Global Analysis of mRNA, Translation, and Protein Localization: Local Translation Is a Key Regulator of Cell Protrusions

1A pre-requisite of cell migration is the establishment of front-back cell polarity. In mesenchymal-like cells, the front is protrusive, characterized by a network of newly generated actin filaments, while the back is retractile, pulling up the body of the cell toward the direction of migration. This asymmetry in morphology and function between the front and back has been shown to be associated with differential localization of proteins to protrusions and the cell body, many of which are associated with actin cytoskeleton. Actin is polymerized at the leading edge by actin nucleating factors such as the ARP2/3 complex and formins. Other actin binding proteins, such as capping, bundling, and severing proteins, further regulate actin cytoskeleton organization by affecting the dynamics of existing filaments. In the cell body, myosin II association with actin filaments generates contractile force, which is crucial for adhesion maturation and tail retraction. The signaling that initiates protrusions is relayed by a wide variety of signaling molecules. These include lipid second messengers such as phosphatidylinositol-phosphates, as well as members of the Rho family of small GTPases, in particular RHO, RAC, and CDC42, which trigger and coordinate front-back polarization and spatial organization of the actin cytoskeleton.

2While much is known about the mechanisms that trigger protrusion formation, little is known about how front-back cell asymmetry is maintained in migrating cells. In addition to differential distribution of proteins, mRNAs are also differentially distributed between the front and back of the cells. Cellular mRNAs are known to be transported via microtubule or actin filaments. Once localized, mRNAs could be subject to local translation, which would result in increased local concentration of their corresponding proteins. Importantly, the mechanisms that regulate mRNA targeting and translation in protrusions are poorly defined. Moreover, to what extent mRNA localization and local translation define front-back asymmetry and protrusion formation remains to be determined.

3Here, we assessed how mRNA localization and local translation contributes toward front-back asymmetry at the proteome level. We used a proteomics approach, quantifying the distribution of cellular proteins between protrusions and the cell body by stable isotope labeling of amino acids in culture (SILAC). Using a pulse labeling variation of the SILAC strategy known as pulsed-SILAC, we also determined the relative translation rates of proteins between protrusions and the cell body. Finally, we quantified the relative distribution of mRNAs between protrusions and the cell body by RNA sequencing (RNA-seq). Comparison of relative protein distributions with mRNA distributions showed little correlation between the two, suggesting that mRNA localization alone does not determine protein localization to protrusions. However, relative translation rates between protrusions and the cell body exhibited a significant correlation with protein distributions, suggesting that localized translation is a significant determinant of front-back cell asymmetry. Accordingly, we show that local protein translation in protrusion is crucial for stabilizing protrusions and maintaining a mesenchymal-like polarized morphology. Analysis of mRNA UTR elements revealed several conserved motifs that were enriched in locally translated mRNAs, suggesting that their absence or presence can control local translation rates. To reveal trans-regulators of local translation that were functionally important for protrusions, we screened RBP categories enriched in protrusions by RNAi. We found several RBPs that significantly regulated protrusions and mesenchymal-like migration. These included many known but also a number of novel regulators of mRNA localization and local translation including the exosome core complex. We demonstrate that the exosome core, but not its catalytic subunits, is highly enriched in protrusions and is essential for protrusion stability but not initiation. Our work establishes local translation as a key regulator of asymmetric distribution of the proteome between protrusions and the cell body, describes cis-regulatory UTR motifs that are associated with localized translation in protrusions, and highlights trans-acting RBPs that are involved in regulation of protrusions and mesenchymal-like migration.

4To study cell protrusions, we used a micro-porous filter-based method to separate protrusions from cell bodies of highly protrusive MDA-MB231 breast cancer cells. In this method, cells are seeded on top of collagen-I-coated 3-μm transwell filters to enable protrusions to form but to prevent the cell bodies from passing through due to the size of the pores, resulting in separation of protrusions and the cell bodies on opposite sides of the filter (Figures 1A and 1B). As expected, these protrusions are dependent on Rho-GTPases RAC1 and CDC42 (Figure 1C) and are enriched in known protein markers of protrusive fronts such as VASP, PAX, ZYX, as well as F-actin (Figures 1D and 1E). In terms of spatial and temporal dynamics, protrusion formation through transwell pores seems to be a good mimic of protrusion formation in 3D matrices, where cells have to similarly protrude through matrix pores and holes. In fact, the transwell pore size of 3 μm is comparable to the average pore size of pepsinized collagen-I 3D matrices, and average diameters of protrusions extended through these transwell pores (∼3 μm wide, ∼10–15 μm long, dictated by pore size diameters) are comparable to the diameters of protrusions in 3D collagen-I (Figures 1F and 1G). In addition, the temporal dynamics of protrusions extended through transwell pores are similar to those extended in 3D collagen-I, with protrusions remaining stable for various lengths of time, from minutes to several hours, with a median length of over 2 hr (Figures 1H and 1I; Movies S1 and S2).

