Exogenous Thyroid Hormone Is Associated with Shortened Survival and Upregulation of High-Risk Gene Expression Profiles in Steroid Receptor–Positive Breast Cancers

The comorbidity of thyroid disease and breast cancer has been recognized for over a century (1–3). Women with hyperthyroidism have an 11% increase in breast cancer risk, whereas those with hypothyroidism have a reduced risk compared with controls (2). Even in cases of benign thyroid disease, elevated serum thyroid-stimulating hormone (TSH) or triiodothyronine (T3) independently or in combination with estrogen, have been associated with a higher rate of breast cancer and more rapid cancer progression in some studies (4–6). A meta-analysis of multinational trials has demonstrated that breast cancer is the most common secondary malignancy in female patients with thyroid cancer (6, 7), and patients with breast cancer have a significantly higher risk of thyroid cancer; thus, the relationship appears to be hormonal in nature and bidirectional (7–9). Preclinical models and in vitro studies have shown that thyroid hormone (TH) promotes breast epithelial cell growth, transformation, and carcinogenesis (1, 10, 11). Despite these observations, the impact of TH replacement therapy (THRT) on outcomes by breast cancer subtype, disease stage, and hormonal treatment have been largely unexplored. Controlled clinical trials of breast cancer development or outcomes for women on THRT, as compared with no THRT, have not been performed.

Over two-thirds of newly diagnosed invasive breast cancers express steroid receptors [SR⁺: estrogen receptors (ER) and/or progesterone receptor (PR)]. The majority of ER⁺ cancers express the isoform ERα. Benign breast epithelial cells and breast cancers also widely express TH receptors (THR; ref. 12), which have been less well studied in breast carcinogenesis (12–15). In ER⁺ breast cancer cells, both genomic and nongenomic mechanisms of ER:THR cross-talk have been reported previously (10, 16–20). Evidence suggests that ligands and receptors in both pathways may promote cancer growth in receptor-bearing cells in vitro and in vivo.

In this study, we interrogated whether TH enhanced estrogen-induced activation of cellular proliferation and pro-oncogenic pathways. We show that TH ± E2 induces nuclear colocalization of ERα and THRα, activates ER target and cell cycle–associated genes, and that THRα and ERα are critical drivers of these interactions. In addition, our clinical and experimental data show that in the presence of TH, tamoxifen (Tam) is converted from a partial ER antagonist to an agonist in breast cancer cells and facilitates disease progression in patients with SR⁺ breast cancer. Our data suggest that TH can drive estrogenic signaling and impact outcomes in patients with SR⁺ breast cancer, even with Tam or aromatase inhibitor (AI) cotreatment.

Human breast cancer cell lines MCF7, T47D, MDA-MB-468, and SKBR3 cells were purchased from ATCC. Cell lines were authenticated by Short Tandem Repeat DNA Profiling (Promega) at the University of Colorado Cancer Center (UCCC) Tissue Culture Core, and tested for Mycoplasma every 3 months. Only cells under five passages were used in this study. PT12 cells (also known as UCD12 cells) were derived from UCD12 patient-derived xenograft (PDX) model (21) and characterized as described in ref. 22. Cell culture conditions are provided in Supplementary methods and described previously (23, 24).

Total RNA was isolated from breast cancer cell lines or animal tumor tissue using the RNeasy Mini Kit (Qiagen). RNA (2,000 ng) was reverse transcribed using the qScript cDNA Synthesis Kit (QuantaBio), and the resulting cDNA used for real-time PCR amplification with Fast SYBR green PCR master mix on a 7500 HT real-time PCR system (Applied Biosystems). GAPDH was used as an endogenous normalization control. Primers are described in Supplementary Methods. Relative levels of gene changes were determined using the ΔΔ/CT analysis protocol.

Xenograft experiments were approved by the University of Colorado Institutional Animal Care and Use Committee (IACUC) and conducted in accordance with NIH guidelines of care and use of laboratory animals. Female NOD-SCID-gamma (NSG) mice purchased from the Jackson Laboratory, grown to 4 to 6 weeks old, implanted with a S.C. E2 pellet [a silastic implant containing cellulose and either 0.25 mg or 1 mg of E2 (MilliporeSigma)], and injected with UCD12 PDX per protocol (23). Mice were treated with l-thyroxine (T1775, Sigma-Aldrich, 5 mg/kg in water), E2 or Tam (T5648, Sigma-Aldrich, 20 mg/mL in peanut oil) two times a week administered via intraperitoneally or fulvestrant (ICI, 5 mg per mouse administered via i.p.) weekly. For additional data, see Supplementary Methods.

Tissues were fixed in 10% neutral-buffered formalin and processed according to standard histologic protocols. For the Massachusetts General Hospital (MGH) cases, IHC was performed on paraffin-embedded tissue sections using antibodies against ER (Clone 1D5) and PR (A0098) purchased from Dako. HER2, Ki-67, and EGFR methods were published previously (25, 26). Additional methods in Supplementary Methods.

