Innovative Diagnostic Methods for Early Prostate Cancer Detection through Urine Analysis: A Review

Prostate cancer is the second most common cause of cancer death among men. It is an asymptomatic and slow growing tumour, which starts occurring in young men, but can be detected only around the age of 40–50. Although its long latency period and potential curability make prostate cancer a perfect candidate for screening programs, the current procedure lacks in specificity. Researchers are rising to the challenge of developing innovative tools able of detecting the disease during its early stage that is the most curable. In recent years, the interest in characterisation of biological fluids aimed at the identification of tumour-specific compounds has increased significantly, since cell neoplastic transformation causes metabolic alterations leading to volatile organic compounds release. In the scientific literature, different approaches have been proposed. Many studies focus on the identification of a cancer-characteristic “odour fingerprint” emanated from biological samples through the application of sensorial or senso-instrumental analyses, others suggest a chemical characterisation of biological fluids with the aim of identifying prostate cancer (PCa)-specific biomarkers. This paper focuses on the review of literary studies in the field of prostate cancer diagnosis, in order to provide an overview of innovative methods based on the analysis of urine, thereby comparing them with the traditional diagnostic procedures.


Introduction
Prostate cancer (PCa) is the most common diagnosed cancer in Europe and America and the second most common cause of cancer death among men [1]. The National Cancer Institute estimates 161,360 new cases of PCa and 26,730 deaths in 2017 [2].
PCa is an asymptomatic tumour, which starts occurring in men aging 20-30 years, but can be detected only in the fourth-fifth decade [3]. Indeed, symptoms appear only when the disease has reached an advanced stage, reducing the number of adoptable treatments and patients' chances of surviving [4].
The long latency period of PCa and its potential curability in early stages make this disease a perfect candidate for screening programs [5]. Nevertheless, PCa diagnosis is challenging because of the late onset of symptoms and the limits of the current diagnostic procedures. and high variability among patients depending on gender, age, hormonal status, diet, or physical activity [50][51][52][53].
This paper focuses on the review of literary studies in the field of medical diagnostics about innovative methods for early PCa detection based on the analysis of urine, considering a timeframe of publication between 2008 and 2017. In particular, this review aims to provide an overview of innovative methods for PCa diagnosis, thereby comparing them with the traditional diagnostic procedures.
The following sections provide a general overview of traditional diagnostic procedures (Section 2) and of the state-of-the-art of alternatives proposed in recent years (Section 3). Innovative techniques considered are grouped by type of study, according to the method adopted: sensorial methods involving the use of high-trained dogs (Section 3.1.1), senso-instrumental methods based on the adoption of electronic noses (Section 3.1.1), and analytical techniques based on the chemical characterisation of urine (Section 3.2).
Given the huge number of possible techniques for the chemical characterisation of urine samples [54], in this review, only approaches based on liquid or gas chromatography and mass spectrometry were considered. These analytic techniques are the most consolidated and allow detecting metabolites at very low concentrations with high sensitivity [54], although recent studies based on magnetic resonance seem to be very promising, especially for the analysis of prostatic fluid or tissue [55][56][57].

Overview of Traditional Diagnostic Methods
This section aims to give a general overview of the traditional diagnostic procedure for PCa detection. The detailed description of the current diagnostic methods falls out of the scope of this paper, besides, it is already object of other exhaustive reviews, papers, and books [58][59][60][61][62]; for this reason, this section is limited to highlight the main issues of current diagnostics for further comparison with the investigated innovative techniques.
Current screening for PCa is based on the measurement of prostate specific antigen (PSA) serum level. The PSA is a 33 kDa serine protease responsible for the controlled release of sperm, which is secreted into the fluid of the glandular ducts [3,63]. In a normal prostate, only small amounts (i.e., <4 ng/mL) of PSA reach the circulation system by leaking backwards into extracellular fluid and diffusing into circulation [3]. Conversely, in case of PCa, PSA serum level is higher, because of the derangement of epithelial cells architecture and polarisation, which causes the loss of normal secretory pathways into the prostatic ducts [3].
Literature studies about PCa development [64,65] demonstrated that the concentration of serum PSA and its doubling time or PSA velocity reflect the grow rate of PCa. However, PSA test is characterised by low diagnostic accuracy (i.e., specificity around 33% and sensitivity around 86%) [9]. Indeed, higher PSA serum levels may be associated with other non-cancer related diseases, such as prostatitis, irritations, benign prostatic hypertrophy (BHP), or also diet alterations [5]. Moreover, PSA falsely diagnoses indolent PCa, which tends to grow and spread slowly. Thus, the lack of diagnostic accuracy associated with this method often results in patients' overtreatment, and increase of health spending [66,67]. Approximately 10% of men over 50 years old have a PSA serum levels over 4 ng/mL, but among them, only 30% have a PCa confirmed by further investigations [3]. Wilt et al. [68] proved that, after positive PSA test, patients who underwent radical prostatectomy did not have a significant reduction in mortality with respect to those who opted for active surveillance over a 12 year follow up.
