A FAM83A Positive Feed-back Loop Drives Survival and Tumorigenicity of Pancreatic Ductal Adenocarcinomas

Pancreatic ductal adenocarcinomas (PDAC) are deadly on account of the delay in diagnosis and dearth of effective treatment options for advanced disease. The insurmountable hurdle of targeting oncogene KRAS, the most prevalent genetic mutation in PDAC, has delayed the availability of targeted therapy for PDAC patients. An alternate approach is to target other tumour-exclusive effector proteins important in RAS signalling. The Family with Sequence Similarity 83 (FAM83) proteins are oncogenic, tumour-exclusive and function similarly to RAS, by driving the activation of PI3K and MAPK signalling. In this study we show that FAM83A expression is significantly elevated in human and murine pancreatic cancers and is essential for the growth and tumorigenesis of pancreatic cancer cells. Elevated FAM83A expression maintains essential MEK/ERK survival signalling, preventing cell death in pancreatic cancer cells. Moreover, we identified a positive feed-forward loop mediated by the MEK/ERK-activated AP-1 transcription factors, JUNB and FOSB, which is responsible for the elevated expression of oncogenic FAM83A. Our data indicates that targeting the MEK/ERK-FAM83A feed-forward loop opens up additional avenues for clinical therapy that bypass targeting of oncogenic KRAS in aggressive pancreatic cancers.

of RAS with phosphodiesterase-delta subunit (PDEδ), altering RAS localization and inhibiting its function 12 . The RAS-mimetic Rigosertib binds to RAS-binding domain (RBD) of RAF blocking its functional interaction with RAS and inhibiting downstream MEK/ERK effector signaling 13 . Moreover, improved analogues of mutant KRAS-G12C-specific inhibitors such as ARS853 are currently being optimized to directly target RAS 14 . Nonetheless, the clinical efficacy of the newer drugs remains uncertain and it is essential that newer 'druggable' targets are identified that can bypass the need to target/inhibit oncogenic RAS.
We discovered the Family with Sequence Similarity 83 (FAM83) proteins in a forward genetic screen for genes that promote cellular transformation. In mammary epithelial cells FAM83 proteins function similarly to RAS, driving the activation of PI3K and MAPK signaling 15,16 . The FAM83 protein family has 8 members varying greatly in size, with each sharing a conserved amino-terminal Domain of Unknown Function (DUF1669 domain). FAM83 proteins are elevated in many human cancers and have been implicated in cancer growth, metastasis and therapy resistance 17,18 . The smallest FAM83 member, FAM83A, was separately identified by the Bissell laboratory, due to its ability to promote EGFR Tyrosine Kinase Inhibitor (TKI) resistance, also in mammary epithelial cells 19 . FAM83A is overexpressed in many human cancers and yet it remains unclear what factors drive elevated FAM83A expression in the transformed cells.
In the current study, we show that FAM83A expression is significantly elevated in human and murine pancreatic cancers. Ablation of FAM83A expression suppresses MEK/ERK survival signalling, resulting in apoptosis and suppression of tumorigenicity. Importantly, FAM83A expression is regulated by the activating protein-1 (AP-1) transcription factors, Jun proto-oncogene B (JUNB) and FBJ murine osteosarcoma viral oncogene homolog B (FOSB), which are activated by MEK/ERK activity. We propose that the MEK/ERK-mediated induction of FAM83A gene expression creates a positive, feed-forward loop involving FAM83A and RAS effectors ultimately drive pancreatic cancer cell survival. Our studies provide new insight into RAS-driven pancreatic cancer development and progression and serve as the foundation for developing new therapies capable of targeting RAS/ FAM83A signalling axis. 20 identified that FAM83A expression is upregulated in human mammary epithelial cells (HMEC) expressing exogenous mutant RAS, but not other oncogenes such as c-sarcoma proto-oncogene (SRC), c-Myc (MYC) and β-catenin (Fig. 1a). To confirm this observation, FAM83A expression was assessed by quantitative real-time PCR (qPCR) of telomerase reverse transcriptase (hTERT)-immortalized human mammary epithelial-1(HME1) cells exogenously expressing mutant Harvey Rat sarcoma (HRAS), Neuroblastoma Rat Sarcoma (NRAS) and KRAS. All three RAS isoforms increased FAM83A gene expression >2.5 fold relative to the control (Fig. 1b). Given that the vast majority of PDAC are driven by oncogenic, mutant KRAS, we surveyed FAM83A expression in PDAC specimens using publically available The Cancer Genome Atlas (TCGA) and University of Texas Southwestern (UTSW) datasets. In three independent studies 21-23 , FAM83A gene expression was significantly elevated in pancreatic tumour tissue as compared to their associated-normal counterparts (Fig. 1c), and 14% of pancreatic cancers exhibit FAM83A gene amplification 24,25 . In three separate patient-derived pancreatic tumour samples, FAM83A expression was elevated compared to matched tumour-associated normal pancreas tissue ( Supplementary Fig S1). In addition, pancreatic cancer cell lines exhibit increased expression of FAM83A when compared to the normal human pancreas (Fig. 2a). In contrast, FAM83B, another FAM83 member, was not elevated in the pancreatic cancer cell lines ( Supplementary Fig. S1). Taken together, these data point at a selective increase in FAM83A expression in pancreatic cancers.