In the first patient cohort, patients with lymph node–negative (LN⁻) breast cancer (n = 820) diagnosed between 1962 and 1993 (median follow-up of 10.2 years) from the MGH (Boston, MA) were identified from the institutional cancer registry in real time and cross-checked for diagnosis and stage using pathology records. A corresponding linked database was constructed with Institutional Review Board (IRB) approval, including clinical history, hormonal therapy including THRT use, surgical, radiologic, and drug treatment, pathology data, recurrence sites and dates, and outcomes. A REMARK (Reporting Recommendations for Tumor Marker Prognostic Studies) flow diagram of breast cancer samples and analyses are shown in Fig. 1A (27).

An additional study group of single institution, consecutively diagnosed postmenopausal stage I breast cancer cases diagnosed between 2006 and 2009 (n = 160) from the University of Colorado Hospital (UCH), Aurora, CO, were identified and collected as study group 2, to evaluate interactions between THRT and AI use. Patient data and outcomes (median follow-up 8.8 years) derived from paper, chart, and electronic records were reviewed, with IRB approval. Clinical and histopathologic data, including treatment with an AI, THRT, or chemotherapy, time to failure and outcomes as described for MGH patients were entered into a separate database.

Studies of MGH patients, as well as the generation and analysis of the corresponding databases were approved by the MGH IRB (1999P-007288). In 2006, all patient identifiers were removed from this patient dataset at University of Colorado Multiple Institutional Review Board (UCMIRB) 11-1816. The inclusion and study of UCH patients was approved by the UCMIRB (00-1094). Experimental animal protocols were also reviewed and approved by the University of Colorado IACUC (#00174).

For some analyses, statistical significance was evaluated using a two-tailed Student t test or ANOVA with Bonferroni or Dunnett multiple comparisons test or nonparametric equivalents in GraphPad Prism (Ver 6, GraphPad Software) or SAS (version 9.4, SAS Institute). Analysis of clinical studies has been published previously (25, 26) or provided in Supplementary Methods.

For details of breast cancer datasets, cell culture, Western blot analysis, qRT-PCR, cell proliferation, cell-cycle analysis, and immunofluorescence assays. See Supplementary Methods.

In our first observational trial, we examined 820 patients with lymph node–negative breast cancer, diagnosed between 1962 and 1993 at MGH, followed for up to 36 years (median of 10.2 years). A REMARK flow diagram of the patients and the observational study is shown Fig. 1A (27). The majority of patients were treated with only surgery (usually a modified radical mastectomy, as was the practice at that time) and were systemically treatment naïve, with Tam given to 111/820 (13.5%) and chemotherapy administered to 64/820 (7.8%). For all patients, only the primary breast cancer proliferative rate (Ki-67) was significantly higher in THRT treated as compared with non-THRT patients (P = 0.04; Supplementary Table S2A). Univariate analysis of all 820 patients showed the following variables were significantly associated with shortened disease-free survival (DFS), including: younger patient age (P = 0.006), high histologic grade (P = 0.01), larger tumor size (P = 0.0003), high Ki-67 (P = 0.004), ER⁻ (P = 0.0008), PR⁻ (P = 0.002), THRT treated (P = 0.001), EGFR⁺ (P = 0.002), SR⁻ (P < 0.0001), and high mitotic rate (P < 0.0001) (Supplementary Table S1). For all patients, many of these same variables were significantly associated with shortened disease-specific survival (DSS), including: high histologic grade (P = 0.0008), large tumor size (P = 0.05), high Ki-67 (P = 0.02), ER⁻ (P = 0.02), PR⁻ (P = 0.02), SR⁻ (P < 0.0003), and high mitotic rate (P < 0.009). THRT was not significantly associated with DSS in all patients by univariate analysis. Multivariate analysis of all patients showed that when tumor size and proliferation by Ki-67 were included in the model, THRT data were independently predictive of DFS (P = 0.03), but not DSS (P = 0.15; Supplementary Table S1).

SR (ER and PR) expression data were available on 764 of 820 breast cancer, of which 576 were SR⁺ and 188 were SR⁻. In subset analysis, THRT-treated SR⁺ breast cancer showed significantly higher proliferation (Ki-67 index; P = 0.03) and upregulation of the estrogen-inducible gene EGFR (P = 0.02) as compared with the non-THRT cohort (Supplementary Table S2B). In the SR− cohort, no variables were significantly different in THRT as compared with non-THRT patients (Supplementary Table S2C). As expected, the proliferation rate (Ki-67 index) was significantly higher in SR⁻ as compared with SR⁺ breast cancer, and it was independent of THRT (mean 37.9 % vs. 20.7 %, respectively; P < 0.0001; Supplementary Table S5).

