Research ArticleCell Biology

The depalmitoylase APT1 directs the asymmetric partitioning of Notch and Wnt signaling during cell division

See allHide authors and affiliations

Science Signaling  02 Jan 2018:
Vol. 11, Issue 511, eaam8705
DOI: 10.1126/scisignal.aam8705

Lipid modification promotes cancer cell heterogeneity

Asymmetric cell division generates daughter cells that have distinct fates and is accomplished through the unequal distribution of cell fate determinants or signaling pathway components. Stypulkowski et al. found that the depalmitoylase APT1 was asymmetrically localized in dividing cancer cells, and its catalytic activity was required for asymmetric localization of the Notch antagonist Numb and the Wnt signaling mediator β-catenin, both of which are palmitoylated. The polarity complex component CDC42 was required for the asymmetric localization of both APT1 and β-catenin. APT1-mediated asymmetric partitioning of Numb and β-catenin restricted Notch and Wnt signaling to one daughter cell or the other, induced a mammary stem cell transcriptional signature in breast cancer cells, and promoted heterogeneity and self-renewal capacity in breast cancer cells. These results identify palmitoylation-dependent asymmetric partitioning of cell fate determinants as a potential driver of tumor cell heterogeneity, which has been associated with tumor progression and metastatic potential.

Abstract

Asymmetric cell division results in two distinctly fated daughter cells. A molecular hallmark of asymmetric division is the unequal partitioning of cell fate determinants. We have previously established that growth factor signaling promotes protein depalmitoylation to foster polarized protein localization, which, in turn, drives migration and metastasis. We report protein palmitoylation as a key mechanism for the asymmetric partitioning of the cell fate determinants Numb and β-catenin through the activity of the depalmitoylating enzyme APT1. Using point mutations, we showed that specific palmitoylated residues on Numb were required for its asymmetric localization. By live-cell imaging, we showed that reciprocal interactions between APT1 and the Rho family GTPase CDC42 promoted the asymmetric localization of Numb and β-catenin to the plasma membrane. This, in turn, restricted Notch- or Wnt-responsive transcriptional activity to one daughter cell. Moreover, we showed that altering APT1 abundance changed the transcriptional signatures of MDA-MB-231 triple receptor–negative breast cancer cells, similar to changes in Notch and β-catenin–mediated Wnt signaling. We also showed that loss of APT1 depleted a specific subpopulation of tumorigenic cells in colony formation assays. Together, our findings suggest that APT1-mediated depalmitoylation is a major mechanism of asymmetric cell division that maintains Notch- and Wnt-associated protein dynamics, gene expression, and cellular functions.

INTRODUCTION

Asymmetric cell division yields two morphologically and functionally distinct daughter cells and serves as a major contributor to cellular heterogeneity during development and tissue homeostasis (1). In dividing stem and progenitor cells, cell fate determinant proteins are unequally segregated along the division axis and inherited by one cell, resulting in the differential activation of transcriptional networks that establish nonidentical daughter cells (25). For example, Drosophila melanogaster neuroblasts divide asymmetrically to produce a self-renewing neuroblast and a differentiating cell (6, 7). Similarly, transformed cells can also exhibit cellular heterogeneity with variations in properties such as signaling activity, tumorigenicity, and drug resistance exhibited by distinct cell populations (810). The cause of tumor heterogeneity has generally been attributed to genomic instability, epigenetic alterations, or interactions with the tumor microenvironment (1113), although it is possible that asymmetric cell division may also play a role.

The molecular mechanisms driving and maintaining asymmetric divisions are poorly understood, but developmental signaling pathways, such as Notch and Wnt, have been shown to be key factors. In Drosophila neuroblasts, polarized cell division results in asymmetric Numb (the Notch antagonist) localization at the plasma membrane, resulting in unequal inheritance by cells fated to differentiate into neurons (6, 7, 1416). Numb is also partitioned asymmetrically in dividing mammalian cells such as mammary epithelial precursors, hematopoietic stem and progenitor cells, and T lymphocyte precursors (2, 17, 18). Likewise, directionally applied Wnt signals restrict β-catenin (a critical intracellular mediator of canonical Wnt signaling) in a polarized manner to mouse embryonic stem cells and Caenorhabditis elegans seam cells that are fated to remain as progenitor cells (4, 19). Differential spatial organization of proteins and the resulting cell polarity are established by the evolutionarily conserved Par-aPKC (atypical protein kinase C)–CDC42 complex (2022). CDC42 is a small Rho-family guanosine triphosphatase (GTPase) that mediates polarized processes such as vesicle budding, trafficking, and directional cell migration by remodeling the actin and microtubule cytoskeleton networks (23). Although the CDC42 polarity complex is involved during asymmetric cell division, the mechanisms directing the asymmetric recruitment and retention of cell fate determinant proteins at the plasma membrane during cell division remain unclear.

Posttranslational lipid modification is a prevalent mechanism of targeting proteins to membranes that alters molecular conformations and increases protein affinity for hydrophobic environments (24). Lipid modifications are largely nonreversible and maintain protein enzymatic function and stability, protein-protein interactions, and intracellular trafficking (25). In contrast, palmitoylation is a reversible lipid modification that modulates subcellular polarity and signaling cascade activity, such as that of the epidermal growth factor receptor (EGFR) and the small GTPase Ras, by rapid protein shuttling between the cytosol and lipid rafts within the plasma membrane (2632). Palmitoylation is modulated by two classes of enzymes: palmitoyltransferases that add palmitate to cysteine residues and depalmitoyltransferases that remove palmitate at the membrane and promote protein localization to the cytosol (33). Thus, palmitoylation is an attractive, but untested, candidate mechanism for the dynamic asymmetric targeting of proteins to the plasma membrane during cell division to drive cellular heterogeneity in embryogenesis, development, and tumorigenesis.

Here, we investigated the role of palmitoylation on the polarized partitioning of cell fate determinants during asymmetric cell division and clarified how palmitoylation contributes to cellular heterogeneity. Using human cancer cell lines derived from malignant tumors that exhibit high cellular heterogeneity, such as triple receptor–negative breast cancers and osteosarcomas (3438), we found that the depalmitoylating enzyme acyl-protein thioesterase 1 (APT1) (3941) directs the asymmetric localization of Numb and β-catenin. In addition, we found that APT1 activity during mitosis restricted Notch-, Wnt-, and Sox2-dependent transcription to one daughter cell. APT1 also maintained the gene expression signature and colony-forming potential of transformed cells. Moreover, our observations indicate that APT1 was required for the transcriptional output of different signaling pathways in cancer cells grown in serial colony formation assays, which correlates with cellular heterogeneity and tumorigenic potential. With these findings, we identify a palmitoylation-mediated mechanism of asymmetric cell division that promotes the generation of functionally heterogeneous cells in tumors.

RESULTS

Activity of the depalmitoylating enzyme APT1 is required for asymmetric localization of Numb and β-catenin

We have previously shown that APT1 promotes the transient and asymmetric localization of cell adhesion molecules during interphase in response to extracellular signals (26). To test the hypothesis that APT1 directs the asymmetric localization of Notch signaling– and Wnt signaling–associated cell fate determinants during cell division, we examined the spatial organization of Numb and β-catenin in fixed cells. Using the MDA-MB-231 human triple receptor–negative breast cancer cell line, dividing cells were identified by immunostaining for acetylated tubulin, a marker of stabilized tubulin structures such as the mitotic spindle and cytokinetic midbody (42), and counterstained for Numb or β-catenin. In the absence of exogenous stimuli, we observed symmetric localization of both proteins (Fig. 1, A and B), as assessed by measuring the percentage difference in the mean fluorescence pixel intensity of Numb or β-catenin across dividing cells (fig. S1A). The distribution of the percentage differences of all quantified cells was plotted, and cells with a percentage difference of 20 or greater were scored as asymmetric (Fig. 1, C and D). Treating cells with Palmostatin B (PalmB), a pharmacological inhibitor of APT enzymes (43), reduced the asymmetric localization of Numb by threefold (29.4% versus 9.9%) and β-catenin by 3.3-fold (26.1% versus 8.0%) (Fig. 1, E and F). These findings suggest that APT enzymes are required for establishing asymmetric localizations of Numb and β-catenin in dividing cells.

Fig. 1 Activity of the depalmitoylating enzyme APT1 is required for asymmetric localization of Numb and β-catenin.

(A and B) Images of dividing MDA-MB-231 cells stained to show endogenous Numb (A) and β-catenin (B) in red, acetylated tubulin in green, and nuclei in blue. Arrowheads indicate asymmetric localization of Numb and β-catenin. Scale bars, 15 μm. (C and D) Distribution dot plots showing the difference in mean fluorescence pixel intensity of endogenous Numb (C) and β-catenin (D) across dividing MDA-MB-231 cells. The distribution of the percentage differences of all quantified cells was plotted, and cells with a difference of >20% (black dotted line) were scored as asymmetric. n = 508 to 582 cells scored for each experimental group. Each dot represents a single cell. Asterisks indicate statistically significant differences between the indicated groups. (E and F) Quantification of dividing MDA-MB-231 cells showing asymmetric Numb (E) and β-catenin (F) localization after treatment with Palmostatin B (PalmB) or dimethyl sulfoxide (DMSO). (G and H) Quantification of the number of dividing MDA-MB-231cells showing asymmetric Numb (G) and β-catenin (H) localization when APT1 was knocked down with shAPT1 and when wild-type APT1 (APT1WT) or the catalytically inactive APT1S119A mutant was coexpressed with shAPT1. Cells expressing a scrambled (Scr) short hairpin RNA (shRNA) sequence were used as a negative control for APT1 knockdown, and cells expressing the empty vector were used as a negative control for the APT1 rescue experiments. (I and J) Quantification of the number of dividing MDA-MB-231 cells showing asymmetric Numb (I) and β-catenin (J) localization when DHHC20 was knocked down with shDHHC20. (K to M) Immunoblots showing biotin-labeled Numb (K), β-catenin (L), and ERK (M) in MDA-MB-231 cell lysates after acyl-biotin exchange (ABE) assays and pulldown on streptavidin beads (PD). Cells were grown in the presence of either PalmB or vehicle control (DMSO). Input lanes show cell lysates before pulldown. Samples without hydroxylamine (−HAM) were negative controls for the ABE reactions. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001, t test (E and F) or analysis of variance (ANOVA) (C, D, and G to J). Error bars indicate SD.

