UPS19 modulates cancer cell migration and invasion and acts as a novel prognostic marker in patients with early breast cancer

Tumor cell dissemination in cancer patients is associated with a significant reduction in their survival and quality of life. The Ubiquitin-Proteasome System (UPS) plays a fundamental role in the maintenance of protein homeostasis both in normal and stressed conditions and its dysregulation has been associated with malignant transformation and invasive potential of tumor cells, thus highlighting its value as a potential therapeutic target. In order to identify novel molecular targets of tumor cell migration and invasion we performed a genetic screen with an shRNA library against UPS-related genes. To this end, we set up a protocol to specifically enrich positive migration regulator candidates. We identified the deubiquitinase USP19 and demonstrated that its silencing reduces the migratory and invasive potential of highly aggressive breast cancer cell lines. We extended our investigation in vivo and confirmed that mice injected with USP19 depleted cells display increased tumor-free survival, as well as a delay in the onset of the tumor formation and a significant reduction in the appearance of metastatic foci, indicating that tumor cell invasion and dissemination is impaired. In contrast, overexpression of USP19 increased cell invasiveness both in vitro and in vivo, further validating our findings. More importantly, we demonstrated that USP19 catalytic activity is important for the control of tumor cell migration and invasion. Finally, we showed that USP19 overexpression is a surrogate prognostic marker of distant relapse in patients with early breast cancer. Altogether, these findings demonstrate that USP19 might represent a novel therapeutic target in breast cancer.


INTRODUCTION
Cell migration plays a crucial role in a wide variety of physiological processes such as development, tissue injury and wound healing [1]. Its activation is highly regulated both spatially and temporarily, contributing to the maintenance of tissue and cellular homeostasis. Therefore, it is not surprising that when deregulated, migration is associated with the development and progress of multiple pathologies, including cancer [2][3][4].
During the development of malignant tumors, transformed cells change and acquire the ability to invade and abandon their original position. In order to do so, they need to pierce the surrounding extracellular matrix (invasion) and reach the circulatory torrent (intravasation). If they survive, some will be able to disseminate and get through distant capillary walls (extravasation), to invade the extracellular matrix in a new host environment, establishing a secondary tumor (colonization). During this process, cells acquire the ability to proliferate in an anchorage-independent manner with elevated invasive potential [5]. Cell motility plays a vital role in many of these events and therefore has a major influence in metastatic cell dissemination. Particularly in cancer, alteration or exacerbation of malignant tumor cell migration and dissemination is the principal cause of death due to solid tumors [6].
In addition, it was observed that decreasing the migratory capabilities of tumor cells can restore a certain level of sensitivity to cytotoxic reagents and increase the susceptibility to chemotherapeutic treatments [7,8]. Consequently, targeting genes that regulate cell motility could be beneficial in the treatment of highly aggressive cancers [9,10], and anti-migratory or anti-invasive activity is usually viewed as a desired attribute for novel anticancer drugs [11].
This complex process requires posttranslational regulation of an innumerable amount of proteins, modifying their biological function, subcellular localization or half-life. Ubiquitination is an important form of protein post-translational modification that consists in the conjugation of ubiquitin polypeptides to target proteins. Ubiquitin tagged proteins are either subjected to destruction or responsible for regulating different processes, including endocytosis, DNA repair, cell cycle regulation, and gene expression [12]. The process of ubiquitin chain conjugation and elongation are regulated by a complex enzymatic cascade [13]. Also, like most post-translational modifications, the addition of poly-ubiquitin chains can be reversed or modified. This process is carried out by deubiquitinating enzymes (DUBs), a family of proteins including approximately 100 members classified into few sub-groups according to their catalytic core domains [14]. The ubiquitination cascade is of vital importance for the maintenance of cellular homeostasis by regulating a wide variety of processes, including cell migration and invasion [12,15]. Therefore, within this framework, we conducted a loss-of-function genetic screen in order to identify novel molecular targets within the Ubiquitin-Proteasome System (UPS) that positively regulate migration. Screens targeting multiple genes significantly shortens experiment duration [16], and therefore they have been widely and successfully used to find genes involved in many physiological and pathological cellular processes [16][17][18][19][20], as they are relatively easy to perform and implement. In this study, we used a pooled shRNA interference library and an immortalized tumorigenic mammary epithelial cell line, MDAMB231, derived from a human triple-negative breast cancer. This type of cancer lacks expression of the estrogen receptor, progesterone receptor and human epidermal growth factor receptor 2 (HER2) and is associated with aggressive behavior and an overall poor prognosis [21]. Currently, there are no specific targeted therapeutic systemic alternatives for its treatment [22][23][24], and optimal chemotherapy regimens have yet to be established.
From our screen, we identified the Ubiquitin specific protease 19 (USP19) as a new candidate gene associated with the regulation of cell migration and invasion. This DUB is associated with protein quality control and cellular homeostasis [25][26][27][28][29].
To validate its novel function as a positive regulator of migration and invasion, we performed a series of in vitro and in vivo experiments analyzing USP19's role in colonization and tumor formation. In addition, we showed that USP19 overexpression is associated with distant relapse in patients diagnosed with early breast cancer (T1-2, N0, M0). Collectively, our data suggest that USP19 plays a crucial role in breast cancer cell dissemination, and we provide novel evidence that it can be a prognostic marker and attractive candidate for the development of new therapeutic strategies in breast cancer.