5Using the transwell-based fractionation method, we first determined the distribution of cellular proteins between protrusions and the cell body by quantitative proteomics (Figure 2A), revealing the relative distribution of 3,334 proteins from two reciprocally labeled SILAC experiments (Data S1A). Overall, the proteome exhibited a normal-like distribution between protrusions and the cell body, with different proteins being enriched or depleted in protrusions to various levels (Figure 2B). Annotation enrichment analysis revealed multiple protein categories that were significantly enriched or depleted in protrusions (Data S1B). As expected, the majority of the protrusion-enriched proteins belonged to known actin-associated categories (Data S1B). These included upstream signaling regulators of protrusion formation, actin polymerizing factors such as the ARP2/3 complex and its upstream regulator WAVE2, as well as other actin binding and adhesion molecules such as VASP, ENAH, and PXN (Data S1B). On the other hand, the majority of cell body-enriched proteins were those of nucleus, ER, mitochondria, and other organelles (Data S1B). Interestingly, several categories of RNA-binding proteins (RBPs), translation factors, and protein folding chaperones were enriched in protrusions (Data S1B), indicating that a significant level of localized protein synthesis may occur in protrusions.

6In order to determine to what level mRNA localization contributes toward protein localization in protrusions, we quantified the distribution of cellular mRNAs between protrusions and the cell body by RNA-seq (Figure S1A; Data S1C). Similar to the proteome distribution, RNA-seq revealed a normal-like distribution, with different mRNAs being enriched or depleted to different degrees between protrusions and the cell body (Figure 2C). Surprisingly, no correlation was detected between mRNA and protein distributions (Figure 2D), suggesting that mRNA localization alone is not a significant predictor of protein localization to protrusions. As a validation of our method, we compared our RNA-seq dataset against a previously published microarray-based protrusion/cell body mRNA localization study in mouse NIH/3T3 fibroblasts. There was a very strong overlap between the two datasets (Figure S1B), suggesting that RNA localization at the global level is likely to be conserved across species and cell types, despite a lack of correlation with protein distributions.

7Next, we assessed whether local translation contributes toward protein localization to protrusions. We quantified relative translation rates of proteins between protrusions and the cell body during a 2 hr transwell protrusion formation assay by utilizing a variation of the SILAC method called pulsed SILAC (pSILAC) (Figure 2E). We determined the relative translation rates of 1,150 proteins between protrusions and the cell body (Data S1D). The lower coverage compared to the proteome analysis was expected, as the medium and heavy peptides that are used for pSILAC quantification are significantly low in abundance after a short 2 hr labeling. Nevertheless, widespread localized translation was evident, with many cellular proteins displaying higher translation in protrusions than the cell body and vice-versa (Figure 2F). Moreover, we observed a significant correlation between relative translation rate and protein localization (Figure 2G). Pearson’s correlation coefficient between protein and translation rate distributions was 0.30. The actual correlation is likely to be even higher, considering that the Pearson’s correlation coefficients between the two SILAC and pSILAC replicates were 0.72 and 0.66, respectively, presumably due to the inherent errors of the mass spectrometry-based measurements (Figures 2B and 2F). These results demonstrate that localized translation is a key determinant of a significant degree of protein localization to protrusions.

