Hormone-dependent regulation of the KLF9 gene in the breast
Based on publicly available RNA-seq datasets from the GTEx and TCGA breast invasive carcinoma, we found that KLF9 expression is significantly downregulated in tumor samples in comparison to normal tissue controls (Fig. 1A). Stratifying tumor samples into molecular subtypes based on transcriptomic data from TCGA and SCAN-B, expression of KLF9 was moderately downregulated in subtypes associated with increasing disease severity and worse prognosis (Fig. 1B). This pattern of expression was reflected in BCa cell lines, with the non-malignant MCF10A cells exhibiting the highest KLF9 mRNA levels (Fig. 1C).
The KLF9 gene is an established direct GR target in neuronal, skin, and lung cells (14, 18, 49). To evaluate the GC-dependent regulation of KLF9 in the breast, we treated BCa lines with increasing doses of CORT (0-300 nM) for 2 hr. In MCF10A and in the highly aggressive MDA-MB-231 cells, KLF9 mRNA expression increased with increasing CORT concentrations (Fig. 1D; MCF10A: EC50 = 21.71, MDA-MB-231: EC50 = 10.93). On the other hand, a significant induction in KLF9 expression was only seen in MCF7 cells treated with 300 nM CORT (Fig. 1D, EC50 = 46.31), which may be attributed to significantly lower NR3C1 (GR) transcript levels in MCF7 cells (S Fig. 1; Additional File 2).
Analysis of transcriptome data in ER + MCF7 cells (39) revealed that 10 nM E2 treatment of MCF7 cells led to a rapid and transient increase in KLF9 mRNA that returned to baseline levels by the first hour of treatment and steadily declined over the course of 24 hr (S Fig. 2A; Additional file 3). To further explore the possible E2-dependent regulation of KLF9, we performed a dose-response experiment on ER + MCF7 cells incubated with increasing concentrations E2 for 24 hr. However, we observed that KLF9 expression remained generally unaltered with E2 treatment (Fig. 1E) in contrast with the E2-dose dependent increase in mRNA level of the direct ER target GREB1 (S Fig. 2B-C; Additional File 3).
KLF9 is directly regulated by GR via GC-responsive enhancers
Treatment of MCF10A with the GR-selective antagonist MIF completely abolished the CORT-induced increase in KLF9 mRNA indicating that CORT-dependent regulation of KLF9 is mediated by and specific to the GR, and not the mineralocorticoid receptor (Fig. 2A). In addition, treatment with the protein synthesis inhibitor CHX did not affect the CORT-dependent increase in KLF9 mRNA (Fig. 2B) and pre-mRNA (Fig. 2C) supporting that GR directly upregulates the transcription of the KLF9 gene in MCF10A cells.
We then sought to identify putative cis-regulatory elements that can mediate response to GC, E2, and circadian signaling. We performed in silico analysis of GR, CLOCK, and ER ChIP-seq data (42, 43, 45), active chromatin marks (H3K27Ac, DNase I hypersensitivity, RNA pol2 binding) (42, 50), evolutionary conservation among vertebrates (50), and predicted long-range interactions with the KLF9 promoter from the GeneHancer database (46), which altogether constitute features of candidate GC-, E2-, and circadian clock-responsive enhancer elements. Based on the GR ChIP-seq data and sequence analysis with LASAGNA 2.0 (48), there was negligible GR localization and no GRE identified in the proximal KLF9 promoter until the region 3 kb upstream of the TSS (S Fig. 3; Additional File 4). With this, we focused on three candidate regulatory elements which show preferential GR localization upon treatment with the synthetic GC analog dexamethasone (DEX) in both MCF10A and MDA-MB-231 cells, along with presence of active chromatin marks within each putative enhancer (Fig. 2D). First, the previously described mouse Klf9 synergy module (KSM) (17), was extended to include three additional CLOCK-binding motifs (E-boxes) and an ERE half site. Herein referred to as eKSM, it is located 4.3kb from the TSS and harbors a canonical GRE and a GRE half-site (Table 1, S Fig. 4; Additional File 5). Upstream of the eKSM located 5.2kb from the TSS and referred to as aKSM is another candidate enhancer enriched for GR and CLOCK localization and contains a non-canonical E-box and a single GRE. Finally, the KDE, a TF binding hotspot located 65.98 kb upstream of the KLF9 TSS, is another potential cis-regulatory element containing two GREs.