Migration-based screen to identify UPS genes with novel regulatory functions
In order to identify novel positive regulators of cell migration within the UPS, we performed an shRNA-based functional selection screen (Fig. 1A). A pooled recombinant lentiviral shRNA library targeting over 400 human UPS related genes (≈ 5 shRNAs per gene) was stably transduced into breast cancer cells. The functional selection consisted in placing the mixed population into the upper compartment of a transwell unit and allowing migration through the perforated membrane to the lower compartment. Cells that exhibited reduced migration were isolated from the upper compartment and amplified. We performed subsequent enrichment cycles until shRNAtransduced cells lost about 80% of their initial migratory potential (Fig. 1B). After every enrichment cycle, we evaluated shRNAs relative abundance in the cell population by PCR amplification and quantitative sequencing from genomic DNA. As shown in Figure 1C, as enrichment cycles increased, we observed a marked reduction in the number of shRNAs, suggesting that the selection process was efficient. As a control, we used an empty vector-transduced cell line.

Selection of candidate genes
After the selection process, we followed an analytical workflow to select candidate genes for further validation ( Fig. 2A). In order to avoid false positives due to off-target effects, we discarded those genes for which only one shRNA targeting its sequence was found in the sequencing results. These criteria allowed us to identify 30 genes whose depletion impaired migration. Half of these genes had already been associated with migration, invasion, metastasis or tumorigenesis, and served as a proof of principle for the efficacy and specificity of our screen. Among the identified candidates, we focused our attention on the study of the deubiquitinase USP19.

Validation of USP19 as a regulator of cell migration
In order to validate USP19 as a potential regulator of cell migration, we established stable MDAMB231 cell lines transduced individually with two different shRNAs targeting USP19 expression, Figure 1| shRNA-based selection of positive regulators of cell migration.
(A) Overview of the selection procedure. The production and infection of a ubiquitination-related lentiviral shRNA library are described in Methods. Two weeks after lentiviral infection and selection, MDAMB231 cells were seeded onto transwell inserts and allowed to migrate across the porous membrane for 24 hours in order to select cells with a decreased migration phenotype. Migrating cells were removed and non-migrating cells were collected from the inserts upper compartment and amplified. Cells were then reseeded onto transwell culture inserts for a subsequent cycle of selection; this procedure was repeated until cells lost 80% of their initial migratory potential. After every cycle of selection, the relative abundance of the different shRNAs was evaluated using Next Generation Sequencing. (B) Transwell assay was used every other enrichment cycle to determine the percentage of migratory cells and monitor the selection process. (C) shRNAs' abundance was estimated after each selection cycle.  Wound healing *** ** **** **** named shRNA#1 and shRNA#2. Our results showed that both caused a significant reduction in USP19 mRNA and protein levels (Fig. 2B).
It is conceivable that shRNAs promoting cell proliferation may have also been enriched during the functional selection, as they gave the cells an advantage during the in vitro amplification step. To discard this possibility, we performed proliferation curves of control and USP19-silenced cell lines and calculated doubling rates. We observed no differences between the different cell lines (Fig. 2C, Supp. Fig. 1), providing evidence for a direct role of USP19 in the control of cell migration.
Next, to confirm the effect of USP19 depletion on cell motility, we used two independent assays: transwell migration and wound-healing assays. As shown in Figures 2D and 2E, USP19 knockdown significantly decreased the migratory potential of cells relative to the control cell line.
Altogether, these experiments indicate that USP19 silencing affects cell migration in vitro. We further confirmed our findings using another highly invasive breast cancer cell line (Supp. Fig. 2).

USP19 knockdown impairs invasion
Cell motility is often associated with increased tumor cell invasion and is a characteristic trait of aggressive tumor cells [30,31]. Based on the observation that modulation of USP19 expression regulates migration, we decided to investigate the effect of USP19 depletion on tumor cell invasion.
To this end, we used three different experimental approaches. We first analyzed the ability of cells to invade agar spots. Our results show that USP19 knockdown significantly reduced the number of invading cells as well as their total displacement, compared to the control cell line (Fig.  3A).
We next performed a 3D growth assay by seeding cells at low confluence into noble agar, an anchorage-independent matrix. After 6 weeks in culture, the control cell line formed bigger colonies than USP19-silenced cell lines (Fig. 3B), indicating that colonization and matrix invasion in these conditions is partially impaired in cells where USP19 expression is reduced.
Finally, to further characterize USP19 depletion on tumor cell invasion, we assessed growth and invasion into a reconstituted extracellular matrix that provides anchorage (Matrigel®). Both cell lines expressing USP19 shRNAs showed colonies with a significantly smaller size than the control cell line (Fig. 3C), indicating that USP19 is required for efficient invasion even when anchorage is provided.
Similarly to the migration experiments, we further validated our results using another breast cancer cell line (Supp. Fig. 2).
Collectively, our results indicate that USP19 knockdown inhibits MDAMB231 cell invasion in vitro.