8We next investigated if the contribution of local translation toward protein localization was uniform across all protein categories or whether certain categories of proteins were more reliant on local translation than others for their localization. We performed a 2D annotation enrichment analysis to determine which categories of proteins show a positive correlation between their protein enrichment levels and their relative translation rates as opposed to those that exhibit a negative correlation (Data S1E). All actin-associated protrusion-enriched categories showed a strong positive correlation, having both high protrusion protein levels and high protrusion translation levels (Figure 2H). These categories included ARP2/3 complex subunits, adhesion-related proteins, ezrin, filamins, alpha-actinin, and other actin-binding protrusion proteins such as actin capping and LIM domain proteins (Data S1E). Most organelle-associated protein categories depleted in protrusions also showed a positive correlation, exhibiting both low protrusion protein and low protrusion translation levels (Figure 2H). Collectively, these results suggest that the localization of these protein categories is likely to be determined by their localized translation, either in protrusions or at the cell body. In contrast, RNA-binding and ribosomal protein categories exhibited a negative correlation, having significantly lower protrusion translation rates compared to their protrusion protein levels (Figure 2H). In fact, removal of ribosomal and RNA-binding proteins from our datasets resulted in a significant increase in the overall correlation from 0.30 to 0.47 (Figure 2I). These results suggest that while the localization of most protein categories, including all actin-associated protrusion proteins, can be explained by the differences in their localized translation rates, RNA-binding and ribosomal proteins do not follow this rule, indicating that these proteins are likely to be transported to protrusions instead.

9Our pSILAC data suggest a significant role for local translation in regulation of the asymmetry between protrusions and the cell body at the proteome level. Thus, we predicted that inhibition of local translation in protrusion should inhibit protrusions. We first confirmed the occurrence of localized translation in protrusions by visualizing nascent protein synthesis in protrusion and cell body fractions, using puromycinylation tagging of newly synthesized proteins. Puromycinylation analysis of protrusion and cell body fractions confirmed that protein translation was widespread in protrusions, with overall levels comparable to that of the cell body (Figure 3A). High levels of translation in protrusions could also be revealed by immunofluorescence staining of transwell protruding cells using anti-puromycin staining following puromycinylation (Figure S2A).

10Next, we tested whether protein translation was needed for protrusion formation. Treatment of MDA-MB231 cells with cycloheximide-abrogated protein translation (Figures S2A and S2B). Temporal analysis of protrusion formation revealed that MDA-MB231 cells still initiated protrusions in presence of cycloheximide, but these protrusions were unstable and retracted back (Figures 3B and 3C; Movie S3), suggesting that translation acts to stabilize protrusions rather than to initiate them. Cycloheximide treatment also destabilized protrusions in a 3D collagen-I matrix, resulting in conversion of the cells from a polarized mesenchymal-like morphology to round (Figures 3D and 3E; Movie S4). The rounding observed with cycloheximide treatment was not due to apoptosis (Figure S2C), suggesting that protein translation is needed for protrusion stability.

11To show that it is localized translation that is needed for protrusion stability, we devised a strategy to inhibit translation specifically in protrusions. We used emetine, an irreversible inhibitor of protein translation. Emetine concentrations as low as 1 μg/ml, and treatment times as short as 5 min could strongly abrogate translation in MDA-MB231 cells (Figures S2D and S2E). To locally inhibit translation in protrusions, a low dose of 1 μg/ml emetine was added to the bottom of transwell filters for 5 min. In order to limit emetine diffusion upward, a positive hydrostatic pressure was generated to push liquid down from the top to the bottom of the filter (Figures S2F and S2G), allowing inhibition of translation in protrusions while leaving the cell-bodies unaffected (Figures 3F and S2H). Importantly, this local inhibition of translation also destabilized protrusions (Figures 3G and 3H; Movie S5), demonstrating that local translation is critical for the stability of protrusions.

12Despite the presence of mRNA in protrusions being a logical pre-requisite for any localized translation, our results showed little correlation between mRNA and protein distributions. We hypothesized that this might be due to some protrusion-localized mRNAs remaining translationally inactive in protrusions. In support of this, mRNA and relative translation rate distributions between protrusions and the cell body negatively correlate with each other (Figure 4A). Furthermore, 2D annotation enrichment analysis revealed that most protein categories also show a negative correlation between their mRNA and translation distributions (Figure 4B), with actin cytoskeleton-associated protein categories having relatively high protrusion translation levels but low mRNA enrichments, while mitochondrial, as well as RNA-binding and ribosomal proteins have high relative protrusion mRNA levels but low relative translation in protrusions (Figure 4B; Data S2A). Collectively, these results suggest that for certain categories of mRNAs, targeting to protrusion may serve as a mechanism for translation repression, while for others, low-level mRNA presence in protrusion maybe enough to support robust local translation.

