Urinary metabolites in association with kidney cancer risk

Abstract

Kidney cancer incidence has increased worldwide in recent decades. While metabolomic studies have shown promise in unveiling mechanisms underlying disease development, few studies have investigated prediagnostic urinary metabolites and kidney cancer risk. We conducted a case-control study nested within the Shanghai Women’s and Men’s Health Studies to prospectively investigate the association between urinary metabolites and kidney cancer risk to understand its etiology and the underlying biological mechanisms. Two hundred primary kidney cancer cases and their individually matched controls were included. A total of 1301 metabolites were evaluated, and 67 metabolites were found nominally associated with kidney cancer using conditional logistic regression. After backward selection, 11 urine metabolites remained significantly associated with kidney cancer: lipids (e.g. picolinoylglycine, odds ratio [OR]; 95% confidence interval [CI]: 2.01 [1.44, 2.79], and pregnanediol-3-glucuronide, OR; 95% CI: 0.56 [0.39, 0.82]), xenobiotics (e.g. beta-guanidinopropanoate, OR; 95% CI: 1.75 [1.32, 2.32] and 4-vinylphenol sulfate, OR; 95% CI: 0.66 [0.49, 0.90]), and nucleotides (e.g. allantoic acid, OR; 95% CI: 0.71 [0.54, 0.92]). Time lag analysis showed that metabolite-kidney cancer associations were stronger for beta-guanidinopropanoate (OR; 95% CI: 8.22 [1.68, 40.18]) and picolinoylglycine (OR; 95% CI: 6.45 [1.28, 32.43]), but weaker for allantoic acid (OR; 95% CI: 0.87 [0.37, 2.06]) and 3-methylglutarate/2-methylglutarate (OR; 95% CI: 0.62 [0.19, 2.00]) when urinary samples were collected within 3 years between urine sample collection and cancer diagnosis (P_interaction < .05 for all). Future metabolomics studies with large sample sizes, particularly from multiple ancestry populations, are needed to validate our findings.

Graphical Abstract

Graphical Abstract.

Introduction

Kidney cancer incidence has increased in many countries worldwide, including China, over the past decade (1). Kidney cancer incidence is higher in men than in women (2). In 2020, a total of 73 587 incident kidney cancer cases occurred in China, contributing to over 17% of the global burden. In terms of mortality, 43 196 out of 212 536 global deaths due to kidney cancer occurred in China, contributing 20% of kidney cancer deaths globally (3). The incidence and mortality rates of renal cell carcinoma (RCC)—a major form of kidney cancer—are higher in non-White individuals than in the White population (4). An analysis of the Surveillance, Epidemiology, and End Results Program between 2007 and 2014 revealed higher incidence of kidney cancer among American Indian/Alaskan Native, Black, and Hispanic/Latino individuals compared with their White counterparts (1,4). Although kidney cancer incidence was found to be lower among Asian American/Pacific Islanders compared with White individuals, the mortality due to kidney cancer was higher among Asian American/Pacific Islanders compared with Whites (6.5 versus 3.6 per 100 000 population) (1). Asian Pacific Islanders also had the highest proportion of patients with advanced-stage RCC (5). In addition, Asian and Pacific islanders were found to have a higher risk of kidney damage (6). Several clinical conditions related to kidney cancer, such as hypertension, chronic kidney disease, chronic renal failure, and end-stage renal disease, were suggested contributors to kidney cancer risk more among the non-White population than their White counterparts (7). While other established risk factors of kidney cancer, such as smoking and Body Mass Index (BMI), showed a consistent association across different racial groups in the United States (USA) (7). Metabolomics is the global analysis of small molecule metabolites which may lead to the identification of drivers of tumorigenesis and understanding disease etiology (8). Metabolomic profiles provide a comprehensive readout of metabolic processes, including alterations at the deoxyribonucleic acid, ribonucleic acid, and protein level. Since minor changes in protein structure and expression can lead to significant changes in the metabolite levels, the information provided by metabolomics can be invaluable (8). A review by Schmidt et al. (8) provides background on metabolomics, particularly showcasing their application in clinical and translational research settings. For example, hypoxia inducible factors (HIFs) are transcription factors consisting of an alpha (α) subunit (HIF1α and HIF2α) which are broken down in the presence of oxygen and stabilized in its absence, allowing cells to adapt to the stress of hypoxic conditions (8). Additionally, HIF2α is involved in anabolic metabolism regulation and represents a therapeutic target in clear cell Renal Cell Carcinoma (ccRCC) (8). Another review by Weiss et al. (9) explored the critical role metabolomics techniques that have been used to further our understanding of the biology of kidney cancer in particular, in addition to its contributions in the identification of novel treatments. Metabolomics has recently gained attention in cancer research for its promise in providing etiological evidence and advancing knowledge of biological mechanisms (10). Studies have shown that dysregulated metabolism can be indicative of kidney cancer incidence and progression (11,12). While metabolism may contribute to the difference in susceptibility to cancer across racial ethnic groups (13), epidemiological evidence is lacking. Overall, few prospective studies have investigated the associations between prediagnostic metabolites and kidney cancer incidence, especially among non-White populations (14,15). To address these knowledge gaps and to further our understanding of kidney cancer etiology and underlying biological mechanisms, we conducted an exploratory pilot study of urinary metabolomics nested within the Shanghai Women’s Health Study (SWHS) and Shanghai Men’s Health Study (SMHS).