We then split the cohort of all patients by SR status, to determine associations between THRT and outcomes in each subgroup. At 10 years of follow-up, THRT-treated SR⁺ patients had a mean DFS of 72.5 ± 5.5 months, as compared with 106.4 ± 1.4 months for non-THRT SR⁺ patients (relative risk, RR 3.1, P < 0.0001; see Fig. 1B). The recurrence rate of SR⁺ patients was 39.5% for patients on THRT and 15.6% without THRT. THRT-treated patients with SR⁺ breast cancer also died of disease earlier and at a much higher rate, with a DSS mean of 84.0 ± 4.5 months in THRT as compared with 114.3 ± 0.9 months in non-THRT-treated patients (RR 3.7, P < 0.0001; Fig. 1D). At 10 years, the death rate from breast cancer was 23.7% for THRT treated, as compared with 8.4% for non-THRT SR⁺ patients. In patients with SR⁻ breast cancer, statistically significant interactions between THRT and outcomes were not detected for DFS (P = 0.49) or DSS (P = 0.53; Fig. 1C and E) or any other time point (data not shown). The recurrence and death rates were also not significantly different in SR⁻ patients by THRT, 30.0% versus 32.1% for DFS and 15.0% versus 20.1% DSS, respectively.

At the time of the last follow-up (up to 36 years), differences in outcomes by THRT were also significantly different in the SR⁺ patients (mean DFS 72.5 ± 5.5 months, no additional THRT patients failed beyond 100 months) compared with untreated patients (207.2 ± 9.3 months, P < 0.0001). The mean DSS in THRT patients, was 84.0 ± 4.5 months, in contrast to non-THRT-treated patients (223.2 months ± 4.4, P < 0.0001) who experienced significantly less death due to breast cancer. Thus in patients with SR⁺ lymph node–negative breast cancer, THRT was associated with a significantly increased risk of recurrence or death (DFS RR 2.9, P < 0.0001; DSS RR 3.4, P = 0.0004, respectively), whereas it was not associated with a worse outcome in patients with SR⁻breast cancer. Because only SR⁺ patients showed significant interactions between THRT and outcomes, we used that subgroup of patients to test the impact of treatment variables, patient or tumor characteristics that may influence the strength of the interaction. We first tested SR⁺ patients, split into two groups based on the surgical treatment they received and found no significant between surgical groups if controlled for THRT. Within each surgical subgroup associations between THRT and outcomes (both DFS and DSS) were statistically significant, see Supplementary Fig. S7 and corresponding legend for detail.

In patients with SR⁺ breast cancer, multivariate analysis showed only size and grade were independent predictors of DFS and DSS among the histologic and other clinical variables at all time points examined (5 or 10 years, or last follow-up). These were used as the base for analysis to test associations between THRT and outcomes. Using this model, THRT was independently associated with both DFS and DSS at all time points in the SR⁺ breast cancer patient cohort (Table 1A). In contrast, in patients with SR⁻ breast cancer, size was the only factor that was an independent predictor of either DFS or DSS. The addition of THRT showed no significant association with outcome in SR⁻ patients (Table 1B).

Next, we investigated the relationship between THRA and THRB expression and patient outcomes in SR⁺ (ER⁺, PR⁺) and SR⁻, HER2⁺ (ER⁻, PR⁻, HER2⁺) lymph node–negative breast cancer samples available through the Kaplan–Meier Plotter (KM Plotter) database (28). High levels of two probes for THRA were significantly associated with a shortened DFS [Supplementary Fig. S1A; HR = 3.05 (1.38–6.72), log-rank P = 0.0038] and [Supplementary Fig. S1C; HR = 2.82 (0.94–8.48), log-rank P = 0.05] in SR⁺ patients. In contrast, THRA expression was not significantly associated with survival in lymph node–negative, SR⁻ HER2⁺ breast cancer for either THRA probes (DFS; Supplementary Fig. S1D and data not shown; HR = 1.61 (0.84–3.05), log-rank P = 0.41; DFS, HR 2.43 (10.58–10.109), log-rank P = 0.21 (NS)]. No significant associations between the two probes for THRB and outcomes were observed in any breast cancer subset examined (data not shown).