13As mRNA enrichment in protrusions does not define local translation rates, we next set out to reveal the properties in mRNA sequence that may be governing whether an mRNA is locally translated or repressed in protrusions. Cis-regulatory elements in the 5′ and 3′ UTRs of mRNAs play a pivotal role in regulation of translation. Thus, we investigated whether presence or absence of specific UTR motifs could be-associated with differences local translation rates. We focused on 58 well-known 5′ and 3′ UTR regulatory elements (Data S2B) and asked whether presence or absence of any of these elements was-associated with local translation in protrusions. We found six elements that were significantly-associated with higher local translation in protrusions (Figure 4C). The best characterized of these elements is Musashi binding element (MBE), the binding element for MSI1, which can act as both activator and inhibitor of translation depending on the context. Interestingly, MSI1 has recently been demonstrated to bind to and regulate ARP2/3 subunits mRNAs in neuronal cells, a role in agreement with its involvement in protrusions. Existence of an association between specific UTR motifs and local translation rates suggests that such motifs are likely to be locally controlling translation in protrusions.

14As translation is mainly governed by different RBPs and translation factors that act upon an mRNA, RBPs that are localized to protrusions are likely to be crucial for local regulation of translation. Thus, to reveal RBPs that are functionally important for protrusions, we performed an RNAi screen for protrusion-enriched RBPs. Category enrichment analysis revealed three families of RBPs that were significantly enriched in protrusions (Data S1B). These were Sm, KH, and RNase PH domain-containing proteins (Figures 5A–5C). As enrichment of multiple structurally related proteins can be indicative of a conserved function, we focused on the members of these categories, depleting each protein by RNAi and assessing the ability of depleted cells to form stable protrusions through transwell filters. Out of a total of 38 RBPs screened, knockdown of 17 significantly impacted on protrusions (Figure 5D). Interestingly, while the majority of knockdowns inhibited protrusions, some enhanced them (Figure 5D), suggesting that different protrusion localized RBPs function in opposite manners to fine-tune protrusion dynamics.

15To further validate the role of these RBPs in mesenchymal-like migration, we assessed the invasion of MDA-MB231 cells through a 3D collagen-I matrix, where they preferentially migrate as mesenchymal-like cells, following siRNA depletion. Strikingly, knockdown of 16 out of the 17 RBP hits significantly impaired the invasive potential of MDA-MB231 cells (Figure 5D), suggesting that most of these RBPs are likely to play a role in regulation of mesenchymal-like cell migration in 3D. Interestingly, RBP knockdown of both negative as well as positive regulators of protrusions had a negative impact on the invasive potential of the cells (Figure 5D), suggesting that a precise fine tuning of protrusion formation is crucial for optimal migration in 3D.

16Several of the identified RBPs such as the fragile X mental retardation protein-1 (FMR1) and its two homologs (FXR1 and FXR2), zipcode-binding proteins (IGF2BP2 and IGF2BP3), heterogeneous nuclear ribonucleoprotein K (HNRPK), and small nuclear ribonucleoprotein core subunits (SNRPD1, SNRPD3, SNRPF, SNRPG), have been reported to regulate mRNA localization or local translation. Others such as vigilin (VGL), partner of NOB1 (PNO1), and ribosomal protein subunit-3 (RPS3), either associate with, or are part of the translation machinery itself. However, a number of the identified proteins were not previously implicated in regulation of local translation, including three subunits of the exosome core complex (EXOSC7, EXOSC8, and EXOSC9), an RNA degrading complex involved in maturation, quality control, and turnover of many types of cellular RNA (Figure 5D), suggesting that they are likely to be novel regulators of local translation in protrusions.

17To further expand on our results, we validated the role of the exosome core in protrusions, as a potential regulator of local translation and protrusion formation. The exosome core consists of nine subunits that assemble into a two-layered barrel-like structure, with the upper layer forming a cap of three subunits that sit on a ring of six RNase PH domain-containing subunits. In mammalian cells, the exosome core is catalytically inactive but peripherally associates with a set of catalytically active 3′-5′ exoribonuclease subunits that mediate RNA processing/degradation events. Although only six subunits of the exosome core were initially included in our screen, we found all of the nine exosome core subunits to be present and more abundant in protrusions than the cell body, with eight out of nine showing at least 2-fold enrichment (Figure 6A). Surprisingly, RRP6 and DIS3, the two catalytic subunits of the exosome that were expressed in our cells, were significantly depleted in protrusions (Figure 6A), suggesting that the activity of the exosome core in protrusions might be independent of the catalytic subunits.