FAM83A expression is elevated in pancreatic cancers. Analysis of gene expression datasets
FAM83A is essential for the growth and tumorigenicity of pancreatic cancer cells. The mitogen activated protein kinase (MAPK) and PI3K signalling pathways downstream of KRAS are major contributors to pancreatic carcinogenesis. Importantly, FAM83A was originally identified as a positive regulator of MAPK and PI3K-signaling pathways 19 . Therefore, we hypothesized that the KRAS-mediated induction of FAM83A described above, creates a positive, feed-forward loop whereby FAM83A increases KRAS effector signalling, which functions to drive increased transcription of the FAM83A gene, ultimately driving pancreatic cancer proliferation and survival. To test this hypothesis, lentiviral constructs expressing two different short hairpin RNA (shRNA) targeting FAM83A (shA2, shA6) or a control shRNA (shGreen Fluorescent Protein/shGFP) were used to knockdown FAM83A in pancreatic cancer cell lines (HPAC, Panc04.03, Panc05.04, Panc08.13). Indeed, suppression of FAM83A expression by either shA2 or shA6 (Fig. 2b) reduced cell number by ~70-80% in 2-dimensional culture (Fig. 2c), and strongly suppressed anchorage-independent growth in soft agar (Fig. 2d) and 3Dimensional (3D)-organoids (Fig. 2e). To determine if the impaired growth following FAM83A ablation had any physiological significance to in vivo pancreatic tumour formation, FAM83A was knocked down in luciferase-expressing HPAC cells and the cells were transplanted orthotopically into the pancreas of immunocompromised mice. Following FAM83A knockdown, pancreatic tumours were significantly smaller when measured by bioluminescence imaging (Fig. 3a) or tumour volume (Fig. 3b,c), resulting in delayed mortality (Fig. 3d). These data clearly demonstrate that FAM83A is an important contributor to in vivo pancreatic tumour growth and that targeting FAM83A or its function has the potential to enhance patient survival.
Suppression of FAM83A induces apoptosis in pancreatic cancer cells. Since suppression of FAM83A resulted in severe growth impairment and inhibited tumorigenesis, we next sought to determine whether FAM83A loss impacted proliferative capacity or survival. First, proliferation was monitored using a membrane-protein labelling dye, 5(6)-carboxyfluorescein diacetate N-succinimidyl ester (CFSE) as a measure of cell division. CFSE labels the cells and is diluted with each cellular division resulting in distinct peaks when assessed by flow cytometry. FAM83A ablation did not alter the rate at which CFSE was reduced with each division, indicating that the proliferative capacity of cells lacking FAM83A is not impacted (Fig. 4a). Similar observations were made using the Bromodeoxyuridine (BrdU) incorporation assay, wherein the incorporation of the nucleoside analogue by actively proliferating PDAC cells was comparable in the presence and absence of FAM83A ( Supplementary Fig. S2). Next, we assessed how FAM83A suppression impacted cell survival. Using propidium iodide (PI) staining and flow cytometry, FAM83A knockdown consistently increased the sub-G1 population to 14% (shA2) and 28% (shA6) when compared to 6% in control cells, indicating that FAM83A suppression results in enhanced cell death (Fig. 4b). Likewise, FAM83A knockdown increased the proportion of Annexin V-positive apoptotic cells from 4.9% to 28.8% (shA2) and 33.1% (shA6; Fig. 4c). Hence, the elevated levels of FAM83A observed in PDAC cells have a pro-survival function.

