Activating Transcription Factor 4 Modulates TGFβ-Induced Aggressiveness in Triple-Negative Breast Cancer via SMAD2/3/4 and mTORC2 Signaling

Breast cancer is the most commonly diagnosed type of cancer in women and it is associated with high incidence and death rates (1, 2). Triple-negative breast cancer (TNBC) is an estrogen (ER), progesterone, and HER2 receptor-negative, very aggressive form of breast cancer, with a poor survival rate. TNBC is characterized by high proliferation, heterogeneity, metastases, drug resistance, and incidence of relapse, and enriched in aggressiveness-related signaling pathways such as TGFβ or mTOR. Currently, no approved targeted therapies exist for its treatment (2, 3). New or improved targeted therapies, patient stratification into responsive-to-treatment subgroups using novel prognostic biomarkers, and the identification of new therapeutic targets are required to ensure an effective personalized therapy (1).

Under stress conditions, including hypoxia, nutrient deprivation, or endoplasmic reticulum stress (ERS), the integrated stress response (ISR) is activated in cells to preserve homeostasis. The activation of the ISR leads to the global protein synthesis reduction through the eukaryotic translation initiation factor 2 alpha (eIF2α, EIF2S1) phosphorylation, driving the translation-regulated activation of activating transcription factor 4 (ATF4) that regulates cell fate. eIF2α phosphorylation is initiated by protein kinase-like endoplasmic reticulum kinase (PERK, EIF2AK3), general control nonderepressible 2 (GCN2, EIF2AK4), protein kinase double stranded RNA-dependent (PKR, EIF2AK2), and heme-regulated inhibitor (HRI, EIF2AK1) in response to the ERS, amino acid deprivation, viral infection, and heme-deficiency, respectively (4, 5).

ATF4 is a transcription factor belonging to the ATF/cyclic adenosine monophosphate response element binding protein (ATF/CREB) family, overexpressed in tumors, including breast cancer and TNBC (6–8). ATF4 regulates tumor growth, autophagy, drug resistance, and metastasis during ISR through PERK and GCN2 pathways (9–17). Independent of the cellular stress, ATF4 regulates cell metabolism (8, 18, 19), osteoblast differentiation (20), drug resistance (21), invasion, and metastasis in esophageal squamous cell carcinoma (22). In the absence of stress, high ATF4 levels correlate with poorer cancer patient survival rate (22). We previously reported increased ATF4 expression in the unstressed MCF10A cells treated with TGFβ1 (23), indicating a potential TGFβ-mediated stress-independent control of ATF4 activity.

Because of these reports, we investigated whether ATF4 can regulate the TGFβ-induced aggressiveness of TNBC and affect patient survival. The identification of the relevant signaling pathway may facilitate the design of combinatorial targeted therapies and provide a gene signature that may improve personalized medicine in breast cancer.

Supplementary Information includes Supplementary Materials and Methods, Supplementary Figure Legends, and Supplementary Tables.

Using the Kaplan–Meier plotter (www.kmplot.com/analysis), the effects of query genes on survival were assessed using 5,143 samples from patients with breast cancer. Gene expression, relapse-free (RFS; n = 3,951) and overall survival (OS) data (n = 1402) were obtained from Gene Expression Omnibus, European Genome-phenome Archive, and The Cancer Genome Atlas (24). Correlations, genomic and transcriptomic alterations, and their impact on patient survival were studied using OncoPrint and Kaplan–Meier analyses of 2,509 patients with breast cancer (25) by using cBioPortal database (26, 27).

Paraffin-embedded tissue from patients with TNBC (n = 35), with pathologic information and follow-up, no previous chemo or radiotherapy, and with previous written informed consents signed by all patients, were obtained from the Jaen's node of the Biobank of the Public Health System of Andalusia (Complejo Hospitalario de Jaén, Spain). All samples and procedures were approved by the Ethical Committee for the Research of Jaén and were conducted in accordance with the Declaration of Helsinki and International Ethical Guidelines for Biomedical Research Involving Human Subjects (CIOMS).