Materials and methods

Study population and design

The SWHS and SMHS are two prospective cohorts of 74 940 female and 61 469 male urban Chinese residents, respectively, aged 40–74 years old, recruited during 1996–2000 and 2002–2006, with a participation rate of 93% and 74%, respectively. The cohorts were established to investigate the etiology of cancers and other non-communicable diseases, as described in detail elsewhere (16,17). Briefly, at baseline, all study participants completed an in-person interview using structured study questionnaires that cover demographics, lifestyle (e.g. smoking, drinking, exercise), occupational history, and personal/family disease history, as well as usual dietary intake. Anthropometrics were taken by trained interviewers (16,17). A spot urine sample was collected from 88% of the SWHS and 89% of the SMHS participants using a sterilized cup with 125 mg of ascorbic acid, transported under a 0–4°C environment, and processed for long-term storage at −75°C within 6 hours of sample collection (16,17).

Cohort participants have been followed-up with through a combination of annual record linkages to the Shanghai Cancer Registry and Shanghai Vital Statistics Registry and in-person interviews every 2–4 years (16,17). The SWHS has completed 5 in-person follow-up surveys with a response rate of 99.8%, 98.7%, 96.7%, 92.2%, and 91.2%, respectively, while the SMHS had 3 in-person follow-up surveys with response rates of 97.6%, 91.9%, and 93.6%. Because of the extremely low out-migration rate and high coverage of the Shanghai Cancer Registry and Vital Statistics Registry, the follow-up rate for the SWHS and SMHS is near 100%.

A total of 439 primary incident kidney cancer cases (ICD-O codes of C64 or C65) were identified from the cohort members after study enrollment (248 cases in the SMHS and 191 cases in the SWHS). Among them, 378 (86%) cases donated urine samples. We randomly selected 100 cases from each of the SWHS and SMHS for our study. Comparison of age, BMI, selected lifestyles, and fasting status between kidney cancer cases included and those not included in the current study showed no significant difference between these two case groups (Supplementary Table S1). We randomly selected one control for each case, using the density-sampling without replacement strategy, from the risk pool containing the remaining cohort members that were alive and were free of cancers at the time of diagnosis for the index kidney cancer case. Each control was matched to the index case on age at enrollment/urine donation (±2 years), urine sample collection date (±less than 30 days), time of urine sample collection (morning/afternoon), time to the last meal (±less than 2 hours), recent antibiotic use (Yes/No), donation of other biospecimens (i.e. blood or buccal cell sample), and menopausal status (SWHS only).

Metabolomics assessment

Urine metabolites were measured using the Metabolon global untargeted platform via ultrahigh performance liquid chromatography/tandem mass spectrometry (UPLC-MS/MS) (Metabolon Inc., Durham, NC, USA) (18). Urine samples from three case-control pairs (SWHS: 2 pairs, SMHS: 1 pair) were excluded from metabolomics analysis due to insufficient sample volume. Urine samples for cases and matched controls were assayed within the same batch and the disease status of the samples was blinded. Urine samples with a metabolite concentration below the minimum detectable threshold were assigned the minimum value for that metabolite. Metabolite data were normalized by urine osmolality concentration and further central standardized to have a median equal to one. After excluding 109 metabolites with constant concentration values for all study participants, a total of 1301 metabolites, 846 named metabolites and 455 metabolites of unknown structural identity, were included in the analysis. Given the low missing rate (1.5%) for any metabolite per participant in our study, all subjects were included in the analysis and missing data were imputed.

Statistical analysis

First, we identified and excluded metabolites with strong concentration outliers on the fourth-root scale beyond the range of [quartile (Q)1 − 1.5 * interquartile range (IQR), Q3 + 1.5 * IQR].

For each metabolite, we then transformed its concentration to the natural log scale and estimated the OR per standard-deviation increment with corresponding 95% confidence intervals (CI) for kidney cancer using conditional logistic regression to accommodate the matched design.

We first screened the metabolites based on their nominal significance in individual associations (P < .05) and the Benjamini–Hochberg false discovery rates (BH-FDR) of 0.20. Because many metabolites are correlated, we then applied stepwise backward selection to identify urine metabolites that were independently or jointly associated with kidney cancer risk. We used entry criteria P-value = .15 and exit criteria P-value = .05 to maximize the chance of inclusion of all potential kidney cancer associated metabolites in the evaluation and keep those retaining their significant associations by the end. Multivariable conditional logistic regression was used to analyze the final selected metabolites and the ORs with 95% CIs were reported. Covariates adjusted for include current smoking, amount of alcohol intake, baseline BMI, hypertension, baseline diabetes, and family history of cancer. We also adjusted for the natural logarithmic daily polyunsaturated fatty acids intake (g/day) because it has been previously linked to kidney function and is associated with increased kidney cancer in our study.

We also conducted analyses stratified by sex and time from urine sample collection to kidney cancer diagnosis (<3 versus ≥3 years between urine sample collection and kidney cancer diagnosis) to investigate potential effect modifications.

