Impact of one-year soy protein supplementation on serum metabolomic profiles: results from a randomized, placebo-controlled trial in prostate cancer patients

Abstract

Purpose –

This study aims to investigate global metabolomic effects of soy protein supplementation.

Design/methodology/approach –

This study was designed as an exploratory study based on residual specimens from a randomized placebo-controlled trial. The authors investigated changes in serum metabolomic and lipidomic profiles of prostate cancer patients who were assigned to soy protein (19.2g/day, 41mg isoflavones, n = 10) or a casein-based placebo (19.8g/day n = 11) for up to one year following prostatectomy. A total of 63 serum samples collected at baseline, 6 and 12months were analyzed by untargeted, global metabolomics.

Findings –

Partial least squares discriminant analysis, based on 493 polar metabolites and 4,706 lipids, showed clear separation of changes in metabolomic profiles between the soy and placebo groups. However, the only metabolite that was significantly different after multiple comparison adjustment was daidzein 4′–sulfate (p = 0.036). A few other metabolites/lipids were weakly associated with soy supplementation; betaine and C18-ceramide derivative increased and L-α-aminoadipic acid and a linolenic acid derivative decreased, compared with the placebo.

Originality/value –

The present study demonstrates that physiologically relevant dietary levels of soy protein supplementation close to real-life levels of consumption lead to a shift in global metabolomic profiles.

Introduction

Soybeans are a key component of plant-based diets, and their potential health benefits have been attributed to bioactive phytochemicals, particularly isoflavones/phytoestrogens (Fan et al., 2022; Chen et al., 2023). Dietary intakes of soybeans, soy products, soy protein and isoflavone/phytoestrogens have been associated with decreased risks of cancer, including prostate cancer, cardiovascular disease and type 2 diabetes mellitus in observational studies (Li et al., 2020). Some randomized clinical trials (RCTs) using soybean products, soy protein, isoflavone and its major component, genistein, have also demonstrated favorable effects on blood pressure, blood lipid profiles, endothelial function and cognitive function (Li et al., 2020).

A broad range of metabolic responses to dietary soy intake may stimulate cellular signaling pathways linked to health-promoting effects of soy. While exposure to dietary soy products can be captured by food frequency questionnaires or dietary records to a limited extent (Rienks et al., 2017), untargeted metabolomic approaches have become increasingly used in objectively identifying endogenous responses to exogenous dietary exposures. In a cohort study of 491 individuals, diet was a strong predictor of serum metabolite profile, explaining almost 50% of the observed variance (Bar et al., 2020).

Because of the weak estrogenic properties of soy phytoestrogens (Cederroth et al., 2012), the preventive effects of soy intake on prostate cancer have been studied more often than other cancers in observational studies (Applegate et al., 2018; Bosland, 2025). The results from a meta-analysis of seven cohort and nine case–control studies suggest that higher intake of soy products may be protective against the development of prostate cancer [risk ratio 0.71, 95% confidence interval (CI) 0.58–85], particularly for localized and low-grade cancer (Applegate et al., 2018). However, direct evidence in support of chemopreventive effects of soy from RCTs is rather limited (Bosland, 2025). An earlier meta-analysis of 11 RCTs did not show consistent effects on prostate-specific antigen (PSA) or testosterone levels (van Die et al., 2014), while the pooled risk estimate of two short-term trials was indicative of prostate cancer incidence reduction in the treatment arm compared to placebo (risk ratio 0.49, 95 CI 0.26, 0.95) (van Die et al., 2014). Our RCTs in men who were at risk of prostate cancer recurrence following radical prostatectomy using an isoflavone-containing soy protein isolate beverage or a casein-based placebo for two years did not find the effect on biochemical recurrence (Bosland et al., 2013).

We postulate that long-term soy exposure leads to systemic metabolic alterations that are not associated directly with PSA synthesis. Leveraging residual serum samples from this RCT, we conducted an exploratory study to identify metabolomic signatures of long-term exposure to soy proteins in men and to address whether any associated alterations in metabolic pathways coincide with metabolic reprogramming associated with prostate cancer (Lasorsa et al., 2023).