To address a specific requirement for APT1, we knocked down APT1 with a short hairpin RNA (shRNA) (fig. S1B). In APT1 knockdown cells, asymmetric localization of Numb was reduced by 6-fold (29.4% versus 4.9%) and asymmetric localization of β-catenin was reduced by 4.1-fold (26.1% versus 6.4%). Ectopic expression of wild-type human APT1 (APT1WT) from a plasmid restored asymmetric localization of Numb and β-catenin to baseline control conditions (Fig. 1, G and H). However, ectopic expression of a catalytically inactive mutant form of APT1 [Ser119 in the catalytic domain is mutated to Ala (APT1S119A) (40, 41)] failed to rescue asymmetric Numb and β-catenin localization (Fig. 1, G and H, and fig. S1B). This indicates that the catalytic activity of APT1 is critical for asymmetric Numb and β-catenin localization.

Of the 23 DHHC palmitoyltransferases found in mammalian cells, DHHC20 is one of three palmitoyltransferases that are localized to the plasma membrane, where we would expect it to be available to palmitoylate proteins such as Numb and β-catenin (44). DHHC20 is also known to be expressed in MDA-MB-231 cells, where it palmitoylates EGFR and attenuates EGFR signaling (28). We asked whether loss of DHHC20 would affect protein localization. Knockdown of DHHC20 reduced asymmetric localization of Numb and β-catenin in a manner similar to knockdown of APT1 (Fig. 1, I and J, and fig. S1, C and D).

The dependence of Numb and β-catenin asymmetric localization on APT1 suggests that Numb and β-catenin may be palmitoylated. To evaluate protein palmitoylation, we used acyl-biotin exchange (ABE) assays (45). In the ABE assay, purified proteins are treated with N-ethylmaleimide (NEM) to block free thiol groups, then the Cys-palmitoyl thioester linkages are cleaved with hydroxylamine (HAM), and the newly exposed thiol groups are coupled to biotin. The modified proteins are then purified with streptavidin. Results of ABE assays were consistent with Numb and β-catenin being palmitoylated under normal growth conditions (Fig. 1, K and L). Treating cells with PalmB increased the total amounts of palmitoylated Numb and β-catenin (Fig. 1, K and L), indicating that palmitoylation of these proteins was inhibited, in part, by APT enzymes. As a negative control, ABE assays did not detect palmitoylation of the ERK (extracellular signal–regulated kinase), which is not palmitoylated (Fig. 1M).

To determine whether palmitoylation-dependent asymmetric localization was specific to Numb and β-catenin, dividing cells were immunostained for CD44 and RhoB, both of which are palmitoylated and unrelated to Wnt or Notch signaling (4649), as well as for green fluorescent protein (GFP) in cells expressing a control GFP plasmid. The percentages of cells with asymmetrically localized GFP, CD44, or RhoB were not higher than the background percentages observed for β-catenin and Numb (about 8%) and were unaffected by PalmB treatment (fig. S1, E to J). Together, these findings uncover a direct role for APT1 in the spatial distribution of the palmitoylated cell fate determinants Numb and β-catenin.

Asymmetric localization of Numb requires palmitoylation of the phosphotyrosine binding domain

Although we found β-catenin to be palmitoylated, β-catenin cortical localization is mediated through association with cadherins at tight junctions (50, 51). Thus, the necessity of palmitoylation for the membrane localization of β-catenin is unclear. However, the conserved phosphotyrosine binding (PTB) domain of Numb is required for association of Numb with the plasma membrane and for asymmetric localization of Numb in Drosophila through mechanisms that are still unknown (52). We sought to directly test whether palmitoylation of Numb is required for its asymmetric localization. Using the palmitoylation prediction algorithm CSS-Palm (53) and through identification of solvent-exposed cysteine residues within the PTB domain crystal structure (54), three conserved and potentially palmitoylated cysteine residues (Cys37, Cys160, and Cys165) were identified and mutated to Ala (Fig. 2A). The Numb triple Cys-to-Ala mutant (NumbAAA) showed reduced palmitoylation, as measured by metabolic labeling of cells with palmitic acid azide (Fig. 2B). Endogenous β-catenin was also metabolically labeled with palmitic acid azide, confirming the efficiency of labeling in all reactions (Fig. 2B). These results demonstrate that Numb and β-catenin are continuously palmitoylated in MDA-MB-231 cells.

Fig. 2 Palmitoylation and APT1 activity drive Numb localization.

(A) Sequence comparison of the N terminus of Numb from fruit fly (D. melanogaster), zebrafish (Danio rerio), mouse (Mus musculus), and human (Homo sapiens). The phosphotyrosine domain (PTB) is highlighted in green, and the putative palmitoylated cysteines are highlighted in yellow. Conserved residues are indicated by an asterisk (*). (B) Immunoblot showing transgenically expressed wild-type Numb (NumbWT) or the NumbAAA mutant and endogenous β-catenin in U2 OS cell lysates after purification of palmitoylated proteins. Cells were metabolically labeled with palmitic acid azide or treated with DMSO (vehicle control), and then lysates were subjected to click chemistry to convert the palmitic acid moiety to biotin, pulled down on streptavidin beads, and used for immunoblotting. Input was taken from cell lysates before pulldown. (C) Time-lapse images of dividing U2 OS cells coexpressing NumbWT-YFP (yellow), APT1WT-CFP (blue), and mCherry–Histone H2B (red). Fluorescence pixel intensity was measured along the division axis (dashed line), and the corresponding pixel values of Numb (yellow line) and APT1 (blue line) along the division axis were plotted on graphs. Red arrowheads on images and graphs indicate the peak Numb and APT1 pixel intensity at the membrane or cytokinetic midbody. Time is shown in minutes (min). a.u., arbitrary units. Scale bar, 15 μm. (D) Quantification of the number of dividing U2 OS cells showing asymmetric localization of NumbWT-YFP (black bar) and NumbAAA-YFP (gray bar) when each was coexpressed with shAPT1. Cells expressing a scrambled (Scr) shRNA sequence were used as a negative control for APT1 knockdown. (E and F) Time-lapse images of dividing U2 OS cells coexpressing either APT1WT-CFP (E) or APT1S119A-CFP (F) with mCherry–Histone H2B (red). Fluorescence pixel intensity was quantified as in (C). Scale bars, 15 μm. (G) Quantification of the number of dividing U2 OS cells showing asymmetric APT1WT-CFP or APT1S119A-CFP localization. Cells expressing an empty green fluorescent protein (GFP) plasmid (Vector) were used as a negative control. n = 102 to 143 cells scored for each group from three independent experiments. *P< 0.05 and **P < 0.01, t test and ANOVA. Error bars indicate SD.

Next, we visualized the dynamic localization and segregation of fluorescently labeled Numb and APT1 during cell division by live-cell imaging. Because MDA-MB-231 cells shifted from a flat, spread morphology to a raised, rounded morphology out of the imaging plane during cell division, we instead used U2 OS human osteosarcoma cells, which maintained a consistent rounded morphology within the imaging plane during division. U2 OS cells were transduced to stably express an mCherry–Histone B (H2B) plasmid, allowing for the unambiguous identification of cells undergoing division in real time. Over the course of cell division, cyan fluorescent protein (CFP)–tagged APT1WT (APT1WT-CFP) and yellow fluorescent protein (YFP)–tagged NumbWT (NumbWT-YFP) exhibited highly dynamic asymmetric localization (Fig. 2C). At the beginning of the cell division cycle, NumbWT was concentrated at one end of the cell at the plasma membrane, but as daughter cells formed, the localization of Numb shifted to membrane regions at and near the cleavage furrow. Finally, as daughter cells separated, Numb was partitioned to the plasma membrane of the cell that emerged from the same side of the mother cell to which Numb was initially concentrated. APT1WT cosegregated with Numb to membrane regions and was retained in daughter cells with high Numb signal, as indicated by line-scan analysis of YFP and CFP pixel intensity along the division axis (Fig. 2C and movies S1 to S3, red arrowheads). This suggests that APT1 either responds to the same spatial cues as Numb or directs Numb localization. Alternatively, Numb may direct APT1 localization. YFP-tagged NumbAAA (NumbAAA-YFP) was live-imaged to determine the contribution of Numb palmitoylation on its localization. NumbAAA showed a 1.5-fold reduced asymmetric localization in dividing cells, as compared to NumbWT (Fig. 2D). Finally, knocking down APT1 reduced asymmetric NumbWT localization but did not further reduce the asymmetry of NumbAAA (Fig. 2D). Our results show that the asymmetric partitioning of Numb is actively maintained by a mechanism that requires both APT1-mediated depalmitoylation and palmitoylation of Cys37, Cys160, and/or Cys165 within the PTB domain.

Asymmetric localization of APT1 during cell division requires APT1 catalytic activity

Having determined that APT1 activity is essential for asymmetric localization of Numb and β-catenin, we next examined whether the catalytic activity of APT1 is required for its own asymmetric localization. Immunostaining fixed cells showed that endogenous APT1 was asymmetrically localized in 22.3% of control cells. Expressing the catalytically inactive form APT1S119A reduced this asymmetric localization by 2.4-fold, to 9.4%, whereas expressing APT1WT had no significant effect (fig. S2, A to C). Knocking down DHHC20 also reduced asymmetric APT1 partitioning, suggesting that DHHC20 promotes asymmetric localization of APT1 (fig. S2, B and D).

