Nucleoporin-93 overexpression overcomes multiple nucleocytoplsamic trafficking bottlenecks to permit robust metastasis


 To identify genetic events driving breast cancer progression, we performed multi-omics integrated analyses that identified overexpression of nucleoporin-93, a nuclear pore component. Here we show that NUP93 overexpression enhances trans-endothelial migration and matrix invasion in vitro, along with metastasis in animal models. These findings were supported by analyses of naturally occurring activating NUP93 mutations and inactivating nephrotic syndrome mutations. Mechanistically, NUP93 activation enhanced the ultimate nuclear transport step shared by multiple signaling pathways, including TGF-beta/SMAD, EGF/ERK and TNF/NF-κB. Likewise, NUP93 can boost nuclear transport of beta-catenin, as well as elevate expression and inhibit degradation of this co-activator. The emerging addiction to nuclear transport exposes vulnerabilities of tumors overexpressing NUP93. Congruently, we report that myristoylated peptides corresponding to the nuclear translocation signals of SMAD and ERK inhibited growth and metastasis. Our study illuminates a previously unappreciated hallmark of advanced tumors, which derive benefit from unblocking a set of nucleocytoplasmic bottlenecks.


Introduction
Breast carcinoma is the leading cause of cancer-related mortality in women 1 . Owing to genomic instability, breast cancers often exhibit somatic copy number aberrations (CNAs) 2 , such as ampli cations of the ERBB2 gene (at 17q12) 3,4 . Together with the epidermal growth factor receptor (EGFR), ERBB2/HER2 acts as a transducer of growth factor signals 5 . Importantly, kinase inhibitors and combinations of antibodies that inhibit HER2 are widely used to treat HER2-overexpressing breast cancers 6,7 , which exempli es the therapeutic potential offered by the identi cation of overexpressed oncogenic proteins. Aneuploidy, CNAs and overexpression of speci c genes are well-known sources of gene dosage imbalance in cancer. Because many cellular functions rely on multi-protein complexes, precise and timely dosage of the individual elements, especially protein stoichiometry, might be critical and de ne true driver genes 8 . For example, limited CNAs affecting diverse breast cancer genes, all regulating receptor endocytosis, confer poor prognosis 9 .
Herein, we demonstrate that breast cancers acquire aggressive phenotypes when another intracellular tra cking pathway, which regulates nuclear-cytoplasm transport, is activated. Usually, proteins shuttle to the nucleus when an intrinsic nuclear localization signal (NLS) interacts with importins, which escort them to the nucleus via the nuclear pore complex (NPC) 10 . However, signaling proteins like ERK translocate, upon stimulation, by means of unique nuclear translocation signals (NTS) 11 . The NPC structure is stabilized by protein complexes arranged in concentric rings. Using a new computational tool, we identi ed a speci c component of the inner NPC's ring, nucleoporin 93 (NUP93), as highly essential for breast cancer. Loss-of-function mutations in NUP93 are recognized as monogenic causes of steroidresistant nephrotic syndrome (SRNS) 12 , whereas rare gain-of-function (GOF) mutations identify relatively aggressive mammary tumors 13 . In line with these observations, we herein report that elevated abundance of NUP93, like GOF mutations, promotes metastasis of mammary tumors, while knockdown markedly retards metastasis in animal models. Subcellular fractionation revealed that NUP93 overexpression enhances the concluding step of transcription factor translocation to the nucleus, an event shared by EGF, TGF-beta, WNT, TNF-alpha and glucocorticoids. In addition, NUP93 overexpression accelerates nuclear transport of beta-catenin and inhibits its degradation. Accordingly, NTS-mimetic decoy peptides inhibited tumor progression in animal models. Altogether, our results identify a robust, yet targetable, mechanism shared by aggressive mammary tumors, which depend on ongoing signaling from the tumor microenvironment to the nucleus.

Results
NUP93 is highly essential for survival of mammary tumor cell lines, and its abundance associates with poor prognosis of patients with breast cancer To identify genes most relevant to progression of breast cancer, we performed computational analyses integrating RNA sequencing data from both METABRIC 14 and The Cancer Genome Atlas, and incorporated both somatic copy number alterations (CNA) and patient survival data (> 3000 patients). In parallel, we added gene essentiality scores derived from whole-genome shRNA screens that used 138 cancer lines 15,16 . Next, we compared essentiality scores of the top 500 clinically relevant genes and the bottom 500 genes. As expected, the top 500 genes were found to be more essential (p-value < 0.01 for pan-cancer cell lines and < 0.02 for breast cancer cell lines). Combining the top 500 clinically relevant genes and the 500 most essential genes, we obtained a ranked list of 20 genes, which are both clinically relevant and phenotypically essential for breast cancer (Supplementary Table 1). Interestingly, two tubulin genes topped the list of 20 genes, in line with reports linking overexpression of tubulins to survival of patients with breast cancer 17 . The scores received by nucleoporin 93 (NUP93) were close to those received by the tubulins. Figure 1A presents the essentiality scores received by NUP93 in a series of breast cancer cell lines. Because anti-tubulin drugs are widely used in chemotherapy, but no approved anti-cancer drugs target NUP93 18 , we focused on the nucleoporin. Additional analyses con rmed that the over-expression of this gene leads to poor prognosis (logrank P < 3.61E-12). Importantly, this trend remained signi cant when age, metastasis to lymph nodes and disease subtypes were controlled (Cox hazard ratio = 1.22, P < 1.96E-4). As an example, we divided the METABRIC dataset into three groups according to NUP93 expression levels. The Kaplan-Meier survival analyses separately performed for each group revealed that NUP93 overexpression is independently and signi cantly associated with shorter disease-free patient survival (Fig. 1B). The same dataset was strati ed also according to the status of estrogen receptor (ER; Fig. 1C), which revealed that the more aggressive group of tumors, which lack ER expression, displayed relatively high NUP93.
