MicroRNA 483‐3p overexpression unleashes invasive growth of metastatic colorectal cancer via NDRG1 downregulation and ensuing activation of the ERBB3/AKT axis

In colorectal cancer, the mechanisms underlying tumor aggressiveness require further elucidation. Taking advantage of a large panel of human metastatic colorectal cancer xenografts and matched stem‐like cell cultures (m‐colospheres), here we show that the overexpression of microRNA 483‐3p (miRNA‐483‐3p; also known as MIR‐483‐3p), encoded by a frequently amplified gene locus, confers an aggressive phenotype. In m‐colospheres, endogenous or ectopic miRNA‐483‐3p overexpression increased proliferative response, invasiveness, stem cell frequency, and resistance to differentiation. Transcriptomic analyses and functional validation found that miRNA‐483‐3p directly targets NDRG1, known as a metastasis suppressor involved in EGFR family downregulation. Mechanistically, miRNA‐483‐3p overexpression induced the signaling pathway triggered by ERBB3, including AKT and GSK3β, and led to the activation of transcription factors regulating epithelial–mesenchymal transition (EMT). Consistently, treatment with selective anti‐ERBB3 antibodies counteracted the invasive growth of miRNA‐483‐3p‐overexpressing m‐colospheres. In human colorectal tumors, miRNA‐483‐3p expression inversely correlated with NDRG1 and directly correlated with EMT transcription factor expression and poor prognosis. These results unveil a previously unrecognized link between miRNA‐483‐3p, NDRG1, and ERBB3‐AKT signaling that can directly support colorectal cancer invasion and is amenable to therapeutic targeting.

In colorectal cancer, the mechanisms underlying tumor aggressiveness require further elucidation. Taking advantage of a large panel of human metastatic colorectal cancer xenografts and matched stem-like cell cultures (m-colospheres), here we show that the overexpression of microRNA 483-3p (miRNA-483-3p; also known as MIR-483-3p), encoded by a frequently amplified gene locus, confers an aggressive phenotype. In m-colospheres, endogenous or ectopic miRNA-483-3p overexpression increased proliferative response, invasiveness, stem cell frequency, and resistance to differentiation. Transcriptomic analyses and functional validation found that miRNA-483-3p directly targets NDRG1, known as a metastasis suppressor involved in EGFR family downregulation. Mechanistically, miRNA-483-3p overexpression induced the signaling pathway triggered by ERBB3, including AKT and GSK3b, and led to the activation of transcription factors regulating epithelial-mesenchymal transition (EMT). Consistently, treatment with selective anti-ERBB3 antibodies counteracted the invasive growth of miRNA-483-3p-overexpressing m-colospheres. In human colorectal tumors, miRNA-483-3p expression inversely correlated with NDRG1 and directly correlated with EMT transcription factor expression and poor prognosis. These results unveil a previously unrecognized link between miRNA-483-3p, NDRG1, and ERBB3-AKT signaling that can directly support colorectal cancer invasion and is amenable to therapeutic targeting.