FAM83A knockdown inhibits activation of the MEK-ERK survival signalling.
To understand how FAM83A contributes to survival signalling in pancreatic cancers, next generation RNA-sequencing (RNA-seq) analysis was performed to examine global gene expression changes in PDAC cells following FAM83A knockdown. Specifically, the knockdown of FAM83A resulted in 542 messengerRNAs that become differentially expressed (≥2 fold change, p < 0.05) (FAM83A knockdown gene signature and gene set enrichment analysis are shown in Supplementary Fig. S3). Importantly, FAM83A suppression clearly decreased the expression of hallmark genes indicative of aggressive pancreatic cancers such as Mesothelin, VEGF, MUC1 and MUC4 (Fig. 5a) 26 . Importantly, FAM83A knockdown upregulated genes associated with apoptosis and cell death, including pro-apoptotic gene caspase-8 (CASP8) and BH3 interacting-domain death agonist (BID) (Fig. 5a,b). BID and Caspase 8 are known regulators of apoptosis that trigger the release of mitochondrial proteins and activate effector caspases like caspase-3 27 . Accordingly, knockdown of FAM83A resulted in increased levels of cleaved caspase-3 comparable to the etoposide-treated positive control (Fig. 5c). The MEK/ERK pathway is known to protect cells from stress-induced caspase-8/BID cleavage 28 . Notably, FAM83A knockdown also reduced MEK/ ERK-responsive genes, including EGR1, FOSB and JUNB (Figs. 5d,e). Early growth response protein 1 (EGR1) and the AP-1 components, FOSB and JUNB are known to regulate transcription in response to active MEK/ ERK signalling, resulting in enhanced cell growth and survival 29,30 . Likewise, suppression of FAM83A reduced MEK-mediated ERK phosphorylation, but did not impact signalling to AKT and mTOR ( Fig. 5f), indicating that FAM83A-mediated MEK/ERK activation is critical for the survival of PDAC cells.
We observed that FAM83A expression was strongly induced by >4-fold upon serum stimulation (Fig. 6a), and inhibition of either MEK or ERK blocked serum-induced FAM83A expression (Fig. 6b). As EGR1 (E), FOSB (F) and JUNB (J) are MEK/ERK regulated transcription factors known to bind serum responsive elements (SREs), we used SMARTpool small interfering RNA (siRNA) targeting each in order to examine their importance in regulating FAM83A expression. While loss of JUNB significantly reduced serum-induced FAM83A expression, the loss of FOSB and EGR1 resulted in slight or no reduction in FAM83A, respectively (Fig. 6c). We also confirmed these data using two different shRNAs against JUNB (shJunB-1 and shJUNB-2; Fig. 6d). Knockdown of JUNB significantly reduced cell growth phenocopying the loss of FAM83A (Fig. 6e). shRNA-mediated knockdown of FOSB failed to reduce FAM83A expression (Fig. 6f). Surprisingly FAM83A was not directly regulated by KRAS, as depletion of KRAS gene expression using two different lentiviral shRNAs (shKRAS#1 and shKRAS#2) and siRNA did not alter FAM83A expression (Supplementary Fig. S4A; Fig. 6c). Taken together, the MEK/ERK-regulated AP-1 transcription factor JUNB controls FAM83A expression in PDAC cells. FAM83A knockdown in turn strongly inhibits MEK/ERK activation and transcription of JUNB as shown in Fig. 5. Our data suggests the existence of a novel positive feed-forward loop involving MEK-ERK-JUNB-FAM83A that promotes pancreatic tumorigenesis.
