Conceived and designed the experiments: CW NAB PEC SS. Performed the experiments: SS MM-F RHA LVL ZAL. Analyzed the data: SS TK ZZ CW NAB. Contributed reagents/materials/analysis tools: NAB TK CW SS. Wrote the paper: NAB CW TK PEC.
The authors have declared that no competing interests exist.
Clinical decision for primary treatment for prostate cancer is dictated by variables with insufficient specificity. Early detection of prostate cancer likely to develop rapid recurrence could support neo-adjuvant therapeutics and adjuvant options prior to frank biochemical recurrence. This study compared markers in serum and urine of patients with rapidly recurrent prostate cancer to recurrence-free patients after radical prostatectomy. Based on previous identification of urinary sarcosine as a metastatic marker, we tested whether methionine metabolites in urine and serum could serve as pre-surgical markers for aggressive disease.
Urine and serum samples (n = 54 and 58, respectively), collected at the time of prostatectomy were divided into subjects who developed biochemical recurrence within 2 years and those who remained recurrence-free after 5 years. Multiple methionine metabolites were measured in urine and serum by GC-MS. The role of serum metabolites and clinical variables (biopsy Gleason grade, clinical stage, serum prostate specific antigen [PSA]) on biochemical recurrence prediction were evaluated. Urinary sarcosine and cysteine levels were significantly higher (p = 0.03 and p = 0.007 respectively) in the recurrent group. However, in serum, concentrations of homocysteine (p = 0.003), cystathionine (p = 0.007) and cysteine (p<0.001) were more abundant in the recurrent population. The inclusion of serum cysteine to a model with PSA and biopsy Gleason grade improved prediction over the clinical variables alone (p<0.001).
Higher serum homocysteine, cystathionine, and cysteine concentrations independently predicted risk of early biochemical recurrence and aggressiveness of disease in a nested case control study. The methionine metabolites further supplemented known clinical variables to provide superior sensitivity and specificity in multivariable prediction models for rapid biochemical recurrence following prostatectomy.
Prostate cancer remains the most common non-cutaneous solid malignancy in the United States, and the second leading cause of cancer specific death in men. Nevertheless, it has become increasingly clear that not all men who are diagnosed with prostate cancer require intervention
A previous study identified sarcosine (N-methylglycine) as a product of methionine catabolism that is elevated in the urine of patients with metastatic prostate disease
In this study we evaluated the serum and urine of radical prostatectomy patients for metabolites to differentiate those who developed early biochemical recurrence (rise in serum PSA≥0.2 ng/ml) within two years of surgery and those who remained recurrence-free after more than five years. We found that the urine of patients in the rapidly recurrent group had significantly higher concentrations of sarcosine and cysteine than those in the recurrence-free group. In addition, significantly greater concentrations of serum cystathionine, homocysteine and cysteine were found in the rapidly recurrence group compared to the recurrence-free group. These products of elevated methionine catabolism in patients with rapidly recurrent prostate cancer represent pre-surgical indicators that augmented serum PSA for the prediction of clinically significant prostate cancer.
This nested case-control study was conducted in accordance with the Institutional Review Board of Vanderbilt University. Written consent was given by the patients for their information to be stored in the hospital database. The board specifically approved the research use of the di-identified information and “on the shelf” specimens to be used for research under a waiver of consent.
The digital medical records of 400 subjects were retrospectively examined using the Vanderbilt University Department of Urologic Surgery registry of radical prostatectomies performed between 2003 and 2007. Several patients were excluded for reasons of compromised renal, heart, or liver function as was determined by electronic records of elevated urinary creatinine, hypertension, cardiac infarction history, and blood markers for hepatic function. Additionally, availability of follow-up data and records of pre-surgical hormone-ablation therapy were reasons for exclusion. Rapidly recurrent patients were identified as those who developed biochemical recurrence following prostatectomy within 2 years (American Joint Committee on Cancer defined as having PSA≥0.2 ng/ml, confirmed at least once two weeks later). The recurrence-free population was defined as having maintained a serum PSA<0.01 ng/ml for five or more years following surgery. Ultimately, for this nested case control study we focused on 54 subjects for analysis of urine and 58 subjects for analysis of serum who developed rapid biochemical recurrence and an age-matched recurrence-free control group who were free of recurrence. The mean age for the subjects was 60 years. All subjects were annotated based on age, pre-surgical serum PSA, biopsy Gleason score, clinical stage, and detection of biochemical recurrence.
