, non-significant clustering of concomitant drug rs10201180 variant of PDE11A gene and rs56127449 variant of therapies was detected between patients that reported or not CYP2D7 gene were significantly more represented in responders ADR, albeit 31.3% of the latter used antihypertensive agents to PDE5i, with an equal relative risk of association of 2.7. On (p  0.0525 vs patients with ADR), ruling out a major role of the other hand, rs1980091, rs392565, and rs426907 variants of the underlying therapy in the onset of ADR. PDE2A gene, together rs7966459 variant of PDE3A, were significantly associated to the presence of ADR, with a relative Results of Molecular Analysis risk of association ranging from 2.7 to 3.2. Interestingly, A total of 2,629 genetic variants were detected in the target rs1799853 and rs9332119 variants of CYP2C9 gene regions. Of the total identified variants, 2,419 were single represented protective factors against ADR, being indeed more nucleotide variants and 210 small insertion/deletions. represented in patients reporting no ADR. Frontiers in Pharmacology | www.frontiersin.org 8 December 2020 | Volume 11 | Article 602369 Rocca et al. Patient Profiling in PDE5i Treatment TABLE 4 | Lipid analysis in serum samples from the 26 out of 28 patients with ED TABLE 4 | (Continued) Lipid analysis in serum samples from the 26 out of 28 included in the metabolomics analysis. patients with ED included in the metabolomics analysis. Lipid Fraction p-value AUROC Log2(FC) Lipid Fraction p-value AUROC Log2(FC) Triglycerides (TG) 0.019 0.793 0.422 LDL Subfractions Cholesterol LDL 3 0.065 0.736 0.481 Cholesterol (Chol) 0.019 0.793 0.417 LDL Subfractions Cholesterol LDL 4 0.019 0.793 0.590 LDL Cholesterol (LDL Chol) 0.028 0.777 0.541 LDL Subfractions Cholesterol LDL 5 0.028 0.777 0.711 HDL Cholesterol (HDL Chol) 0.401 0.612 0.163 LDL Subfractions Cholesterol LDL 6 0.040 0.760 0.582 Apo A1 0.193 0.669 0.193 LDL Subfractions Free Cholesterol LDL 1 0.057 0.744 0.399 Apo A2 0.212 0.661 0.294 LDL Subfractions Free Cholesterol LDL 2 0.171 0.678 0.424 Apo B100 0.040 0.760 0.480 LDL Subfractions Free Cholesterol LDL 3 0.040 0.760 0.361 Lipoprotein Main Fractions Triglycerides 0.007 0.835 0.675 LDL Subfractions Free Cholesterol LDL 4 0.034 0.769 0.529 VLDL LDL Subfractions Free Cholesterol LDL 5 0.040 0.760 0.619 Lipoprotein Main Fractions Triglycerides IDL 0.023 0.785 0.797 LDL Subfractions Free Cholesterol LDL 6 0.047 0.752 0.593 Lipoprotein Main Fractions Triglycerides LDL 0.108 0.707 0.349 LDL Subfractions Phospholipids LDL 1 0.401 0.612 0.190 Lipoprotein Main Fractions Triglycerides HDL 0.974 0.508 0.005 LDL Subfractions Phospholipids LDL 2 0.562 0.579 0.225 Lipoprotein Main Fractions Cholesterol VLDL 0.023 0.785 0.590 LDL Subfractions Phospholipids LDL 3 0.065 0.736 0.398 Lipoprotein Main Fractions Cholesterol IDL 0.023 0.785 0.912 LDL Subfractions Phospholipids LDL 4 0.019 0.793 0.520 Lipoprotein Main Fractions Cholesterol LDL 0.028 0.777 0.541 LDL Subfractions Phospholipids LDL 5 0.023 0.785 0.654 Lipoprotein Main Fractions Cholesterol HDL 0.401 0.612 0.163 LDL Subfractions Phospholipids LDL 6 0.034 0.769 0.518 Lipoprotein Main Fractions Free Cholesterol 0.016 0.802 0.512 LDL Subfractions Apo B LDL 1 0.293 0.636 0.203 VLDL LDL Subfractions Apo B LDL 2 0.401 0.612 0.253 Lipoprotein Main Fractions Free Cholesterol 0.013 0.810 0.951 LDL Subfractions Apo B LDL 3 0.040 0.760 0.427 IDL LDL Subfractions Apo B LDL 4 0.019 0.793 0.576 Lipoprotein Main Fractions Free Cholesterol 0.019 0.793 0.443 LDL Subfractions Apo B LDL 5 0.028 0.777 0.691 LDL LDL Subfractions Apo B LDL 6 0.047 0.752 0.522 Lipoprotein Main Fractions Free Cholesterol HDL 0.699 0.554 0.089 HDL Subfractions Triglycerides HDL 1 0.470 0.595 −0.311 Lipoprotein Main Fractions Phospholipids 0.019 0.793 0.594 HDL Subfractions Triglycerides HDL 2 0.652 0.562 −0.067 VLDL HDL Subfractions Triglycerides HDL 3 0.431 0.603 0.159 Lipoprotein Main Fractions Phospholipids 0.015 0.810 0.927 HDL Subfractions Triglycerides HDL 4 0.148 0.686 0.202 IDL HDL Subfractions Cholesterol HDL 1 0.847 0.529 0.005 Lipoprotein Main Fractions Phospholipids 0.034 0.769 0.450 HDL Subfractions Cholesterol HDL 2 0.844 0.529 0.108 LDL HDL Subfractions Cholesterol HDL 3 0.438 0.603 0.221 