Tumor tissue was stained for ATF4 (Abcam; ab28830) at 1:50 dilution as reported (28). ATF4 was assessed blindly by three different pathologists. Both staining intensity and extent in neoplastic cells were considered by using semiquantitative scores: (i) staining intensity (granular, cytoplasmic) was graded as 0: no staining; 1+: weak, 2+: moderate, 3+: intense (Fig. 1B). (ii) staining extent was assigned with a value of 0 to 3 by the following criteria based on % of stained tumor cells: 0–25% = 0; 26–50% = 1; 51–75% = 2; 76–100% = 3. Finally, an integrated score was obtained by ponderation of the results as follows: values of staining extent were multiplied by the value of its corresponding intensity score. Therefore, score 0 was multiplied by 0, score 1+ was multiplied by 1, score 2+ was multiplied by 2, and score 3+ was multiplied by 3. The sum of these values (from 0 to 7) was the final score. Example: negative = 10%, 1+ = 50%, 2+ = 30%, 3+ = 10% are assigned with the values 0, 1, 1, 0, respectively. ATF4 score: (0×0)+(1×1)+(1×2)+(0×3) = 3.

TNBC cell lines, MDA-MB-231 and BT549, were purchased from the ATCC, whereas SUM159PT cells were obtained from Asterand Bioscience. SBE (SMAD binding element) reporter-HEK293 (SBE-HEK293) cell line was purchased from BPS Bioscience. All cells were maintained in DMEM (Gibco) supplemented with 10% FBS (Thermo Fisher Scientific) and 1% antibiotic–antimycotic (Gibco). SBE-HEK293 cells were cultured under Geneticin selection (Sigma), following the manufacturer's instructions.

The cells were transiently transfected with siRNAs targeting ATF4 (25 nmol/L), SMAD2/3, SMAD4, PERK, PKR, GCN2, HRI, eIF2α (EIF2S1), RPTOR, RICTOR, TAK1 (MAP3K7), and RAS (50 nmol/L) using Lipofectamine RNAiMAX (Invitrogen). TGFβ1 (10 ng/mL) was added 48-hour posttransfection, and the samples were incubated for 24 or 72 hours, depending on the experiment.

All animal procedures were approved by the Methodist Hospital Research Institute Animal Care and Use Review Office. Experiments were conducted using two human TNBC-derived patient-derived xenografts (PDX), BCM-4664 and BCM-3887 (basal intrinsic subtype; ref. 29). PDXs were transplanted into the cleared mammary fat pad of 4- to 5-week-old NOD.Cg-Prkdc^scidIl2rg^tm1Wjl/SzJ (NSG) mice maintained in the standard conditions (28). When tumors reached 150 to 200 mm³ in size, the mice were randomly assigned to four treatment groups (n = 8/group): (i) noncoding siRNA (SCR), (ii) ATF4-siRNA (siRNA#2), (iii) SCR plus docetaxel (Chemo+SCR, 20 mg/kg), and (iv) siRNA#2 plus docetaxel (Chemo+siRNA#2, 20 mg/kg) groups. siRNAs were injected twice weekly for 6 weeks at 5 μg/mouse, and docetaxel was administered once per week on days 1, 14, and 28. Tumor volumes and body weights were recorded every 2 days. Tumors were calipered and volume was calculated as previously described (30). Mice were euthanized 24 hours after the last injection and tumors were collected for further analyses. For tumor relapse, docetaxel was given at 33 mg/kg dose to BCM-4664-bearing mice, and tumor volume was recorded until the appearance of morbidity, loss of 20% of body weight, or when tumors reached 2 cm³ in size. The metastatic PDX model of TNBC is detailed in Supplementary Materials and Methods.

Differences between two groups were analyzed by two-tailed Student t test. Correlation between ATF4 staining and OS after diagnosis in tumor tissue of patients with TNBC was analyzed by the Kaplan–Meier method and further log-rank test with SPSS 21.0. Patient samples were stratified by computing all ATF4 scores by a ROC curve analysis, and the best performing threshold was used as a cutoff of positive staining in the analysis. Tumor volume was assessed by two-way ANOVA and Bonferroni's post hoc test. Median survival posttreatment in mice was analyzed using log-rank (Mantel–Cox) test and the hazard ratio with 95% confidence intervals were calculated. Each dead animal was assigned with a number 1, and each surviving mouse with a 0. The last day of treatment (day 42) was considered as day 0 for the survival analysis. A P-value <0.05 was considered significant.