Murine KRAS-G12D p53-R172H cells require sustained FAM83A expression for tumor formation. Murine pancreatic cancer models such as the KPC (KRAS-G12D p53-R172H Pdx1-Cre) mice offer a critical resource for the testing of potential therapeutic targets 31 . As observed in human pancreatic tumours and tumour cell lines, FAM83A expression was significantly elevated in pancreatic tumours isolated from the KPC mice (Fig. 7a). Moreover, pancreatic epithelial cells from the LSL-KRAS-G12D LSL-p53-R172H (KP) mice transformed in vitro by infection with retroviral particles encoding Cre recombinase (KP-Cre) also expressed higher FAM83A levels relative to the control KP-pB cells (infected with control pBabe vector-expressing retrovirus; Fig. 7c). Expression of the mutant KRAS protein and stable p53 protein were confirmed in the Cre-transformed cells (KP-Cre) along with their ability to grow anchorage independently in soft agar (Fig. 7b). Similar to our results with the human PDAC cells, inhibitors to both MEK and ERK significantly inhibited FAM83A expression www.nature.com/scientificreports www.nature.com/scientificreports/ in the KP-Cre cells (Fig. 7d). Targeting FAM83A expression in the KP-Cre cells using shRNA against murine FAM83A (shA4) significantly reduced in vitro cell growth (Fig. 7e), anchorage-independent growth in soft agar (Fig. 7f) and in vivo tumour formation (Fig. 7g).
In summary, we have identified a MEK/ERK-AP1-FAM83A positive feed-back loop that promotes pancreatic cancer growth and tumorigenesis (Fig. 8). Dismantling the positive feed-back loop by targeting FAM83A expression or function would block essential MEK/ERK survival signalling and the MEK/ERK-regulated transcriptional activation in a tumour-specific manner. This strategy will have the additional advantage of a wider therapeutic window and lower side effect/toxicity to normal non-tumour cells.

Discussion
Identifying the genetic drivers responsible for carcinogenesis will help uncover molecular targets for future cancer therapies, screen for individuals at risk, and predict a patient's response to cancer therapies. We demonstrate here that the oncogene FAM83A is upregulated in pancreatic cancer cells, and functions to promote cancer cell survival. Importantly, we identify a MEK/ERK-FAM83A feed-forward signalling loop as the driver of oncogenic FAM83A expression and propose that patients with elevated FAM83A may benefit from therapeutic targeting of the MEK/ERK-FAM83A axis.
The FAM83 family was identified based on its transforming potential and ability to induce resistance to EGFR TKIs, solidifying its role in EGFR/MAPK signaling 15,19 . Since our initial discoveries, numerous reports have identified roles for various FAM83 proteins in cancer [32][33][34] . Moreover, genomics analyses have identified consistent upregulation of FAM83 member expression across the majority of tumour types, with high levels of FAM83 gene expression frequently associated with poor prognosis 18 . The smallest FAM83 member and focus of the current study, FAM83A, is not frequently mutated in cancer (except in skin cutaneous melanoma) 24,25 . Rather, it is overexpressed in lung, breast, ovary, colon and pancreas cancers. Beyond simply being overexpressed, FAM83A expression is an unfavourable prognostic gene in lung cancers 35,36 , and was documented as a prognostic biomarker (together with Keratin-19, Squamous carcinoma antigen (SCCA) and other proteins) capable of identifying circulating breast and lung cancer cells that predicted for shorter patient survival time 37 . More recently, Snijders et al. used a multi-omics approach to reveal increased expression of a FAM83 gene family was predictive of poor survival in many human cancers including pancreatic cancer 18 . Consistent with our findings, Chen et al. show that FAM83A expression is elevated in PDAC patients and high expression correlates with significantly www.nature.com/scientificreports www.nature.com/scientificreports/ poor overall and disease free survival, due in part to the gain of cancer stem cell features and chemoresistance to Gemcitabine 38 .