To gain detailed insights into how APT1 activity promotes and maintains the dynamics of its own asymmetric localization, we compared the spatiotemporal distribution of ectopically expressed APT1WT-CFP versus APT1S119A-CFP by live-cell imaging (Fig. 2, E to G, and movies S4 to S7). In 52.6% of dividing cells, APT1WT-CFP asymmetry was maintained at the plasma membrane through cytokinesis (Fig. 2, E and G, and movies S4 and S5). The asymmetric localization of APT1S119A-CFP was significantly reduced to 20.6% of cells, similar to the asymmetric localization of GFP-vector (16.9%) (Fig. 2, F and G, and movies S6 and S7). In addition, the catalytically inactive mutant APT1S119A-CFP was not discretely localized to the plasma membrane at the start of division and appeared to be stuck at the cytokinetic midbody during cytokinesis (Fig. 2F and movies S6 and S7). The data up to this point demonstrate that asymmetric localization of APT1 requires its catalytic activity and suggest a role for protein depalmitoylation activity at the site of asymmetric protein localization.

Palmitoylating and depalmitoylating enzymes localize asymmetrically with palmitoylated proteins during cell division

To clarify the purpose of APT1 accumulation at sites of asymmetrically localized proteins, we hypothesized that APT1 localizes to regions of high protein palmitoylation. Fixed MDA-MB-231 cells were immunostained for DHHC20, which was asymmetrically partitioned to the same region of a dividing cell as APT1 (Fig. 3A). Knocking down DHHC20 reduced asymmetric APT1 partitioning (fig. S2D). In addition, the distribution of APT1 and DHHC20 overlapped with that of caveolin, a palmitoylated protein (55), at the plasma membrane (Fig. 3, B and C). This would suggest that DHHC20-mediated palmitoylation of substrates could recruit APT1 to the membrane.

Fig. 3 Palmitoylating enzymes and depalmitoylating enzymes are asymmetrically partitioned during cell division.

(A) Images of dividing MDA-MB-231 cells stained to show endogenous APT1 (red), DHHC20 (green), and nuclei (blue). (B and C) Images of dividing MDA-MB-231 cells stained to show endogenous APT1 (B) or DHHC20 (C), caveolin, and nuclei. Asymmetric localization is indicated by arrowheads (A to C). (D) Images of dividing MDA-MB-231 cells treated with PalmB or DMSO and stained to show biotin-labeled palmitoylated proteins (red), acetylated tubulin (green), and nuclei (blue) by ABE immunofluorescence. Samples without HAM (−HAM) were negative controls for the ABE reaction. (E) Distribution dot plots showing the difference in mean fluorescence pixel intensity of biotin-labeled palmitoylated proteins. The distribution of the percentage differences of all quantified cells was plotted, and cells with a difference of >20% (dotted line) were scored as asymmetric. n = 91 to 101 cells scored for each experimental group. Each dot represents a single cell. Asterisks indicate statistically significant differences between the indicated groups. (F) Quantification of the number of dividing MDA-MB-231 cells showing asymmetric palmitoylated proteins after treatment with PalmB or DMSO control. (G) Confocal images of ABE immunofluorescence in nondividing MDA-MB-231 cells showing all palmitoylated proteins (green), APT1 or DHHC20 (red), and nuclei (blue) by ABE immunofluorescence. White dotted boxes indicated magnified areas shown directly below (zoom). Samples without HAM (−HAM) were negative controls for the ABE reaction. Scale bars (including zoom), 15 μm. **P < 0.01, ***P < 0.001, and ****P < 0.0001, ANOVA. Error bars indicate SD.

The presence of both the depalmitoylating enzyme APT1 and the palmitoylating enzyme DHHC20 in plasma membrane–associated domains led us to test whether these regions were also enriched for palmitoylated proteins. Because there are currently no palmitoylation-specific antibodies to visualize the localization of palmitoylated proteins, we modified the ABE assay for immunofluorescence and detected asymmetric palmitoylated protein localization at the cortex of dividing cells (Fig. 3D). This asymmetric localization depended on the activity of APT1 because PalmB treatment decreased the enrichment of palmitoylated proteins at the cortex and reduced the percentage of cells showing asymmetric localization of palmitoylated proteins by 4.1-fold (31.4% versus 7.6%). The immunofluorescence signal generated by the modified ABE assay was specific to palmitoylated proteins; negative control staining of samples in which HAM was omitted from the ABE reaction showed greatly reduced signal and symmetric localization (Fig. 3, D to F). To determine the localization of palmitoylated proteins relative to APT1 and DHHC20 and whether this asymmetric localization was only observed during mitosis, we performed the modified ABE immunofluorescence assay on nondividing cells. Both APT1 and DHHC20 puncta were localized to regions enriched in palmitoylated proteins at membrane ruffles, as assessed by confocal microscopy (Fig. 3G). The data thus far indicate that both APT1 and DHHC20 localize to regions enriched in palmitoylated proteins.

A constitutively active CDC42 mutant promotes asymmetric localization of APT1, Numb, and β-catenin during cell division

We next examined whether APT1-mediated asymmetric protein partitioning functions independently of known polarity-establishing mechanisms. The Par-aPKC-CDC42 polarity complex promotes the asymmetric subcellular distribution of cell fate determinants (15, 56). We knocked down CDC42 and PARD3 (the mammalian homolog of Par3) to assess the requirement of these factors for APT1, Numb, and β-catenin localization (fig. S3, A and B). Asymmetric localization of endogenous APT1 was reduced by 2.6-fold (22.3% versus 8.5%) in CDC42 knockdown cells and by 4.3-fold (22.3% versus 5.2%) in PARD3 knockdown cells (Fig. 4, A and B, and fig. S4C). PARD3 knockdown reduced the asymmetric localization of β-catenin, but not that of Numb (fig. S3, D to G). We next asked whether CDC42 and APT1 double knockdown would completely abolish asymmetric Numb and β-catenin partitioning. The percentage of cells showing asymmetric partitioning of Numb and β-catenin under the double-knockdown condition did not further decrease, as compared to APT1 or CDC42 single knockdown (fig. S3, D, E, H, and I). This suggests that there might be a basal level of asymmetric distribution for both of these proteins that is independent of both APT1 and CDC42.

Fig. 4 The reciprocal interaction between APT1 and CDC42 establishes and maintains asymmetric protein partitioning during cell division.

(A and B) Quantification of the number of dividing MDA-MB-231 cells showing asymmetric localization of APT1 when CDC42 was knocked down with shCDC42 (A) or PARD3 was knocked down with shPARD3 (B). Cells expressing a scrambled (Scr) shRNA sequence were used as a negative control for knockdown. (C and D) Quantification of the number of dividing U2 OS cells showing asymmetric APT1WT-CFP or APT1S119A-CFP in cells expressing constitutively active (CDC42V12 and CDC42F28L) or dominant-negative (CDC42N17) forms of CDC42 (C) or expressing isoforms of CDC42 that are prenylated (CDC42Pren) or both palmitoylated and prenylated (CDC42Palm) (D). (E and F) Time-lapse images of dividing U2 OS cells coexpressing CDC42Palm-YFP (yellow) and either APT1WT-CFP (E) or APT1S119A-CFP (F) (blue) and mCherry–Histone H2B (red). Overlapping CFP and YFP signal in the merge appears green. Fluorescence pixel intensity was measured along the division axis (dashed line), and the corresponding pixel values of CDC42 (yellow line) and APT1 (blue line) along the division axis were plotted. Red arrowheads on images and corresponding graphs mark the peak asymmetric CDC42 and APT1 accumulation at the membrane or cytokinetic midbody. Time is shown in minutes (min). Scale bars, 15 μm. n = 152 to 288 cells scored for each group from four independent experiments. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001, t test (A and B) and ANOVA (C and D). Error bars indicate SD.

We next investigated whether APT1 asymmetric localization depended on CDC42 by altering CDC42 activity. Work from other groups has shown that expression of either guanosine triphosphate (GTP)– or guanosine diphosphate (GDP)–locked mutant forms of CDC42 can interfere with CDC42-dependent cellular processes (20, 23, 5760). We used live-cell imaging of APT1WT-CFP and YFP-tagged CDC42 mutants to test the function of CDC42 activity on APT1 localization. Cells expressing a GTP-locked, constitutively active CDC42 mutant (CDC42V12) or a GDP-locked, inactive CDC42 mutant (CDC42N17) showed reduced asymmetry of APT1WT during cell division, as compared to cells expressing a control GFP-vector (Fig. 4C). In fixed cells, knocking down CDC42 and expressing shRNA-resistant CDC42V12 or CDC42N17 also significantly reduced asymmetric APT1 and β-catenin localization. Compared to CDC42 knockdown, which had minimal effect, expressing CDC42V12 or CDC42N17 resulted in a strong reduction of asymmetric Numb localization (fig. S4, A to G).

To demonstrate the requirement of CDC42 cycling activity on APT1 localization in cells, we expressed a YFP-tagged constitutively active CDC42 mutant that retains GTP-GDP cycling (CDC42F28L). CDC42F28L was previously shown by other groups to mimic CDC42-mediated effector binding and subcellular localization (57). Expressing shRNA-resistant CDC42F28L in CDC42 knockdown cells was sufficient to rescue β-catenin and APT1 asymmetric localization but only partially rescued asymmetric Numb localization in CDC42 knockdown cells (fig. S4, A to G). In addition, expression of CDC42V12, CDC42N17, or CDC42F28L did not significantly alter the low percentage of asymmetrically localized catalytically inactive APT1S119A as compared to APT1WT by live-cell imaging (Fig. 4C). These results demonstrate a requirement for the polarity complex and, specifically, CDC42 activity in promoting the asymmetric partitioning of APT1 and β-catenin to the membrane. However, although Numb asymmetric localization required CDC42 activity, overall, it appears to be less dependent on the canonical Par-aPKC-CDC42 polarity complex than APT1 and β-catenin.