Pathway analysis was done using Metaboanalyst 6.0 by including all metabolites that were significantly associated with kidney cancer (P < .05).

Results

A total of 197 case-control pairs were included in the study, 98 pairs of women and 99 pairs of men, and the mean age was 58.1 (SD = 9.3). Matching factors were well-balanced between cases and controls (Supplementary Table S2). The prevalence of smoking (prior or current) and family history of cancer were higher among men than women (Table 1). The median time between urine sample collection and kidney cancer diagnosis was 4.75 years with a range of 0.14–13.45 years.

Table 1.

Distributions of baseline characteristics by sex and case-control status in the SWHS and SMHS.

	SWHS (N = 98 pairs)a			SMHS (N = 99 pairs)a			Overall P-valueb
	Control (n = 98)	Case (n = 98)	P-valueb	Control (n = 99)	Case (n = 99)	P-valueb
BMI at baseline (kg/m 2 )
Mean (SD)	25.2 (3.72)	24.6 (3.36)	.281	24.1 (2.87)	24.6 (2.73)	.180	.934
Occupation, n (%)
Professional	25 (25.5%)	34 (34.7%)	.165	37 (37.4%)	29 (29.3%)	.389
Clerical	25 (25.5%)	16 (16.3%)		20 (20.2%)	20 (20.2%)		.521
Manual work	48 (49.0%)	48 (49.0%)		42 (42.4%)	50 (50.5%)
Leisure time physical activities (MET-h/day)
Median [Min, Max]	0 [0, 20.9]	0.20 [0, 7.97]	.493	0 [0, 15.0]	0 [0, 11.5]	.332	.233
Current Smoking, n (%)
Yes	1 (1.0%)	2 (2.0%)	1.000	48 (48.5%)	43 (43.4%)	.508	.606
Ever Smoking, n (%)
Yes	1 (1.0%)	3 (3.1%)	.625	62 (62.6%)	59 (59.6%)	.686	.896
Current Drinking Alcohol, n (%)
Yes	3 (3.1%)	2 (2.0%)	1.000	27 (27.3%)	19 (19.2%)	.186	.163
Family history of cancer, n (%)
Yes	27 (27.6%)	34 (34.7%)	.265	31 (31.3%)	36 (36.4%)	.457	.192
Hypertension, n (%)
Yes	33 (33.7%)	42 (42.9%)	.183	40 (40.4%)	42 (42.4%)	.752	.234
Diabetes, n (%)
Yes	7 (7.1%)	6 (6.1%)	1.000	9 (9.1%)	8 (8.1%)	1.00	.856
Daily dietary consumption (Kcal)
Mean (SD)	1672 (430)	1685 (430)	.828	1950 (447)	1964 (602)	.833	.762
Total fat intake (g/day)
Mean (SD)	27.2 (12.8)	31.0 (13.1)	.028	34.2 (15.2)	35.6 (20.2)	.599	.093
Saturated fatty acids (g/day)
Mean (SD)	7.68 (3.99)	8.65 (3.69)	.076	10.32 (4.71)	10.54 (6.08)	.776	.202
Monounsaturated fatty acids (g/day)
Mean (SD)	11.67 (6.03)	13.44 (6.58)	.034	15.03 (7.75)	15.64 (10.01)	.627	.114
Polyunsaturated fatty acids (g/day)
Mean (SD)	7.45 (3.31)	8.40 (3.76)	.033	8/36 (3.47)	8.80 (4,48)	.438	.054

^a N = 2 and N = 1 pairs of matched cases and controls were excluded due to insufficient baseline urine sample volumes from the SWHS and SMHS cohorts, respectively.

^b P-values for continuous variables are calculated from paired t-test if distribution is near normal or from Friedman’s Two-way Nonparametric ANOVA if distribution is skewed, and P-values for categorized variables were calculated from conditional logistic regression.

In analyses of all samples combined, 67 urine metabolites were significantly associated with kidney cancer risk, with unadjusted P-values < .05 (Table 2). Most of the urine metabolites that were negatively associated with kidney cancer in the combined cohort were lipids, nucleotides, and amino acids; most urine metabolites positively associated with kidney cancer belonged to xenobiotics and amino acids (Fig. 1A). In the female cohort (SWHS), some carbohydrates were found to be negatively associated with kidney cancer besides lipid, nucleotide, and amino acids. The most positively associated urine metabolites were xenobiotics and amino acids, similar to findings in the combined cohort (Fig. 1B). In the male cohort (SMHS), we found that the majority of negatively associated urine metabolites were xenobiotics, amino acids, and nucleotides; xenobiotics, amino acids, carbohydrates, and cofactors/vitamins comprised the majority of urine metabolites that were positively associated with kidney cancer (Fig. 1C).

Table 2.

