Abstract Background Prostate cancer (PCa) is one of the most common malignancies affecting men, with primary treatments involving surgery, radiotherapy, and hormonal therapy. The introduction of precision medicine and next-generation sequencing (NGS) has profoundly influenced the clinical management of PCa, particularly by enabling the assessment of genetic alterations that guide treatment decisions. Liquid biopsy using diverse sample types, including plasma, urine, and semen, offers non-invasive alternatives to tissue biopsies. This study sought to compare the performance of NGS-based mutation detection across various sample types in PCa patients. Methods Thirty-seven PCa patients, diagnosed with intermediate to advanced stages (II-IV), were enrolled. All collected samples, including tissues (n = 34), plasma (n = 37), urine (n = 32), and seminal fluids (n = 9), underwent targeted NGS of 437 cancer-related genes. The detection sensitivity, mutational landscape, and maximum variant allele frequencies (MVAFs) were compared across different sample types. Results Tissue samples, serving as the gold standard, achieved a 100% mutation detection rate. Plasma and urine samples demonstrated high detection sensitivities, reaching 67.6% and 65.6%, respectively, while semen samples showed a lower detection rate of 33.3%. Mutations in FOXA1, SPOP, and TP53 were commonly detected across most sample types with comparable prevalence. AR mutations were observed with similar frequencies in plasma and semen samples, but were absent in tissue and urine samples. The average MVAFs were at similar levels among tissue, plasma, urine, and semen, although urine sediment samples exhibited the lowest MVAFs. Advanced disease stages correlated with increased circulating tumor DNA (ctDNA) detection in both plasma and urine samples. No significant survival advantage associated with ctDNA negativity was observed, likely due to the small sample size. Conclusions This study validates the utility of urine and plasma samples as non-invasive and sensitive liquid biopsy options for PCa, showing comparable ctDNA detection rates. Seminal fluid samples also demonstrate potential, despite current sampling challenges. These findings offer insights into the advantages of different sampling methods for PCa detection and reinforce the clinical utility of liquid biopsies in PCa management. Keywords: Prostate cancer, Next-generation sequencing, Liquid biopsy, Urine, Seminal fluid Introduction Prostate cancer (PCa) is one of the most common malignancies of the male genitourinary system, with primary treatment modalities including surgery, radiotherapy, and hormonal therapy [[48]1]. With the advent of precision medicine, next-generation sequencing (NGS) technologies have become increasingly integrated into PCa clinical management. NGS has been widely applied to aid in treatment decision-making and evaluate genetic predisposition. Metastatic castration-resistant prostate cancer (mCRPC) harboring homologous recombination repair gene mutations can be treated with poly (ADP-ribose) polymerase inhibitors, as approved by the U.S. Food and Drug Administration. Additionally, PCa patients with mismatch repair deficiency and high microsatellite instability can benefit from immune checkpoint blockade therapies. NGS also plays a crucial role in predicting hereditary tumor risks, as genomic variations are significantly related to PCa development, with 15.6% of patients identified to carry pathogenic germline mutations in genes such as BRCA1, BRCA2, HOXB13, MLH1, MSH2, PMS2, MSH6, EPCAM, ATM, CHEK2, NBN, and TP53 [[49]2]. Moreover, NGS holds enormous potential in predicting immunotherapy outcomes, such as the likelihood of response or hyperprogression, through the identification of established predictive markers, and can aid in guiding clinical decisions post-drug resistance. Patient tumor tissues, including fresh and formalin-fixed paraffin-embedded (FFPE) tissue samples, along with circulating tumor DNA (ctDNA), can be used for the simultaneous detection of both germline and somatic mutations [[50]3]. Although tissue-based testing remains the gold standard for NGS-based gene analysis, the majority of newly diagnosed PCa patients present with mid to late-stage disease, resulting in a higher proportion of non-surgical cases and making tissue acquisition more challenging. In cases where tissue biopsy is not feasible, ctDNA offers a non-invasive alternative by detecting tumor-derived DNA fragments in various bodily fluids, with plasma being a commonly used source. The sensitivity and specificity of ctDNA detection in plasma depend largely