Dual mTOR Kinase Inhibitor MLN0128 Sensitizes HR⁺/HER2⁺ Breast Cancer Patient-Derived Xenografts to Trastuzumab or Fulvestrant

Hormone receptor–positive (HR⁺)/HER2⁺ tumors account for approximately 10% of all breast cancers. Activation of estrogen receptor alpha (ERα) promotes the growth of HR⁺ tumors. HR⁺/HER2⁻ patients are treated with endocrine therapies, such as aromatase inhibitors (AI) and ERα antagonists (1–3). HER2 belongs to receptor tyrosine kinase (RTK) family. In the clinical setting, trastuzumab is the current standard treatment for HR⁻/HER2⁺ patients. A combination of HER2-targeting agents and chemotherapies is recommended for HER2⁺ advanced breast cancer (4). HR⁺/HER2⁺ breast cancer is heterogeneous and bears the concurrent activity of ER and HER2 (3–6). The HR⁺ status implies that the ER pathway plays a critical role in directing the clinical features of HR⁺/HER2⁺ breast cancer. While chemotherapy is typically applied, the increased HR level in HR⁺/HER2⁺ tumors is correlated to a better response to endocrine therapies (4–6). Therefore, the use of chemotherapy may overtreat this subgroup of breast cancer patients, causing potentially unnecessary adverse events and economic cost (6). Additionally, tumor response to trastuzumab may be reduced by concurrent activity of the ER pathway (7). Clinical studies show that the addition of anti-HER2 therapies significantly improves the clinical outcomes of HR⁺/HER2⁺ patients treated with AIs (8). Clinical and laboratory studies also suggest that cross-talk between the ER and RTK signaling pathways plays a critical role in resistance to endocrine therapies and anti-HER2 agents in HR⁺/HER2⁺ breast cancers (3, 5). The molecular mechanisms of acquired and de novo resistance to anti-HER2 agents vary in different biological contexts. Several studies reveal that ERα signaling is augmented in the presence of anti-HER2 agents (9, 10). Notably, cancer cell lines resistant to anti-HER2 agents are often sensitive to the pure antiestrogen fulvestrant. In addition, activation of the PI3K/AKT/mTOR signaling has been reported in the establishment of endocrine resistance, leading to estrogen-independent ERα activation. In the clinic, the mTORC1 inhibitor everolimus has been prescribed with endocrine agents for advanced and AI-resistant HR⁺ breast cancers (11–16).

The PI3K/AKT/mTOR axis, as the downstream of the RTK cascade, is considered to be a promising therapeutic target in HR⁺ and HER2⁺ breast cancers (14–16). The protein kinase mTOR, which exists in two complexes (mTORC1 and mTORC2) and serves as the catalytic subunit, transduces proliferation cues to the signaling components of the PI3K/AKT/mTOR axis. While activation, PI3K and mTORC2 activate AKT, leading to upregulation of mTORC1 and the downstream S6K1 and 4EBP1 for promoting cell proliferation and survival (16–20). Due to the PI3K/AKT rebound effect, long-term treatment of mTORC1 inhibition has been known to alleviate AKT activation by mTORC2. Therefore, allosteric mTOR inhibitors (e.g., everolimus) do not fully inhibit mTORC1 activity (19). To overcome this barrier, the ATP-competitive dual mTOR kinase inhibitors, such as MLN0128, are developed to simultaneously target both mTORC1 and mTORC2 (20).

Presently, there are a few HR⁺/HER2⁺ breast cancer cell lines available for preclinical research (9), which are limits in understanding the clinical features of patient tumors. Instead, PDXs mimicking the features from the tumors of origin can offer a realistic solution (21). In this study, we established two PDX models from two individual tumors to gain insight on HR⁺/HER2⁺ breast cancers. Via the two PDXs with distinct clinical features, we investigated the role of estrogen/ER in controlling tumor growth and dissected molecular characteristics of this breast cancer subtype. Integrative omics analyses coupled with in vivo drug efficacy examination further defined the therapeutic effects of simultaneous suppression on ER signaling and mTOR pathway by fulvestrant and MLN0128, respectively, in HR⁺/HER2⁺ breast cancers.