ORs of kidney cancer associated with urine metabolites from univariable conditional logistic regression model.^a

Metabolite name	Crude OR	95% CI		P-valuea	P-FDR
5,6-dihydrouracil	0.68	0.540	−0.848	<.001	<.001
Inosine	0.72	0.574	−0.900	.004	<.001
3-methyl-2-oxobutyrate	0.73	0.577	−0.920	.008	.002
4-vinylphenol sulfate	0.73	0.577	−0.927	.010	.002
glycochenodeoxycholate glucuronide (1)	0.75	0.593	−0.936	.011	.002
N-acetylaspartate (NAA)	0.74	0.578	−0.936	.012	.003
diacetylspermidine*	0.75	0.595	−0.939	.013	.003
S-methylcysteine	0.76	0.607	−0.945	.014	.003
pregnen-diol disulfate*	0.73	0.567	−0.939	.014	.003
5,6-dihydrothymine	0.75	0.600	−0.947	.015	.003
2,3-dihydroxypyridine	1.30	1.046	−1.609	.018	.004
4-methoxyphenol sulfate	0.76	0.599	−0.953	.018	.004
allantoic acid	0.78	0.631	−0.961	.020	.004
N(1) + N(8))-acetylspermidine	0.77	0.611	−0.964	.023	.005
3-methylglutarate/2-methylglutarate	0.77	0.607	−0.964	.023	.005
3alpha,21-dihydroxy-5beta-pregnane-11,20-dione 21-glucuronide	0.78	0.629	−0.972	.027	.006
beta-guanidinopropanoate	1.26	1.026	−1.551	.027	.006
homocarnosine	1.29	1.028	−1.606	.027	.007
glycerophosphorylcholine (GPC)	0.78	0.628	−0.977	.030	.007
hexanoylglycine (C6)	0.78	0.627	−0.976	.030	.007
picolinoylglycine	1.26	1.023	−1.550	.030	.007
isocaproylglycine	0.78	0.627	−0.977	.030	.008
Sucrose	0.78	0.628	−0.980	.033	.008
Sedoheptulose	0.78	0.626	−0.981	.033	.008
4-methylhexanoylglycine	0.79	0.640	−0.982	.034	.009
N2,N2-dimethylguanine	0.80	0.643	−0.984	.035	.009
Xanthine	0.76	0.585	−0.981	.035	.009
Benzoate	0.78	0.623	−0.985	.036	.009
Hypoxanthine	0.79	0.639	−0.986	.037	.010
dihydrobiopterin	0.80	0.644	−0.987	.038	.010
2’-O-methyluridine	0.79	0.636	−0.989	.039	.010
Cortolone	0.80	0.647	−0.992	.042	.010
N-methylhydantoin	0.81	0.661	−0.993	.043	.011
17alpha-hydroxypregnanolone glucuronide	0.75	0.571	−0.997	.048	.012
dehydroepiandrosterone glucuronide	0.81	0.651	−0.998	.048	.012
3-hydroxybutyroylglycine	0.80	0.646	−0.999	.049	.012
pregnanediol-3-glucuronide	0.81	0.653	−0.999	.049	.013
X—21 825	0.73	0.596	−0.892	.002	<.001
X—25 457	0.70	0.556	−0.889	.003	<.001
X—18 838	0.71	0.556	−0.895	.004	<.001
X—23 161	0.74	0.602	−0.917	.006	.001
X—24 362	0.74	0.600	−0.920	.006	.001
X—24 361	0.73	0.576	−0.916	.007	.002
X—11 470	0.75	0.594	−0.937	.012	.002
X—23 655	1.35	1.068	−1.716	.012	.003
X—24 495	0.77	0.620	−0.958	.019	.004
X—13 553	0.76	0.600	−0.959	.021	.004
X—12 729	0.76	0.608	−0.961	.022	.005
X—15 806	0.78	0.625	−0.966	.023	.005
X—21 258	0.78	0.624	−0.966	.023	.005
X—24 356	0.78	0.626	−0.967	.024	.006
X—25 271	0.74	0.568	−0.962	.025	.006
X—15 807	0.79	0.643	−0.972	.026	.006
X—24 947	0.76	0.601	−0.971	.028	.007
X—23 668	0.79	0.634	−0.978	.030	.007
X—23 787	0.79	0.635	−0.980	.033	.008
X—17 351	0.77	0.606	−0.979	.033	.008
X—24 411	1.25	1.014	−1.531	.037	.010
X—23 314	0.78	0.615	−0.985	.037	.010
X—12 127	0.79	0.632	−0.993	.044	.011
X—21 851	0.80	0.651	−0.994	.044	.011
X—24 334	0.80	0.650	−0.995	.045	.011
X—25 435	0.79	0.621	−0.995	.045	.011
X—24 494	0.79	0.632	−0.995	.045	.011
X—24 546	0.79	0.623	−0.996	.046	.012
X—25 109	0.79	0.625	−0.996	.046	.012
X—24493	1.23	1.000	−1.503	.0495	.013

^a P-values were calculated based on the Wald-statistics from conditional logistic regression model.

P-FDR: False-discovery rate adjusted P-value.

Figure 1.

Volcano plots illustrating the P-values and unadjusted ORs associated with one standard deviation (SD) increase in the natural logarithms of each metabolite’s (N = 1301) concentration in urine and renal cell carcinoma, among all (A), female (B), and male (C). Each urine metabolite is color-coded by its corresponding super pathway. The y-axis represents the negative natural logarithms of P-values before multiple comparison adjustments (smaller P-values are higher on the y-axis). The red horizontal line marks a P-value of 0.05, and the orange line marks a P-value of 0.01. A total of 67 metabolites had P-values < .05 (above the red horizontal line) in the female and male combined cohort.