Since PSA occurs in serum in various molecular forms (i.e., free PSA or complexes with inhibitors), many efforts have been made to increase the diagnostic accuracy of PSA test through the quantification of these different forms [69][70][71]. Although the consideration of different PSA molecular forms (ratio free to total PSA, isoforms p2PSA) increases cancer detection rates [72,73], they are not useful for the diagnosis of prostate cancer by itself [74]. Therefore, in case of high PSA serum level, the patient usually undergoes digital rectal examination (DRE). As most cancers are located in the peripheral posteriori area, DRE may detect bumps, soft or hard spots, or abnormal masses [75]. However, DRE is not capable of detecting early PCa, because a minimum tumour volume of 0.2 mL is needed for a possible diagnosis: DRE will detect a cancer only in the 0.1-4% of asymptomatic men considered [75]. Patients having a normal DRE outcome, though PSA serum level is markedly altered, will watchfully wait, repeat PSA and DRE periodically, and undergo further invasive tests (i.e., the biopsy).
Transrectal ultrasound-guided prostate biopsies (PBs) provide the confirmation of PCa presence. This exam is invasive, expensive, and risky. It may cause subsequent infections, erectile dysfunction, and urinary incontinence [76]. Many studies have demonstrated that there is a significant increase of the hospitalisation rate due to infections following biopsy [8,77]. Moreover, the overall cancer detection rate is 30% at first biopsy [19,78,79]. Therefore, in general, repeated biopsies are performed to improve cancer detection rate. Although the use of MRI has led to the implementation of new diagnostic strategies, including the development of equipment that facilitates MRI/TRUS fusion-guided biopsies, and consequently, the increase of the positive biopsy detection rate, there remains a need for standardisation, technical improvements, and appropriate training of radiologists to guarantee sufficient quality and reproducibility [80]. The current PCa diagnostic procedure is summarised in Figure 1. Considering the limits of the actual diagnostic procedure for PCa detection, the need of a more accurate prognostic method able of reducing overtreatment of low risk patients, unnecessary biopsies and radical prostatectomies is evident.

Innovative Techniques
Given the abovementioned issues related to the current diagnostic procedures for PCa, in recent years, many researches [19,69,81] have focused on the development of more accurate diagnostic tests compared to traditional ones. The main goal of this research field is the identification of cancer during its early stage, in order to decrease mortality and treatment costs related to PCa. For this purpose, new tests should be capable of providing accurate information about the prognosis and the most adequate treatment plan to be adopted.
This review aims to provide an overview of the state-of-the-art of innovative diagnostic techniques for early PCa detection based on urine analysis. The in-depth analysis of literature highlighted that this research field is still in progress, and different approaches have been proposed as alternatives of the current diagnostic procedure. Considering the differences between those approaches, for clarity of exposure, we decided to group the considered literary works by the type of analysis proposed, which are • sensorial analysis, which relies on the mammalian sense of smell; • senso-instrumental analysis, which tries to gather information about the olfactory properties of the analysed sample (urine) by means of specific instruments (i.e., electronic noses); • chemical analysis, which relies on analytical techniques for the identification and quantification of chemical compounds (e.g., GC-MS).
Sensorial and senso-instrumental analyses have, in common, the characterisation of the odour emanated from biological samples aimed to the identification of a cancer-characteristic "odour fingerprint", thus considering the olfactory properties of urine as a whole [82][83][84][85]. As already mentioned, sensorial methods are based on the direct characterisation of odours relying on the general human or animal sense of smell. In the particular field of diagnostics, some researchers [15,86] proved the capability of highly trained dogs to detect alterations of body odours associated with specific illnesses.
Senso-instrumental methods are based on instruments that mimic the mammalian olfactory system. Those instruments, commonly named electronic noses (EN), provide a global characterisation of the odorous mixture. EN typically comprise an array of electronic chemical sensors with partial specificity, and an appropriate pattern recognition system capable of recognising simple or complex odours [87].
Conversely, chemical analyses are based on the chemical characterisation of liquid urine or its headspace aiming to the detection of PCa biomarkers and the quantification of their amounts. They provide a detailed chemical mapping of urine through the adoption of different analytic techniques, such as gas or liquid chromatography coupled with mass spectrometry, solid phase microextraction or ion exchange liquid chromatography [27,30,36].

Olfactory Fingerprint Investigation
It has long been known that VOCs emanating from biological fluids contain information about the internal biochemistry of the human body. In ancient medicine, the diagnosis of different diseases relied on the sensorial analysis of biological fluids. Hippocrates [88] attributed all diseases to disorders of fluids, and proposed a diagnostic protocol that included observing skin and urine colour or urine tasting.
In recent years, the diagnostic usefulness of biological fluid has been revaluated. Indeed, several research studies [86,89,90] have been published regarding the investigation of biological fluid odours with the aim of identifying the presence of VOCs that are specifically related to different diseases. In the field of PCa diagnosis, some research groups proved the capability of highly trained dogs to detect alterations of body odours associated with cancer, while others tried to develop instrumental methods, proposing ENs as diagnostic tools.

Trained Dogs
Dogs are widely employed by police for detecting explosives and drugs and locating missing persons [91,92]. Dogs' olfactory system is capable of detecting odours as low as part per trillion [93], thanks to characteristic anatomic factors, such as the increased dimension of their olfactory epithelium, the huge number of olfactory receptors and the dense innervations of their olfactory mucosa [94].
