Exploiting codon usage identifies intensity-specific modifiers of Ras/MAPK signaling in vivo

Signal transduction pathways are intricately fine-tuned to accomplish diverse biological processes. An example is the conserved Ras/mitogen-activated-protein-kinase (MAPK) pathway, which exhibits context-dependent signaling output dynamics and regulation. Here, by altering codon usage as a novel platform to control signaling output, we screened the Drosophila genome for modifiers specific to either weak or strong Ras-driven eye phenotypes. Our screen enriched for regions of the genome not previously connected with Ras phenotypic modification. We mapped the underlying gene from one modifier to the ribosomal gene RpS21. In multiple contexts, we show that RpS21 preferentially influences weak Ras/MAPK signaling outputs. These data show that codon usage manipulation can identify new, output-specific signaling regulators, and identify RpS21 as an in vivo Ras/MAPK phenotypic regulator.


Introduction
Conserved signal transduction pathways are employed throughout nature during diverse processes such as cell fate decisions and tissue growth. These same pathways can be aberrantly phenotypes that are specific to only strong or only weak Ras/MAPK signaling. Our screen specifically looked for modifiers unique to specific Ras signaling states, by leveraging the differential signaling phenotypic output driven by rare versus common codons in the Ras gene. Importantly, the Ras gene enriched in rare codons used in our screen models more closely the rare codon-enriched sequence of human KRAS [27], which is the most frequently mutated RAS family member in human cancers [28]. Our screen enriched for genomic regions not previously ascribed to Ras phenotypic modification. Of the 15 Dfs identified, we successfully mapped the modification of Df(2L)BSC692, an enhancer of the rough-eye phenotype driven only by weak Ras/MAPK signaling, to the ribosomal protein S21 gene (RpS21). We show that RpS21 negatively regulates Ras protein levels in several contexts, the effect of which is preferentially manifested at low levels of MAPK signaling. This approach highlights the usefulness of codon manipulation as a viable approach to identify signal output-specific signaling regulation and introduces new genetic reagents to explore weak Ras signaling regulation in Drosophila. Our uniquely identified modifiers include those specific to Ras with rare codons, like that of human KRAS.

Exploiting codon usage to control MAPK signaling output
To identify Ras/MAPK molecular regulators that differentially impact strong or weak signaling outputs, we required a platform to tightly control the strength of MAPK signaling. To activate the pathway, we expressed a highly conserved, mutant active (G12V) Drosophila Ras transgene (termed Ras V12 here for convenience). To control MAPK signaling strength during fly development, we opted for the new approach of simply changing the codon usage of a Ras V12 transgene. Codons that occur infrequently in a given genome (rare codons) are known to impede translation, including in Drosophila [29][30][31][32][33][34][35]. By engineering a gene enriched in rare codons for each given amino acid, it is possible to create an mRNA that is poorly translated without altering the amino acid sequence of the encoded protein [36,37]. This has the distinct advantage that control of protein expression is embedded in the DNA and requires no additional factors or experimental variables. We used established data on Drosophila codon usage (see Methods) and created four distinct versions of Drosophila Ras transgenes: 1) we altered none of the codons (Ras V12 Native), 2) we made all codons the most commonly occurring in the genome (Ras V12 Common), 3) we made all codons the most rare in the genome (Ras V12 Rare), and 4) we created a control wild-type version lacking the V12 mutation and also lacking codon alteration (Ras WT Native). To monitor expression, all four transgenes were epitopetagged at the N-terminus with a 3XFLAG-tag sequence and expressed under the control of a Gal4-inducible UAS promoter (Fig 1A, see Methods). We note that Ras V12 Native has primarily common codons and a similar Codon Adaptation Index (CAI [38]) to Ras V12 Common [24], while the CAI for Ras V12 Rare is much lower (S1A and S1B Fig). To control for position effects, all transgenes were integrated at the same site in the genome (see Methods). Our altering of the codon sequence yielded a Drosophila Ras V12 Rare transgene that has a closer nucleic acid identity to the human KRASB isoform than the endogenous Drosophila Ras85D sequence (S1C- S1E Fig).
To measure signaling output strength of our transgenes, we first chose to use an in vivo phenotypic readout rather than a biochemical readout, an approach validated by quantitative studies of MAPK activation in Drosophila embryos [20,21]. For genetic screening of Ras/MAPK phenotypic regulators, the Drosophila eye is a highly accessible model. Driving expression of Ras V12 in the developing eye with an eye-specific promoter such as sevenless (sev) dysregulates the proper differentiation of the R7 photoreceptor cell, leading to an easily scored 'rough-eye' phenotype [39,40]. This phenotype relies on Ras action through the conserved MAPK pathway [2,41].
We assayed the phenotypic output of each Ras transgene in vivo by driving their expression in the developing fly eye using sevenless (sev)-Gal4. As expected [39], expression of Ras WT Native in this manner does not result in a rough-eye phenotype (S2A Fig). However, when we expressed the constitutive-active versions of Ras (Ras V12 ), we found a range of rough-eye phenotypes ( Fig 1B). We binned these phenotypes into one of three classes: severe, moderate, or mild. Each class was assigned an increasing numeric score, based on the incidence and severity of eye phenotypes such as necrotic spots and discoloration (Fig 1B, see Methods). We then calculated an average severity score for each Ras transgene. Ras V12 Native and Ras V12 Common animals exhibit a similar phenotypic score, reflecting their similar CAI. Further, this phenotypic score is, on average, approximately 2-fold more severe than that of Ras V12 Rare (Fig 1C).