Methods

This study was a secondary exploratory analysis of stored serum samples from a double-blind, randomized, casein-based placebo-controlled clinical trial with soy protein isolate (registered at clinicaltrials.gov; identifier: NCT00765479) as described previously (Bosland et al., 2013). Prostate cancer patients who underwent radical prostatectomy within past 4months and were at high risk of recurrence because of unfavorable pathology were enrolled between July 1997 and November 2005 at the Manhattan VA Medical Center or May 2009 at New York University School of Medicine. A total of 159 men were randomized to soy protein (19.2g per day, n = 81) or placebo (calcium caseinate 19.8g per day, n = 78), and ultimately evaluable for the previously reported study (Bosland et al., 2013). The project has been approved by University of Illinois Chicago Institutional Review Board (#2006–0706).

In the parent study, study participants were asked about their medical history and use of medications, nutritional supplements and soy intake at baseline and at follow-up visits every 2months in year 1 and every three months in year 2 of the study. At each study visit, nonfasting blood samples were taken. For this metabolomic study, we utilized serum samples collected at baseline, 6 months and 12months from subjects who did not develop biochemical recurrence and were fully compliant with the treatment (>90% consumption). Samples were available at all time points from a total of 21 patients (soy n = 10; placebo n = 11, see flowchart in Supplemental Methods).

These 63 samples were analyzed using comprehensive two-dimensional liquid chromatography–mass spectrometry (2DLC-MS) for polar metabolites and for lipids. Differential analysis was conducted between groups based on the slopes (i.e. changes) from baseline (i.e. 0month) to 6months, baseline to 12months and across all three time points. Details concerning laboratory and statistical analyses are described in Supplemental Methods.

Results

Both soy and placebo groups primarily consisted of White men and subjects in the soy group were approximately 6.0years younger than those in the placebo group. Otherwise, characteristics of both groups were similar (Table 1).

Table 1.

Characteristics of patients randomized to soy protein or placebo

Subject characteristics	Treatment arm		p-valuea
	Soy protein	Placebo
Number of evaluable subjects	10	11
Age, years (mean±SD)	57.6±5.5	63.6±7.0	0.04
Baseline
Body weight (lbs.) (mean±SD)	190±37	194±46	0.838
Body height (inch) (mean±SD)	69.9±3.2	70.3±3.6	0.807
Body mass index (kg/m2) (mean±SD)	27.3±4.3	27.5±5.4	0.911
Race
White (Caucasian)	10 (100%)	10 (91%)	1.000
Black (African American)	0	1 (9%)
Location
Tisch Hospital	10 (100%)	10 (91%)	1.000
Manhattan Veterans’ administration	0	1 (9%)
Subjects taking any medication during trial	6 (60%)	10 (91%)	0.149
Baseline serum genistein levels (ng/mL) (mean±SD)b	1.84±3.21	3.95±10.89	0.552
Six month serum genistein levels (ng/mL) (mean±SD)b	142.7±91.3	6.07±9.02	0.001
12 Month serum genistein levels (ng/mL) (mean±SD)b	109.3±47.3	5.20±7.59	<0.0001

Note(s):

^ap-values obtained from two-sided χ², Fisher exact or t-tests, whichever was appropriate.

^bMeasured in the parent study using targeted high-performance liquid chromatography (HPLC)

Source(s): Authors’ own work

The 2DLC-MS positive mode (+) detected 171–292 (median 265) annotated peaks (i.e. identified metabolites) for both treatment groups and all time points combined, while the 2DLC-MS negative mode (−) detected 78–173 (median 166) annotated peaks for all samples. Lipid 2DLC-MS (+) identified 2,698–3,138 (median 2,898) annotated peaks and lipid 2DLC-MS (−) 867–1,227 (median 1,012) annotated peaks. In total, 43% of MS (+) lipids and 38.2% of MS (−) lipids were linked to single lipid species, while the remaining annotation peaks included multiple lipid species.