The first trial investigating the feasibility of dogs' adoption for early PCa detection through the analysis of urine samples was published by Gordon et al. [16] in 2008. They trained four dogs of different breeds to discriminate men affected by PCa from healthy volunteers by the clicker training method. The training was performed by dogs' owners, who followed the same general outline for training protocol. During the training phase, cancerous urines were progressively presented to dogs against empty test tubes, water, diluted control urines, and finally, full-strength control urines in order to progressively complicate the system. In the latter training phase, dogs were presented with one positive urine against six control samples. Then, dogs' ability was tested in 33 blind runs. Each run contained six control samples and one cancerous urine placed in screw-top vial in randomised order. Dogs' performance in blind runs was worse than expected. Sensitivity achieved was around 20% for all dogs involved, while specificity was higher than 60% only for two dogs. This outcome may be explained, as suggested by the authors themselves, considering protocol deficiencies of training, i.e., absence of professional team of trainers, of a central training site and of a standardised procedure.
Cornu et al. [17] tested the ability of a Belgian Malinois shepherd to discriminate PCa and control urines. The dog was trained by the clicker training method for 16 months. Then, it was involved in a double-blind procedure that consisted of consecutive runs. For each run, the dog was presented five controls and one cancer anonymised sample. For the analyses, urine samples, stored at −4 • C, were slowly heated to 37 • C. The dog had to scent, successively, the six samples, and after a mean time of 30 s, it had to sit in front of the cancer sample. In case of success, the sample was classified as true positive; otherwise, it was considered as a false positive. Their study involved 59 men affected by PCa and 49 control subjects, and no exclusion criteria regarding medical history, diet, drugs, or tobacco consumption was considered in the selection of participants. All patients were classified as controls or PCa after undergoing PBs. The descriptive analysis for dogs' performance evaluation was performed with XLStat for Windows (Addinsoft, Paris, France) and the sensitivity and specificity achieved were of 91%.
Elliker et al. [18] attempted to train ten dogs of seven different breeds for the PCa detection adopting a two-stage training procedure. During the first stage of training, dogs learnt to find and indicate PCa samples, while during the second phase, they became able to discriminate PCa samples from controls. Urine samples were frozen at −20 • C within 10 min after collection, and defrosted in a 37 • C water bath before dogs' examination. Following training, three double-blind tests were performed for the two dogs that gave the best performances. During the blind test, the dogs were presented with 15 arrays, containing one PCa and three controls. The specificity achieved was 71% and 75% for the two dogs, respectively, while the sensitivity was 13% and 25%. Those data were lower than expected by chance. Probably the drop in sensitivity and specificity registered was due to the procedure adopted for training and blind tests. In particular, during the training phase, samples from the same donors were presented to dogs several times, and this may have led dogs to memorise sample-specific odour fingerprint that they did not rediscover during the double-blind tests, since only samples from new donors were considered. Indeed, comparison of the urine sample choices made by the dogs in different tests suggested that each dog was using different odour signature for the sample selection.
This outcome emphasises the importance of using different urine samples for the training and the double-blind tests. However, this outcome does not exclude the possibility that dogs could learn to generalise based on a common PCa with an optimised training procedure.
In order to reach this goal, Taverna et al. [19] defined a rigorous procedure for dogs' training aimed at the identification of a pool of VOCs specific for PCa emanating from urine samples. Two German Shepherd explosion detection dogs were trained using the clicker training. The dogs were taught to sit in front of the cancerous sample after sniffing a set of six urine samples, including one PCa sample and five controls. Urine samples were stored at −20 • C. For the analysis, 2 mL of each sample were defrosted and housed in circular perforated metal containers. Metal containers were placed in thermally sealed plastic packets to avoid any contamination.
Taverna's research involved a huge and multifaceted population (i.e., 902 participants), including also men and women suffering from different tumours. Diagnostic test performance was evaluated, considering the whole population, after excluding females, and considering only control men older than 45 years. In all cases, sensitivity was higher than 98% and specificity was over 96%.
In the attempt to resume and order the main information present in the scientific literature regarding the use of trained dogs for PCa detection in a table, according to a common logic for the readers' use, the following categories were defined and used: involved population and trained dogs, sample preparation, training methods, and diagnostic accuracy achieved in terms of sensitivity and specificity [95][96][97][98] (Table 1).

Electronic Nose
Given the promising results reported in studies regarding the dogs' capability of detecting PCa by sniffing urine samples, some research groups [20][21][22][23][24] started investigating the possibility to transfer those results to an instrumental method based on the analysis of urine samples through electronic noses (EN).
Electronic noses are already adopted in the food and pharmaceutical industries [87,99,100], in the environmental field [101] and for indoor air quality monitoring [102]. Their application to diagnostics has been studied with promising results for discrimination of bacteria cultures [103,104], or the detection of urinary tract infections [105][106][107], diabetes [108], kidney diseases [109,110], bowel diseases [111,112] by means of urine analysis, and lung and colon cancers by means of exhaled breath analysis [113].
In the field of PCa diagnosis, Bernabei et al. [20] investigated the ability of an EN based on 8 quartz crystal microbalances coated with different metalloporphyrins, i.e., the ENQBE, to characterise urine headspaces for the detection of PCa and bladder cancers (BC). The research involved 113 patients, including other urological diseases in the control group and PCa post-surgical patients. For the creation of urine headspace, the sample was kept at 25 • C for the time necessary to obtain a steady gaseous mixture. Then, 10 mL of the enriched headspace were extracted and injected in a 2 L bag pre-filled with N 2 .
