Functions of the Nonsense-Mediated mRNA Decay Pathway in Drosophila Development

Nonsense-mediated mRNA decay (NMD) is a cellular surveillance mechanism that degrades transcripts containing premature translation termination codons, and it also influences expression of certain wild-type transcripts. Although the biochemical mechanisms of NMD have been studied intensively, its developmental functions and importance are less clear. Here, we describe the isolation and characterization of Drosophila “photoshop” mutations, which increase expression of green fluorescent protein and other transgenes. Mapping and molecular analyses show that photoshop mutations are loss-of-function mutations in the Drosophila homologs of NMD genes Upf1, Upf2, and Smg1. We find that Upf1 and Upf2 are broadly active during development, and they are required for NMD as well as for proper expression of dozens of wild-type genes during development and for larval viability. Genetic mosaic analysis shows that Upf1 and Upf2 are required for growth and/or survival of imaginal cell clones, but this defect can be overcome if surrounding wild-type cells are eliminated. By contrast, we find that the PI3K-related kinase Smg1 potentiates but is not required for NMD or for viability, implying that the Upf1 phosphorylation cycle that is required for mammalian and Caenorhabditis elegans NMD has a more limited role during Drosophila development. Finally, we show that the SV40 3′ UTR, present in many Drosophila transgenes, targets the transgenes for regulation by the NMD pathway. The results establish that the Drosophila NMD pathway is broadly active and essential for development, and one critical function of the pathway is to endow proliferating imaginal cells with a competitive growth advantage that prevents them from being overtaken by other proliferating cells.


Introduction
Nonsense-mediated mRNA decay (NMD) is a cellular surveillance pathway in eukaryotes that recognizes and degrades transcripts with premature termination codons (PTCs). Such transcripts arise as a consequence of genomic mutation, as in numerous human genetic diseases [1,2], and from errors in transcription and aberrant RNA splicing. Destruction of PTC-containing transcripts by NMD prevents production of truncated, potentially harmful proteins that can interfere with normal cellular processes (e.g., [3]). The NMD pathway has also been found to influence expression of a variety of wild-type transcripts (reviewed in [4]), implying that the pathway has regulatory roles beyond its surveillance function. In this paper, we describe Drosophila mutants that affect NMD.
NMD pathway genes were discovered by genetic studies in yeast (up-frameshift suppressor [Upf] genes; [5]) and Caenorhabditis elegans (suppressor with morphogenetic effect on genitalia [smg] genes; [6]), and their functions and mechanisms of action have been characterized by molecular genetic and biochemical analysis of the proteins and target RNAs in yeast [7] and cultured mammalian and Drosophila cells [8][9][10]. There are three conserved core components of the pathway, Upf1 (smg-2), Upf2 (smg-3), and Upf3 (smg-4) (reviewed in [11]). Upf1 is an RNA helicase that associates with the translation termination complex at PTCs and, at least in yeast, targets the RNA to cytoplasmic RNA processing centers called P bodies [12]. Upf1 is proposed to recruit Upf2 and Upf3 to these termination complexes, which leads to activation of decapping enzymes and nucleases that degrade the target RNA.
Additionally, in metazoans, Upf1 undergoes a phosphor-ylation cycle (reviewed in [13]). Upf1 is phosphorylated on serine residues by Smg1, a PI3K-related kinase. The phosphates are subsequently removed by complex(es) containing Smg5, Smg6, and/or Smg7, three similar proteins that are thought to recruit the phosphatase PPA2. The Upf1 phosphorylation cycle is apparently necessary for Upf1 and NMD activity at least in some organisms, because NMD function is abrogated when Smg1, Smg5, Smg6, or Smg7 activity is reduced [6,9,10,14]. One intriguing mechanistic question is how the NMD machinery distinguishes a PTC from a normal termination codon. In mammals, an important feature appears to be the relationship between the termination codon and splice junctions in the mRNA [15]. Most normal termination codons are located beyond the last splice junction, in the final exon of the mRNA. Termination codons that lie upstream of an exon-exon boundary are generally recognized as premature and target the mRNA for destruction by NMD. Such boundaries are marked after splicing by deposition of a multiprotein complex, the exon junction complex (EJC), which includes Upf2 and Upf3. One current model proposes that EJCs along an mRNA are normally all displaced by the translocating ribosome, but if the ribosome encounters a termination codon before the last EJC, this EJC remains and promotes delivery of Upf2 and Upf3 to Upf1 at the termination complex to activate NMD [8,12]. However, this cannot be the sole mechanism of PTC recognition because there are mammalian transcripts with stop codons in the last exon that are subject to NMD, as well as transcripts with stop codons upstream of introns that are resistant to NMD (reviewed in [15]). Furthermore, PTC recognition in yeast and cultured Drosophila cells can occur in the absence of introns and splice junctions [7,10]. In yeast it appears that the distance between the stop codon and a special site in or near the 39 UTR, or the ability of proteins bound at these sites to efficiently associate, marks a termination codon as premature and targets the mRNA for destruction by NMD [16], and this also appears important for NMD targeting of certain mammalian transcripts [17].