To determine whether Ras protein levels track with the difference in Ras-driven rough-eye phenotype, we isolated heads from flies encoding common and rare Ras V12 transgenes and performed serial dilution immunoblotting with an anti-FLAG antibody (Figs 1D and S2D). Separately, flies expressing all three active Ras transgenes were again immunoblotted with an anti-FLAG antibody, and protein levels were normalized to a loading control (S2B and S2C  Fig). In both experiments, we found Ras V12 Common flies express roughly 1.5 to 1.9-fold more Ras protein than flies expressing Ras V12 Rare (Figs 1D, and S2B-S2D). Additionally, Ras V12 protein levels are similar between Ras V12 Native and Ras V12 Common flies (S2B and S2C Fig), which is consistent with the similar codon content between these transgenes (S1 Fig). These experiments established that codon usage can be manipulated to examine an in vivo, Ras signal-driven output (eye phenotype), and identified both weak (Ras V12 Rare) and strong (Ras V12-Common) versions of this output. Further, these results are consistent with Ras V12 Rare serving as a model of the rare codon bias of human KRAS, the most commonly mutated RAS family member in human cancers.
We next assessed the impact of codon content in the Ras gene on Ras signaling and Ras GTPase activity. To examine the effect of expression of Ras V12 Rare versus Ras V12 Common transgenes on MAPK signaling, we measured the level of phosphorylated Mek (p-Mek, Fly-Base: Dsor) and Erk (p-Erk, FlyBase: rolled) compared to the total level of these proteins by immunoblot analysis. Ras V12 Common animals exhibit elevated levels of p-Erk and p-MEK compared to Ras V12 Rare fly heads (Figs 1E-see figure legend for quantitation and S2E). We independently verified this difference in cultured S2 and KC insect cells (see Methods), again finding that Ras V12 Common is expressed higher and more robustly activates the MAPK pathway compared to Ras V12 Rare (S1F Fig). Further, using a Ras binding domain (RDB) pulldown assay [42], we found that S2 cells expressing Ras V12 Common contain a higher total level 10,20, and 30 ug of lysates derived from the heads of flies expressing the indicated versions of transgenic Ras V12 were immunoblotted with an anti-FLAG antibody. Bottom: quantification and protein loaded. (e) Immunoblot detection of phosphorylated (p-) and total Mek and Erk, and actin as a loading control from lysates derived from the head of flies with the indicated versions of transgenic Ras V12 . The ratio of pErk/Erk and pMek/Mek for Ras V12 Rare is 0.89 and 0.94, respectively. The ratio of pErk/Erk and pMek/Mek for Ras V12 Rare is 1.43 and 2.05, respectively. (f) Quantitative RT-PCR, measured using 2^ΔΔCt, of animals expressing the indicated versions of transgenic Ras V12 . Paired T-test. Data represent three independent replicates per condition, with 10-40 animals/replicate. One-way ANOVA and Tukey's multiple comparisons test. (g) Percentage of animals surviving to adulthood after larval induction of FLP-out somatic clones of a Ras transgene using Tubulin-Gal4 (3 replicate experiments, N = 32-55 animals/genotype/ replicate). One-way ANOVA and Tukey's multiple comparisons test were used for statistical comparisons. (h) Images representing the leg imaginal disc Tubulin-Gal4 FLP-out clone sizes generated in the indicated genotype backgrounds. Scale bars = 20 um. (i) The mean +/-SEM leg imaginal disc clone size in pixels for each genotype ( of active Ras than cells expressing Ras V12 Rare (S1G Fig). In sum, our findings establish Ras V12-Rare and Ras V12 Common as two distinct transgenes that either weakly or strongly activate Ras/ MAPK signaling output (as measured by Ras V12 protein expression, Ras activity, and MAPK activation), and that transgene-driven signal strength tracks with an observable difference in phenotypic output.
Codon bias has been shown to impact not only translation fidelity and efficiency [43][44][45][46][47] but also pre-translational processes, including transcription [48,49] and mRNA stability [50][51][52][53]. To assess whether our codon-altered Ras transgenes impact Ras protein levels and Erk signaling through pre-translational processes, we performed quantitative RT-PCR (Methods). Paralleling our findings with Ras protein, Ras V12 Common mRNA is significantly higher than Ras V12 Rare in adult heads (Fig 1F). These findings are consistent with the model that altering codon usage of Drosophila Ras impacts Ras RNA, Ras protein, and Erk signaling.
We next examined the impact of Ras V12 Rare and Ras V12 Common at single cell resolution. Using the FLP-out system [54], we generated mosaic clones of cells throughout developing larvae that expressed these transgenes under a ubiquitous Tubulin-Gal4 driver. Clones generated in this system are marked with GFP. We used heat shock to control the frequency of FLP-out events, and used a level of heat shock that resulted in 1-2 discrete clones in leg imaginal discs (see Methods). In these animals, Ras V12 Common, but not Ras V12 Rare or Ras WT Native, mosaic expression leads to animal lethality (Fig 1G). We note that Ras V12 expression is connected to animal lethality in other contexts, including when induced transiently or in somatic clones [55][56][57]. In surviving animals, Ras V12 Rare significantly increases clone size relative to controls (Fig 1H and 1I), which is consistent with the well-known role of Ras/Erk signaling in promoting cell proliferation. Interestingly, in surviving Ras V12 Common animals, clones are no bigger than in RasNative controls (Fig 1H and 1I). Taken together with the organismal death and frequent necrotic spots seen in the eyes of sev-Gal4, UAS-Ras V12 Common animals, we interpret this result to likely reflect the increased apoptosis or cellular senescence that can result from increased Ras expression [58,59]. Given that Ras V12 Rare and Ras V12 Common have such differing effects on cell proliferation in leg discs, our results underscore the critical importance of signal output levels on Ras-driven phenotypes and highlight that lower Ras levels can actually drive more cell proliferation in specific contexts.