To assess the treatment effects of soy supplementation on serum metabolites, we calculated a slope for changes in metabolite level over time (baseline, 6 and 12months) for individual subjects. To identify the overall compositional changes in metabolites by treatment group, we performed principal component analysis (PCA) [Figure 1(a)] and partial least squares discriminant analysis (PLS-DA) [Figure 1(b)] using these slope data for all polar and lipid metabolites combined. While unsupervised PCA did not show any clustering by treatment, supervised PLS-DA yielded clear separation/clustering of the two treatment groups. These observations were replicated when metabolites were grouped into lipid, polar metabolites, positive and negative modes (Supplemental Figure 1).

Figure 1.

Quantitative metabolite profiling of slopes across 0, 6 and 12months between groups. (a) Principal component analysis (PCA) plot. (b) Partial least square-discriminant analysis (PLS-DA) plot. The circles represent 95% confidence ellipses. (c and d) Longitudinal plots with mean log₂ abundance and the change across months 0, 6 and 12 for the two polar metabolites, daidzein 4′-shulfate_1 and daidzein 4′-shulfate_2

Note(s): Legend/group: orange: soy; blue: placebo

Source: Authors’ own work

Table 2 presents a list of polar metabolites that showed a statistically significant difference by treatment at the conventional p-value cutoff level of 0.05. The only metabolite that was statistically significant after multiple comparison adjustment by false discovery rate (FDR) was daidzein 4′-sulfate_2, a soy isoflavone metabolite (p = 0.036), which was detected by two peaks at different retention times (RT). As illustrated in Figure 1(c) and (d), the relative abundance of daidzein 4′-sulfate increased more than 70 times in the soy supplementation group compared with placebo after 6 and 12months. On the other hand, we found that an aglycone of a secondary isoflavone metabolite, dihydrogenistein, showed a declining trend over time with soy supplementation compared with the placebo. Two amino acids, which could be either endogenous or exogenous, were marginally altered by soy supplementation, higher betaine and lower L-α-aminoadipic acid relative abundances compared with the placebo (both p = 0.044). Other metabolites whose relative abundance showed a declining trend with soy supplementation include those of exogenous (dietary) origins, i.e. L-threonic acid, (2S)-1-hydroxy-3-(pentadecanoyloxy)-2-propanyl (9Z,12Z,15Z)-9,12,15-octadecatrienoate, (25S)-11α,20,26-trihydroxyecdysone and 3-aminobutanoic acid, as well as the glycerophospholipids, 1-oleoyl-sn-glycero-3-phospho-_D-myo-inositol and (2R)-2-(acryloyloxy)-3-hydroxypropyl 2-(trimethylammonio) ethyl phosphate.

Table 2.

