HMGA2/TET1/HOXA9 signaling pathway regulates breast cancer growth and metastasis

Abstract

The ten–eleven translocation (TET) family of methylcytosine dioxygenases initiates demethylation of DNA and is associated with tumorigenesis in many cancers; however, the mechanism is mostly unknown. Here we identify upstream activators and downstream effectors of TET1 in breast cancer using human breast cancer cells and a genetically engineered mouse model. We show that depleting the architectural transcription factor high mobility group AT-hook 2 (HMGA2) induces TET1. TET1 binds and demethylates its own promoter and the promoter of homeobox A (HOXA) genes, enhancing its own expression and stimulating expression of HOXA genes including HOXA7 and HOXA9. Both TET1 and HOXA9 suppress breast tumor growth and metastasis in mouse xenografts. The genes comprising the HMGA2–TET1–HOXA9 pathway are coordinately regulated in breast cancer and together encompass a prognostic signature for patient survival. These results implicate the HMGA2–TET1–HOX signaling pathway in the epigenetic regulation of human breast cancer and highlight the importance of targeting methylation in specific subpopulations as a potential therapeutic strategy.

Epigenetic changes play an important role in cancer progression as well as development (1). Recent studies indicate that DNA demethylation can be catalyzed by a class of methylcytosine dioxygenases termed the ten–eleven translocation (TET) family (2–5). TET1 promotes DNA demethylation by catalyzing conversion of 5-methylcytosine (5mC) primarily to 5-hydroxymethylcytosine (5hmC) as well as 5-formylcytosine or 5-carboxylcytosine (3, 5). The modified cytosines are then removed through active or passive mechanisms (2–6). While TET1 is highly expressed in embryonic stem (ES) cells (5, 7–10), loss of TET1 protein and decreased 5hmC levels have been recently shown in solid tumors relative to normal epithelial cells (2, 11–14). However, the mechanism by which TET1 is suppressed in solid tumors has not been identified. Furthermore, the downstream targets by which TET1 regulates growth and metastasis in cancer are largely unknown.

High mobility group AT-hook 2 (HMGA2), a chromatin-remodeling factor (15), binds to AT-rich regions in DNA, altering chromatin architecture to either promote or inhibit the action of transcriptional enhancers. HMGA2 is highly expressed in ES cells but is generally low or lacking in normal somatic cells. Interestingly, HMGA2 is highly expressed in most malignant epithelial tumors, including breast (16), pancreas (17), oral squamous cell carcinoma (18), and non-small-cell lung cancer (19). HMGA2 overexpression in transgenic mice causes tumor formation, whereas Hmga2-knockout mice have a pygmy phenotype indicative of a growth defect (20). We have reported that HMGA2 promotes tumor invasion and metastasis in breast cancer in part through regulation of prometastatic genes, including Snail, osteopontin, and CXCR4 (21, 22).

To systematically identify critical downstream mediators of HMGA2 that regulate invasion and metastasis, we performed gene expression array analysis by knocking down HMGA2 in breast cancer cells. Here we show that TET1 is an important effecter of HMGA2 in breast cancer. We further show that TET1 regulates homeobox A (HOXA) genes, including HOXA7 and HOXA9. Both TET1 and HOXA9 suppress breast tumor growth and metastasis. Our study reveals a regulatory pathway that stratifies breast patient survival.

Results

TET1 and HOX Gene Expression Are Dramatically Induced upon Depletion of HMGA2 in both Invasive Human Breast Cancer Cells and MMTV–Wnt1 Mouse Breast Tumors.

Gene expression arrays were conducted using two invasive human breast cancer cell lines expressing either HMGA2 shRNA or control scrambled shRNA (SCR sh): 1833, a bone-tropic derivative of the human breast cancer cell line MDA-MB-231 (23), and MDA-MB-436. We found a dramatic induction of HOX genes, particularly at the HOXA loci in the more invasive 1833 cells (Fig. 1 A–C and Fig. S1). The HOX genes comprise four clusters (A, B, C, and D) located on different human chromosomes. This transcriptional factor family with 39 members controls posterior–anterior patterning during embryogenesis and the development of specific organs (reviewed in ref. 24). We validated the induction of HOXA gene expression including HOXA4, HOXA5, HOXA6, HOXA7, HOXA9, and HOXA11 in HMGA2-depleted 1833 cells (Fig. 1 D, F, and G).

