The Effects of Soy Supplementation on Gene Expression in Breast Cancer: A Randomized Placebo-controlled Study

Moshe Shike; Ashley S. Doane; Lianne Russo; Rafael Cabal; Jorge S. Reis-Filho; William Gerald; Hiram Cody; Raya Khanin; Jacqueline Bromberg; Larry Norton

Disclosures

J Natl Cancer Inst. 2014;106(9) 

In This Article

Results

Patients

A total of 140 women with invasive breast adenocarcinoma (stage T1, T2, or T3) were randomized to participate in this study from 2003 to 2007.[32] Eight women dropped out (three elected to have surgery elsewhere, two withdrew, and three refused surgery). When available, blood and tumor tissues were analyzed form 132 remaining women (Figure 1). Median durations of soy or placebo supplementation were 14 and 15 days, respectively (P = .70). There were no side effects or complications related to the intervention or placebo. Measurements consisted of: plasma isoflavones (n = 125), tumor IHC (n = 104), NanoString (n = 14), gene expression profiling (n = 51), and qPCR (n = 46) (Figure 1).

Figure 1.

CONSORT diagram for study design and availability of samples for analysis.

Demographics and clinicopathological characteristics including age, race, menopausal status, TNM stage, tumor estrogen receptor (ER) status, HER2 status (by IHC and FISH) showed no differences between the two groups (Table 1). There were no statistically significant differences in baseline weight, BMI, dietary components, or alcohol consumption (Supplementary Table 1, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1).

Plasma Isoflavones

To assess adherence to soy consumption, paired plasma samples were compared before and after treatment for all participants (n = 125). The soy group had a seven-fold increase in plasma genistein (from median 1.6 [range = 0.4 to 64.6] to 11.6 [0.3 to 387.9] ng/mL, P < .001) and a four-fold increase in plasma daidzein (from median 1.5, [range = 0.1 to 55.9] to 6.7 [range = 0.5 to 291.5] ng/ml, P < .001). No statistically significant changes in isoflavone levels were observed in the placebo group (Figure 2). A strong positive correlation was observed between genistein and daidzein levels in the soy group (r = 0.94, P < .001). These results indicated a strong adherence to the assigned treatment.

Figure 2.

Plasma genistein and daidzein following soy intake. Post-treatment isoflavone levels increased compared with pretreatment levels following intake of soy (P < .001) but not placebo. The changes (post-pre) in plasma isoflavones were statistically significantly greater for women receiving soy compared with those receiving placebo (P < .001). * Indicates within-group statistical significance by the Wilcoxon matched-pairs signed rank test, P < .001. Indicates statistical significance for the comparison of the fold-change between treatment groups by the Wilcoxon rank-sum test (P < .001).

The dispersion in posttreatment isoflavone levels in the soy group was large. While in most patients there was a marked increase, in a few levels changed minimally (Figure 2). This may be explained by differences in adherence to treatment, absorption, metabolism, and clearance of soy and its metabolites.[22,32]

NanoString Analysis of Gene Expression Before and After Soy or Placebo

We measured expression of 202 BC-related genes by NanoString analysis in matched tumor samples obtained before and after intervention from 14 BCs. The availability of pretreatment core biopsy tissue limited the sample size to eight patients in the soy, and six in the placebo group. There were no statistically significant between-group differences in patient or tumor characteristics, including ER status (Supplementary Table 2, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1). We identified genes that were statistically significantly changed post-intervention and compared the magnitude and direction of gene expression changes between the two groups (Table 2). Fourteen genes changed in the soy group: 10 increased, and four decreased expression. In the placebo group, 10 genes changed, five increased, and five decreased. Three of these 10 genes were among those that changed in the soy group in the same direction. Thus a total of 21 (out of 202) genes in both groups demonstrated changes. The expression of these genes in pre- and posttreatment tumor samples from both groups is represented in Supplementary Figure 1 (available online) http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1.