Surgical resections (2 × 2 mm²) from consented HR⁺/HER2⁺ breast cancer patients were orthotopically engrafted into the mammary fat pad of 6- to 8-week-old female NOD-scid/IL2Rγ^−/− (NSG) mice to derive parental tumors. To examine estrogen-dependent growth, tumor slices (2–3 mm) from the parental tumors and in-house-made estrogen pellets containing 1 mg of E2 each were simultaneously implanted into the mammary fat pad of 6-week-old ovariectomized NSG mice (22). For in vivo drug efficacy examination, tumor slices (2–3 mm) from the parental tumors were engrafted into the intact NSG mice. Once tumors were established to be 100 to 200 mm³ in size, mice were randomized and then subjected to 4-week treatment of control (150 μL sterile saline, twice weekly of intraperitoneal injection), trastuzumab (10 mg/kg in sterile saline, twice weekly of intraperitoneal injection), fulvestrant (5 mg in 100 μL sterile saline, once weekly/subcutaneous injection), and/or MLN0128 (0.3 mg/kg in 5% N-methyl-2-pyrrolidone, 15% polyvinyl pyrrolidone in water, 6 days weekly of gavage; ref. 15). Tumor growth and body weight were monitored twice per week and tumor volume was calculated as length × width² × π/6 (23). Fulvestrant (Faslodex, AstraZeneca) and trastuzumab (Genentech) were obtained from the City of Hope National Medical Center Pharmacy. The City of Hope Institutional Review Board approved this study and all patients provided written informed consent prior to tissue collection. All animal experiments were done under a protocol approved by the Institutional Animal Care and Use Committee. Facilities are credited by the Association for Assessment and Accreditation ×of Laboratory Animal Care and operated according to NIH guidelines.

Immunohistochemistry (IHC) and hematoxylin and eosin (H&E) staining of formalin-fixed tumor tissues were performed by the Pathology Core facility at City of Hope. Antibodies used in IHC included ERα (ab1660; Abcam), PR (PA0312; Leica Biosystems Inc.), HER2 (A0485; DAKO/Agilent Technologies), and Ki-67 (M7240; DAKO). Hormone receptors were interpreted and scored according to joint American Society of Clinical Oncology/College of American Pathologists guidelines (24, 25). Slides were reviewed first at 10× magnification to identify areas of positive staining, followed by confirmation and quantitation at 20 to 40× magnifications. Ki-67 was scored by identifying areas of most abundant positivity at low magnification, followed by manual counting of 10 high-power (40×) fields. Representative images and scoring were acquired using an Olympus BX46 microscope with DP27 camera and Olympus CellSens software (Olympus).

Total RNA from six COH-SC1 and three COH-SC31 PDX tumors, respectively, and two biological replicates from an ER⁺/HER2− PDX established in our laboratory (indicated as REF in Fig. 3A) were extracted using the RNeasy Extract Kit (Qiagen) and then subjected to RNA sequencing conducted by the Integrative Genomics Core at City of Hope. The details of sequencing processing, sequence read mapping, and further systems-level pathway analyses using Ingenuity Pathway Analysis (IPA; Ingenuity Systems) and gene set enrichment analysis (GSEA) were summarized in Supplementary Materials and Methods. All sequencing data were submitted to the GEO database.

Three snap-frozen tumor samples from COH-SC1, COH-SC31, and an ER⁺/HER2⁻ PDXs were subjected to RPPA analysis probed a total of 232 antibodies conducted by the MD Anderson Cancer Center Functional Proteomics RPPA Facility as described previously (26). To define the relatively activated signaling in either the COH-SC1 or COH-SC31 model, RPPA data of an ER⁺/HER2⁻ PDX was used as the reference (indicated as REF in Fig. 3C).

To decipher the biology and potential therapeutic strategy of HR⁺/HER2⁺ breast cancers, two PDXs named as COH-SC1 and COH-SC31 were established. As shown in Fig. 1A, surgical resections from consented HR⁺/HER2⁺ breast cancer patients were engrafted into the mammary fat pad of immunodeficient NSG mice to derive parental tumors (step 1). Once established, parental tumors were divided into pieces and subsequently implanted to 30 to 40 mice for the following molecular and pharmacologic analyses. Samples from individual clones were subjected to RNA-seq transcriptome and RRPA proteome analyses to dissect the molecular features of the two PDXs. “Same-patient-on-all-arms” testing cohorts were generated for drug efficacy examination (step 2). Drug-mediated tumor growth and phenotypic/molecular changes were then investigated (step 3).