Fig. 1.

Induction of TET1 and HOX gene expression upon depletion of HMGA2 in 1833 cells, a bone-tropic derivative of human breast cancer cell line MDA-MB-231, or in MMTV–Wnt1 transgenic mouse breast tumors. (A, B, and D–H) We stably transduced 1833 cells with HMGA2 shRNA (shHMGA2) or control SCR sh. (A) Gene expression array analysis showing up-regulation of TET1 and 20 of 39 HOX genes in HMGA2-depleted cells. (B) The expression levels of HOXA genes are shown. *Fold change (fc) < 2; **fc > 2 based on the signal intensity of gene expression arrays. (C) Genomic transcription units of human HOXA genes on chromosome 7 viewed using the UCSC Genome Browser (40). HOXA genes are transcribed from right to left with the following order: 5′UTR (thin blue bar), Coding Sequence (thick blue bar), and 3′UTR (thin blue bar). Bar length is proportional to length of DNA sequence. (D–H) qRT-PCR and immunoblotting analyses validated induction of TET1 and HOXA gene expression in HMGA2-depleted cells. (D–F) HMGA2 (D), TET1 (E), or HOXA4/5/6/7/9/11 (F) mRNA analyzed by qRT-PCR (GAPDH as normalization control). (G) HMGA2, TET1, and HOXA9/7 protein analyzed by immunoblotting (GAPDH as control). (H) Genome-wide 5hmC levels analyzed by dot blot assay. (I and J) Loss of Hmga2 in MMTV–Wnt1 transgenic mouse breast tumors induced Tet1 and Hoxa9/7 expression. Wnt1 transgenic mice were crossed with Hmga2-specific knockout mice. Mouse primary breast tumors were obtained from Hmga2 wild-type (Hmga2^+/+), heterozygous (Hmga2^+/−), or null (Hmga2^−/−) mice. (I) Murine Hmga2, Tet1, and Hoxa9/7 mRNA analyzed by qRT-PCR (with mouse Gapdh as normalization control). (J) Murine Tet1 and Hoxa9 protein and 5hmC levels analyzed by immunostaining. (D–F, H, and I) Data are means ± SEM; n = 3. *P < 0.05; **P < 0.01; ***P < 0.001.

Among the other genes induced by HMGA2 depletion was TET1 (Fig. 1A and Fig. S1). We confirmed that HMGA2 is a negative regulator of TET1 in 1833 cells by quantitative RT-PCR (qRT-PCR) and immunoblotting (Fig. 1 E and G). Consistent with increased TET1 expression, we observed elevated 5hmC levels in HMGA2-depleted 1833 cells (Fig. 1H). We also observed similar induction of TET1 and HOXA gene expression following HMGA2 depletion by shRNA in MDA-MB-436 cells, although the effects were not as robust, consistent with their relatively higher basal levels of TET1 and HOXA protein and the less invasive phenotype of these cells (Fig. S2 A–D). Consistent with these observations, analysis of gene expression in a cohort of 75 human breast tumors (25) showed a significantly negative correlation between HMGA2 and TET1 expression (Fig. S2E), and this relationship existed in both estrogen receptor (ER)-negative (Fig. S2F) and ER-positive (Fig. S2G) patient subpopulations.

To validate the regulation of TET1 and HOXA genes by HMGA2 in vivo, we used an MMTV–Wnt1 transgenic mouse model for breast cancer. Deletion of Hmga2 by crossing MMTV–Wnt1 mice with Hmga2-specific knockout mice reduced tumor incidence (26) and decreased tumor cell proliferation as assessed by immunostaining of antigen Ki67 (Fig. S3). Analysis by qRT-PCR or immunohistochemistry showed a strong induction of Tet1, 5hmC, and Hoxa gene expression including Hoxa9 and Hoxa7 in tumors from the MMTV–Wnt1/Hmga2^−/− mice (Fig. 1 I and J). Moreover, expression of HMGA2 in both 1833 and MDA-MB-436 cells inhibited TET1 expression (Fig. S4). These results indicate that induction of TET1 by depletion of HMGA2 is not an off-target effect and raise the possibility that loss of HMGA2 suppresses breast tumor growth by inducing TET1 and HOXA genes.