Serum and urine obtained at the time of radical prostatectomy were rapidly processed and stored at −80°C. We evaluated serum and urine for the metabolites, sarcosine, dimethylglycine, methionine, homocysteine, cystathionine, cysteine, methylmalonic acid and methylcitrate by gas-liquid chromatography/mass spectrometry
Patients' baseline demographic and clinical variables were assessed using Wilcoxon rank sum tests for continuous variables and Fisher exact tests for categorical (including binary) variables. All marker values, as well as PSA levels, were logarithmically transformed to achieve normality. Correlations among the markers were assessed using Spearman's rank correlation. Logistic regression models were used to analyze incidence of recurrence. The base model includes serum PSA, biopsy Gleason score, and clinical stage, clinical variables that are available prior to surgery. The post-surgical variables (e.g., lymph nodes, surgical margins, pathologic Gleason scores) were not considered. For multiplicity control, p≤0.007 (p-value less than 5%/7 = 0.7%) was considered statistically significant. To avoid further overfitting of the data, no variable selection was performed in the subsequent analyses based on logistic regression models. We used a likelihood ratio test to compare the simpler model (without the metabolites) and the full model (with the individual metabolites). Receiver operating characteristics (ROC) curves were generated for each logistic regression model, where the area under ROC curve (AUC) was determined. Integrated discrimination improvement (IDI) and Net reclassification index (NRI)
Urine metabolites were initially measured in fifty-four patients who developed biochemical recurrence (N = 25) and those that remained recurrence-free (N = 29). These patients were matched for age and pre-surgical serum PSA.
Recurrent-free (29) | Recurrent (25) | P value | |
|
59 (53, 64) | 62 (58, 67) | 0.10 |
|
5.2 (4.3, 6.5) | 6.0 (5.0, 8.2) | 0.08 |
0.09 | |||
|
15 (94%) | 12 (67%) | |
|
1 (6%) | 6 (33%) | |
|
0 | 0 | |
0.050 | |||
|
1 (6%) | 0 | |
|
2 (12%) | 0 | |
|
9 (56%) | 4 (22%) | |
|
3 (19%) | 8 (44%) | |
|
1 (6%) | 3 (17%) | |
|
0 | 2 (11%) | |
|
0 | 1 (6%) | |
190 (168, 212) | 221 (189, 252) | 0.007 | |
|
2.7 (2.2, 3. 2) | 2.8 (2.4, 4.0) | 0.40 |
|
27.3 (22.1, 38.5) | 25.4 (17.6, 33.7) | 0.34 |
|
3.7 (3.1, 5.7) | 5.4 (4.1, 6.7) | 0.03 |
Values for sarcosine, homocysteine, dimethylglycine and cysteine are expressed as µmoles/mg creatinine. Wilcoxon rank sum tests for continuous variables and Fisher exact tests for categorical (including binary) variables are indicated. Normal values for metabolites (nmole/mg creatinine) are: cysteine, 140–579; homocysteine, 0.974–7.17; dimethylglycine, 10.1–108.2 and sarcosine, 2.65–8.67. Median values with quartiles were used to summarize the distributions of the continuous variables.