Data were processed by means of principal component analysis (PCA) and discriminant analysis solved by partial least squares (PLS-DA). The leave one out cross validation (LOOCV) was adopted to evaluate the classification performance. The PCA score plot shows a good discrimination between PCa, BC samples and controls. A very interesting result was the migration of post-surgical patients from the PCa cluster to the healthy cluster, suggesting the ability of the EN not only to detect urological diseases, but also to monitor the response to treatments. D'Amico et al. [21] conducted a pilot study for PCa diagnosis using an EN equipped with 8 non-selective gas sensors, coated with metalloporphyrins. Urine samples were provided before prostate biopsy by 21 volunteers, including men suffering from PCa and healthy subjects. Each control participant provided two urine samples, thus executing two different trials. For the EN analysis, urine was put in a sterile urine box with a dedicated top to extract the headspace to be analysed. Data were processed by PLS-DA, and only a qualitative plot is reported for evaluating the discrimination achieved. This study needs to be enlarged in terms of population involved. In particular, the number of sick participants should be increased in order to create two olfactory classes, approximately including the same number of subjects, otherwise, the classification may be biased towards the class with most representatives. Moreover, the experimental procedure should be standardised in order to confirm results.
Asimakopoulos et al. [22] evaluated the efficacy of PCa detection by an EN equipped with 8 non-selective quartz crystal microbalance gas sensors coated with different metalloporphyrins, based on the analysis of urine samples collected before prostate biopsy. Each participant was asked to collect the initial part of the urination and the midstream in two different sterile vials. The measurements were performed without knowing biopsy outcomes. A first aspect to be highlighted is the different EN outcome related to the analyses of different parts of the urination. In particular, the analysis of midstream urine did not provide useful information for PCa detection. Conversely, the analysis of the first part of urine correlated with prostate biopsy outcomes. The author attributed this result to an increased content of elements of prostatic secretions in the first part of the urines. In this study, the EN performance achieved a sensitivity of 71.4% (CI 42-92%) and a specificity of 92.6% (CI 76-99%).
Santonico et al. [23] resumed the study of D'Amico et al. [21] and analysed urine headspaces of men suffering from PCa searching for specific odour fingerprints with the EN developed at the University of Rome "Tor Vergata". Measurements were performed at room temperature and data were elaborated by means of PLS-DA using LOOCV. The test achieved a specificity of 93%. However, the PLS-DA score plot showed only a partial discrimination of positive samples from controls.
Roine et al. [24] evaluated the EN ability of discriminating PCa from BPH by means of the analysis of urine headspaces. They involved 50 patients with confirmed PCa, and 24 control subjects suffering from BPH, among them, 15 patients provided urine preoperatively and 9 patients 3 months postoperatively. Urines were collected in the morning, and samples were stored at −70 • C.
The EN used was a commercial model (i.e., ChemProR 100, Environics Inc., Mikkeli, Finland), equipped with ion mobility cell consisting of 8 electrode strips and a metal oxide-based semiconductor cell. For the analysis, urine was defrosted and pipetted to a polystyrene culture plate, which was heated and maintained at 37 • C. Each measure lasted 25 min: 15 min for urine analysis and 10 min for recovery. Sensitivity and specificity were, respectively, 78% and 67%, when using LOOCV, and 82% and 88% when using LDA. Urine samples were collected only from men who required surgical operation. The authors in the conclusions suggested extending the study, considering patients with mild symptoms and the effects of other factors responsible for urine odour alterations, such as diet, medications, or hydration.
Also for this technology, we tried to resume and schematise the main information gathered from the scientific literature in a table. The logic adopted in this case involved the definition of the following categories-besides authors and year of publication-population involved, sample preparation methods, statistical methods adopted, and results achieved in terms of classification performance ( Table 2).

Chemical Analysis
Recent years have seen remarkable progresses in the characterisation and the quantification of biological molecules and the explanation of their roles in cells, providing new pathways for diagnostic purposes [114]. Metabolic alterations may be indicative of disease incidence, and may allow for identification of cancer biomarkers [7], defined by National Cancer Institute as biological molecules found in body fluids or tissues that are signs of normal or abnormal biological process or of a condition or disease [115].
The discovery of new and more efficacious PCa biomarkers may improve current diagnostic procedure, allowing determination of which patients will develop an aggressive tumour, prediction of recurrence, and monitoring of response to treatments.
In this section, we tried to provide a general overview of literature works focused on the chemical characterisation of liquid or gaseous urine aimed at early PCa detection.
Jentzmik et al. [26] involved 45 healthy subjects and 107 PCa patients. They determined sarcosine levels in urine by GC-MS, using a commercial amino acid assay, and discovered that the median sarcosine/creatinine was 13% lower in PCa patients than in controls. However, ROC analyses proved the inefficacy of sarcosine as PCa biomarker in comparison with total PSA, since the discrimination between PCa patients and controls was significantly worse. Authors recognised a limit of their study in the higher proportion of PCa patients than healthy subjects.
Jiang et al. [27] developed a novel method for the quantification of six urinary metabolites suitable as PCa biomarkers, i.e., sarcosine, proline, kynurenine, uracil, glycerol-3-phosphate, and creatinine, and reported that their average concentrations were higher in PCa urine than controls.