TET1 Is Involved in an Autoregulation in Human Breast Cancer Cells.

Because the Tet1 protein may bind directly to its own promoter region as suggested by the ChIP-sequencing (ChIP-seq) data for Tet1 in mouse ES cells (9, 10, 27, 28), we investigated whether TET1 was also involved in regulating its own expression in human breast cancer cells. Conventional ChIP assays using 1833 cells expressing TET1 or vector control showed that TET1 bound to its own promoter region (Fig. 2A). Consistent with increased expression of TET1, ChIP assays also showed that 1833 cells expressing TET1 exhibited increased binding of H3K4Me3, a histone marker for transcriptional activation, to the TET1 promoter region (Fig. 2A). Furthermore, because TET proteins are typically involved in DNA demethylation pathways (2–5, 29, 30), we analyzed the effect of TET1 on the methylation status of its own promoter regions. Our DNA methylation-specific digestion combined with qPCR showed that ∼70% of the TET1 promoter region within ±1 kb from the transcription start site (TSS) in the parent 1833 cells contained methylated CpG islands, whereas the fraction decreased to 9% following HMGA2 depletion (Fig. 2B). Similarly, bisulfite sequencing within the same region of the TET1 promoter showed an increase in demethylated CpGs from 38% in 1833 cells to 86% in HMGA2-depleted cells (Fig. 2C). The 1833 cells treated with 5-azacytidine, a demethylation reagent that inhibits DNMTs, also showed an increase in TET1 expression (Fig. 2D). Together, our results suggest that HMGA2 depletion in 1833 cells causes extensive demethylation of the TET1 promoter and therefore results in a robust induction of TET1 expression.

Fig. 2.

TET1 is involved in an autoregulation in human breast cancer cells. (A) TET1 binds to its own promoter. We analyzed 1833 cells expressing TET1 or control by ChIP assay with anti-TET1 or H3K4Me3 antibody followed by qPCR analysis: TET1 and H3K4Me3 binding to the CpG island proximal to the TSS of TET1 (see site-1 and -2 in SI Materials and Methods). Site-3 is a negative control. (B and C) HMGA2 depletion causes demethylation of CpG islands at the TET1 promoter region. We analyzed 1833 cells stably expressing shHMGA2 or SCR sh for CpG island methylation status by multiple approaches. (B) TET1 promoter region was analyzed within ±1 kb from the TSS. Methylation-specific digestions followed by qPCR distinguished between methylated CpGs vs. unmethylated or other modified (e.g., 5hmC) CpGs. The percentage of methylation vs. unmethylation (includes unmethylated or other modified C) is indicated. (C) Bisulfite sequencing of specific CpGs (see SI Materials and Methods for primers) at the TET1 promoter proximal to the TSS. Results show unmethylated CpGs (open circles) vs. methylated or modified CpGs (filled circles) in 10 or more independent clones encompassing the region of interest. (D) We treated 1833 cells with 5-azacytidine followed by qRT-PCR analysis for TET1 mRNA expression (GAPDH as normalization control). (A, B, and D) Data are means ± SEM; n = 3. *P < 0.05; **P < 0.01.

TET1 Directly Induces HOXA Gene Expression in Breast Cancer Cells Through Binding to the Promoter Regions of HOXA Genes and Contributing to Local Demethylation.

Previous ChIP-seq analyses of Tet1 in mouse ES cells also implied that Hoxa genes might be downstream targets of Tet1 because their promoter regions are enriched with Tet1 protein binding (9, 10, 27, 28). To investigate whether TET1 is an upstream regulator of the HOXA cluster in human breast cancer cells, we transfected HMGA2-depleted 1833 cells with TET1 siRNA and observed a significant decrease in HOXA gene expression and 5hmC levels along with decreased TET1 expression (Fig. 3 A and B). Conversely, ectopic expression of TET1 in the parent 1833 cells dramatically increased HOXA9 expression and 5hmC levels (Fig. 3 C and D).

Fig. 3.