Similar trends were observed using publicly available Oncomine database, where THRA and THRB were significantly higher in SR⁺ breast cancer samples relative to SR⁻ breast cancer. Consistent with our clinical cohort results, ESR1 was significantly higher in SR⁺ (ESR1; 4.87 ± 0.04, n = 1347, P < 0.0001) relative to SR⁻ breast cancer (ESR1 −1.76 ± 0.35, n = 43) in the TCGA dataset (Supplementary Fig. S1B). THRA was also significantly higher in SR⁺ breast cancer (THRA; 3.22 ± 0.06 n = 1324, P < 0.0001) relative to SR⁻ breast cancer (0.08 ± 0.16 n = 43). Using the Curtis Breast2 dataset, similar findings were observed. SR⁺ breast cancer had a significantly higher expression of THRA (3.486 ± 0.05923, n = 1190, P < 0.0001; Supplementary Fig. S8A) relative to SR⁻ breast cancer tumors (1.01 ± 0.07, n = 133). Using the Finak database, we observed that the log₂ median-centered ratio of THRA in invasive breast cancers (1.52 ± 0.13, n = 53, P = 0.002) was statistically higher than in normal breast (−0.001 ± 0.06, n = 6; Supplementary Fig. S1E). Similarly, using the Finak database (29), THRB was also higher in the invasive breast samples (0.69 ± 0.10, n = 53, P = 0.013) compared with normal breast (−0.01 ± 0.12, n = 6; Supplementary Fig. S1F). Finally, we used publicly available breast cancer databases to validate differences in the relative levels of ESR1 and THRA among SR⁺ as compared with SR⁻ HER2⁺ breast cancer. In TCGA breast cancers, ESR1 was 2.81-fold higher (P = 6.41E-57) and THRA was 1.54-fold higher (P = 5.19E-7) in the ER⁺ breast cancer as compared with ER⁻ subset cancers. Similar findings were observed in an alternate dataset Curtis Breast (P = 3.81E-27 for ESR1 and P = 2.21E-13; Supplementary Fig. S1G and S1H).

Amongst the 820 patients with stage I breast cancer, 97 were treated Tam of which nine also took daily THRT. We compared a number variables including prognostic risk factors in Tam-treated patients, as compared with Tam-untreated patients including age, grade, size, ER/PR/HER2 status, and others (Supplementary Table S3). Tam was more frequently given to women over age 50, as compared with those less than 50, although the difference was not significant (P = 0.08). Other risk factors were not significantly different either, including: grade (P = 0.61), size (P = 0.09), ER status (P = 0.68), PR status (P = 0.21), chemotherapy (P = 0.24). The only variables that were significantly associated with Tam use were a high Ki-67 score (>24%, P = 0.0003) and EGFR immunopositivity (P = 0.0185).

We also show characteristics of THRT-treated SR⁺ patients, treated or untreated by Tam (Supplementary Table S4). A number of risk factors were not significantly different in the THRT versus non-THRT-treated patients, including grade (P = 0.59), size (P > 0.99), ER status (P > 0.99), PR status (P > 0.99), chemotherapy (P = 0.55). The contingency tables show that the variables were not statistically different in Tam-treated versus Tam-untreated patients (Supplementary Table S3) as well as in the Tam treated with or without THRT patients (Supplementary Table S4). Patients with SR⁺ breast cancer treated with THRT+Tam experienced the shortest DFS survival of all cohorts studied at 10 years (mean 52.1 ± 7.0 months) as compared with 64.8 ± 1.5 months in the Tam only (Fig. 1F; P = 0.0049). Interactions between Tam and THRT failed to reach significance for DSS, although a similar trend was observed (THRT+Tam estimated mean 24.0 months, Tam no THRT estimated mean 69.5 ± 0.09 months, P = 0.09; Fig. 1G). For DSS, the number of deaths (events) was very limited (3 of 9 patients died from disease).

On the basis of clinical observations showing a worse outcome for patients with early-stage SR⁺ breast cancer on THRT, we examined the effects of TH±E2 on breast cancer cell proliferation by dose in breast cancer cell lines. ER⁺ cells, including MCF7, PT12, and T47D cells treated with TH [tetra-iodothyronine (T4) + tri-iodothyronine (T3)]±E2 in media with charcoal-stripped serum. TH and TH+E2 significantly enhanced cell growth as compared with vehicle or vehicle +E2 controls. Significant growth induction was observed across all dose levels, including the physiologic range for both hormones (Fig. 2A). Similar results were observed in T47D and PT12 cells (Supplementary Figs. S2 and S8B). Other dose ratios of T4:T3, up to 10:1 showed similar results. The effects of TH±E2 on cell proliferation were consistent using alternative methods (Supplementary Fig. S2A and S2B; MCF7 and T47D, respectively). Transcription of the ER target gene, “growth regulation by estrogen in breast cancer 1” (GREB1), was not upregulated by TH alone at several doses in MCF7 or T47D cells. These same cells were cultured in media containing E2 and TH (at multiple concentrations, including the physiologic range). GREB1 was significantly upregulated by E2+TH as compared with E2 only controls (Supplementary Fig. S2C and S2D). Under similar conditions, proteins associated with estrogenic signaling and cell-cycle regulation were also significantly upregulated in a TH dose-dependent manner in E2-pretreated MCF7 cells, including ERα, E2F1, cyclin A, cyclin B, cyclin D, cyclin E, and MYC (Fig. 2B). The proproliferative effects mediated by TH±E2, tested at a single dose and time point, were observed only in ER⁺ breast cancer cells (MCF7, PT12, and T47D) and not in ER⁻ breast cancer cells (SKBR3, MDA-MB-468, and MDA-MB-231; Fig. 2C). TH+E2 enhanced activation of ERα (Ser118) and upregulation of the cell-cycle regulatory proteins E2F1, cyclin A, Cyclin B, and Cyclin E were observed in both MCF7 and T47D cells, although MCF7 cells that express more ER had a more robust response (Fig. 2D). TH±E2 also enhanced proliferation of ER⁺ breast cancer cells MCF7 (Fig. 2E), PT12 (Fig. 2F), and T47D (Supplementary Fig. S2E) over time, but not in ER⁻ breast cancer cell SKBR3 (Supplementary Fig. S2F). Using similar methods, we showed that TH+E2 enhanced expression of GREB1, throbospondin1 (THBS1), and the proto-oncogene FOS (FOS) in both MCF7 and PT12 ER⁺ breast cancer cells (Fig. 2G and H).