COH-SC1 was derived from an ER⁺/PR⁻/HER2⁺ breast tumor prior to treatment and the patient later responded well to trastuzmab plus chemotherapies; COH-SC31 was established from a trastuzumab-resistant ER⁺/PR⁺/HER2⁺ metastatic breast tumor (Supplementary Table S1). H&E staining indicated that cell morphologies between PDXs and the corresponding patients’ samples are similar in both models, which COH-SC1 tumors are moderately differentiated but poor differentiation in COH-SC31 (Supplementary Fig. S1A). IHC analyses of ER, PR, and HER2 expression patterns independently supported the recapitulation capability of PDXs by showing that COH-SC1 tumors harbor 1% to 2% of ER⁺, no PR expression, and strong HER2⁺ expression (scored as 3⁺); 40% to 50% of ER⁺, 2% to 5% of PR⁺, and scored as 3⁺ of HER2⁺ staining in COH-SC31 (Supplementary Fig. S1B). To further confirm the differential expression levels of ERα between the two PDXs, Western blot analysis conducted on three biological replicates per model and MCF-7 cancer cells as the positive control pointed out the relatively higher ERα expression in COH-SC31 tumors (Supplementary Fig. S1C). In addition, there was no mutation found in the ESR1 locus in both COH-SC1 and COH-SC31. Taken together, the above pathological observations on PDXs and the corresponding patients’ tumors were similar, supporting that the established PDXs is capable of recapitulating the clinical features of individual breast tumors.

To clarify whether estrogen/ERα signaling drives HR⁺/HER2⁺ tumor growth, PDX tumors were implanted into ovariectomized mice accompanied with or without estrogen pellets. Uterus weight was then measured to assess in vivo estrogen responses and showed significant enlargement of uteri in the presence of estrogen pellets (n = 9, P < 0.001; Supplementary Fig. S2A), supporting the fact the implanted estrogen pellets are the main source of estrogen supplementation. Tumor volumes and weights were simultaneously monitored to indicate that estrogen remarkably contributes to tumor growth of both two PDXs (P ≤ 0.01; Fig. 1B and Supplementary Fig. S2B). Additional estrogen-dependency comparison verified that COH-SC31 tumor growth relies on estrogen (n = 5, P = 0.0087; Fig. 1B, bottom), but the presence of estrogen potentially acts as a growth accelerator in COH-SC1 tumors (n = 9, P < 0.01; top). For example, at 77 day after tumor implantation (Fig. 1B, top), tumor volumes of COH-SC1 in the absence and presence of estrogen were 438.5 ± 159.1 mm³ (see “CTRL”) and 994.8 ± 333.2 mm³ (see “E2”), respectively. Contrastingly, after 91 days in posterior to COH-SC31 tumor implantation (Fig. 1B, bottom), tumor volume was ∼100 mm³ under estrogen deprivation (see “CTRL”) and 619.0 ± 164.6 mm³ upon E2 supplement (see “E2”). IHC of Ki-67 (a cellular proliferation marker; ref. 27) expression levels was subsequently determined to verify estrogen-driven proliferation by showing approximately 1.4 and 1.6 times higher in Ki-67–positive staining in COH-SC1 (60.4% ± 3.5% vs. 43.1% ± 1.9%; left) and COH-SC31 (29.8% ± 4.4% vs. 18.8 ± 3.7%; right), respectively (Fig. 1C). Notably, Ki-67–positive cells in COH-SC1 (43.1% ± 1.9%) were relatively higher than that of COH-SC31 (18.8% ± 3.7%) under estrogen-deprived conditions (see “CTRL” in Fig. 1C), supporting the estrogen independency in COH-SC1 tumor initiation. These findings demonstrated that estrogen/ERα signaling modulates HR⁺/HER2⁺ tumor growth in different manners, which it acted as a crucial initiator in controlling tumor growth of COH-SC31 with higher ER expression, but an accelerator in facilitating growing of COH-SC1 tumors with lower-expressed ER (Fig. 1B and C and Supplementary Fig. S1C).

Trastuzumab is the first-line monotherapy of HER2⁺ breast cancer (4–7). To validate its efficacy against HR⁺/HER2⁺ tumors in vivo, the intact NSG mice bearing either COH-SC1 or COH-SC31 tumors were subjected to 4-week drug treatment. As shown in Supplementary Fig. S3A, either trastuzumab (TRA) or placebo (CTRL) treatment triggered less than 15% reduction in body weights of COH-SC1 and COH-SC31 PDXs, supporting the well-tolerated treatment scheme in mice. In the COH-SC1 model (n = 7), compared with the CTRL group, TRA treatment significantly caused ∼45% reduction in tumor volume (P = 0.006; Fig. 2A, top and Supplementary Fig. S6A), but not tumor weight (P = 0.62; Fig. 2B, top; Supplementary Fig. S6C). Consistent with the clinicopharmacological history of the corresponding patient (Supplementary Table S1), COH-SC31 PDX derived from a trastuzumab-relapsed breast tumor was found to be resistant to TRA treatment with no considerable changes in either tumor volume (P = 0.0436) or weight (P = 0.8889; n = 5; Fig. 2, bottom plots). These findings exhibited the distinct trastuzumab response in the two established HR⁺/HER2⁺ PDXs as well as suggested the insufficiency of trastuzumab monotherapy in treating HR⁺/HER2⁺ tumors and the need of defining alternative therapeutic optimization.