TET1 induces HOXA gene expression. (A and B) Depletion of TET1 by siRNA partially countered induction of HOXA genes. We transfected 1833 cells stably expressing HMGA2 shRNA with control or TET1 siRNA. (A) Analysis of TET1 and HOXA gene mRNA by qRT-PCR. (B, Upper) Analysis of TET1 and HOXA9/7 protein by immunoblotting. (Lower) Analysis of 5hmC levels by dot-blot assay. (C and D) Expression of TET1 dramatically induced HOXA9 expression. We analyzed 1833 cells expressing constitutive Tet1 (Flag–Tet1) by qRT-PCR for HOXA9 mRNA (C) and by immunoblotting for Tet1 (Flag–M1) and HOXA9 protein (D, Upper) and dot-blot assay for 5hmC levels (D, Lower). (E and F) Induced expression of TET1 in breast xenograft tumors significantly induced HOXA9 expression. The 1833 cells stably expressing an inducible Tet1 expression vector were orthotopically injected into the mammary fat pad of nude mice. Tumor tissues were collected and analyzed after 6 wk with (+DOX) or without (−DOX) doxycycline treatment. (E) Tet1 and HOXA9 mRNA analyzed by qRT-PCR. (F) Tet1 and HOXA9 protein and 5hmC levels analyzed by immunostaining. (G) Significant positive correlation between TET1 and HOXA9/7 expression in breast cancer patients (see Table S3 for patient information). Correlations were determined by Pearson’s correlation coefficient. P value was determined by Student t test. (A–E) GAPDH served as normalization control. Data are means ± SEM; n = 3. **P < 0.01; ***P < 0.001.

To test whether TET1 can regulate HOXA gene expression in vivo, we stably transduced 1833 cells with an inducible Tet1 expression vector. Cells were orthotopically injected into the mammary fat pad of mice. qRT-PCR and immunohistochemistry analysis of the mouse xenografts after 6 wk of treatment with doxycycline showed a dramatic induction of Tet1 and HOXA9 expression (Fig. 3 E and F), as well as increased 5hmC levels (Fig. 3F). Gene expression analysis in a cohort of 54 human breast tumors (25) also showed a strong positive correlation between TET1 and HOXA gene expression (Fig. 3G).

We next investigated whether the induction of HOXA gene expression could also be attributed to direct binding of TET1 and subsequent demethylation of the HOXA promoter regions. Our ChIP-qPCR assays confirmed that TET1 and H3K4Me3 also bound to the HOXA gene promoter regions (Fig. 4 A and B). Our DNA methylation-specific digestion combined with qPCR assay showed that, in the parent 1833 cells, only a small fraction of the promoter region of the HOXA4–A11 genes lacked methylation (Fig. 4C), accounting for the low expression of HOXA transcripts; by contrast, HMGA2 depletion caused dramatic loss of methylation at the HOXA gene loci (Fig. 4C). Bisulfite sequencing of the 14 CpGs near the HOXA7 TSS showed that only 4% were unmethylated, whereas HMGA2 depletion caused >80% demethylation (Fig. 4D). Similarly, bisulfite sequencing of the 15 CpGs in the upstream 1-kb locus of the HOXA9 promoter showed that only 5% of the CpG sites were unmethylated, whereas unmethylated CpGs increased to 91% upon loss of HMGA2 in 1833 cells (Fig. 4E). The 1833 cells treated with 5-azacytidine to demethylate DNA also showed a dramatic increase in HOXA7 and HOXA9 expression (Fig. 4 F and G). These results suggest that TET1 binds directly to the HOXA promoter regions and contributes to local demethylation, inducing activating histone binding and gene transcription in breast cancer cells.

Fig. 4.