We then performed a nested case control study with pre-surgical serum. Fifty-eight age-matched prostatectomy patients were stratified by pre-surgical PSA, clinical stage, and biopsy Gleason grade as well as pathologic variables (
Recurrent-free (30) | Recurrent>(28) | P value | |
|
59 (54, 64) | 61 (59, 64) | 0.07 |
|
5.4 (4.0, 8.1) | 6.8 (5.2, 8.9) | 0.02 |
|
0.30 | ||
|
24 (80%) | 18 (64%) | |
|
6 (20%) | 9 (32%) | |
|
0 | 1 (4%) | |
|
0.006 | ||
|
1 (3%) | 0 | |
|
2 (7%) | 0 | |
|
18 (60%) | 6 (20%) | |
|
6 (20%) | 13 (46%) | |
|
2 (7%) | 4 (15%) | |
|
1 (3%) | 4 (15%) | |
|
0 | 1 (4%) | |
|
346 (321, 377) | 419 (367, 452) | <0.001 |
|
9.0 (8.0, 10.2) | 11.7 (9.4, 13.4) | 0.003 |
4.6 (3.8, 4.7) | 4.9 (4.2, 5.4) | 0.21 | |
1.3 (1.1, 1.4) | 1.3 (1.1, 1.7) | 0.67 | |
24.8 (21.7, 30.6) | 27.6 (23.9, 33.7) | 0.08 | |
44.8 (25.2, 52.8) | 42.3 (31.3, 51.5) | 0.72 | |
|
126 (102, 144) | 135 (117, 167) | 0.13 |
|
167 (145, 220) | 164 (146, 211) | 0.91 |
149 (130, 176) | 186 (148, 239) | 0.007 | |
|
0 (0%) | 6 (21%) | 0.01 |
|
0 (0%) | 8 (29%) | 0.002 |
|
1 (3%) | 8 (29%) | 0.01 |
|
3 (10%) | 21 (75%) | <0.001 |
|
0.002 | ||
|
2 (7%) | 0 (0%) | |
|
15 (50%) | 4 (14%) | |
|
10 (33%) | 14 (50%) | |
|
3 (10%) | 4 (14%) | |
|
0 (0%) | 6 (21%) |
Wilcoxon rank sum tests for continuous variables and Fisher exact tests for categorical (including binary) variables are indicated. Normal values for metabolites are: cysteine, 203–369 µM homocysteine, 5.4–13.9 µM; dimethylglycine, 1.4–5.3 µM; sarcosine, 0.6–2.7 µM; methionine, 11.3–42.7 µM; folate, >3.0 ng/ml; methylcitrate, 60–228 nM; methylmalonate, 73–271 nM; cystathionine, 44–342 nM. Median values with quartiles were used to summarize the distributions of the continuous variables.
Correlation coefficient | P value | n | |
|
0.19 | 0.34 | 28 |
|
0.12 | 0.53 | 28 |
|
0.33 | 0.06 | 33 |
|
0.13 | 0.48 | 34 |
All correlations are rank based “Spearman's rho”.
The relevance of these newly identified markers to patient recurrence status were illustrated in Kaplan-Meier plots for homocysteine, cystathionine, and cysteine as compared to pre-operative serum PSA levels, and time-to-recurrence (
The patients were separated into two groups, divided at median tissue level for (A) PSA, (B) homocysteine, (C) cystathionine, and (D) cysteine as significantly associated with time to recurrence (
The clinical value of these methionine metabolites as biomarkers would be to significantly increase the ability to predict aggressive prostate cancer features and early biochemical recurrence over and above existent clinical variables including serum PSA, biopsy Gleason score, and clinical stage. We developed a multiple logistic regression model for the prediction of biochemical recurrence based on serum methionine metabolites and the pre-surgical predictor variables, serum PSA and biopsy Gleason grade. Since majority of patients in both cohorts had clinical stage T1c disease, this variable had little discriminative power and was dropped from the model. Serum cysteine, cystathionine, and homocysteine were the top three predictors for recurrence in 70% of the patients, so further analysis of methionine metabolites focused on these three metabolites. Correlations between cysteine and homocysteine were the highest among all pair-wise correlations (R2 = 0.65, p<0.01), and cysteine was also highly correlated with cystathionine (R2 = 0.39, p<0.01,
Serum PSA is compared to the added value of serum (A) homocysteine, (B) cystathionine, and (C) cysteine. In the ROC curve the probability with greater Area Under the Curve (AUC) support increased specificity and sensitivity over random guess, represented by the dotted diagonal line.
Dimethylglycine | Sarcosine | Cysteine | Cystathionine | |
|
0.28, 0.05 n = 50 | 0.28, 0.04 n = 50 | 0.65,<0.01 n = 57 | 0.22, 0.10 n = 55 |
|
0.35, 0.01 n = 50 | 0.40, <0.01 n = 50 | 0.16, 0.26 n = 48 | |
|
0.35, <0.01 n = 50 | 0.08, 0.60 n = 48 | ||
|
0.39, <0.01 n = 54 |
All correlations are rank based “Spearman's rho”, presented as correlation, p-value, and n.