Wu et al. [28] adopted microwave-assisted derivatisation (MAD) combined with GC-MS to analyse urine samples from 20 PCa patients, 8 patients with BHP, and 20 healthy men, and compared metabolic information. PCa patients' average sarcosine levels were 13% higher than healthy controls and 19% higher than BPH samples. Propenoic acid, dihyroxybutanoic acid, creatinine, xylonic acid, and dihyroxybutanoic acid were proposed as PCa biomarkers, since their concentrations were higher in PCa patients with respect to controls.
Stabler et al. [29] compared markers in serum and urine of patients with rapidly recurrent prostate cancer to recurrence-free patients after radical prostatectomy. They tested methionine metabolites in urine and serum as pre-surgical markers for aggressive disease. They reported that urinary dimethylglycine and homocysteine of the groups did not differ significantly, while urinary sarcosine and cysteine were significantly higher in recurrent patients.
Bianchi et al. [30] developed and validated a SPME-GC/MS method for the analysis of urine and urinary sediments. Their results showed that sarcosine could be adopted as PCa biomarker. Correspondence of a cut-off of 179 µg sarcosine /g creatinine sensitivity of 79% and specificity of 87% were achieved. Shamsipur et al. [31] combined dispersive derivatisation liquid-liquid microextraction (DDLLME) with GC-MS and LC-MS to define a method for the determination of PCa metabolite biomarkers, including sarcosine, alanine, leucine, and proline. They proved that sarcosine mean concentration was higher in PCa patients, while leucine mean concentration was lower.
Heger et al. [33] monitored the level of potential non-invasive PCa biomarkers responsible for the genesis and the progression of the tumour. PSA and free PSA were determined by immunoenzymometric assay too. They detected statistically significant differences in concentrations of amino acids (i.e., aspartic acid, threonine, methionine, isoleucine, leucine, tyrosine, arginine, sarcosine, proline) and biochemical parameters (i.e., concentrations of K + , uric acid, urea, and creatinine). In particular, amino acids, urea, and creatinine were more abundant in PCa patients, while K+ and uric acid concentrations were higher in controls.
Khalid et al. [34] investigated the VOCs emanating from urine samples. They proposed 2,6-dimethyl-7-octen-2-ol, pentanal, 3-octanone, 2-octanone as suitable biomarkers for PCa detection. Except for pentanal, all of these compounds were downregulated and/or less frequently present in the urine samples from PCa patients compared to healthy subjects. The accuracy of the model based on four biomarkers discovered was 63-65%, while it was 74% (RF) and 65% (LDA), if combined with PSA level.
Tsoi et al. [35] evaluated the potential role of urinary polyamines, i.e., putrescine (Put), spermidine (Spd), and spermine (Spm), in PCa development. Their levels were determined by UPLC-MS/MS. Spm demonstrated a good discrimination performance between PCa patients and BPH patients: normalised Spm showed a significant decrease in PCa patients compared to non-cancerous cases, including BPH patients. Correlations between urinary Spm had also been performed with patients' pathologic characteristics, like age, serum PSA, creatinine content and prostate volume. However, all of them showed weak correlation with correlation coefficients <0.1.
Sroka et al. [36] proposed the LC-ESI-QqQ-MS for the quantification of amino acids and amine concentrations in urine samples from PCa patients and men with diagnosed BPH. They aimed to determine whether amino acids and amine could be used to discriminate between PCa and BPH. Arginine, homoserine, and proline were more abundant in samples from PCa patients compared to patients with benign growth. Their study underlined also that patients classified with Gleason score (GS) 7 had significantly higher concentrations of proline, homoserine, and tyramine compared with those classified as GS 6 or GS 5. They also showed the inefficacy of sarcosine as PCa indicator, by determining its levels before and after the prostate massage.
Gkotsos et al. [38] measured sarcosine, uracil, and kynurenic acid concentrations by UPLC-MS/MS and LC-ESI-MS/MS in urine from PCa patients, men with elevated PSA serum levels, and healthy subjects. Decreased median sarcosine and kynurenic acid and increased uracil concentrations were observed in PCa samples compared to controls. The ROC curve analysis showed that sarcosine and uracil did not correlate with the clinical status of subjects considered. Contrarily, kynurenic acid seemed a suitable PCa biomarker, especially in cases where the urine was collected after prostatic massage. However, this metabolite cis not useful in order to monitor disease progression.
Derezinski et al. [39] presented a comprehensive analysis of amino acids in urine. Their study provided strong evidence those branched-chain amino acids metabolic pathways can be a valuable source of markers for prostate cancer. The univariate statistical analyses performed showed that, in PCa samples, taurine was present at significant higher level, while γ-amino-n-butyric acid, phosphoethanolamine, ethanolamine, homocitrulline, arginine, δ-hydroxylysine, and asparagine occurred at significantly lower levels with respect to healthy samples. Moreover, γ-amino-n-butyric acid, phosphoethanolamine, ethanolamine, homocitrulline, arginine, δ-hydroxylysine, asparagine, cystathionine, and methionine had AUC higher than 75%. The PLS-DA model built on urine amino acid levels achieved sensitivity and specificity of 89.47% and 73.33%, respectively, whereas the total accuracy of classification was 82.35%.