Lipoprotein Main Fractions Phospholipids HDL 0.606 0.570 0.099 HDL Subfractions Cholesterol HDL 4 0.101 0.711 0.270 Lipoprotein Main Fractions Apo A1 HDL 0.243 0.653 0.180 HDL Subfractions Free Cholesterol HDL 1 0.949 0.512 0.148 Lipoprotein Main Fractions Apo A2 HDL 0.243 0.653 0.280 HDL Subfractions Free Cholesterol HDL 2 0.606 0.570 0.223 Lipoprotein Main Fractions Apo B VLDL 0.034 0.769 0.537 HDL Subfractions Free Cholesterol HDL 3 0.332 0.628 0.403 Lipoprotein Main Fractions Apo B IDL 0.023 0.785 0.742 HDL Subfractions Free Cholesterol HDL 4 0.082 0.723 0.352 Lipoprotein Main Fractions Apo B LDL 0.047 0.752 0.481 HDL Subfractions Phospholipids HDL 1 0.478 0.595 −0.153 VLDL Subfractions Triglycerides VLDL 1 0.016 0.802 0.566 HDL Subfractions Phospholipids HDL 2 0.949 0.512 0.043 VLDL Subfractions Triglycerides VLDL 2 0.007 0.843 0.866 HDL Subfractions Phospholipids HDL 3 0.669 0.558 0.193 VLDL Subfractions Triglycerides VLDL 3 0.004 0.851 0.980 HDL Subfractions Phospholipids HDL 4 0.171 0.678 0.215 VLDL Subfractions Triglycerides VLDL 4 0.007 0.835 0.833 HDL Subfractions Apo A1 HDL 1 0.606 0.570 −0.194 VLDL Subfractions Triglycerides VLDL 5 0.797 0.537 0.082 HDL Subfractions Apo A1 HDL 2 0.797 0.537 0.100 VLDL Subfractions Cholesterol VLDL 1 0.438 0.603 0.264 HDL Subfractions Apo A1 HDL 3 0.438 0.603 0.176 VLDL Subfractions Cholesterol VLDL 2 0.022 0.793 0.677 HDL Subfractions Apo A1 HDL 4 0.076 0.727 0.247 VLDL Subfractions Cholesterol VLDL 3 0.006 0.851 1.068 HDL Subfractions Apo A2 HDL 1 0.844 0.529 0.073 VLDL Subfractions Cholesterol VLDL 4 0.019 0.793 0.889 HDL Subfractions Apo A2 HDL 2 0.699 0.554 0.328 VLDL Subfractions Cholesterol VLDL 5 0.768 0.541 0.048 HDL Subfractions Apo A2 HDL 3 0.401 0.612 0.344 VLDL Subfractions Free Cholesterol VLDL 1 0.088 0.719 0.721 HDL Subfractions Apo A2 HDL 4 0.040 0.760 0.300 VLDL Subfractions Free Cholesterol VLDL 2 0.019 0.793 0.741 VLDL Subfractions Free Cholesterol VLDL 3 0.009 0.835 1.058 A positive Log2(FC) value means higher level in patients that experienced adverse drug reactions. Abbreviations: AUROC, Area under the receiver operating characteristics curve. VLDL Subfractions Free Cholesterol VLDL 4 0.013 0.818 1.040 Associations with p values <0.05 are reported in bold. VLDL Subfractions Free Cholesterol VLDL 5 0.554 0.579 −0.174 VLDL Subfractions Phospholipids VLDL 1 0.047 0.752 0.605 VLDL Subfractions Phospholipids VLDL 2 0.005 0.843 0.803 VLDL Subfractions Phospholipids VLDL 3 0.003 0.876 1.012 Evaluation of the Metabolic Profiles by VLDL Subfractions Phospholipids VLDL 4 0.017 0.806 0.720 Metabolomics Analysis VLDL Subfractions Phospholipids VLDL 5 0.430 0.603 0.134 Principal Component Analysis of Serum and Urine LDL Subfractions Triglycerides LDL 1 0.652 0.562 0.114 LDL Subfractions Triglycerides LDL 2 0.040 0.760 0.319 Samples LDL Subfractions Triglycerides LDL 3 0.622 0.566 0.147 Principal component analysis (PCA) was used as first exploratory LDL Subfractions Triglycerides LDL 4 0.101 0.711 0.395 analysis. Two serum samples shown to be outliers in all the three LDL Subfractions Triglycerides LDL 5 0.033 0.773 0.534 kind of NMR spectra acquired (Figures 1A–C). Both serum LDL Subfractions Triglycerides LDL 6 0.049 0.752 0.354 samples (Figure 1D) had peculiar high levels of lipids and LDL Subfractions Cholesterol LDL 1 0.300 0.636 0.261 LDL Subfractions Cholesterol LDL 2 0.519 0.587 0.257 lipoproteins and, interestingly, both patients showed the highest (Continued in next column) BMI values (BMI > 31). Consequently, both samples were removed Frontiers in Pharmacology | www.frontiersin.org 9 December 2020 | Volume 11 | Article 602369 Rocca et al. Patient Profiling in PDE5i Treatment TABLE 5 | Serum metabolite univariate analysis. TABLE 6 | Metabolite analysis in urine samples. P-value AUROC Log2(FC) p-value AUROC Log2(FC) Alanine 0.308 0.632 0.077 Acetic acid 0.671 0.548 0.940 Creatinine 0.687 0.554 0.109 Acetoacetic acid 0.448 0.557 −1.339 Glutamic acid 0.528 0.579 0.145 Acetone 0.098 0.685 −1.030 Glutamine 0.375 0.616 0.133 Alanine 0.865 0.523 0.032 Glycine 0.717 0.550 0.064 Betaine 0.604 0.563 −0.202 Histidine 0.344 0.620 0.144 Citric acid 0.368 0.608 0.218 Isoleucine 0.612 0.566 0.065 Creatinine 0.481 0.585 0.064 Leucine 0.146 0.686 0.241 D-Glucose 