Introduction
Metastatic colorectal cancer (CRC) is the third most common cause of cancer-associated death worldwide, with 5-year survival rates < 15% [1,2]. Increasing evidence suggests that cancer stem-like cells have a central role in recurrence after both conventional and targeted therapy and in metastatic dissemination of many cancer types, including CRC [3,4].
From a cohort of metastatic CRC propagated as patient-derived xenografts (PDX or xenopatients) [5], we previously derived stem-like cells that were longterm propagated as cultures hereafter named mcolospheres [6,7]. These cells faithfully retained the genetic traits of the original tumors, together with the genetically driven response to targeted therapy [6,7]. A subset of xenopatients and m-colospheres without genetic alterations of KRAS, NRAS, PIK3CA, or BRAF, henceforth defined as 'quadruple wild-type (WT)', and including the majority of cases, were found to be strongly relying on epidermal growth factor receptor (EGFR) and MET signaling, and thus sensitive to targeted inhibition of these receptors [5][6][7]. However, within this tumor subset, a subgroup of xenopatients and m-colospheres displayed proliferative ability independent of EGFR (resistance to EGFR inhibition), associated with overexpression and actionability of insulin-like growth factor 2 (IGF2) [7,8].
Interestingly, the IGF2 locus, located on chromosome 11p15.5 and amplified in 7% of CRCs, encodes also miRNA-483 [9,10]. In CRCs harboring an amplified locus, both IGF2 and miRNA-483 were found overexpressed and miRNA-483 was validated in organoids as the dominant driver oncogene [9,11]. More recently, the importance of the IGF2/miRNA-483 locus in CRC was confirmed by the evidence that the locus undergoes 'enhancer hijacking', by formation of an aberrant contact domain with a lineage-specific super-enhancer, resulting in transcriptional upregulation [12]. Moreover, the IGF2/miRNA-483 locus undergoes parental epigenetic imprinting, the loss of which unleashes transcription and correlates with increased CRC risk [13].
The molecular mechanisms by which miRNA-483 exerts its protumorigenic functions in CRC are still obscure. So far, only miRNA-483-3p was partly characterized with the identification, by in silico prediction of 3'UTR RNA binding, of putative targets. These include the BH3-containing protein BBC3/PUMA, whose downregulation prevents apoptosis in cell lines of various origins [14], and PARD3, whose targeting increases TGF-b1-induced invasion in anaplastic thyroid cancer [15].
To investigate the miRNA-483-3p biological effects and targets in CRC we took advantage of mcolospheres endogenously expressing low or high miRNA-483-3p levels, which we subjected to forward and reverse genetic approaches. We found that miRNA-483-3p expression correlates with the induction of cell invasion, epithelial-mesenchymal transition [16], and increased stem properties. By transcriptomic analysis, we identified N-myc downstream-regulated gene (NDRG1) as a new and prominent miRNA-483-3p target. This is known as a metastasis suppressor capable of downregulating the EGFR family [17,18]. We then found that metastatic CRC overexpressing miRNA-483-3p hyperactivates EGFR/ERBB3 signaling in a ligand-independent manner, leading to constitutive activation of EMT via the AKT/GSK3b/EMT transcription factors axis.
These results contribute to explaining the increased aggressiveness and resistance to specific EGFR targeting of CRC overexpressing the IGF2/miRNA-483 locus and provide a rationale for therapeutic strategies aimed at extensive EGFR family inhibition.

Human patients
Patients were recruited in a prospective observational trial (ClinicalTrials.gov, https://clinicaltrials.gov/ct2/ show/NCT03347318?term=001-IRCC-00IIS-10&draw= 2&rank=1, No. NCT03347318) approved by the Candiolo Cancer Institute Review Board on human experimentation (protocol No. 001-IRCC-00IIS-10). The study was conducted according to the standards set by the Declaration of Helsinki. Written informed consent was obtained from all patients and all samples were deidentified. Patients' data were treated according to ethical requirements and GCP.

Animal models
Animal models, including xenopatients, i.e., mice transplanted with patient-derived xenografts (PDX) of CRC liver metastasis surgical samples, and spheropatients, i.e., mice transplanted with m-colospheres (see also Section 2.18), were generated according to ethical regulations, methodologies and protocols approved by the Italian Ministry of Health (permissions No. 223/2015-PR and No. 806/2016-PR). 5-to 6-week-old male NOD.CB17-Prkdc scid /NcrCr mice (NOD/SCID, RRID: IMSR_CRL:394, Charles River Laboratories, Calco, Italy), were used for all in vivo studies. Mice were housed at a maximum of 6 per cage with a 14-h light/10-h dark cycle with food and water ad libitum. Mice were monitored at a minimum of twice weekly for general performance status and euthanized when the volume of xenografts reached 1600 mm 3 , or they displayed signs of distress, or weight loss ≥ 20%.
Gene expression was normalized vs. UBC, GADPH, TBP, B2M, and ACTB as endogenous controls. For miRNA-483-3p analysis, starting from total RNA, reverse transcription was performed with specific primers (Table S1) with a TaqMan microRNA reverse-transcription kit (Thermo Fischer Scientific) according to manufacturer's instructions. The diluted RT-PCR products were amplified using TaqManÒ Universal Master Mix No Amperase UNG and TaqManÒ miRNA assay (hsa-miR-483-3p; Thermo Fisher Scientific) with ABI PRISM 7900 HT (Applied Biosystem). The miRNA-483-3p gene expression was normalized to RNU48 expression (Thermo Fischer Scientific). Relative expression was calculated by subtracting scaled CT values from the total 40 cycles or as fold change (2 ÀDDct ). Numerical results were expressed as means AE SEM (n ≥ 3 independent experiments, Student t-test).