18First, we confirmed the enrichment of exosome core in protrusions of MDA-MB231 cells by western blotting. In agreement with the proteomics data, western blot analysis of protrusion and cell body fractions with antibodies against exosome core confirmed a high level of enrichment in protrusions, while blotting for the catalytic subunits showed a strong enrichment in the cell body (Figure 6B). Specificity of the antibodies was validated by RNAi (Figure S3A). Accordingly, immunofluorescence analysis showed a fraction of exosome core that was localized to the leading edge, while the two catalytic subunits of the exosome were enriched in the nucleus (Figure 6C). Next, using RNAi, we depleted each exosome core subunit and assessed the ability of depleted cells to form protrusions through 3-μm transwell filters. Efficient depletion of different exosome core subunits was evaluated by qPCR (Figure S3B), and mass spectrometry (Figure S3C). No apoptosis or change in cell proliferation could be detected upon depletion of exosome core subunits over 72 hr (Figures S3D and S3E), suggesting that transient depletion of the exosome core does not impact cell viability. In these settings, knockdown of seven out of nine exosome core subunits significantly inhibited protrusions (Figures 6D and 6E). Depletion of exosome core subunits also inhibited protrusions in HT-1080 fibrosarcoma cells (Figures S3F and S3G), without impacting cell viability (data not shown), suggesting that the role of the exosome in protrusions is not cell type-specific. Importantly, analysis of protrusion dynamics by time-lapse microscopy revealed that similar to inhibition of local translation, exosome core depletion did not affect protrusion initiation, but resulted in destabilization of already initiated protrusions (Movie S6). Exosome depletion also destabilized protrusions in 3D collagen-I matrix, resulting in conversion of the cells from a polarized mesenchymal-like morphology to round (Figures 6F and 6G; Movie S7). As expected, the invasion of MDA-MB231 cells through 3D collagen-I matrix was also significantly impaired upon depletion of several exosome core subunits (Figures 6H and 6I). Together, these results reveal that the exosome core subunits, but not the catalytic subunits, are enriched in protrusions and are necessary for protrusion stabilization, mesenchymal-like morphology, and migration in 3D.

19By comparing the differences between protrusions and the cell body in distribution of proteins, mRNAs, and their relative translation rates, we show that localized translation contributes significantly toward protein localization to protrusions (Figures 2G and 2I). In fact, for the majority of actin-related protein categories enriched in protrusions, such as drivers of actin polymerization, adhesion regulators, and most actin-binding proteins, localized translation seems to be a crucial determinant of protein localization to protrusions (Figure 2H; Data S1E). However, unlike actin-related proteins, RBPs and ribosomal proteins exhibit much higher relative protein levels in protrusions compared to their relative local translation rates (Figure 2H; Data S1E), suggesting that their localization to protrusions must be independent of local synthesis and thus defined by transport. In case of the majority of ribosomal proteins, this is not surprising as regardless of where they may be synthesized in the cytoplasm, they would have to be first transported to the nucleus where they associate with rRNAs in order to form the ribosome subunits. Mature ribosome subunits then need to be transported to other cellular destinations to carry out translation. By analogy, the fact that RBPs are not locally translated either suggests that they may be pre-associated with their target mRNAs before transport to protrusions. Considering this, and as we show that the role of localized translation is to stabilize already initiated protrusions (Figures 3B–3H; Movies S3, S4, and S5), we propose a two-step model for protrusion formation in which ribosomes and specific RBPs in complex with their associated mRNAs are transported to sites of protrusions following protrusion initiation, where their localized activity increases translation of actin cytoskeletal proteins, resulting in further growth and stabilization of already initiated protrusions (Figure 7).