To determine gene expression changes, we focused on fold change (posttreatment/pretreatment ratio) for each of the 21 genes, and compared this value between treatment groups. Expression of FANCC and UGT2A1 increased in 87.5% of tumors following soy intake (mean FC = 1.27 and 1.57, P < .05) and decreased (mean FC = -1.26 and -1.33, P value not statistically significant) in the placebo group (Figure 3, A-D). While these fold changes were modest, they were consistent and in opposing direction in the two groups (P < .05). Genes with altered expression in the soy group also included SERPINE1 (mean FC = 2.7, P = .006) (Figure 3E). However, this increase was not statistically significant between soy and placebo groups (P = .26) (Figure 3F).

Figure 3.

Expression of FANCC and UGT2A1. Gene expression was measured in paired samples (pre/post) using NanoString. Expression levels of FANCC (A), UGT2A1 (C), and SERPINE1 (E) were increased following consumption of soy (P < .05), but not placebo. The expression fold changes (FC) were statistically significantly greater in tumors exposed to soy compared with placebo for FANCC (B) and UGT2A1 (D) (P < .05), but not SERPINE1 (P = .26) (F). A heat map and hierarchical clustering of the post/pre FC of the 21 genes DE between paired samples (P < .05) (G). Positive FC are colored red, negative FC are colored blue, and treatment group is indicated below the sample dendogram. *Indicates within-group statistical significance by paired t test P < .05, Indicates statistical significance for the comparison of fold change between treatment groups by unpaired t test (P < .05).

To evaluate the patterns of gene expression changes in the matched tumors, we performed hierarchical clustering of the paired samples using the pre/postexpression fold change of the 21 differentially expressed (DE) genes (Figure 3G). Clustering showed a tendency to organize samples by soy or placebo group, and the heat map showed groups of genes correlated by expression fold change in response to soy or placebo. Genes related to cell cycle functions, including CCNA2, CCNE2, and CDKN1B, were closely grouped by cluster analysis, and demonstrated a pattern of expression with increases in soy and decreases in placebo group samples. These data suggest an effect of soy intake characterized by subtle yet consistent and statistically significant alterations in BC gene expression.

Genome-wide Expression Analysis in Posttreatment Specimens

We performed genome-wide expression analysis of 51 specimens (39 ER+ and 12 ER-) from surgically resected tumors, 28 from soy group, and 23 from placebo. There were no statistically significant between-group differences in demographics or clinicopathological criteria (Supplementary Table 3 and Table 4, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1). We identified 131 differently expressed (DE) genes between the two groups (absolute fold change ≥2 and P < .01). Of these, 11 were overexpressed and 120 were underexpressed in tumors of the soy relative to the placebo group.

We next considered the possibility that genistein plasma levels, rather than assignment to the soy group per se, may be a more relevant marker of genistein effects on BC gene expression. Therefore, we examined differential gene expression as a function of plasma genistein. Median genistein concentration in the soy group was 6.3ng/mL, and 25% demonstrated very low levels (<0.5ng/mL). We therefore limited tumors of the soy group to those from patients with serum genistein greater than 16ng/mL, which corresponded to the 95th percentile concentration of the placebo group. The resulting analysis consisted of posttreatment expression profiles of 11 tumors of the soy group with elevated plasma genistein, and 23 tumors from the placebo group with low plasma genistein, referred to as high and low-genistein subsets, respectively. Tumor characteristics, including ER status, were similar in high- and low-genistein subsets (Supplementary Figure3, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1).

One hundred and twenty-six genes were differentially expressed in the high-genistein vs low-genistein subsets and defined a high-genistein expression signature (P < .01, FC ≥2; 47 overexpressed and 79 underexpressed) (Supplementary Table 5, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1). Hierarchical clustering of the DE genes was performed to investigate patterns of relative expression among tumor subsets (Figure 4). Tumors clustered in the sample dendogram according to plasma genistein levels as expected, reflecting the selection of DE genes by genistein subset.

Figure 4.

Hierarchical clustering of DE genes in high versus low- genistein subsets. Clustering of samples and 126 DE genes between high- and low-genistein samples (fold change > 2 and t-test P < .1) as identified by microarrays. Clustering was performed using Euclidian distance. Gene expression values are log2 transformed and standardized. ER status, menopausal status, and genistein plasma concentrations for each sample are indicated. Twelve tumors of the soy group with high (>16ng/mL) plasma genistein defined the high-genistein subset, and 22 tumors of the placebo group with low (<6.8ng/mL) genistein defined the low-genistein subset. FGFR2 was overexpressed in three of 12 tumors of the high-genistein group (arrow, P < .01).