M-colosphere transduction
M-colospheres were dissociated to single-cell suspensions and 1 9 10 5 cells were transduced in 6-wells plates with lentiviral vectors encoding pre-miR483-3p or antagomiRNA-483-3p (System Biosciences, Mountain View, CA, USA) or shRNAs-hNDRG1 (Vector Builder GmbH, Neu-Isenburg, Germany; Table S1) at a Multiplicity Of Infection (MOI) of 5. All constructs included a GFP reporter under the control of CMV and H1 promoters and infection efficiency was verified by fluorescent microscopy or facs analysis, and by qPCR, 72 h after infection.

Cell viability
M-colospheres were dissociated and seeded in 96microtiter wells at the concentration of 700-1000 cellsÁ100 lL À1 in basal medium supplied with EGF, FGF2, neuregulin 1 or without factors, as indicated.
In some experiments, MM121 [also known as seribantumab, Creative Biolabs (Shirley, NY, USA), 50 ngÁmL À1 ] [19] was added after seeding and every 48 h. ATP consumption was measured at days 0 and 4 with Cell Titer GloÒ and GloMax 96 Microplate Luminometer (Promega) according to manufacturer's instructions. In each experiment the average relative luminescence values (n ≥ 6 technical replicates) were normalized vs. day 0 and fold changes were reported (n ≥ 3 independent experiments, mean AE SEM, ANOVA, Bonferroni Multicomparison test).

3D-spheroid invasion assay
3D-spheroid assays were performed by seeding mcolospheres in a matrix composed of growth factor reduced matrigel (BD Biosciences, Franklin Lakes, NJ, USA) and collagen type I [20]. Invasion was monitored using Cytation 3 Cell Imaging Multi-Mode Reader (Bio-Tek Instruments, Agilent Technologies, Santa Clara, CA, USA). Morphometric analyses were performed with IMAGEJ software (RRID:SCR_003070, NIH, Bethesda, MD, USA). Invaded area was normalized vs. day 0 and fold changes were reported (n ≥ 3 independent experiments, mean AE SEM, ANOVA, Bonferroni multicomparison test and Kolmogorov-Smirnov test).

Western blot analysis
Total proteins were extracted using RIPA buffer supplemented with a protease inhibitor cocktail (Roche Life Science, Saint Louis, MO, USA), NaVO 3 1 mM and NaF 1 mM, sonicated, quantified by BCA (Pierce, Thermo Fischer Scientific, Waltham, MA, USA). Proteins (12-20 lg) were separated on SDS/PAGE 4-12% or 4-20% (Invitrogen, Thermo Fischer Scientific, Waltham, MA, USA) and blotted onto nitrocellulose membrane. After blocking, primary antibodies were incubated at the indicated concentrations (Table S2). After incubation with HRP-conjugated secondary antibodies (Jackson Lab, Thermo Fischer Scientific, Bar Harbor, ME, USA; Table S2), enhanced chemiluminescence (Biorad, Segrate, Italy) was used for detection according to manufacturer's instructions and images were acquired with the Chemi-Doc Touch TM Imaging System (Biorad) with IMAGE LAB software. The results shown are representative of n ≥ 3 independent experiments.