An association between NUP93 and more virulent mammary tumors emerged from analysis of the 10 integrative clusters (IC) of breast cancer 2 . IC10, which incorporates mostly triple negative tumors, showed highest NUP93 expression. Likewise, NUP93's transcripts were relatively high in IC5, which identi es almost all cases with ERBB2/HER2 ampli cation. In addition, we found that high NUP93 was signi cantly associated with CNA (Fig. 1E). As expected, the highest gains of the gene were found in the basal subtype, and this was followed by the HER2-enriched subtype. Furthermore, high NUP93 expression was typical to high-grade tumors (Fig. 1F). In conclusion, by developing a novel computational pipeline, we learned that a component of the NPC is highly essential for survival of breast cancer cells and can predict shorter survival of a subset of patients with breast cancer.
Experimental strategies applying induced overexpression and downregulation reveal involvement of NUP93 in proliferation, migration and invasiveness Because NUP93 is overexpressed in the basal subtype, we selected two basal models: (i) MDA-MB-231 basal B breast cancer cells, and (ii) MCF10A, a non-tumorigenic immortalized basal B line. Firstly, we expressed an inducible allele of NUP93 in MDA-MB-231 cells and veri ed relatively high expression on induction with tetracycline (Tet; Fig. 2A). Reciprocally, we established sublines expressing different doxycycline (DOX) inducible shRNAs (iSh; Fig. 2B). As predicted, exposing cells to Tet enhanced their ability to incorporate a radioactive nucleoside into DNA (Fig. 2C), whereas DOX-induced downregulation reduced DNA synthesis (Fig. 2D). Next, we applied a cell viability assay, which supported the ability of ectopic NUP93 to enhance viability (Figs. S1A and S1E). Similar conclusions were derived from experiments that used subclones constitutively expressing shNUP93 (Fig. S1B and S1F; left panels) and cells expressing siRNAs speci c to NUP93 (Fig. S1B and S1F; right panels). A parallel set of experiments, which made use of genetically modi ed MCF10A cells (Figs. S1C and S1D) con rmed association of higher viability signals with overexpression ( Fig. S1G) and lower signals in NUP93-depleted cells (Fig.  S1H).
According to a recent breast cancer study, predictors of migration, rather than proliferation, are strongly associated with patient survival 19 . When we placed inducible MDA-MB-231 cells on the upper compartment of cell culture inserts and treated them with the inducer, they migrated signi cantly faster than untreated cells (Fig. 2E). The reciprocal approach, which used DOX-inducible shRNAs, con rmed lower migration rates relative to control shRNA (Fig. 2F). Moreover, when the intervening membrane was coated with an extracellular matrix we similarly observed increased and decreased invasion rates with cells exposed to Tet (overexpression) or DOX (knockdown), respectively (Figs. 2G and 2H). Additional migration (Fig. S1I) and invasion assays (Fig. S1J) using MDA-MB-231 cells, as well as a similar set of MCF10A cell experiments (Figs. S1K-S1N) provided further support to the notion that NUP93 increases cellular motility. In conclusion, in line with the high essentiality and clinical signi cance, high abundance of NUP93 associated in vitro with increased viability, mitosis, migration and matrix invasion.