The reciprocal interaction between APT1 and palmitoylated CDC42 is sufficient to promote asymmetric protein partitioning during cell division

In addition to cycling between GTP and GDP, CDC42 function is also maintained, in part, by membrane association through a polybasic region and lipid modification of the C-terminal tail (23, 59). There are two known naturally occurring exon splice variants of CDC42 with distinct lipid modifications: a solely prenylated splice variant and a dually palmitoylated and prenylated splice variant (61). We asked whether a CDC42 splice variant could rescue asymmetric protein localization in CDC42 knockdown cells during division by expressing shRNA-resistant constructs of the dually lipid modified isoform (CDC42Palm) or the solely prenylated CDC42 (CDC42Pren) splice variants (fig. S4A) and examining the localization of Numb, β-catenin, and APT1 in fixed cells. In CDC42 knockdown cells, expressing CDC42Palm was sufficient to fully rescue APT1 asymmetry, whereas expression of CDC42Pren appeared to inhibit asymmetric APT1 localization. CDC42Palm and CDC42Pren had little effect on asymmetric Numb localization in CDC42 knockdown cells. CDC42Palm rescued asymmetric β-catenin localization to baseline control conditions, but CDC42Pren had no observable effect (fig. S4, B to D and H to J). These results suggest that palmitoylated CDC42 promotes asymmetric localization of specific downstream proteins, and APT1 may reciprocally promote asymmetric localization of palmitoylated CDC42 during cell division.

To spatiotemporally visualize CDC42 and APT1 localization during cell division, we measured the distribution of APT1WT-CFP and CDC42Pren-YFP or CDC42Palm-YFP by live-cell imaging. CDC42Palm expression increased the percentage of asymmetric APT1WT-CFP in dividing cells by 1.5-fold (52.6% versus 82.7%), whereas CDC42Pren suppressed asymmetric partitioning of APT1WT-CFP by 1.9-fold (52.6% versus 27.2%) (Fig. 4D). In contrast, neither isoform sufficiently altered asymmetric partitioning of catalytically inactive APT1S119A (Fig. 4D). This led us to ask whether APT1 was sufficient to promote asymmetric CDC42Palm localization. Both APT1WT-CFP and CDC42Palm-YFP were asymmetric at the plasma membrane early during division and asymmetrically redistributed to the membrane of one daughter cell upon cytokinesis (Fig. 4E and movies S8 to S10). As expected, expression of APT1S119A inhibited asymmetric redistribution of CDC42Palm-YFP and retained CDC42Palm-YFP at the cytokinetic midbody (Fig. 4F). We next asked whether CDC42Palm was endogenously expressed in MDA-MB-231 cells using exon-spanning primers. We detected the endogenous transcript for this palmitoylated CDC42 splice variant by polymerase chain reaction (PCR) (fig. S4, K and L). Furthermore, we detected CDC42 palmitoylation in U2 OS cells by the ABE assay and demonstrated that treatment with PalmB increased the amount of palmitoylated CDC42, whereas treatment with the palmitoyltransferase inhibitor 2-bromopalmitate decreased palmitoylation (fig. S4M). These results suggest that palmitoylated CDC42 could function with APT1 in these cells either at a basal level or in a subset of cells to promote asymmetric protein localization.

APT1 restricts Wnt and Notch transcriptional activity to one daughter cell

One downstream effect of asymmetrically partitioning cell fate determinants during cell division is the activation of different transcriptional networks in the two daughter cells, resulting in cells with unique transcriptional profiles (3, 4, 6). We hypothesized that APT1 could also mediate the partitioning of asymmetric transcriptional activity of the Notch and Wnt–β-catenin signaling pathways to one daughter cell. We expressed GFP reporter transgenes containing Notch-activated RBPJ (recombination signal binding protein for immunoglobulin κ J region) binding sites (pGF1-Notch) (62, 63) or Wnt-activated TCF and Lef1 binding sites (pGF1-TCF/Lef1) in MDA-MB-231 cells, which were then immunostained for GFP and acetylated tubulin in daughter cells (Fig. 5, A and B). After plotting the distribution of asymmetric divisions as described in fig. S1, we observed asymmetric Notch and TCF/Lef1 reporter signal in 22.7 and 31.7% of daughter cells, respectively. A control GFP reporter containing a minimal CMV promoter (mCMV) lacking pathway-specific promoter elements showed symmetric signal in most cells and confirmed that the observed asymmetries were dependent on the TCF/Lef1 and Notch enhancer elements (Fig. 5, C to H).

Fig. 5 APT1 restricts Wnt and Notch transcriptional activity to one daughter cell.

(A and B) Images of cytokinetic MDA-MB-231 cells stained to show the expression of the pGF1-Notch GFP reporter (A) or pGF-1 TCF/Lef1 GFP reporter (B) (red), acetylated tubulin (green), and nuclei (blue). Arrowheads indicate asymmetric localization. Scale bars, 15 μm. (C and D) Distribution dot plots showing the difference in mean fluorescence pixel intensity of pGF1-Notch reporter (C) or pGF1-TCF/Lef1 reporter (D) across dividing cells. Cells expressing an empty pGF1-mCMV GFP reporter were used as a negative control for the reporters. The distribution of the percentage differences of all quantified cells was plotted, and cells with a difference of >20% (dotted line) were scored as asymmetric. n = 784 to 822 cells scored for each experimental group. Each dot represents a single cell. Asterisks indicate statistically significant differences between the indicated groups. (E and F) Quantification of dividing MDA-MB-231 cells showing asymmetric localization of pGF1-Notch GFP reporter (E) or pGF1-TCF/Lef1 GFP reporter (F) (black bars) after treatment with PalmB or DMSO vehicle control. Cells expressing an empty pGF1-mCMV reporter (gray bars) were used as a negative control for reporter expression. (G and H) Quantification of the number of dividing MDA-MB-231 cells showing asymmetric localization of the pGF1-Notch GFP reporter (G) or pGF1-TCF/Lef1 GFP reporter (H) (black bars) when coexpressed with shAPT1, shCDC42, shAPT1 and shCDC42 (DKD), and shDHHC20. Cells expressing an empty pGF1-mCMV GFP reporter (gray bars) were used as a negative control for reporter expression, and cells expressing a scrambled (Scr) shRNA sequence were used as a negative control for knockdown. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.001, t test (between reporters and pGF1-mCMV) or ANOVA (C to H). Error bars indicate SD.

Pharmacologically inhibiting APT1 with PalmB treatment reduced asymmetric Notch reporter signal by 3.1-fold (31.7% versus 10.2%) and Wnt reporter signal by 2.6-fold (22.7% versus 8.7%) (Fig. 5, E and F). Knocking down APT1 or DHHC20 also reduced asymmetric Notch and Wnt reporter signal, indicating that palmitoylation can restrict Notch- and Wnt-dependent transcription to one daughter cell. Furthermore, knocking down CDC42 significantly disrupted asymmetric Wnt reporter signal, whereas the Notch reporter signal was unchanged, consistent with observations of asymmetric Numb localization (fig. S4). In addition, CDC42 and APT1 may function in the same pathway that stimulates signal-activated transcription because double knockdown of APT1 and CDC42 did not appreciably decrease asymmetric Wnt or Notch reporter signal over APT1 or CDC42 single knockdown (Fig. 5, G and H). Together, these findings suggest that palmitoylation directs the localization of a Wnt activating factor and a Notch inhibitory factor to restrict transcription to one daughter cell.

APT1 induces Wnt, Notch, and mammary stem cell transcriptional signatures in MDA-MB-231 triple receptor–negative breast cancer cells

To date, APT1 has not been shown to drive transcriptional changes in the contexts of development or disease. To further investigate the observed influence of APT1 on asymmetric Notch and Wnt reporter activity, we asked whether APT1 influences a Notch or Wnt gene signature in cells. We performed RNA sequencing (RNA-seq) and gene set enrichment analysis (GSEA) on wild-type control, APT1 knockdown, or APT1WT-expressing MDA-MB-231 cells (table S1). We performed GSEA using a ranked list of differentially expressed genes and compared APT1WT-expressing cells versus control cells, APT1 knockdown cells versus control cells, and APT1WT-expressing cells versus APT1 knockdown cells. In APT1 knockdown cells compared to control cells, we identified a high-scoring signature [NES (normalized enrichment score), 1.46; false discovery rate (FDR) q value, 0.073] that positively correlated with genes reported to increase in cells overexpressing active β-catenin (BCAT_UP.V1_UP), suggesting that APT1 depletion promotes β-catenin signaling (Fig. 6A and fig. S5A). When comparing APT1WT-expressing cells to control cells, we identified a negative correlation (NES, 1.25; FDR q value, 0.302), with a gene signature that includes genes reported to decrease in cells treated with a Notch inhibitor (NOTCH_DN.V1_DN), suggesting that APT1 suppresses Notch signaling (Fig. 6B and fig. S5A). When APT1WT-expressing cells were compared to APT1 knockdown cells, we identified a high-scoring signature (NES, 2.34; FDR q value, 0.0001) for genes reported to increase in mammary stem cells (PECE_MAMMARY_STEM_CELL_UP) (Fig. 6C). Visualization of the leading-edge genes from the mammary stem cell signature also indicated a strong decrease in expression with APT1 knockdown when normalized to control wild-type cells (Fig. 6D).