To determine whether ER was required for TH to induce proliferation in ER⁺ breast cancer cells, we studied the effects of Tam or an AI, ICI (Fulvestrant, ICI,182,780) in combination with TH and E2 in ER⁺ and ER⁻ breast cancer cells. ICI significantly blocked the induction of proliferation by TH alone, TH enhancement of E2-induced proliferation, as well as E2 in MCF7 (Fig. 3A) and T47D (Supplementary Fig. S3A) cells, compared to their respective controls. In contrast, Tam blocked E2-induced proliferation, showed no effect on growth associated with TH alone and promoted proliferation in TH- and E2-treated MCF7 cells as compared with control (Fig. 3A). Similarly, the expression of the estrogen inducible gene GREB1 expression was significantly increased in TH alone or with E2. In contrast, ICI reduced estrogenic signaling even if TH was present, but Tam has limited effects (Fig. 3B). Similar findings were also observed in T47D and PT12 cells (Supplementary Fig. S3B). None of these estrogenic effects were observed in ER⁻ cells (MDA-MB-468 or SKBR3; Supplementary Fig. S3C and S3D). To further investigate mechanistic differences between the effects of Tam or ICI on TH-associated estrogenic and cell-cycle signaling in ER⁺ breast cancer cells, we studied both MCF7 (Fig. 3C) and T47D cells (Supplementary Fig. S3E). E2-treated cells showed significantly higher levels of receptor activation (Ser118 ERα) as well as increased expression of E2F1, cyclin A, cyclin B, cyclin E, and myc proteins when they were cotreated with TH. These effects were abrogated by ICI but not by Tam.

To investigate cross-talk between ER and THR pathways, we evaluated receptor localization in ER⁺ breast cancer cells in response to hormone and antagonist treatment. The combination of TH+E2 enhanced nuclear colocalization of THRα and ERα in MCF7 (Fig. 3D–F), PT12 and T47D (Supplementary Fig. S4A–S4F) cells. Consistent with the aforementioned observations of gene expression and proliferation changes, ICI significantly attenuated receptor expression, nuclear translocation, and a colocalization of both THRα and ERα, whereas Tam did not. TH+E2 robustly activated and upregulated the receptors THRα and ERα, which in turn activated estrogenic and thyroidogenic signaling in these hormone-responsive breast cancer cells. ICI blocked these effects, whereas Tam did not.

We postulated that the pro-oncogenic effects of TH on SR⁺ breast tumors and ER+ breast cancer cells involved their cognate receptors and tested this hypothesis using knockdown studies. On the basis of our initial studies, database mining, and the literature, we postulated that ER (ESR1) and THR (THRA) were key modulators of TH+E2 hormonal cross-talk in ER⁺ breast cancer. We used short hairpin RNA lentivirus to generate two constructs against THRA and ESR1 (ShTHRA#1, ShTHRA#2, ShESR1#1, ShESR1#2) to study the function of THRA and ESR1 in ER⁺ breast cancer. We efficiently knocked down both THRA and ESR1 as shown by mRNA (Fig. 3G and H; Supplementary Fig. S4G and S4H) and protein expression (Fig. 3I and J; Supplementary Fig. S4I and S4J) in both MCF7 and PT12 cells. The knockdown of THRA did not affect the expression of ESR1 mRNA levels and vice versa. Next, we explored the effects of THRA and ESR1 knockdown on ER⁺ breast cancer growth using IncuCyte Zoom. Our results showed that TH alone or TH+E2 enhanced ER⁺ breast cancer growth in the lentiviral control cells, but knockdown of either ESR1 or THRA significantly attenuated the proliferation in MCF7, PT12, and T47D cells (Fig. 3K–M; Supplementary Fig. S4K–S4M; T47D cells; Supplementary Fig. S8C–S8E). Thus, blockade of THRA or ESR1 was sufficient to inhibit the growth of ER⁺ breast cancer. These data indicated that both receptors were involved in TH+E2 induced cross-talk and growth induction in ER⁺ breast cancer. Supporting these studies, we also used methimazole (a TH inhibitor) under similar conditions, with attenuation of TH enhanced growth in E2-treated patients with ER⁺ breast cancer cells (Supplementary Fig. S8F and S8G).