Polar metabolites that differentially changed over treatment

Compound name	Description (sources)	Mode	m/z	RT	Relative abundance at time 0		Relative abundance at 6 months		Relative abundance at 12 months		Slope over 0–12 months		p-valuesa for differences in slopes
					Soy	Placebo	Soy	Placebo	Soy	Placebo	Soy	Placebo
Daidzein 4′-sulfate_2	Isoflavone	Negative	333.01	15.28	−9.64	−9.63	−3.61	−9.83	−3.55	−10.19	0.51	−0.05	0.000
Daidzein 4′-sulfate_1	Isoflavone	Negative	333.01	4.80	−9.64	−9.69	−5.20	−10.08	−5.20	−10.67	0.37	−0.08	0.002
3-Hydroxy-N-[(3S)-2-oxotetrahydro-3-furanyl] octanamide	Fatty amides (fungi)	Positive	244.15	15.73	0.48	1.66	0.85	1.39	1.45	−0.12	0.08	−0.15	0.033
Betaine	Amino acid	Positive	118.09	1.17	9.71	9.97	9.74	9.73	9.71	9.47	0.00	−0.04	0.044
l-α-Aminoadipic acid	Amino acid	Negative	160.06	1.37	−0.46	−0.39	−0.15	−0.39	−0.62	0.03	−0.01	0.03	0.044
1-Oleoyl-sn-glycero-3-phospho-d-myo-inositol	Glycerophospholipids	Negative	597.30	5.49	−3.05	−3.85	−3.55	−3.46	−3.35	−3.05	−0.02	0.07	0.013
6α-Hydroxy-castasterone	Plant steroid hormone	Negative	465.36	2.26	−0.68	−0.76	−1.07	−0.84	−1.01	−0.63	−0.03	0.01	0.017
2,3-Bis(octanoyloxy)propyl 10-methyldodecanoate	Glycerolipids	Negative	539.43	2.25	−1.11	−1.34	−1.69	−1.31	−1.64	−1.17	−0.04	0.01	0.015
l-Threonic acid	Vitamin C metabolite	Negative	135.03	1.21	7.07	6.63	6.64	6.77	6.48	6.74	−0.05	0.01	0.021
3-Aminobutanoic acid (BABA)	Amino acid (dairy)	Positive	104.07	6.34	4.12	4.18	3.84	4.10	3.44	3.99	−0.06	−0.02	0.049
β-d-Glucopyranuronic acid	Carboxylic acids, endogenous conjugation metabolites	Negative	193.04	7.57	0.56	−0.49	−0.22	−0.14	−0.36	−0.24	−0.08	0.02	0.021
(25S)-11α,20,26-trihydroxyecdysone	Plant steroid hormone	Negative	511.29	5.14	−2.18	−2.79	−2.67	−3.09	−3.17	−2.98	−0.08	−0.02	0.014
Dihydrogenistein	Isoflavone	Negative	271.06	17.67	−6.11	−8.99	−7.10	−8.44	−7.68	−7.18	−0.13	0.15	0.002
6-Hydroxysphing-4E-enine	Sphingolipid	Positive	316.29	4.16	−4.66	−9.15	−8.67	−7.28	−7.32	−4.18	−0.22	0.41	0.001
(2S)-1-Hydroxy-3-(pentadecanoyloxy)-2-propanyl (9Z,12Z,15Z)-9,12,15-octadecatrienoate	Lineolic acid derivative DAG(15:0 / 18:4)	Positive	577.48	2.24	−0.96	−1.74	−2.68	−1.09	−3.83	−0.99	−0.24	0.06	0.022
(2R)-2-(Acryloyloxy)-3-hydroxypropyl 2-(trimethylammonio) ethyl phosphate	Glycerophospholipid	Positive	312.12	5.54	−5.52	−9.90	−7.16	−5.25	−8.90	−6.94	−0.28	0.25	0.023

Note(s):

^aRaw p-values not adjusted for multiple comparisons. With multiple comparison adjustment (false discovery rate correction), only daidzein 4′-sulfate_1 was significant at <0.05

Source(s): Authors’ own work

Lipid metabolites were first analyzed by major class (n = 49), separated by positive and negative mode. Six major classes showed an association with soy treatment at p < 0.05 level, including acidic glycosphingolipids, other glycerolipids, oxidized glycerophospholipids, phosphosphingolipids, steroids and steroid conjugates; all were from the positive mode (Table 3). We further examined individual lipid species within these four groups (Table 4). Lipid species significantly altered by supplementation in other glycerolipids and phosphosphingolipid groups consisted of those isolated from marine organisms and insects, thus possibly contaminants (data not shown). The overall trends of steroid lipid abundance showed an increase over soy supplementation, but those trends were primarily limited to exogenous steroid hormone compounds. Among endogenous sex hormones, a major metabolite group designated as epitestosterone, dehydroandrosterone, 1-testosterone and testosterone showed a 70% decline with soy supplementation while virtually no change occurred in the placebo group, but the between-group difference was not statistically significant (p = 0.10).

Table 3.