Although the mechanism of the NMD pathway has been studied extensively, its developmental functions have received less attention. Yeast and C. elegans NMD pathway mutants are viable as homozygotes and have only subtle or no effects on development and differentiation. The most conspicuous defects in C. elegans mutants are the swollen bursa in the tail of adult males and swollen vulva of hermaphrodites, both of which apparently result from an effect on morphogenesis rather than cell lineage [6]. By contrast, mouse UPF1 À/À mutants are not viable [18], and RNA interference knockdown of Upf1, Smg1, or other NMD pathway genes in cultured Drosophila S2 or mammalian cells causes a cell cycle arrest, implicating the NMD pathway in cell cycle progression [19,20]. However, two recently described mutations in the Drosophila homolog of Smg1, the only extant mutations in Drosophila NMD genes, are homozygous viable and do not appear to affect NMD, raising questions about the function of Smg1 and the NMD pathway in Drosophila [21].
Here, we describe the isolation and characterization of Drosophila mutants that enhance expression of green fluorescent protein (GFP) and other transgenes. We demonstrate that these are loss-of-function mutations in three NMD pathway genes, Upf1, Upf2, and Smg1. We show that Upf1 and Upf2 are required for NMD pathway activity, whereas Smg1 has a variable, gene-selective role potentiating the pathway. We then use the Upf1 and Upf2 mutations to characterize the functions and endogenous RNA targets of the pathway in Drosophila development. We find that the NMD pathway is broadly active during development and required for proper expression of dozens of endogenous genes and for larval viability, and that one critical function of the pathway is to enhance the ability of cells to compete with other cells during proliferative growth.

Identification of Mutations That Increase Transgene Expression in Drosophila
We conducted a genetic mosaic screen of ethane methyl sulfonate-induced mutations on the X chromosome for tracheal (respiratory) system mutants. Details of the screen will be described elsewhere. In the screen, we used the S. cerevisiase FLP1 recombinase (FLP)/FLP1 recombinase target (FRT) system [22] to generate homozygous mutant cell clones in otherwise heterozygous animals. Tracheal clones were identified by labeling them with GFP using the MARCM system [23] in which a btl-GAL4 transgene drives tracheal expression of a UAS-GFP transgene but is kept off in heterozygous cells by a btl-GAL80 transgene present on the wild-type X chromosome ( Figure 1A). In a screen of 749 ethane methyl sulfonate-induced X-linked lethal lines, we identified six mutations (13D, 14J, 25G, 26A, 29AA, and 32AP) that caused markedly increased GFP signal in homozygous mutant tracheal clones compared to control wild-type clones examined as third instar (L3) larvae ( Figure 1B). We named this the ''photoshop'' phenotype because it increased visualization of clones like that achieved by digital enhancement with Photoshop software (Adobe, http://www.adobe. com). The photoshop phenotype was not dependent on the btl-GAL80 repressor in the MARCM system: homozygous photoshop tracheal cell clones in btl-GAL4, UAS-GFP larvae showed a similar enhancement of GFP signal, and viable hemizygous 25G and 32AP larvae and adults (see below) carrying the same transgenes showed increased GFP signal throughout the tracheal system ( Figure 1C and 1D and data not shown). The photoshop phenotype was not specific to GFP, because mutant clones showed similar enhancement of DsRed1 signal ( Figure 1E). All tracheal cells examined showed the photoshop phenotype, and all six mutations gave similar GFP enhancement.
All but two of the mutations (14J, 29AA, 13D, and 26A) lead to hemizygous male lethality before L3. One exception was 25G: 25G/Y hemizygous males and 25G homozygous females developed to L3 larvae at approximately normal frequencies

Synopsis
Cells possess a variety of surveillance mechanisms that detect and dispose of defective gene products. One such system is the nonsense-mediated mRNA decay (NMD) pathway, which degrades aberrant mRNAs containing nonsense mutations or other premature translation stop signals. In a genetic screen in Drosophila, the authors identified a set of mutations they call ''photoshop'' mutations because they increase expression of green fluorescent protein transgenes such that cells expressing green fluorescent protein are more easily visualized. They found that the photoshop mutations are mutations in three different genes implicated in NMD. Using these mutations, they show that the NMD pathway not only degrades mutant mRNAs but also influences expression of many transgenes and dozens of endogenous genes during development and is essential for development beyond the larval stage. One important function of the pathway is to provide proliferating cells with a competitive growth advantage that prevents them from being overtaken by other proliferating cells during development. Thus, the Drosophila NMD pathway has critical cellular and developmental roles beyond the classical surveillance function of eliminating mutant transcripts.
and produced a few percent of escaper adults. These adults appeared morphologically normal but had greatly reduced fertility. The other exception was 32AP, which after removal of extraneous linked lethal loci (see Materials and Methods) was found to be hemizygous-male and homozygous-female viable and fertile. We generated a deficiency uncovering 32AP and found that 32AP/Df showed the same GFP enhancement as 32AP/Y ( Figure 1D), implying that 32AP is an amorphic (null) allele.
