MYC regulation of a “poor-prognosis” metastatic cancer cell state

Results

Although these 13 “poor-prognosis” signatures were derived in various ways, many without any relation to breast cancer, most are prognostic across different breast tumor expression data sets. A few of these signatures also are coordinately expressed in the same high-risk human breast tumors (13). Importantly, many poor-prognosis signatures contain genes whose expression correlates with estrogen receptor (ER) status in breast cancer, and estrogen signaling is known to drive the growth and survival of breast cancer cells from the earliest stages of disease (14). These observations led to the hypothesis that estrogenic signaling may regulate the expression of both signatures derived from breast tumors and other diverse signatures derived in non–breast cancer contexts. To test this hypothesis, we assayed the expression of the 13 poor-outcome tumor signatures using Affymetrix microarrays in estrogen-dependent ERα⁺/β⁻ MCF7 breast cancer cells after experimental manipulation of estrogenic signaling (15).

Detailed statistical analysis revealed that a significant proportion of genes in each of these 13 different transcriptional signatures was indeed expressed by MCF7 cells and was tightly and coordinately regulated by signaling through the ER (Fig. 1B , Figs. S2 and S3, and Table S2). Different signatures overlapped by no more than 10% with respect to individual component genes (Fig. S1). Furthermore, even this minimal overlap between different signatures could not account for the coordinate pattern that we observed, which was far beyond chance based on permutation analysis using both random and unrelated, experimentally derived signature sets (Table S3). Overall, approximately 40% of all poor-outcome signature genes were estrogen-regulated in MCF7 cells, with proportions in individual signatures ranging from around 30% to nearly 100% (Table S2). These initial findings in a cancer cell–autonomous experimental system demonstrated that the ER can functionally regulate the coordinate expression of a majority of poor-prognosis signatures.

Beyond ER, we also observed similar transcriptional effects with other oncogenes. For example, ERBB2 overexpression and EGF stimulation in MCF10A breast epithelial cells, which are immortalized but untransformed, both induced coordinated expression of all 13 poor-outcome signatures (Fig. 1 C and D ), and stimulation of LNCaP prostate cancer cells with dihydrotestosterone (DHT) via the androgen receptor (AR) produced a similar transcriptional pattern, albeit somewhat more weakly than ERBB2 or EGF signaling with respect to the “DOWN” metagenes from different signatures (Fig. 1E ). These results suggested that increased activity of four different oncogenic pathways known to independently regulate cancer cell growth and survival in different cellular contexts could push different epithelial cell types toward a common transcriptional state defined by the coordinate expression of 13 different poor-outcome signatures.

We next investigated whether poor outcome signature regulation reflected direct or indirect effects of these oncogenes on transcription. ERBB2 and EGFR, as membrane-bound receptor tyrosine kinases, almost certainly induced poor-outcome signature expression indirectly by signaling through downstream transcription factors. In contrast, ERα and AR are nuclear hormone receptors whose transcriptional activity can be modulated by the EGFR and ERBB2 pathways, raising the possibility that these transcription factors might directly regulate poor-outcome signature expression. But treatment of MCF7 cells with 17β-estradiol in the presence of cycloheximide (CHX), a general inhibitor of new protein translation, destroyed coordinated signature induction by estrogen (Fig. 1F ) (16). These results demonstrated that ERα might modulate poor-outcome signature expression indirectly, at least in part through a common transcriptional regulator downstream of the ERα, ERBB2, EGFR, and AR pathways.

Consequently, we analyzed the ERα-, ERBB2-, EGFR-, and AR-regulated transcriptional profiles in further detail, to identify candidate genes from the 13 poor-outcome signatures whose expression was commonly induced in response to these four different stimuli. Overall, approximately 50% of the poor-outcome signature genes were regulated by these different pathways across different cellular contexts. Although some poor-outcome genes were regulated uniquely by one of these pathways, others were commonly regulated by multiple pathways. In particular, ERα, ERBB2, EGFR, and AR commonly induced the expression of a “core gene set” of 20 poor-outcome genes, which was far beyond random (P = 1.6 × 10⁻¹²) (Fig. 2A ).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

MYC-centered core interactome. (A) Venn diagram illustrating derivation of the core gene set. (B) A highly connected network of the 20 genes in the core gene set (orange) and 13 associated genes identified using the IPA algorithm (17). MYC is highlighted in red. (C) Statistical analysis of genome-wide MYC-binding site data obtained from MCF7 cells after stimulation with 17β-estradiol, generated using ChIP-Chip (chromatin immunoprecipitation and tiling microarrays spanning the entire nonrepetitive human genome) (26). UP and DOWN poor-outcome signature genes with at least one MYC binding site within ± 5 kb of their transcription start site (orange) versus the same for all other genes on the tiling array (green).