USP19 overexpression enhances migration and invasion
In order to complement our analysis on the putative role of USP19 as a positive regulator of migration and invasion, we analyzed the effect of USP19 overexpression in a poorly migratory and non-invasive breast cancer cell line (MCF7).
For this purpose, we stably transfected MCF7 cells with a USP19 overexpressing plasmid (Fig.  4A), and then performed wound healing assays. As shown in Figure 4B, USP19 overexpression induced a significant increase in the gap covered area, compared to the control cell line.
As a control, we overexpressed a catalytically mutant version of USP19, and in contrast to USP19 wild type, we did not detect an increase in migration compared to the control cell line.
This result further supports the hypothesis that USP19 is a positive regulator of migration, and it provides evidence that this phenotype is dependent on its catalytic activity.
Doubling time was calculated for control and USP19 silenced cell lines on the bottom (n³ 3, Kruskal-Wallis, Dunn's multiple comparison test. shRNA#1 p> 0.9999 and shRNA#2 p> 0.9999). Migratory potential was evaluated by two different experiments. (D) Transwell assay: After 24 hours of incubation, USP19-depleted cells were stained for microscopic examination and the number of migratory cells was compared to control cells. The graph shows the number of migratory cells per transwell membrane (n= 4, one-way ANOVA, Dunnett's multiple comparison test. shRNA#1 p= 0.0187 and shRNA#2 p= 0.0001). (E) Wound healing assays: scratching with a pipette tip made a gap on a monolayer of the different cell cultures, and time-lapse imaging monitored the number of migrating cells across the border. After 8 hours, cells exhibited different levels of migration. The graph on the left shows the gap covered area (mm 2 ) after 8 hours (n= 16, one-way ANOVA, Dunnett's multiple comparison test. shRNA#1 p< 0.0001 and shRNA#2 p< 0.0001) and the images on the right show representative areas in a wound healing experiment at the indicated time points. Scale bar= 100 μm. Next, we analyzed the effect of USP19 overexpression on invasion and growth into a reconstituted extracellular matrix (Matrigel®), using similar experimental settings as described before. We observed a significant increase in colony areas when comparing wild type USP19 overexpressing cells to the control cell line (Fig. 4C). In accordance with our previous results, the USP19-dependent increase in invasion is also determined by its catalytic activity (Fig. 4C).

USP19 regulates invasion in vivo
To further characterize USP19-dependent control of cell invasion in vivo, we performed subcutaneous orthotopic xenotransplants and experimental metastasis assays in immunocompromised mice (NOD/SCID).
First, we injected MDAMB231 control or USP19-silenced cells subcutaneously in the mammary fat pad of female mice and monitored tumor growth every 2-3 days. Tumor growth curves analysis indicated that those generated from MDAMB231 control cells were significantly more volumetric than the ones originated from USP19silenced cells (Fig. 5A, left and Supp. Fig. 3). Moreover, Kaplan-Meier curves for tumor-free survival indicated that cells expressing either of the shRNAs targeting USP19 generated fewer tumors compared to the control cell line (Fig. 5A, right and Supp. Table 1). In addition, we observed similar results using another breast cancer cell line (Supp. Fig. 4 and Supp. Table 1).
Second, we analyzed USP19's role in the regulation of tumor cell lung colonization. For that purpose, we inoculated control or USP19-silenced MDAMB231 cells through tail vein injection and harvested the lungs two months later. As shown in Figure 5B, USP19 depletion inhibits tumor foci formation in vivo, as evaluated by human DNA quantification (left) and metastatic load quantification in Hematoxylin & Eosin stained lung sections (right). We observed the same trend when another breast cancer cell line was used (Supp. Fig.  4).
Last, we repeated the same type of tests using MCF7 cells in similar experimental conditions. We subcutaneously injected control cells or cells expressing either wild type or catalytically mutant versions of USP19 in female mice.
In agreement with our in vitro experiments, wild type USP19-expressing cells formed tumors in all injected mice, whereas mice injected with cells expressing the catalytic mutant did not show signs of tumor growth (Fig. 5C, Supp. Fig. 3 and Supp. Table 1). This latter observation is consistent with the fact that the MCF7 cell line does not usually form tumors unless an external estrogen source is supplied, highlighting the importance of USP19 for tumor development and onset.
Altogether, we concluded that USP19 is important for in vivo colonization and tumor growth.
In addition, our results indicate that USP19 catalytic activity is required for its stimulatory effect on cell motility.