To study the effect of the mutations on GFP signal in other tissues and other stages of development, we used the viable alleles 25G and 32AP along with da-GAL4 and UAS-CD8:GFP transgenes, which give ubiquitous expression of GFP [24].
Compared to control animals, 25G/Y male and homozygous 25G female larvae and adults had greatly increased GFP in all tissues examined, including epidermis, salivary glands, fat body, and eyes ( Figure 1F and data not shown). 32AP mutants also showed enhanced GFP signal in all tissues examined, although enhancement appeared slightly weaker than in 25G mutants. 32AP also enhanced levels of a nuclear localized DsRed2 construct expressed in the larval epidermis ( Figure  1G). Thus, photoshop mutations increase transgene signal in many, if not all, larval and adult tissues.
To test whether the increased signal seen in the photoshop mutants was due to increased expression of the transgene, we performed quantitative RT-PCR experiments on RNA (C) GFP signal is increased throughout tracheal system in photoshop mutant 25G larvae. A wild-type (y w FRT 19A /Y; btl-GAL4, UAS-GFP) control larva is shown above, and a hemizygous 25G/Y photoshop mutant (25G/Y; btl-GAL4, UAS-GFP) larva is shown below. Bar, 1 mm. (D) GFP signal is increased throughout tracheal system in photoshop 32AP larvae. Above is a control heterozygous (32AP/þ; btl-GAL4, UAS-GFP/þ) larva showing weak GFP signal throughout tracheal system. In the middle is a 32AP/Df mutant female (32AP/Df; btl-GAL4, UAS-GFP/þ) larva showing enhanced GFP signal throughout. Below is a 32AP/Y mutant male (32AP/Y; btl-GAL4, UAS-GFP/þ) larva: GFP signal is indistinguishable from that observed in a 32AP/ Df larva, indicating 32AP is an amorphic (null) allele. Bar, 1 mm. (E) Effect of a photoshop mutation on a DsRed transgene. Photoshop 14J tracheal clone in y w 14J FRT 19A / w FRT 19A , FLP122 larva carrying btl-GAL4 and UAS-DsRed1 transgenes. Two unicellular tracheal tubes are indicated by brackets; 14J À clone has higher DsRed1 signal than neighboring wild-type (14J þ ) cell. Bar, 25 lm. (F) Effect of photoshop 25G mutation outside the tracheal system. A mixture of wild-type and 25G/Y mutant animals carrying da-GAL4, UAS-CD8:GFP is shown. GFP signal is increased throughout all tissues of the 25G mutants. 25G genotype of high-signal larvae was confirmed by PCR. Bar, 2 mm. (G) Effect of photoshop 32AP mutation in epidermis. A wild-type larva is shown above, and below is a 32AP/Y mutant larva (bottom) carrying e22c-GAL4, UAS-nls:DsRed2, which expresses a nuclear localized DsRed2 in epidermis and other epithelial tissues. Animals were photographed next to each other. The nls:DsRed2 signal is enhanced in the epidermis of the 32AP/Y mutant. Bar, 1 mm. (H) Effect of photoshop mutants on GFP RNA levels. Quantitative real time RT-PCR was done on the GFP transcript in wild-type (y w FRT 19A /Y), 25G/Y, and 32AP/Y L3 larvae carrying one copy of da-GAL4 and UAS-CD8:GFP. Results of replicate reactions were individually normalized to parallel reactions on rp18LA transcript. Values shown are mean 6 2 standard errors of the mean normalized to value in wild type. doi:10.1371/journal.pgen.0020180.g001 derived from hemizygous 25G, hemizygous 32AP, or control wild-type male L3 larvae, each also containing da-GAL4 and UAS-CD8:GFP. GFP RNA levels were increased ;5-fold in 25G mutants and ;2.5-fold in 32AP mutants compared to the wild type ( Figure 1H). Hence, the increased GFP signal in photoshop mutants is due at least in part to increased GFP RNA levels.

Photoshop Mutations Are Loss-of-Function Alleles of Three NMD Pathway Genes
We used meiotic recombination to map the lethality and the mosaic tracheal GFP enhancement phenotype of the photoshop mutations. 14J and 29AA mapped between ct and v ( Figure 2A). Complementation analysis (see below) showed that 14J and 29AA and the semi-lethal allele 25G formed a single complementation group. 13D and 26A mapped between v and g ( Figure 2D), and complemented 14J for lethality. 32AP was mapped by its enhancement of GFP expression in hemizygous males, and localized between cv and ct ( Figure 2F). Thus, the six photoshop mutations define at least three loci.