TET1 induces HOXA gene expression through binding to the promoter regions of HOXA genes and contributing to local demethylation in human breast cancer cells. (A and B) TET1 binds to the HOXA gene promoters. We analyzed 1833 cells expressing TET1 or control by ChIP assay with anti-TET1 or H3K4Me3 antibody followed by qPCR analysis: TET1 and H3K4Me3 binding to the CpG islands proximal to the TSS of HOXA7 (see site-1 and -2 in SI Materials and Methods) where site-3 is a negative control (A); or HOXA9 (see site-1 in SI Materials and Methods) where site-2 is a negative control (B). (C–E) HMGA2 depletion causes demethylation of CpG islands at HOXA gene promoter regions. We analyzed 1833 cells stably expressing shHMGA2 or SCR sh for CpG island methylation status by multiple approaches (see Fig. 2 B and C for the specificity of each method). (C) HOXA promoter regions were analyzed within −5 to +3 kb from the TSS. The percentage of methylation vs. unmethylation is indicated. (D and E) Bisulfite sequencing of specific CpGs (see SI Materials and Methods for primers) at HOXA7 (D) and HOXA9 (E) promoters proximal to the TSS. Results show unmethylated CpGs (open circles) vs. methylated or modified CpGs (filled circles) in 10 independent clones encompassing the region of interest. (F and G) We treated 1833 cells with 5-azacytidine followed by qRT-PCR analysis for expression of HOXA7 (F) or HOXA9 (G) mRNA (GAPDH as normalization control). (A–C, F, and G) Data are means ± SEM; n = 3. **P < 0.01; ***P < 0.001.

HMGA2/TET1/HOXA Pathway Regulates Breast Cancer Cell Invasion.

To assess the pathological significance of this HMGA2/TET1/HOXA signaling cascade, we determined the effect of manipulating these genes on cell invasion. HMGA2 depletion in 1833 cells decreased cell invasion (Fig. 5A); this effect was reversed in part by siRNA depletion of TET1, HOXA9, or HOXA7 (Fig. 5 B and C and Fig. S5 I–K). The 1833 cells treated by demethylation reagent 5-azacytidine or decitabine showed a similar decreased cell invasion and a partial rescue in invasion followed by siRNA depletion of HOXA9 (Fig. 5D and Fig. S5 A–C and E–G). These data are consistent with a previous study showing that HOXA9 is a breast cancer inhibitor (31). Together, our results reveal a signaling cascade whereby HMGA2 promotes breast cancer cell invasion in part through inhibition of TET1-mediated demethylation and HOXA gene expression.

Fig. 5.

Both TET1 and its target, HOXA9, suppress breast tumor growth, invasion, and metastasis. (A–D) HMGA2/TET1/HOXA pathway regulates breast cancer cell invasion. (A) Inhibition of cell invasion in 1833 cells with HMGA2 depletion. (B) Transfection of TET1 siRNA into HMGA2-depleted 1833 cells increases invasion. (C) Transfection of HOXA7 or HOXA9 siRNA into HMGA2-depleted 1833 cells increases invasion. (D) Decitabine (5-aza-dC) treatment of 1833 cells decreases cell invasion, and transfection of HOXA9 siRNA into treated cells partially reversed cell invasion. (A–D) Data are means ± SEM; n = 3. (E–K) We orthotopically injected 1833 cells stably expressing an inducible control, Tet1, or HOXA9 expression vector into the mammary fat pad of nude mice. Mice were administered drinking water with (+DOX) or without (−DOX) addition of doxycycline. (E–G) Both TET1 and HOXA9 suppress xenograft breast tumor growth. (E) Representative bioluminescence images of mice bearing 1833 cells treated as indicated. (F) Photograph of representative xenograft breast tumors of 1833 cells treated as indicated. (G) Xenograft breast tumors of 1833 cells treated as indicated and analyzed for tumor weight. (F and G) Tumors were dissected at 6 wk after implantation. (H and I) Both TET1 and HOXA9 suppress the proliferation in xenograft breast tumors: immunostaining showing Ki67-positive cells in tumor sample of 1833 cells with induced (+DOX) vs. noninduced (−DOX) expression of Tet1 (H) or HOXA9 (I). (J and K) Both TET1 and HOXA9 inhibit intravasation of 1833 cells. Cells isolated from the blood after 6 wk were analyzed for GAPDH/Gapdh transcripts derived from human (tumor) or mouse (control) by qRT-PCR: intravasation of 1833 cells with induced (+DOX) vs. noninduced (−DOX) expression of Tet1 (J) or HOXA9 (K). Data are means ± SEM; n = 8 per group. (L–N) Both TET1 and HOXA9 suppress bone metastasis of 1833 cells. We injected 1833 cells stably expressing an inducible Tet1 or HOXA9 expression vector into the left ventricle of mice. Mice were administered drinking water with (+DOX) or without (−DOX) addition of doxycycline and imaged for luciferase activity after 3 wk. (L) Representative bioluminescence images of mice with bone metastasis. (M) Quantification of bone colonization by 1833 cells with induced (+DOX) vs. noninduced (−DOX) expression of Tet1 or HOXA9. Data are means ± SEM; n = 7–9 per group. (N) Kaplan–Meier survival analysis of mice over 8 wk after injection of the tumor cells.