Fig. 6 Altering APT1 expression changes β-catenin and Notch gene signatures in MDA-MB-231 cells.

(A to C) Gene set enrichment analysis (GSEA) of a top-scoring β-catenin overexpression transcriptional signature in MDA-MB-231 cells when APT1 was knocked down and compared against control cells [false discovery rate (FDR) q value, 0.073] (A), a Notch inhibition signature when APT1WT was overexpressed and compared against control cells (FDR q value, 0.302) (B), and a mammary stem cell signature when APT1WT was overexpressed and compared against shAPT1 cells (FDR q value, 0.0001) (C). (D) Heat map of leading-edge genes obtained from the mammary stem cell signature shown in (C). Data are grouped by APT1 knockdown (kd) cells, APT1WT-overexpressing (oe) cells, and control (wt) cells. (E) Images of cytokinetic MDA-MB-231 cells stained to show the pGF1-SRR2 GFP reporter (red), acetylated tubulin (green), and nuclei (blue). Arrowheads indicate asymmetric localization of pGF1-SRR2. Scale bar, 15 μm. (F) Distribution dot plots showing the difference in mean fluorescence pixel intensity of the pGF1-SRR2 reporter across dividing MDA-MB-231 cells. Cells expressing an empty pGF1-mCMV GFP reporter were used as a negative control for the reporters. The distribution of the percentage differences of all quantified cells was plotted, and cells with a difference of >20% (dotted line) were scored as asymmetric. n = 959 cells scored for the experimental group. Each dot represents a single cell. Asterisks indicate statistically significant differences between the indicated groups. (G) Quantification of dividing MDA-MB-231 cells showing asymmetric localization of the pGF1-SRR2 GFP reporter after treatment with PalmB or DMSO control (black bars). Cells expressing an empty pGF1-mCMV reporter (gray bars) were used as a negative control for reporter expression. (H) Quantification of the number of dividing MDA-MB-231 cells showing asymmetric localization of pGF1-SRR2 GFP reporter (black bars) when coexpressed with shAPT1, shCDC42, shAPT1 and shCDC42 [double knockdown (DKD)], or shDHHC20. Cells expressing an empty pGF1-mCMV GFP reporter (gray bars) were used as a negative control for reporter expression, and cells expressing a scrambled (Scr) shRNA sequence were used as a negative control for knockdown conditions. *P < 0.05, ***P < 0.001, and ****P < 0.0001, t test (between reporters and pGF1-mCMV) or ANOVA (H). Error bars indicate SD.

Two other high-scoring signatures found in APT1WT-overexpressing cells compared to APT1 knockdown cells were signatures for Myc and Cyclin D1 overexpression, both of which are maintained by Wnt and Notch signaling (NES, 2.08; FDR q value, 0.0001; NES, 1.69; FDR q value, 0.014) (fig. S5, B and C). Target genes with high log2FC (fold change) values were validated by quantitative real-time PCR (qRT-PCR). Notable APT1-driven genes included DKK1, BMP4, GATA6, and KLF5 (fig. S5, D and E, and table S1) (6466).

On the basis of the high-scoring gene expression signature for mammary stem cells observed in APTWT-overexpressing cells, it is possible that APT1 asymmetrically restricts signaling pathways that are activated in stem cells. The transcription factor Sox2 has been implicated in mammary stem cell function (67). Expression of a Sox2-responsive reporter (pGF1-SRR2) in MDA-MB-231 cells revealed that its asymmetric signal was reduced with APT1 knockdown or PalmB treatment. DHHC20, CDC42, or APT1 and CDC42 double knockdown reduced the SRR2 asymmetric signal (Fig. 6, E to H). We also observed asymmetric APT1 localization in dividing mouse embryonic stem cells expressing an APT1WT plasmid (fig. S6). These results implicate a role for palmitoylation in promoting cell fate–related transcriptional signatures and maintaining asymmetric cell division in progenitor cells.

APT1 and CDC42 maintain unique cell populations in MDA-MB-231 colonies

Tumors are heterogeneous populations of cells that have various degrees of proliferative and self-renewal potential, and these properties contribute to tumorigenicity (35). Triple receptor–negative breast tumors, such as those from which MDA-MB-231 cells were derived, contain subpopulations of cells that vary in proliferative and self-renewing, tumor-initiating potential (17, 34, 36). Whether asymmetric cell division contributes to the generation of functionally heterogeneous cells in tumors is unclear. We used the colony formation assay, which measures the growth of transformed cells under anchorage-independent conditions and correlates with tumorigenicity. Because only a subset of transformed cells can form colonies, this assay may also indicate the degree of heterogeneity within a cell population (17, 68). Knocking down APT1 reduced the colony-forming potential of MDA-MB-231 cells by 2.3-fold (average colonies counted, 39 versus 90), whereas expressing APT1WT increased colony numbers by 1.7-fold (average colonies counted, 152 versus 90), suggesting that APT1 maintains anchorage-independent growth and tumorigenic potential of transformed cells (fig. S7A).

Self-renewing cells within the colony are expected to indefinitely form new colonies upon serial dissociation and replating. To test whether APT1-mediated asymmetry maintains a self-renewing population within colonies, we examined the self-renewal potential of APT1 knockdown or CDC42 knockdown MDA-MB-231 cells in colonies over three rounds of dissociation and replating. APT1 knockdown significantly reduced the number of colonies on the second and third plating, suggesting depletion of colony-initiating cells (Fig. 7A). Unexpectedly, CDC42 knockdown increased the number of colonies formed with each plating compared to control cells, suggesting an expansion of a colony-initiating cell population (Fig. 7B). Under normal two-dimensional culture conditions, we observed a slight lag in proliferation for APT1 knockdown, CDC42 knockdown, and double-knockdown cells. Because we did not see a significant impairment of proliferation, this result would suggest that the reduced colony-forming potential is not due to decreased proliferation under normal growth conditions (fig. S7, B to D). Colonies from APT1 and CDC42 double-knockdown cells showed reduced replating ability (fig. S7E), again demonstrating that the expansion of this colony-forming population is dependent on APT1.

Fig. 7 APT1 and CDC42 maintain tumorigenic cell populations in MDA-MB-231 colonies.

(A and B) Quantification of the average number of colonies formed from MDA-MB-231 shAPT1 cells (A) or shCDC42 cells (B) over three serial replatings. Cells expressing a scrambled (Scr) shRNA sequence were used as a negative control for knockdown. Each graph shows means taken from three independent experiments. *P < 0.05 and ***P < 0.001, as measured by t test. Error bars indicate SEM. (C) Representative flow cytometry analysis showing gating strategy of CD44+/CD24lo cells (red box) in cells dissociated from colonies or adherent. The population was gated off of ALDH+ cells, as shown in fig. S7. The percentages inside the red box indicate the relative proportion of the CD44+/CD24lo cell population. Cells were stained with phycoerythrin (PE)–conjugated anti-CD24 (CD24-PE) (x axis) and allophycocyanin (APC)–conjugated CD44-APC (y axis). Flow cytometry plots are representative of results from six independent experiments.

Because the increase in colony-forming potential of CDC42 knockdown cells was unexpected, we hypothesized that this could be caused by increased symmetric divisions of colony-initiating cells. A subset of highly tumorigenic cells in basal breast cancers have been shown to reactivate developmental signaling pathways such as those dependent on Notch, Wnt, or Sox2 (34, 6971). Pharmacological inhibition of Wnt (with the Porcupine inhibitor IWP2) or Notch (with the γ-secretase inhibitor compound E) reduced the self-renewal potential of colonies, similar to APT1 knockdown or APT1 and CDC42 double knockdown (fig. S7E). Furthermore, APT1 knockdown increased Wnt reporter signal and decreased Notch and Sox2 reporter signals in colonies (fig. S7F). This is consistent with our GSEA analysis that indicated that APT1 knockdown suppresses Notch signaling and promotes β-catenin signaling and pathways active in mammary stem cells.

Knocking down CDC42 reduced the Notch reporter signal but increased the Sox2 reporter signal in most of the cells within the colony (fig. S7F). The Sox2 reporter has been previously shown to correlate with increased tumorigenicity (71); therefore, these results suggest that the increase in colony formation caused by CDC42 knockdown is due to an increase in Sox2 transcriptionally active cells.

To address the possibility that APT1 maintains a specific subpopulation of tumorigenic cells, we dissociated colonies and sorted the cells by flow cytometry to test for the cell surface marker profile that is associated with highly tumorigenic cells in breast cancers: CD44 high, CD24 low, and aldehyde dehydrogenase high (CD44+/CD24lo/ALDH+) (34). We first examined the size of the ALDH+ population, which was higher in APT1 knockdown cells (16.7%), as compared to control (7.3%) or CDC42 knockdown (7.8%) cells on the first replating (fig. S8). Further gating the ALDH+ population for CD44+/CD24lo cells revealed that knocking down APT1 reduced CD44+/CD24lo cells (2.6%), whereas knocking down CDC42 increased this population (14.4%), as compared to control cells (5.3%) (Fig. 7C). The trend of the number of CD44+/CD24lo/ALDH+ cells being reduced under the APT1 knockdown condition and increased under the CDC42 knockdown condition was sustained on second and third platings (Fig. 7C). This pattern was observed to a lesser extent in adherent cells (Fig. 7C and fig. S8). These results are consistent with the observed colony phenotypes and may explain why CDC42 knockdown cells form many colonies, whereas APT1 knockdown cells are unable to do so. Together, these findings demonstrate that in an anchorage-independent setting, an APT1-CDC42 axis maintains the expansion of a self-renewing, tumorigenic cell population and activation of transcriptional profiles required to maintain this population.