We cloned each of the identified putative enhancer regions into a luciferase reporter construct and evaluated CORT-mediated transactivation in the three mammary epithelial cell lines. In MCF10A cells, a robust increase in luciferase activity in cells transfected with the aKSM and KDE constructs was observed upon CORT treatment and the induction was abolished with the addition of MIF (Fig. 2E). Notably, transactivation of the eKSM construct, although significant, was only at 1.19-fold relative to ~ 3-fold activation for the other enhancer-reporter constructs. In MCF7 cells, GR-specific transactivation of aKSM and KDE is conserved, while the eKSM was not induced by CORT (Fig. 2F). Lastly, in the MDA-MB-231 line, both aKSM and KDE exhibited CORT-dependent transactivation, although the fold increase in luciferase activity for the KDE construct was only 1.7-fold in comparison to ~ 3-fold induction in normal MCF10A and luminal MCF7 cells (Fig. 2G). The EDE (33) was induced in all three cell lines and served as the CORT-responsive positive control.
Owing to the localization of RNA pol II at the aKSM, eKSM, and KDE (Fig. 2D), we also investigated whether these candidate enhancers are transcribed into non-coding RNA. In MCF10A cells, hormone response of all three eRNAs was similar to that of the KFL9 mRNA, such that eKSM (S Fig. 5A, B; Additional File 6), aKSM (S Fig. 5C, D; Additional File 6), and KDE (S Fig. 5E, F; Additional File 6) eRNA expression was induced by CORT in a GR-specific manner and was resistant to protein synthesis inhibition.
Circadian expression of KLF9 in mammary epithelial cells
KLF9 is expressed in a circadian fashion in the mouse liver (15), human skin (14), and mouse and human hippocampus (16). To determine whether KLF9 oscillates in the breast and recapitulate circadian gene expression in vitro, we synchronized mammary epithelial cell lines through incubation with 1 µM CORT pulse for 2 hr after which cells were harvested every 4 hr for 48 hr (40). In MCF10A cells, CORT pulse resulted in rhythmic oscillation of KLF9 mRNA that was antiphase with BMAL1 transcript levels (Fig. 3A-C), with maximal KLF9 mRNA levels occurring at around 24 hr corresponding with the BMAL1 mRNA nadir (Fig. 3A). Cosine wave regression analysis determined the period to be at 25.8 hr and 28.1 hr for BMAL1 (Fig. 3B) and KLF9 (Fig. 3C), respectively. This is consistent with in vivo time-series transcriptomic data from Yang and colleagues (51) (E-MTAB-5330) where the mouse Klf9 gene was included in the list of genes with circadian expression. Klf9 expression in the mouse mammary gland is indeed rhythmic, peaking at circadian time (CT) 11 and CT35 (S Fig. 6A-B; Additional File 7). As details on the period, amplitude, and phase shift were unavailable in the study by Yang et al. (51), we applied the same cosine-wave fit analysis on their microarray data and predicted the period to be 26.14 hr (S Fig. 6B; Additional File 7). The BMAL1 target PER1 also exhibited an antiphase relationship of expression with that of the master clock regulator in normal MCF10A cells (S Fig. 6C-E; Additional File 7).
We employed the CORT pulse method to synchronize MDA-MB-231 cells (Fig. 3D-F). Oscillation of the BMAL1 transcript in this cell line is still evident with the period determined to be 27.43 hr (Fig. 3D, E), although expression did not decrease after the expected peak at t = 36 hr as observed in MCF10A cells. While BMAL1 mRNA had rhythmic expression in MDA-MD-231 similar to that in MCF10A, circadian oscillation of KLF9 mRNA was absent in this BCa line (Fig. 3D, F). After decreasing at the 4-hr timepoint after CORT withdrawal, KLF9 transcript levels remained stable until t = 40 hr and thereafter abruptly increased to baseline levels by the 48-hr timepoint. Moreover, the curve fitting failed to detect any period for the KLF9 time-series dataset in MDA-MB-231 cells (Fig. 3F). PER1 oscillation was likewise slightly irregular in the MDA-MB-231 line, with period determined to be 21.4 hr (S Fig. 6F-H; Additional File 7).