The heatmap describing the correlations among metabolites with a nominal significant association with kidney cancer is presented in Supplementary Fig. S1. As expected, many urine metabolites were highly correlated. Therefore, we used conditional logistic regression with stepwise backward selection to identify independent associations. We included all urine metabolites with P-values less than .05 from individual association analyses in this analysis selection with model entry criteria P-value being .15 and exit criteria P-value being .05. Eleven urine metabolites, including five lipids, two xenobiotics, one nucleotide, and three unidentified metabolites, remained in the final model. Metabolites associated with increased kidney cancer risk are picolinoylglycine (OR [95% CI]: 2.01 [1.44, 2.79]), beta-guanidinopropanoate (OR [95% CI]: 1.75 [1.32, 2.32]), glycerophosphorylcholine (GPC) (OR [95% CI]: 1.60 [1.16, 2.22]), X-24 493 (OR [95% CI]: 1.47 [1.11, 1.94]), and X-24 411 (OR [95% CI]: 1.41 [1.08, 1.84]) (Table 3). Metabolites associated with decreased kidney cancer risk are pregnanediol-3-glucuronide (OR [95% CI]: 0.56 [0.39, 0.82]), 3alpha,21-dihydroxy-5beta-pregnane-11,20-dione 21-glucuronide (OR [95% CI]: 0.61 [0.44, 0.85]), 3-methylglutarate/2-methylglutarate (OR [95% CI]: 0.63 [0.45, 0.85]), 4-vinylphenol sulfate (OR [95% CI]: 0.66 [0.49, 0.90]), allantoic acid (OR [95% CI]: 0.71 [0.54, 0.92]), and X—21 825 (OR [95% CI]: 0.68 [0.53, 0.88]). Adjustment for known and suggested risk factors for kidney cancer, including daily polyunsaturated fatty acid intake, did not have a substantial impact on these associations (Table 3).

Table 3.

ORs of kidney cancer associated with urine metabolites from conditional logistic regression model.^a

Super pathway	Metabolites (Cases = 197, Controls = 197)d	Crude OR	95% CI		P-valuec	Adjustedb OR	95% CI		P-valuec
Lipids	Picolinoylglycine	2.01	1.44	−2.79	<.001	1.97	1.38	−2.80	<.001
	glycerophosphorylcholine (GPC)	1.60	1.16	−2.22	.005	1.53	1.07	−2.19	.021
	3-methylglutarate/2-methylglutarate	0.63	0.45	−0.90	.011	0.66	0.45	−0.95	.027
	3alpha,21-dihydroxy-5beta-pregnane-11,20-dione 21-glucuronide	0.61	0.44	−0.85	.003	0.52	0.35	−0.77	.001
	pregnanediol-3-glucuronide	0.56	0.39	−0.82	.003	0.59	0.40	−0.88	.010
Xenobiotics	beta-guanidinopropanoate	1.75	1.32	−2.32	<.001	1.76	1.29	−2.40	<.001
	4-vinylphenol sulfate	0.66	0.49	−0.90	.008	0.61	0.43	−0.86	.005
Nucleotides	allantoic acid	0.71	0.54	−0.92	.010	0.71	0.53	−0.94	.018
Unidentified	X—24493	1.47	1.11	−1.94	.008	1.62	1.18	−2.21	.003
	X—24411	1.41	1.08	−1.84	.012	1.47	1.10	−1.97	.010
	X—21825	0.68	0.53	−0.88	.003	0.64	0.49	−0.85	.001

^aApplied stepwise backward selection for metabolites presented in the model (P_entry = .15, P_stay = .05).

^bAdjusted for current smoking, drinking alcohol, baseline BMI, hypertension, baseline diabetes, family history of cancer, and the natural logarithmic daily polyunsaturated fatty acids intake (g/day).

^c P-values were calculated based on the Wald-statistics from conditional logistic regression model.

^dThis applies to all 11 metabolites.

The association between metabolite X-24 493 and kidney cancer significantly differed by sex (adjusted OR [95% CI] among females: 2.05 [1.30, 3.23] versus 0.93 [0.55, 1.59] among males; P-value for interaction = .043, Supplementary Table S3). We did not find effect modification of sex on other metabolite associations. Analyses stratified by time interval between urine collection and cancer diagnosis found that the positive metabolite-kidney cancer associations were more pronounced during the period shortly following baseline enrollment (<3 years) versus those diagnosed later (Table 4). For example, ORs of 6.45 [1.28, 32.43], 5.45 [1.13, 26.39], 8.22 [1.16, 40.18], and 4.39 [1.31, 14.37] for picolinoylglycine, GPC, beta-guanidinopropanoate, and X—24 411 for cases diagnosed within 3 years versus 1.78 [1.20, 2.62], 1.46 [0.97, 2.20], 1.49 [1.05, 2.12], and 1.19 [0.86, 1.64] for cases diagnosed after 3 years (P-values for interaction < .05 for all named metabolites). In contrast, we observed the inverse associations among 3-methylglutarate/2-methylglutarate and allantoic acid, which were weaker among cases diagnosed soon after baseline (<3 years) than those diagnosed later (P-values for interaction < .02 for both metabolites).