0.249 0.605 1.042 Lysine 0.817 0.533 −0.199 Formic acid 0.276 0.625 −0.428 Ornithine 0.573 0.574 0.090 Glycine 0.336 0.614 −0.242 Phenylalanine 0.402 0.607 0.215 Hippuric acid 0.486 0.580 0.127 Threonine 0.009 0.831 0.621 Methanol 0.535 0.571 −0.099 Tyrosine 0.893 0.521 0.022 Methylmalonic acid 0.812 0.523 0.113 Valine 1.000 0.504 0.040 N-Isovaleroylglycine 0.812 0.523 0.216 Acetic acid 0.916 0.517 −0.038 N-N-Dimethylglycine 0.297 0.622 −0.739 Citric acid 0.895 0.521 0.074 Oxaloacetic acid 0.059 0.696 −1.742 Formic acid 0.503 0.587 0.135 Oxypurinol 0.120 0.614 3.862 Lactic acid 0.393 0.612 0.020 Proline betaine 0.520 0.557 −1.698 3-Hydroxybutyric acid 0.099 0.702 −1.728 Propylene glycol 0.497 0.551 −2.066 Acetone 0.588 0.570 0.142 Succinic acid 0.110 0.676 −0.968 Pyruvic acid 0.387 0.612 −0.096 Tartaric acid 0.314 0.614 −1.097 Glucose 0.278 0.640 0.145 Trigonelline 0.974 0.506 −0.048 Uracil 0.480 0.580 0.467 A positive Log2(FC) means higher level in patients that experienced adverse effects. Valine 0.361 0.608 0.325 Associations with p values <0.05 are reported in bold. A positive Log2(FC) means higher level in patients that experienced adverse effects. from the study group for the following analyses. On the contrary, no evident clustering neither outlier was present in urine and thus, all samples were considered in the modeling. Univariate Analysis for the Discrimination Between Patients That Reported or did not Report Adverse Multivariate Discrimination Between Patients with Drug Reactions Erectile Dysfunction Associated with Organic Causes Univariate metabolite analysis enables the identification of Compared to All Other Causes several differences between patients that reported or not ADR Random Forest models built using serum NOESY (54.5% accuracy, after the administration of PDE5i in serum lipid fractions. Results 54.5% sensitivity, 54.5% specificity), CPMG (59.1% accuracy, 54.5% are reported in Table 4. Conversely, only the levels of threonine sensitivity, 63.6% specificity), Diffusion (63.6% accuracy, 63.6% showed statistically significant differences in serum and no sensitivity, 63.6% specificity) spectra, and urine NOESY spectra significant difference was observed for low molecular weight (55.6% accuracy, 33.3% sensitivity, 73.3% specificity) did not show metabolites in urine (respectively, Tables 5, 6). any obvious difference between patients with ED associated with Since the serum lipid fractions showed the best prognostic organic causes and those for whom the disease was related to other values for ADR, hazard ratios (HR) for the seven main lipid causes (Figures 2A–D). The bins that mainly contribute to the group fractions (namely triglycerides, total cholesterol, LDL-cholesterol, discrimination in each model are reported in Figure 3. These results HDL-cholesterol, ApoA1, ApoA2, ApoB100) were calculated, suggest that the patient’s metabolic profiles, evaluated by metabolomics setting each respective median value as arbitrary threshold. As approach, are not significantly influenced by the disease etiology. reported in Table 7, triglycerides, total cholesterol, and LDL- cholesterol showed the most significant association, with LDL- Multivariate Discrimination Between Patients That cholesterol having the highest hazard ratio. Reported or did not Report Adverse Drug Reactions Random Forest models (Figures 2A–C) built using NOESY (72.7% accuracy, 81.8% sensitivity, 63.6% specificity), CPMG (72.7% DISCUSSION accuracy, 72.7% sensitivity, 72.7% specificity), and Diffusion (77.3% accuracy, 81.8% sensitivity, 72.7% specificity) spectra This is the first study focusing on the early identification of showed statistically significant discrimination accuracy between markers of efficacy and safety of PDE5i, used as the first-choice patients that reported or not ADR after the administration of treatment of ED, through the analysis of the clinical phenotype of PDE5 inhibitors. Conversely, only slight discrimination (59.3% patients in association with their genetic profile, concerning the accuracy, 36.4% sensitivity, 75.0% specificity) between these two genes involved in PDE5i pharmacodynamics and drug groups was present in the urine analysis (Figure 4D). The bins that metabolism, and the metabolic profile of serum