Survival analysis of USP19 expression in early breast cancer patients
Finally, we analyzed USP19 protein expression in a cohort study of early breast cancer patients (T1-2, N0, M0; n= 168) with long-term follow-up. Kaplan-Meier plots showed that overexpression of USP19 was associated with a significantly higher frequency of distant relapse (distant relapse free survival, DRFS), while no significant correlation with disease free survival (DFS) was observed ( Fig. 6A and B).
Multivariate analysis of DRFS, adjusted for other prognostic factors, revealed that USP19 High was an independent prognostic predictor of DRFS (Table 1).
Altogether these findings indicate that USP19 is a new predictor of distant metastasis formation in early breast cancer patients.

DISCUSSION
Migration occurs in a wide variety of physiological conditions, and alterations in its regulation are associated with different pathologies, including cancer [2,3,32]. In this disease, mortality is associated primarily with tumor growth at secondary sites, and effective therapies to block the metastatic cascade are lacking [6]. Tumor cells need to migrate, invade and colonize new niches prior to metastasis, making it a vital trait of malignancy. Indeed, a recent work indicated that migration, rather than proliferation, is strongly associated with breast cancer patient survival [33]. Therefore, modulation of genes that regulate migration and invasion could find application for the treatment of cancer. In line with this reasoning, we chose to screen for genes that positively regulate motility within the UPS, as this cascade is currently emerging as an attractive therapeutic target in drug development [34][35][36][37]. Here we report the identification of USP19, a deubiquitinating enzyme, as a positive regulator of migration. USP19 was initially characterized as a DUB predominantly localized in the cytosol in association with Hsp90 and other chaperones [28]. USP19 has been associated with the regulation of the half-life of several proteins that participate in different cellular processes [29,[38][39][40][41][42][43][44][45][46][47][48][49][50]. To our knowledge, it has not been associated with migration or invasion yet.
Our in vitro validation experiments showed that USP19 depletion did not affect cell proliferation in agreement with Lu et al. [51], but directly inhibited cellular migration. In addition, we observed that USP19 knockdown impaired invasion, as evaluated by agar drop assays and three-dimensional basement membrane cultures (Figs. 2, 3 and Supp. Fig. 2). When we analyzed growth into noble agar, the size of the resulting colonies was smaller than control cells, indicating that USP19 silencing also reduced anchorage-independent growth (Fig. 3). In all cases, we observed a correlation between the extent of USP19 silencing and the reduction of migration and invasion potential. These results are in agreement with USP19-silencing deleterious effects in growth and development in Zebrafish embryos [52].
To further confirm our results, we analyzed how USP19 overexpression affected migration and invasion, using a poorly migratory cell line. In agreement with our depletion experiments, USP19 overexpression induced an increase in cellular migration, invasion and growth in three-dimensional basement membrane cultures. These effects were dependent on the presence of a highly conserved cysteine at the catalytic site and mutation of this residue abolished USP19-induced migration and invasion. Taken together, these in vitro results suggest that USP19 expression levels are associated with the regulation of motility, invasion, and anchorage-independent growth in breast cancer cell lines.
Our in vivo studies using immunocompromised mice demonstrated that USP19 silencing decreased cell engraftment and tumor growth, as well as colonization into the lungs (Fig. 5A and Supp. Fig. 4B). On the contrary, overexpression of wild type USP19, but not its catalytically deficient mutant version, promoted tumor growth (Fig. 5B). This is compatible with the requirement of USP19 catalytic activity for local invasion and growth in three dimensions, both in vitro and in vivo.
Finally, a retrospective study conducted on human breast tumor samples indicated that high USP19 protein levels are associated with high-risk for metastatic relapse in patients diagnosed with early breast cancer (Fig. 6).
Altogether these results provide evidence indicating that USP19 has great potential as a therapeutic target for drug development in breast cancer treatment.
In this regard, there is considerable scientific evidence demonstrating that DUBs exhibit strong substrate selectivity, which can be advantageous to ensure high efficacy and low adverse effects. Moreover, the design and development of a selective enzyme inhibitor is easier than generating an enzyme activator due to competitive inhibition and modeling of substrates [53]. In fact, numerous inhibitors for DUB activities have been already identified [54][55][56][57][58]. All this evidence indicates that drug development to inhibit USP19 activity is plausible. Indeed, an orally bioavailable compound has been developed, but it yet remains to be preclinically tested for the treatment of cancer [59].
USP19 molecular mechanism of action in the regulation of migration and invasion remains to be determined. USP19 could either affect migration and invasion directly, or alternatively, it could induce an invasive mesenchymal-like phenotype, both compatible with the observed aggressive behavior in tumor growth and invasion.
Supporting the latter, we analyzed e-cadherin protein expression levels in the same breast cancer samples described before and observed an inverse correlation between USP19 and e-cadherin expression (n= 168, Spearman correlation analysis, rho= -0.180, p= 0.032). In agreement with our results, previous reports demonstrated that low e-cadherin expression holds a prognostic value as a predictor of poorer prognosis and more aggressive phenotypes in breast cancer [60,61]. Moreover, we performed an in silico analysis on breast cancer mRNA expression publicly available datasets, which revealed that high USP19 expression levels correlate with the activation of the Wnt pathway (Supp. Fig. 7). This is consistent with a recent work which showed that USP19 regulates LRP6 stability, a co-receptor of the Wnt signaling cascade [44]. Aberrant activation of this pathway is associated with cancer, and it correlates with poor prognosis and metastasis development in breast cancer [62][63][64][65]. On the other hand, USP19 was identified as a putative regulator of the transforming growth factor-β (TGF-β) type I receptor (TβRI) stability [66], whose deregulation is associated with numerous pathologies, including breast cancer [67][68][69][70][71][72].
For all the reasons expressed before, we conclude that USP19 is relevant for the regulation of breast cancer cell dissemination and its expression levels correlate with high risk of metastases development, and therefore it could represent a novel target for the management of breast cancer metastatic disease.