NUP93 overexpression enhances trans-endothelial migration and 3D invasion, as well as remodels focal adhesion sites According to a recent report, NUP93 is involved in 3D migration and this correlates with an altered actin cytoskeleton 20 . In line with this report, when pre-formed spheroids were induced with tetracycline to elevate NUP93, many cells migrated outward to form invasion zones (Fig. S2A). Reciprocally, NUP93 silencing reduced zone area (Fig. S2B). Increased invasiveness emerged also from TEM (trans-endothelial migration) assays that used endothelial monolayers overlaid by cells expressing inducible shNUP93 (Fig.   S2C). Because VEGF might be involved, we used ELISA and found that siNUP93 reduced VEGF secretion (Fig. S2D). Interestingly, depletion of NUP93 also increased cell area (Fig. S2E) and reduced by 4-fold the ability of cells to form invadopodia (Fig. S2F). Notably, invadopodia contain microdomains with spatiotemporal dynamics of actin-rich adhesion sites 21 . Indeed, we found that depletion of NUP93 was associated with remodeling of the actin cytoskeleton (Fig. S2G) and up-regulation of integrin alpha5 and integrin beta1 (Fig. S2H). Next, we extended the analysis to additional cell adhesion molecules, such as two paxillin family members-paxillin and Hic-5, which have been widely implicated in turnover of adhesion sites 22 . Imaging and quanti cation of the respective adhesion areas unveiled redistribution along with up-regulation of the respective areas in siNUP93-treated cells (Fig. S2I). In summary, the overexpressed NUP93 increases 3D cell invasion, as well as enhances TEM, invadopodium assembly and remodels adhesion sites containing molecules like paxillin, fascin and zyxin.
High NUP93 abundance associates with increased rates of tumor growth and metastasis, along with matrix reorganization Next, we implanted MDA-MB-231 cells, expressing a Tet-inducible NUP93 allele, in the fat pad of female mice and added the inducer to the drinking water of one of 2 groups of mice. Both the volumes and weights of tumors we harvested 3 weeks later were signi cantly larger in the treated group (Fig. S3A). A reciprocal experiment that made use of cells expressing a DOX-inducible shNUP93 reinforced the ability of NUP93 to accelerate tumor growth (Fig. S3B). This conclusion was independently supported by using cells stably expressing shNUP93 and determining tumor growth (Fig. S3C). As an initial test of metastasis, lungs were excised and metastases were quanti ed (Fig. S3D). The results re ected strong inhibitory effects, which prompted studies employing inducible expression and two different metastasis assays. Cells inducibly overexpressing NUP93 were injected either into the tail vein (Fig. 3A, left panel) or into the subaxillary mammary fat pad ( Fig. 3A; right panel). Seven days later, we added tetracycline to the drinking water and mice were sacri ced 3 weeks later. The results indicated that NUP93 overexpressors are endowed with > 3-fold stronger capacity to colonize lungs. Further, when the same protocol was applied on tumors inducibly expressing shNUP93, we observed remarkably fewer lung metastases (Fig. 3B). Together, these observations assigned to NUP93 an important role in metastasis.
To resolve the identity of protein mediators, we sequenced RNA from MDA-MB-231 cells that were pretreated with siNUP93. The major differentially expressed (DE) genes are shown in Fig. 3C and Supplementary Excel File 1. Along with up-regulation of the cysteine-rich angiogenic inducer 61 23 , we observed downregulation of several mitochondrial genes, such two subunits of the respiratory chain I, which drives ATP generation, and RNR1, a subunit of ribonucleotide reductase, which catalyzes dNTP production. These alterations might re ect the metabolically active state of NUP93-overexpressing cells.
We also noted TGF-beta, keratins and extracellular matrix (ECM) components 24 . The heatmap shown in Fig. 3D lists the major matrisome DE genes, including several types of collagen. These observations were supported using PCR (Fig. S3E) and immunoblotting for collagen IVa6 (Fig. S3F). Because collagen enables cell-to-matrix adhesion through binding with integrins, we predicted accompanying changes in substrate adhesion. Experiments using DOX-inducible shNUP93 (Fig. S3G), or an inducible NUP93 (Fig.  S3H), con rmed that NUP93 reduces adhesion to substrate. To examine collagen involvement, we stained collagen IVa6 (Fig. S3I). Importantly, collagen brils were observed only in NUP93-depleted cells. For in vivo assays, tumors excised from DOX-treated mice were stained with either a collagen dye, picrosirius red (Fig. 3E), or an anti-collagen IVa6 antibody (Fig. S3J). Evidently, tumors from mice treated with shNUP93 showed strong staining only if the animals were pre-treated with Doxycycline. Since collagen is induced by TGF-beta 25 , we used various methods to probe for the ligand and the receptor, TGFBRII. The results indicated co-induction, along with localization of the receptor to lamellipodia (Figs. 3F-3G, S3K-S3M). In conclusion, high abundance of NUP93 confers accelerated growth and metastatic spread upon mammary tumor cells. The underlying mechanism involves weakening ECM adhesion due to inhibition of collagen deposition, lower expression of ECM-modifying enzymes and reduced secretion of TGF-beta.