Both TET1 and Its Target, HOXA9, Suppress Breast Tumor Growth, Intravasation, and Metastasis.

To determine whether TET1 or HOXA9 can reverse the tumorigenic phenotype in breast cancer cells transformed by HMGA2, we injected 1833 cells expressing inducible Tet1 or HOXA9 into the mammary fat pad of mice followed by doxycycline treatment and tested their effect on xenograft tumor growth. Consistent with our in vitro observation (Figs. S5D and S6 A–C), induced expression of Tet1 (Fig. 3 E and F) or HOXA9 (Fig. S6 D and E) significantly suppressed xenograft tumor growth (Fig. 5 E–G) and tumor cell proliferation (Fig. 5 H and I).

To test TET1 or HOXA9 regulation of invasion in vivo, we determined their effect on tumor cell intravasation from a primary site in a murine orthotopic model. The 1833 cells expressing inducible Tet1 or HOXA9 were injected into the mammary fat pad of mice. After 6 wk of treatment with doxycycline, cells isolated from the blood were lysed and analyzed for human (tumor) or mouse (control) GAPDH transcripts. qRT-PCR analysis showed that both TET1 and HOXA9 significantly inhibited tumor cell intravasation (Fig. 5 J and K).

Because HMGA2 depletion suppresses breast tumor cell bone metastasis (21, 22), we determined whether its downstream effecters TET1 and HOXA9 similarly inhibit tumor metastasis. Luciferase-labeled 1833 cells expressing inducible Tet1 or HOXA9 were injected into the left ventricle of mice that were subsequently treated with doxycycline. After 3 wk, mice were imaged for luciferase activity. TET1 or HOXA9 expression caused a dramatic decrease in bone metastasis (Fig. 5 L and M) and a significant increase in overall survival rate (Fig. 5N and Fig. S6 F–H).

HMGA2/TET1/HOXA9 Regulate a Common Set of Important Genes and Encompass a Prognostic Signature for Patient Survival.

To identify and compare target genes of HMGA2, TET1, and HOXA9, we performed additional microarray assays for cells expressing induced TET1 and HOXA9. Compared with the parental 1833 cells, there were 1,012; 7,220; and 7,132 genes differentially expressed [P < 0.05; false discovery rate (FDR) < 0.05; and fold change > 1.25) upon HMGA2 depletion, TET1 induction, or HOXA9 induction, respectively (Fig. 6A). Interestingly, >60% of the genes differentially regulated by TET1 or HOXA9 were the same (4,510 genes; Fig. 6A), indicating that HOXA9 is a major downstream effecter of TET1. There were 214 genes that overlapped among all three sets, including 144 up-regulated and 70 down-regulated genes (Fig. 6A and Table S1). Gene annotation enrichment analysis (Database for Annotation, Visualization and Integrated Discovery) (32) indicated that the 144-gene up-regulated set was enriched in genes that have functions such as binding, catalytic activity, transcription regulator activity, and developmental processes, whereas the 70-gene down-regulated set was enriched in genes related to epithelial cell proliferation (P = 0.041) and the extracellular matrix (P = 0.0077) (Table S2). Gene set enrichment analysis (33) indicated that the down-regulated set was also enriched in genes that promote tumor growth and comprise metastasis signatures, such as CCL2, EFEMP1, IL7R, PPAP2B, and STX3 (34). This pattern of gene regulation is consistent with a role for TET1 through its effecter HOXA9 in the suppression of breast tumor growth and metastasis.

Fig. 6.