Although this core gene set did not contain any obvious candidate transcriptional regulators, all of the genes in this set were linked into a single and highly-connected “core interactome” using the IPA algorithm and database of literature-based molecular interactions (Fig. 2B ) (17). This core interactome was highly connected to the MYC transcription factor, which was not a member of any poor-outcome signature and thus was not in the core gene set itself. This correlation fit with the observation that at least 10 of 20 core interactome genes are known direct MYC targets (www.myccancergene.org). It also was consistent with previous reports implicating MYC as a regulator of poor-prognosis signature expression in breast cancer (18 –22). Taken together, this evidence raised the hypothesis that MYC, which is downstream of the ERα, ERBB2, EGFR, and AR pathways, may act as a common transcriptional regulator for expression of the 13 different poor-outcome signatures that we studied.

Consistent with this hypothesis, overexpression of MYC in MCF7 cells was sufficient to induce coordinated poor-prognosis signature expression (Fig. 1G ) (23). In contrast, overexpression of ZIP (a MYC mutant lacking the two N-terminal transactivation domains) in MCF7 cells failed to induce this expression pattern (Fig. 1H ). Furthermore, siRNA knockdown of endogenous MYC in MCF7 (ER⁺) reversed this coordinated pattern of signature expression (Fig. 1I ) (24, 25). Overall, these transcriptional changes involved 20% of all poor-outcome signature genes in MCF7 cells, ranging from 10% to 40% across individual signatures (Table S4). In addition, analysis of genome-wide MYC chromatin immunoprecipitation microarray (ChIP-Chip) profiling data, obtained from MCF7 cells after 17β-estradiol treatment (26), demonstrated a highly significant enrichment for MYC binding within 5 kb of the transcriptional start sites of up-regulated poor-outcome signature genes (P = 6 × 10⁻⁸) (Fig. 2C ). In contrast, we found no similar enrichment in MYC binding for down-regulated poor-outcome genes, suggesting that many down-regulated genes are in fact indirect MYC targets. Finally, siRNA knockdown of endogenous MYC in MDA-MB-231 (ER⁻PR⁻ERBB2⁻) breast cancer cells also reversed coordinated signature expression (Fig. 1J ). These results demonstrated that MYC could powerfully regulate the coordinate expression of many, but not all, genes from 13 different poor- outcome transcriptional signatures in breast cancer cells.

MYC is known to regulate a wide spectrum of cancer cell behaviors, including proliferation, survival, differentiation, and genetic stability, which are general features of tumorigenesis (27). Therefore, one explanation for our findings was that MYC activity within cancer cells, as reflected by poor-prognosis signature expression, might indirectly influence the metastatic propensity of cancer cells (by, e.g., promoting cellular proliferation and survival). Another possibility was that MYC actually might regulate more specific aspects of metastatic cell behavior. To test this hypothesis, we focused on the MDA-MB-231 cell line, which is highly metastatic and stereotypically forms distant lung metastasis when injected into immunocompromised mice. This well-credentialed model has been used to experimentally identify and manipulate molecular pathways that regulate various aspects of metastatic cell behavior (28). Furthermore, MYC knockdown in these metastatic breast cancer cells resulted in the coordinate down-regulation of 13 different poor-outcome transcriptional signatures, supporting the relevance of this model system to human cancer metastasis (Fig. 1J ). We therefore investigated whether inhibiting MYC activity would specifically inhibit the metastatic behavior of MDA-MB-231 cells.

Stable MYC knockdown with two different short-hairpin RNAs introduced using a lentiviral vector [hairpin 1 (HP1) ≈95%; hairpin 2 (HP2) ≈ 70%] did not significantly alter the growth kinetics of MDA-MB-231 cells after serial passage in vitro (Fig. 3 A and B ). It must be noted, however, that immediately after MYC knockdown, HP1 and HP2 cells did show mildly decreased proliferation that then normalized after 1–2 weeks of passage, suggesting that these cells depended on MYC for proliferation to some degree but were still able to preserve their proliferative drive in the face of stable MYC inhibition, presumably through compensatory changes. Importantly, stable MYC knockdown also did not significantly affect the kinetics of primary tumor formation after orthotopic injection into the mammary fat pad of immunocompromised NOD/SCID mice (Fig. 3C ). Nine of 10 control mice and 8 of 10 experimental mice had similar-sized tumors 8 weeks after injection, whereas the remaining animals did not form tumors.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

MDA-MB-231 primary tumorigenesis and metastasis in vivo with stable MYC knockdown. (A) Western blot for MYC protein after stable, lentiviral-mediated knockdown with two different MYC hairpins (HP1 and HP2) compared with pLKO empty vector control after serial passage in MDA-MB-231 cells. (B) Proliferation curves for empty vector control (pLKO), HP1, and HP2 cells after serial passage over 4 days, showing absorbance as a measure of cell number (n = 3, 95% CI). (C) Tumor growth of control (pLKO) and HP1 cell xenografts in NOD/SCID mice (pLKO, n = 9; HP1, n = 7; 95% CI). (D) Boxplot of the number of lung metastases per mouse in control and MYC HP1 knockdown tumor-bearing mice (pLKO, n = 9; MYC HP1, n = 7; P value calculated from the two-sided, two-sample Wilcoxon test). (E) Representative immunohistochemistry for MYC protein expression in primary and metastatic tumor sections from the outlier animal in the MYC HP1 knockdown group. (F) Representative light microscopy (H&E) and immunohistochemistry (MKI67, TUNEL, vimentin, and e-cadherin) in pLKO control and MYC HP1 knockdown primary tumors. (Scale bars: 20 μm.)