DISCUSSION

Our results uncover a molecular mechanism through which palmitoylation restricts the localization of the cell fate determinants β-catenin and Numb and the downstream transcriptional responses to Wnt and Notch signaling to one daughter cell during cell division. Furthermore, palmitoylation directly promotes asymmetric Numb localization because point mutations in conserved cysteine residues in Numb inhibit not only its palmitoylation but also its asymmetric localization. Inhibition of APT1 reduces the percentage of cells with asymmetrically localized cell fate determinants, indicating that it facilitates a controlled, rather than a stochastic, process. The observation that decreased palmitoylation of the NumbAAA mutant or increased palmitoylation of Numb upon APT1 inhibition blocks asymmetric localization would suggest that it is the dynamic association with the plasma membrane that promotes asymmetry. The enrichment of both the palmitoyltransferase DHHC20 and the depalmitoylating enzyme APT1 to a discrete membrane domain could reinforce this process. APT1 inhibition did not completely block the asymmetric localization of β-catenin and Numb but resulted in a residual asymmetry in about 8% of cells. This is similar to the percentage of cells with asymmetrically localized GFP that does not respond to APT1 inhibition and is most likely independent of palmitoylation. This palmitoylation-independent asymmetric localization could be the effect of previously reported asymmetric partitioning of machinery that maintains protein abundance such as the proteasome (3, 72) or the protein translation machinery (73, 74), which could result in localized differences in degradation or synthesis of proteins during cell division.

Our data show an interdependence of APT1 and CDC42 activity for their own asymmetric localization. This interdependence could occur through the association of two distinct compartments of the plasma membrane: the lipid bilayer and the membrane-associated, cortical actin cytoskeleton. CDC42 establishes polarity in cells by spatially reorganizing actin to direct vesicle traffic to and fuse with the plasma membrane (23, 75, 76). Because active CDC42 is membrane-associated, APT1 could promote the recruitment of palmitoylated proteins that also stimulate CDC42 function, such as the CDC42 activator Intersectin, which is also activated by Numb (77), to the plasma membrane. This is consistent with current models of diffusible, membrane-bound molecules establishing distinct polarized domains at the plasma membrane in a feedforward manner (78, 79). It has also been hypothesized that the formation of polarized domains can be inhibited if the association rate outpaces the dissociation rate or vice versa (78, 80). This could explain why polarized membrane domains are sensitive to changes in CDC42 protein abundance, activity, and lipid modification, as demonstrated by our findings.

Whereas the asymmetric distributions of CDC42 and APT1 appear to be interdependent, the depletion of either protein individually has distinct effects on the Wnt and Notch pathways. Knocking down CDC42 significantly reduces both APT1 and β-catenin asymmetric localization without affecting the localization of Numb. This would suggest that although asymmetric β-catenin and Numb localization require APT1 activity, only the asymmetric localization of β-catenin requires asymmetric localization of APT1. CDC42 is known to promote canonical Wnt signaling activity by sequestering adenomatous polyposis coli (APC), a component of the β-catenin destruction complex, to the plasma membrane in an actin-dependent manner (81, 82). Thus, maintenance of the actin cytoskeleton by CDC42 may be required for asymmetric β-catenin localization, but not for Numb. Hence, the requirement for CDC42 may be necessitated by unique mechanisms of Wnt signal activation distinct from a requirement for APT1 activity.

By manipulating APT1, we demonstrate that we can alter the asymmetric activation of Notch, Wnt, and Sox2 transcriptional reporters without directly altering the expression of transcription factors or upstream signaling factors. The fact that β-catenin and Notch gene signatures were not as high-scoring as compared to those of mammary stem cells, Myc, and Cyclin D1 signatures may indicate that APT1 directs combinatorial signaling to determine cell identity. Myc and Cyclin D1 are canonical downstream transcriptional targets of both Wnt and Notch and are also involved in determining mammary stem cell identity (8385). Within the mammary stem cell signature is a notable abundance of ribosomal proteins. A similar abundance in ribosomal proteins is also present in the Myc signature, consistent with its role in driving ribosomal biogenesis and protein translation in the mammary gland, among other tissues, and in cancer (86, 87). Because Myc is a critical factor for mammary stem cell identity and function (84), our results would suggest that APT1 has a role in translation in mammary stem cells. What remains unclear is whether altering APT1 protein amounts induces the expansion of a population of cells expressing a mammary stem cell transcriptional signature or induces a de novo transcriptional signature in most cells. We also have evidence of asymmetric APT1 localization in dividing mouse embryonic stem cells expressing APT1WT, suggesting that this mechanism may also have a broader and conserved role in development. Future experiments, including single-cell RNA-seq, may address these questions.

Consistent with the concept of changing cell populations, our results suggest that APT1 is required to maintain a tumorigenic population of breast cancer cells, whereas CDC42 appears to restrict the size of this population. Our findings indicate that APT1 may contribute to the activation of transcriptional programs that promote colony-forming potential, whereas CDC42 could restrict APT1 activity to one daughter during asymmetric division. This would be consistent with the increase in the CD44+/CD24lo/ALDH+ population in CDC42 knockdown colonies. The implications of this study may be relevant to human disease because APT1 is amplified in various cancers, and this amplification is associated with poor patient prognosis, suggesting that APT1 could function in the development and progression of human cancers (88, 89). Although the importance of self-renewing, tumor-initiating, or stem-like cells in cancer is still unclear, the question of how tumors maintain and generate cells with diverse properties, such as metastatic potential, dormancy, and drug resistance, is still a critical one. Further exploration of the mechanisms that establish and direct asymmetric cell division and promote cellular heterogeneity may help us understand tumor development and progression.

MATERIALS AND METHODS

Cell culture

MDA-MB-231 and U2 OS cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) + glutamax (catalog no. 10566-016, Thermo Fisher Scientific) and 10% fetal bovine serum. For drug treatment, cells were treated with dimethyl sulfoxide (DMSO) (catalog no. D2650, Sigma-Aldrich), 10 μM PalmB (catalog no. 178501, EMD Millipore) prepared in DMSO, or 5 μM 2-bromopalmitate (catalog no. 21604-1G, Sigma-Aldrich) prepared in DMSO for 16 hours before staining or harvesting for cell lysates. Cells were treated with puromycin (0.5 μg/ml) for selection. E14 mouse embryonic stem cells (ESCs) were cultured in KnockOut DMEM (catalog no. 10829-018, Thermo Fisher Scientific), 15% fetal bovine serum, 1% l-glutamine, 1% penicillin-streptomycin, 1% nonessential amino acids, 0.1 mM 2-mercaptoethanol, leukemia inhibitory factor (1000 U/ml; catalog no. L5158, Sigma-Aldrich), 1 μM MEK1/2 Inhibitor (catalog no. 444966, EMD Millipore), and 3 μM GSK3 Inhibitor XVI (catalog no. 361559, EMD Millipore) on gelatin-coated tissue culture plates.

Stable cell lines

Human embryonic kidney cells were transfected with 0.69 μg/μl of a GAG, Rev, and Vsvg mix and 1.42 μg/μl of the following plasmids: Scramble control, GFP-PRRL, APT1WT-CFP-FLAG-PRRL, APT1S119A-CFP-FLAG-PRRL, FLAG-CDC42Pren-PRRL, FLAG-CDC42V12-PRRL, FLAG-CDC42N17-PRRL, FLAG-CDC42F28L-PRRL, FLAG-CDC42Palm-PRRL, pGF-SRR2-mCMV-GFP-puro (catalog no. SR20071-PA-P, System Biosciences), pGF-Notch-mCMV-GFP-puro (catalog no. TR020PA-P, System Biosciences), pGF-TCF/Lef-mCMV-GFP-puro (catalog no.TR013PA-P, System Biosciences), or pGF-mCMV-GFP (catalog no. TR011PA-1, System Biosciences) for 24 hours with LT-1 Transfection Reagent (catalog MIR2300, Mirus Bio). The aforementioned APT1 and CDC42 plasmids were designed with short hairpin–resistant sequences for rescue studies. Virus was collected 72 hours after infection, with 0.5 to 1 ml of virus used for stable cell line generation. U2 OS cells were infected with APT1WT-CFP-FLAG or APT1S119A-CFP-FLAG lentivirus for 24 hours and then recovered in complete DMEM for 48 hours before cell culture. E14 ESCs were infected with APT1WT-CFP-FLAG lentivirus for 24 hours and then recovered in complete KnockOut DMEM for 48 hours before cell culture. E14 ESCs were infected with APT1WT-CFP-FLAG lentivirus for 24 hours and then recovered in complete DMEM for 48 hours before cell culture. MDA-MB-231 cells were infected with the aforementioned lentivirus for 24 hours and recovered in complete DMEM for 48 hours before cell culture.