Overexpression of NUP93 enhances nuclear translocation of SMAD, ERK, STAT3, GR and p105 (NF-kB) in response to TGF-beta, EGF, glucocorticoid and TNF-alpha, respectively Steroid resistant nephrotic syndrome is caused by NUP93 mutations, which interfere with BMP7-induced SMAD transcriptional activity 12 . In addition, it was shown that insect NUP93 is needed for nuclear import of active SMADs 26 . Hence, we assumed that NUP93 overexpression enhances signals initiated by BMP7/TGF-beta and other stimuli. Probing endogenous importin7 and NUP93 revealed a perinuclear ring of importin7, similar to the pattern of endogenous NUP93 (Fig. 4A). Importantly, importin7's perinuclear localization was lost in mammary cells pre-treated with siNUP93. To validate the prediction that NUP93, by recruiting speci c importins, mediates TGF-beta induced nuclear translocation of SMADs, we depleted NUP93 in MCF10A cells and followed the kinetics of nuclear import (Fig. 4B). While SMAD2/3 translocated to the nucleus of control cells within 20 minutes, we observed only limited translocation in NUP93-depleted cells. Assuming that NUP93 controls translocation of additional importin7's cargos, we examined the extracellular signal-regulated kinase (ERK). Upon stimulation with EGF, ERK undergoes phosphorylation that exposes a nuclear translocation signal (NTS), which facilitates binding to importin7 27 . Accordingly, when we stimulated control cells with EGF the phosphorylated form of ERK (pERK) clearly translocated to the nucleus, but cells depleted of NUP93 displayed much weaker translocation (Fig. 4C).
This difference was veri ed by fractionation of cell extracts into cytoplasmic and nuclear fractions (Fig. 4D). Consistent with these results, we observed reciprocal changes in MCF10A cells engineered to inducibly overexpress NUP93 (Fig. 4E).
In similarity to ERK, STAT3 undergoes phosphorylation and nuclear translocation upon stimulation with EGF, but alpha importins, rather than importin7, have been implicated 28,29 . Nevertheless, analyses of both total STAT3 (Fig. S4A) and pSTAT3 (Fig. S4B) indicated that NUP93 can regulate EGF-induced import of STAT3. These observations raised the possibility that NUP93 anchors different importins and translocates them via the NPC, along with the respective cargos. Hence, we followed two additional cargos, the glucocorticoid receptor (GR) and the p105 subunit of the nuclear factor kappa B (NF-κB; Figs. S4C and S4D). Using NUP93-depleted cells, in both cases we observed strong inhibition of ligand-induced translocation following stimulation with either dexamethasone (DEX), a synthetic GR ligand, or with the tumor necrosis alpha (TNFa). Notably, importin alpha/beta, as well as importin7, translocate GR 30 , whereas NF-κB is translocated by importin alpha3 and alpha4 31 . In conclusion, apart from the known functions of NUP93 as a scaffold nucleoporin involved in NPC assembly, this molecule emerges as a broad-spectrum transporter of the active forms of signaling proteins.
Myristoylated peptides corresponding to the NTS of ERK and SMAD inhibit progression of NUP93overexpressing tumors Next, we attempted blocking nuclear translocation of ERK by preventing binding of importin7 to ERK's NTS. The strategy used a previously described NTS-derived peptide fused to myristic acid 32 . As expected, the ERK-derived peptide inhibited nuclear accumulation of ERK, but a control peptide exerted no effect (Fig. 5A). In addition, the peptide reduced incorporation of radioactive thymidine into DNA, only in cells overexpressing NUP93 (Figs. S5A and S5B). Likewise, the S-phase fraction observed with cells treated with the peptide was severely reduced (Fig. S5C). Similarly, the ERK peptide reduced the ability of cells to form colonies, migrate and invade (Figs. S5D-S5F). Hence, we implanted NUP93-overexpressing cells in the subaxillary mammary fat pad of mice, and once tumors became palpable intravenously treated animals with the ERK-derived peptide. RFP-uorescence and size of the primary tumors were measured (Figs. 5C and 5D), along with tumor weight (Fig. 5E). In addition, we cut out the lungs to assay metastases (Fig. 5F). Evidently, tumors overexpressing NUP93 grew faster and colonized lungs better than the control tumors. Moreover, while the ERK peptide weakly inhibited growth and metastasis of control tumors, the inhibitory effects observed upon treatment with the ERK peptide were signi cantly stronger, implying that NUP93-overexpressing tumors acquire dependence on ERK's nuclear transport.
Because NUP93 mutations causing a renal disease fail SMAD signaling 12 and tumors frequently corrupt the TGF-beta pathway 33 , we synthesized a similar SMAD peptide, based on a previously identi ed Ser-Pro-Ser triad 27 . The peptide nearly completely inhibited TGF-beta-induced nuclear import of SMAD2/3 (Fig. 5B). In addition, it reduced incorporation of radioactive thymidine (Fig. S5B), markedly lowered the Sphase fraction (Fig. S5C), as well as inhibited the ability of cells to form colonies, migrate and invade (Figs. S5D-S5F). As with the ERK peptide, we implanted NUP93-overexpressing cells in the fat pad of mice and delivered the SMAD peptide three times per week. RFP-uorescence, as well as tumor volumes (Figs. S5G and S5H) and weights were determined (Fig. S5I) and lungs metastases were counted (Fig. S5J). We observed weak inhibition of control tumors, but stronger effects were observed when treating NUP93overexpressing tumors. Moreover, treatment with the SMAD peptide only weakly inhibited metastasis of control tumors, but the numbers of micro-metastases formed by NUP93-overexpressing cells were reduced by 75-85% (Fig. S5J). In conclusion, the ERK and SMAD peptides exemplify the potential therapeutic scenario offered by targeting the cargo-importin-NUP93 axis, speci cally in NUP93overexpressing breast cancers. Importantly, it has recently been reported that inhibition of NPC formation causes selective cancer cell death, while normal cells undergo a reversible cell cycle arrest 34 . In conclusion, the myristoylated peptides we tested might effectively and selectively inhibit cancer cells overexpressing NUP93.