High ATF4 expression was shown to correlate with poorer OS (n = 1,402, P = 0.0095) and RFS (n = 3,951, P = 8.4e−6) in all breast cancer cases (All_BC), and RFS in ER^– (n = 801, P = 0.0058), ER⁺ (n = 2,061, P = 0.0011), and TNBC (n = 255, P = 0.016) patients (Fig. 1A; Supplementary Fig. S1A). We next investigated the ATF4 expression in 35 patients with TNBC by IHC staining. The frequency of ATF4-positive staining was of 66% (intensity ≥1+; Fig. 1B). Our results showed that patients with a score ≥1 (determined by ROC curve analysis and considered as positive cases for the Kaplan–Meier analysis; Supplementary Fig. S1B) had less OS after diagnosis (37 month) than patients with score <1 (considered as negative cases; 46 months) starting at a 24-months follow-up (Fig. 1C); however, it was not significant (P = 0.125).

We previously reported increased ATF4 levels in TGFβ1-treated MCF10A cells (23). Because TNBC microenvironment is often enriched in TGFβ ligands (31), we analyzed TGFβ activation effects on ATF4 expression, and demonstrated that it increases in BT549 and SUM159PT cells treated with TGFβ1, which was abrogated by the TGFβR1 kinase inhibitor LY2157299 treatment (P < 0.001; Fig. 1D), suggesting that ATF4 represents its downstream target. ATF4 expression induced by TGFβ1 and thapsigargin was similar in SUM159PT and BT549, and lower in MDA-MB-231 (Supplementary Fig. S1C). Knockdown experiments demonstrated that ATF4 is regulated by SMAD2/3/4 (Fig. 1E). To analyze whether SMAD2/3 directly regulates ATF4, we analyzed the human ATF4 promoter region for SBEs and found the conserved CAGAC, CAGA, GTCT, GGCGC, GGCCG motifs (Supplementary Fig. S1D; ref. 32). Careful inspection of ChIP-Seq data of SMAD2/3 in BT549 cells treated with TGFβ1 for 1.5 hours (33) showed specific binding of SMAD2/3 to SERPINE1 and MMP2 (positive controls), ATF4, and the TGFβ1 responsive genes ID1, JUN, and CDKN1A promoters, but not to HBB and HPRT1 (negative controls; Supplementary Fig. S1E). Further, we carried out ChIP-qPCR analysis in BT549 cells treated with TGFβ1 for 1.5 hours and found that SMAD2/3 bind to the ATF4 promoter, comparable to the SERPINE1 and MMP2 promoters (Fig. 1F), what suggests that SMAD2/3 can bind and regulate the ATF4 gene transcription.

To ascertain the importance of ATF4 effects on the TGFβ pathway, we inhibited ATF4 expression and SBE activity was tested. The most effective siRNA sequence, siRNA#2 (Supplementary Fig. S2A), decreased SBE activity in HEK-293 cells (Fig. 1G), and phosphorylated (p)-SMAD2/3, SMAD2/3, and SMAD4 levels in BT549 and SUM159PT cells (Fig. 1H), indicating a positive TGFβ feedback. In patients with breast cancer, coexpression of ATF4 and the canonical TGFβ pathway members correlated with poorer OS (P = 0.0038; Fig. 1I), with a shift from positive to negative effects on survival when coexpressed with SMAD4 or SMAD3 in All_BC group (Fig. 1J). LOOCV results showed that ATF4 overexpression induces a significant decrease in OS (Supplementary Table S1).

ATF4 depletion in the TNBC cells decreased their wound-healing ability independently of the treatment with TGFβ1 (Fig. 2A; Supplementary Fig. S2B). According to its capacity to silence ATF4 (Supplementary Fig. S2A), siRNA#2 was more efficient to reduce the tumor cell migration. The migration index was reduced in BT549, SUM159PT, and MDA-MB-231 cells, with TGFβ1 (41%, 50%, and 45%, respectively) and without it (42%, 61%, and 65%, respectively; Fig. 2A). ATF4 knockdown with siRNA#2 in BT549, SUM159PT, and MDA-MB-231 cell lines reduced the number of invading cells with (67%, 50%, and 46%, respectively) and without TGFβ1 (38%, 23%, and 54%, respectively; Fig. 2B). In absence of chemoattractant, less number of invading cells was seen upon TGFβ1 treatment in BT549 and SUM159PT cell lines (50% and 42%, respectively). Such a decrease was seen in MDA-MB-231 regardless the absence (44% decrease) or presence (55% decrease) of TGFβ1 in the medium. These changes were accompanied with the downregulation of epithelial–mesenchymal transition (EMT)-related transcription factors (ZEB1, TWIST1, SNAIL, and SLUG) in all cells after TGFβ1 treatment, and TWIST1 and SNAIL without TGFβ1. N-cadherin levels were decreased in BT549 and SUM159PT cells, but they were not detected in MDA-MB-231 cells (Fig. 2C). Cell proliferation diminished after ATF4 knockdown (Fig. 2D), which was followed by a reduction in BCL2 and MCL1 in these cells (Fig. 2E).