Short hairpin design

The following shRNA primers were used: PARD3: 5′-CCGGGCCATCGACAAATCTTATGATCTCGAGATCATAAGATTTGTCGATGGCTTTTTG-3′ (forward) and 5′-AATTCAAAAAGCCATCGACAAATCTTATGATCTCGAGATCATAAGATTTGTCGATGGC-3′ (reverse), 5′-CCGGGCCATCGACAAATCTTATGATCTCGAGATCATAAGATTTGTCGATGGCTTTTTG-3′ (forward) and 5′-AATTCAAAAAGCCATCGACAAATCTTATGATCTCGAGATCATAAGATTTGTCGATGGC-3′ (reverse), and 5′-CCGGAGTCAATTGGATTTCGTTAAACTCGAGTTTAACGAAATCCAATTGACTTTTTTG-3′ (forward) and 5′-AATTCAAAAAAGTCAATTGGATTTCGTTAAACTCGAGTTTAACGAAATCCAATTGACT-3′ (reverse); CDC42: 5′-CCGGCGGAATATGTACCGACTGTTTCTCGAGAAACAGTCGGTACATATTCCGTTTTTG-3′ (forward) and 5′-AATTCAAAAACGGAATATGTACCGACTGTTTCTCGAGAAACAGTCGGTACATATTCCG-3′ (reverse), 5′-CCGGTGCTTGTTGGGACTCAAATTGCTCGAGCAATTTGAGTCCCAACAAGCATTTTTG-3′ (forward) and 5′-AATTCAAAAATGCTTGTTGGGACTCAAATTGCTCGAGCAATTTGAGTCCCAACAAGCA-3′ (reverse), and 5′-CCGGAGATTACGACCGCTGAGTTATCTCGAGATAACTCAGCGGTCGTAATCTTTTTTG-3′ (forward) and 5′-AATTCAAAAAAGATTACGACCGCTGAGTTATCTCGAGATAACTCAGCGGTCGTAATCT-3′ (reverse); APT1: 5′-CCGGTAGGCCTGTTACATTAAATATCTCGAGATATTTAATGTAACAGGCCTATTTTTG-3′ (forward) and 5′-AATTCAAAAATAGGCCTG‑TTACATTAAATATCTCGAGATATTTAATGTAACAGGCCTA-3′ (reverse).

Alignment of Numb

Numb sequences from D. melanogaster (P16554), mouse (Q9QZS3), zebrafish (Q5FBC1), and human (P49757) were chosen from UniProt canonical sequences. Alignment was performed with Clustal Omega. Conserved domains were identified by the Conserved Domain Database (National Center for Biotechnology Information).

Mutagenesis

Site-directed mutagenesis of Numb C37, C160, and C165 to alanine and that of APT1 (S119) to alanine were performed using QuikChange Multi Site-Directed Mutagenesis kit (catalog no. 210515, Agilent). Potential palmitoylated cysteines on Numb were identified using CSS-Palm 3.0 developed by Zhou et al. (53) and analysis of the PTB domain crystal structure. Mutants were sequence-verified by the DNA Sequencing Facility at the Perelman School of Medicine, University of Pennsylvania.

Live-cell imaging

MDA-MB-231 cells were not conducive to studying the dynamics of asymmetric localization at the plasma membrane by live-imaging because these cells frequently divided out of the focal imaging plane. Thus, U2 OS cells stably expressing mCherry–Histone H2B facilitated protein tracking over time and allowed us to study the dynamics of asymmetric localization at the plasma membrane over the course of cell division. U2 OS cells stably expressing APT1WT-CFP-FLAG-PRRL or APT1S119A-CFP-FLAG-PRRL were transfected with 2 μg of YFP-CDC42Pren-PCDNA3.1, YFP-CDC42Palm-PCDNA3.1, YFP-CDC42V12-PCDNA3.1, YFP-CDC42N17-PCDNA3.1, YFP-CDC42F28L-PCDNA3.1, or NUMB-YFP-FLAG-PRRL wild type or mutant for 24 hours using LT-1 (catalog no. MIR2300, Mirus Bio) according to the manufacturer’s protocol. Cells were imaged 48 hours after transfection in Hanks’ balanced salt solution (catalog no. 14175079, Life Technologies) containing 2% fetal bovine serum, glutamine (1 mg/ml), and 20 mM Hepes (pH 7.4) at 37°C. Cells were imaged using the Leica DMI6000 B inverted microscope.

Immunofluorescence

MDA-MB-231 cells were plated on glass coverslips and treated as described. Cells were fixed in 10% formalin; blocked in 5% BSA in tris-buffered saline containing 0.1% Triton X-100 (Roche); incubated in 1:500 primary antibody (β-catenin, catalog no. 9581S, Cell Signaling Technologies), PARD3 (ab64646)/Numb (ab14140)/APT1 (ab91606)/GFP (ab290)/caveolin (ab17052) (1:1000; Abcam), DHHC20 (1:1000; catalog no. HPA014483, Sigma-Aldrich), and acetylated tubulin (1:1000; catalog no. sc23950, Santa Cruz Biotechnology) for 1 to 2 hours at room temperature; incubated in secondary antibody [1:1000; Alexa Fluor 488 goat anti-mouse (A11001)/Alexa Fluor 594 goat anti-rabbit (A11012), Life Technologies] for 1 hour at room temperature; and mounted in DAPI mount (catalog no. 0100-20, SouthernBiotech). ESCs were cultured without LIF (leukemia inhibitory factor), MEK1/2 (mitogen-activated protein kinase kinase), or GSK3 (glycogen synthase kinase 3) inhibitors for 24 hours before staining with 1:1000 anti-GFP (ab290), as described. Cells were imaged using the Leica DMI6000 B inverted microscope on ×40 magnification, and colonies were imaged on ×20 magnification.

Western blotting

A total of 200,000 MDA-MB-231 cells were plated on 60-mm tissue culture dishes and were lysed in tris lysis buffer [50 mM tris (pH 7.4) buffer, 150 mM NaCl, 2% Triton X-100, leupeptin (1 μg/ml), aprotinin (1 μg/ml), and pepstatin A (2 μg/ml)]. Proteins were run out on 10% acrylamide gel, probed with PARD3 (1:1000), CDC42 (1:500) (catalog no. ACD03, Cytoskeleton), Flag (1:500) (catalog no. F3165, Sigma-Aldrich), APT1 (1:500), and DHHC20 (1:1000) at 4°C overnight and then incubated in secondary anti-rabbit horseradish peroxidase (HRP) (catalog no. 211-032-171, Jackson ImmunoResearch) or anti-mouse HRP (catalog no. 115-035-003, Jackson ImmunoResearch) for 1 hour at room temperature. Blots were developed in Pierce ECL Chemiluminescence solution (catalog no. 32106, Thermo Fisher Scientific).

ABE assay

The protocol was adapted from Wan et al. (45). A total of 200,000 MDA-MB-231 cells were plated on 60-mm tissue culture dishes and were harvested by scraping in ABE lysis buffer [50 mM Hepes (pH 7.4), 1% Triton X-100, 150 mM NaCl, 5 mM EDTA, 50 mM NEM, leupeptin (1 μg/ml), aprotinin (1 μg/ml), and pepstatin A (2 μg/ml)]. Lysates were clarified by centrifugation at 15,000 rpm for 10 min and incubated with NEM overnight at 4°C. The samples were methanol/chloroform (m/c)–precipitated twice and then resuspended in 80 μl of 4% SDS buffer. The samples were split in half, and 160 μl of HAM buffer [0.7 M HAM (pH 7.4), 50 mM Hepes (pH 7.4), 0.2% Triton X-100, 150 mM NaCl, and 5 M EDTA] was added to one-half of the sample and control 0.2% Triton X-100 buffer [50 mM Hepes (pH 7.4), 0.2% Triton X-100, 150 mM NaCl, and 5 mM EDTA] was added to the remaining sample and incubated at room temperature for 1 hour. The samples were m/c-precipitated and resuspended in 40 μl of 4% SDS buffer containing 10 μM Biotin-HPDP (catalog no. 21341, Thermo Fisher Scientific). One hundred sixty microliters of 0.2% Triton X-100 buffer + 10 μM Biotin-HPDP was added and incubated at room temperature for 1 hour. The samples were m/c-precipitated and resuspended in 40 μl of 4% SDS buffer containing 10 μM Biotin-HPDP (Pierce). One hundred sixty microliters of 0.2% Triton X-100 buffer + 10 μM Biotin-HPDP was added and incubated at room temperature for 1 hour. The samples were m/c-precipitated and resuspended in 20 μl of 4% SDS buffer, followed by addition of 800 μl of 1% Triton X-100 buffer (50 μl was removed for analysis as “input”). Thirty microliters of streptavidin agarose beads (catalog no. 20349, Thermo Fisher Scientific) was added to the samples and incubated overnight at 4°C while rotating. The samples were washed in 1% Triton X-100 buffer and analyzed by SDS–polyacrylamide gel electrophoresis.

ABE assay for immunofluorescence

Washes were performed in 1× ABE buffer + 0.2% Triton X-100 + 0.1% SDS. Cells were seeded onto coverslips. Cells were fixed in 10% formalin + 50 mM NEM for 10 min at room temperature and washed once before they were incubated overnight in 1× ABE buffer + 0.2% Triton X-100 + 0.1% SDS + 50 mM NEM at 4°C. The following day, three 15-min washes were performed. Cells were incubated in hydroxylamine+ or hydroxylamine buffer for 2 hours at room temperature, and then, three 15-min washes were performed. Cells were incubated in Biotin-HPDP buffer for 1 hour at room temperature, and three 20-min washes were performed. Cells were incubated in primary antibody (biotin, 1:500; catalog no. ab53494, Abcam) at 4°C overnight, and immunofluorescence was performed as described above.

Click chemistry assay for palmitoylation

A total of 200,000 U2 OS cells were plated on 60-mm tissue culture dishes and were transfected with 2 μg of NumbWT or NumbAAA YFP-Flag for 24 hours, labeled with 100 μM palmitic acid azide (catalog no. C10265, Life Technologies). Cells were prepared using the Click-iT Protein Reaction Buffer (catalog no. C10276, Life Technologies) according to the manufacturer’s protocol and analyzed with Western blot, as described above.

Quantification and line-scan analysis

Cells were quantified by drawing around the mitotic spindle poles or around each daughter cell in cytokinesis using the Leica LAS AF software, as shown in fig. S1A. Percent difference was calculated from the mean gray values generated in Leica LAS AF and calculated as described in fig. S1A. The distribution of acquired percentage differences for each experimental condition was plotted as dot plots. Cells with a percentage difference of 20 or greater were counted as asymmetric and plotted in a bar graph. All graphs were generated with Prism software. Line-scan analysis of pixel intensity was performed on still frames from live-cell movies using ImageJ software.