Natural gain-and loss-of-function mutations con rm NUP93's roles in metastasis Three NUP93 mutations, E14K, Q15X and R327C, were identi ed by a screen aimed at putative driver mutations escalating the risk of metastasis 13 . Hence, we stably expressed two mutant alleles in MCF10A (Fig. S6A) and in MDA-MB-231 cells (Fig. 6A). Both mutants, especially R327C, superseded the ability of wildtype NUP93 to seed colonies (Figs. 6B and S6B). In addition, we performed DNA synthesis (Figs. 6C and S6C), migration and invasion assays (Fig. 6D, 6E, S6D and S6E), which indicated that the mutants were more active than the wildtype form. Next, we implanted the respective MDA-MB-231 cells in the fat pad of animals and followed tumor growth (Figs. 6F and 6G). Both mutations enhanced tumor growth relative to WT and control (EV) cells. Similarly, quanti cation of metastatic lung nodules indicated that both mutant alleles were consistently more active than WT in either metastasis format (Fig. S6F). Notably, we occasionally observed metastases in other organs. Focusing on liver metastases, we observed striking differences ( Fig. 6H): Whereas cells overexpressing either mutant colonized livers, we were unable to detect any lesion in livers from mice injected with WT cells. In conclusion, the gain-offunction NUP93 mutations enhance oncogenic attributes and confer organ-speci c metastatic colonization.
Analyses of patients with nephrotic syndromes detected homozygous missense NUP93 mutations 12 . To study oncogenic effects, we rstly disrupted the endogenous NUP93 gene using Crispr-CAS9. Several knockout (KO) clones were established (Fig. 6I). Next, we generated a series of KO cells expressing WT or individual mutants ( supported the LOF phenotypes: stable expression of neither mutant was able to reconstitute the relatively high rates of tumor growth/metastasis displayed by KO cells re-expressing WT. In summary, by studying GOF and LOF mutations of NUP93 we obtained independent evidence in support of the critical roles played by NUP93 in breast cancer progression. NUP93 overexpression increases GTP-loading onto RHO GTPases and enhances transcriptional outcome of the WNT and other signaling pathways RHO family GTPases are switches involved in mammary tumorigenesis and metastasis 35 . Hence, we utilized the ability of effectors to bind with the active, GTP-bound forms. Overexpression of NUP93 increased CDC42-GTP and RAC1-GTP, but decreased RHOA-GTP (Fig. S7A). Congruently, siNUP93 reduced active RAC1 and CDC42, but increased RHOA-GTP. Notably, RAC1 and CDC42 collaborate with KRAS. Accordingly, we observed signi cant downregulation of RAS-GTP in siNUP93-treated cells (Fig.   S7B). Because RAS is activated upon stimulation of numerous pathways, we undertook a multiple promoter-reporter strategy. HEK293 cells were co-transfected with a NUP93 expression vector and various luciferase plasmids containing different DNA response elements (REs), including the glucocorticoid RE (GRE; Fig. 7A). This revealed that NUP93 overexpression activated several promoters, including GRE, SMAD4 and FLI1. Yet, the strongest signal (> 50-fold) was observed with, a beta-catenin responsive reporter containing the binding site for T cell factor (TCF)/lymphoid enhancer factor (LEF). Nuclear accumulation of beta-catenin is induced by WNT; once in the nucleus, beta-catenin associates with TCF/LEF and activates target genes 36 . As expected, co-transfection of the reporters and siNUP93 decreased several signals, but two collagen promoters were activated (Fig. S7C). In addition, cotransfecting NUP93's oncogenic mutants increased reporter activity beyond the WT signal (Fig. 7B). Accordingly, transfection of NUP93, either WT or mutants, strongly elevated the beta-catenin protein ( Fig. 7C). This translated to increased expression of multiple beta-catenin target genes (three are shown in Fig. S7D). Because EGFR can transactivate the beta-catenin pathway 37 and SMAD3 can occupy WNTresponsive elements 38 , we examined transactivation by EGF and TGF-beta. The results con rmed transactivation ( Fig. 7D), which might enhance the relatively large effect of WNT. Next, we addressed stability of beta-catenin. Cell treatment with cycloheximide, a protein synthesis inhibitor, followed by immunoblotting, con rmed short half-life of beta-catenin (Fig. 7E). In contrast, when overexpressed, both WT NUP93 and the oncogenic mutants strongly prolonged the half-life, implying multiple NUP93/betacatenin interactions.