While little is known about what regulates FAM83A expression, it is clear that when overexpressed, FAM83A can lead to aberrant activation of MEK/ERK signalling through its interaction with C-RAF 16 , by potentially stabilizing signalling complexes. In publically available gene expression data 23 , elevated FAM83A expression correlates with an increased MAPK gene expression signature (Supplementary Fig. S5). While the impact of FAM83A expression on MEK/ERK signalling is of interest, even more intriguing is our finding that FAM83A gene expression is positively regulated by the AP1 transcription factors JUNB and FOSB downstream of MEK/ ERK signalling. Further promoter-binding studies are needed to confirm the direct role of JUNB and the identity of its partners. However, the consequence of this regulatory cascade is that, increased MEK/ERK activity increases FAM83A expression, which further increases MEK/ERK activity in a positive feed-forward signalling loop. Importantly, as shown here, suppression of FAM83A expression in human or murine pancreatic cancer cells can break this feed-forward loop, induce apoptosis, and thereby reduce the size of pancreatic tumours. Similar findings have been reported in human epidermal growth factor receptor 2 (Her2)-positive as well as triple negative breast cancer, both of which are aggressive tumour sub-types 16,39 . We therefore propose that, interfering with Importantly, the novel MEK/ERK-FAM83A self-sustaining loop described here may be independent of mutant KRAS expression, as knockdown of KRAS did not affect FAM83A expression, likely due to continued MEK/ERK activation independent of mutant KRAS. One possibility is the contribution of the remaining HRAS and NRAS. Hayes et al. reported enduring activation of ERK signalling in pancreatic cancer cells despite KRAS knockdown, reaffirming that in some pancreatic cancer cells, MEK/ERK activation is independent of KRAS 40 . In KRAS-independent pancreatic tumours, the continued activation of MEK/ERK signalling by other mechanisms may still drive the MEK-FAM83A positive feed-back loop. A recent kinome siRNA screen identified a remarkable diversity of kinases that contributed to resistance to MEK/ERK pathway inhibition 41 . These findings suggest that once cells are transformed, they can evolve a striking heterogeneity in proliferative and pro-survival signalling. For example, as in many other cancers, amplification and overexpression of EGFR activates oncogenic MEK and PI3K signalling in pancreatic cancers, even those harbouring mutant RAS 42 . Although the EGFR inhibitor Erlotinib has modestly improved patient survival, acquired resistance to this drug remains a major stumbling block. Given that elevated FAM83A expression results in acquired resistance to EGFR TKIs 19 , FAM83A expression may help predict the response to EGFR inhibitor therapy in patients with pancreatic cancer.
Interestingly, FAM83A was also reported to be one of the highly tyrosine-phosphorylated proteins in Trastuzumab-resistant HER2-positive breast cancer cells, and again, FAM83A ablation suppressed the growth of Trastuzumab-resistant cells 43 . The possibility that FAM83A acts as a scaffold, promoting or stabilizing oncogenic avian erythroblastosis oncogene B (ErbB)-MAPK signalling pathways and that its function may be regulated upon phosphorylation by the ErbB family of receptors such as EGFR and HER2 opens up novel targeting avenues. Moreover, we postulate that the MEK/ERK-FAM83A feed-forward loop may be a novel therapeutic target in pancreatic cancer cells, including those that remain dependent on mutant RAS, as well as those that are independent of mutant RAS.
FAM83A is mostly undetectable in normal tissues, thus, tumours with elevated FAM83A expression are postulated to have a wide therapeutic window 17 . Finding methods to suppress FAM83A expression or prevent FAM83A from serving as a signalling scaffold are leading potential therapeutic approaches. Recently, small molecules have been successfully designed to block protein-adaptor interactions in the RAS/RAF pathway such as Rocaglamide blocking prohibitin/C-RAF 44 and Rigosertib blocking RAS-RAF interaction 13 . Inhibitors or www.nature.com/scientificreports www.nature.com/scientificreports/ peptides to weaken the FAM83A/C-RAF interactions or the FAM83A/EGFR interaction by specifically targeting the amino-terminal DUF1669 responsible for the signalling interactions are potential avenues to directly target FAM83A function. Another feasible option is employing nanoparticles encapsulating siRNA targeting oncogenic FAM83A mRNA. Diverse kinds of non-immunogenic nanoparticles with increased in vivo stability and armed with tumour-homing peptides are currently in clinical trials for targeting oncogenic mRNAs in many solid tumors [45][46][47] . The means of targeting oncogenic FAM83A in pancreatic cancers may still be elusive but it is very likely to overcome current therapeutic roadblocks with much less harm to normal healthy cells.