and urine mainly contribute to the group discrimination in each model are assessed by unbiased metabolomics analysis. We found that reported in Supplementary Figure S3. the pharmacological response to Sildenafil is essentially related Frontiers in Pharmacology | www.frontiersin.org 10 December 2020 | Volume 11 | Article 602369 Rocca et al. Patient Profiling in PDE5i Treatment FIGURE 5 | Bins that mainly contribute to the discrimination between patients that experienced and not experienced adverse effects to the administration of PDE5 inhibitors calculated on the NMR spectra acquired: (A) serum NOESY; (B) serum CPMG; (C) serum DIFFUSION; (D) urine NOESY. The importance is measured by the Gini index. to the genetic background of the individual, particularly to the activity towards other cGMP-dependent phosphodiesterases pattern of variants localized in genes belonging to the PDE family. (Bischoff, 2004). In particular, the expression of PDE3A in the On the other hand, the occurrence of ADR showed a mixed gastrointestinal tissue, cardiovascular system, and brain has been correlation both with the genetic profile, particularly with genes linked with headache, flushing, and dyspepsia associated with the involved in hepatic drug metabolism, and importantly with the use of PDE5i (Goldstein et al., 1998; Wallis et al., 1999). We found clinical phenotype of the patients, being largely related with the significant clustering of genetic variants rs392565 and rs426907 of serum LDL-lipid profile. PDE2A gene and rs7966459 variant of PDE3A gene, respectively, In this study, the possible genetic base of the efficacy/safety in patients reporting ADR, further supporting this model. profile PDE5i was assessed through the sequencing of genes Interestingly, the rs10201180 variant in PDE11A gene, involved in pharmacodynamics and pharmacokinetics of this corresponding to a splice region variant, was significantly class of drugs. The occurrence of ADR associated with the use associated with the responding phenotype in our cohort of ED of Sildenafil has been classically ascribed to the side inhibitory patients. PDE11A owns a mixed activity both on cAMP and Frontiers in Pharmacology | www.frontiersin.org 11 December 2020 | Volume 11 | Article 602369 Rocca et al. Patient Profiling in PDE5i Treatment TABLE 7 | Association between the main lipid fractions, obtained by pharmacokinetics both in animal and in human models (Wójcicki metabolomics analysis, and the occurrence of adverse drug reactions in 26 et al., 1999; Wang et al., 2019). Therefore, we could speculate that the out of 28 patients with ED included in the analysis. altered hepatic function, underlying the increased lipid profiles, Main Lipid Fraction HR p-value associates with individual variations of the drug bio-availability and, in turn, with the occurrence of ADR as recently observed in Triglycerides, TG ≥ 80 7.111 0.040 Cholesterol, Chol ≥ 190 7.111 0.040 human models (Muirhead et al., 2002; Gao et al., 2015). In the LDL Cholesterol, LDL Chol ≥ 100 17.500 0.019 present study, serum sampling was performed only at the patient’s HDL Cholesterol, HDL Chol ≥ 50 0.686 0.665 recruitment and prior to any prescription of the drug, thus it was Apo A1, Apo A1 ≥ 140 1.458 0.665 impossible to retrieve any data of sildenafil concentration from Apo A2, Apo A2 ≥ 30 1.458 0.665 Apo B100, Apo B100 ≥ 80 3.200 0.200 available serum samples. Further studies will warrant the validation of this model through the quantification of serum levels of sildenafil Abbreviations: HR, hazard ratio. in patients with altered metabolic phenotype. Importantly, PDE5i Risk threshold for each lipid fractions were arbitrarily set at each median value. Associations with p values <0.05 are reported in bold. have well-defined pharmacodynamics and are prescribed for the use on demand (Hackett, 2005). This reduces any possible bias resulting from the mutual influence of the drug activity on the metabolic cGMP, and the side inhibition on PDE11 has been typically phenotype of patient. associated with the back and muscle pain ascribed to the The treatment of ED represents a peculiar case of a treatment with Tadalafil, another PDE5i (Bischoff, 2004). This pharmacological success. The rising incidence of metabolic evidence