We further refined the position of mutation 14J by single nucleotide polymorphism (SNP) mapping and localized it to an ;200-kb interval (Figure 2A). DNA sequencing of predicted genes in this interval identified changes in the Drosophila homolog of the NMD pathway gene Upf2 in all three alleles of the 14J complementation group. The 14J mutation is a 14-bp deletion approximately halfway through the 1,241-residue coding sequence of Upf2 ( Figure 2B), which causes a frame shift at codon 562 followed by a stop 40 codons later. 29AA is a nonsense mutation at codon 324. Both 14J and 29AA truncate the Upf2 coding sequence and are likely to be null alleles. A transgene comprising the Upf2 gene ( Figure 2B) rescued 14J hemizygous males and homozygous females to viability and fertility; 29AA could not be tested because of a linked lethal mutation. We conclude that 14J and 29AA are loss-of-function mutations in Upf2.
25G hemizygous males and homozygous females survived to L3 and beyond (see above), whereas 25G/14J trans hetero- zygotes were not viable after L2, suggesting that 25G is a hypomorphic allele of Upf2. The Upf2 þ transgene rescued 25G hemizygous males and homozygous females to viability and fertility, confirming this assignment. The 25G allele is a curious mutation that alters the natural Upf2 stop codon to an arginine codon (TGA to AGA). The next in-frame termination codon is 45 bp downstream, so 25G encodes a Upf2 protein with a 15-residue C-terminal extension ( Figure  2C). The C-terminus of yeast Upf2p is required for its interaction with Upf1p [25], so the Upf2 extension might interfere with this function. The 25G mutation also creates a consensus 39 splice site, but RT-PCR experiments on 25G mutant RNA did not detect any novel splice forms involving this site.
The map positions of 13D and 26A were refined using SNP markers, and the lethality associated with each allele localized to an ;400-kb interval that contained the Drosophila homolog of Upf1 ( Figure 2D). Sequencing of the Upf1 gene in 26A revealed a nonsense mutation (CAG to TAG; Q637stop) in the middle of the 1,180-residue coding sequence ( Figure 2E), and sequencing of 13D identified a missense mutation (TGC to TAC) that alters a conserved cysteine (C186Y) in a domain required for interaction between yeast Upf1p and Upf2p [25] ( Figure 2E). We also found that 26A and 13D failed to complement each other for lethality. Thus, 26A and 13D appear to be loss-of-function alleles of Upf1, and we refer to them as Upf1 26A and Upf1 13D .
32AP was mapped to a 2-Mb interval containing 122 predicted genes, one of which is the Drosophila homolog of Smg1 ( Figure 2F). We generated an ;46-kb deficiency that removes ten genes including Smg1, and found that this deficiency failed to complement the photoshop phenotype of 32AP ( Figure 1D). DNA sequencing identified a nonsense mutation (CAG to TAG) in codon 1651 of the 3,218-residue Smg1 coding sequence, which truncates the Smg1 protein before most of the conserved domains, including the kinase domain ( Figure 2G). Thus, 32AP is most likely a null Smg1 allele, consistent with the gene dosage experiments described above ( Figure 1D). We refer to it as Smg1 32AP .

Photoshop Mutations Abolish or Reduce NMD of a Mutant Transcript In Vivo
Upf1, Upf2, and Smg1 are required for NMD in yeast and C. elegans and have been shown to be involved in NMD in cultured Drosophila cells [10], although recent data question the in vivo role of Smg1 in Drosophila [21]. To determine whether photoshop genes are required for NMD in vivo, we tested the effects of photoshop mutations on mRNA levels of Adh n4 , a nonsense mutation in Adh [26] that subjects the mRNA to NMD in S2 cells [10]. We isolated RNA from adult wild-type males (y w FRT 19A /Y; Adh n4 /Adh þ ) and from Upf2 (Upf2 25G /Y; Adh n4 /Adh þ ) and Smg1 (Smg1 32AP /Y; Adh n4 /Adh þ ) mutant males, amplified Adh mRNA by RT-PCR, and quantitated the Adh þ and Adh n4 products to assess the relative levels of the two transcripts ( Figure 3). In wild-type animals, steady state transcript levels from the Adh n4 allele were reduced 13-fold relative to Adh þ transcripts, presumably because of increased turnover of the mutant transcript by the NMD pathway. In Upf2 25G animals, Adh n4 levels increased 13-fold, such that Adh n4 and Adh þ transcript levels were equivalent, implying that the 25G mutation abolishes NMD pathway function. By contrast, Smg1 32AP had only a modest effect, increasing Adh n4 levels by 30%.