Given the high expression of both ESR1 and THRA in SR⁺ breast cancer cells and the ability of both TH and E2 to enhance cell proliferation, and activation of cell-cycle regulatory proteins (in vitro), we next determined whether these effects would also occur in vivo. We evaluated TH and E2 effects on tumor growth of novel ER⁺ PDX (UCD12) grown in the quadrilateral mammary fat pads of NSG ovary-intact mice with estrogen supplementation (Fig. 4). A cohort of mice were treated with exogenous TH (LT4, 5 mg/L) for 30 days before stratification by tumor size. Each cohort was then treated ± Tam (1 mg/kg) for the duration (R_X, 60 days). E2 alone promoted tumor growth and increased tumor mass (Fig. 4A and B). TH+E2 further increased these endpoints. Tam inhibited tumor growth and burden in E2-treated mice. However, in the mice treated with both E2 and TH Tam was ineffective (Fig. 4A and B). These interactions were confirmed by analyzing pretreatment versus posttreatment changes in tumor burden for each of the cohorts. As compared with E2 alone, a combination of TH+E2 resulted in a significant increase in tumor burden (Fig. 4C). Tam attenuated the effect in animals treated with only E2, whereas this effect was not shown in the E2+TH cohort. IHC staining for proliferation (Ki-67) revealed consistent results. E2+TH-treated animals showed a significant increase in tumor proliferation, as compared with the control E2 tumors (Fig. 4D and E).

To identify the molecular mechanisms by which TH enhanced E2-mediated tumor growth, we performed RNA sequencing (RNA-seq) analysis on UCD12-PDX tumors treated with E2 or E2+TH ± Tam. An unsupervised hierarchical clustering heatmap was generated to compare E2 versus E2+TH, on UCD12 tumors using a threshold cutoff (FDR⇐0.05 and |log2FC|> = 1; Supplementary Fig. S5A and S5B). TH significantly upregulated 330 genes and downregulated 119 genes compared with E2 alone. A comparison of gene sets enriched in the E2+TH-treated tumors relative to E2 alone was examined using GSEA analysis. A number of pathways involved in cell cycle, mismatch repair, homologous recombination, and DNA replication were enriched (Supplementary Fig. S5C). Volcano plot analysis of E2 versus E2+TH-treated tumors also showed significant enrichment of thyroid-specific genes (Supplementary Fig. S5D) and estrogen-mediated signatures (Supplementary Fig. S5E). From these gene signatures, a number of genes involved in either EMT or cell-cycle regulation were modulated by TH, includingTGFB3, CCND1, GREB1, E2F1, SNA1, and MYC. In 42 cell-cycle regulatory, genes were significantly upregulated by E2+TH relative to E2 alone, including MCM2, PTTG1, CCNB2, E2F2, ORC1, CCNE1, E2F1, MYC, CCNA2, and PCNA (Fig. 4F).

Four gene expression panels [GEP, Oncotype Dx (ref. 30; Fig. 4G), EndoPredict (ref. 31; Fig. 4H), MammaPrint (ref. 32; Fig. 4I; Supplementary Fig. S6A and S6B)], and PAM50 [Prosigna (ref. 33; Supplementary Fig. S6C–S6F)], used clinically to predict prognostic risk based on gene expression profiles were used examine the relative fold expression of specific patented gene set in E2 versus E2+TH-treated human-derived zenograft tumors implanted in mice. The majority of GEPs in the E2+TH-treated samples had remarkably high expression of the prognostic gene targets, as compared with the non-TH-treated controls (Fig. 4G–I). Our data showed that breast cancer in TH-treated mice showed marked upregulation of genes associated with aggressive biological characteristics including proliferation, invasion, metastasis, evasion of apoptosis, cell division, insensitivity to antigrowth signals, limitless replicative potential, and sustained angiogenesis.