Change in major classes of lipid metabolites over treatment

Lipid major class	No. of lipids included	Relative abundance at time 0		Relative abundance at 6 months		Relative abundance at 12 months		Slope over 0–12 months		p-valuesa for differences in slopes
		Soy	Placebo	Soy	Placebo	Soy	Placebo	Soy	Placebo
Acidic glycosphingolipids	59	3.276	3.500	3.146	3.300	3.565	3.139	0.024	−0.030	0.026
Other glycerolipids	3	−4.734	−4.607	−4.934	−3.387	−5.466	−3.689	−0.061	0.077	0.023
Oxidized glycerophospholipids	79	1.408	1.649	1.465	1.669	1.661	1.401	0.021	−0.021	0.042
Phosphosphingolipids	245	6.621	6.964	6.659	6.936	6.771	6.639	0.012	−0.027	0.027
Steroid conjugates	150	1.527	1.917	1.752	2.089	1.824	1.685	0.025	−0.019	0.029
Steroids	69	−1.778	−1.374	−1.779	−1.365	−1.540	−1.726	0.020	−0.029	0.020

Note(s):

^aRaw p-values not adjusted for multiple comparisons

Source(s): Authors’ own work

Table 4.

Change in selected lipid species over treatment

Lipid names	m/z	RT	Relative abundance at time 0		Relative abundance at 6 months		Relative abundance at 12 months		Slope over 0–12 months		p-valuesa for differences in slopes
			Soy	Placebo	Soy	Placebo	Soy	Placebo	Soy	Placebo
(3′-Sulfo)Galbeta-Cer (d18:1/16:0(2OH))	796.524	40.42	−3.05	−1.22	−2.80	−3.20	−2.75	−3.25	0.02	−0.17	0.043
	796.524	78.81	−1.73	−4.75	−2.80	−2.23	−2.85	−1.46	−0.09	0.27	0.097
	796.524	8.20	−4.69	−3.26	−3.59	−2.89	−3.44	−2.74	0.10	0.04	0.641
	796.524	60.45	0.19	−0.11	0.55	0.70	0.50	−1.29	0.03	−0.10	0.582
	796.524	41.68	−0.05	−1.60	−1.43	−2.46	−1.46	−2.51	−0.12	−0.08	0.804
	813.550	78.00	−4.52	−4.64	−3.47	−2.84	−2.44	−2.22	0.17	0.20	0.901
	813.550	55.59	1.26	0.57	0.31	−0.31	−0.43	−1.19	−0.14	−0.15	0.970
GlcAbeta-Cer(d18:1/18:0)	742.583	78.97	6.55	6.49	6.65	5.99	6.87	6.18	0.03	−0.03	0.028
	759.609	61.75	1.74	1.46	1.62	1.30	2.04	0.78	0.02	−0.06	0.134
	742.583	60.81	10.11	10.05	9.53	9.58	10.30	9.79	0.02	−0.02	0.398
	759.609	44.21	−4.80	−3.82	−5.01	−4.73	−4.86	−4.23	−0.01	−0.03	0.744
OOB-PS	608.319	53.32	−0.24	0.51	0.20	0.42	0.44	0.50	0.06	0.00	0.049
Glycocholic acid 3-glucuronide	659.375	53.70	0.88	−1.23	−1.28	0.13	−2.66	−0.71	−0.30	0.04	0.024
	642.348	20.95	−4.88	−5.29	−4.95	−5.16	−3.45	−5.25	0.12	0.00	0.064

Note(s):

^aRaw p-values not adjusted for multiple comparisons

Source(s): Authors’ own work

In the acidic glycosphingolipids class, two ceramide derivatives from acidic glycosphingolipids showed increasing relative abundances with soy supplementation, (3′–sulfo)Galbeta-Cer(d18:1/16:0(2OH)) and GlcAbeta-Cer (d18:1/18:0). The former had seven peaks and the latter four peaks at different retention times. While all of the four metabolites of GlcAbeta-Cer (d18:1/18:0) showed an increasing trend, those for (3′-sulfo)Galbeta-Cer(d18:1/16:0(2OH)) were widely variable among the seven metabolites. In the steroid conjugate class, we found associations with soy supplementation for glycocholic acid 3-glucuronide, a bile acid conjugate, at two different retention times, but the directions of the change at these two time points were opposite, offsetting effects of each other. The relative abundances of OOB-PS, an oxidized glycerophospholipid, increased with soy supplementation. Additional statistical adjustments for age in assessing differences in slopes between soy and placebo arms did not make any measurable changes in the p-values reported in Tables 2–4.