Using WNT3A, we found that R327C-NUP93, more than WT, enhanced ligand-stimulated promoter activation (Fig. 7F). To test if this cooperative effect was due to nuclear transport, we rstly fractionated control cells and detected only a small fraction of beta-catenin in the nucleus (Fig. S7E). Importantly, this fraction was erased by NUP93 knockdown. Moreover, immuno uorescence indicated that treatment with WNT3A induced relatively weak nuclear translocation of beta-catenin, which was enhanced by wildtype NUP93, and further increases were observed in cells expressing oncogenic NUP93 (Fig. 7G). Thus, both fractionation and immuno uorescence supported the cooperative effect of WNT and NUP93. To validate functionality, we knocked-down LEF1 and stimulated cells with WNT3A (Figs. 7H and 7I). While siLEF1 inhibited cell migration/invasion by approximately 30% in WT-expressing cells, this increased to 90% in cells expressing R327C-NUP93, in support of cooperativity. In conclusion, when overexpressed, NUP93 enhances loading of GTP onto RAS, RAC1 and CDC42, molecular switches of cell migration. In line with this, several signaling pathways were found to be constitutively active in NUP93-overexpressing cells, including the WNT pathway, hyperactivation of which has been implicated in metastasis 39 . WT NUP93 elevated expression and inhibited degradation of beta-catenin, while the oncogenic mutants, better than WT, enhanced WNT-induced nuclear transport of beta-catenin.
Proteome-and transcriptome-wide analyses identify podocalyxin and multiple RNA transcripts as potential mediators of NUP93-induced metastasis To complement the promoter analyses, we applied liquid chromatography-mass spectrometry (LC-MS/MS) on cytoplasmic and nuclear fractions. As expected, siNUP93 signi cantly reduced beta-catenin in the nucleus (Figs. 7J and S7F; Supplementary Excel File 2). Alongside, the cytoplasmic fraction showed downregulation of speci c MAPK pathways, whereas the nuclear fraction displayed elevated SP1, which up-regulates collagen 40 , bystin, which regulates cell adhesion 41 , and cullin-3, which is essential for collagen export 42 . All three proteins were up-regulated in NUP93-depleted cells and underwent nuclear-cytoplasm translocations (Figs. S7G and S7H). The proteomic analyses also indicated that the mucin podocalyxin (PODXL) underwent up-regulation in the cytoplasm of siNUP93-treated cells.
Along with transport of proteins, the NPC mediates selective exchange of RNA molecules between the nucleus and cytoplasm 43 . To explore transport of speci c RNAs, we depleted NUP93 in MCF7 breast cancer cells and examined the subcellular localization of polyadenylated transcripts (Fig. S7I), essentially as previously described 44 . NUP93 depletion signi cantly affected the localization of hundreds of RNAs, led to a 3-fold increase in cytoplasmic enrichment of transcripts from 200 genes, and conversely, to 3-fold increase in nuclear enrichment of transcripts from 153 other genes (see Supplementary Excel File 3). Because the lincRNA called NORAD (LINC00657) was identi ed also by the RNAseq analysis presented in Fig. 3C, we focused on this molecule. Notably, NORAD enhances TGF-beta signaling and metastasis 45 . Hence, we employed single-molecule FISH (smFISH) to examine sub-cellular distribution of this intronless, mostly cytoplasmic lncRNA. In line with the RNA-seq analysis, upon NUP93 depletion NORAD exhibited strong nuclear enrichment in both MCF7 and MDA-MB-231 cells (Fig. S7J). Conceivably, nuclear retention inactivates NORAD, along with similar RNAs, to retard metastasis of NUP93-low breast cancers.
In summary, by integrating clinical and laboratory lines of evidence, we concluded that high abundance of NUP93 is associated with highly aggressive breast cancers. Correspondingly, the overexpressed NUP93 enhanced TEM and matrix invasion, while reorganizing the matrisome. These attributes translated to increased rates of tumor growth and metastasis in animal models. Mechanistically, overexpression of NUP93 boosts the ultimate nuclear transport step activated by diverse extracellular signals, including WNT, EGF and TGF-beta. Collectively, these lines of evidence identify NUP93 as a driver of metastasis, which hijacks the beta-catenin and other signaling pathways.