In a second animal model, we used a PT12 ER⁺ breast cancer cell line that was injected into both quadrilateral mammary glands of female NSG, ovary-intact mice with estrogen pellet supplementation. Mice were treated with or without LT4 (5 mg/L) for 30 days prior to stratification by tumor size. Each cohort (vehicle control or THRT treated) groups were then treated with or without Fulvestrant (ICI, 5 mg/kg), an AI, for the duration of the experiment (66 days). Tumor volume was measured over time. The E2+TH treatment group (358.6 ± 117 mm³, n = 8, P < 0.0001) showed significant increases in tumor volume relative to E2 alone (162.2 ± 43.2 mm³, n = 8; Fig. 5A). ICI significantly attenuated E2 increased tumor volume (73.73 ± 11.9 mm³, n = 8, P = 0.003) and diminished E2+TH mediated tumor growth (95.27 ± 16.1 mm³, n = 8, P = 0.02). These results confirmed, that attenuation of estrogenic signaling by ICI could reduce the pro-oncogenic effects mediated by E2+TH on ER⁺ breast cancer. TH significantly increased tumor weight when combined with E2 (1,090 ± 92.2 mg, n = 16, P < 0.0001) relative to E2 alone (780.9 ± 78.7 mg, n = 16; Fig. 5B). ICI treatment significantly reduced tumor weight (E2+ICI, 392.7 ± 55.94 mg, n = 16 vs. E2+TH+ICI, 471.9 ± 68.18 mg, n = 16) at necropsy. ICI inhibited E2+TH (P < 0.0001) and E2+TH+ICI tumor weights (P = 0.005). PT12 tumors treated with E2+TH grew at a faster rate (696.2 ± 63.09 mm³, n = 16, P < 0.0001), relative to E2 alone (293 ± 37.15 mm³, n = 16; Fig. 5C). In contrast, ICI attenuated E2+TH tumor burden (E2+TH+ICI, 113.6 ± 16.62 mm³, n = 16; E2+ICI, 93.74 ± 24.89 mm³, n = 16, NS). ICI significantly reduced tumor burden in E2±TH group (P < 0.0001; Fig. 5C). This was in contrast to the agonistic effects we showed in our first animal model (E2+TH+Tam; Fig. 4). We examined tumor growth over time using IVIS bioluminescent imaging and showed that TH enhanced growth relative to E2 treatment alone. This tumor growth was attenuated by the addition of ICI (Fig. 5D). Ki-67 data confirmed that TH+E2 significantly increased proliferation (Fig. 5E), whereas ICI inhibited TH+E2 associated proliferation (Fig. 5E). Immunostaining of Ki-67 was also quantified by digital imaging (Fig. 5F). TH+E2 significantly enhanced Ki-67 staining (42.76 ± 1.828%, n = 16, P = 0.0192) as compared with E2 alone (36.34 ± 1.837%, n = 16). When TH was added to E2+ICI, Ki-67 staining was not significantly increased (30.53 ± 1.638%, n = 16, NS) relative to E2+ICI alone (30.17 ± 2.021%, n = 16). The addition of ICI to TH+E2 also significantly attenuated the increased proliferation of Ki-67 (P < 0.0001; Fig. 5F).

To test interactions between an AI and THRT, we identified 160 additional patients with stage I, LN⁻ postmenopausal SR⁺ breast cancer for a second observational study (median follow-up 8.8 years). Of these, 21 patients were treated with AI + daily THRT and 52 took an AI alone. We could not assess the impact of THRT alone due to the limited patients in this group (n = 29). By univariate analysis, AI+THRT-treated patients had a significantly shorter DFS as compared with AI alone (Fig. 5G; P = 0.0419). Patients with SR⁺ breast cancer treated with AI+THRT showed a 14% relapse rate as compared with a 2% relapse rate in non-THRT patients treated with AI alone at 10 years (Fig. 5G). We saw no significant differences in DSS in this limited database, with only 1 patient dead of disease who received AI alone (data not shown). These data are consistent with our in vivo and in vitro data that suggest more effective abrogation of estrogenic signaling by AI, as compared with Tam, in the presence of TH.

An increase in the coprevalence of thyroid disease and breast cancer was first reported over a century ago (3). Numerous studies have confirmed that a majority of patients with breast cancer have preexisting thyroid disease (34) and up to one-third take daily THRT. The relationship is bidirectional as discussed in the introduction, with patients with thyroid disease experiencing higher rates of breast cancer. Meta-analytic data from several international trials of women with resected thyroid cancer, treated with TH to inhibit cancer recurrence, show a significant increase in early and aggressive breast cancer independent of radiotherapy (2, 35–38).

Our work focused on the impact of THRT on breast cancer aggression, progression, and outcomes. Because the data are exclusively derived from two cohorts of patients with only stage I LN⁻ breast cancer, we cannot be assured of similar interactions with earlier or later disease stage patients based on our work. The majority of patients in the first cohort did not receive systemic therapy, in contrast to the second cohort. Of the patients who received chemotherapy, the agents and regimens were different than current practice, hence, we cannot predict if modern chemotherapy may impact these interactions. Interactions between THRT and antihormonal agents (Tam or AI) were also observed in each cohort, and the extremely poor outcomes of cotreated women are especially worrisome. We caution, however, that because of the very limited number of patients cotreated in both cohorts, the data must be considered preliminary and requires further validation.