To decode molecular features of HR⁺/HER2⁺ breast cancers and explore potential therapeutic targets, whole-genome RNA-seq transcriptome and RPPA proteome analyses were performed on COH-SC1 and COH-SC31 tumors. To identify HR⁺/HER2⁺ PDX-specific gene signature, another RNA-seq dataset from an ER⁺/HER2⁻ PDX tumor was included as the reference (see “REF” in Fig. 3A). Compared with two biological replicates of REF results, a heat map plotted to summarize the expression patterns of 7,014 differentially expressed loci (P < 0.05) in either COH-SC1 (n = 6) or COH-SC31 (n = 3) datasets individually showed a distinct transcriptome pattern of these two PDX models (Fig. 3A). To explore the major signaling controlling HR⁺/HER2⁺ tumor growth, a total of 549 common upregulated loci (fold change ≥1.5, P < 0.05) shown in the box areas were subjected to IPA to uncover that RTK (i.e., EGFR, HER2, and HER3) and PI3K/AKT/mTOR pathways are the mostly significant networks in the two studied models (P < 0.001; Supplementary Fig. S4A). Differential expression profiling between COH-SC1 or COH-31 and REF analyzed by GSEA supportively identified that HER2, mTOR signaling, and ERα target-Cyclin D1, are remarkably activated in the two examined PDXs (Fig. 3B). RPPA results of 232 pan-/phosphor-proteins independently revealed that expression levels of the active forms of signaling components in the PI3K/AKT/mTOR pathway (e.g., AKT, 4E-BP1, and S6K), as well as those of HER2 (e.g., HER2 and HER3) and ER signaling (e.g., Cyclin D1, c-MYC, and PR) are relatively higher at the posttranslational level (P < 0.05; Fig. 3C and Supplementary Fig. S5A and S5B). Western blot analysis of the phosphorylation status of AKT at serine 473 and ERα at serine 118 further validated activated AKT signaling in the two examined PDXs, but relatively low levels of activated ERα (Supplementary Fig. S5C). RPPA comparison and IPA of differential gene expression between COH-SC1 and COH-SC31 tumors additionally pointed out that expression levels of ERα signaling-involved loci or proteins are relatively higher in COH-SC31 (Supplementary Figs. S4B, S5A, and S5B). For example, expression levels of ESR1 (10.64-fold) and well-known ERα-targeted loci, such as TFF1 (657.11-fold) and BMP7 (203.42-fold), were higher in COH-SC31 in relation to that of COH-SC1 (Supplementary Table S2). Summarization of these findings proposed that PI3K/AKT/mTOR signaling is a critical oncogenic pathway in both PDXs but ERα signaling plays a more dominated role in the COH-SC31 model.

To address the clinical relevance of COH-SC1 and COH-SC31 PDXs, we conducted in silico expression analysis of a published breast cancer cohort containing 266 samples (Fig. 4; ref. 28). Based on PAM50 (prediction analysis of microarray 50) subtype characterization (29), the transcriptome profiling of COH-SC1 tumors was similar to that in luminal-B/basal-like tumors (Fig. 4A). Nine genes highlighted in luminal-B/basal-like patient samples were coincidently highly expressed in the COH-SC1 model and involved in PI3K/AKT/mTOR signaling. Kaplan–Meier analysis of those identified gene signature found that the activated PI3K/AKT/mTOR axis is correlated to poor prognosis of luminal-B breast cancer patients (P = 0.0136; Fig. 4C). However, PAM50 analysis of COH-SC31 transcriptome elucidated the association with luminal-A breast cancer subtype (Fig. 4B). The highlighted sections of 21 highly expressed loci in luminal-A samples were associated with ER and PI3K/AKT/mTOR signaling components, which were also highly expressed in COH-SC31 tumors. Kaplan–Meier survival analysis of those identified genes with increased expression signature was linked to poor prognosis of luminal-A breast cancer patients (P = 0.0393; Fig. 4D). These clinically relevant analyses suggested that: (i) blockade of the PI3K/AKT/mTOR signaling would potentially suppress COH-SC1 tumors; (ii) simultaneous inhibition of ER and PI3K/AKT/mTOR pathways can be a more efficient strategy to attenuate COH-SC31 tumor progression.