Immunofluorescence and immunohistochemistry
Samples undergoing immunofluorescence were either Formalin-Fixed, Paraffin-Embedded tumor specimens or growing agnospheres. The latter were harvested, fixed 10 min with PFA 4% at 4°C, washed in PBS, and suspended in bio-agar for cyto-inclusion (Bio-Optica, Milano, Italy) at 42°C, processed for inclusion in paraffin, and stained. Briefly, 3 lm sections were kept for 1 h at 65°C, then deparaffinized, rehydrated, and subjected to antigen retrieval for 40 min at 95°C in TRIS/EDTA buffer pH 9.5. After cool down, sections were permeabilized in TBS/Tween20 0.1% 10 min and blocked in BSA 5%/normal donkey serum 5% in TBS Tween 0.1% for 1 h as previously described [21]. Primary antibodies were incubated O/N at 4°C at the indicated concentrations (Table S2). For immunofluorescence, fluorescent-conjugated secondary antibodies (Alexa Fluor, Thermo Fischer Scientific) were incubated for 40 min at RT in TBS/Tween20 0.1%. Nuclei were counterstained with DAPI (Dako, Merck KgaA Darmstadt, Germany) and mounted with gel mount. Images were acquired using a LEICA SPEII confocal microscope, equipped with a 209 oil objective. Optical single sections were acquired with a scanning mode format of 1024 9 1024 pixels. Fluorochromes unmixing was performed by the acquisition of the automated-sequential collection of multi-channel images, to reduce spectral crosstalk between channels. For immunohistochemistry, an additional peroxidase blocking was performed in H2O2 3%/methanol 50% incubated for 20 min in the dark. Secondary antibodies were HRP-conjugated (Dako, Agilent), and diaminobenzidine (DAB) substrate chromogen kit (Dako) was used for detection. Nuclei were counterstained with Hematoxylin and images were acquired through LASV4.2 software (Leica Microsystem, Wetzlar, Germany). Images are representative of at least three independent immunostainings.

RNA sequencing
Quant-seq 3' mRNA sequencing was performed using total RNA from m-colospheres. RNA was purified with RNeasy Micro Kit (Qiagen, Dusseldorf, Germany) and quantified with QUBIT 2.0 fluorimetric assay (Thermo Fisher Scientific). Libraries were prepared from 100 ng of total RNA with Quant-Seq 3' mRNA-Seq Library Prep Kit FWD for Illumina (Lexogen GmbH, Wien, Austria), followed by library quality control by screen tape High sensitivity DNA D1000 (Agilent Technologies). Libraries were sequenced on a NovaSeq 6000 sequencing system (Illumina Inc., San Diego, CA, USA) as previously described [23].

Analysis of Quant-Seq RNA data
Analysis of RNA-sequencing data was performed as previously described [23]. Briefly, Illumina novaSeq base call (BCL) files were transformed into fastq files through BCL2FASTQ (version v2.20.0.422, Illumina Inc.), and sequence reads were trimmed using BBDUK software (bbmap suite 37.31, Joint Genome Institute, Walnut Creek, CA, USA). Alignment was performed on hg38 reference assembly (Ensembl Assembly 93) with STAR 2.6.0a (GPL v3, open source) and gene expression levels were determined with htseq-count 0.9.1. Genes with an average number of cpm (counts per million) < 5 and Perc of duplicated reads > 20% were excluded from the ensuing analysis. Gene expression normalization and differentially expressed genes were identified by using EDGER [24]. Volcano plots were generated by using GGPLOT2 package in R statistical environment. RNA-sequencing data were released in a public repository (GEO accession number: GSE209535) and are available in Table S3.

In situ proximity ligation assay (PLA) on m-colospheres
In situ proximity ligation assay (PLA) assays were performed using the Duolink TM In Situ Red Starter Kit Mouse/Rabbit (Sigma) according to the manufacturer's instructions. Briefly, m-colospheres were harvested, fixed 10 min with PFA 4% at 4°C, washed in PBS, suspended in bio-agar for cyto-inclusion (Bio-Optica) at 42°C, paraffin-embedded and sectioned as previously described [19]. Sections were incubated with primary anti-EGFR, anti-ERBB2, and anti-ERBB3 (Table S2) at 4°C overnight followed by species-specific secondary antibodies conjugated with oligonucleotides (PLA probes). Negative control was performed by the sole EGFR primary antibody. After ligation and amplification, the signal from each pair of PLA probes in close proximity (< 40 nm) was visualized as an individual red spot and analyzed by LEICA SPEII confocal microscope and LASV software. PLA Multicolor Reagent Pack PLA kits DUO96000 and In Situ Red Starter Kit Mouse/Rabbit DUO92101 were purchased from Sigma-Aldrich, Duolink TM .