on tumor burden and shedding, as indicated by the variant allele frequencies (VAFs) of mutations present in the sample. A study of 45 mCRPC patients demonstrated that mutations with < 2% VAFs could not be reliably detected, whereas those with ≥ 35% VAFs reached an over 90% concordance with tissue samples [[51]4]. Localized PCa has a smaller tumor burden, resulting in a lower abundance of gene mutations in plasma ctDNA, often below the detection threshold. Consequently, the likelihood of identifying clinically significant mutations via liquid biopsy is reduced [[52]5]. In a study including 53 newly diagnosed metastatic hormone-sensitive prostate cancer patients, the average ctDNA VAF was reported to be 11%, with 80% consistency between liquid biopsy and tissue testing under an average sequencing depth of 927 ×. Following short-term (22-day) androgen deprivation therapy, the ctDNA VAFs in these patients decreased to approximately 1% [[53]6]. For mCRPC patients, over 70% had ctDNA VAFs > 2%, achieving a 90% concordance with tissue-based testing [[54]7]. In addition to plasma, ctDNA can also be released into other bodily fluids, such as urine, cerebrospinal fluid, pleural effusion, ascites, and semen. Among these, urine and semen have garnered significant attention in PCa studies due to their direct association with the prostate gland. Elevated concentrations of circulating free DNA (cfDNA) in seminal fluid have been significantly correlated with PCa compared to non-PCa patients (median 428.45 vs. 25.4 ng/ml, P < 0.001). Significantly higher levels of cfDNA have been detected in the seminal fluid compared to plasma [[55]8]. However, the collection of semen samples poses considerable challenges, particularly for patients undergoing androgen deprivation therapy, which can result in erectile dysfunction, and for those who have undergone post-brachytherapy with reduced semen volume. In addition, patients who have undergone radical prostatectomy and seminal vesicle removal are unable to provide semen for ctDNA testing. PCa occurring in the urinary tract also releases DNA fragments directly into the urine [[56]9]. Urine sampling is more convenient and has gained significant attention in the screening of PCa-related biomarkers. In a study that included 29 PCa patients and 25 healthy individuals, real-time PCR was used to detect fragments of MYC, HER2, and AR in urinary cfDNA, yielding a sensitivity of 79% and a specificity of 84% [[57]8]. This underscores the clinical potential of using urine for NGS-based ctDNA testing in PCa. The current National Comprehensive Cancer Network Guidelines for Prostate Cancer [[58]1] endorses liquid biopsy methodologies when tissue testing fails or is unattainable. However, further research is needed to validate the concordance between these methods. This study aims to compare the consistency of NGS-based mutation detection using tissue, plasma, urine, and semen samples, evaluating their sensitivity and specificity to provide guidance for future clinical applications. Materials and methods Patient inclusion and data collection This study retrospectively included 37 PCa patients, diagnosed at intermediate to advanced (II-IV) stages, at the Second Affiliated Hospital of Dalian Medical University (Dalian, China) between May 2018 and December 2023. Those with other primary cancers in addition to PCa, with a history of alcohol abuse or substance abuse, or with a previously diagnosed mental disorder, were excluded from the study. Tumor tissue, blood, urine, and semen samples were collected for subsequent NGS analysis. Clinical information including age, medical history, date of diagnosis, clinical staging, pathological type, and treatment history, was also collected. For patients with sufficient clinical follow-up data, progression-free survival (PFS) was defined as the time from the initiation of treatment to the date of progressive disease. Informed written consent was obtained from each subject or the subject’s family member upon sample collection. Ethic approval is waived due to the retrospective nature of the study. DNA extraction and sequencing Genomic DNA from tissue samples was extracted using the DNeasy Blood and Tissue Kit (Qiagen, Germany) for fresh tissue samples, and the QIAamp DNA FFPE Tissue Kit (Qiagen) for FFPE tissues, in accordance with the manufacturer’s protocols. ctDNA was isolated from body fluid samples with the QIAamp Circulating Nucleic Acid Kit (Qiagen). Genomic DNA from white blood cells was also extracted using the DNeasy Blood and Tissue Kit (Qiagen) to serve as a control for identifying germline variations. Sequencing libraries