Integration of transcriptome and proteome results discovered the activated PI3K/AKT/mTOR pathway as the oncogenic driver in both HR⁺/HER2⁺ PDXs (Fig. 3 and Supplementary Figs. S4 and S5). To assess its therapeutic potential, in vivo efficacy examination of a dual mTOR inhibitor MLN0128 (MLN) was performed on the two preclinical models using the intact NSG mice bearing either COH-SC1 or COH-SC31 tumors. After 4-week oral administration, less than 5% decline in body weight observed in the two PDXs supported the well-tolerated toxicity of treatment dosage (see “MLN” in Supplementary Figs. S3B and S6B). MLN treatment profoundly caused 41% to 66% and 38% decrease in tumor volumes of COH-SC1 (n = 7, P ≤ 0.0021) and COH-SC31 (n = 5, P = 0.0042) PDXs, respectively (Fig. 5A and Supplementary Fig. S6A). These observations demonstrated that MLN0128 monotherapy results in partial growth inhibition of both HR⁺/HER2⁺ PDXs, suggesting that alternative therapeutic strategies are needed for more efficient against HR⁺/HER2⁺ tumors.

While TRA single-drug treatment triggered ∼45% tumor volume reduction in COH-SC1 PDX (Fig. 2A, top and Supplementary Fig. S6A), we hypothesized that multidrug treatment scheme brings better therapeutic efficacy against trastuzumab-sensitive HR⁺/HER2⁺ cancers. To verify this hypothesis, we used the same preclinical cohort, which was applied to test trastuzumab efficacy, to evaluate the effectiveness of TRA-plus-MLN combination treatment in vivo. Compared with TRA alone, dual-therapy scheme with the acceptable drug tolerance (∼14.9% decrease in body weight; Supplementary Fig. S6B) efficiently prohibited the majority of tumor growth (∼80.5%, P < 0.0001) and reduced tumor weight (P < 0.001) (n = 7; see “Comb” in Supplementary Fig. S6A and S6C). Interestingly, MLN single-drug treatment in the same cohort induced ∼66% growth inhibition (P < 0.0001) in relation to the CTRL group (Supplementary Fig. S6A), implying the better potency of MLN against trastuzumab-sensitive COH-SC1 tumors in relation to that in the TRA group. However, monotherapy using either TRA or MLN cannot completely attenuate mitogenic signaling (see “Mitotic counts” in Supplementary Fig. S6D), suggesting the superior therapeutic potential of the combination strategy against HR⁺/HER2⁺ tumors.

To corroborate this concept, we examined the drug efficacy in vivo on both COH-SC1 and COH-SC31 PDXs. In the trastuzumab-sensitive COH-SC1 model, the TRA-plus-MLN combination significantly suppressed ∼80.5% tumor growth (see “Comb” in Supplementary Fig. S6A). But, it concurrently increased the ER-expressing cell population (see “Comb” in Supplementary Fig. S6D), implying that simultaneous blockade of the ERα and mTOR pathway is another therapeutic scheme against trastuzumab-sensitive HR⁺/HER2⁺ cancer, which can potentially lower the recurrence risk resulting from the increasing ER expression. Because the activated PI3K/AKT/mTOR and ERα pathways were found to be involved in tumor progression of HR⁺/HER2⁺ cancers via integrative omics analysis (Fig. 3 and Supplementary Figs. S4 and S5), we, thus, tested the efficacy of combined MLN with antiestrogen fulvestrant (FUL) on the two preclinical cohorts bearing COH-SC1 or COH-SC31 tumors as aforementioned in Fig. 5. Within the acceptable drug tolerance (P < 0.8933; Supplementary Fig. S3B), the MLN-plus-FUL combination resulted in significant 50% to 60% tumor volume reduction in COH-SC1 (60%, P = 0.0021) and COH-SC31 (50%, P < 0.0001) models and downregulation of the associated PI3K/AKT/mTOR and ERα signaling (see “Comb” in Fig. 5A and C), whereas either MLN (41% and 38% decrease in COH-SC1 and COH-SC31, respectively) or FUL (17% increase in COH-SC1 and 43% decrease in COH-SC31) single-drug treatment induced less than 50% suppression. To validate these in vivo observations, MLN or FUL treatment was independently applied on an in vitro culture system including three HR⁺/HER2⁺ breast cancer cell lines—MCF-7aro/HER2 derived from MCF-7aro cells with stable overexpression of HER2 (30), BT-474 (HR⁺/amplified HER2), and BT/Her0.2R derived from BT-474 cells with long-term trastuzumab exposure (31)—which was previously found to be sensitive to trastuzumab treatment in the presence of tamoxifen (32), and HR⁺ MCF-7aro cancer cells. We observed that the four examined cancer cell lines respond to either MLN or FUL in a dose-dependent manner (Fig. 6A and B). Compared with HR⁺ MCF-7aro cancer cells, HER2-overexpressing MCF-7aro/HER2 cancer cells were less sensitive to either MLN or FUL treatment (Fig. 6C), implying that HER2 overexpression likely activates mTOR and ER signaling as consistent as the integrative omics observations. However, MCF-7aro/HER2 and BT-474 cancer cells were more sensitive to FUL treatment in relation to that in BT/Her0.2R cells (Fig. 6C). Further combination treatment performed on BT-474 and BT/HER0.2R cancer cells revealed that the presence of MLN enhances the potency of FUL in treating HR⁺/HER2⁺ cancer cells (Fig. 6D and E), supporting that dual-therapy treatment using MLN-plus-FUL potentially increase drug potency against trastuzumab-sensitive as well as trastuzumab-resistant HR⁺/HER2⁺ tumors.