In vivo m-colospheres transplantation (spheropatients)
To assess tumorigenic potential, m-colospheres were dissociated as single-cell suspensions and counted with trypan blue to exclude dead cells. 10 5 cells were resuspended in 50 lL of basal medium mixed 1 : 1 with growth factor reduced matrigel (BD Bioscience, Franklin Lakes, NJ, USA) and injected subcutaneously in the flank of 5-6 weeks old male NOD.CB17-Prkdcscid/J mice (referred to as 'spheropatients'). During the procedure anesthesia with 2.5% isofluorane in 100% oxygen at a flow rate of 1 LÁmin À1 was delivered to mice. Tumor growth was measured by caliper using the formula (d) 2 9 (D)/2, where d and D are the minor and the major tumor axis, respectively. Tumors grown at the injection site were explanted, formalin-fixed, and paraffin-embedded to undergo histopathological evaluation.

Statistical analysis
RNAseq analysis was performed with the fgsea package in R statistical environment version 3.6. All the other statistical analyses were performed using GRAPHPAD

miRNA-483-3p is overexpressed in a subset of CRC tissues and m-colospheres
We previously generated a large cohort of human metastatic CRC (mCRC) xenografts (or xenopatients), from which we derived m-colospheres (n = 58), also called 'xenospheres', i.e. cultures enriched in cells with cancer-initiating properties (Fig. 1A,B) [5,6]. Previous characterization showed that m-colospheres retain the same driving genetic alterations as the matched xenopatients (Table S4) [6,7]. To unveil novel genetic mechanisms that may support tumor aggressiveness, we focused on the 'quadruple WT' case subgroup, which is devoid of mutations in the RAS/phosphoinositide 3-kinase (PI3K) pathway, and where correlations between oncogenic pathways and mechanisms of tumor progression are partly obscure [5][6][7]. Interestingly, in a previous study, we found that a subset of quadruple WT xenopatients displayed IGF2 overexpression and was only partially sensitive to EGFRtargeted therapy with the EGFR antibody cetuximab [8]. In these tumors, IGF2 was shown to provide a direct contribution to bypass EGFR activity requirement [8]. Among available quadruple WT mcolospheres (n = 17), 8/17 were derived from IGF2 overexpressing xenopatients and were confirmed to overexpress IGF2 ( Fig. 1C and Table S4). In the remaining quadruple WT cases (n = 9/17), low levels of IGF2 expression reported in xenopatients were consistently observed in matched m-colospheres ( Fig. 1C and Table S4). Moreover, by extending IGF2 analysis to the entire cohort of available m-colospheres (n = 33), including also those harboring RAS-PI3K pathway mutations, we found that IGF2 expression levels remained consistent between m-colospheres and matched xenopatients ( Fig. S1A and Table S4).
First, we investigated whether miRNA-483-3p overexpression affected the response to growth factors included in the standard medium conventionally used for colosphere selection and propagation (EGF and FGF2) and to the ERBB3 ligand neuregulin 1, which we previously identified as capable of sustaining longterm propagation of quadruple WT colospheres [7]. After growth factor deprivation by culture in basal medium for 48 h, CRC264miR and CRC327miR displayed a remarkably increased proliferative response to EGF, supplied either at a minimally effective (0.2 ngÁmL À1 ) or standard (10 ngÁmL À1 ) concentration, and to neuregulin 1 (10 ngÁmL À1 ) compared with their matched parental m-colospheres, while they did not change their response to FGF2 (Fig. 1G).
To investigate the invasive response, we challenged m-colospheres in 3D assays with a matrix composed of growth factor reduced matrigel and collagen type I [20]. Notably, CRC264miR and CRC327miR could grow and invade the surrounding matrix even in the complete absence of exogenous growth factors (i.e., in basal medium), while their matched parental m-colospheres were unable to invade the matrix, either in the absence or in the presence of growth factors, and CRC327 even showed signs of internal necrosis (Fig. 1H and Fig. S1F, G). The addition of exogenous growth factors (EGF and FGF2 + neuregulin 1) significantly increased miRNA-483-3p overexpressing m-colospheres invasiveness at the early time-point ( Fig. 1H and Fig. S1F), but it was almost ineffective on parental m-colospheres ( Fig. 1H and Fig. S1G).
Altogether, these data show that miRNA-483-3p overexpression fosters m-colosphere invasive growth, by promoting the proliferative response to EGFR family ligands, and conferring the ability to invade the extracellular matrix either in the presence or in the absence of growth factors.
Altogether, these data unearth a correlation between miRNA-483-3p overexpression and EMT upregulation in mCRC.
These data indicate that miRNA-483-3p overexpression can sustain the stem-like traits of m-colospheres.
Altogether, these data support the conclusion that NDRG1 is a direct target of miRNA-483-3p.