deserves further insights, suggesting a possible role of diseases together with the growing base of the geriatric this protein as a co-target in the pharmacodynamic of Sildenafil. population, both representing important risk factors for ED, On the other hand, the association of the occurrence of ADR with strongly nourished drug demand from patients. As a result, genetic variants of cytochromes involved in drug metabolism the market size for ED drugs in 2014 accounted for more than appears as novel in regard of the treatment with PDE5i. This is 4.39 billion USD (Grand View Research, 2016). However, because particularly the case of rs1799853, a missense variant of CYP2C9 of the aforementioned unfavorable efficacy/safety profile, patients gene, predicted to be pathogenic by PolyPhen-2, and previously frequently chose to prematurely discontinue or switch to other associated with a reduced drug metabolic function of the CYP2C9 drugs for ED, with unnecessary resource utilization (Harnett cytochrome (Helin et al., 2019; Calderon-Ospina et al., 2020). et al., 2006). Since ED burdens a severe impact on the quality of Although the potential pharmacokinetic impact of the identified life and interpersonal relationships, the proper tailoring of its genetic variants on the phenotype goes beyond the aim of this treatment has been claimed as a primary health issue (Mulhall study, this evidence suggests that the occurrence of ADR may rely et al., 2005; Corona et al., 2011; Corona et al., 2016). In this on individual fluctuations of the PDE5i serum levels due to context, our results suggest the evaluation of serum lipid profile as functional variations of the enzymes involved in drugs a work-up procedure to be added in the clinical evaluation of ED metabolism. patients. This would reasonably help the identification of patients Analogous considerations can be drawn from the results of at risk of ADR and likely addresses toward lower dosages of metabolic profiling. To date, few studies have correlated the Sildenafil or drive the choice toward other PDE5i. serum lipid profile with the response to PDE5i with non- Although the small sample size of the study cohort and the conclusive results. In a study performed on 45 patients with ED, exclusive inclusion of Sildenafil as PDE5i represent limitations of Solomon et al. 2006 showed that the acute vascular response to this study, there are nonetheless some strengths to underline. Sildenafil, evaluated by photoplethysmography-based technique, These are the thorough clinical characterization of patients and negatively correlated with basal lipoprotein levels (Solomon et al., the independent evaluation of both the genetic and phenotypic 2006). Furthermore, a placebo-controlled randomized clinical trial profile through two new high-throughput techniques. To the best including 118 non-responders to Sildenafil showed that that the of our knowledge, this renders the present study unprecedented. adjunct treatment with atorvastatin, resulting in serum cholesterol In conclusion, by the use of an integrated evaluation reduction, was associated with a significant increase of the IIEF-15ED comprising the clinical phenotype and the genomic/metabolic score, particularly in those patients with moderate/severe ED profiling of ED patients, in this study we provide evidence that the (Dadkhah et al., 2010). Conversely, dyslipidemia did not occurrence of ADR associated with the use of PDE5i is strictly significantly impact the response to Vardenafil as described by linked to the serum lipid pattern, in addition to the known side Miner et al. in a Post-hoc subgroup analysis designed to address inhibition effect on other phosphodiesterase. These elements can the influence of serum lipid levels and the presence of metabolic help clinicians in the tailoring of individual therapies for ED syndrome (Miner et al., 2010). However, neither of the three studies patients, corroborating the favorable effect of personalized considered the occurrence of ADR in the overall response to PDE5i medicine. that, in the present study, was mostly predicted by the elevated serum levels of LDL, as depicted by both clinical data and the untargeted analysis of serum metabolites by NMR. Of note, the DATA AVAILABILITY STATEMENT