Cancer stem cells (CSC) contribute to metastasis, tumor growth, and treatment resistance. To determine whether the role of ATF4 in the TNBC aggressiveness is affected by CSC alterations, we assessed ATF4 expression in mammospheres as a surrogate marker of CSCs versus the attached cells. Protein levels were shown to increase with time and mammosphere generation stage (Fig. 3A; Supplementary Fig. S2C). We investigated the effects of ATF4 depletion on mammosphere-forming efficiency (MSFE), which was reduced after ATF4 knockdown in all cells (Fig. 3B). Because ATF4 expression was induced by oxidative stress in suspension cultures (13), we measured the expression levels of stemness markers after ATF4 inhibition in the attached cells, to determine whether our results were due to the modulation of stem-like properties or they represent a consequence of detachment. NANOG, SOX2, OCT4, and CXCL10 levels were decreased in BT549 and SUM159PT cells (Fig. 3C). These results were confirmed at protein levels as well, except for CXCL10, which was not detected (Fig. 3D). With TGFβ1, cleaved NOTCH1, OCT4, and CD44 expression levels were consistently decreased in all cells. Without TGFβ1, NANOG, NOTCH1, and CD44 proteins were inhibited by ATF4-siRNA (Fig. 3D).

We selected the BCM-3887 and BCM-4664 PDX models for our analyses, with high and medium ATF4 expression, respectively, by RNA-Seq and IHC (Fig. 4A and B). To determine the effects of ATF4 silencing on metastases, we used a highly metastatic PDX model (3887-LM) in mice. After the primary tumor removal, mice were treated with DOPC-conjugated ATF4-siRNA#2 and SCR twice weekly for 6 weeks. siRNA#2-treated animals had less metastatic nodules in liver and lungs (P < 0.05; Fig. 4C and D). Metastatic lesions were confirmed microscopically by Ki67 staining (Fig. 4E) and they were positive for ATF4 (Fig. 4F).

We assessed tumor growth, ALDF+ subpopulation number, and recurrence following the mouse treatment with ATF4-siRNA and/or docetaxel. In vivo ATF4 silencing significantly reduced the tumor growth alone (P < 0.01) or in combination with docetaxel (P < 0.05; Fig. 5A), whereas the ALDF+ subpopulation number decreased in BCM-3887 model (Fig. 5B). In the BCM-4664 model, ATF4 inhibition restrained the tumor growth (Fig. 5C) and the ALDF+ subpopulation number (Fig. 5D) compared with those in the controls. To investigate tumor relapse after chemotherapy, we co-administered ATF4-siRNA and docetaxel (33 mg/kg) twice per week for 6 weeks to mice bearing BCM-4664 tumors. Chemo+SCR tumors reached the minimum volume (124 mm³) at day 24, whereas the regrowth was initiated at day 28 (128 mm³), showing a 2.4-fold increase at day 38. In contrast, tumor volume in Chemo+siRNA#2 mice was 63 mm³ at day 24, and they started to regrow at day 28 (78 mm³), reaching a 1.4-fold increase at day 38. At day 56, tumor volume in Chemo+SCR was 2083 mm³, whereas it was shown to reach 548 mm³ in Chemo+siRNA#2 mice (P < 0.001; Fig. 5C). Median survival posttreatment was 28 days in Chemo+siRNA#2 and 15 days in Chemo+SCR (P < 0.0001; Fig. 5E). To confirm that ATF4 targeting was successful, we determined ATF4 expression in BCM-3887 (Fig. 5F) and BCM-4664 tumors (Fig. 5G).

To analyze whether TGFβ activates ISR-dependent ATF4 expression, we knocked down PERK, PKR, GCN2, HRI, and eIF2α (EIF2S1) in the presence of TGFβ1, and demonstrated that their inhibition did not downregulate ATF4 expression consistently across the cell lines (Fig. 6A). PERK, PKR, and GCN2 depletion in SUM159PT cells, and GCN2 in MDA-MB-231, inhibited ATF4, indicating a cell line-dependent relevance of these pathways on ATF4 expression. Unexpectedly, ATF4 levels were increased upon eIF2α (EIF2S1) knockdown. Interestingly, after PERK knockdown, p-eIF2α was inhibited only in BT549 and MDA-MB-231 cells, what did not correlate with a decrease of ATF4.