In 2015, Aggio et al. [40] came up with a particular approach based on a "hybrid" system. In particular, they proposed a GC-MOS system, comprising a GC oven fitted with a commercial capillary column interfaced with a MOS sensor working at 450 • C, for classifying urine samples from patients with urological symptoms. Their study included 58 men with PCa, 24 with BC, and 73 with haematuria and/or poor stream, without cancer. The headspace was injected into the inlet of the GC-sensor system, and the gas sensor signals were processed by PCA to visualise the discrimination achieved. LDA and support vector machine (SVM) were used as statistical models to diagnose unknown samples. The performance of the classifiers was validated by LOOCV, repeated 10FoldCV, repeated DoubleCV, and Monte Carlo permutations. The first two principal components of the PCA performed on dataset relevant to PCa and control samples show a good discrimination between the two classes. In particular, the sensitivity reached was higher than 93%, while the specificity was above 95%.
The same work of schematisation of the significant data and information contained in the-in this case quite rich numbers of-scientific literature, was done for the papers regarding the application of chemical analyses for PCa detection. Table 3, besides authors and years, reports the population involved, the sample preparation methods, the analytical methods, the statistical methods, and the biomarkers proposed. Urine collection: no info; Storage and pre-treatments: Samples were frozen at −80 • C; Sample preparation: Samples were thawed at room T and diluted 3 times using water; 10 µL of diluted urine were mixed with 10 µL of the internal standard solution and 1480 µL of 0.1% formic acid in water; those samples were diluted 450 times and injected for HPLC/MS/MS analysis HPLC: An LC system working at 25 • C under a flow rate of 250 µL/min using a gradient system with the mobile phase consisting of (A) 0.1% formic acid in water and (B) 0.1% formic acid in acetonitrile (100%) was used for metabolite separation. The gradient program was initial 98% A and 2% B, linear gradient to 60% A and 40% B in 5 min, and return to initial conditions in 0.1 min at a flow rate of 250 µL/min, followed by equilibration for 10 min. MS/MS: An API 4000Q trap MS/MS system operated in multiple-reaction monitoring mode with ESI-positive ionisation was used. Turbo Spray was used as the ion source. The capillary voltage was set at 5.5 kV. Nitrogen gas was used as the curtain gas and cone gas. The cone gas flow was 50 L/h, and the desolvaation gas flow was 800 L/h. Optimal detection conditions were determined by direct infusion of each standard solution (20 ppb) in solvent A using a syringe pump. Parent-ion and daughter-ion scans were performed using nitrogen as the collision gas at a pressure of 3.8 × 10 3 millibar and a flow of 0.2 mL/min.    Urine collection: no info; Storage and pre-treatments: Samples were stored at −20 • C; Sample preparation: Each sample was defrosted by immersing the vial in a water bath at 60 • C for 30 s. One single aliquot of urine sample per patient was used for VOC analysis. Thereafter, each sample was treated with an equal volume (0.75 mL) of sodium hydroxide 1 M. The mixture was equilibrated at 60 • C in a water bath for 30 min prior to SPME. SPME: The SPME fibre was 85 µm thick and consisted of carboxen/polydimethylsiloxane. It was exposed to the headspace above the urine mixture for 20 min. GC-MS: VOCs were thermally desorbed from the fibre at 220 • C in the injection port of the GC/MS for 5 min. Injection was made in splitless mode and a split of 50 mL/min was turned on two minutes into the run. It was used helium as carrier gas (99.996% purity). Capillary column consisted of 94% dimethyl polysiloxane and 6% cyanopropyl-phenyl. The GC/MS T program of the run was as follows: initial oven T was held at 40 • C for 2 min then T was ramped up at a rate of 5 • C/min to 220 • C, with a 4 min hold at this T to give a total run time of 42 min. The mass spectrometer was run in electron impact (EI) ionisation mode, scanning the mass ion range 10-300 at 0.05 scan/s. A 4 min solvent delay was used at the start of the run. Urine collection: after lunch prior PBs; Storage and pre-treatments: −20 • C; Sample preparation: Firstly, urine samples were thawed and centrifuged for 5 min at 13,000 rpm at room T. Urine sample supernatant (120 µL) and 60 µL of internal standard working solution were mixed with 420 µL of water. Of this well-mixed solution, 550 µL was passed through SPE, which had been conditioned and equilibrated with 1 mL of methanol and water respectively. Water (450 µL) was passed through the cartridge afterwards to elute out all polyamines. Of these SPE-treated samples, 400 µL were then mixed with 100 µL of 10% HFBA, and the final mixture was ready for instrumental analysis UPLC-MS/MS: The column used was an Agilent EclipsePlus C18 RRHD (2.1 × 50 mm, 1.8 µm) protected with an Agilent SB-C18 guard column (2.1 × 5 mm, 1.8 µm). The LC elution profiles were optimised as follows: Eluent A was water with 0.1% HFBA while eluent B was acetonitrile with 0.1% HFBA. Eluent A was decreased from 95% to 60% in 10 min, and from 60% to 10% in 1 min. Afterwards the gradient was held constant for 5 min. The gradient was then increased from 10% to 95% in 1 min, and held constant for 8 additional minutes. The autosampler and column temperatures were set at 4 and 35 • C respectively. Injection was achieved by 5-s needle wash in Flush Port mode for 3 times with eluent B. Ten microlitres was injected each time. For the source parameter, drying gas (N 2 ) temperature was set as 300 • C with 5 L/min flow rate. Nebuliser pressure was 45 psi. Sheath gas temperature was set as 250 • C with 11 L/min flow rate. Capillary voltage was set as 3500 V.