Table 4.

ORs of kidney cancer associated with urine metabolites from the conditional logistic regression model, by time between urine sample collection and kidney cancer diagnosis.

		Within 3-years between urine sample collection and diagnosis (N = 49 pairs)				More than 3-years between urine sample collection and diagnosis (N = 148 pairs)				P-valuec for interaction
Super Pathway	Metabolites (Cases = 197, Controls = 197)d	Adjustedb OR	95% CI		P-valuec	Adjustedb OR	95% CI		P-valuec
Lipid	Picolinoylglycine	6.45	1.28	−32.43	.024	1.78	1.20	−2.62	.004	.0131
	glycerophosphorylcholine (GPC)	5.45	1.13	−26.39	.035	1.46	0.97	−2.20	.071	.0274
	3-methylglutarate/2-methylglutarate	0.62	0.19	−2.00	.421	0.56	0.37	−0.85	.006	.0166
	3alpha,21-dihydroxy-5beta-pregnane-11,20-dione 21-glucuronide	0.08	0.01	−0.70	.023	0.57	0.39	−0.85	.006	.1409
	pregnanediol-3-glucuronide	0.36	0.06	−2.23	.274	0.56	0.36	−0.88	.012	.1522
Xenobiotics	beta-guanidinopropanoate	8.22	1.68	−40.18	.009	1.49	1.05	−2.12	.025	.0212
	4-vinylphenol sulfate	0.68	0.27	−1.75	.428	0.65	0.44	−0.96	.031	.1023
Nucleotide	Allantoic acid	0.87	0.37	−2.06	.751	0.64	0.46	−0.90	.009	.0194
Unidentified	X—24 493	2.29	0.67	−7.86	.189	1.89	1.30	−2.74	<.001	.0591
	X—24 411	4.39	1.31	−14.72	.017	1.19	0.86	−1.64	.291	.0300
	X—21 825	0.75	0.38	−1.45	.387	0.72	0.52	−0.99	.044	.5149

^aApplied stepwise backward selection for metabolites presented in the model.

^bAdjusted for the log transformation of daily Polyunsaturated fatty acids intake (g/day).

^c P-values were calculated based on the Wald-statistics from conditional logistic regression model.

^dThis applies to all 11 metabolites.

Pathway analysis showed that the 67 metabolites with significant ORs for kidney cancer were overly presented in two pathways, including purine metabolism (P = .001) and pantothenate biosynthesis (P = .01).

Discussion

In this nested case-control study involving 197 kidney cancer cases and individually matched cancer free controls, we evaluated 1301 metabolites in prediagnostic urine and identified 67 metabolites nominally associated with kidney cancer. Among these, 11 metabolites remained significantly associated with kidney cancer in multivariable modeling with stepwise backward selection. These include metabolites in the lipid, xenobiotics, nucleotides pathways, and three metabolites with unknown identity.

Picolinoylglycine and GPC, two lipid metabolites, were associated with 60%–100% increased odds of kidney cancer, and increment reached to more than 400% within 3 years of urine sample collection, although the latter estimates were imprecise due to the small sample size. A review by Weiss et al. (9) summarized the current literature highlighting multiple pathways undergoing metabolic reprogramming in kidney cancer, such as glycolysis, fatty acid oxidation, and lipid synthesis, among others. Previous studies in ccRCC, which is the most common type of kidney cancer, showed that renal cancer cells have increased fatty acid utilization (19). Moreover, a study by Horiguchi et al. found that the enzyme used in synthesizing fatty acids and cholesterol, fatty acid synthase (FASN) has higher expression in ccRCC. They also found an association between FASN and elevated tumor severity and unfavorable prognosis (20). Additionally, multiple studies found notably lower lipogenesis in various cancer models, including RCC (21,22). We also found another three lipids associated with kidney cancer represented by 3-methylglutarate/2-methylglutarate, 3alpha, 21-dihydroxy-5beta-pregnane-11, 20-dione 21-glucuronide, and pregnanediol-3-glucuronide. This is consistent with two previous analyses conducted among the Chinese population that found several lipids associated with kidney cancer such as 17-methyltestosterone, 2-hydroxylauroylcarnitine, and 2- hydroxymyristoylcarnitine, although they did not overlap with the kidney cancer associated metabolites found in our study (23,24).

A distinctive hallmark of cancer is the metabolism of choline, depicted by high levels of various metabolites such as phosphocholine (PC), GPC, and total choline-containing compounds. In fact, tumors exhibiting heightened choline metabolism have been termed as possessing a “cholinic phenotype.” In our study, GPC was found to be associated with higher odds of kidney cancer, which aligns with the literature (25,26). The association between urine choline and kidney cancer was also previously described in two different studies conducted in Poland (27) and the USA (11).