ACKNOWLEDGEMENTS
This work was supported by grants from the Agencia Nacional de Promocioń Científica y Tecnoloǵica (ANPCyT) PICT 2011-2783, PICT2016-2620 and PICT 2018-03688 awarded to MR. JEM was supported by NIH R01CA117907, NIH R01GM120109, NSF MCB-1817582 and NIH P30CA046934 grants, and GS was funded by an AIRC grant: IG2016 id18467. FAR and JHES are postdoctoral fellows of Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET), and EHCR was recipient from a fellowship from Instituto Nacional del Cańcer (INC).

AUTHOR CONTRIBUTIONS
FAR conceived, designed, and administered the study; acquired, interpreted, and analyzed data, and wrote the manuscript with assistance from JHES and MR. JHES and EHCR were responsible for acquisition, analysis, and interpretation of data.
MJ prepared the shRNA library. AP processed screen sequencing data. MAC performed in silico pathway activation analysis. BD and GS performed mouse tissue staining, measurements, and data analysis. SB, VDL, GS and RL performed patients' biopsies staining, data collection and analysis. JEM provided resources, helped in funding acquisition and coordinated the NGS sequencing and analysis processes.
MR conceptualized, designed, supervised, and administered the study, reviewed the manuscript, and acquired financial support for the project leading this publication.

Cell lines and cell culture
The human breast cancer cell lines MCF7, MDAMB231 and MDAMB436 and the Hek293T cells were obtained from the ATCC and cultured in Dulbecco's modified Eagle's medium (DMEM) (Gibco) supplemented with 10% fetal bovine serum (FBS) (Natocor, Córdoba, Argentina), 50U/ml penicillin-streptomycin and 200µM L-glutamine at 37°C and 5% CO2 in a humidified incubator. ATCC uses morphology, karyotyping, and PCR based approaches to confirm the identity of human cell lines. Mycoplasm contamination was evaluated monthly by PCR, and cell lines were cultured less than three months.

shRNA screening and plasmid transfections
A pool of plasmids encoding 1,885 shRNAs targeting 407 different genes related to the ubiquitination pathway (UB/DUB library) in the pLKO.1 backbone produced by The RNAi Consortium (TRC, Sigma-Aldrich, St. Louis, MO) were obtained from the University of Colorado Cancer Center Functional Genomics Shared Resource. 1 mg of the shRNA library plasmid DNA at 100 ng/mL was mixed with 4 mg of packaging plasmid mix (pD8.9 and pCMV-VSVG lentiviral packaging plasmids at a 1:1 ratio) and incubated with 30 mg of polyethylenimine for 15 min at RT. The entire mixture was then added to a 100 mm dish containing Hek293T packaging cells at 75% confluence. 6 hours after transfection, media on cells was replaced with complete DMEM and 48 hours after media replacement, the supernatant from each dish of packaging cells (now containing lentiviral library particles) was filtered through 0.45 mm cellulose acetate filters and stored at -80ºC until use. Before performing the screen, MOI determination of the lentivirus stock was carried out using different dilutions. Target cells were seeded at 8x10 4 cells/well in 6-well plates and then transduced with the lentivirus. After 48 hours the infective media was removed, and target cells were selected for 5 days with a 0.5 μg/ml puromycin DMEM medium. The amount of virus required to maintain 10% survival was used. These infection conditions were essential for each cell to be transduced with less than one lentivirus particle expressing a single shRNA. Target cells were transduced with a 1:27 dilution of the lentivirus stock and were then combined at the time of harvest to reach a starting number of 4,4x10 6 cells per condition (~300X coverage of the library complexity). 48 hours after transduction, the media was replaced with puromycin selection media and cells were then propagated for 14 days before use, in order to select out those shRNA targeting essential genes.
For single shRNA transduction, SHC001, TRCN0000051715 and TRCN0000051716 PLKO.1 vectors were used (obtained from the University of Colorado Cancer Center Functional Genomics Shared Resource); no MOI determination was performed.
For overexpression experiments, transfections were performed using Lipofectamine 2000 reagent (Invitrogen, Carlsbad, CA) according to the manufacturer's protocol. GFP tagged USP19 plasmids were a kind gift of Dr. Sylvie Urbé (University of Liverpool, UK).