A genome-wide screen uncovers differential phenotypic regulation between strong and weak Ras/MAPK signaling states We next sought to use our codon alteration system to gain insight into how the Ras/MAPK pathway can be differentially regulated in different signal-strength states. To do so, we screened for molecular regulators that modify Ras/MAPK phenotypes driven only by strong or only by weak signaling states. We first confirmed that Ras V12 Common and Ras V12 Rare rough-eye phenotypes were both in the range that can be modified. Specifically, two different heterozygous loss-of-function mutations known to suppress active Ras phenotypes, namely the S-627 allele of kinase suppressor of ras, (FlyBase: ksr) [9], and the S-2554 allele of beta subunit of type I geranylgeranyl transferase, (FlyBase: betaggt-I) [3]. As with previous work, we find these mutations suppress the rough-eye phenotype for Ras V12 Common and Ras V12 Rare (S3A Fig). Next, we examined heterozygous mutants of the yan-XE18 allele of anterior open, or aop, which is known to enhance the active Ras phenotype [60,61] Although we did not observe clear eye enhancement for aop yan-XE18 /+, we did observe a marked decrease in another phenotypic readout-animal survival. As for our FLP-out experiments with Tubulin-Gal4, sev-Gal4 expression of Ras V12 Common leads to considerably more organismal death than with Ras V12 Rare (S2B Fig). This sev-Gal4-driven lethality likely reflects the expression of sevenless-Gal4 in other tissues [62]. Survival is lower for aop yan-XE18 /+ animals expressing both Ras V12 Common and Ras V12 Rare transgenes (S3B Fig). These results establish that codonaltered Ras V12 transgenes are subject to phenotypic modification, including by dose-sensitive heterozygous mutations.
Previous modifier screens, including in the eye, employed the native Ras cDNA to express activated Ras [2,8,40,63]. This sequence has a strong common-codon bias (S1B Fig) and is similar to Ras V12 Common in terms of MAPK biochemical and phenotypic outputs (Fig 1). To find unidentified modifiers that may be specific to weaker (or stronger) Ras/MAPK-driven phenotypes, we conducted a genome-wide unbiased heterozygous mutant screen to specifically identify modifiers of the rough-eye phenotype driven by only Ras V12 Rare, (or only Ras V12 Common), (Fig 2A). We used the Bloomington Deficiency (Df) Kit, which covers 98.3% of the euchromatic genome [64] . In a primary screen (Fig 2B and S1 Table), we crossed 470 Dfs representing 99.1% of the Df collection to animals with Ras V12 Rare or Ras V12 Common expressed in the eye by sev-Gal4, and scored the resulting eye severity in an average of 30 (Ras V12 Common) or 60 (Ras V12 Rare) progeny animals per cross. We also factored animal lethality into our scoring (see Methods).
As expected, we found general Ras modifiers that either enhance or suppress eye phenotypes driven by both Ras V12 transgenes (Fig 2C and 2D and S1 Table). Interestingly, we identified more enhancers than suppressors (16% versus 7%, Fig 2C). The reason for this remains to be determined, but we note that our calculation of phenotypic modification (see Methods) included scoring animal lethality, which may identify strong enhancers of Ras V12 Common not identified in previous screens based solely on a rough-eye phenotype. Of great interest, we also identified Dfs whereby Ras V12 Common and Ras V12 Rare are differentially modified (Fig 2A), meaning they scored as only modifying the eye phenotype driven by a single signaling state (Ras V12 Common or Ras V12 Rare, not both). Using a low-stringency cutoff score (see Methods), we identified 178 putative differential modifier Dfs in our primary screen (Fig 2B and S1  Table). To filter our hits to those that were the most robust, these Dfs were then re-tested in a secondary screen (Fig 2B) by crossing them a second time to sev-Ras V12 Common and sev-Ras V12 Rare. In this screen, we used a more stringent cutoff score to ensure repeatability to define a robust differential modifier (see Methods). This scoring and replicate analysis reduced the number of candidates to 15 Dfs, or 3% of the tested Dfs (Fig 2E and 2D and S1 Table), that reproducibly differentially modify either only Ras V12 Common or only Ras V12 Rare (Fig 3A). Among these differential modifiers, we again recovered more enhancers than suppressors, although importantly we recovered both enhancers of Ras V12 Common and suppressors of Ras V12 Rare, arguing that our screen had the dynamic range to modify both strong (Ras V12-Common) and weak (Ras V12 Rare) Ras/MAPK signaling outputs (Fig 3B).
We next queried both the general (signal output-independent) and differential (signal output-dependent) modifiers against a FlyBase database of all reported Ras genetic enhancers and suppressors (see Methods). 56% of our general modifier Dfs covered regions of the genome containing reported Ras enhancers or suppressors. These data support the idea that our approach can identify Ras eye modifiers. Additionally, we note that among our identified differential modifier Dfs, most (73%) do not encompass known Ras modifiers, supporting the idea that our signal strength-specific modifier hits are enriched in new Ras enhancers and suppressors (Fig 3C). To explore possible relationships amongst these 15 differential modifier Dfs, we queried the genes within differential versus enhancer and suppressor Dfs against the established list of FlyBase Gene Groups (FBGG). Interestingly, the gene groups enriched in the differential Dfs do not overlap with those in the general enhancer/suppressor Dfs (Fig 3D), suggesting that the differential modifiers may represent a distinct class of Ras modifiers. Unlike the general modifier Dfs, differential modifier regions are enriched for basic Helix Loop Helix (bHLH) transcription factors, potentially reinforcing their distinct regulation of Ras/MAPK signaling. In summary, by controlling Ras/MAPK signal output strength through codon usage and using a phenotypic output screen, we successfully identified Dfs that alter a Ras/MAPK phenotype in a signaling output-specific fashion.