Picolinoylglycine is an N-acylglycine formed by acyltransferases transferring a fatty acid group from picolinilic acid to the N-terminal of the amino acid glycine. A review by Prakash et al. (28) found that N-arachidonoylglycine (NArG), which belongs to the N-acylglycine family, exhibited anti-inflammatory activity in vivo. NArG was shown to be involved in inflammatory cell death in addition to its role in the dissolution of inflammation. We hypothesize that inflammation during cancer development induces host defense systems by increasing metabolites, such as picolinoylglycine, with anti-inflammatory activity and immune cell death properties. This hypothesis will need to be tested in in vivo and human studies.

Beta-guanidinopropanoate, a xenobiotic, was found to be positively associated with kidney cancer risk in our study, while another xenobiotic, 4-vinylphenol sulfate, was found to be associated with lower odds of kidney cancer. Derangements of xenobiotics in urine have been observed in several cancers, especially the ones associated with p-glycoprotein mutations (29,30). These mutations impact the cells’ ability to excrete drugs and metabolites, which might explain the derangement in xenobiotics we found in our study (29,30). The stronger association we found for beta-guanidinopropanoate shortly before cancer diagnosis supports this hypothesis.

Allantoic acid is a product of purine metabolism in the urine (31) of most mammals, except in humans where it is produced from uric acid oxidation (32). This metabolite was significantly associated with several kidney diseases, including chronic kidney disease, diabetic nephropathy, and polycystic kidney diseases (33). It is important to note that all the aforementioned diseases are associated with increased risk of kidney cancer (34). Here, we found new evidence that it may also be associated with kidney cancer risk.

Previous urine (11,27,35), plasma (23,36), and serum (36,37) metabolomics studies demonstrated that several amino acids were associated with kidney cancer. Although we found several amino acids to be associated with kidney cancer in univariate analysis that overlap with the previously reported metabolites, such as aspartate and glycine derivatives, none of these remained significant in the multivariate models. A study by Kim et al. used metabolomics methods to analyze urine samples from 29 kidney cancer patients and 33 control patients. After adjusting for false discovery rate (FDR), they reported three metabolites to be significantly different between cases and controls at an FDR of 0.26, with quinolinone being higher in cancer patients’ urine samples, and 4-hydroxybenzoate and gentisate being lower (38). In our study, we also found benzoate to be associated with kidney cancer risk in univariate analysis, but this association did not remain in multivariate analysis.

Previous studies have reported that alterations in lipid metabolism and nucleotide metabolism were involved in kidney cancer pathogenesis (39,40). In fact, lipid and glycogen-rich deposits in the cytoplasm are some of the histologically defining features of ccRCC (40). One study identified the importance of lipid deposition and the regulation of fatty acid metabolism by HIF as a requirement for ccRCC tumorigenesis (40). They identified carnitine palmitoyltransferase 1A, as a direct HIF target gene and reported that CPTA1 inhibition enables lipid deposition and tumorigenesis (40). Another study found that changes in genes related to one-carbon metabolism, which is the foundation of nucleotide metabolism, were associated with ccRCC risk in a Chinese population (39). They reported strong associations between ccRCC risk and two of the one-carbon metabolism-related genes: rs706209 (P = .006) in CBS (gene encoding cystathionine beta-synthase enzyme) and rs9332 (P = .027) in MTRR (gene encoding methionine synthase reductase enzyme) where those carrying one or more variant alleles in these two genes had a significantly decreased risk of ccRCC (OR; 95% CI: 0.73 [0.06–0.90]) (39).

Few prospective studies were conducted to investigate metabolomics and kidney cancer. One such example is a study by Guida et al. which conducted a case-control study nested within multiple prospective cohorts (the MetKid Consortium). They identified 25 metabolites in blood samples robustly associated with kidney cancer risk; with glycerophospholipids being inversely associated and several amino acids being positively associated with kidney cancer risk (15). Another case-control study nested within the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial by McClain et al. was done to prospectively identify changes in metabolism associated with kidney cancer. Six serum metabolites, including two glycerophospholipids, one sphingolipid, and an organic nitrogen compound, were found to be inversely associated with kidney cancer risk, while two acylcarnitines were positively associated with risk (14). In our study, we found that an overlapping metabolite, GPC, which is a glycerophospholipid, was positively associated with kidney cancer risk. The little overlap of findings between current and previous studies could be due to the different sources of biological samples (urine versus blood) used. Racial differences may also play a role, as both aforementioned prospective studies were conducted in a predominantly European/White population, while our study was conducted among Asians. Variations in metabolomic profiles between different races and ethnicities have been previously suggested by several studies (41). For example, a study by Hu et al. examined 472 metabolites, over 50% of which differed significantly between Black and White women, in the Women’s Health Initiative-Observational Study and WHI-Hormone Therapy trials after adjusting for several factors. These differences could contribute to the inconsistent findings.

Pathway analysis showed that two significant metabolic pathways: purine metabolism and pantothenate and coenzyme A biosynthesis, were associated with kidney cancer risk. Previous studies have highlighted the importance of the nucleotide pathway in several kidney diseases (33). The pantothenate pathway was also previously shown to be associated with both liver and renal cell cancer (25). On the other hand, previous metabolomics studies (23,24,27) suggested other pathways were involved, including lipid and amino acids metabolism, yet we did not find these pathways to be significantly enriched in our pathway analysis, although 5 out of the 11 metabolites found on backward selection in our study were lipid metabolites.