Previous studies showed that KLF9 can be induced by CLOCK/BMAL1 in keratinocytes and neuronal cells through multiple E-boxes upstream of its TSS (14, 16). To determine if CLOCK/BMAL1 can influence KLF9 gene transcription in mammary epithelial cell lines, we transiently overexpressed BMAL1 and CLOCK in MCF10A cells and evaluated changes in KLF9 expression. Transfection of increasing concentrations of CLOCK (Fig. 3G) plus BMAL1 (Fig. 3H) plasmid DNA resulted in a trend of increasing KLF9 transcript levels (Fig. 3I). Notably, while CLOCK mRNA was increased in a dose-dependent manner with increasing amounts of transfected CLOCK expression vector, there was only a slight increase in in BMAL1 mRNA with increasing amounts of transfected BMAL1 vector (Fig. 3G-I), which might partially account for sub-optimal induction of KLF9 observed in the assay.
Feedback of KLF9 to the breast circadian oscillator
KLF9 has been previously demonstrated to co-localize with the core circadian regulator CLOCK on promoters of core clock and clock output genes in mouse hippocampal neurons (16). To determine if the phenomenon is conserved in mammary epithelia, we evaluated publicly available KLF9 and CLOCK ChIP-seq datasets in MCF7 as there were no available datasets for MCF10A cells. As with hippocampal neurons, we observed a similar pattern in MCF7 cells where we found overlapping KLF9 and CLOCK ChIP-seq peaks at the loci of core clock genes PER1-3 (Fig. 4A, S Fig. 7; Additional File 8) and CRY1-2 (TTFL #1) (S Fig. 7; Additional File 8), accessory loop genes NR1D1-2 (TTFL #2) (S Fig. 7; Additional File 8), and auto-regulatory loop genes DEC2 (TTFL #3) (Fig. 4A), as well as the clock output gene TEF (S Fig. 7; Additional File 8). We also observed prominent KLF9 peaks in DEC1, WEE1 (S Fig. 7; Additional File 8), and DBP (Fig. 4A) although they did not overlap with a CLOCK peak at these loci.
With this, we further probed the regulatory impact of KLF9 on baseline expression of clock genes and their response to both GC and E2 signaling. To this end, we stably expressed KLF9-specific shRNAs and effectively reduced KLF9 expression by 60–80% in the three model cell lines using shKLF9 type #3 (sh3, S Fig. 8A, C, E; Additional File 9). Since knockdown upon transduction of shKLF9 #3 was consistently higher compared to shKLF9 #4, we opted to use the cells transduced with this shRNA type for downstream functional analyses. To complement our knockdown experiments, we also overexpressed KLF9 coding sequence in all three cell lines (S Fig. 8B, D, F; Additional File 9).
We then evaluated changes in the expression of components of the core clock machinery (BMAL1, CLOCK, PER1/2/3, CRY1/2, NR1D1/2, and BHLHE40/41) and three clock output genes (WEE1, TEF, DBP) upon knockdown or overexpression of KLF9 in normal MCF10A cells and in the highly aggressive MDA-MB-231 cancer line. Genetic perturbation generally led to moderate changes in clock gene expression, with KLF9 knockdown in MCF10A resulting in increased levels of mRNAs for BMAL1 and DEC2 (Fig. 4B), as well as CRY2, NR1D2, and DEC1 (S Fig. 9A; Additional File 10). In the MDA-MB-231 line, KLF9 knockdown led to a consistent upregulation of DEC2 (Fig. 4C) and CRY2 (S Fig. 9B; Additional File 10), but not in the other genes studied. On the other hand, KLF9 overexpression in MCF10A reduced BMAL1 and DBP transcript levels (Fig. 4D), whereas other clock genes like DEC2 (Fig. 4D), CRY1, NR1D1, and NR1D2 increased (S Fig. 9C; Additional File 10). In MDA-MB-231 cells, overexpression of KLF9 resulted in an increase in BMAL1, DEC2 (Fig. 4E), CRY2, NR1D1, and TEF mRNAs (S Fig. 9D; Additional File 10), but reduced the levels of DBP (Fig. 4E) and DEC1 (S Fig. 9D; Additional File 10). Notably, BMAL1, which increased upon KLF9 knockdown in MCF10A, was reduced upon forced expression in the line. In addition, DBP which is downregulated by KLF9 in mouse hippocampal neurons (16) was unchanged upon KLF9 knockdown in MCF10A, but exhibited a trend of upregulation in MDA-MB-231 cells. Overexpression of KLF9 led to the reduction in DBP transcript levels in both cell lines (Fig. 4D- E).