NDRG1 downregulation by miRNA-483-3p enhances the EGFR family/AKT axis signaling
The miRNA-483-3p target NDRG1 is particularly intriguing as it was identified as a metastasis suppressor in CRC [37] and in several other cancer types [38]. NDRG1 is a pleiotropic molecule with still unclear mechanistic functions; it has been shown to inhibit various oncogenic pathways, causing, among other events, decreased expression, activation, and heterodimerization of EGFR family receptors [39]. In particular, NDRG1 was shown to promote ERBB3 protein degradation by increasing its interaction with NEDD4 ubiquitin ligase [38,40]. Therefore, we investigated whether NDRG1 downregulation in m-colospheres transduced with miRNA-483-3p correlated with ERBB3 protein stabilization and activation of the downstream signaling pathways. Indeed, upon miRNA-483-3p ectopic overexpression, NDRG1 reduction correlated with increased ERBB3 protein levels ( Fig. 3D and Fig. S3F) and stabilization at the plasma membrane, as observed by immunofluorescence (Fig. 3E), without effects on ERBB3 mRNA levels (Fig. S3G). Total EGFR protein or mRNA levels were not significantly increased (Fig. 3D and Fig. S3G), but EGFR was stabilized at the membrane, as well (Fig. 3E). CRC264miR and CRC327miR displayed intense EGFR tyrosine phosphorylation (Tyr1068), as compared with parental m-colospheres (Fig. 3D,E and Fig. S3F), despite growth factors (EGF and FGF2) had been withdrawn for 24 h before the experiment. EGFR activation was accompanied by intense activating phosphorylation of AKT (Ser473) and its target GSK3b (Fig. 3D and Fig. S3F). The latter occurred at Ser9, responsible for GSK3b inhibition [41], which is known to prevent SNAI1 and SNAI2 targeting to protein degradation, thereby leading to EMT upregulation [42,43]. Interestingly, phosphorylation of p38 mitogen-activated protein kinase (MAPK), but not p42/44 MAPK, was increased as well (Fig. 3D).
To assess whether NDRG1 targeting is essential for the biochemical and biological activities associated with miRNA-483-3p expression, we silenced NDRG1 expression in parental CRC264 and CRC327 by shRNA lentiviral transduction, obtaining CRC264shNDRG1 and CRC327shNDRG1 ( Fig. 3G and Fig. S3J). We then investigated whether NDRG1 knockdown reproduced the same phenotype observed after miRNA-483-3p transduction. Indeed, compared with parental m-colosphere controls, CRC264shNDRG1 and CRC327shNDRG1 showed (a) increased expression of ERBB3 and pronounced phosphorylation of EGFR and AKT ( Fig. 3G and Fig. S3J); (b) marker expression switch from epithelial to mesenchymal ( Fig. 3G and Fig. S3J); and (c) increased sensitivity to EGF family ligands. Such response was more pronounced to neuregulin 1 in both NDRG1silenced m-colospheres, either in viability (Fig. 3H) or invasive growth assays (Fig. 3I and Fig. S3K). Therefore, in m-colospheres, miRNA-483-3p overexpression is phenocopied by NDRG1 silencing, indicating that NDRG1 is an essential miRNA-483-3p target.