Transwell migration assay
All cell lines were starved during 24 hours in assay medium (growth medium containing 0.1% FBS). The starved cells were trypsinized, 5x10 4 cells were added to the top chamber of 24-well transwells (8 μm pore size membrane; BD Bioscience, Bedford, MA), and assay medium was added to the bottom chambers and incubated for 24 hours. After non-migratory cells removal, membranes were fixed, stained with 4′,6-diamidino-2-phenylindole, and mounted. The whole membrane was then imaged using a Zeiss Axio Observer Z1 Inverted Epi-fluorescence microscope with montage function. Image analysis was performed with Fiji software, using an automated analysis macro to measure the number of nuclei per transwell.
For the migration screen, 6.6x10 5 starved MDAMB231 cells expressing the shRNA library (or control cell line) were plated onto 6-well transwell chambers (8 μm pore size membrane; BD Bioscience, Bedford, MA) in assay medium. After a 24-hour incubation, the non-migratory cells were collected from the upper chamber, propagated and allowed to re-migrate eleven times for enrichment purposes (the non-migratory cells of each migration experiment were used for the subsequent cycle of enrichment). The non-migratory cells of 8 transwells were combined per each selection cycle to ensure a > 700 library coverage. Simultaneously, the percentage of non-migratory cells in each cycle was determined in 24-well plates as described before, using 8 μm pore size membrane transwells (BD Bioscience, Bedford, MA).

shRNA library preparation and sequencing
The library preparation strategy uses genomic DNA and two rounds of PCR in order to isolate the shRNA cassette and prepare a single strand of the hairpin for sequencing by means of an XhoI restriction digest in the stem-loop region. We used barcoded half-hairpin sequences for the identification of shRNAs from every enrichment cycle. The procedure was performed as described previously [73]. After purity analysis of the sequencing library, barcode adaptors were linked to each sample to allow a multiplexing strategy. A HiSEQ 2500 HT Mode V4 Chemistry Illumina instrument was used for that purpose and each sample was quantified and mixed together at a final concentration of 10 ng/mL. Samples were sequenced with a simple 1x50 run and on average 1.2x10 6 reads were obtained per sample (> 600X shRNA library complexity).

shRNA screen analysis
shRNA data were analyzed in a similar fashion to RNA-seq data. Briefly, quality control was performed with FastQC, reads were trimmed to include only shRNA sequences using FASTQ trimmer and filtered with the FASTQ Quality Filter. Reads were then aligned to a custom reference library of shRNA sequences using TopHat2.

Quantitative PCR
Total RNA was extracted from cell lines using TRIzol reagent (Invitrogen, Thermo Fisher Scientific) according to the manufacturer's instructions and complementary DNA synthesis was carried out using M-MLV reverse transcriptase in the presence of RNasin RNase inhibitor (Promega) and an oligo(dT) primer (Invitrogen).
Total DNA was extracted from cell lines or fixed tissues using Qiagen DNeasy Blood and Tissue kit with the optional RNAse A treatment step and following the manufacturer's instructions.
Quantitative real-time PCR amplification, using specific primer sets at a final concentration of 300 nM, was carried out using the FastStart Essential DNA Green Master kit (Roche) at an annealing temperature of 60°C for 35 cycles, and a CFX96 PCR Detection System (Biorad). Expression was calculated for each gene by the comparative CT (ΔCT) method with GAPDH for normalization.
Sequences and expected product sizes are as follows:

Crystal violet proliferation assay
1x10 4 cells were seeded in 96-well plates at time= 0 hours in quadruple. Every 24 hours, medium was removed at each time point, cells were rinsed with 1X PBS and fixed with methanol for 15 minutes. After 4 days, all wells were staining simultaneously with 150 μl of a 0.05% crystal violet solution for 15 minutes. Plates were then rinsed with water, dried out and the remaining crystal violet was solubilized in 150 μl methanol. The absorbance was measured at 595 nm in a plate reader.

Area-based analysis of proliferation rate
This experiment was performed as described previously, with minor modifications [74]. Briefly, 1x10 3 cells were seeded in 96-well plates and incubated overnight to allow cell attachment. Images were then acquired under bright field illumination every 6 hours for 3 days using a 10X air objective and Zeiss Zen Blue 2011 software for image acquisition. Image analysis was performed with Fiji software, using an automated analysis macro to measure the occupied area by cells.