We inhibited the noncanonical TGFβ pathways MEK/ERK, PI3K, TAK1, and P38-MAPK (34) using the pharmacologic inhibitors and TGFβ1. ATF4 expression was decreased after PI3K and TAK1 inhibition (Supplementary Fig. S3A). PI3K, mTOR, and SGK1/2 were shown to represent the upstream ATF4 regulators, independent of AKT and PDK1 (Supplementary Fig. S3B). A second PI3K inhibitor treatment excluded possible inhibitor-dependent off-target effects on ATF4 expression (Supplementary Fig. S3C). To test whether the crosstalk between TGFβ and RAS, upstream of PI3K, represents the leading signal, we transfected BT549 and SUM159PT cells with RAS-siRNA, accompanied or not by the treatment with TGFβ1. RAS inhibition failed to decrease ATF4 levels, independent of p-AKT levels (Supplementary Fig. S3D).

Rapamycin inhibits mTORC1 and mTORC2 in a dose- and time-dependent manner, together with SGK1 expression, which is activated by mTORC2 (35). To determine whether ATF4 is a downstream target of mTORC1 and/or mTORC2 with active TGFβ, TNBC cells were transfected with RPTOR or RICTOR-siRNAs and treated with TGFβ1. We observed decreased ATF4 levels only following the RICTOR inhibition in all analyzed cells (Fig. 6B). Because we showed that SNAIL expression considerably decreases after ATF4 knockdown, it was used as a surrogate for ATF4 inhibition. SNAIL levels were decreased after RICTOR silencing and TGFβ1 treatment in all analyzed cells (Fig. 6C).

mTOR signaling activity is modulated by several feedback loops (35), and therefore, we analyzed a potential feedback loop between ATF4 and mTORC1/2. In 2,509 patients with breast cancer, ATF4 expression was shown to correlate with the expression of mTORC1 (EIF4E, R = 0.463; RPS6, R = 0.380) and mTORC2 targets (NDRG1, R = 0.213; RHOA, R = 0.320; P < 0.0001; Fig. 6D). The positive feedback between ATF4 and mTORC1 and mTORC2 activity was further confirmed by demonstrating that ATF4-siRNA treatment inhibited the downstream targets of mTORC2 (p-NDRG1, RHOA, RAC1) and mTORC1 pathway (p-AKT, p-P70S6K) in SUM159PT and BT549 cells (Fig. 6E). Interestingly, RHOA and RAC1 levels were consistently reduced after TGFβ1 treatment, and RAC1 expression inhibition was maintained at different time points in all cell lines (Fig. 6E; Supplementary Fig. S3E).

Collectively, our data suggest that ATF4 is involved in and regulates both the canonical, SMAD2/3/4, and noncanonical, PI3K/mTORC2/RHOA-RAC1, TGFβ signaling pathways to modulate metastasis, stemness, and tumor cell survival (Fig. 6F).

To help improve the prognosis and treatment decision-making in patients with breast cancer, using Kaplan–Meier plotter database, we studied the impact of different members of the TGFβ/SMAD/ATF4 and PI3K/mTOR/ATF4 pathways on the breast cancer patient RFS (Supplementary Table S2) using multivariate analysis and LOOCV. Here, we identified an eight-gene signature, including ATF4, TGFBR1, SMAD4, PIK3CA, RPTOR, EIF4EBP1, RICTOR, and NDRG1 genes, that predicts a poorer RFS in the high-expression cohort of All_BC (61-fold-change decrease; n = 1,764, P < 0.005), ER^– (81-fold-change decrease, n = 347, P < 0.005; Fig. 6G), and basal intrinsic subtype (n = 618, P < 0.005; Supplementary Table S3). In All_BC group, this signature predicts a 27-time poorer RFS compared with that predicted by using ATF4 expression alone, as the single gene associated with the highest significant decrease in RFS of this group. In the ER^– group, multigene signature predicts a 53-time poorer RFS compared with that predicted using NDRG1 expression alone, which was the gene associated with the highest significant decrease in the RFS of this group (P < 0.005; Fig. 6G). LOOCV results demonstrated that the lower RFS in ER⁺ and patients with TNBC depended on the NDRG1 expression, although its impact on RFS could not be determined in patients with TNBC (P < 0.008; Supplementary Table S2).