Materials and Methods
Mice. LSL KRAS G12D (01XJ6) 48 , p53 LSL R172H (01XAF) 49 and Pdx-1-Cre/+(01XL5) 50  Cell culture. All PDAC cell lines were obtained from ATCC (Manassas, VA, USA) and grown in a humidified atmosphere containing 5% CO 2 at 37 °C. HPAC, Panc 04.03, Panc 08.13 and Panc 05.04 cells were grown in medium described previously 51 . Cells were used from frozen stocks within 10 passages upon receipt from ATCC. Cells in culture were routinely tested for mycoplasma using the MycoAlert Detection kit (LT07-218; Lonza, Basel, www.nature.com/scientificreports www.nature.com/scientificreports/ Switzerland). Normal pancreatic tissue was obtained from a pancreatic resection biopsy and the tissue identified as normal by the pathologist. All human tissue samples were obtained under discarded tissue protocol (IRB number CASE12Z12) in full accordance with the guidelines and regulations approved by Case Western Reserve University Institutional Review Board. The institutional review board waived informed consent for this discarded tissue protocol. (e-g) KP-pB and KP-Cre cells were subjected to FAM83A knockdown using lentivirus encoding shRNA targeting mouse FAM83A (shA4) or control (shGFP) and assessed for FAM83A expression by qRT-PCR (e; left), growth in culture (e; right). KP-pB and KP-Cre cells with FAM83A knockdown were also assessed for anchorage independent growth in soft agar (f) and for in vivo tumorigenicity upon subcutaneous injections (n = 4) in the flanks of immunocompromised mice (g). Images of the tumours were from day 21 post inoculation. All data in (b-g) are representative of two independent experiments. All bar graphs show mean ± SD. Student's t-test was used to assess statistical significance with ** and *** representing p < 0.01 and p < 0.001 respectively. (2019) 9:13396 | https://doi.org/10.1038/s41598-019-49475-5 www.nature.com/scientificreports www.nature.com/scientificreports/ For pancreatic ductal epithelial cell cultures, pancreas removed from KRAS G12D p53 R172H (KP) mice were minced with scalpels and digested in digestion medium containing DMEM/F-12, Penicillin/Streptomycin (#30-009-CI; Corning), Collagenase (#C0130; Sigma Aldrich), Hyaluronidase (#H3506; Sigma Aldrich), Insulin, FCS and Polymyxin-B sulphate (#P4932; Sigma Aldrich) for 20 min at 37 °C. The digested solution was spun at 300 g for 5 min and pellet resuspended in M87 culture medium (composition as described in 52 ). This suspension was filtered through 100 μ filter and the filtrate was passed through 40 μ filter. The ductal organoids stuck on the 40 μ filter were washed and collected in a fresh tube and spun at 300 g for 5 min. KP ductal organoids were grown on Corning TM Biocoat TM Collagen I-coated multiwell plates in M87 medium. The cells that grew out onto the dishes were trypsinized and passaged in culture as needed.
The MEK Inhibitor (U0126; #70970) and PI3K inhibitor (LY294002; #70920) were both used at 50 μM and obtained from Cayman Chemicals (Ann Arbor, MI, USA); ERK inhibitor used at 10 μM (SCH772984; #S7101) was from Selleck Chemicals (Houston, TX, USA). Cells were pretreated for 2 h with the inhibitors prior to use in experiments. For FAM83A expression, cells serum starved for 16 h were stimulated with 10% serum for 4 h following which cells were harvested for RNA extraction.
Short hairpin RNA and Small interfering RNA mediated knockdown. Lentiviruses were packaged and used to infect target cells as previously described 53  pBabe-puro-control and pBAbe-puro-cre have been described previously 15 . Retroviruses were generated using protocol previously described [54][55][56] . KP murine pancreatic ductal epithelial cells were transduced with retroviral particles encoding either pBABE-puro-control or pBabe-puro-cre. The Dharmacon (Lafayette, CO, USA) ON-TARGET plus Human FAM83A (84985); KRAS (3845); EGR1 (1958); FOSB (2354); JUNB (3726) siRNA SMARTpools, were obtained from GE Healthcare. siRNA transfections were done as per manufacturer's instructions for Lipofectamine RNAiMAX (#13778-075; Invitrogen, Thermo Fisher Scientific, Waltham, MA, USA) and siRNA transfected cells were used for experiments 24 h post-tranfection. pFLUG-GFP-LUC (Firefly) fusion construct used to build cells for bioluminescence measurement has been described previously 51,57 . Growth assays and three dimensional organotypic cultures. For growth assays, 25 000 cells/well were plated in six-well plates (in triplicates) for the indicated days and total cell number for each well was quantified using a Beckman Coulter counter (Beckman Coulter, Brea, CA, USA). Average cell number per well for each www.nature.com/scientificreports www.nature.com/scientificreports/ triplicate on day 4 is shown. For three-dimensional cultures, soft agar assays and organotypic culture in matrigel were as described previously 54,56 . All data presented are representative of three independent, biological replicates. Error bars represent standard deviation.