Pathway analyses with FDR correction, including all individual polar metabolites and lipids, revealed that the glycerophospholipid metabolism pathway was overrepresented with soy supplementation (p < 0.0001, total of four hits, phosphatidylethanolamine; phosphatidylcholine; phosphatidate and phosphatidylserine, impact factor 0.38583). No other pathways were statistically significantly associated with soy supplementation. Interrelationships among metabolites that showed changes in relative abundance in response to soy supplementation were further examined in network analysis, including both treatment groups (Figure 2, Supplemental Table S1). This analysis revealed that nodes 1 and 37, (2S)-1-hydroxy-3-(pentadecanoyloxy)-2-propanyl (9Z,12Z,15Z)-9,12,15-octadecatrienoate and GlcAbeta-Cer(d18:1/18:0), were two main hubs with a large number of edges, followed by node 89, betaine. Daidzein 4′-sulfate nodes 130 and 88 were connected to these hubs through either L-α-aminoadipic acid (node 43) or several glycerophospholipids (nodes 25, 10 and 56).

Figure 2.

Metabolic subnetwork with 11 metabolites of interest. Dashed edges represent negative edge weights. Solid edges represent positive edge weights. Nodes colored purple represent 11 compounds of interest. Detailed node information can be found in Supplemental Table S1

Source: Authors’ own work

Discussion

In this post hoc comprehensive untargeted metabolomic analysis of residual serum samples from a one-year placebo-controlled clinical trial with soy protein consumption vs a casein placebo among men who were successfully surgically treated for prostate cancer, we obtained several intriguing findings that warrant further investigation. Despite the modest, physiologically relevant soy dose, which was close to that in people with regular dietary soy intake, the intervention produced metabolomic profiles distinguishable between the soy and placebo arms. However, because PCA did not indicate a clear separation of the two groups, the separation we found by PLS-DA may be highly dependent on a few selected metabolites altered by soy supplementation. PCA is a powerful unsupervised tool, but PCA can fail to accurately capture information in high-dimensional data. This may occur particularly in the presence of a myriads of noises/signals, as it assumes a linear relationship among the independent variables and solely depends on the variance of the independent variables not taking variance between the dependent and independent variables into consideration.

The results of the present study indicate that daidzein 4′-sulfate is a good biomarker for soy intake. In a randomized crossover trial among 11 postmenopausal women consuming a soybean-based diet for six weeks, elevation of plasma levels of the soy isoflavone plasma metabolites daidzein 4′-sulfate, genistein and daidzein were significant predictors for the soy diet (Huang et al., 2022). A more recent controlled feeding study in 153 women also reported urinary daidzein sulfate to be a marker of soy intake (Playdon et al., 2024). In both studies, major isoflavone conjugates documented by others, such as genistein sulfoglucuronide and diglucuronide, dihydrogenistein glucuronide and sulfate (Soukup et al., 2016; Kato et al., 2012), were not detected (Huang et al., 2022; Playdon et al., 2024), similar to our findings. This discrepancy may demonstrate a limitation of untargeted metabolomics in identifying low-abundance metabolites that could be undetected due to overlapping peaks of high-abundance metabolites. Thus, targeted metabolomics remains an irreplaceable approach to gaining pharmacokinetic data on soy metabolites.