were constructed using the KAPA Hyper DNA Library Prep Kit (Roche, Switzerland), and were subsequently enriched for 437 cancer-related genes utilizing probe-based hybridization capture panels designed by Geneseeq Technology Inc., as previously outlined [[59]10]. Following each procedural step, DNA concentration was measured using a Qubit Fluorometer (Life Technologies, USA). Sequencing was conducted on an Illumina HiSeq4000 platform (Illumina, USA) in a Clinical Laboratory Improvement Amendments-certified and College of American Pathologists-accredited laboratory at Nanjing Geneseeq Technology Inc., China. Mutation calling For quality control of the FASTQ files, bases with an'N'or low quality (quality score below 20) were removed using Trimmomatic [[60]11], and PCR duplicates were removed using Picard (available at: [61]https://broadinstitute.github.io/picard/). The filtered paired-end reads were then aligned to the human genome reference (GRCh37, hg19) using the Burrows-Wheeler Aligner [[62]12], followed by realignment around indels performed with GATK3 [[63]13]. Single nucleotide variants (SNVs) and insertions/deletions (indels) were identified using VarScan2 [[64]14] and annotated through ANNOVAR [[65]15]. The identification of SNVs/indels was based on a minimum VAF threshold of 0.01 and a p-value threshold of 0.05. Considering the variations in tumor purity between tumor tissue and bodily fluid samples, the following thresholds were applied to filter SNVs/indels from different sources: (i) Read depth ≥ 10; (ii) For tissue samples: unique variant-supporting reads ≥ 5 and variant allele frequency ≥ 1%; (iii) For plasma samples: unique variant-supporting reads ≥ 3 and variant allele frequency ≥ 0.3%; (iv) Supporting reads had to be mapped to both strands; (v) Variants with a prevalence > 1% in the 1000 Genomes Project or in the ExAC database were excluded. NGS results from matched white blood cells of each patient were used to further discard germline variants and clonal hematopoiesis-related variants. Fusion events were detected using Delly [[66]16], while copy-number variations (CNVs) were identified using CNVkit [[67]17], employing default parameters. Somatic CNVs were additionally corroborated by comparing paired normal and tumor samples for each exon, applying a cut-off value of 0.65 for copy-number loss and 1.60 for copy-number gain. Statistical analysis For the functional enrichment analysis of mutated gene sets, we utilized the KEGG REST API ([68]https://www.kegg.jp/kegg/rest/keggapi.html) to obtain the latest KEGG pathway gene annotations. These annotations served as the background, against which we mapped the genes. Enrichment analysis was then conducted using the R package ‘clusterProfiler’ to identify enriched gene sets. A p-value threshold of < 0.05, and a false discovery rate of < 0.25 were considered statistically significant. The R package'survival'was employed to perform Kaplan–Meier (KM) survival estimates, and log-rank tests were used to compare survival probabilities between groups. Unless otherwise specified, a two-tailed p-value of less than 0.05 was considered statistically significant. Results Patient overview A total of 37 PCa patients were included in the study. Tissue samples were available for 34 patients, with 30 collected at baseline prior to treatment. Plasma samples were obtained from all 37 patients, including 16 at baseline, 14 during or after treatment, and seven post-progression. Urine samples were collected from 32 patients, comprising 16 baseline and five post-progression samples. Additionally, urine sediment samples were collected from eight patients (including four at baseline), and semen samples were available from nine patients, seven of which were baseline collections. A total of 28 patients had three specimen types available — tissue, plasma, and urine — whereas 31 patients had matched tissue and plasma samples. 28 patients had sufficient clinical information, with prognostic data available for eight of them, as summarized in Table [69]1. 