Wound healing assays
Cells were seeded onto 24-well culture plates at 2,5x10 5 per well in growth medium. Confluent monolayers were starved during 24 hours in 0.1% FBS supplemented DMEM medium and a single scratch wound was created using a micropipette tip. Cells were washed with PBS to remove cell debris, supplemented with 3% DMEM medium, and incubated at 37°C under 5% CO2 to enable migration into wounds. Images were acquired with a Zeiss Axio Observer Z1 Inverted Epi-fluorescence microscope equipped with an AxioCam HRm3digital CCD camera; a Stage Controller XY STEP SMC 2009 scanning stage and an Incubator XLmulti S1 and Heating Unit XL S1 for live imaging incubation. Images were acquired under bright field illumination every 15 minutes for 8 or 12 hours for MDAMB231 or MCF7 cells, respectively, using a 10X air objective and Zeiss Zen Blue 2011 software for image acquisition. Image analysis was performed with Fiji software, using an automated analysis macro to measure the occupied area by cells, and the results are presented as the wound covered area at the end of the experiment, relative to time= 0 hours.

Agar invasion assay
The procedure was performed as described previously, with minor modifications [75][76][77]. A 1% noble agar solution was heated until boiling, swirled to facilitate complete dissolution, and then taken off of the heat. When the temperature cooled to 50°C, 5 μL spots were pipetted onto 96 well cell culture plates and allowed to cool for 20 min at RT under the hood. At this point, 5x10 3 cells were plated into spot-containing wells in the presence of 10% FCS cell culture media supplemented with 1 μg/ml Hoechst 33258 (Thermo Fisher Scientific) and allowed to adhere for 1 hour. Fluorescent images of the edges of each spot were taken every 20 minutes during 18 hours on an Axio Observer Z1 (Zeiss) Fluorescence Microscope using a 10X magnification air objective, equipped with CCD Axio Cam HRm3 digital Camera, and a XL multi S1 (D) incubation unit plus a XL S1 (D) temperature module to maintain cell culture conditions at 37°C and 5% CO2. Acquisition was controlled with Zen Blue 2011 (Zeiss) Software. Graph construction and statistical analysis were performed using GraphPad Prism. The number and position of cells were determined using image analysis software, ImageJ/Fiji -'Trackmate' plug-in [78].

Noble agar assay
This experiment was performed as previously, with minor modifications [79]. A 2 ml mixture of 5,000 cells in assay medium and 0.3% noble agar was seeded onto a 4 ml-solidified bed of 0.6% noble agar on six-well plates. The plates were allowed to solidify and were then incubated at 37°C. The cultures were fed once a week with assay medium. The cultures were imaged, and the number of colonies was counted after 6 weeks.

Matrigel three-dimensional cell culture
Experiments were carried out based on experimental settings described before [80][81][82][83]. Briefly, 1,000 cells were cultured during the length of the experiment in 100 μl basement membrane gels composed of 9.2 μg/μl phenol red-free growth factor reduced Matrigel (BD Bioscience) in 96 well plates. 50 μl of fresh medium were added on top of each gel every three days. For in vivo mouse tumor studies, 5x10 5 transduced cells were suspended in 100 μl of sterile 1X PBS and subcutaneously injected in the mammary fat pads of female mice. Tumors were measured every 3 days and tumor volumes were calculated using the following formula: Vol (volume)= ½ (width 2 * length). Area Under Curve analysis was performed using measurements from mice that were alive at the end of the experiment.

Mouse tumorigenesis and metastasis models
For the experimental metastasis assay, 1x10 6 cells were suspended in 200 μl of sterile 1X PBS and injected in the lateral tail vein of male mice. Lungs were harvested 60 days post-injection, fixed in buffered formalin and then stored in 70% ethanol until use for DNA quantification (as described before [84]) or paraffin embedding and Hematoxylin and Eosin staining, or insufflated with a 15% India Ink solution and counterstained with Fekete's solution for macrometastasis exposure and imaging.

Metastatic load quantification of lung Hematoxylin & Eosin (H&E) stained slides
The presence of tumor nodules was identified by scanning individual lung H&E stained slides with an optical microscope. Digital image files were acquired for each specimen. For the analysis of the lung metastatic area, Adobe Photoshop software was used to determine the percentage of the lung section that was occupied by the tumor. Furthermore, since the cell-surface glycoprotein CD44s is constitutively expressed by MDAMB231 cells, immunohistochemical staining with the antihuman CD44s (HCAM) antibody (clone DF1485; heat induced epitope retrieval in citrate buffer pH 6.0; dilution 1:50, incubation time: 60') was used to confirm the tumor origin of the lung nodules (Supp. Fig. 5).