Flow cytometry, cell proliferation and apoptosis assays. All assays used HPAC cells 6 days after transduction with lentiviral particles expressing shFAM83A. 1-2 × 10 6 HPAC cells were labelled with 1 μM CFSE (#21888; Sigma) for 15 min at 37 °C, washed well with 1x PBS and cultured in replicates of 1 × 10 5 cells that were monitored for CFSE dilution by flow cytometry over 5-7 days. For BrdU assays, cells were labelled in culture with BrdU at 37 °C for 3 h and incorporation measured by staining with anti-BrdU FITC antibody as per manufacturer's protocol for BD Pharmingen BrdU FITC flow kit (#51-2354AK, San Jose, CA, USA).
For cell cycle analysis, cells were washed with PBS, fixed with cold 70% ethanol for an hour at −20 °C. Fixed cells were washed with PBS and treated with 20 μg/ml of RNase A for 30 min at 37 °C. Cells on ice were stained with 50 μg/ml of Propidium Iodide for 60 min prior to flow cytometry. BD Biosciences LSR II was used for CFSE and BrdU analysis and Attune NxT (Thermo Fisher Scientific) was used for the cell cycle analysis. Data were analysed using FACSDiva software version 6.2 (BD Biosciences, San Jose, CA, USA) or FlowJo for Windows Version 10.1 (Ashland, OR, USA).

Mouse xenografts. Athymic NCr (nude) mice were bred and maintained at the Athymic Animal and
Xenograft Core facility at the Case Western Reserve University Case Comprehensive Cancer Center. For tumorigenicity assays, 1 × 10 6 of HPAC-shGFP or HPAC-shA2 cells were resuspended in 50 μl of 1:1 mix of matrigel:media mix. Orthotopic pancreatic injections were performed on 8-12 week old male or female Nude mice as described 51 previously. KPC-shGFP and KPC-shA2 cells were injected in 100 μl matrigel-media mix into the flanks of the mice. For all experiments, tumour volume was also measured manually using calipers and tumour weight assessed following euthanasia and tumour retrieval. All animals were used in compliance with the guidelines approved by the Case Western Reserve University Institutional Animal Care and Use Committee.
RNA sequencing and analysis. Quality control of total RNA samples was executed using Qubit (Invitrogen) for quantification and Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA) analysis to assess quality using a cut-off of RIN > 7.0 to select specimens for further analysis. For library preparation, the Illumina TruSeq Stranded Total RNA kit (Illumina Inc, San diego, CA, USA) with Ribo Zero Gold for rRNA removal was used. This protocol starts by using the Ribo-Zero kit to remove ribosomal RNA (rRNA) from 0.10-1 µg of Total RNA using a hybridization/bead capture procedure that selectively binds rRNA species using biotinylated capture probes. The resulting purified mRNA is used as input for the Illumina TruSeq kit in which libraries are tagged with unique adapter-indexes. Final libraries were validated on the Agilent 2100 Bioanalyzer, quantified via qPCR and pooled at equimolar ratios. Pooled libraries were diluted, denatured and loaded onto the Illumina HiSeq. 2500 using a paired-end Rapid Run flowcell. Following sequence alignment, the differentially expressed genes that are in common between the two distinct shRNAs were intersected and only annotated coding genes were selected. There were 542 differentially expressed genes that become differentially expressed ( ≥ 2 fold change, p < 0.05). Heatmaps of selected signature genes were generated using Hierarchical Clustering and Hierarchical ClusteringImage modules available from GenePattern (Broad Institute, University of California, San Diego, CA, USA). Gene set enrichment analysis was performed using the g:Profiler database (http://biit.cs.ut.ee/ gprofiler/index.cgi) 59,60 . Data and statistical analysis. GraphPad Prism (La Jolla, CA, USA) and Microsoft Excel were used to plot data and Student's t test or non-parametric Mann-Whitney test were used to calculate statistical significance.

Data Availability
The datasets analysed during the current study are available from the corresponding author on reasonable request.