RpS21 negatively regulates Ras/MAPK signaling in a signal strengthspecific manner
To identify a differential modifier from our screen at the single gene level, we focused on Df (2L)BSC692 as it was one of the smallest deficiencies, encompassing only 12 genes, that specifically enhanced Ras V12 Rare (Figs 3A and S4A). Of these 12 genes, Ribosomal protein S21, or RpS21 (also known as overgrown hematopoietic organs 23B/oho23B), represented a plausible candidate modifier. RpS21 stands out among small ribosomal subunits for its reported negative regulation of hematopoietic and imaginal disc hyperplasia [65]. To determine if RpS21 is a responsible gene in Df(2L)BSC692 for specifically enhancing Ras V12 Rare, we assessed the rough-eye phenotype of Ras V12 Common and Ras V12 Rare in the background of the mutant RpS21 03575 . Indeed, only the sev-Ras V12 Rare rough-eye phenotype is enhanced in the RpS21 03575 /+ background (Fig 4A). RpS21 03575 /+ did not score as a hit by our animal lethality criteria (see Methods), suggesting our comparison of eye phenotypes between control and mutant animals was not impacted by animal viability. We also note that the RpS21 03575 chromosome also carries a mutation in cinnabar (cn). However, cn mutations were also present in 4 other Dfs in our screen, only one of which was a hit. Therefore, RpS21 and not cn is the likely modifier on the RpS21 03575 mutant chromosome. Similar to our findings in the eye, RpS21 03575 /+ preferentially impacts the phenotype of Ras V12 rare leg imaginal disc clones. We observe smaller average clone sizes in RpS21 03575 /+, Ras V12 Rare animals relative to Ras V12 Rare alone, whereas RpS21 03575 /+ does not impact clone size in Ras V12 Common animals (Fig 4B  and 4C). Together, these findings identify RpS21 as a responsible modifier of Ras V12 Rare in one Df from our Ras/MAPK signal strength-specific screen.
From our genome-wide screen and follow-up mapping efforts, we were able to identify both an RpS21 mutant allele and a small deficiency encompassing this gene (Df(2L)BSC692) as differential Ras V12 eye phenotype modifiers. We next examined the molecular alterations of Ras signaling that underlie this signal intensity-specific modification. To this end, we assessed Ras V12 levels and/or MAPK pathway activation by immunoblot analysis in three distinct cellular and signal output settings: ectopic Ras activation in adult fly heads, ectopic Ras activation in cultured S2 cells, and endogenous MAPK signaling in ovaries. Our results overall show that while RpS21 reduction impacts Ras/MAPK in numerous settings, there is a more pronounced effect in cases where signaling output is weaker.
In the heads of Ras V12 Rare flies, transgenic Ras protein levels increase in RpS21 03575 /+ animals relative to wild type. This result is consistent with the enhanced Ras V12 Rare eye phenotype in RpS21 03575 /+ animals. However, unlike our lack of an observable phenotypic enhancement of Ras V12 Common in the eye, at the biochemical output level we also observe an increase in the level of Ras V12 Common in the RpS21 03575 /+ background (Figs 4D and S4B). This result shows that RpS21 03575 modifies both sevenless-driven Ras V12 Rare and  Ras V12 Common protein levels in the adult fly head, but only Ras V12 Rare modification leads to an observable phenotypic output in this setting. This difference between eye phenotype and protein level effects could suggest that a large difference in Ras protein change is needed to cause a detectable change at the eye phenotype level. Alternatively, our adult head assay focuses on Ras levels in the adult animal, whereas our eye assay focuses on the effect of RpS21 reduction during eye development. RpS21 03575 /+ does not impact Ras V12 Rare or Ras V12 Common RNA levels in adult heads, suggesting RpS21 acts at the translational level to impact Ras signaling (Fig 4E).
Next, we examined the impact of RpS21 on Ras signaling in additional cellular contexts. We first transduced S2 cells with an expression vector encoding either Ras V12 Common or Ras V12 Rare, and then used RNAi to reduce RpS21 levels. As in the fly head, RpS21 RNAi elevates Ras V12 Rare protein levels. However, unlike in the head, Ras V12 Common protein levels in S2 cells are unaffected by RpS21 RNAi (Figs 4F and S4C). We note that, in these cells, our expression system led to particularly robust expression of the Ras V12 Common protein (Figs 4F and  S4C). We also examined MAPK activation in S2 cells. Whereas p-Mek and p-Erk are noticeably increased in RpS21 RNAi S2 cells expressing Ras V12 Rare, we see no overt increase in these MAPK activation readouts upon RpS21 RNAi in S2 cells expressing Ras V12 Common (Figs 4F  and S4C). Taken together, our results in the head and in S2 cells suggest that when Ras signaling is above a particular threshold (e.g., Ras V12 Common expression in S2 cells), RpS21 reduction does not impact pathway output.
We also assessed whether endogenous MAPK signaling can be regulated by RpS21 in vivo. To do so, we examined the effect of disrupting one allele of the RpS21 gene on endogenous MAPK signaling in the ovaries of flies, a tissue where EGFR/Erk signaling has a well-defined role [66,67] and where phosphorylated Mek and Erk are readily detected (Fig 4G). Of note, RpS21 03575 /+ animals have no obvious female fertility defects. In this tissue, endogenous p-Mek and p-Erk levels increase in both Df(2L)BSC692/+ and RpS21 03575 /+ animals relative to control w 1118 animals (Fig 4G). Although we were not able to successfully determine endogenous Ras levels in the ovary with existing reagents (not shown), our overall findings are consistent with RpS21 negatively regulating endogenous Ras/MAPK signaling in this tissue. Collectively, we find that loss of RpS21 elevates Ras/MAPK signaling in multiple contexts.
