Abstract

   Genetic approaches in Drosophila have successfully identified many
   genes involved in regulation of growth control as well as genetic
   interactions relevant to the initiation and progression of cancer in
   vivo. Here, we report on large-scale RNAi-based screens to identify
   potential tumor suppressor genes that interact with known
   cancer-drivers: the Epidermal Growth Factor Receptor and the Hippo
   pathway transcriptional cofactor Yorkie. These screens were designed to
   identify genes whose depletion drove tissue expressing EGFR or Yki from
   a state of benign overgrowth into neoplastic transformation in vivo. We
   also report on an independent screen aimed to identify genes whose
   depletion suppressed formation of neoplastic tumors in an existing
   EGFR-dependent neoplasia model. Many of the positives identified here
   are known to be functional in growth control pathways. We also find a
   number of novel connections to Yki and EGFR driven tissue growth,
   mostly unique to one of the two. Thus, resources provided here would be
   useful to all researchers who study negative regulators of growth
   during development and cancer in the context of activated EGFR and/or
   Yki and positive regulators of growth in the context of activated EGFR.
   Resources reported here are available freely for anyone to use.

   Keywords: Tumorigenesis, Neoplasia, Drosophila, EGFR, Hippo pathway
     __________________________________________________________________

   Studies in genetic models of tissue growth have identified networks of
   signaling pathways that cooperate to control growth during animal
   development (reviewed in ([52]Harvey et al. 2013; [53]Richardson and
   Portela 2017). Normal tissue growth involves controlling the rates of
   cell proliferation and cell death, as well as cell size, cell shape,
   etc. Signaling pathways mediate hormonal and neuroendocrine regulation
   of growth, which depend on nutritional status. Cell interactions also
   contribute to coordinating growth of cells within a tissue.

   Growth regulatory pathways include both positive and negative elements
   to allow for feedback regulation. These feedback systems confer
   robustness to deal with intrinsic biological noise, and with a
   fluctuating external environment ([54]Herranz and Cohen 2010). They
   also provide the means for different regulatory pathways to interact
   ([55]Ren et al. 2010; [56]Herranz et al. 2012a; [57]Reddy and Irvine
   2013). In the context of tumor formation, this robustness is reflected
   in the difficulty in generating significant misregulation of growth - a
   twofold change in expression of many growth regulators seldom has a
   substantial effect on tissue size in Drosophila genetic models. More
   striking is the difficulty in transitioning from benign overgrowth to
   neoplasia: hyperplasia does not normally lead to neoplasia without
   additional genetic alterations (e.g., ([58]Huang et al. 2005;
   [59]Herranz et al. 2012b [60]2014).

   Cancers typically show mis-regulation of multiple growth regulatory
   pathways. Mutational changes and changes in gene expression status
   contribute to driving cell proliferation, overcoming cell death and
   cellular senescence, as well as to allowing cells to evade the
   checkpoints that normally serve to eliminate aberrant cells. These
   changes alter the normal balance of cellular regulatory mechanisms,
   from initial cellular transformation through disease progression
   ([61]Stratton 2011; [62]Alexandrov et al. 2013). For many tumor types,
   specific mutations have been identified as potent cancer drivers, with
   well-defined roles in disease ([63]Kandoth et al. 2013; [64]Zehir et
   al. 2017). However, most human tumors carry hundreds of mutations,
   whose functional relevance is unknown. The spectrum of mutation varies
   from patient to patient, and also within different parts of the same
   tumor ([65]McGranahan and Swanton 2017). Evidence is emerging that some
   of these genetic variants can cooperate with known cancer drivers
   during cellular transformation or disease progression. The mutational
   landscape of an individual tumor is likely to contain conditional
   oncogenes or tumor suppressors that modulate important cellular
   regulatory networks.

   Sequence-based approaches used to identify cancer genes favor those
   with large individual effects that stand out from the ‘background
   noise’ of the mutational landscape in individual cancers ([66]Stratton
   2011; [67]Alexandrov et al. 2013). In vivo experimental approaches are
   needed to assign function to candidate cancer genes identified by tumor
   genome sequencing, and to identify functionally significant
   contributions of genes that have not attracted notice in genomics
   studies due to low mutational frequency, or due to changes in activity
   not associated with mutation. In vivo functional screens using
   transposon mutagenesis of the mouse genome have begun to identify
   mutations that cooperate with known cancer driver mutations, such as
   K-Ras, in specific tumor models ([68]Copeland and Jenkins 2010;
   [69]Pérez-Mancera et al. 2012; [70]Takeda et al. 2015). Genetic
   approaches using Drosophila models of oncogene cooperation have also
   been used to identify genes that act together with known cancer drivers
   in tumor formation ([71]Brumby and Richardson 2003; [72]Pagliarini and
   Xu 2003; [73]Wu et al. 2010; [74]Brumby et al. 2011; [75]Herranz et al.
   2012b [76]2014; [77]Eichenlaub et al. 2016; [78]Richardson and Portela
   2017; [79]Song et al. 2017). The simplicity of the Drosophila genome,
   coupled with the ease of large-scale genetic screens and the high
   degree of conservation of major signaling pathways with humans, make
   Drosophila an interesting model to identify novel cancer genes and to
   study the cellular and molecular mechanisms that underlie tumor
   formation in vivo (reviewed in ([80]Gonzalez 2013; [81]Herranz et al.
   2016; [82]Sonoshita and Cagan 2017; [83]Richardson and Portela 2018).