Pathway analysis of the high-genistein signature revealed over-representation of pathways that regulate cell growth and proliferation in tumors of the high-genistein group (P < .001). DAVID analysis of overexpressed genes in the high-genistein group revealed that 18 of 23 categories represented cell cycle functions (Bonferroni P < .001, FDR < 0.01) (Supplementary Table 6, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1).[33] Similarly, IPA revealed that the Top Biological Functions and Network Modules of the 126 DE genes were cellular growth and proliferation, cell cycle, cell death and survival, cell development, and nucleic acid metabolism (P < .001) (Supplementary Figure 2, A and B, and Supplementary Table 7, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1). Results of a "downstream effect analysis," which focuses on a gene's function, were concordant with the network analysis, indicating enrichment of genes that regulate cell proliferation (Supplementary Table 8, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1). Specifically, genes in the top-ranked IPA network that were overexpressed in the high-genistein tumors included those which coordinately regulate G1/S and G2/M cell cycle processes, such as E2F5, BUB1, CCNB2 (Cyclin B2), MYBL2, CDK1, and CDC20 (Supplementary Figure 2B and Supplementary Table 8, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1). The receptor tyrosine kinase FGFR2, a known regulator of the cell cycle, was found to be overexpressed in the downstream effect analysis (Supplementary Table 8, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1).

Gene Set Analysis performed on fold changes of all genes between high- and low-genistein groups revealed a higher level of expression of numerous cell cycle gene sets, including RB1 cell cycle targets and E2F-family target genes (Supplementary Table 9, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1). Similar results were obtained when ER(-) samples were excluded from the analysis (Supplementary Table 10, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1).

To assess whether increased expression of cell cycle–related genes in the soy group was associated with BC molecular subtypes, we evaluated the distribution of the PAM50 subtypes in soy and placebo groups.[34,35] We implemented a nearest centroid molecular classification model based on the expression of the PAM50 genes to predict a breast cancer's intrinsic subtype as luminal A, luminal B, HER2-enriched, or Basal.[35] Although all intrinsic subtypes were represented in the high and low-genistein groups, there was a trend for luminal A in the low-genistein group and luminal B in the high-genistein group (P = .06) (Supplementary Figure 3, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1). These data demonstrate enrichment for luminal A BCs in the placebo group and for luminal B tumors in the soy group. Although this may constitute a selection bias despite randomization, we cannot exclude the possibility that soy might increase expression of genes associated with luminal B, including proliferation and cell cycle–related genes.

Expression of the protumorigenic growth factor receptor FGFR2 was elevated in the high- compared with the low-genistein group (FC = 2.4, P = .006) (Figure 5A; Supplementary Table 5, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1), with much higher expression in three out of 12 tumors with very high genistein (46.9, 84.5, 155ng/ml). To confirm overexpression of FGFR2, we performed quantitative real-time PCR (qPCR) in 27 tumors of the soy and 19 of the placebo group, revealing its overexpression by 2.3-fold in tumors of the soy vs placebo group (P = .03) (Figure 5B). The three cases with FGFR2 overexpression by microarray also demonstrated an increase by qPCR. Two of the three samples with FGFR2 overexpression were included in the NanoString paired analysis; one tumor demonstrated a three-fold, and the second a 7.7-fold increase in FGFR2 following soy treatment (Figure 5C). Taken together, these data raise the concern that FGFR2 expression was increased by soy in a subset of BCs.

Figure 5.

Differential expression of FGFR2. FGFR2 expression was increased 2.4 fold in the high-genistein subset of the soy group by microarray (P = .006) (A). FGFR2 overexpression by qPCR in soy vs placebo (P = .03) (B). Nanostring analysis of pre/post samples. Although there were no statistically significant differences for the group as a whole, in two samples from the soy group there was a marked increase in FGFR2 expression (C). *Indicates statistically significant overexpression (P < .05).