20Importantly, we show that mRNA distributions do not correlate with protein distributions between protrusions and the cell body (Figure 2D). In fact, several categories of locally translated proteins in protrusions do not seem to exhibit any mRNA enrichment (Figure 4B). Conversely, several categories of protrusion-enriched mRNAs have low relative translation rates in protrusions, suggestive of their local repression (Figure 4B). These include mitochondrial as well as RNA-binding and ribosomal protein categories, which show a strong enrichment of their mRNAs in protrusions, but these localized mRNAs seem to be translationally repressed (Figure 4B). Regulated translational repression of localized mRNAs has been shown before. For instance, zipcode binding protein IGF2BP1, which is crucial for targeting β-actin mRNA, also represses translation, but this translational repression can be relieved with the right signal, which in case of IGF2BP1 is phosphorylation at the leading edge by the tyrosine kinase SRC. A similar model may apply to the categories of locally repressed mRNAs defined in this study, which would mean that their local translation could be switched on, following yet unidentified cellular signals. Alternatively, for certain categories of mRNAs, localization to protrusion could be functioning as a means of global translation suppression by keeping them away from their correct site of translational activity in the cell body.

21Nevertheless, the lack of a positive correlation between mRNA localization and local translation (Figures 4A and 4B) suggests that the level of mRNA accumulation may not be a good predictor for local translation. A few copies of certain localized mRNAs may be able to support robust levels of local translation, while high levels of localization for other mRNAs may not lead to significant local translation. Although it is possible that our findings may mostly apply to immortalized, transformed, or malignant cells such as the one that was used here, these findings still have important ramifications for RNA localization studies as they highlight that mRNA localization alone may not be an indicator for local protein expression by default.

22As mRNA translation is chiefly regulated by cis-regulatory elements within 5′ and 3′ UTRs of mRNAs, we investigated the association of different known UTR elements with regulation of local translation in protrusions. Our study reveals that presence of specific 3′ UTR elements is associated with higher local translation rates in protrusions (Figure 4C). The exact functions of these regulatory motifs in control of local translation in protrusions remain to be determined. It also remains to be determined if there are yet unknown UTR elements associated with local translation in protrusions, in addition to the known UTR elements investigated here.

23To reveal functionally important RBP regulators of localized translation in protrusion, we performed an RNAi screen, depleting members of RBP categories significantly enriched in protrusions and assessing whether their depletion modulated protrusions. Out of 38 RBPs screened, we identified 17 belonging to different categories that significantly regulated protrusions. Sixteen out of 17 also significantly affected migration in 3D collagen-I matrix (Figure 5D). Many of these RBPs have been reported to regulate RNA localization or local translation. However, a number of proteins we identified, including exosome core subunits, have not been implicated in regulation of RNA targeting or local translation. Previously, the exosome core was shown to be present in both the nucleus and cytosol of eukaryotic cells, although tagging approaches in Drosophila had also suggested presence of certain exosome subunits in the periphery of the cells. Our data reveal that, at least in a mesenchymal-like highly invasive cancer cell-line, the exosome core, but not its catalytic subunits, is enriched in protrusions (Figures 6A–6C) and acts to stabilize protrusions and promote 3D invasion (Figures 6D–6I; Movies S6 and S7). As cell migration and invasion are believed to be intimately linked with metastasis, the exosome-mediated regulation of mesenchymal-like cell migration described here may impact cancer metastasis. In support of this view, we found that, in human breast cancers, increased expression of the exosome core subunits correlates with bad prognosis. This was the case with every exosome core subunit for which expression data were available (Figures S3H–S3L). This result implicates the exosome core in cancer progression and raises the possibility that regulation of cell migration via the exosome could be resulting in a more aggressive, metastatic phenotype.

24It is now evident that many types of non-coding RNAs such as microRNAs, long non-coding RNAs, and anti-sense RNAs play a significant role in post-transcriptional regulation of mRNA expression. It is therefore likely that in addition to RNA-binding proteins, non-coding RNAs may also be involved in regulation of local translation in protrusions. In fact, two of the known cis-regulatory elements (K-BOX, BRD-BOX) identified here as associated with higher protrusion translation rates, are not known to bind to any RBPs but instead seem to associate with microRNAs. As our RNA-seq analysis was limited to mRNAs, it remains to be determined whether certain non-coding RNAs also show enrichment in protrusions, and more importantly, whether they play a functional role in regulating protrusions.