Our immunoblot analysis validates our genetic screen finding that RpS21 can negatively regulate Ras and/or MAPK signaling, in a manner that potentially depends on the strength of Ras/MAPK signaling. One interpretation of these data is that RpS21 has a minimal effect on MAPK signaling output above a certain threshold of MAPK signaling. Such a model would predict that experimentally reducing the amount of Ras V12 Common expression should render fly eye development sensitive to the RpS21 03575 /+ mutant background. To experimentally test this threshold model, we took advantage of the well-known fact that expression of transgenes using the Gal4-UAS system is responsive to temperature, with higher temperature resulting in higher expression over the physiological range of 18˚C-29˚C. We thus evaluated the rough-eye phenotype of sev-Ras V12 Common versus sev-Ras V12 Rare flies in a wild-type versus RpS21 03575 / + mutant background, only this time at 18˚C. At this lower temperature, RpS21 03575 /+ now (RpS21 03575 /+) RpS21 backgrounds. Data represent three independent replicates per condition, with 10-40 animals/replicate. One-way ANOVA and Tukey's multiple comparisons test. (f) Immunoblot detection of transgenic Ras V12 (with an anti-FLAG antibody), phosphorylated (p-) and total Mek and/or Erk, RpS21, and actin as a loading control from lysates derived from S2 cells stably transduced with expression vectors expressing the indicated Ras V12 transgenes in the absence (-) and presence (+) of RpS21 RNAi. (g) Immunoblot detection of indicated proteins derived from lysates of the adult ovaries of either wild-type (+/+) or mutant (RpS21 03575 /+) flies. (h) The mean ± SEM eye severity score of the genotypes from three replicate experiments at 18˚C. Tukey's multiple comparisons test was used for statistical comparisons in a and e. ���� p<0.0001. �� p<0.01. n.s., not significant. (i) Model depicting how a signal intensity modifier such as RpS21 may be ineffective above a specific signaling intensity threshold (dotted line).
https://doi.org/10.1371/journal.pgen.1009228.g004 acts as an enhancer of Ras V12 Common (Fig 4H). Interestingly, RpS21 03575 /+ no longer enhances Ras V12 Rare, underscoring the sensitivity of RpS21/+ to Ras/MAPK signaling strength. Therefore, RpS21 regulation of the Ras pathway appears to be signal-strength dependent, rather than codon-dependent. Collectively, these results demonstrate that while RpS21 negatively regulates Ras-MAPK signaling in diverse contexts, at the phenotypic level this regulation preferentially impacts weak Ras/MAPK signaling. These findings are consistent with a model whereby above a certain signaling intensity threshold, regulators that impact Ras signaling at weaker intensity levels are no longer effective (Fig 4I).

RpS21 downregulation does not alter expression of a codon-altered GFP reporter
Our above results suggested that it is Ras/MAPK signaling strength, and not codon manipulation specifically, that determine whether RpS21 heterozygosity impacts protein expression. To test this idea further, we generated an additional pair of transgenes with identical protein sequence but distinct codon usage. Specifically, we generated two GFP transgenes-one with GFP containing 100% common codons, and one where the same GFP had 50% synonymous substitutions of rare codons dispersed throughout the protein. Both transgenes were expressed under a ubiquitin promoter and were integrated into the same site in the genome (Figs 5A and S5, see Methods). Consistent with our results for altering codon content of the Ras gene, GFPCommon protein is expressed at a higher level in adult animals than GFPRare protein (Fig 5B and 5C). Given this, we next tested whether RpS21 downregulation alters GFP protein expression in a codon-dependent manner. RpS21 03575 /+ animals exhibit similar GFPRare protein expression as wild type animals (Fig 5D and 5E). Additionally, RpS21 03575 /+ animals exhibit similar GFPCommon protein expression as wild type animals (Fig 5F and 5G). These results indicate that RpS21 downregulation does not impact translation of at least one other tested transgene pair, suggesting that RpS21 may, to some degree, act specifically to regulate the Ras/MAPK pathway at specific signaling intensity levels. Overall, our findings highlight the ability of our approach to reveal new Ras/MAPK regulators that preferentially impact specific signaling outputs.

Discussion
Here, we revisit a well-proven strategy to identify Ras/MAPK modifiers (a heterozygous mutant screen in the Drosophila eye) but do so with the new angle of altering codon usage in a core signaling component to find signal strength-dependent regulators. We show here that changing codon usage in a signaling pathway component can be an effective strategy to find signal strength-dependent modifiers, as evidenced by our identification of 15 Df from a wholegenome screen that only modify the rough-eye phenotype driven by either a common or rare codon-enriched Ras V12 transgene, but not both. From these efforts, we identify the RpS21 gene as a negative regulator of a weak or low-level Ras phenotype in the in vivo context of eye development. These findings are further supported by our finding that RpS21 reduction in other contexts also impacts (low) endogenous Ras signaling in the ovary, but not higher Ras signaling in S2 cells.
Our results show that altering codon usage can serve as a valuable platform to stably alter protein production to undertake signal strength-specific screens. Clearly, there are other ways that one can modulate signal output strength, such as modulating gene expression strength as we also do here, or through use of an allelic series [68]. However, an advantage of altered codon usage is that it can be hard-wired into the genome, and thus no additional (and potentially confounding) experimental parameters such as altering temperature, inducing genes with drugs, and so forth are required. Our approach should be applicable to any signal transduction pathway. The utility of our approach is underscored in the fact that signal strengthspecific modifiers found in our screen appear to be enriched for genome regions not previously linked to Ras genetic modification. The causative genes contained within 14 of these differential Df hits remain to be mapped and represent a potentially rich source of new genes modulating Ras/MAPK signaling. Previous work found that different levels of MAPK activity impact different biological processes [68]. Intriguingly, our differential hits appear to be enriched in bHLH transcription factors. Of note, the bHLH transcription factor Myc is a wellknown Erk target [69][70][71][72], and it will be interesting to explore whether specific bHLH transcription factors are preferentially targeted by this pathway in signal strength-dependent contexts.