Soy Effects on Tumor Proliferation (Ki67) and Apoptosis (Cas3)

Markers of apoptosis (Cas3) and proliferation (Ki67) were examined in paired pre- and posttreatment tumor samples from 54 patients from the soy and 50 from the placebo group, and the percentage of positive staining tumor cells was assessed (Table 3). A comparison of changes (between pre- and posttreatment) in the placebo and soy groups showed no statistically significant differences (P = .2 for Ki67 and P = .3 for Cas3) (Table 3).

NanoString Analysis of Gene Expression Before and After Soy or Placebo

We measured expression of 202 BC-related genes by NanoString analysis in matched tumor samples obtained before and after intervention from 14 BCs. The availability of pretreatment core biopsy tissue limited the sample size to eight patients in the soy, and six in the placebo group. There were no statistically significant between-group differences in patient or tumor characteristics, including ER status (Supplementary Table 2, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1). We identified genes that were changed postintervention, and compared the magnitude and direction of gene expression changes between the two groups (Table 2). Fourteen genes changed in the soy group: 10 increased, and four decreased expression. In the placebo group, 10 genes changed, five increased, and five decreased. Three of these 10 genes were among those that changed in the soy group in the same direction. Thus a total of 21 genes in both groups demonstrated changes. The expression of these genes in pre- and posttreatment tumor samples from both groups is represented in Supplementary Figure 1 (available online) http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1.

To determine gene expression changes, we focused on fold change (posttreatment/pretreatment ratio) for each of the 21 genes, and compared this value between treatment groups. Expression of FANCC and UGT2A1 increased in 87.5% of tumors following soy intake (mean FC = 1.27 and 1.57, P < .05), and decreased (mean FC = -1.26 and -1.33, P value not statistically significant) in the placebo group (Figure 3, A-D). While these fold changes were modest, they were consistent and in opposing direction in the two groups (P < .05). Genes with altered expression in the soy group also included SERPINE1 (mean FC = 2.7, P = .006) (Figure 3E). However, this increase was not significant between soy and placebo groups (P = .26) (Figure 3F).

To evaluate the patterns of gene expression changes in the matched tumors, we performed hierarchical clustering of the paired samples using the pre/postexpression fold change of the 21 differentially expressed genes (Figure 3G). Clustering showed a tendency to organize samples by soy or placebo group, and the heat map showed groups of genes correlated by expression fold change in response to soy or placebo. Genes related to cell cycle functions, including CCNA2, CCNE2, and CDKN1B, were closely grouped by cluster analysis, and demonstrated a pattern of expression with increases in soy and decreases in placebo group samples. These data suggest an effect of soy intake characterized by subtle, yet consistent alterations in BC gene expression.

Genome-wide Expression Analysis in Posttreatment Specimens

We performed genome-wide expression analysis of 51 specimens (39 ER+ and 12 ER-) from surgically resected tumors, 28 from soy group and 23 from placebo. There were no statistically significant differences between group demographics or clinicopathological criteria (Supplementary Table 3 and Table 4, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1). We identified 131 differently expressed (DE) genes between the two groups (absolute fold change ≥2, P < .01) (Supplementary Table 5, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1). Of these, 11 were overexpressed, and 120 were underexpressed in tumors of the soy relative to the placebo group.

We next considered the possibility that genistein plasma levels, rather than assignment to the soy group per se, may be a more relevant marker of genistein effects on BC genes expression. Therefore, we examined differential gene expression as a function of plasma genistein. Median genistein concentration in the soy group was 6.3ng/mL, and 25% demonstrated very low levels (<0.5ng/mL). We therefore limited tumors of the soy group to those from patients with serum genistein greater than 16ng/mL, which corresponded to the 95th percentile concentration of the placebo group. The resulting analysis consisted of posttreatment expression profiles of 11 tumors of soy group with elevated plasma genistein, and 23 tumors from the placebo group with low plasma genistein, referred to as high- and low-genistein subsets, respectively. Tumor characteristics including ER status were similar in high- and low-genistein subsets.

One hundred and twenty-six genes were differentially expressed in the high-genistein vs low-genistein subsets and defined a high-genistein expression signature (FC ≥2, P < .01; 47 overexpressed and 79 underexpressed). Hierarchical clustering of the DE genes was performed to investigate patterns of relative expression among tumor subsets (Figure 4). Tumors clustered in the sample dendogram according to plasma genistein levels as expected, reflecting the selection of DE genes by genistein subset.