Upon stratification by sex, we found that the association between metabolite X-24 493 and kidney cancer significantly differed by sex, where it was positively associated kidney cancer risk in females but not associated with kidney cancer risk in males. Several studies examined sex-specific differences in kidney cancer, such as a review by Peired et al. (2) which examined the role of hormones, genes, comorbidities, and other risk factors between males and females. Males have about a 2-fold higher risk of developing kidney cancer in their lifetime compared with females, which may be linked to several factors such as the higher incidence of smoking and exposure to occupational hazards in males (2). Results from the Women’s Health Initiative, a large prospective study in the USA, revealed an association between a hysterectomy and higher kidney cancer risk (2). Interestingly, high parity was also found to be associated with kidney cancer risk (2). The authors suggested that might be due to the myriad of physiological and hormonal changes in pregnant females, such as increased estrogen levels, renal hyperfiltration, and weight gain (2). Future validation studies and additional research on the identity of metabolite X-24 493 would shed more light on its potential sex specific association with kidney cancer risk.

Strengths of our study include its prospective study design, comprehensive evaluation of metabolites, and adjustment for multiple potential confounders. The main limitation of our study is that metabolomics was done using a spot urine sample which may be subject to the influence of day to day urine volume and concentration variation (42). In our study, we adjusted for urine osmolality concentration to minimize the influence of urine concentration. Previous studies have also shown comparable results when using spot urine samples and 24-hour urine collection in identifying and measuring metabolites for diagnosing cancers such as neuroblastoma, pheochromocytomas, and paragangliomas (43). However, the change of metabolite concentrations in urine over time cannot be captured in a single urine sample collected at baseline. Thus, some disease-related metabolites may have been missed (44). Moreover, analyses stratified by sex and time to diagnosis yielded large ORs with wide CIs. This imprecision is mainly due to the limited sample size within each stratum which could result in an overestimation of the association. Thus, these results should be interpreted with caution. We found several metabolites with unknown identity being significantly associated with kidney cancer risk. Future studies are warranted to reveal their structural identity in order to facilitate the understanding of their biological relevance to kidney cancer etiology.

There is a potential concern of overfitting and unstable estimates in our study due to the backward regression approach and a small sample size involved, which highlights the need for replications (45). To mitigate these concerns, we only considered metabolites for inclusion in the backward stepwise selection if their associations with kidney cancer were of nominal significance (P < .05) and the BH-FDR was less than 0.20 to minimize chance findings due to multiple comparisons and overfitting (46). It should be acknowledged that our study is exploratory in nature, with the aim being hypothesis generation, seeking clues on biological mechanisms of kidney carcinogenesis, and laying a foundation for future validation studies by our own group and/or the broader scientific community. Our study was conducted in an Asian population. These results, thus, may not be generalized to other populations. Future validation studies should include ethnically and geographically diverse populations. Nevertheless, combining evidence from studies conducted predominantly in the European/White population, we found a strong justification for conducting further investigations on the associations of metabolites with kidney cancer risk.

In summary, this pilot study provides the first evidence suggesting that lipid and nucleotide disturbances may play an important role in kidney cancer development. These findings showcase the potential of urine metabolomics in the etiological research for kidney cancer and unraveling of the underlying biologic mechanisms. Further studies with a sufficiently large sample size and inclusion of multiple ethnic groups are needed to validate our findings and extend our understanding of the role of metabolomics in kidney cancer.

Glossary

Abbreviations

α

Alpha

BH-FDR

Benjamini–Hochberg false discovery rates

BMI

Body Mass Index

ccRCC

clear cell Renal Cell Carcinoma

CI

confidence intervals

FASN

fatty acid synthase

FDR

False Discovery Rate

GPC

glycerophosphocholine; Hypoxia Inducible Factor

IQR

interquartile range

NArG

N-arachidonoylglycine

OR

Odds Ratio

PC

phosphocholine

Q

quartile

RCC

Renal Cell Carcinoma

SD

Standard Deviation

SMHS

Shanghai Men’s Health Study

SWHS

Shanghai Women’s Health Study

UPLC-MS/MS

ultrahigh performance liquid chromatography/tandem mass spectrometry

US

United States

Author contributions

T.A.-S. and S.S. were involved in data analysis, visualization, and drafting the manuscript. X.-O.S. conceptualized, designed, and supervised the study. H.C., Y.G., Q.C., W.Z., and X.-O.S. were involved in data curation and biospecimen procurement/management. X.-O.S. and W.Z. obtained the funding for the research. All authors have critically reviewed and approved manuscript.

Ethics Statement

Not applicable.

Conflict of interest

The authors declared no potential conflicts of interest.

Funding

This work was supported by grants from the National Institutes of Health (NIH) (UM1 CA182910, to W.Z., UM1 CA173640 to X.S.) and in part by the Vanderbilt-Ingram Cancer Center supporting grant (P30 CA68485 for sample preparation which was conducted at the Survey and Biospecimen Shared Resources). T.A.-S. is supported by the Vanderbilt Training Program in Molecular and Genetic Epidemiology of Cancer (T32CA160056 to X.S.). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the study.

Data availability

Data available from the corresponding author upon reasonable request.