energy metabolism phenotype has been shown to significantly alter the local expression of cytochromes and transporters involved in Sequence data are available at SRA repository with the following hepatic drug metabolism, impacting the global pattern of the drug url link. https://www.ncbi.nlm.nih.gov/bioproject/PRJNA674473. Frontiers in Pharmacology | www.frontiersin.org 12 December 2020 | Volume 11 | Article 602369 Rocca et al. Patient Profiling in PDE5i Treatment ETHICS STATEMENT FUNDING The studies involving human participants were reviewed and This project has received funding from the European Union’s approved by Ethic Committee for Clinical Trials of the Horizon 2020 research and innovation programme under grant University-Hospital of Padova (Italy). The patients/participants agreement No 654248. Alessia Vignoli is supported by an AIRC- provided their written informed consent to participate in Foundation for Cancer Research fellowship for Italy. this study. ACKNOWLEDGMENTS AUTHOR CONTRIBUTIONS We thank Dr Antonella Di Mambro, Pierfrancesco Palego, Nicola MSR performed the analysis of genomic data and drafted the Caretta and Gabriel Cosmin Petre for patient management and manuscript, AV performed NMR experiments and drafted the samples processing. manuscript, LT critically discussed NMR data, MG, and MDRP performed patients recruitment and evaluation, GV and DT performed NGS sequencing, RP, and CF critically evaluated SUPPLEMENTARY MATERIAL the manuscript, LDT conceived the study, performed the analysis of clinical data and drafted the manuscript. All The Supplementary Material for this article can be found online authors contributed to the manuscript. All authors read and at: https://www.frontiersin.org/articles/10.3389/fphar.2020.602369/ approved the final manuscript. full#supplementary-material. Goldstein, I., Lue, T. F., Padma-Nathan, H., Rosen, R. C., Steers, W. D., and Wicker, REFERENCES P. A. (1998). Oral sildenafil in the treatment of erectile dysfunction. Sildenafil Study Group. N. Engl. J. Med. 338 (20), 1397–1404. doi:10.1056/ Bischoff, E. (2004) Potency, selectivity, and consequences of non selectivity of PDE NEJM199805143382001 inhibition. Int. J. Impot. Res. 16 (Suppl. 1), S11–S14. doi:10.1038/sj.ijir.3901208 Grand View Research (2016). Erectile dysfunction drugs market analysis by Breiman, L. (2001) Random forests. Mach. Learn. 45, 5–32. doi:https://doi.org/10. product (Viagra, Levitra/Staxyn, Stendra/Spedra, Zydena, Vitaros), and 1023/A:1010933404324 segment forecasts to 2022. San Francisco, CA, United States Available at: Calderon-Ospina, C. A., Galvez, J. M., López-Cabra, C., Morales, N., Restrepo, C. https://www.grandviewresearch.com/industry-analysis/erectile-dysfunction- M., Rodríguez, J., et al. (2020). Possible genetic determinants of response to drugs-market/methodology (Accessed July 2016). phenytoin in a group of Colombian patients with epilepsy Front. Pharmacol. 11, Greenberg, D. R., Richardson, M. T., Tijerina, J. D., Bass, M. B., and Eisenberg, 555. doi:10.3389/fphar.2020.00555 M. L. (2019). The quality of systematic reviews and meta-analyses in erectile Caretta, N., De Rocco Ponce, M., Minicuci, N., Palego, P., Valente, U., Garolla, dysfunction treatment and management published in the sexual medicine A., et al. (2019). Penile doppler ultrasound predicts cardiovascular events literature. J. Sex. Med. 16 (3), 394–401. doi:10.1016/j.jsxm.2019.01.009 in men with erectile dysfunction. Andrology 7 (1), 82–87. doi:10.1111/andr. Hackett, G. I. (2005). Patient preferences in treatment of erectile dysfunction: the 1256 continuing importance of patient education Clin. Cornerstone 7 (1), 57–65. Corona, G., Rastrelli, G., Burri, A., Serra, E., Gianfrilli, D., Mannucci, E., et al. doi:10.1016/s1098-3597(05)80049-3 (2016). First-generation phosphodiesterase type 5 inhibitors dropout: a Harnett, J. P., McLaughlin, T. P., Mulhall, J. P., Russell, D., and McLean, S. (2006). comprehensive review and meta-analysis. Andrology 4 (6), 1002–1009. The economic effect of switching from sildenafil to another phosphodiesterase doi:10.1111/andr.12255 type 5 inhibitor. Manag. Care Interface 19 (11), 22–36. Corona, G., Mondaini, N., Ungar, A., Razzoli, E., Rossi, A., and Fusco, F. (2011). Hawksworth, D. J., and