   In Drosophila, overexpression of the Epidermal Growth Factor Receptor,
   EGFR, or Yorkie (Yki, the fly ortholog of the YAP oncoprotein) cause
   benign tissue over-growth ([84]Huang et al. 2005; [85]Herranz et al.
   2012a [86]2014). Combining these with additional genetic alterations
   can lead to neoplastic transformation and eventually metastasis
   ([87]Herranz et al. 2012b [88]2014; [89]Eichenlaub et al. 2016,
   [90]2018; [91]Song et al. 2017). Here, we report results of large-scale
   screens combining UAS-RNAi transgenes with EGFR or Yki expression to
   identify negative regulators of these growth regulatory networks that
   can lead to aggressive tumor formation in vivo. We also performed an
   independent screen to identify factors that could suppress EGFR-driven
   neoplasia. These screens have identified an expanded genomic repertoire
   of potential tumor suppressors that cooperate with EGFR or Yki. We have
   also identified few positive regulators of growth in the context of
   activated EGFR. Interestingly, there was limited overlap among the
   genes that cooperated with EGFR and those that cooperated with Yki.
   Gene intractome analysis and analyses of cancer databases for human
   orthologs of positives of these screens suggest that a large number of
   them have strong correlations to many clinical parameters. The output
   of this screen would, therefore, be useful to all researchers who study
   negative regulators of growth during development and cancer in the
   context of activated EGFR and/or Yki. Resources reported here are
   freely available for anyone to use.

Materials and Methods

RNAi Screens

   The KK transgenic RNAi stock library was obtained from the Vienna
   Drosophila RNAi Center ([92]www.vdrc.at; also listed in Table S1)
   carrying inducible UAS-RNAi constructs on Chromosome II. For each
   cross, 5 males from the KK transgenic RNAi stock were crossed
   separately to 10-15 virgins from each of the following three driver
   stocks (see Supplemental Fig. S1A for the schematics of fly stocks):
   w*, ap-Gal4, UAS-GFP/CyO; UAS-Yki, tub-Gal80^ts/TM6B (Yki driver;
   [93]Song et al. 2017); w*; ap-Gal4, UAS-GFP/CyO; UAS-EGFR,
   tub-Gal80^ts/TM6B (EGFR driver; [94]Herranz et al. 2012b); and w*;
   ap-Gal4, UAS-GFP/CyO; and w*; ap-Gal4, UAS-GFP, Socs36E^RNAi/CyO;
   UAS-EGFR, tub-Gal80^ts/TM6B (EGFR driver +SOCS36E^RNAi). The
   combination of UAS-EGFR and UAS- SOCS36E^RNAi inducing tumorous growth
   is reported in [95]Herranz et al. (2012b).

   Virgin female flies were collected over 4-5 days and stored at 18° in
   temperature-controlled incubators on medium supplemented with dry
   yeast, prior to setting up crosses. Virgin females were mated to KK
   stock males (day 1) and the crosses were stored at 18° for 4 days to
   provide ample time for mating before starting the timed rearing
   protocol used for the screen. On day 5, the crosses were transferred
   into new, freshly yeasted vials for another 3 days at 18°. On day 8,
   the adult flies were discarded, and larvae were allowed to develop
   until day 11, at which time the vials were moved to 29° incubators to
   induce Gal4 driver activity. Crosses were aged at 29° for a further 8-9
   days, after which larvae were scored for size and wing disc overgrowth
   phenotypes for Yki and EGFR driver screen crosses. Flies were scored
   for suppression of the tumor phenotype for the EGFR driver
   +SOCS36E^RNAi crosses (see Supplemental Fig. S1B for the screen
   workflow).