OncoPrint in 2,509 patients with breast cancer showed that the gene-signature expression is altered in 1,138 patients (45; Supplementary Fig. S3F). The percentages of alterations ranged from 4% to 25% for individual genes (ATF4, 4%; TGFBR1, 4%; SMAD4, 7%; PIK3CA, 7%; RPTOR, 8%; EIF4EBP1, 16%; RICTOR, 6%; NDRG1, 25%). Kaplan–Meier analysis and LOOCV results showed that patients with the altered expressions (n = 1,055) of these genes have poorer survival (143 months) compared with that of the patients without alterations (n = 925, 173 months, P = 0.00005; Fig. 6H).

ATF4 has been proposed as a potential contributor to the pathogenesis and development of breast cancer; however, the underlying mechanisms and the impact on patient survival remain unclear. In patients with breast cancer, infiltrating carcinoma had higher p-ATF4 than normal breast tissue, what was associated with lymph node metastases (7). Recently, a gene expression analysis revealed that ATF4 is overexpressed in TNBC patient tissues (8). Here, we investigated the potential of ATF4 as a prognostic marker and therapeutic target in breast cancer and showed that high ATF4 RNA expression correlates with poorer survival in patients with All_BC, ER⁺, ER^–, and TNBC. In a cohort of 35 patients with TNBC, we found a trend showing that ATF4 protein expression correlates with a poorer OS. Our results demonstrate that ATF4 positiveness starts to have a negative impact on survival of patients with TNBC at 24 months of follow-up. A higher follow-up period and a bigger cohort would be necessary to show statistically significant results.

During tumor invasion and metastases, active pathways like TGFβ or NOTCH induce EMT, a shift from the epithelial into mesenchymal phenotype, induced by transcription factors such as SNAIL, SLUG, TWIST1, and ZEB1 (36). TGFβ-induced EMT leads to the generation of CSCs with increased self-renewal and tumor-initiating capabilities, resistance to apoptosis and chemotherapy, decreased proliferation, and enhancing tumor recurrence (31). TNBC samples exhibit gene expression profiles observed in CSCs and during EMT, such as increased TGFβ and mTOR expression (3), together with a more frequent expression of CSC markers, which is associated with poorer patient outcomes (37). Similar to earlier reports in osteoblasts and pancreatic adenocarcinoma cells (11, 20), we reported previously enhanced ATF4 levels in MCF10A cells treated with TGFβ1 (23), which suggests that ATF4 expression is regulated by TGFβ. This pathway is commonly upregulated and necessary in tumor progression and EMT in patients with TNBC (3, 31, 38). Our results showed that ATF4 is expressed after TGFβ1 treatment in TNBC cells, and its expression is inhibited by the treatment with LY2157299, suggesting a direct effect of TGFβ on ATF4 expression. We showed that SMAD2/3/4 were at least partially responsible for the regulation of ATF4 expression after TGFβ1 treatment. Further analysis of previously published ChIP-Seq data (33) and subsequent ChIP-qPCR in TGFβ1-treated BT549 cells demonstrated for the first time that SMAD2/3 bind and regulate ATF4 transcription. Previous reports show that ATF4 was dependent of SMAD3 in mouse adipocytes (39) but independent of SMAD4 in osteoblasts (40). As previously described for ATF3 (41), ATF4 depletion reduced TGFβ activity and SMAD2/3/4 expression, indicating the presence of a feedback loop between ATF4 and TGFβ pathway. In patients with breast cancer, coexpression of ATF4/TGFBR1, ATF4/SMAD2, ATF4/SMAD4, and ATF4/SMAD3 resulted in poorer OS, shown to depend on ATF4 overexpression. Together, these results demonstrate that TGFβ/SMAD2/3/4 are upstream of ATF4, which regulate the signaling through a positive feedback with the TGFβ pathway, and it may be involved in the TGFβ-associated aggressiveness of TNBC.