The NMD Pathway Targets the SV40 39 UTR of Drosophila Transgenes
To investigate how the NMD genes influence transgene expression, we examined the effect of photoshop mutations on transgenes containing different reporter genes, promoters, and 39 UTRs (Table 1; Figure 4). All of the transgenes whose steady state expression levels were found to be upregulated in photoshop mutants were constructed in the Left, agarose gel analysis of PvuII-digested products of RT-PCR of Adh RNA extracted from heterozygous Adh n4 /Adh þ adult males that were wild-type (WT) for NMD genes (lanes 1 and 2), hemizygous for Smg1 32AP (lanes 3 and 4), or hemizygous for Upf2 25G (lanes 5 and 6). Adh n4 is a nonsense mutation that also disrupts a PvuII site that divides the 671-bp PCR product into 484-bp and 187-bp fragments. For each genotype, two independent RNA samples were analyzed. Right, quantification of Adh n4 and Adh þ RNA levels by capillary electrophoresis of PvuII-digested RT-PCR products. Similar results were obtained by directly sequencing the same RT-PCR products and quantitating the level of each product from the chromatogram. doi:10.1371/journal.pgen.0020180.g003 Actin GFP drs À (b) His:GFP [55] His2AvD his:GFP His2AvD À (b) srcGFP [29] UAS/hsp70 src:GFP hsp70 pUAST vector, the standard vector for gene misexpression in Drosophila [27]. This vector contains multiple binding sites for the yeast transcription factor GAL4 followed by a core promoter derived from the hsp70 gene upstream (59) of the reporter insertion site, and the SV40 39 small t antigen intron and polyadenylation signal (henceforth referred to as the SV40 39 UTR) downstream (39) of the insertion site [28]. Hence, it was possible that the photoshop phenotype was due to increased GAL4 expression or activity, increased hsp70 promoter activity, an interaction with the SV40 39 UTR, or some special feature of the GFP and DsRed coding sequences. Upf2 25G enhanced expression of the pUAST-eGFP transgene ( Figure 4B), which has a metazoan optimized codon bias, suggesting that the unusual codon bias of native GFP and DsRed genes is not critical for the photoshop effect. The GFP and DsRed reporters do not contain introns in their coding sequences, so we inserted a 61-bp intron from Drosophila gene CG3585 into pUAST-eGFP to make pUAS-eGFPþI. This transgene did not show significantly increased eGFP signal in wild type and was still sensitive to photoshop mutations ( Figure  4C), suggesting that the photoshop effect is not through surveillance of intronless coding regions.
One tested transgene, src:GFP, which was not upregulated in Smg1 32AP animals, contains the same GAL4 binding sites and hsp70 promoter as the photoshop-sensitive pUAST constructs, but its 39 UTR and polyA signal are derived from Drosophila hsp70 instead of SV40 [29]. This implicated the SV40 39 UTR in the photoshop effect. To test the role of the SV40 39 UTR directly, we replaced the SV40 39 UTR in pUAST-GFP and pUAST-GFPþI constructs with the hsp70 39 UTR. Expression of these constructs was insensitive to photoshop mutations ( Figure 4D and 4E), confirming the importance of the 39 UTR. Interestingly, expression levels of GFP from the constructs containing the hsp70 39 UTR were somewhat lower than those of the corresponding pUAST-eGFP or pUAST-eGFPþI transgenes in a wild-type background and much lower than those of pUAST-eGFP or pUAST-eGFPþI transgenes in a photoshop mutant background. The significance of this finding is discussed below.
A small intron present in the SV40 39 UTR was an appealing candidate for sensitizing transcripts to the photoshop effect. A model for recognition of PTCs in vertebrates is that they occur 59 of the site of an intron, the same arrangement of the termination codons of GFP and DsRed with respect to the SV40 intron. RT-PCR experiments confirmed that the SV40 intron was indeed recognized and spliced in Drosophila larvae. However, deletion of this intron in the pUAST-eGFP and pUAST-eGFPþI constructs (to make pUASTDI-eGFP and pUASTDI-eGFPþI) did not eliminate the photoshop effect ( Figure 4F and 4G). Thus, targeting of the SV40 39 UTR by the NMD pathway does not require splicing of this intron, the only known intron in the UTR.

Targets of the NMD Pathway during Development
To identify candidate endogenous targets of the NMD pathway, we compared steady state RNA levels in hemizygous Upf2 25G male larvae to those in wild-type male larvae using a whole genome microarray. To avoid confounding effects of the Upf2 mutation on developmental progression, we focused our analysis on a set of 954 genes that do not undergo significant changes in expression levels during development or differ in expression between male and female larvae (see Materials and Methods). Among this set, we found 14 genes whose expression was upregulated 2-fold or more, and 26 genes that were downregulated 2-fold or more in the mutant compared to wild type (Tables S1 and S2). Genes whose expression was downregulated could be novel targets whose expression is paradoxically enhanced by the NMD pathway, or they could be indirect effects of NMD pathway inactivation. The affected genes encode proteins of diverse classes, including signal transduction molecules, proteases, and proteins involved in cell metabolism. Most of the affected genes differed from ones identified recently in Drosophila S2 cells depleted of NMD-gene function by RNA interference [19], suggesting tissue-specific regulation.