Discussion
The results presented herein identify NUP93 as a potent driver of metastasis in aggressive subtypes of breast cancer. Unlike other subtypes, the group overexpressing HER2, a functional partner of EGFR, is driven by signals mimicking cellular stimulation by growth factors, whereas a fraction of the basal subtype is propelled either by an overexpressed EGFR, or by mutations 46 . In analogy, our results indicate that the aberrantly overexpressed NUP93 simultaneously ampli es multiple growth factor signals by means of activating the ultimate nuclear transport step shared by the majority of signal transduction pathways. This entails nuclear translocation of transcription co-activators, such as STAT, SMAD, ERK and beta-catenin. Consistent with this scenario, two lines of computational evidence identi ed NUP93 as a putative driver of breast cancer progression: the rst analyzed recurrent CNAs 47 , while the other employed addressed driver mutations 13 . Three rare NUP93 point mutations were identi ed by this study, and they were characterized in vitro. The animal studies we performed assigned metastasis driver roles to two of the three mutant forms, as well as uncovered their ability to enhance basal and ligand-induced nuclear translocations of beta-catenin, a well-characterized driver of metastasis 48  as a factor required for beta-catenin transport 50 . Regardless of the exact mechanisms enabling NUP93 to control beta-catenin import, our promoter-reporter assays indicated that NUP93 more strongly regulates the beta-catenin pathway, as compared to other pathways. This superiority of the LEF1/TCF promoter might be due to two mechanisms, which are not mutually exclusive. The rst involves the recently discovered ability of NUP93 to drive expression of cell identity and other genes through interactions with super-enhancers 20,51 . The other mechanism might relate to the rich crosstalk between the WNT pathway and other signaling routes. For example, SMAD3 colocalizes with LEF1 at WNT-responsive elements 38 , and EGFR can transactivate the beta-catenin pathway 37 .
In summary, the integrative computational analysis we developed has assigned NUP93 with important roles in progression of mammary cancer, which we con rmed in animals. Surprisingly, NUP93 emerges from these studies as a master driver of metastasis. This conclusion was supported by functional characterization of two groups of naturally occurring mutant alleles. Remarkably, mammary cells overexpressing NUP93 display over-activation of multiple growth factor pathways, including the WNT/beta-catenin pathway. This route has previously been associated with colorectal tumors, but recent network-based analysis linked it to breast cancer 52 . We conclude that NUP93 integrates and enhances not only WNT but also additional growth factor signals. Thus, by seizing an essential step common to multiple signaling routes, NUP93 acquires robust oncogenic attributes. Beyond the identi cation of augmented nuclear transport as a new hallmark of cancer, our study exposes tumor vulnerabilities and offers not only a new target for intervention but also a pharmacological strategy tailored to the subset of Images were acquired using the LSM 800 software.

Immunohistochemistry and Picrosirius red staining
Formalin-xed tumor sections were de-para nized and rehydrated. Antigen retrieval was performed in a microwave oven using a citric acid solution (pH 9.0). Slides were blocked in saline containing 20% horse serum, followed by treatment (15 minutes) with a blocking solution, and an overnight incubation with the primary antibody. Thereafter, sections were incubated for 90 minutes with a biotinylated secondary antibody, followed by a Cy3-conjugated Streptavidin. DAPI was used to visualize nucleal. All slides were examined using a uorescence microscope. Positive cells were counted using the Image Pro Plus software. For Picrosirius red staining, sections from tumor tissues were stained with 0.1% Picrosirius red and quanti cation was done as above.

Invadopodia assays
Gelatin was labeled using Alexa Fluor 488 Protein Labeling Kit (Molecular Probes, Thermo Fisher Scienti c). Glass-bottomed well plates were coated as mentioned above, with ratios of 1:10 labeled gelatin:nonlabeled gelatin. Cells were plated on the gelatin matrix and cultured for varying lengths of time (on average, 5-6 hours). Next, cells were xed and stained for actin and DAPI, and degraded areas were assessed. Cells in every eld of view were counted and the degradation area was calculated by the uorescently labeled gelatin channel, using Analyze Particles plug-in (ImageJ software). Total degradation area/cell (μm 2 ) was used to assess invadopodia.
Rho GTPase activation assays G-LISA Small GTPase activation assays were performed by following the protocol in the activation assay kit (from Cytoskeleton).

ELISA tests
Human VEGF DuoSet, TGFB1-2 DuoSet and Human TGF-beta RII DuoSet ELISA's were purchased from R&D Systems and assays were carried out as per the manufacturer's instructions.

Luciferase-reporter assays
Assays was performed as previously described 53 . Brie y, cells were co-transfected with a luciferase reporter plasmid, along with control plasmids (Promega, Madison, WI). Luciferase activity was determined using the dual-luciferase reporter assay system (Promega). Fire y luciferase luminescence values were normalized to Renilla luminescence.