In line with evidence that hormones can modulate the local mammary circadian clock by directly impacting expression of some core clock genes, we also investigated whether KLF9 can alter the GC or E2 response of established hormone-regulated clock genes. We treated the KLF9-overexpressing MCF10A cells with 100 nM CORT for 2 hr and evaluated changes in the expression of clock genes, PER1 and DEC2, known to be induced and repressed by CORT, respectively (52, 53). CORT treatment induced PER1 mRNA, and KLF9 overexpression augmented the CORT-mediated increase in PER1 transcript (Fig. 4F). On the other hand, CORT treatment decreased DEC2 mRNA levels by 20%, and repression was further enhanced by ectopic expression of KLF9 by 31% relative to vehicle-treated cells. (Fig. 4G).
In addition, since E2 can directly impact the expression of CLOCK (11) and PER2 (10) in MCF7, we also investigated whether KLF9 can influence the E2-mediated induction of these genes in this cell line. We treated empty vector- and KLF9-overexpressing MCF7 cells with 1 µM E2 for 24 hr. Remarkably, KLF9 overexpression abrogated the E2-dependent induction of CLOCK (Fig. 4H) and diminished the magnitude of E2-mediated increase in expression of the direct ERα target GREB1 (S Fig. 10; Additional File 11), consistent with its demonstrated antagonism of ERα-mediated signaling (19). The transcript levels of PER2 were refractory to E2 treatment in MCF7 but decreased upon concomitant KLF9 overexpression (Fig. 4I).
Functional role of KLF9 in breast cancer pathogenesis
To investigate the impact of the GC signaling and KLF9 in BCa etiology, we treated the KLF9 knockdown and overexpression lines with CORT prior to functional analysis using assays of cancer hallmarks including survival, proliferation, apoptosis, and migration. The colony formation assay was utilized to determine the effects of CORT treatment and KLF9 on cellular survival and neoplastic transformation of breast epithelial cell models, while the resazurin-based assay that quantifies the changes in the bulk metabolism of the cells was used as an indirect measure of proliferation (54). Non-malignant MCF10A cells formed very few colonies (Fig. 5A; S Fig. 11A-B; Additional File 12). CORT treatment enhanced colony formation and proliferation (Fig. 5A-B; S Fig. 11G-H; Additional File 12) in MCF10A, an effect that was unaffected by KLF9 knockdown but diminished by KLF9 overexpression. In the luminal MCF7 line, CORT instead reduced long-term cell survival (Fig. 5C; S Fig. 11C-D; Additional File 12) and viability (Fig. 5D; S Fig. 11I-J; Additional File 12). Knockdown of KLF9 reduced the anti-survival effects of CORT but did not influence CORT-mediated inhibition of proliferation. In addition, KLF9 overexpression consistently augmented the anti-tumorigenic effects of CORT in this cell line. For the aggressive TNBC MDA-MB-231line, CORT treatment generally enhanced survival (Fig. 5E; S Fig. 11E-F; Additional File 12) and proliferation (Fig. 5F; S Fig. 11K-L; Additional File 12) of the cells. Knockdown of KLF9 enhanced survival but did not further affect proliferation. On the other hand, ectopic expression of KLF9 reversed the pro-oncogenic effects of CORT.
Next, we assessed the influence of KLF9 on DOX-induced apoptosis in the luminal MCF7 and aggressive TNBC MDA-MB-231 lines through a fluorescence-based assay that measures caspase-3/7 activation. DOX-mediated cytotoxicity in MCF7 cells was unaltered by CORT treatment and KLF9 knockdown (Fig. 6A). In MDA-MB-231 cells, CORT treatment by itself protected against baseline apoptosis in both scrambled and shKLF9-transduced cells. However, this protective effect was not apparent upon apoptotic induction by DOX. When KLF9 was knocked down, however, the MDA-MB-231 cells were overall less susceptible to DOX-induced apoptosis (Fig. 6B).
Finally, the role of KLF9 in cell migration was evaluated through the scratch-wound assay. In MCF7 cells, CORT treatment did not affect migration while KLF9 knockdown in vehicle-treated cells slightly enhanced wound closure rates (Fig. 6C). In MDA-MB-231 cells, CORT treatment augmented the migration rate of scrambled shRNA-transduced cells. KLF9 knockdown resulted in a similar magnitude of increase in vehicle-treated cells but this was not further enhanced with CORT treatment (Fig. 6D).