Burnett, A. L. (2015). Pharmacotherapeutic management Phosphodiesterase type 5 (PDE5) inhibitors in erectile dysfunction: the proper of erectile dysfunction. Clin. Pharmacol. Ther. 98 (6), 602–610. doi:10.1002/cpt. drug for the proper patient. J. Sex. Med. 8 (12), 3418–3432. doi:10.1111/j.1743- 261 6109.2011.02473.x Helin, T. A., Joutsi-Korhonen, L., Asmundela, H., Niemi, M., Orpana, A., and Dadkhah, F., Safarinejad, M. R., Asgari, M. A., Hosseini, S. Y., Lashay, A., and Lassila, R. (2019). Warfarin dose requirement in patients having severe Amini, E. (2010). Atorvastatin improves the response to sildenafil in thrombosis or thrombophilia. Br. J. Clin. Pharmacol. 5 (8), 1684–1691. hypercholesterolemic men with erectile dysfunction not initially responsive doi:10.1111/bcp.13948 to sildenafil. Int. J. Impot. Res. 22 (1), 51–60. doi:10.1038/ijir.2009.48 Issue Information-Declaration of Helsinki (2018). Issue Information-Declaration De Toni, L., De Rocco Ponce, M., Franceschinis, E., Dall’Acqua, S., Padrini, R., of Helsinki. J. Bone Miner. Res. 33 (12), BM i–BMs ii. doi:10.1002/jbmr.3277 Realdon, N., et al. (2018). Sublingual administration of sildenafil oro-dispersible Jiménez, B., Holmes, E., Heude, C., Tolson, R. F., Harvey, N., Lodge, S. L., et al. film: new profiles of drug tolerability and pharmacokinetics for PDE5 (2018). Quantitative lipoprotein subclass and low molecular weight metabolite inhibitors. Front. Pharmacol. 9, 59. doi:10.3389/fphar.2018.00059 analysis in human serum and plasma by 1H NMR spectroscopy in a Dieterle, F., Ross, A., Schlotterbeck, G., and Senn, H. (2006). Probabilistic quotient multilaboratory trial. Anal. Chem. 90 (20), 11962–11971. doi:10.1021/acs. normalization as robust method to account for dilution of complex biological analchem.8b02412 mixtures. Application in 1H NMR Metabonomics. Anal. Chem. 78 (13), Lacchini, R., and Tanus-Santos, J. E. (2014). Pharmacogenetics of erectile 4281–4290. doi:10.1021/ac051632c dysfunction: navigating into uncharted waters. Pharmacogenomics 15 (11), Gao, X., Ndongo, M. N., Checchio, T. M., Cook, J., Duncan, B., and LaBadie, 1519–1538. doi:10.2217/pgs.14.110 R. R. (2015). A randomized, open-label 3-way crossover study to Miner, M. M., Barnes, A., and Janning, S. (2010). Efficacy of phosphodiesterase investigate the relative bioavailability and bioequivalence of crushed type 5 inhibitor treatment in men with erectile dysfunction and dyslipidemia: a sildenafil 20 mg tablets mixed with apple sauce, extemporaneously post hoc analysis of the vardenafil statin study. J. Sex. Med. 7 (5), 1937–1947. prepared suspension (EP), and intact sildenafil 20 mg tablets in healthy doi:10.1111/j.1743-6109.2010.01766.x volunteers under fasting conditions. Clin. Pharmacol. Drug Dev. 4 (1), Muirhead, G. J., Wilner, K., Colburn, W., Haug-Pihale, G., and Rouviex, B. (2002). 74–80. doi:10.1002/cpdd.146 The effects of age and renal and hepatic impairment on the pharmacokinetics of Frontiers in Pharmacology | www.frontiersin.org 13 December 2020 | Volume 11 | Article 602369 Rocca et al. Patient Profiling in PDE5i Treatment sildenafil. Br. J. Clin. Pharmacol. 53 (Suppl. 1), 21S–30S. doi:10.1046/j.0306- within two years after acute myocardial infarction in the AMI-Florence II 5251.2001.00029.x cohort. BMC Med. 17 (1), 3. doi:10.1186/s12916-018-1240-2 Mulhall, J. P., McLaughlin, T. P., Harnett, J. P., Scott, B., Burhani, S., and Russell, D. Wallis, R. M., Corbin, J. D., Francis, S. H., and Ellis, P. (1999). Tissue distribution of (2005). Medication utilization behavior in patients receiving phosphodiesterase phosphodiesterase families and the effects of sildenafil on tissue cyclic type 5 inhibitors for erectile dysfunction. J. Sex. Med. 2 (6), 848–855. doi:10. nucleotides, platelet function, and the contractile responses of trabeculae 1111/j.1743-6109.2005.00149.x carneae and aortic rings in vitro. Am. J. Cardiol. 83 (5A), 3C–12C. doi:10. Nehra, A., Jackson, G., Miner, M., Billups, K. L., Burnett, A. L., Buvat, J., et al. 1016/s0002-9149(99)00042-9 (2012). The princeton III consensus recommendations for the management of Wang, Z., Yang, H., Xu, J., Zhao, K., Chen, Y., Liang, L., et al. (2019). Prediction of erectile dysfunction and cardiovascular disease. Mayo Clin. Proc. 87 (8), atorvastatin pharmacokinetics in high-fat diet and low-dose streptozotocin- 766–778. doi:10.1016/j.mayocp.2012.06.015 induced diabetic rats using a semiphysiologically based pharmacokinetic model Nieuwlaat, R., Wilczynski, N., Navarro, T., Hobson, N., Jeffery, R., Keepanasseril, involving both enzymes and transporters. Drug Metab. Dispos. 