Colony assays

MDA-MB-231 cells were plated on six-well, low-adhesion plates (catalog no. 3471, Corning) in 2 ml of WIT-P + serum-free supplement growth medium (catalog no. 00-0045-500, Cellaria) containing FGF (fibroblast growth factor) (20 ng/ml; catalog no. PHG0024, Life Technologies), epidermal growth factor (20 ng/ml; catalog no. PHG0311, Life Technologies), and heparin (10 μg/ml; catalog no. 07980, STEMCELL Technologies). Each well contained 4000 cells. Cells were treated with 1 μM gama-secretase inhibitor (GSI) (compound E; catalog no. 565790, EMD Millipore), 5 μM IWP2 (catalog no. 72124, STEMCELL Technologies), or DMSO (catalog no. D2650, Sigma-Aldrich) and grown for 7 days before counting. For replating, cells were grown into colonies as described. On day 7, colonies were spun down at 1000 rpm for 5 min, resuspended in 0.05% trypsin-EDTA, reconstituted in growth medium, and plated as described above. The process was repeated for three replatings in duplicate.

Proliferation assays

A total of 15,000 MDA-MB-231 cells were plated on a 12-well dish. At 24, 48, and 72 hours, cells were washed with 1× phosphate-buffered saline, trypsinized, and spun at 1000 rpm for 5 min. Pellets were resuspended in DMEM and counted by a hemocytometer.

RNA isolation and qRT-PCR

Total RNA was isolated from cells using RNeasy Mini kit (catalog no. 74104, Qiagen), and complementary DNA (cDNA) was synthesized from 2000 ng of total RNA using SuperScript III First-Strand Synthesis SuperMix (catalog no. 18080400, Life Technologies). qRT-PCRs were performed in triplicate using standard SYBR green reagents and protocols on a StepOnePlus Real-Time PCR system (Applied Biosystems). Target genes with high log2FC values were chosen for validation. The target mRNA expression was quantified using the ΔΔCt method and normalized to glyceraldehyde-3-phosphate dehydrogenase (GAPDH) expression. The following primers were used for validation: hDKK1: 5′-CAGGCGTGCAAATCTGTCT-3′ (forward) and 5′-AATGATTTTGATCAGAAGACACACATA-3′ (reverse); hFGF5: 5′-CCCAGAATCAGCCCTACAAG-3′ (forward) and 5′-GAGGAGGAAGGACAAGCTCA-3′ (reverse); hGATA6: 5′-GCAAAAATACTTCCCCCACA-3′ (forward) and 5′-TCTCCCGCACCAGTCATC-3′ (reverse); hKLF5: 5′-CTGCCTCCAGAGGACCTG-3′ (forward) and 5′-TCGTCTATACTTTTTATGCTCTGGAAT-3′ (reverse); hITGB4: 5′-TCAGCCTCTCTGGGACCTT-3′ (forward) and 5′-TCCTTATCCACACGGACACA-3′ (reverse); hBMP4: 5′-TCCACAGCACTGGTCTTGAG-3′ (forward) and 5′-TGGGATGTTCTCCAGATGTTCT-3′ (reverse); hPTK7: 5′-CAGAGGACTCACGGTTCGAG-3′ (forward) and 5′-TACCAGGGTCTCTGCCACTC-3′ (reverse); hGAPDH: 5′-ACACCATGGGGAAGGTGAAG-3′ (forward) and 5′-AAGGGGTCATTGATGGCAAC-3′ (reverse); hCDC42-palm: 5′-TGGAGTGTTCTGCACTTACA-3′ (forward) and 5′-GAATATACAGCACTTCCTTTTGGG-3′ (reverse); and hCDC42-prenyl: 5′-AGGCTGTCAAGTATGTGG-3′ (forward) and 5′-TAGCAGCACACACCTGCG-3′ (reverse).

RNA isolation and analysis for RNA-seq

Total RNA was harvested from MDA-MB-231 control, APT1WT-overexpressing, or shAPT1 cells using the RNeasy RNA Isolation kit (catalog no. 74104, Qiagen). One hundred nanograms of RNA was used to generate cDNA libraries with the Illumina TruSeq mRNA Library Prep kit for the NeoPrep Library Preparation system (catalog no. NP-202-1001, Illumina) and analyzed on a NextSeq 500 sequencer. Analysis was prepared by the DNA Sequencing Facility at the Perelman School of Medicine, University of Pennsylvania, as follows: Estimated transcript levels were ranked with Salmon, TX Import was used to condense transcript levels to gene intensity, and DESeq2 was used to calculate statistical levels for each condition. The scaled values (determined by DESeq2) were input into GSEA software (Broad Institute) and analyzed against the C2 chemical and genetic perturbations, C6 oncogenic, and hallmark signatures gene matrix applying classic enrichment statistic. Heat maps were generated by taking the log2 values with an offset of 1 for all conditions, and targets were chosen by taking the leading-edge targets from GSEA BCAT, NOTCH, and mammary stem cell sets. The average of the three control wild-type values was subtracted from each individual value. Values were clustered for samples and genes using Euclidian similarity measure with average linkage.

Polymerase chain reaction

To confirm the expression of FLAG-CDC42 plasmids in shCDC42 MDA-MB-231 cells, we isolated RNA and prepared cDNA from FLAG-CDC42Pren, FLAG-CDC42V12, FLAG-CDC42N17, FLAG-CDC42F28L, and FLAG-CDC42Palm cell lines. PCR reaction was performed using 2 μg of cDNA, 10× PCR buffer (catalog no. P2192, Sigma-Aldrich), 10 mM deoxynucleotide triphosphates (dNTPs) (catalog no. 18252015, Life Technologies), Taq polymerase (catalog no. D1806, Sigma-Aldrich), and 10 μM hCDC42-palm and hCDC42-prenyl primers mentioned above, which are designed to span the alternatively spliced exon. The reaction was run out on a 1.4% agarose gel.

Flow cytometry

Adherent nonconfluent MDA-MB-231 cells or colony-dissociated cells were treated to detect ALDH activity using the ALDEFLUOR assay (catalog no. 01702, STEMCELL Technologies) according to the manufacturer’s protocol. Diethylaminobenzaldehyde was used as a negative control to set ALDH+ gates. Cells were stained for surface markers with CD44-APC (catalog no. 560890, BD Biosciences), CD24-PE (catalog no. 560991, BD Biosciences), and Live/Dead Violet (catalog no. L34963, Thermo Fisher Scientific) for viability. Compensation was performed using UltraComp beads (catalog no. 01-2222-41, Thermo Fisher Scientific). Experiments were run on the Attune NxT Flow Cytometer system (Life Technologies) and analyzed with FlowJo software. ALDH+ cells (fig. S7) were gated for CD44+/CD24lo (Fig. 7C) for comparison.

SUPPLEMENTARY MATERIALS

www.sciencesignaling.org/cgi/content/full/11/511/eaam8705/DC1

Fig. S1. Scoring method for determining asymmetric divisions.

Fig. S2. DHHC20 and APT1 localization in MDA-MB-231 cells.

Fig. S3. Effect of CDC42 and PARD3 knockdown on asymmetric Numb and β-catenin localization.

Fig. S4. CDC42 activity and lipidation promote asymmetric APT1, Numb, and β-catenin localization.

Fig. S5. Validation of RNA-seq and additional GSEA analyses relating to Fig. 6.

Fig. S6. Staining of asymmetric APT1 in mouse embryonic stem cell.

Fig. S7. Colony counts, growth curves, and reporter expression relating to Fig. 7.

Fig. S8. Gating scheme for ALDH+ cells on dissociated colonies or adherent cells.

Table S1. Excel spreadsheet of RNA-seq annotated genes.

Movie S1. Single channel of a U2 OS cell ectopically expressing APT1WT-CFP (blue).

Movie S2. Single channel of a U2 OS cell from movie S1 ectopically expressing NumbWT-YFP (yellow).

Movie S3. Merge of movie S1 of a U2 OS cell ectopically expressing APT1WT-CFP (blue), NumbWT-YFP (yellow), and mCherry–Histone H2B (red).

Movie S4. Single channel of a U2 OS cell ectopically expressing APT1WT-CFP (blue).

Movie S5. Merge of movie S4 of a U2 OS cell ectopically expressing APT1WT-CFP (blue) and mCherry–Histone H2B (red).

Movie S6. Single channel of a U2 OS cell ectopically expressing APT1S119A-CFP (blue).

Movie S7. Merge of movie S6 of a U2 OS cell ectopically expressing APT1S119A-CFP (blue) and mCherry–Histone H2B (red).

Movie S8. Single channel of a U2 OS cell ectopically expressing APT1WT-CFP (blue).

Movie S9. Single channel of a U2 OS cell ectopically expressing YFP-CDC42Palm (yellow).

Movie S10. Merge of movie S1 of a U2 OS cell ectopically expressing APT1WT-CFP (blue), YFP-CDC42Palm (yellow), and mCherry–Histone H2B (red).

REFERENCES AND NOTES

Acknowledgments: We thank C. J. Lengner and D. C. Brady for insightful comments on this manuscript. We also thank J. Tobias and the Molecular Profiling Core at the University of Pennsylvania for analysis of the RNA-seq data and generation of heat maps and GSEA signatures. The results shown here are based (in whole or in part) on data generated by the The Cancer Genome Atlas Research Network (http://cancergenome.nih.gov/). Funding: This work was supported by the NIH (grant R01CA181633) and the American Cancer Society (grants RSG-15-027-01 and IRG-78-002-34). Author contributions: E.S. designed and conducted the experiments. I.A.A. performed the RNA-seq experiment. E.S. and E.S.W. wrote the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: RNA sequencing data are available at GEO (Gene Expression Omnibus) (ID: GSE105486).
View Abstract

Navigate This Article