Given the importance of Ras/MAPK signaling in many settings across evolution, our identified modifiers may shed insight into how this pathway is controlled at different signal strengths. While our focus here is on Drosophila eye development, signal strength dependencies of the Ras/MAPK pathway are appreciated to play a role in human disease. Activating mutations in the MAPK pathway of humans underlie a class of human diseases termed RASopathies [73]. Further, relevant to our approach here, of the three human RAS genes, KRAS, is the most enriched in rare codons [27] and is the most commonly mutated RAS isoform in human cancers [28,74]. Changing the rare codons to more common codons in a single exon of the mouse KRAS gene leads to fewer tumors following carcinogen exposure [22], which is in line with current thinking on a "sweet spot" level of Ras/MAPK signaling required to initiate tumorigenesis [28]. We argue that the larger clone size that we observe in leg imaginal discs of animals expressing Ras V12 Rare vs. Ras V12 Common reflects this same concept. As such, the new tools we report here may provide valuable reagents to more accurately model KRAS-relevant regulation in Drosophila and ultimately in KRAS-driven disease.
Our approach found that RpS21 functions as a negative regulator of weak Ras/MAPK signaling. While one might expect that a codon-based approach would pull out ribosomes as hits, we show here that codon-independent manipulation of Ras signaling, through temperature change, confirms that RpS21 is responding to specific signaling levels rather than specific codons. As Ras/MAPK signaling is known to drive tissue growth in diverse settings, this may suggest that RpS21 can function as a negative regulator of tissue or tumor growth. Interestingly, downregulation of RpS21 was previously shown to cause excessive hyperplasia in hematopoietic organs and imaginal disc overgrowth during larval development, suggesting RpS21 acts as tumor suppressor in Drosophila [65]. Although this finding may seem paradoxical given that ribosomal mutants in flies are well-known to cause minute phenotypes, characterized by short bristles, small body size, and delayed growth [75][76][77][78], a subset of ribosomal proteins including RpS21 have been identified to have a growth suppressive role [65,[79][80][81][82][83]. Further, heterozygosity of many ribosomal proteins is reported to be tumorigenic in zebrafish [84], and heterozygous inactivating mutations of ribosomal proteins have been described in human cancers [85,86]. Several mechanisms have been proposed to account for this apparent tumor suppressor activity of ribosome protein downregulation, including activation of p53 [87][88][89], inhibition of NF-KB [90], E2F [91], MYC [92], and CDK8 [93]. Thus, RpS21 joins the ranks of an emerging number of ribosomal proteins with roles in growth suppression, although whether RpS21 acts as a tumor suppressor in mammals awaits investigation.
The mechanism underlying the negative regulation of Ras/MAPK signaling by RpS21 remains to be determined. Future work can explore how direct the regulation is, and whether RpS21 acts in a cell autonomous or, has been shown for a subset ribosomal subunits, a nonautonomous manner to regulate tissue growth [83]. Future work can also explore whether other signaling pathways connected to eye development are also impacted by RpS21 reduction.
In our work, we found that RpS21 downregulation promotes elevated levels of Ras V12 protein in multiple settings. The effect of RpS21 on Ras V12 protein level could potentially be through RpS21's canonical ribosomal function or through an extra-ribosomal function. Dose-dependent ribosome dysfunction is linked to the human disease Diamond-Blackfan anemia, where heterozygous mutations in specific ribosomal subunits are linked, at least in part, to compromised ribosome biogenesis and translation [94][95][96][97]. A defect in RpS21 ribosomal function may trigger ribosomal biogenesis defects that alter translational fidelity or promote generation of oncoribosomes to preferentially express subset of mRNA pools [98,99]. Alternatively, RpS21 might participate in other cellular processes independent of its canonical ribosomal function, as has been shown for other ribosomal subunits [100][101][102][103].
We note that RpS21 has been connected to positive regulation of Ras/MAPK in other contexts. While this manuscript was in review, a recent study revealed that downregulation of human RPS21 inhibits metastatic behavior of osteosarcoma cells in a MAPK-dependent manner [104], underscoring the potential human relevance of our findings here. Further, in contrast to our screen results revealing negative regulation by RpS21 in multiple contexts, numerous ribosomal proteins (RpS21 included) were found among 1,162 genes to positively regulate Erk phosphorylation in a previous primary screen in cultured Drosophila S2R+ cells [15]. Unlike this Erk activation screen, we note that our Ras V12 eye modifier screen hits were not preferentially enriched for ribosomal subunits, and that ribosomes in general are not enriched among known FlyBase Ras genetic enhancers/suppressors. We hypothesize that the addition of insulin to the growth media, required for Erk activation in the context of the S2R + cell screen, revealed a dependency for cell growth, which is dependent on both ribosomes and Erk activation. S2R+ cells have known differences from S2 cells in response to external signaling, and this could reflect differences in MAPK regulation in this context as well [105], underscoring the need to understand signaling dynamics and regulation in a given biological context.
Another question for future investigation is why RpS21 regulation of Ras signaling is nonfunctional in contexts of heightened Ras/MAPK signaling, as we observed in S2 cells with strong Ras/MAPK biochemical output, as well as at the phenotypic output level where Rps21/+ failed to noticeably modify the eye phenotype of Ras V12 Common. One possible explanation is that different MAPK signaling strengths activate a different host of MAPK targets, and this impacts the degree of negative regulation by RpS21. To that end, it will be important to further mine our screen to identify single gene modifiers in the other 14 Dfs, which may similarly yield new regulatory insight into the Ras/MAPK pathway.