In summary, these data suggest that the link between miRNA-483-3p and EMT is embodied by a cascade of events originating from the inhibition of NDRG1 protein translation, resulting in ERBB3 protein stabilization and upregulation of EGFR/ERBB3 signaling, followed by increased AKT activity, GSK3b inhibition, SNAI1 and SNAI2 protein stabilization, and E-cadherin transcriptional repression (Fig. S3L).
In cell viability assays, MM121 fully abolished the hyperproliferative effect of neuregulin 1, observed in CRC264miR and CRC327miR, reducing it to levels detectable in their matched parental m-colospheres (Fig. 4A); moreover, MM121 significantly reduced also the response to EGF, which, like the response to neuregulin 1, was conceivably mediated by EGFR/ERBB3 heterodimers (Fig. 4A). In addition, MM121 fully prevented invasive growth promoted by miRNA-483-3p in m-colospheres cultured in the absence of any growth factor, and induced the appearance of inner necrotic areas, as observed in 3D assays ( Fig. 4B and Fig. S4B). Consistently, in miRNA-483-3p overexpressing m-colospheres, MM121 downregulated ERBB3 from the cell surface, inhibited EGFR and AKT phosphorylation, and it induced E-cadherin re-exposure on the cell surface, as assessed by immunofluorescence (Fig. 4C) and western blot (Fig. 4D). Notably, together with inactivation of the EGFR/ERBB3/AKT pathway, MM121 caused robust induction of cleaved-Caspase3 (Fig. 4D), indicating that not only proliferation and invasive growth associated with miRNA-483-3p were prevented (Fig. 4A,B), but apoptosis was induced as well.
These data support the key role of the ERBB3/ EGFR axis in the phenotype of miRNA-483-3p overexpressing m-colospheres, and indicate that ERBB3-specific inhibition is sufficient to counteract this phenotype (Fig. S4A).
Finally, based on in vitro evidence, we investigated whether antagomiRNA-483-3p could counteract CRC729 growth and invasion in vivo. Therefore, we subcutaneously transplanted 5 9 10 5 CRC729Antago or their parental counterpart in NOD/SCID mice (spheropatients). After 6 weeks, all 14 CRC729 spheropatients developed palpable grafts (average volume = 231 mm 3 ), six of which (irrespectively of tumor volume) displayed underlying muscle invasion (Fig. 5G,H). Strikingly, none of the 14 CRC729Antago spheropatients developed a palpable tumor until 24 weeks after m-colosphere injection (Fig. 5G). At this time-point, two mice displayed tumors with a poorly aggressive aspect, featuring a more differentiated adenomatous morphology, the presence of a well-defined connective capsule, and lack of peritumoral invasion (Fig. 5H).