Despite our knowledge of the apparent therapeutic targets of HR⁺/HER2⁺ breast cancer, the optimal therapeutic approach for this subtype remains evolutionary. Currently, endocrine therapy is the preferred option for HR⁺ cancer (2–5). While utilizing a collection of HR⁺ breast PDXs to identify therapeutic targets and evaluate drug response has been reported (33), the HR⁺/HER2⁺ subgroup has not been specifically characterized. In this study, we molecularly established two PDXs from two individual tumors to comprehensively translate the biology of HR⁺/HER2⁺ breast cancers and subsequently explore targeted-therapy strategies. COH-SC1 PDX was derived from a tumor removed from a patient with positive response to trastuzumab plus chemotherapy, whereas the COH-SC31 model was established from a trastuzumab-resistant tumor. According to dissection of molecular features using omics approaches and PAM50 analyses, we found that COH-SC1 PDX with high expression of MKI167 (encoding Ki-67) and low levels of ERα is associated with luminal-B subtype (Fig. 4A); COH-SC31 with high expression of ESR1 (encoding ERα) and BCL2 presented as luminal-A subtype (Fig. 4B). Taken together, the two PDXs revealed that different levels of ERα and dissimilar trastuzumab responses in HR⁺/HER2⁺ breast cancers might link to diverse treatment scheme.

ER signaling has been reported to drive HR⁺/HER2⁺ breast tumorigenesis in cancer cell lines (3–5). Here, we used COH-SC31 PDX harboring high expression of ERα to confirm the crucial role of ER signaling in modulating HR⁺/HER2⁺ tumor growth. Long-term estrogen exposure as the driving force of estrogen/ERα signaling directed COH-SC31 tumor growth at a larger extent than that in COH-SC1 (Fig. 1B). The transcriptome analysis additionally pointed to the relative elevation of ER signaling and it function readouts, such as SERPINA1 (34), in the COH-SC31 model compared with that of COH-SC1 (Supplementary Figs. S1C and S4B). Moreover, the expression patterns of SERPINA1, SIAH2, and EZH2 identified from the transcriptome profiling reflected the pharmacological behaviors of COH-SC1 and COH-SC31 to fulvestrant treatment. It has been documented that high levels of SERPINA1 predict good response to endocrine therapy as similar as our previous observation (34); low expression of SIAH2 is associated with resistance to endocrine therapies in vitro and in the clinic (35, 36); reduction of EZH2 expression resensitizes cancer cells to antiestrogens (37). In vivo drug efficacy examination supportively confirmed that COH-SC31 is more responsive to fulvestrant in relation to COH-SC1 does (see “FUL” in Fig. 5A). As a result, endocrine therapies should be included to treat HR⁺/HER2⁺ breast cancers expressing high levels of ERα or activated estrogen-dependent ERα signaling, especially for those developed trastuzumab resistance.