In contrast, MYC knockdown greatly inhibited the formation of distant lung metastasis (Fig. 3D ). Whereas two animals had no lung metastases, four other animals had a small number of metastases compared with control animals at 8 weeks. One other MYC knockdown animal died of unknown causes at 5.5 weeks and so was not fully evaluable, but only had a small primary tumor and no signs of macroscopic metastasis at autopsy. Curiously, the last MYC knockdown animal was an outlier and was similar to animals in the control population in forming numerous lung metastases (Fig. 3D ). To address this paradoxical finding, we examined MYC protein expression in the primary and metastatic tumors from this outlier by immunohistochemistry. Whereas the primary tumor was negative for MYC expression, confirming stable knockdown in cells at the point of orthotopic injection, metastatic cells in the lungs of this animal were strikingly positive for MYC expression, suggesting that rare cells had somehow reactivated MYC expression, enabling metastasis of these cells from the mammary fat pad to the lungs (Fig. 3E ). Further histopathologic examination of primary tumors from MYC knockdown versus control animals revealed that both tumor types were high-grade carcinomas containing round polygonal cells with comparable levels of MKI-67, TUNEL, vimentin, and E-cadherin staining, which mark cell proliferation, apoptosis, and mesenchymal/epithelial cell state, respectively (Fig. 3F ). Taken together, these in vivo results demonstrated that MYC activity was specifically required to maintain the metastatic state rather than the proliferative, apoptotic, or differentiation state of these highly metastatic breast cancer cells.

Given these findings in vivo, we finally examined whether MYC activity was necessary for fundamental and hallmark features of metastatic cell behavior, namely, cancer cell invasion and migration. We used a novel microfluidic device developed to study cell migration and invasion at the single-cell level, where locomotion is mechanically constrained along predefined paths (29). Each device contained an array of linear microcapillaries each 10 μm tall, 20 μm wide, and 600 μm long. Individual MDA-MB-231 cells were visualized using live-cell time-lapse videomicroscopy after seeding into microcapillaries either filled with Matrigel (to simulate some of the three-dimensional microenvironmental conditions that cancer cells encounter in vivo) or simply coated with collagen IV (to study unimpeded migration). We specifically performed these experiments in the absence of a chemogradient, which allowed us to examine intrinsic cell behavior. Quantitative analysis of single-cell velocities in this device revealed that MYC knockdown greatly impeded MDA-MB-231 invasion in Matrigel but had only mild effects on migratory velocities compared with control cells (Fig. 4 A–F , and Movies S1–S4). As was most evident in the time-lapse movies of migrating cells, however, knockdown cells were rounded, moved less efficiently and persistently, exhibited frequent and disordered changes in direction compared with control cells, and did not exit microcapillaries efficiently (Movies S3 and S4). We also assessed the quantitative effect of MYC knockdown on MDA-MB-231 cell invasion in more traditional Boyden chamber assays. We found that MYC knockdown in this experimental system led to a ≈50% decrease in invasion (Fig. S4). These findings demonstrated that in highly metastatic MDA-MB-231 breast cancer cells, MYC is necessary to specifically maintain an invasive and migratory state.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

MDA-MB-231 invasion and migration in vitro with stable MYC knockdown. (A) Representative light microscopy images of control (pLKO) and MYC knockdown (MYC HP1) MDA-MB-231 breast cancer cells migrating through Matrigel-filled microcapillaries. (B) Position of individual pLKO and MYC HP1 cells invading through Matrigel-filled microcapillaries over 24 h plotted against time. (C) Bar plot representing the average invasion speed through Matrigel for pLKO and MYC HP1 cells, 13.9 ± 1.1 μm/h for pLKO (n = 33) and 9.1 ± 0.9 μm/h for MYC HP1 (n = 31) and 10.9 ± 0.7 μm/h for MYC HP2 (n = 33). P values were computed by the two-sided Walsh t test. (D) Representative light microscopy images of pLKO and MYC HP1 MDA-MB-231 breast cancer cells migrating through collagen IV–coated microcapillaries. (E) Position of individual pLKO and MYC HP1 cells migrating along collagen IV–coated microcapillaries over 12 h plotted against time. (F) Bar plot representing the average migration speed through collagen-coated microcapillaries for pLKO and MYC HP1 knockdown cells, 82.5 ± 7.5 μm/h for pLKO (n = 17) and 70.4 ± 6.4 μm/h for MYC HP1 (n = 21). P values were computed by the two-sided Walsh t test. Also see Movies S1–S4.