47 (10), A., et al. (2014). Interventions for enhancing medication adherence. Cochrane 1066–1079. doi:10.1124/dmd.118.085902 Database Syst. Rev. 2014 (11), CD000011. doi:10.1002/14651858.CD000011 Widmann, P., Reverter, A., Fortes, M. R., Weikard, R., Suhre, K., Hammon, H., Osterberg, L., and Blaschke, T. (2005). Adherence to medication. N. Engl. J. Med. et al. (2013). A systems biology approach using metabolomic data reveals genes 353 (18), 1972–1974. doi:10.1056/NEJMra050100 and pathways interacting to modulate divergent growth in cattle. BMC Genom. R Core Team (2014). R: a language and environment for statistical computing. 14, 798. doi:10.1186/1471-2164-14-798 Vienna, Austria: R Foundation for Statistical Computing. Wilcoxon, F. (1945). Individual comparisons by ranking methods. Biom. Bull. 1, Solomon, H., Wierzbicki, A. S., Lumb, P. J., Lambert-Hammill, M., and Jackson, G. 80. doi:10.1007/978-1-4612-4380-9_16 (2006). Cardiovascular risk factors determine erectile and arterial function Wójcicki, J., Sulzyc-Bielicka, V., Kutrzeba, J., Gawrońska-Szklarz, B., Droździk, M., response to sildenafil. Am. J. Hypertens. 19 (9), 915–919. doi:10.1016/j. and Sterna, Z. (1999). Studies on the pharmacokinetics and pharmacodynamics amjhyper.2006.02.010 of propranolol in hyperlipidemia. J. Clin. Pharmacol. 39 (8), 826–833. doi:10. Taylor, J., Baldo, O. B., Storey, A., Cartledge, J., and Eardley, I. (2009). Differences 1177/00912709922008498 in side-effect duration and related bother levels between phosphodiesterase type Yafi, F. A., Jenkins, L., Albersen, M., Corona, G., Isidori, A. M., Goldfarb, S., et al. (2016). 5 inhibitors. BJU Int. 103 (10), 1392–1395. doi:10.1111/j.1464-410X.2008. Erectile dysfunction. Nat. Rev. Dis. Primers 2, 16003. doi:10.1038/nrdp.2016.3 08328.x Zullig, L. L., Blalock, D. V., Dougherty, S., Henderson, R., Ha, C. C., Oakes, M. M., Touw, W. G., Bayjanov, J. R., Overmars, L., Backus, L., Boekhorst, J., Wels, M., et al. (2018). The new landscape of medication adherence improvement: where et al. (2013). Data mining in the life sciences with random forest: a walk in the population health science meets precision medicine. Patient Prefer. Adherence park or lost in the jungle? Brief. Bioinform. 14 (3), 315–326. doi:10.1093/bib/ 12, 1225–1230. doi:10.2147/PPA.S165404 bbs034 van Dulmen, S., Sluijs, E., van Dijk, L., de Ridder, D., Heerdink, R., and Bensing, J. Conflict of Interest: The authors declare that the research was conducted in the (2007). Patient adherence to medical treatment: a review of reviews. BMC absence of any commercial or financial relationships that could be construed as a Health Serv. Res. 7, 55. doi:10.1186/1472-6963-7-55 potential conflict of interest. Verikas, A., Gelzinis, A., and Bacauskiene, M. (2011). Mining data with random forests: a survey and results of new tests. Pattern Recognit. 44 (2), 330–349. Copyright © 2020 Rocca, Vignoli, Tenori, Ghezzi, De Rocco Ponce, Vatsellas, doi:10.1016/j.patcog.2010.08.011 Thanos, Padrini, Foresta and De Toni. This is an open-access article distributed Vignoli, A., Ghini, V., Meoni, G., Licari, C., Takis, P. G., Tenori, L., et al. (2019a). under the terms of the Creative Commons Attribution License (CC BY). The use, High-throughput metabolomics by 1D NMR. Angew. Chem. Int. Ed Engl. 58 distribution or reproduction in other forums is permitted, provided the original (4), 968–994. doi:10.1002/anie.201804736 author(s) and the copyright owner(s) are credited and that the original publication Vignoli, A., Tenori, L., Giusti, B., Takis, P. G., Valente, S., Carrabba, N., et al. in this journal is cited, in accordance with accepted academic practice. No use, (2019b). NMR-based metabolomics identifies patients at high risk of death distribution or reproduction is permitted which does not comply with these terms. Frontiers in Pharmacology | www.frontiersin.org 14 December 2020 | Volume 11 | Article 602369