miRNA-483-3p expression directly correlates with increased EMT and poor prognosis in CRC patients
To further investigate the correlation between miRNA-483-3p expression and aggressive tumor behavior, in our metastatic CRC xenopatient cohort we compared a panel of cases displaying high miRNA-483-3p expression levels with a panel of cases expressing low levels, as defined by qPCR ( Fig. S1B and Table S4). miRNA-483-3p-high tumors showed higher expression of EMT-TFs such as SNAI2, ZEB1, and YAP, and weaker E-cadherin expression, as compared with miRNA-483-3p-low cases (Fig. 6A,B). Strikingly, NDRG1 expression showed an inverse correlation with EMT markers, being low in miRNA-483-3p-high xenopatients and high in miRNA-483-3p-low xenopatients (Fig. 6A,B). Accordingly, high miRNA-483-3p expression correlated with loss of epithelial and adenomatous morphology, high cellularity, and poorly differentiated features (Fig. 6A,C).
Finally, we analyzed whether miRNA-483-3p expression levels correlated with different clinical outcomes of CRC patients. Indeed, in the TCGA cohort group including stage III and IV quadruple WT patients for which miRNA data were available, we observed that patients overexpressing miRNA-483-3p (n = 14) displayed a significantly reduced overall survival compared with those expressing normal/low levels of miRNA-483-3p (n = 23; P = 0.0085; Fig. 6D).
Overall, these results suggest that miRNA-483-3p expression is associated with aggressiveness and poor prognosis in metastatic CRC patients. Given the tight relationship between miRNA-483-3p expression and upregulation of the ERBB3/AKT/EMT pathway, these patients could be further investigated for eligibility to personalized therapies targeting the overall EGFR family or, specifically, ERBB3.
Therefore, selective inhibition of EGFR binding to EGF ligands, as attained by clinically approved EGFR antibodies (such as cetuximab), may easily encounter bypass mechanisms and primary resistance involving other members of the EGFR family [51].
On these premises, not surprisingly CRC may evolve multiple mechanisms to sustain constitutive EGFR family signaling. Here we provide evidence that one of such mechanisms can be miRNA-483-3p overexpression, which mainly results in upregulated expression and signaling activity of ERBB3, a prominent regulator of stem-like cells in CRC [7] and in glioblastoma, where ERBB3 specifically sustains the PI3K/AKT axis [46]. As ERBB3 is a heterodimerization partner lacking intrinsic tyrosine kinase activity [44], its oncogenic activation should preferentially rely on mechanisms controlling its expression, rather than on structural alterations, and, to such purpose, aberrant oncogenic or oncosuppressor miRNAs are attractive candidates.
Interestingly, forced miRNA-483-3p expression not only enhanced the m-colosphere basal and growth factor-dependent proliferative rate but also EMT and the ability to invade 3D extracellular matrices, a recognized harbinger of metastatic ability [25].
Addressing the link between a microRNA and a biological outcome such as EMT and increased invasiveness may imply to consider a plethora of direct and indirect microRNA targets. However, we offer evidence that the thread leading from miRNA-483-3p to EMT chiefly involves NDRG1. First, by providing the first analysis of the global transcriptomic response to miRNA-483-3p expression and functional experiments, we show that NDRG1 is prominently, robustly, and directly downregulated among other genes. Most importantly, we provide evidence that NDRG1 silencing mimics the effects of miRNA-483-3p expression. This evidence strongly associates the miRNA-483-3pinduced phenotype, featuring increased sensitivity to EGF ligands, EMT, and invasive growth, to the specific activity of the miRNA-483-3p target NDRG1.
NDRG1 has been strikingly associated with cancer metastasis and thus defined as a metastasis-suppressor gene in several carcinomas, including CRC [38,52]. However, the mechanisms by which NDRG1 sustains metastasis have been so far elusive and can be at least in part elucidated by findings presented in this study. NDRG1 was known to promote protein degradation, in particular of EGFR family members [38]. By the use of a specific ERBB3 antibody (MM121), which was able to fully revert EMT and invasion induced by miRNA-483-3p forced expression and the ensuing NDRG1 downregulation, we show that, in this context, ERBB3 is essential for EMT and invasion induction. This is not surprising, considering the preferential ability of this receptor to activate the PI3K/AKT pathway, leading to GSK3b inhibition and escape of EMT-core transcription factors, such as SNAI1 and SNAI2, from the degradative pathway [41], as observed in m-colospheres.
The modality of miRNA-483-3p overexpression, which is genetically fixed through IGF2 locus amplification, or by other inheritable mechanisms, as indicated by overexpression preservation from the patient to the xenopatient, and from the latter to mcolospheres, indicates that the mechanism is selectable under evolutive pressure. Overall, these findings provide additional evidence that the EGFR signaling pathway is essential for CRC progression and suggest that a possible cooperation between miRNA-483-3p and IGF2 in fostering EGFR family activity deserves further investigation. Notably, this study shows that ERBB3 signaling upregulation can sustain the escalation from local proliferation to invasive growth and metastasis. In this perspective, an extended EGFR family targeting strategy could be more beneficial than the inhibition of one single family member.

Conclusions
In this study, we reconstruct for the first time the chain of mechanistic events that link miRNA-483-3p to the upregulation of invasive growth in metastatic colorectal cancer. We show that miRNA-483-3p targets NDRG1, a known 'metastasis suppressor', thereby activating the ERBB3-AKT signaling pathway. The latter sustains EMT and stem properties, hence promoting colorectal cancer invasive growth, a critical prerequisite for metastasis. Notably, the proinvasive activity of miRNA-483-3p can be countered by ERBB3 blockade through selective antibodies. These findings not only present a new mechanism that can fuel colorectal cancer progression but also provide a rationale for targeted therapeutic intervention.