Here, we report a more important role of ATF4 in the constitutive (average of 56% decrease) than in the TGFβ1-induced tumor cell migration (average of 45% decrease). However, ATF4 was more relevant in the TGFβ1-induced than in the basal tumor cell invasiveness (average of 54% and 38% decrease, respectively). In addition, different EMT transcription factors and stemness markers were inhibited by ATF4 silencing when TGFβ1 was added or not to the medium. Our results suggest that, under nonstressing conditions, ATF4 is involved in the aggressiveness of TNBC cells mediated not only by TGFβ, but also by other signaling pathways. In TNBC PDX mouse models, ATF4 depletion resulted in a reduced lung and liver metastasis rate, tumor growth, ALDF+ CSC-like population numbers, delayed tumor relapse, and increased mouse survival. Accordingly, independent of the ISR-induced ATF4 expression, the effects of ATF4 on the cell functions regulated by growth-factors were shown to be important (8, 18–22). Taken together, our findings indicate that ATF4 modulates the aggressiveness of TNBC through the regulation of ISR-independent key signaling pathways, suggesting a potential usefulness of this gene as a therapeutic target.

ATF4 is regulated at both transcriptional and translational levels by different signals (6). Our results show that ATF4 expression depends on the canonical SMAD-dependent TGFβ pathway; however, there are not evidences in literature showing that SMADs are responsible to modulate protein translation. Numerous stress types induce the ISR-regulated ATF4 activation mediated by p-eIF2α (4). The ISR controlled by PERK-GCN2/eIF2α/ATF4 mediates EMT and metastasis (10, 11, 13), tumorigenesis (9, 14), and chemoresistance (16, 17). In nonstressing conditions and presence of TGFβ1, we found that the ISR did not drive ATF4 expression for all the cell lines tested herein; however, it was important in SUM159PT and MDA-MB-231 cells. Contrary to previous reports (9–11, 13, 15, 16), eIF2α depletion induced ATF4 expression and, noteworthy, when PERK was inhibited, reduced eIF2α phosphorylation was only observed in BT549 and MDA-MB-231, what did not reduce ATF4 levels. These results suggest that neither PERK nor eIF2α are responsible for the ATF4 translation in absence of stress and presence of TGFβ1. Therefore, we sought to investigate the eIF2α-independent regulator mechanism of ATF4 activation that could be shared by the three TNBC cell lines. TGFβ activates noncanonical pathways such as PI3K, MAPK, and TAK1 (34). Pharmacologic inhibitor treatment showed that ATF4 expression in the presence of TGFβ1 is also regulated by the PI3K/mTOR pathway independent of AKT activity. In colorectal cancer, ATF4 stabilized by mutant PI3K was found downstream of PDK1/RSK2, and shown to reprogram glutamine metabolism independently of AKT (19). We showed that ATF4 expression does not depend on the presence of AKT, PDK1, RSK2, or PIK3CA mutations, as only SUM159PT cells harbored a mutation in PIK3CA (3). In contrast to previous reports (18, 40), our findings revealed mTORC2 to be the leading upstream regulator of ATF4 upon TGFβ1 treatment (although mTORC1 was also important in SUM159PT and BT549 cells). mTORC2 has been reported as a necessary mediator in the TGFβ-induced EMT through AKT phosphorylation (Ser473; ref. 42), a well-known feedback loop that activates mTORC1 (35), as well as in protein translation by direct interaction with ribosomal proteins (43). Similarly, mTORC1 modulates not only ATF4 transcription but also translation independently of eIF2α (5). Whether mTORC2 can directly regulate ATF4 translation and transcription remains elusive, but it would explain why ATF4 expression is independent of AKT, PDK1, RSK2, or PIK3CA mutations. According to previous studies (5, 15), we observed that ATF4 also regulates mTOR signaling by a feedback loop on mTORC1, what may regulate cell survival and drug resistance induced by MCL1 and BCL2 (21, 44), and mTORC2. We hypothesize that this dual regulation on mTOR may be attributed to the ATF4-mediated regulation of RAC1 that further affects mTORC1 and mTORC2 activity in response to growth-factor stimulation (45), which may potentiate the mTORC2/AKT feedback loop on mTORC1. Because TGFβ signaling activates both mTORC1 (46) and mTORC2 (47) in a SMAD-dependent way by inhibition of DEPTOR (47), we suggest that TGFβ could activate mTORC2 (and mTORC1 in some cell lines) through a SMAD-dependent signaling, what would induce ATF4 expression to mediate EMT, motility, metastasis, pluripotency, and self-renewal. Active mTORC2 could also feedback on AKT to enhance mTORC1-dependent ATF4 translation and transcription. This circuit would be maintained by the feedback of ATF4 on mTOR and TGFβ signaling through the regulation of SMAD2/3/4 and mTORC1/2-RHOA-RAC1 pathways (48, 49).