In silico analysis of USP19 mRNA expression among the TCGA-BRCA dataset
Pre-processed USP19 expression levels among 800 primary breast carcinomas with intrinsic subtype data and their integrated pathway activities (pathway activity -z score of 1387 constituent PARADIGM pathways) were obtained from the TCGA Breast Cancer (BRCA) dataset at UCSC Xena browser (http://xena.ucsc.edu/). The PARADIGM algorithm integrates pathway, expression and copy number data to infer activation of pathway features within a superimposed pathway network structure extracted from NCI-PID, BioCarta, and Reactome [85].
Briefly, Luminal A/B primary breast cancer group (n= 600) was divided into low (n= 77) or high (n= 209) USP19 expression levels according to the StepMiner one-step algorithm (http://genedesk.ucsd.edu/home/public/StepMiner/ ). These two groups were then compared at their integrated pathway activities to identify the most relevant signaling pathways associated with USP19 expression using the SAM test (p< 0.01; Fold Change> 1.5) with MultiExperiment Viewer Software (MeV 4.9).

Patients and immunohistochemistry
We retrospectively extracted the eligible patients for the study from a consecutive cohort of cases (year range, 1983-2001) diagnosed with primary unilateral breast carcinoma at the Regina Elena National Cancer Institute, Rome, Italy. From the original series, only N0 patients with T1/T2 tumors were included in this study (n= 168). The patient and tumor characteristics can be found in Supplementary Tables 2 and 3. This study was reviewed and approved by the Ethics Committee of the Regina Elena National Cancer Institute, and written informed consent was obtained from all patients. All the patients were treated with quadrantectomy and received radiation therapy, while 90 (53,6%) received chemotherapy associated or not to hormonal therapy, and 58 (34.5%) underwent only hormonal therapy. Patients with HER2-positive tumors did not receive trastuzumab because it was not available during the study period. The median follow-up was 91.5 months (range, 6 -298 months). Follow-up data were collected from institutional records or referring physicians. During the follow-up, distant relapse was seen in 29 (17.3%) of the patients.
Tissue microarrays (TMA) were constructed by punching 2-mm-diameter cores from invasive breast carcinoma areas, as previously described [86]. TMA sections were incubated overnight with the rabbit anti-USP19 polyclonal antibody (LifeSpan, Cat# LS-C353286, 1:50 dilution) after applying the MW antigen retrieval technique at 750 W for 10 min in 10 mM Sodium Citrate Buffer (pH 6.0). The anti-rabbit EnVision kit (Agilent, CA) was used for signal amplification. For the negative control, the primary antibody was substituted with a rabbit non-immune serum.
The anti-E-cadherin mouse monoclonal antibody (clone HECD-1, 1:50 dilution, 30 min, Zymed Laboratories Inc., San Francisco, CA) was also used. Antigen retrieval was performed as described in the previous paragraph.
The immunohistochemical analysis was carried out by two pathologists (R.L., S.B.) by agreement, with both blinded to the clinicopathological information. The proportion of USP19-positive cells that showed cytosolic positivity was in the range of 4-100%, with a mean ± S.E. of 63.2% ± 3.6. E-cadherin positivity was defined as a membrane-associated, linear pattern of immunoreactivity which decorated the cell membrane entirely.
The immunohistochemical results for the estrogen receptor (ER), progesterone receptor (PR), Ki67, and HER2 status were obtained from the patient hospital records.

Statistical analysis
Results are presented as Box-and-whisker plots with median interquartile ranges plus minimum to maximum. n indicates the number of independent biological replicates. The one-way ANOVA with Dunnett's multiple-comparison test as well as non-parametric Kruskal-Wallis and Dunn's Tests were used to compare treatments to their corresponding control, and adjusted p-values are indicated. Pvalue differences of < 0.05(*), < 0.01(**), < 0.001(***) or < 0.0001(****) were considered statistically significant. GraphPad Prism and SPSS (SPSS version 15.0, Chicago, IL) statistical software were used for the analysis.
The expression of the USP19 protein in patients' samples was reported as percent of positive cells and dichotomized (high vs. low) according to the ROC analysis. The optimal cut-off parameter for USP19 positive expression was 50%. Consequently, tumors were identified as USP19 High (n= 62) with a score above the cut-off threshold, while it was USP19 Low (n= 106) with a score below the threshold (Supp. Fig. 6). Pearson's χ 2 or Fisher's exact tests were used to assess the relations between the tumor USP19 protein expression and the patient clinicopathological parameters. Disease-free survival (DFS) was defined as the interval from surgery to the first of the following events: tumor relapse at local or distant sites. Distant relapse-free survival (DRFS) was defined as the time from surgery to the occurrence of distant relapse. The Log-Rank (Mantel-Cox) test was used to analyze differences between the survival curves, and Cox's proportional hazard model was used to evaluate the association of USP19 expression with survival time, using covariates. The following covariates were computed in the multivariate model: tumor size, tumor grade, and ER, PR, Ki-67, HER2, and USP19 status. An area-based analysis of proliferation rate was used to determine cell growth over time. Cells were seeded onto wells, allowed to attach and imaged at the indicated time points. Left: relative mean occupied area at each time point ± S.E., and right: doubling time calculated for control and USP19 silenced cell lines (n= 10, one-way ANOVA, Dunnett's multiple comparison test. shRNA#1 p= 0.1953 and shRNA#2 p= 0.2824).