Two well characterized genes upregulated in the photoshop mutant were analyzed further. The mRNA of orthinine decarboxylase antizyme (oda, also called gut feeling) contains a naturally occurring coding sequence frame shift [30] that causes the transcript to contain early termination codons, and is an NMD target in cultured Drosophila S2 cells [19]. oda transcript is also a target of the NMD pathway during development: oda RNA levels were increased in hemizygous Upf2 25G larvae as determined by microarray analysis (3-fold) and by quantitative RT-PCR (2-fold; Figure 5A). However, oda transcript levels were not significantly increased in Smg1 32AP mutants ( Figure 5A).
The sex determination gene transformer (tra) [31] is also an NMD pathway target during development, as in S2 cells [19]. tra transcript levels were increased 3-to 4-fold in hemizygous Upf2 25G larvae, as determined by microarray analysis and quantitative RT-PCR ( Figure 5A). Smg1 32AP also increased tra levels, but to a lesser extent ( Figure 5A). tra might be expected to be an NMD pathway target because the primary transcript is spliced in males to produce a long mRNA (traL) that contains an early termination codon that would likely be recognized as premature, whereas in females some of the primary transcript is alternatively spliced to produce a shorter transcript (traS) lacking the early termination codon and encoding full-length protein ( Figure 5B). Indeed, RT-PCR experiments indicate that the increase in tra levels in Upf2 25G mutant males and females was due to selective stabilization of traL ( Figure 5B and data not shown).

Drosophila NMD Genes Are Dispensable for Many Developmental Processes but Provide Cells a Competitive Growth or Survival Advantage
The above results show that the core NMD gene Upf2 is broadly active during development and influences expression of dozens of genes, including a key sex determination gene, and is required for larval viability. To identify specific cellular and developmental functions of the NMD pathway, we analyzed the effect of photoshop mutations on sex determination and cell growth and differentiation. We did not detect any defects in sex determination in adult males hemizygous for Upf2 25G , despite the observed increase in traL levels: sexspecific splicing of downstream gene dsx was normal, as was that of the sex determination genes Sxl and msl-2; sex combs and genitalia appeared normal; and males made sperm and were capable of mating. Homozygous Upf2 25G females also appeared normal. We also did not detect any defects in larval cell differentiation. Larval tracheal cell clones lacking Upf1 or Upf2 displayed wild-type morphology at all levels of branching, including tracheal terminal cell clones, which showed normal branching patterns and luminal structures ( Figure 6A and 6B). Likewise, type IV da neurons, another morphologically complex cell type [32], appeared normal when homozygous for Upf1 or Upf2 null mutations ( Figure 6C and 6D and data not shown).
The only prominent defect we detected in cell clones mutant for NMD genes was an inability to contribute to adult (imaginal) structures. Whereas homozygous clones of all photoshop mutants were readily obtained in the larval tracheal system, we did not recover clones in the adult tracheal system, which is generated by proliferation of imaginal tracheal precursor cells during metamorphosis. We also did not recover large clones in adult epidermis or eyes, Values shown are means 6 2 standard errors of the mean relative to the results with y w FRT 19A /Y. GFP RNA levels in these larvae (see Figure 1H) are included for comparison. EK, photoshop-independent control gene defined by expressed sequence tag EK161155. (B) Effect on tra RNA levels in females. Above are shown sex-specific splice patterns in the coding portion of the tra transcript. Males use splice pathway shown below line to produce traL, whereas females use both splice pathways to produce traL and traS. traS encodes a 197residue protein, whereas traL contains an early termination codon and encodes a 37-residue protein. Arrows, position of PCR primers used to simultaneously amplify traL (364-bp product) and traS (189 bp). Bottom left shows RT-PCR on RNA from heterozygous Upf2 25G /þ and homozygous Upf2 25G female larvae; aliquots of reaction were taken at PCR cycles indicated. Note increased traL in Upf2 25G homozygotes. Bottom right shows quantification of results after 30 PCR cycles. Areas of the peaks are indicated, normalized to traS peak. These and similar experiments (not shown) indicate traL is increased ;10-fold in Upf2 25G mutants. This was greater than the value measured for Upf2 25G mutant males (;4-fold; Table S1; Figure 5A), perhaps due to sex-specific differences in NMD or differences in sensitivity of the assays. doi:10.1371/journal.pgen.0020180.g005 although small peripheral eye clones were occasionally observed ( Figure 6E and 6F). However, when the GMR-hid technique [33] was used to eliminate all heterozygous and wild-type cells in the developing eye, we found that the remaining photoshop mutant cells could proliferate, differentiate, and form an eye, although the eyes were smaller and more disorganized (rougher) than those of controls ( Figure  6G and 6H). These results suggest that Drosophila NMD genes are not required for cell proliferation, survival, or differentiation, but provide proliferating cells with a competitive growth or survival advantage during development.