Identifying patterns that predict signaling pathway activation, by using gene signatures and considering the target interactions, has been demonstrated to be a viable approach to the personalized TNBC treatment (1). Because ATF4 is involved in TGFβ/SMAD and TGFβ/PI3K/mTOR pathways, we identified an eight-gene prognostic signature, including ATF4, TGFBR1, SMAD4, PIK3CA, RPTOR, EIF4EBP1, RICTOR, and NDRG1 genes that can be used for the prediction of patient survival in all breast cancer, ER^–, and the basal subtype groups. The expression of these signature genes was shown to be altered in 45% of 2,509 patients with breast cancer, with lower survival rates observed in these patients. Breast cancer patient stratification, especially ER^– patients, according to this gene signature may provide a useful strategy for designing effective signaling pathway-guided combinatorial targeted therapies aimed at the reduction of tumor growth, metastases, and relapse risk, and may allow the identification of potentially responsive patients.

In conclusion, we demonstrate here for the first time the potential of ATF4 as a prognostic biomarker and a therapeutic target in patients with TNBC. Furthermore, we showed that ATF4 is involved in the regulation of signaling pathways associated with tumor metastasis, proliferation, and drug resistance, which induce the aggressiveness of TNBC. In contrast to the previous reports, ATF4 activity was shown to be independent of the ISR, integrating and modulating TGFβ/SMAD2/3/4 and TGFβ/PI3K/mTORC1/2 pathways. We identified a signaling pathway-guided prognostic gene signature in patients with breast cancer that may help design combinatorial targeted therapies, identify potential responsive patients, and predict and overcome drug resistance. Precision medicine may benefit from our approach by improving the treatment decision-making in breast cancer patients.

No potential conflicts of interest were disclosed.

Conception and design: A. González-González, J.A. Marchal, D. Landeira, P. Sanchez-Rovira, J.C. Chang, S. Granados-Principal

Development of methodology: A. González-González, E. Muñoz-Muela, F.E. Cara, M.P. Molina, M. Cruz-Lozano, G. Jiménez, A. Ramírez, M.D. Martín-Salvago, R.J. Luque, C.L. Ramirez-Tortosa, C. Rodriguez-Aguayo, A.K. Sood, P. Sanchez-Rovira, J.C. Chang, S. Granados-Principal

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E. Muñoz-Muela, W. Qian, W. Chen, A.J. Kozielski, M.D. Martín-Salvago, R.J. Luque, M. Quintana-Romero, R.R. Rosato, M.A. García, H. Kim, C. Rodriguez-Aguayo, G. Lopez-Berestein, P. Sanchez-Rovira, J.C. Chang, S. Granados-Principal

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. González-González, E. Muñoz-Muela, F.E. Cara, M.P. Molina, M. Cruz-Lozano, G. Jiménez, A. Verma, A. Ramírez, O. Elemento, M.D. Martín-Salvago, R.J. Luque, C. Rosa-Garrido, D. Landeira, M. Quintana-Romero, R.R. Rosato, C.L. Ramirez-Tortosa, H. Kim, A.K. Sood, P. Sanchez-Rovira, J.C. Chang, S. Granados-Principal

Writing, review, and/or revision of the manuscript: A. González-González, E. Muñoz-Muela, F.E. Cara, G. Jiménez, D. Landeira, R.R. Rosato, H. Kim, G. Lopez-Berestein, A.K. Sood, J.A. Lorente, P. Sanchez-Rovira, S. Granados-Principal

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): W. Qian, C. Rosa-Garrido, P. Sanchez-Rovira, J.C. Chang, S. Granados-Principal

Study supervision: R.R. Rosato, J.A. Lorente, P. Sanchez-Rovira, S. Granados-Principal

This work was funded by Instituto de Salud Carlos III (to S. Granados-Principal: CP14/00197, PI15/00336, MV16/00005; to J.A. Marchal: PIE16/00045), European Regional Development Fund (to S. Granados-Principal), the Chair “Doctors Galera-Requena in Cancer Stem Cell Research” (to J.A. Marchal), Spanish Ministry of Economy and Competitiveness and Andalusian Regional Government (to D. Landeira: RYC-2012-10019; BFU2016-75233-P; PC-0246-2017). M. Quintana-Romero is funded by the “Garantia Juvenil” program (REF-2813; Andalusian Regional Government and University of Granada). We thank Dr. Pedro Carmona-Sánchez for the assistance on statistics of the meta-analysis and Editage (www.editage.com) for English language editing.

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.