89.3% of the patients were aged 60 years or older and 67.9% were classified at stage IV. The levels of baseline prostate-specific antigen (PSA) varied, ranging from 4.54 to 4000.00 ng/mL, with a median of 51.45 ng/mL. Among the eight patients with prognostic data, six (75.0%) had experienced disease progression. Table 1. Patient clinical characteristics Patients (n = 28) Number Percentage (%) Age  ≥ 60 25 89.29  < 60 3 10.71 Stage  II 4 14.29  III 4 14.29  IV 19 67.86  NA 1 Baseline PSA  Median 51.45  Range 4.54–4000.00 Progression  Yes 6 75.00  No 2 25.00  NA 20 [70]Open in a new tab NA not available, PSA prostate-specific antigen Mutational Landscape across various sample types Mutations were identified in all 34 tissue samples (Fig. [71]1A), revealing a total of 229 mutations across 115 unique genes, with the number of unique mutations per patient ranging from 1 to 46. The most frequently mutated genes included FOXA1 (35%), SPOP (32%), and TP53 (29%). Pathway analysis indicated that the identified mutations were primarily enriched in pathways involved in PI3K-AKT signalling, human papillomavirus infection, thyroid hormone signalling, endocrine resistance, and resistance to EGFR tyrosine kinase inhibitors (Fig. [72]1B). Fig. 1. [73]Fig. 1 [74]Open in a new tab Mutational landscape showcasing genomic variations detected across various sample types in the study cohort (N = 37). A Tissue samples (n = 34) with a list of the top 30 most frequently mutated genes identified. B Pathway enrichment analysis from tissue samples indicating five significantly enriched pathways associated with prostate cancer. C Plasma samples in which genomic variations were detected (n = 25), visualizing the top 30 most frequently mutated genes. D Urine samples with detected mutations (n = 21), illustrating the top 30 most frequently mutated genes. E Seminal fluid samples with detected mutations (n = 3). Variants are classified by their clinical significance, with P/LP indicating pathogenic or likely pathogenic variants, and VUS indicating variants of uncertain significance In plasma samples, mutations were detected in 25 out of 37 samples, yielding a mutation detection rate of 67.6% (Fig. [75]1C). A total of 155 mutations across 98 unique genes were identified, with the number of mutations per patient ranging from 1 to 45. Consistent with the findings from tissue samples, the most mutated genes in plasma were TP53 (36%), SPOP (24%), AR (20%), and FOXA1 (16%). Of the 32 urine samples analyzed, 21 demonstrated detectable mutations, resulting in a detection rate of 65.6% (Fig. [76]1D). These samples contained a total of 121 mutations across 84 unique genes, with the number of unique mutations per sample ranging from 1 to 28. The most frequently mutated genes in urine were TP53 (24%), SPOP (19%), FGFR3 (14%), and FOXA1 (14%). Among the eight urine sediment samples, only one sample (12.5%) had detectable mutations in APC, GNAS, and SPOP. In the nine semen samples, mutations were detected in three samples, showing a detection rate of 33.3% (Fig. [77]1E). These samples collectively harbored 42 mutations across 34 unique genes, with one sample exhibiting a relatively high count of 39 mutations. Sensitivity of ctDNA detection The sensitivity for detecting genomic variations across different sample types was evaluated (Fig. [78]2A). Using tissue samples as the gold standard with a 100% mutation detection rate, liquid biopsies utilizing plasma and urine samples demonstrated ctDNA detection sensitivities of 67.6% and 65.6%, respectively. The maximum variant allele frequency (MVAF) levels for the identified mutations were compared among the sample types (Table [79]2 and Fig. [80]2B), with tissue samples exhibiting the highest average MVAF of 28.5%. The average MVAF detected from urine samples (6.72%) was slightly lower that for plasma samples (10.48%), but the median MVAF for urine samples (0.73%) were higher than that for plasma samples (0.58%), indicating similar detection potential of urine biopsies compared to plasma. Seminal fluid samples also displayed a 4.4% average MVAF, while the value was the lowest in urine precipitation samples (0.22%). However, no statistically significant differences were observed among different groups. Fig. 2. [81]Fig. 2 [82]Open in a new tab Comparative analyses of mutation detection across various sample types and clinical stages in the studied cohort. A Comparison of the mutation detection rates among different sample types, highlighting the variation in detection efficiency. B Comparison of the maximal variant allele frequencies (MVAF) across sample types, indicating no significant differences (P > 0.05). ns, non-significant. C Comparison of mutation detection rates categorized by patient disease stages (II, III, IV) across different sample types, illustrating the changes in mutation prevalence with disease progression Table 2. Distribution of maximal variant allele frequencies in each sample type Tissue (N = 34) Plasma (N = 37) Urine (N = 32) Urine Precipitation (N = 8) Seminal Fluid (N = 9) 25% Percentile 6.53 0 0 0 0 Median 22.15 0.58 0.73 0 0 75% Percentile 43.82 11.06 2.58 0 8.65 Maximum 85.55 84.36 77.99 1.75 19.08 Mean 28.50 10.48 6.72 0.22 4.04 [83]Open