Student's t-test; ROC analysis putrescine (Put), spermidine (Spd) and spermine (Spm) Normalised Spd was significantly lower in PCa than in BHP patients and controls The AUC for normalised Put, Spd and Spm were found to be 0.63 ± 0.05, 0.65 ± 0.05 and 0.83 ± 0.03 respectively LC-ESI-QqQ-MS: Mobile phase consisted of (A) 0.1% formic acid in water (v/v) and (B) 0.1% formic acid in ACN (v/v). Flow rate was set to 300 µL min −1 .
Separation was performed at 30 • C with monitored pressure below 400 bar. Analysis time was 19 min. The gradient was run from 0-2 min using 1% solvent B, then linearly raised over 7 min from 1% to 15% solvent B. then raised to 30% solvent B over 5 min and dropped to 1% for re-equilibration which lasted 5 min.
Concentrations were quantified using Agilent 1200  LC-QTOF: A Mediterranea Sea C18 analytical column thermostated at 25 • C was used. The initial mobile phase was a mixture of 98% phase A (0.1% formic acid in water) and 2% phase B (0.1% formic acid in ACN). After injection, the initial mobile phase was kept under isocratic conditions for 1 min; then, a linear gradient of phase B from 2% to 100% was applied within 16 min. The flow rate was 0.6 mL/min. The total analysis time was 17 min, and 5 min were required to re-establish the initial conditions. The injected volume was 5 µL. The autosampler was kept at 4 • C to increase sample stability. A, then rising to 15% B linearly over the next 2 min, finally reaching 40% B over 2 min and returning to initial conditions over 5 min. The column was equilibrated for 6 min in the initial conditions. Flow rate was 0.5 mL/min

Discussion and Conclusions
In the previous paragraphs, we tried to give an overview of the innovative techniques that are being studied in the field of PCa diagnosis. This section has the aim of critically discussing the different methods proposed, and to compare them in terms of advantages, drawbacks, and future perspectives.
In general, innovative tools proposed in recent years can be grouped into two macro-categories, according to the method adopted. Many authors explored the possibility of analysing urine odour fingerprint, while others preferred the chemical characterisation of liquid urine or of its gaseous headspace, trying to identify potential PCa biomarkers. All approaches propose the comparative analysis of urine samples from healthy men and PCa patients with the aim of discriminating the two classes.
Studies reporting the adoption of trained dogs for PCa detection proved its feasibility and the diagnostic accuracy achieved in terms of sensibility and specificity was promising. However, they boosted further investigation, since some critical issues about the definition of the experimental protocol emerged. In particular, the type of training and blind test, the training site, the team of trainers involved, the frequency and duration of training, the method adopted for the sample somministration and the number of blind-test runs influence dogs' discriminative ability. In addition, it is worth considering the high costs related to dogs' training and the effect of the training procedure on the classification performance.
Considering the limits related to the adoption of trained dogs for the development of a large-scale diagnostic tool, researchers started investigating the possibility to transfer those experimental observations to an instrumental method based on the analysis of urine samples through electronic noses.
All literary works, summarised in Table 2, reporting the adoption of the EN for PCa diagnosis, focused on the characterisation of urine headspace. In general, published results confirm the capability of the EN to distinguish between urine samples collected from men suffering from PCa and healthy subjects, with very promising diagnostic accuracy in terms of specificity and sensitivity, although there is no uniformity concerning sample preparation, analysis, and data processing techniques among the different studies proposed.
It is worth considering that none of the abovementioned studies address the problem of sensor response drift over time, and repeatability among different instruments. Sensor drift (i.e., non-deterministic temporal variations of the sensor response when exposed to the same analytes under identical conditions) is recognised as one of the main problems associated with gas sensor [116]. This aspect limits the EN ability to operate over long time periods in all fields of application of ENs, thus, it is one of the key criticalities to be solved for the development and spread of an innovative PCa diagnostic tool based on EN analysis.
Recent advances in the understanding of cancer genesis and progression and in the characterisation and the quantification of biological molecules boosted the research in the field of urine chemical characterisation for PCa diagnostic purposes. Indeed, many research groups started working on the identification of novel biomarkers able to improve the diagnostic and prognostic accuracy of the traditional tests.
Most of investigated literary works, summarised in Table 3, proposed the comparative analysis of samples from PCa patients and controls, since authors agreed that PCa development caused alterations of different metabolic pathways, such as amino acid, fatty acid, and carbohydrate metabolism. They developed diverse methods, combining different analytical techniques, for the detection and quantification of changes in metabolites levels in PCa samples compared to healthy ones. Sometimes, different stages of the tumour were considered to evaluate also the efficacy of staging of proposed PCa biomarkers.
Nevertheless, the in-depth analysis of these literary works highlighted that no exhaustive results have been published until now, since many different metabolites were proposed as suitable PCa biomarkers, and divergent opinions upon the same metabolites emerged in different studies.
Many literary works considered proposed a quantitative characterisation of urine samples. Table 4 reports concentrations trends of proposed biomarkers in cancerous samples compared to controls, as reported in literature. Among the proposed biomarkers, the most debated is sarcosine. Indeed, many authors [25,[28][29][30][31] reported that its level in urine from PCa patients is higher than in control samples, and that its classification performance is good, whereas other researchers [36,38] showed that changes in its concentrations between PCa and healthy men were not statistically significant.