The HMGA2/TET1/HOXA pathway regulates breast tumorigenesis. (A) Comparison of the genes regulated by HMGA2, TET1, or HOXA9 in 1833 cells (human breast cancer cells; hBrCa). (B) Scheme illustrating HMGA2/TET1/HOXA signaling pathway in breast tumorigenesis. (C) Kaplan–Meier analysis of gene expression data from 101 breast tumor patients (see Table S3 for patient information). Patients were stratified for survival using HMGA2, TET1, HOXA9, HOXA7, or the complete pathway as indicated. (Right) Red line, high HMGA2 and low TET1/HOXAs (n = 34); blue line, low HMGA2 and high TET1/HOXAs (n = 35); P, χ² P value.

These data illustrate a signaling cascade in human breast cancer progression, by which expression of the oncogene HMGA2 leads to TET1 suppression. Because TET1 binds and demethylates itself as well as HOXA genes, including HOXA7 and HOXA9, decreased TET1 causes further inhibition of TET1 in addition to loss of HOXA gene expression. Suppression of TET1 and HOXA9 then enables expression of genes that promote breast tumor growth and metastasis (Fig. 6B). When considered individually, gene expression of HMGA2, TET1, HOXA7, or HOXA9 does not significantly predict survival in a heterogeneous group of breast cancer patients (Fig. 6 C, Left and I, Left). By contrast, Kaplan–Meier analysis using the complete HMGA2–TET1–HOXA pathway (HMGA2 high and TET1/HOXA9/7 low vs. HMGA2 low and TET1/HOXA9/7 high) or a combination of HMGA2 and HOXA genes was able to stratify patients and predict survival (Fig. 6 C, Right and I, Right). There are no significant differences in the composition of cancer subtypes between the two stratified groups of patients (Table S3), suggesting that this regulatory mechanism exists in a variety of breast cancer subtypes.

Discussion

We identify here not only an important upstream regulator (HMGA2) of TET1, but also a previously undescribed downstream regulatory pathway for TET1 involving HOXA genes; in addition, we provide compelling evidence that TET1 and HOXA9 play an important role not only in breast tumor invasion and growth, but also in metastasis through commonly regulated genes. Because HMGA2 is a genomic architectural factor, we do not exclude the possibility that HMGA2 might be able to regulate TET1 expression by direct binding to the TET1 promoter or by alteration of its chromatin structure. Nonetheless, a recent investigation of HMGA2 binding sites in HCT116 cells that highly express HMGA2 protein showed only 49 binding sites (35). None of them are even close to the genomic locus of TET1 gene. Thus, it is very likely that HMGA2 regulates TET1 expression indirectly via other mediators.

While this work was under completion, it was reported that TET1 inhibits growth and metastasis in prostate and breast cancer (36). In that report, TET1 was shown to inhibit invasion in culture in part through tissue inhibitors of metalloproteinases (TIMPs) (36). By contrast, we did not observe significant induction of TIMP expression by TET1. Instead, we identify a group of genes commonly altered by HMGA2 depletion or induction of either TET1 or HOXA9, including a subset of induced genes that promote development and a subset of inhibited genes that promote cell proliferation, consistent with a role for TET1/HOXA9 in suppression of breast tumor growth and metastasis.

The TET1/HOXA9 signaling pathway that we identify here also highlights the importance of cell context in determining the pathological function of TET1. In contrast to our results for breast cancer, the mixed-lineage leukemia–TET1 fusion protein and the HOXA9 protein both promote leukemogenesis (37, 38). Recently, HOX family members were reported to play key roles in regulating tumorigenesis, including the epithelial/mesenchymal transition, invasion, and apoptosis (24). Highly methylated HOXA gene loci have been reported in human breast cancer (39), although mutations in these genes are not common. Whether these genes function in similar ways or promote different phenotypes is an interesting question that requires further investigation.

Finally, this report identifies a gene signature comprising three mechanistically linked genes (HMGA2, TET1, and HOXA9) that is prognostic for breast cancer survival. This signature has the potential to identify patients harboring breast or other tumors with suppressed TET1/HOXA9 signaling who might benefit from DNA demethylation agents currently used in the clinic.

Materials and Methods

HMGA2-depleted cell lines were generated by lentiviral transduction with HMGA2 shRNA. TET1 and HOXA gene knockdown cell lines were generated by transfection with relative siRNA. Tet1- and HOXA9-inducible expression cells were generated by lentiviral transduction with relative inducible expression vector. Detailed descriptions are in SI Materials and Methods.