253D collagen invasion assay and 3D collagen morphology assays were done as described before with modifications (see also Supplemental Experimental Procedures). Transwell protrusion formation assays were performed on 3-μm pore transwell filters (6.5 mm insert size, 24-well format), coated with 5 μg/ml collagen. A total of 100,000 cells were seeded in 100 μl serum-free DMEM on top of the filter and 600 μl of DMEM with 10% serum was placed in the bottom chamber. Cells were imaged live, or fixed with 4% formaldehyde at indicated time points before imaging. Multiple tile scans, from at least three independent experiments, each composed of 3 × 3 contiguous fields of view, were taken at two z planes (top and bottom of the filter). For quantification of protrusions, fluorescence intensity at the bottom and top of the filter was then measured by ImageJ and bottom to top ratio was calculated.

26For puromycinylation of nascent proteins, cells were treated with 10 μg/ml puromycin for 10 min prior to lysis. For cycloheximide treatments, 10 μg/ml cycloheximide was added to the cells for indicated times. For emetine treatments, indicated dose of emetine was added to the cells for indicated times, followed by a single wash in 10% FBS containing DMEM. For specific inhibition of translation in protrusions, cells were seeded on transwells and allowed to form protrusions for 2 hr. The transwells were transferred to a reservoir dish with 1 μg/ml emetine (or vehicle for control treatment) in 10% FBS containing DMEM. The level of liquid in the reservoir was always significantly lower than on top of the filter, creating a positive hydrostatic pressure that limited emetine diffusion upward. After 5 min, the filters were taken out, washed by dipping the filter in 10% FBS containing DMEM, and transferred back to the original well for downstream analysis.

27Purification of cell protrusions was performed using 3-μm pore polycarbonate transwell filters (75 mm membrane inserts). Filters were coated with 5 μg/ml collagen before being seeded with 10 million cells in serum-free DMEM on top, with DMEM containing 10% serum added to the bottom. For western blot analysis, a single transwell was used per condition. For proteomics analysis, four transwells were used per condition. After allowing the cells to form protrusions for 2 hr, transwells were washed in PBS, fixed with −20°C methanol for 20 min, washed again with PBS, and the protrusions from the bottom of the filter were shaved off using a glass coverslip, with the coverslip being dipped in 2% SDS sample buffer. The cell body fraction was then prepared by direct addition of sample buffer to the top of the filter. A similar procedure was used for RNA purification from protrusions, but from one transwell per condition, no methanol fixing, and lysing in RLT buffer from QIAGEN’s RNeasy Mini Kit. RNA was purified according to the manufacturer’s instructions, followed by spectrophotometric measurement of the quality and concentrations.

28For SILAC labeling, cells were grown for seven doublings in heavy or light SILAC DMEM supplemented with 10% dialyzed FBS and 600 mg/l L-Pro. For pSILAC, light-labeled cells were transiently switched to medium or heavy SILAC DMEM during a 2-hr transwell protrusion formation assay. Reciprocally mixed SILAC or pSILAC samples were resolved by SDS-PAGE, stained with Gel-code blue Coomassie dye (Pierce), and cut into 23 sections before Trypsin digestion and peptide extraction. Liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis was performed by the Institute of Cancer Research (ICR) proteomics core facility. Mass spectrometry search and quantifications were done by Maxquant software. For pSILAC quantifications, the L label was ignored and H/M ratios, which measure relative nascent protein levels, were used. For RNA-seq, protrusions and the cell body total RNA preps were generated in duplicates from two biological filter-fractionated replicate experiments. Sample preparations and Illumina sequencings were performed by ICR’s tumor profiling unit. All proteomics/transcriptomics data analyses were performed by Perseus software (see also Supplemental Experimental Procedures). Mass spectrometry proteomics raw data and search results were deposited to the ProteomeXchange Consortium, via the PRIDE partner repository, with the dataset identifiers PRIDE: PXD000914 and PRIDE: PXD002649. RNA-seq data were deposited to the ArrayExpress database (http://www.ebi.ac.uk/arrayexpress) under the accession number ArrayExpress: E-MTAB-2546.

29F.K.M. and C.J.M. designed all experiments and wrote the manuscript. A.P. assisted with mass spectrometry sample preps and LC-MS/MS runs. H.P. generated the MDA-MB231 mKate CAAX cell line and assisted with confocal microscopy. A.S. assisted with immunofluorescence. S.K. assisted with 3D live-cell imaging. A.M. assisted with exosome validation experiments. Y.Y. performed the gene expression-based stratification of the breast cancer patient survival data. All other experiments and data analyses were performed by F.K.M.