   In order to verify the integrity of the driver stocks during the course
   of the screen, we examined their expression patterns in conjunction
   with setting up screen crosses each week. For each driver, 2-3 of the
   bottles used for virgin collection were induced at 29° for 24 hr and
   analyzed using fluorescence microscopy for apterous-Gal4 specific
   expression in wandering 3-instar larvae (see Supplemental Fig. S2 for
   larval images of quality control). Any batch that showed tumorous
   growth on its own without a cross with KK-RNAi line (in case of SOCS
   stocks, if the batch didn’t show tumorous growth) were discarded and
   new batches were made from the original clean stock.

   Positive hits form the initial screen were retested by setting up 2 or
   more additional crosses. The hits were scored as verified if 2 out of 3
   tests scored positive. Wandering third instar larvae of confirmed
   positives were imaged and documented using fluorescence microscopy.

Genomic DNA PCR 40D landing site occupancy test

   Genomic DNA from a select number of Drosophila KK transgenic RNAi
   library stocks was isolated following a protocol available at the VDRC
   ([96]www.vdrc.at). The presence or absence of the KK RNAi transgene at
   the 40D insertion site on the second chromosome was determined by
   multiplex PCR using the following primers:

   40D primer (C_Genomic_F): 5′-GCCCACTGTCAGCTCTCAAC-3′

   pKC26_R: 5′-TGTAAAACGACGGCCAGT-3′

   pKC43_R: 5′-TCGCTCGTTGCAGAATAGTCC-3′

   PCR amplification was performed using GoTaq G2 Hot Start Green Master
   Mix kit (Promega) in a 25 µL standard reaction mix and the following
   program: initial denaturation at 95° for 2 min, followed by 33 cycles
   with denaturation at 95° for 15 sec, annealing at 58° for 15 sec and
   extension at 72° for 90 sec. One final extension reaction was carried
   out at 72° for 10 min. Reactions were stored at -20° prior to gel
   loading. PCR using these primers generate an approximately 450 bp
   product in case of a transgene insertion or a 1050 bp product in case
   of no transgene insertion site at 40D.

Screen database

   Results from the three screening projects were added to a screen
   management database, [97]http://www.iiserpune.ac.in/rnai/, including
   images of positive hits and background information such as RNAi line
   ID, corresponding gene information from the Flybase etc. The database
   was developed by Livetek Software Consultant Services (Pune,
   Maharashtra, INDIA).

Pathway and gene set enrichment analysis

   Gene set enrichment analysis was performed using genes that upon down
   regulation induced tumor formation (EGFR, YKI background) or suppressed
   tumor formation (EGFR+SOCS background). For D. melanogaster enrichment
   analysis all D. melanogaster protein coding genes were used as the
   “gene universe” together with organism specific datasets. For human
   ortholog enrichment analysis all human protein coding genes were used
   as the “gene universe” together with organism specific datasets. The
   algorithm packages and databases used in analysis are listed in
   Supplemental Tables S2 and S3. Unless otherwise specified, pathway
   databases included in these packages were used. The KEGG database was
   downloaded directly from source on 10.10.2018. Organ system specific
   and disease related pathway maps were excluded from this analysis.
   Minimum and maximum number of genes per pathway or gene set,
   significant criteria, minimum enriched gene count and annotated gene
   counts for each test and database are indicated in Supplemental Tables
   S2 and S3. GO results were filtered for level >2, to eliminate broad
   high-level categories and <10 to minimize duplication among
   subcategories. A representative term was selected in the cases were
   identical set of genes mapped to multiple terms within the same
   database. After filtering, the top 10 terms from each database were
   used for clustering analysis.

   Pathway and gene set enrichment analysis results were visualized as
   enrichment map with appropriate layout based on gene overlap ration
   using igraph. Gene overlap ratio was set as edge width. Edges with low
   overlap were deleted, filtering threshold was based on a number of
   “terms” in the results table – from 0 to 50 by 10; increasing filtering
   thresholds from 0.16 to 0.26 by 0.2. Clusters were detected using “Edge
   betweenness community” algorithm. Similar biological processes were
   color-coded.

R packages

   clusterProfiler (3.8.1) - ([98]Yu et al. 2012).

   ReactomePA (1.24.0) - ([99]Yu and He 2016).

   [100]http://pubs.rsc.org/en/Content/ArticleLanding/2015/MB/C5MB00663E.

   graphite (1.26.1) - Sales G, Calura E, Romualdi C (2018). graphite:
   GRAPH Interaction from pathway Topological Environment. R package
   version 1.26.1.

   igraph (1.2.2) - Csardi G, Nepusz T: The igraph software package for
   complex network research, InterJournal, Complex Systems 1695. 2006.
   [101]http://igraph.org

Database references