SERUM HOMOCYSTEINE MODEL | ||||
Variable | Comparison Q3∶Q1 | Odds | 95% Confidence Int. | P value |
Pre-surgery PSA |
|
|
|
|
Biopsy GS |
|
|
|
|
Serum homocysteine |
|
|
|
|
SERUM CYSTATHIONINE MODEL | ||||
Variable | Comparison Q3∶Q1 | Odds | 95% Confidence Int. | P value |
Pre-surgery PSA |
|
|
|
|
Biopsy GS |
|
|
|
|
Serum cystathionine |
|
|
|
|
SERUM CYSTEINE MODEL | ||||
Variable | Comparison Q3∶Q1 | Odds | 95% Confidence Int. | P value |
Pre-surgery PSA |
|
|
|
|
Biopsy GS |
|
|
|
|
Serum cysteine |
|
|
|
|
IDI | 95% CI | P-value | NRI | 95% CI | P-value | |
|
0.14 | 0.05–0.24 | 0.003 | 1.03 | 0.52–1.55 | <0.001 |
|
0.12 | 0.004–0.20 | 0.003 | 0.81 | 0.28–1.34 | 0.003 |
|
0.14 | 0.04–0.23 | 0.005 | 0.64 | 0.13–1.16 | 0.015 |
To define the efficacy of the markers in predicting recurrence-free survival, Cox proportional hazard regression models were fit showing that cysteine, cystathionine, and homocysteine were each independent predictors of recurrence-free survival when adjusting for pre-operative serum PSA and biopsy Gleason score (
SERUM HOMOCYSTEINE MODEL | ||||
Variable | Comparison Q3∶Q1 | Hazard | 95% Confidence Int. | P value |
Pre-surgery PSA |
|
|
|
|
Biopsy GS |
|
|
|
|
Serum homocysteine |
|
|
|
|
SERUM CYSTATHIONINE MODEL | ||||
Variable | Comparison Q3∶Q1 | Hazard | 95% Confidence Int. | P value |
Pre-surgery PSA |
|
|
|
|
Biopsy GS |
|
|
|
|
Serum cystathionine |
|
|
|
|
SERUM CYSTEINE MODEL | ||||
Variable | Comparison Q3∶Q1 | Hazard | 95% Confidence Int. | P value |
Pre-surgery PSA |
|
|
|
|
Biopsy GS |
|
|
|
|
Serum cysteine |
|
|
|
|
Current risk stratification of patients prior to surgery involves variables including serum PSA, clinical stage, and biopsy grade. Independent serum markers in conjunction with PSA could help distinguish patients with aggressive prostate cancer. In the current era of PSA testing, clinical staging has reduced relevance when tumor volumes are relatively small. In our study, the highest biopsy Gleason score in ≥8-core biopsies provided a significant independent predictor comparable to serum cysteine and homocysteine. However, routine ultrasound directed first biopsies are reported to miss nearly a quarter of the prostate cancers
Pathways of methionine metabolism involve two mechanisms for sarcosine formation (
Methionine is first converted to SAM, the donor of methyl groups in all but one methyltransferase reaction. SAM may transfer the methyl group to a variety of compounds, X, by a group of specific enzymes to yield the methylated compounds, CH3-X (eg. methylated lipids, DNA, or proteins). Alternatively, SAM may transfer the methyl group to glycine to form sarcosine via the enzyme glycine N-methyltransferase (GNMT. After transfer of the methyl group SAM is converted to S-adenosylhomocysteine (SAH), which is broken down further to homocysteine, cystathionine and cysteine. Sarcosine may also be formed by breakdown of choline to betaine, which, after loss of a methyl group, is converted to dimethylglycine. A dehydrogenase converts dimethylglycine to sarcosine.
The majority of the sarcosine produced in the body is made in the liver as a downstream product of SAM and homocysteine. Studies using homozygous mice with GNMT knocked out have plasma SAM levels 50% greater than that of wild type. The SAM levels in the livers of the
To our knowledge, no previous study has correlated an entire arm of a metabolic pathway in the aggressiveness of cancer. In our study the comparison was made between patients with proven cancer, not between subjects with proven cancer and benign prostatic disease. These results were obtained with only a relatively small group of patients but the results are significant and suggest that further studies are needed. The underlying biology supports the robustness of these markers. Higher serum homocysteine, cystathionine, and cysteine improved the utility of currently used clinical variables in predicting early recurrence and suggest a greater flux of methyl groups through the enzyme GNMT.