In summary, we show here the value of manipulating codon usage of one component of a pathway to modulate the corresponding signaling output, and the use thereof to screen for modifiers of specific signaling intensities. This approach proved successful, identifying a novel regulator of the Ras/MAPK pathway, RpS21. As such, this approach may find value in similarly interrogating other signaling pathways.

Generation of codon-altered genes in Drosophila
Codon-altered exon sequences for Ras V12 Common, Ras V12 Rare, GFPRare, and GFPCommon were created using the Kazusa codon usage database (https://www.kazusa.or.jp/codon/) and subsequently generated by Gene Synthesis (ThermoFisher Scientific, Invitrogen GeneArt). A cDNA clone (LD17536, Drosophila Genomics Resource Center) was used as a template to generate the non-altered Ras85D sequence. To generate Ras V12 Native, the QuikChange II Site-Directed Mutagenesis Kit (Agilent) was used to change codon 12 in Ras85D from GGA (glycine) to GTA (valine). Subsequently, primers (sequences available upon request) were designed to amplify Ras sequences and the Invitrogen Gateway BP Clonase II Enzyme Mix (ThermoFisher Scientific) was used to insert these sequences into the Gateway entry vector pDONOR221 (ThermoFisher Scientific). Subsequently, the Invitrogen LR Clonase Enzyme Mix (ThermoFisher Scientific) was used to insert the Ras WT , and Ras V12 Native, Common, and Rare sequences into the Gateway destination vector pBID-UASC-FG (Addgene Plasmid #3520 [106]), which has a N-terminal FLAG tag and a PhiC31 site for site-directed genomic insertion. pBID-UASC-FG-Ras plasmids were prepared with a ZymoPURE II Plasmid Midiprep Kit (Zymo Research) and sent to Model System Injections (Durham, NC, USA) for injection into attP40 (2L) flies. GFP sequences were cloned into a pBID plasmid (modified from Addgene Plasmid #3520), and DNA and transgenic flies were prepared as for Ras transgenes. For cell culture, Ras V12 Common and Ras V12 Rare transgenes were cloned into pMKInt-Hyg vectors, which were sequenced to confirm the correct sequence.
Fly stocks. All flies were raised at 25˚C on standard media unless noted otherwise (Archon Scientific, Durham NC). FlyBase (http://FlyBase.org) describes full genotypes for all stocks used in this study. See S1 Table for  Fly genetics and deficiency screen. To examine mitotic clones in leg imaginal discs and associated animal survival in such experiments, flies containing UAS Ras transgenes were crossed to hsflp;; UAS-GFP, tubulin-FRT-STOP-FRT-Gal4 animals. F1 larvae were collected 96 hours after egg laying and heat shocked at 37 degrees for 20 minutes. After 24 hours, leg imaginal discs were dissected from living larvae. Discs were fixed as done previously for imaginal discs [107] and probed with DAPI for DNA. Images were taken on a Nikon A1 confocal microscope. Clone sizes were determined using FIJI's Tracing and measuring tools.
To examine eye phenotypes and associated animal survival in such experiments, the Ras transgenes were combined with a sev Gal4 driver and subsequently crossed to Df/Balancer flies. After 16-18 days after egg laying, the rough eye phenotype of the resulting progeny was scored (both males and females). The scoring system was as follows (category = numerical score, qualitative description): Mild = 1, no discoloration or necrotic tissue; Moderate = 3, discoloration and no necrotic tissue; Severe = 5, discoloration and necrotic tissue (see Fig 1B and  1C). Severity scores for each genotype was calculated as follows: (#Mildx1+#Moderatex3 +#Severex5)/Total # of flies. To determine if heterozygosity for a subset of genes altered the rough eye phenotype the following two genotypes for each deficiency (Df) were compared: Ras transgene only and Ras transgene + Df (used as an internal comparison to control for background effects). Then, we calculated a fold change score for both Ras V12 Common and Ras V12-Rare for each deficiency: Ras transgene + deficiency/Ras transgene. We note that none of the Df animals on their own had detectable eye phenotypes. For the primary screen, the fold change score was defined as follows: enhancer (fold change �1.35 or 5X less flies eclosed); suppressor (fold change �0.65 or 5X more flies eclosed). For the secondary screen, the fold change score was defined as follows: enhancer (fold change �1.95 or 5X more flies eclosed); suppressor (fold change �0.50 or 5X less flies eclosed). The final phenotype for a deficiency was defined as follows: not a modifier (neither Ras V12 Common or Ras V12 Rare + Df were modified); enhancer (both Ras V12 Common and Ras V12 Rare + Df were enhanced); suppressor (both Ras V12 Common and Ras V12 Rare + Df were enhanced); differential (only Ras V12 Common or Ras V12 Rare + Df were modified). We note that overall eye size was relatively unaffected by different Ras transgenes. Images of fly eyes were obtained using a Leica MZ10F microscope with a PlanApo 1.6X objective, Pixel Shift Camera DMC6200, and LASX software.
RT-PCR. Animals were aged 3-7 days at 25˚C on standard fly medium. RNA was extracted from adult fly heads using TRIzol™ reagent (ThermoFisher, cat#15596026) according to the manufacturer's protocol (10-40 heads per sample in 500ul TRIzol™ reagent). Purified RNA was resuspended in molecular grade water. RNA was DNase treated with DNase I at room temp for 15 minutes, then the reaction was terminated by adding 25mM EDTA and incubating at 65˚C for 10 minutes. DNase efficiency was confirmed using a positive control. DNase treated RNA was reverse transcribed into cDNA using iScript cDNA synthesis kit (BIO-RAD, cat#170-8891) according to the manufacturer's protocol. Subsequent cDNA was treated with RNase H prior to use in qPCR reactions. The concentration of the RNA was quantified on a NanoDrop spectrophotometer and samples were diluted with molecular grade water to match the concentration of the lowest concentration sample. Luna Universal qPCR Master Mix (NEB #M3003) was used to run the qPCR reaction according to the manufacturer's specifications. Primers for the detection of Ras constructs were designed against an identical region containing the 3xFLAG sequence shared by both Ras V12 Common and Ras V12 Rare transcripts. Primers were designed against Drosophila Beta Tubulin 56D as a reference gene. Ras qPCR FW primer: TGGACTACAAAGACCATGACGGT, Ras qPCR RV primer: ACTTGTATACCGGTGCTTGTCAT, Tubulin qPCR FW primer: GGACGAGACCTACTG CATCG, Tubulin qPCR RV primer: GGTCACCGTATGTGGGTGTC.