Discussion
We have isolated to our knowledge the first mutations affecting NMD in Drosophila based on their ability to enhance expression of a GFP transgene, an effect we call the photoshop phenotype. Mapping of the mutations, complementation tests, and molecular analysis demonstrate that the photoshop mutations identify three genes, the Drosophila orthologs of NMD pathway genes Upf1, Upf2, and Smg1. The results show that Upf1 and Upf2 are essential genes, required for NMD and, at least in the case of Upf2, for proper expression of dozens of native mRNAs during development, including oda and the sex-nonspecific form of tra that contain early termination codons. By contrast, Smg1 is dispensable and only potentiates the NMD pathway. Genetic mosaic analysis of the Upf genes showed that they are not required for cell proliferation, survival, or complex cell differentiation events such as tracheal and neuronal growth and sprouting, but they provide proliferating imaginal cells with a competitive growth or survival advantage during development. We also mapped the cis-acting signal that confers sensitivity to the NMD pathway in the transgenic reporter assay, and discovered that it resides in the heterologous 39 UTR present in the reporter construct. Below, we discuss the implications of these results for our understanding of the functions and mechanism of the Drosophila NMD pathway during development, and compare and contrast them with what has been found for NMD pathway function in other organisms.

Roles of NMD Genes in Drosophila Development
The finding that mutations in Drosophila NMD genes Upf1 and Upf2 cause lethality during larval development contrasts with the minor effect of mutations in the homologous genes in yeast (mutations have almost no discernable effect on growth or survival [5]) and in C. elegans (mutants are viable and have only morphogenetic defects late in development [6]). Why are Upf1 and Upf2 essential in Drosophila? One possibility is that they are required to eliminate mutant transcripts with PTCs that encode truncated, deleterious protein products. Such PTC-containing alleles could be present in the background of our Upf1 and Upf2 mutants, and in the absence of NMD activity, these mutations become lethal. However, all our Upf1 and Upf2 mutations were independently isolated, and the lethality in these lines segregates with the Upf mutations, not with other genomic regions. Thus, if this explanation is correct, there would have to be multiple, potentially lethal PTC mutations distributed throughout the genome.
A second possibility is that the NMD pathway has a more general surveillance function that also eliminates naturally occurring transcripts resulting from aberrant splicing events or repetitive DNA elements, and accumulation of such transcripts is toxic. However, many aspects of cell biology and development appear normal in Upf mutants. Indeed, (E and F) Control (y w) wild-type clones (E) and Upf1 13D clones marked with w À in adult eyes. Wild-type clones (E) are easily detected as w À patches (white areas, demarked with dotted lines) in the w þ (red) heterozygous eye, whereas only small w À clones or single w À cells are detected in the Upf1 13D heterozygous eye (F). Similar results were obtained with Upf2 14J (not shown). (G and H) Similar experiment with Upf2 14J mutant except eye cells not part of the clone were eliminated by GMR-hid technique. Upf2 þ control clones proliferate (G) to form an eye (slight eye roughness is common with GMR-hid technique), as do Upf2 14J clones (H), although the latter are somewhat smaller and rougher than the controls. Similar results were obtained with Upf1 26A and Upf1 13D (not shown). doi:10.1371/journal.pgen.0020180.g006 sensitive assays examining individual tracheal cells and neurons show that loss of Upf function does not lead to cell death or impairment of complex cell morphogenesis events, implying that NMD inactivation does not cause general cellular toxicity.
A third possibility, which we favor, is that the NMD pathway modulates the activity of specific native transcripts, whose misregulation leads to lethality. An initial microarray survey identified several dozen genes of diverse functional classes whose expression was altered in NMD mutant larvae (Tables S1 and S2). Some of the affected transcripts contain early stop codons that are interpreted as bona fide PTCs, as for tra and oda genes ( Figure 5A); other affected transcripts could harbor other cis-acting signals that are interpreted as aberrant by the NMD machinery, like the SV40 39 UTR discussed below. In the absence of the NMD pathway, overexpression of such transcripts would perturb the development or function of select cells or tissues and lead to lethality. Indeed, although many cells appear to develop normally in photoshop mutants, we never observed Upf1 À and Upf2 À clones in adult tracheae or epidermis, and only small clones were found in the eye, implying that the NMD pathway promotes growth or survival of the proliferating imaginal cells that give rise to these tissues. Because Upf genes are broadly expressed [34] and broadly active ( Figure 1F and 1G and unpublished data) throughout development, identification of the tissue focus of Upf lethality will be an important first step towards identifying the critical cellular targets of NMD gene regulation.