The bidirectional interaction between ER and RTK pathways frequently conferred resistance to single-drug treatment of endocrine therapy or HER2-targeted therapy. PI3K/AKT/mTOR signaling contained multiple components (e.g., AKT) involved in these processes (3–5, 9). Compared with expression profiling of the reference samples in transcriptional or translational levels (Fig. 3 and Supplementary Figs. S4 and S5), a highly activated PI3K/AKT/mTOR axis was found in the two examined HR⁺/HER2⁺ PDXs. Further in silico expression and Kaplan–Meier survival analyses supported the therapeutic value in targeting PI3K/AKT/mTOR signaling (Fig. 4C and D). Two phase III clinical trials (BOLERO-2 and HORIZON) evaluated the addition of mTORC1 inhibitors to endocrine therapies in treating advanced or metastatic HR⁺/HER2⁻ breast cancers (13, 38, 39). But, overall survival (OS) was not benefited from the addition of mTORC1 inhibitors, which may be due to the PI3K/AKT rebound effect, resulting in the alleviation of AKT (40). Instead, MLN0128 targeting to both mTORC1 and mTORC2 was developed to treat HR⁺ as well as HER2⁺ breast cancers in vitro and in vivo (15, 20, 41). In ER⁺ breast cancer, dual mTORC inhibitors (e.g., AZD8055 and AZD2014) combined with endocrine agents exhibited superior inhibition effects on in vitro culture system or cell-based xenografts compared to that of single-drug treatment (14, 16). Additionally, MLN combined with and the dual tyrosine kinase inhibitor lapatinib has synergistic effects on HER2⁺ cancer cell lines and PDXs, which are resistant to anti-HER2 drugs (15). At the present time, fulvestrant with either MLN or AZD2014 are being investigated in HR⁺/HER2⁻ advanced metastatic breast cancer in phase II clinical trials. To sum up, combination strategy with utilization of MLN may shed a light to treat HR⁺/HER2⁺ cancers.

In this study, we first observed that COH-SC1 PDX is more responsive to the first-line TRA treatment compared with the COH-SC31 model (Fig. 2A), consistent with the pharmacological features of the corresponding tumors. Nonetheless, TRA therapy only contributed to ∼45% abolishment of tumor volumes, suggesting that alternative treatment schemes are in need to treat trastuzumab-sensitive HR⁺/HER2⁺ breast cancers. Via integrative omics analysis, we uncovered that, instead of the HER2 pathway, PI3K/AKT/mTOR and ER signaling are relatively upregulated in the two studied PDXs and attenuation of these two pathways has therapeutic value against both trastuzumab-sensitive and -resistant HR⁺/HER2⁺ breast cancers (Fig. 3 and Supplementary Figs. S4 and S5). In trastuzumab-sensitive and lower ER expressed COH-SC1 PDX (Fig. 2A and Supplementary Fig. S1C), we conducted a dual-therapy treatment scheme using TRA and MLN to prove that combination therapy is more potent against tumor growth in relation to TRA or MLN monotherapy does (Supplementary Fig. S6A). However, TRA-plus-MLN combination treatment as same as TRA triggered an increase of ER-positive cell population (see “Comb” in Supplementary Fig. S6D), indicating that diminishing both PI3K/AKT/mTOR and ER pathways provides better therapeutic potential against HR⁺/HER2⁺ cancer. To address this assumption, in vivo drug efficacy examination using MLN and/or FUL revealed that MLN alone as like MLN-plus-FUL combination potently reduces COH-SC1 tumor growth (Fig. 5A, top), whereas dual-therapy treatment scheme results in more significant tumor growth inhibition of COH-SC31 PDX (P < 0.0001; Fig. 5A, bottom), supporting the applicable potential of MLN in treating HR⁺/HER2⁺ tumors. Moreover, we simultaneously targeted ER signaling and one of the major signaling outputs in the HER2 pathway– PI3K/AKT/mTOR axis, which proved the concept of treating HR⁺/HER2⁺ breast cancers via dual inhibition of ER and HER2 signaling cross-talk. The different pharmacological behaviors of the two HR⁺/HER2⁺ PDX models with different molecular signatures suggested the importance of estrogen response and ER function/levels in guiding treatment decisions. Due to the complicated and redundant signaling transduction in the HER2 signaling network, the treatment backbone may need to be further fine-tuned for a more complete suppression of ER and RTK signaling. However, it is important to figure out the possibility to treat HR⁺/HER2⁺ cancers with targeted strategies such as drugs targeting ER, HER2, and PI3K/AKT/mTOR, rather than chemotherapies associated with severe side effects. Clearly, mTOR-targeting drugs, such as MLN0128, work synergistically together with anti-HER2 drugs (e.g., trastuzumab) or endocrine therapeutic agents (e.g., fulvestrant) against ER⁺/HER2⁺ breast cancers.