We also found relative increases in the amino acid betaine, for which soy was the possible dietary source, although it can be endogenously synthesized from choline (Lever and Slow, 2010). Betaine functions as a methyl donor, methylating homocysteine to methionine, and it may alter epigenetic modification of DNA, influencing gene expression (Lever and Slow, 2010). Betaine insufficiency has been associated with the metabolic syndrome, some lipid disorders, diabetes (Lever and Slow, 2010) and coronary artery disease (Shea et al., 2024). Interestingly, an untargeted plasma metabolomic study among 7,012 population samples in Japan found that betaine was an excellent marker for soy product intake (Shibutami et al., 2021). While antioxidant properties of soy isoflavones are well recognized, it is widely acknowledged that compounds with antioxidant activity can exhibit prooxidant behavior under certain conditions and this is the case for genistein and daidzein (Osborn-Barnes and Akoh, 2003; Eren-Guzelgun et al., 2018; Chen et al., 2014). This notion may be relevant to the increased oxidized glycerophospholipids by soy supplementation observed in the present study.

Although we found nonsignificant modest effects on plasma testosterone metabolites, 12months of soy supplementation may not have been sufficiently long to see a significant difference, as our previous study from the same RCT demonstrated that serum testosterone measured by targeted immunoassay progressively declined with soy supplementation up to 18months (Bosland et al., 2022).

Dysregulation of lipid metabolism, particularly alterations in glycerophospholipids, has been linked to prostate cancer development and progression, in addition to potential pro- or anticarcinogenic effects of specific lipid species such as C16-, C18-ceramides (Grösch et al., 2012; Sedić et al., 2019; Saddoughi and Ogretmen, 2013). However, the roles of individual lipids in prostate cancer progression remain inconclusive with mixed results reported of both positive and inverse associations. While large European prospective cohort studies based on plasma lipidomics reported inverse associations of certain glycerophospholipids with prostate cancer incidence, particularly of aggressive cancer (Schmidt et al., 2017, 2020), a tissue-based lipidomic study showed increased several glycerophospholipid contents in cancer tissue compared with adjacent normal tissue (Lima et al., 2022). Furthermore, glycerophospholipid metabolism was associated with drug resistance and high metastatic potential of prostate cancer cells (Ingram et al., 2021; Ding et al., 2025).

Our study had several strengths, including the design of the RCT study, availability of repeated measures over an extended treatment period of 12months, the use of a physiologically relevant soy dose and comprehensive two-dimensional LC-MS techniques. On the other hand, there were several limitations. First, the sample size was small because the study was designed as a post hoc analysis of limited residual samples available for at least three time points, and most of the associations did not remain significant after multiple comparison adjustment. In addition, because this study was based on a subset of the originally randomized samples, there was some imbalance in the basic demographics of study subjects. Although body weight and total cholesterol levels did not show a difference between treatment arms over time in the parent study (Bosland et al., 2021), other unmeasured dietary changes may have involved in the global metabolomic shift that we observed. Furthermore, this analysis was based on nonfasting blood, and we did not strictly standardize the timing of blood draw with respect to timing of supplement intake within and between individuals, which may have increased intra- and inter-individual variabilities of metabolite abundance and reduced detection of soy isoflavone metabolites because of their relatively short half-lives (Bloedon et al., 2002). In addition, the primary endpoint in this study was the slope of changes in metabolite abundance over time. Compared with targeted metabolomics with internal standards, the slope calculated based on relative abundances in untargeted metabolomics is likely to be more prone to measurement errors. Finally, we cannot rule out the possibility that observed associations reflect the effects of casein, rather than soy. Although there is a report of anticarcinogenic effects of casein over soy for experimental colon cancer in rats (McIntosh et al., 1995), there are no data supporting anticarcinogenic effects of casein in humans. Besides, recent metabolomic and lipidomic studies from clinical trials using casein in humans did not show alterations in the metabolites that we observed in this study (Chen et al., 2022; Schmedes et al., 2018). Thus, the results of this study likely represent the effects of soy supplementation.

Conclusions

Despite several limitations discussed above, the present study demonstrates that soy protein supplementation close to real-life levels of consumption in non-Western populations leads to a shift in global metabolomic profiles. However, because of exploratory nature of the study, the clinical translation of our findings is depending on further external validation and laboratory studies to delineate underlying biological mechanisms.

Supplementary Material

The supplementary material for this article can be found online.