in a new tab Moreover, we observed that for both plasma (P = 0.158) and urine (P = 0.569) samples, ctDNA detection rates increased with advancing tumor stages of the patients, although lacking statistical significance (Fig. [84]2C). Such trend was not observed for semen or urine precipitation samples, likely due to the small sample size and the absence of certain stage groups with the exact sample types. MVAF levels were comparable across different stages for each sample type (P > 0.05); however, plasma and urine samples with higher MVAFs were mostly observed in patients with more advanced tumor stages (Fig. [85]3). Fig. 3. [86]Fig. 3 [87]Open in a new tab Comparison of maximal variant allele frequencies (MVAF) for patients of different disease stages (II, III, IV) among different sample types. ns, non-significant Survival analysis Despite the availability of survival data for only eight patients, we explored the association between ctDNA positivity and patient outcomes. Plasma samples were obtained from all eight patients, with genomic variations identified in seven (87.5%) individuals. Urine samples were available for three patients, of which two (66.7%) tested positive for ctDNA. KM analysis indicated that the presence of detectable mutations in plasma (hazard ratio = 0, 95% confidence interval [95% CI] = 0-Inf, P = 1.00; Fig. [88]4A) or urine (HR = 1.17, 95% CI = 0.11–12.98, P = 0.90; Fig. [89]4B) did not impact PFS probabilities. Fig. 4. [90]Fig. 4 [91]Open in a new tab Kaplan–Meier plots predicting the progression-free survival probabilities of patients with (ctDNA +) or without (ctDNA-) detectable mutations in their plasma (A) or urine (B) samples. No significant survival difference was noted. CI, confidence interval; HR, hazard ratio Discussion This study was purely observational and primarily aimed at investigating the potential of semen and urine as new media for NGS liquid biopsy in PCa. We investigated their ctDNA detection rates, mutational profiles, variant abundances, and the concordance of these findings with those yielded from the ‘gold-standard’ tissue samples, or from the well-established liquid biopsy medium of plasma. Firstly, in terms of ctDNA detection rates, while the tissue samples achieved a 100% detection rate, the mutation detection rates for urine (65.6%) and plasma (67.6%) were comparable. Instances where mutations were detected in the tissue sample but not the liquid biopsy sample may suggest a localized tumor or effective treatment control, leading to low tumor burden and, consequently, reduced ctDNA release. Similar findings have been reported for PCa plasma biopsies, suggesting a favorable prognosis for those with undetectable ctDNA [[92]18, [93]19]. However, due to the limited number of patients with survival data in this study, we could not demonstrate significant survival advantage associated with ctDNA absence. Notably, detection rates in both blood and urine samples increased with advancing stages of PCa, indicating the clinical significance of ctDNA detection in monitoring tumor progression. For semen samples, the relatively low detection rate (33.3%) could be attributed to sample size limitations, or it might reflect a limitation in using NGS mutation calling to detect tumor burden, as previous studies have mainly identified CTCs [[94]20, [95]21] or higher levels of cfDNA in seminal fluid of PCa patients [[96]22, [97]23]. Despite this, one semen sample exhibited a high number of unique mutations, suggesting the potential of semen to detect rare mutations and serve as an alternative liquid biopsy medium. This also led to the highest average mutation count detected in semen samples among all sample types. Additionally, we evaluated the use of urine sediment, which was proposed to contain shed PCa cells [[98]24], and found its limited capability in detecting genomic variations, with a lowest mutation detection rate of only 12.5%. Previous studies have identified nucleic acids enabling PCa detection in both urine supernatant and sediment, but have also noticed that the supernatant fraction exhibited higher sensitivity in detecting somatic mutations than the sediment [[99]24]. Regarding the detected mutations, the most frequently observed mutations in tissue samples were FOXA1 (35%), SPOP (32%), and TP53 (29%), all of which are well-documented in PCa [[100]25, [101]26]. The pathways significantly enriched with these mutations are also crucial in the oncogenesis and progression of PCa [[102]27–[103]30]. The reported frequencies of 17–29% for TP53 alterations and 12–17% for SPOP mutations from previous studies or from the TCGA dataset (Firehose legacy; [104]https://www.cancer.gov/tcga, accessed on 2 June 2025), are slightly lower compared to the results from our study, with urine detection rates (TP53: 24%, SPOP: 19%) and plasma detection rates (TP53: 36%, SPOP: 24%), but show a similar mutually exclusive mutation patterns of these two tumor suppressor genes [[105]25]. In semen, SPOP mutation was detected in one sample (33.3%). The higher mutation frequencies observed in our cohort may be attributed to different sample sizes and distribution of patient stages, as we sampled fewer patients and included those at more advanced stages, whereas the TCGA cohort predominantly focused on primary tumors. FOXA1 mutations, implicated in AR signalling and PCa progression [[106]31], have been reported with prevalence ranging from 8 to 40%, for different histological subtypes and clinical stages of PCa [[107]26, [108]32]; in our study, FOXA1 alteration was also a high-frequency mutation in blood (16%) and urine (14%) samples, although slightly lower than the 35% prevalence in tissue. Notably, a high frequency of AR mutations (20%) detected in plasma was not observed in tissue, presumably because AR mutations commonly present as secondary resistance mutations following androgen receptor signalling inhibitor treatment [[109]33]. While tissue samples of the prostate tumor predominantly indicate the mutational profile of the primary tumor clone, ctDNA in the plasma may reflect that of resistant or metastatic clones [[110]34, [111]35]. Prior studies have also reported lower AR mutation prevalence in localized or primary PCa compared to metastatic PCa [[112]36]. Consequently, the absence of AR mutations in urine might indicate that it mainly reflects the primary tumor's mutational profile, while the detection of AR mutation in one semen sample (33.3%) suggests its potential to reveal resistance mutations, though this requires further validation with larger sample sizes. In terms of mutation abundance, the MVAF did not show significant differences across sample types, with average MVAFs in urine and semen samples comparable to that in the plasma, highlighting their clinical utility. Consistent with the lowest mutation detection rate, urine sediment samples also exhibited the lowest MVAFs, suggesting limited ctDNA levels. The MVAF across different stages also showed similar patterns within each sample type, without significant inter-stage differences. This could be due to the small sample sizes of stage II and III patients, and may also be attributed to influence of intertumoral heterogeneity among patients [[113]37]. It may be more interesting to evaluate the dynamic changes in MVAF over time within the same patient as a comparison metric. Despite these findings, the study has several limitations associated with the novel sampling methods employed. Firstly, the sample sizes—particularly for semen samples—were limited, in part due to the practical challenges of collecting such specimens in patients with reproductive system cancers like PCa. In addition, our assay did not incorporate unique molecular identifiers (UMIs), which may affect the overall sensitivity and specificity. Although ctDNA detection in liquid biopsy samples is well-validated for SNV and indel, it often struggles to effectively identify CNVs and structural variants due to challenges such as DNA fragmentation and low ctDNA abundance. Moreover, the samples were not uniformly collected at the same clinical time point, which may have contributed in part to differences in the observed mutation profiles across specimen types. These factors may have collectively impacted our ability to accurately compare mutation frequencies accurately across different sample types. Future larger-scale studies with standardized sampling protocols, UMI-based error correction, and refined bioinformatic pipelines may enhance mutation detection performance, particularly in the context of unconventional sample types. Finally, the retrospective nature of this study resulted in incomplete clinical and survival data, precluding paired analyses and yielding non-significant survival results. Future studies employing these media are expected to corroborate our findings. Conclusions In summary, this study compared the performance of urine and semen in NGS liquid biopsy for PCa against the results from the ‘gold-standard’ tissue sample and from the well-established plasma sample, demonstrating their potential application as alternative or supplementary sampling sources. It is hoped that these findings could provide more insights into the diagnosis and treatment of prostate cancer. Acknowledgements