In HR⁺/HER2⁺ breast cancers, the ER signaling pathway is frequently activated while suppressing HER2 signaling (9, 10, 42), which potentially leads to resistance to anti-HER2 drugs and resensitization to endocrine therapies; this scenario was similar to that observed in the COH-SC1 and COH-SC31 models (Fig. 5A and Supplementary Fig. S6D). We considered poorly differentiated COH-SC31 as a good trastuzumab-resistant model in which estrogen-mediated ER activation became the major driving force of growth (Fig. 1B). Previous studies had shown that the concurrent pharmacological inhibition of the ER and RTK pathways achieved better effect on tumor regression than the anti-HER2 regimens alone in HR⁺/HER2⁺ cells that are resistant to anti-HER2 regimens (9, 10, 42). Our in vivo study showed the MLN-plus-FUL combination treatment synergistically inhibits COH-SC31 tumor growth, providing clinical implication that this therapeutic strategy can serve as a second-line treatment for HR⁺/HER2⁺ breast cancer patients relapsed from trastuzumab therapy.

With the advance of genomic profiling studies on breast cancers, an increasing number of druggable genomic aberrations are discovered, suggesting the importance of personalized medicine in effective breast cancer therapy (43). However, cancer cell line models are limited in adequately reflecting tumor heterogeneity and morphology in vivo, which depreciates the predictive value of these models in drug discovery (21). Alternatively, we utilize PDX technology to explore therapeutic strategies for individual HR⁺/HER2⁺ breast cancer. Similar to observations reported by Cottu and colleagues (44), COH-SC1 and COH-SC31 PDXs retain biomarker status and pathological characteristics of their original patient tumors. The preclinical phenotype of COH-SC31 is consistent with its clinical history, supporting that this PDX model can recapitulate patient clinical features (Supplementary Fig. S1A and S1B). This study presents an example of examining drug efficacy in the “same-patient-on-all-arms” trials and provides a prototype for designing future “co-clinical” trials to tailor personalized therapeutic decisions in clinical practice (45). One caveat is that the inclusion of more PDX breast samples would enhance the clinical implication of this study in understanding and managing HR⁺/HER2⁺ breast cancers. However, a low success rate to generate HR⁺/HER2⁺ breast PDXs and time-consuming endpoint evaluation in testing drug response of individual PDXs hamper the scaling up of this research module. Moreover, because PDXs are expected to be heterogeneous by nature, molecular characterizations need to be performed with multiple biological replicates similar to what we experienced in this study. Also, while it is difficult to expand the PDX lines in a timely manner, the study has to be carefully prioritized.

In summary, we have generated two HR⁺/HER2⁺ breast PDX models that exhibit different molecular and pharmacological features, compared with the currently available cell line models. The establishment and characterization of these two PDX models broadens the representation of HR⁺/HER2⁺ breast cancer research tools with more diverse genetic profiles and clinical relevance. We have provided new preclinical models to clinicians and scientists in tailoring therapeutic strategies for this 10% of the total breast cancer population, by demonstrating that (i) a dual mTORC inhibitor MLN sensitizes HR⁺/HER2⁺ breast cancers to TRA or FUL treatment; (ii) the level of estrogen-mediated ER activity in affecting pharmacological behaviors in HR⁺/HER2⁺ breast cancers; (iii) tumors with higher ER activity in tending to be more responsive to FUL. Noteworthy, targeting to trastuzumab-resistant breast cancers with high ERα expression, MLN in combination with MLN approves to be a potential therapeutic solution. Furthermore, PDX technology has the potential to be implemented into breast cancer disease management. Although performing therapeutic screening in PDXs is labor-intense and time-consuming, it can provide invaluable information in guiding therapeutic strategies for late-stage breast cancer patients.

No potential conflicts of interest were disclosed.

Conception and design: P.-Y. Hsu, V.S. Wu, S. Chen

Development of methodology: P.-Y. Hsu, V.S. Wu, P. Chu, S. Chen

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P.-Y. Hsu, V.S. Wu, N. Kanaya, K. Petrossian, H.-K. Hsu, D. Nguyen, D. Schmolze, M. Kai, H. Lu, P. Chu, C.A. Vito, L. Kruper, J. Mortimer

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P.-Y. Hsu, V.S. Wu, D. Schmolze, S. Chen

Writing, review, and/or revision of the manuscript: P.-Y. Hsu, V.S. Wu, C.-Y. Liu, L. Kruper, S. Chen

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P.-Y. Hsu, V.S. Wu, S. Chen

Study supervision: S. Chen

The authors thank Dr. Susan Kane for providing the BT/Her0.2R cancer cell line and Dr. Nicola Solomon for assistance with editing the manuscript. This work was supported by the Panda Charitable Foundation (S. Chen) and the NCI (P30 CA033572). Research reported in this publication included work performed in the Integrative Genomics Core, Pathology Research Service Core, and Animal Resource Center supported by the NCI (P30CA033572).

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.