RNA Isolation, real-time PCR analysis and RNA sequencing
The TRIzol reagent (Life Technologies) and the PerfectPure RNA Cultured Cells kit (5 Prime) were used for RNA puri cation and real-time quantitative PCR. Generation of cDNA was performed using either the qScript cDNA Synthesis kit (Quanta), High-capacity cDNA Reverse Transcription kit (Applied Biosystems), or RevertAid Reverse Transcriptase (Thermo Scienti c). Real-time qPCR analysis was performed using Fast SYBR Green Master Mix (Applied Biosystems). Primers were designed using PrimerBlast. Transcripts encoding beta-2 microglobulin (B2M) and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) were used for normalization. Illumina HiSeq 2500v4 was used for RNA sequencing (~40 million reads per sample). Poly-A/T stretches and Illumina adapters were trimmed from the reads using Cutadapt. Reads were mapped to the Homo Sapiens GRCm38 reference genome using STAR 54 . Expression levels were quanti ed using htseq-count 55 . Differentially expressed genes were identi ed using DESeq2 56 while employing betaPrior, cooksCutoff and independent Filtering parameters set to False. Raw P values were adjusted for multiple testing using the procedure of Benjamini and Hochberg 57 .
Extraction and sequencing of cytoplasmic and nuclear RNA RNA extraction and analysis were performed as previously described 44 . Brie y, Cells were washed in cold PBS and detached from plates using 10mM EDTA. A fraction was transferred to a new tube and RNA was extracted with TRIREAGENT (MRC). Remaining cells were washed in cold PBS, resuspended in RLN buffer (50mM Tris-HCl pH8, 140mM NaCl, 1.5mM MgCl 2 , 10mM EDTA, 1mM DTT, 0.5% NP-40, 10U/ml RNase inhibitor), and incubated on ice for 5 min. The extract was centrifuged for 5 min at 300Xg in a cold centrifuge, the supernatant was transferred to a new tube and centrifuged again for 1 min at 500g in a cold centrifuge. The supernatant (cytoplasmic fraction) was transferred to a new tube and RNA was extracted using TRIREAGENT. The nuclear pellet was washed once in RLN buffer, resuspended in 1ml of buffer S1 (250mM Sucrose, 10mM MgCl 2 , 10U/ml RNase inhibitor), layered over 3ml of buffer S3 (880mM sucrose, 0.5mM MgCl 2 , 10U/ml RNAse inhibitor), and centrifuged for 10 min at 2800Xg in a cold centrifuge. The supernatant was removed and RNA was extracted from the nuclear pellet using TRIREAGENT. Fractionation quality was validated by qRT-PCR using primers for ACTB and MALAT1, which are expected to be enriched in the cytosolic and the nuclear fractions, respectively. WCE, cytosolic and nuclear fractions obtained after fractionation (from two biological replicates) were used to generate cDNA libraries using the SENSE mRNA-Seq Library Preparation kit (Lexogen) according to manufacturer's protocol and sequenced on a NextSeq 500 machine to obtain 75 nt single-end reads. RNA-seq reads were mapped to the human genome (hg19 assembly) with STAR and gene expression levels were quanti ed using Bowtie2 and RSEM 58 . GENCODE v26 annotations were used for further analysis. Differential expression in WCE samples was computed using DESeq2 with default parameters. Subcellular localization was quanti ed similarly using DESeq2, using gene-level RSEM output.

Single Molecule FISH
The protocol and probes we used have previously been described 44 59 . We performed automatic 2D projections as suggested in FishQuant documentation, followed by automatic cell segmentation using CellPro ler 60 . Hoechst signal was used to segment nuclei and the oligo-dT signal was used to segment cell bodies. Following batch analysis, we manually examined segmentation and removed incorrectly segmented cells from further analysis using Fiji (ImageJ) software. Quanti cation of cytoplasmic and nuclear signals was performed with default parameters and recommended lters of FishQuant.
In vitro knockout of the gene encoding for NUP93 The CRISPR system was used as described 61 to create a double-stranded break next to the ProtospacerAdjacent Motif (PAM) sequence. The target site was selected from the ENSEMBL database in a way that targeted the transcript of NUP93. The selected target was 21bp long, including the PAM sequence in exon 5, which was ltered to minimize off-target cross-reactivity. The sequence was cloned in pX458 and cells were sorted using cytometry.

Proteomic analysis
Proteins were digested and puri ed on carboxylated beads according to the SP3 protocol 62 . Peptides were analyzed by liquid-chromatography using the EASY-nLC1000 HPLC coupled to high-resolution mass spectrometric analysis on the Q-Exactive HF mass spectrometer (
The data was ltered to include proteins with valid values in at least 75% of the samples. Missing values were then imputed to represent low abundance proteins by replacing them with random, low intensity values that form a normal distribution. Student t-test was performed (permutation-based FDR 0.05) to distinguish the signi cantly regulated proteins between the control and the NUP93 knockdown samples.     In addition, lungs were removed and evaluated for metastases (J). Image analysis utilized ImageJ. The data are presented as average numbers of micro-metastases ± SEM. Scale bar, 0.5 cm.    for TGF-beta and extracts were tested for TGFBRII using ELISA.   High NUP93 stabilizes beta-catenin as well as enhances its transcription and nuclear transport, while low

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.