Proline, citrulline, and homocitrulline seemed to be the most suitable biomarkers for PCa detection, since many authors [27,31,33,[36][37][38] agreed that their levels were higher in PCa samples compared to control ones, and classification performance of models built on these metabolites was encouraging.
Future works should focus on those disparities among different studies, in order to adopt chemical analyses with the objective of improving traditional diagnostic procedures.
The key aspects of different approaches discussed in this section were schematically compared in terms of pros and cons as reported in Table 5. Considering the specific and different difficulties associated with each of the discussed innovative diagnostic approaches, possibly the evolution towards "hybrid" systems combining two or more different approaches, as proposed by Aggio et al. [40], might represent an answer to those contradictory outcomes in the next future.
It is realistic to think of a possible future development of a "hybrid" system based on the combination of the odour analysis performed by the EN with the chemical characterisation of urine samples, which should differ from previous studies in this field by focusing on odour analysis, and take unique advantage of the olfactory differences highlighted by EN analysis. This might possibly focus on the identification of those compounds responsible for the alteration of urine odour, thus simplifying the task of the chemical characterisation of urine and provide an innovative pathway for the discovery of new and more efficient biomarkers specific for the PCa.
In general, one aspect that is common to all innovative diagnostic methods is the importance of the size of the population involved. In particular, the classes considered (i.e., PCa patients and healthy subjects) should include approximately the same number of samples, otherwise, the output of the innovative tests tend to be biased towards the class with most representatives [117]. Almost all literary works here presented involved small populations and, in many cases, the control and the PCa groups were not numerically comparable.
Another very important critical point is that the majority of studies do not discuss the specificity of the method towards PCa with respect to other pathologies. Indeed, it would be very interesting to investigate the specificity to other types of tumours, and especially, tumours associated with the urinary tract (i.e., bladder and kidney cancers). This aspect is particularly important for the development of a specific diagnostic tool whose answer should be ideally positive only in cases of PCa, and negative for any other disease, as obtained by Taverna et al. [19] in his experience with trained dogs.
Only Bernabei et al. [20] and Aggio et al. [40] tested the their methods towards another urological tumour, i.e., the bladder cancer. Bernabei et al. [20] reported a PLS-DA score plot that showed the complete discrimination between healthy and sick subjects and a PCA score plot, where the gradual discrimination between different cancers is visible. On the contrary, Aggio et al. [40] reported only the performance of the EN in distinguishing between controls and BC or PCa separately.
Roine et al. [24], Wu et al. [28], Bianchi et al. [30], Tsoi et al. [35] and Sroka et al. [36] included in their chemical comparative studies men suffering from BPH, which typically causes a high number of false positives in traditional procedures. However, their results should be confirmed by considering larger populations.
One aspect common to the different methods proposed, which is worth highlighting here as part of the critical discussion, is the importance of the training phase, which should be intended as the effective training of dogs for developing the discriminative ability between different urine samples, and for the development of a suitable model for data processing and classification for senso-instrumental and chemical approaches.
According to the scientific literature here examined, it is not possible to identify a data processing or classification algorithm whose classification performance clearly prevails compared to others. Indeed, the proposed algorithms are sometimes very different from each other (e.g., PCA, PLS-DA, ROC analysis, Mann-Whitney analysis, Pearson correlation) and this result in the variability of classification performances reported. However, it is worth highlighting that mathematics, even when involving extremely elaborate and complex algorithms, can never adjust bad data [118]. Therefore, it is very important to optimise input data, especially when samples are characterised by high variability, as with urine. This is a crucial aspect of this research field aimed at the development of a large-scale PCa diagnostic tool.
Last but not least, a fundamental aspect for the development of a method that might become of widespread use in clinical diagnosis is the switch from a complex laboratory apparatus to an easy-to-use instrument. The training of dogs, the use of electronic noses or of chemical analysers, require highly specialised personnel, whereas the field of clinical diagnosis is rapidly evolving towards point-of-care tests, which ideally should be used by non-specific staff. This is a great challenge for current research in the field of medical diagnostics.
Despite the difficulties associated with the development of innovative and reliable diagnostic techniques, a significant increase of the research in this field-and hopefully the successful introduction of some of these techniques in clinical diagnosis-in the near future is to be expected, due to the high social and economic impact that new technologies for early diagnosis of cancer might have in today's culture.
As a last consideration, it might be worth highlighting that, given the ambitious purpose, only a multidisciplinary team, that includes clinicians, engineering, biologists, physicians, and biochemical scientists, collaborate together to understand the complexity of human beings. This way of thinking may further help to clarify concepts and indicate alternative experiments in order to develop appropriate diagnostic methods [42].
Author Contributions: Carmen Bax was responsible of bibliographic research and conceived the structure and the outline of the paper. Lidia Eusebio and Selena Sironi helped with the interpretation of the literature data and the critical comparison of literary works. Fabio Grizzi and Giorgio Guazzoni contributed with the understanding of medical aspects. Gianluigi Taverna conceived the experiments with trained dogs and thus helped in revising the part concerning dog analysis. Laura Capelli was the responsible of the research project, coordinated the work and revised the paper.

Conflicts of Interest:
The authors declare no conflict of interest.