Cell culture. KC and S2 cell lines were obtained from Bloomington (Indiana University DGRC Bloomington) and as a gift from Dr. David MacAlpine (Duke University) respectively. These cells were cultured in Schneider's Drosophila medium (Invitrogen) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin-L Glutamine (Invitrogen) at 25˚C. FBS was heated for 60 minutes in 58˚C and then cooled down before being added to the medium. These cells were confirmed to be free of mycoplasma infection, as measured by the Duke Cell Culture Facility using MycoAlert PLUS test (Lonza). S2 and KC cell lines were stably transduced with the pMKInt-Hyg vector encoding Ras V12 Common and Ras V12 Rare cDNAs using 1000 ng of DNA in 6 well plates per manufacturer instructions (Effectene transfection reagent, Qiagen). The following day, Schneider's media was changed, and cells were seeded in a coated culture dish (100x20 mm). Four days later, cells were passaged with fresh Schneider's medium and 200 μg/ ml hygromycin (Invitrogen) was added. The stably transfected cells were selected within a month growing in media containing hygromycin. Three days prior to any experiment, these cells were grown in media without hygromycin. Four million S2 cells that were stably transduced with Ras V12 Common and Ras V12 Rare plasmids were seeded into coated tissue culture dishes (60x15mm, VWR) with 2 ml of Schneider's media (without FBS). Sixty micrograms of RpS21 dsRNA were added on top of these cells. One hour later, two ml Schneider's media containing 20% FBS were added on top of 2 ml Schneider's media without FBS resulting in medium with 10% FBS concentration in total media of this culture. Within 16-24 hours after RNAi treatment, expression of Ras V12 Common and Ras V12 Rare transgenes were induced by CuSO4 for another 12 hours. Finally, these cells were collected 30-36 hours after dsRNA treatment.
dsRNA synthesis. S2 cell DNA was used to produce a PCR template for RpS21 dsRNA production using the forward primer "TAATACGACTCACTATAGGGTTACTGACCAGCC GATACCC" and reverse primer "TAATACGACTCACTATAGGGCCACGCTTAGAAGTTC CTGC". Next, 500 ng of RpS21 PCR template was used for an in vitro production of dsRNA as instructed in the MEGAscrip T7 transcription kit (ThermoFisher). The dsRNA solution was cleared using MegaClear kit (ThermoFisher). Finally, the concentration of RpS21 dsRNA was measured and stored in -80˚C for future use.
Gene enrichment analysis and statistical analyses. To determine the Codon Adaptation Index (CAI), sequences were entered at the CAIcal web-server (http://genomes.urv.es/CAIcal [108]. For gene enrichment, deficiency sequence boundaries were defined using coordinates available through FlyBase [109] and the Bloomington Drosophila Stock Center website. Deficiencies were then uploaded as a custom BED track to the UCSC Genome Browser (Reference Assembly ID: dm6). Genes overlapping the deficiency coordinates were then extracted using BEDtools for additional analysis [110]. A deficiency was determined to contain known Ras modifiers if any of the deficiency covered genes known as Drosophila Ras85D genetic interactors (332 interactors, FlyBase). Enhancers and suppressor deficiencies were analyzed using the same metric against known Ras85D interactors of the same respective modifier type. Statistical analysis (chi-square) was performed using Graphpad Prism v8.1. FlyBase Gene Group Enrichment analysis was performed by comparing deficiency covered genes with pre-defined FlyBase Gene Groups. Analysis and statistical tests were performed in R using Gene Overlap package (https://rdrr.io/bioc/GeneOverlap/) and results are reported as adjusted p-values (False Discovery Rate [111], using Benjamini Hochberg correction). Graphs and statistical analyses were generated using GraphPad Prism 7. Statistical tests and adjusted P-values are detailed in figure legends. For all tests, adjusted P-value reporting is as follows: (P>0.05, n.s.; P<0.05, � ; P<0.01, �� ; P<0.001, ��� , P<0.0001, ���� ). were immunoblotted with an anti-FLAG antibody, demonstrating differential expression of Ras V12 common and rare. Bottom: quantification and protein loaded. (d) Immunoblot detection of transgenic Ras V12 (with an anti-FLAG antibody), phosphorylated (p-) and total Mek and Erk, and actin as a loading control from lysates derived from (e) the head of flies with the indicated versions of transgenic Ras V12 or (f) S2 and KC cells stably transduced with expression vectors expressing the indicated Ras V12 transgenes. First lane is S2 cells without any transfection. (g) Levels of GTP-bound Ras V12 common versus rare. GTP-bound Ras from lysates derived from S2 cells stably expressing Ras V12 common versus rare (or no transgene as a control) were affinity captured with a Ras Binding Domain (RBD IP) and immunoblotted with an anti-FLAG antibody to detect the ectopic active portion of the expressed Ras V12 protein.
Whole cell lysates (WCL) were immunoblotted with an anti-FLAG antibody to detect total ectopic Ras V12 protein and Actin as a loading control. One representative blot from multiple replicates is shown.