Pathway analysis of the high-genistein signature revealed overrepresentation of pathways that regulate cell growth and proliferation in tumors of the high-genistein group (P < .001). DAVID analysis of overexpressed genes in the high-genistein group revealed that 18 of 23 categories represented cell cycle functions (Bonferroni FDR < 0.01, P < .001) (Supplementary Table 6, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1).[33] Similarly, IPA revealed that the Top Biological Functions and Network Modules of the 126 DE genes were cellular growth and proliferation, cell cycle, cell death and survival, cell development, and nucleic acid metabolism (P < .001) (Supplementary Figure 2, A and B, and Supplementary Table 7, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1). Results of a "downstream effect analysis," which focuses on a gene's function, were concordant with the network analysis, indicating enrichment of genes that regulate cell proliferation (Supplementary Table 8, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1). Specifically, genes in the top-ranked IPA network that were overexpressed in the high-genistein tumors included those which coordinately regulate G1/S and G2/M cell cycle processes, such as E2F5, BUB1, CCNB2 (Cyclin B2), MYBL2, CDK1, and CDC20 (Supplementary Figure 2B and Supplementary Table 8, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1). The receptor tyrosine kinase FGFR2, a known regulator of the cell cycle, was found to be overexpressed in the downstream effect analysis (Supplementary Table 8, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1).

Gene Set Analysis performed on fold changes of all genes between high- and low-genistein groups revealed higher level of expression of numerous cell cycle gene-sets, including RB1 cell cycle targets and E2F-family target genes (Supplementary Table 9, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1). Similar results were obtained when ER(-) samples were excluded from the analysis (Supplementary Table 10, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1).

To assess whether increased expression of cell cycle–related genes in the soy group was associated with BC molecular subtypes, we evaluated the distribution of the PAM50 subtypes in soy and placebo groups.[34,35] We implemented a nearest centroid molecular classification model based on the expression of the PAM50 genes to predict a breast cancer's intrinsic subtype as luminal A, luminal B, HER2-enriched, or Basal.[35] Although all intrinsic subtypes were represented in the high- and low-genistein groups, there was a trend for luminal A in the low-genistein group and luminal B in the high-genistein group (P = .06) (Supplementary Figure 3, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1). These data demonstrate enrichment for luminal A BCs in the placebo group and for luminal B tumors in the soy group. Although this may constitute a selection bias despite randomization, we cannot exclude the possibility that soy might increase expression of genes associated with luminal B, including proliferation and cell cycle–related genes.

Protumorigenic growth factor receptor FGFR2 expression was elevated in the high- compared with the low-genistein group (FC = 2.4, P = .006) (Figure 5A; Supplementary Table 5, available online http://jnci.oxfordjournals.org/content/106/9/dju189/suppl/DC1), with much higher expression in three out of 12 tumors with very high genistein (46.9, 84.5, 155ng/mL). To confirm overexpression of FGFR2, we performed quantitative real-time PCR (qPCR) in 27 tumors of the soy and 19 of the placebo group, revealing its overexpression by 2.3-fold in tumors of the soy vs placebo group (P = .03) (Figure 5B). The three cases with FGFR2 overexpression by microarray also demonstrated an increase by qPCR. Two of the three samples with FGFR2 overexpression were included in the NanoString paired analysis; one tumor demonstrated a three-fold, and the second a 7.7-fold increase in FGFR2 following soy treatment (Figure 5C). Taken together, these data raise the concern that FGFR2 expression was increased by soy in a subset of BCs.

Soy Effects on Tumor Proliferation (Ki67) and Apoptosis (Cas3)

Markers of apoptosis (Cas3) and proliferation (Ki67) were examined in paired pre- and posttreatment tumor samples from 54 patients from the soy and 50 from the placebo group, and the percentage of positive staining tumor cells was assessed (Table 3). A comparison of changes (between pre- and posttreatment) in the placebo and soy groups showed no statistically significant differences (P = .21 for Ki67 and P = .35 for Cas3) (Table 3).

Comments

3090D553-9492-4563-8681-AD288FA52ACE

processing....