We have shown important interactions only in SR⁺ (and not SR⁻) breast cancer cells and between TH and other antihormonal agents in two cohorts of patients we well as in vivo, in vitro, and in silico model systems, Interactions between THRT and outcomes only in patients with SR⁺ breast cancer are critical to identifying patients at risk and suggest a mechanistic pathway underlying the cross-talk. Bidirectional crosstalk involving ER and THR, their low molecular weight ligands E2 and TH, and ligand-activated transcription factors (enhancer elements) for estrogenic or thyroidogenic target genes have been widely reported and shown to have differential activity in various organs and selective cell types (1, 14, 17, 36, 39–43). Our data show that TH activates THRA (THRα), upregulates and activates ERS1 (ERα) through MAPK activation and a feedback loop, inducing downstream signaling and colocalization of the receptors in the nuclear compartment and transcriptionally activating a number of estrogenic and cell-cycle regulatory genes and proteins. Tam has previously been shown to alter thyroid function tests in postmenopausal women with breast cancer (44). We have shown that in the presence of TH, Tam upregulated and enhanced nuclear colocalization of both THRα and ERα in ER⁺ breast cancer cells (Fig. 3D–F; Supplementary Fig. S4A–S4F). Others have shown that TH (primarily T4) signals through αVβ3 integrins to drive cell growth of ovarian cancer, and this axis may also be important in breast cancer cells (45). While this was observed in our RNA-seq data, nuclear receptor (ER and THR) cross-talk appeared to be the major driver of breast cancer growth in our model systems and patient studies, given the influence of anti-estrogenic agents and knockdown studies we performed.

On the basis of our abundant experimental and mechanistic data, we provide a gene expression and transcription model depicted in Fig. 5H. Using tumors derived from human SR⁺ breast cancer cells implanted in mice, treated or untreated with TH, we further show remarkable upregulation of the patented gene expression profiles that serve as the basis for the Oncotype Dx, EndoPredict, MammaPrint, and PAM50 tests applied clinically to SR⁺ early-stage breast cancer to subset cases by prognostic categories to inform treatment decisions. These data further support our cohort based observations that THRT promotes SR⁺ LN⁻ breast cancer aggression, as upregulation of the patented gene sets used by these tests are associated with prognostic risk.

An increased incidence of breast cancer observed in women with hyperthyroid states (endogenous or exogenous), many (but not all) studies of the impact of thyroid disease on breast cancer outcomes and our anecdotal cases of early-stage, otherwise low-risk patients with breast cancer on THRT who experienced an unexpectedly poor outcome prompted these studies. Our work narrows the breast cancer patient pool at risk to only those with SR⁺ disease, consistent with the evolving field of nuclear receptor interactions, physiological flexibility, and molecular specificity (14). Our work also suggests that patients who experienced poor outcomes as a result of THRT-induced hormonal cross-talk also demonstrated complex and unexpected actions of Tam, apparently modulating its activity to be an agonist or inducing resistance at the cellular level. According to the American Thyroid Association Management Guidelines (46), thyroid hormone suppression (THST) has been associated with significant medical risks such as tachycardia, osteopenia, osteoporosis, and atrial fibrillation. We suggest, based on a growing body of data as well as our own, that THRT and THST should be considered as a risk and prognostic factors for breast cancer. Furthermore, TH administration in the absence of clinical or biochemical hypothyroidism, such as for weight loss, mental focus or menopausal symptoms, should not be used, especially in females.

No disclosures were reported.

R.S. Wahdan-Alaswad: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. S.M. Edgerton: Conceptualization, data curation, formal analysis, validation, investigation, writing-review and editing. H. Salem: Data curation, validation, writing-review and editing. H.M. Kim: resources, software, validation, writing-review and editing. A.C. Tan: Data curation, software, validation, writing-review and editing. J. Finlay-Schultz: Methodology, writing-review and editing. E.A. Wellberg: Methodology. C.A. Sartorius: Methodology, writing-review and editing. B.M. Jacobsen: Methodology, writing-review and editing. B.R. Haugen: Conceptualization, writing-review and editing. B. Liu: Conceptualization, resources, methodology, writing-review and editing. A.D. Thor: Conceptualization, resources, data curation, supervision, funding acquisition, validation, writing-review and editing.

The authors thank Dr. John Prioleau and Mrs. Zeying Fan for their contributions. Support for the University of Colorado Cancer Center (UCCC) Genomics Shared Resource, UCCC DNA Sequence and Analysis Shared Resource Core, UCCC Biostatistics and Bioinformatics Shared Resource, UCCC Biorepository Core Facility, UCCC Tissue Biobanking and Histology Shared Resource, and UCCC flow cytometry core facility is funded in part by the NIH/NCI Cancer Center Support Grant P30CA043934. Grant support provided in part by Susan G. Komen for the Cure K100575 to R.S. Wahdan-Alaswad, S.M. Edgerton, and A.D. Thor; ACS-IRG 16-184-56 to R.S. Wahdan-Alaswad from the American Cancer Society; CCL-C92110 to R.S. Wahdan-Alaswad and A.D. Thor from the Colorado Cancer League; and Mary Kay Ashe Foundation 051-04 to A.D. Thor.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.