The sole cellular defect we identified in NMD mutants was the absence or small size of mutant clones in the adult tissues described above, which is reminiscent of the cell cycle arrest observed in cultured Drosophila S2 cells depleted of Upf function [19]. Although these results suggest a function for the NMD pathway in cell proliferation or survival, the requirement for the pathway in these processes is not absolute: Upf1 À and Upf2 À cells were able to proliferate and form an eye when competing wild-type eye progenitor cells were eliminated ( Figure 6). This implies that the NMD pathway provides proliferating imaginal cells with a competitive growth advantage that prevents them from being overtaken by other proliferating cells during development. This could be a cell autonomous effect, like Minute mutations [35], or a cell nonautonomous effect, e.g., if the pathway influences expression of a signal secreted from proliferating cells that affects growth of neighboring cells (e.g., [36]). The NMD pathway may play a similar, or more extreme, role in mice, because a mouse UPF1 knockout is lethal and attempts at establishing UPF1 À/À embryonic stem cells were unsuccessful [18].
Recently, it has been suggested that Upf1 and Upf2 participate in other aspects of gene regulation besides NMD, such as stimulating translation [37] and in translational termination [38] (reviewed by [39]). It is important to note that for neither mouse nor our Drosophila mutants is it established that the lethality associated with Upf1 and Upf2 mutations derives from their roles in NMD. Indeed, our analysis of Upf2 25G suggests the opposite possibility. This allele appears completely compromised for NMD, as assessed by expression of an Adh mRNA carrying a PTC (Figure 3), yet unlike Upf2 null alleles, some Upf2 25G mutants survived to adulthood. Thus, either the essential function of Upf2 is in a process other than NMD, or this allele retains residual NMD function sufficient for regulation of its essential target genes but not of others like the mutant Adh mRNA.
The differing molecular, cellular, and developmental requirements for NMD-pathway genes in yeast, C. elegans, Drosophila, and mice make clear that the function of this pathway has diversified during evolution. Perhaps the ancestral function of the pathway was in some general process like translation termination, and only later did the pathway evolve roles in monitoring transcripts for PTCs and more specialized regulatory roles. Alternatively, the ancestral function could have been regulation of RNAs involved in a specific cellular process such as cell growth regulation, and only later did the pathway acquire a more general role in RNA surveillance.

Smg1 Potentiates the Drosophila NMD Pathway
Our genetic analysis demonstrated a striking difference in the developmental requirements of Smg1 compared to those of Upf1 and Upf2. First, Upf1 and Upf2 are essential genes, whereas an amorphic Smg1 allele resulted in viable and fertile animals. Second, a Upf2 mutation abolished NMD of an Adh PTC allele, whereas the amorphic Smg1 mutation only modestly reduced NMD efficiency. Third, the magnitude of the Smg1 mutant effect differed at different targets. At some targets, such as oda, there was little or no effect of the Smg1 mutation, whereas at other targets, such as tra and a GFP transgene, the Smg1 mutant effect was up to half that of the Upf2 mutant. The small and gene-selective effect of Smg1 could explain why a recent genetic analysis failed to detect a role for Drosophila Smg1 in NMD [21], whereas earlier Drosophila cell culture studies suggested an important role for the gene [10]. The small and gene-selective function of Drosophila Smg1 contrasts with genetic results in C. elegans, which did not identify differences in the requirements of smg-1 and the Upf1 and Upf2 homologs smg-2 and smg-3 [6]. One possibility is that Drosophila has another protein with activity similar to that of Smg1. This seems unlikely because Smg1 is the only sequence ortholog of Smg1-family genes in the Drosophila genome, although there are other genes that encode proteins with PI3K-related kinase domains. Another possibility is that phosphorylation of Upf1 by Smg1 is not absolutely required for Upf1 activity in Drosophila but only enhances its activity or reactivates spent protein after a catalytic cycle. This would be more similar to the NMD pathway in yeast, which lacks a Smg1 ortholog and is thought to function without a Upf1 phosphorylation cycle, than to the NMD pathways in C. elegans and vertebrates, where the Upf1 phosphorylation cycle is thought to be essential for pathway activity.

Targeting of a Specific 39 UTR by the Drosophila NMD Pathway
We found that a variety of reporter constructs in the pUAST transformation vector were upregulated when NMD pathway function was abrogated ( Table 1), implying that transcripts derived from this vector are recognized as aberrant by the RNA surveillance machinery. Strictly speaking, this is not an NMD process, because all of the transgenic constructs contain full-length coding sequences with no PTCs. However, because multiple NMD genes are involved in this regulation, and because the observed increase in reporter activity is associated with increased reporter mRNA, it suggests that an NMD-related RNA decay process normally limits expression of pUAST transgenes in Drosophila. The signal that targets transgenes for regulation by the NMD pathway appears to lie in the SV40 39 UTR of the pUAST vector: all reporter constructs that were sensitive to the NMD pathway contain this UTR, and swapping it for one derived from the hsp70 gene rendered the transcript insensitive to NMD, the first example to our knowledge of a change in NMD pathway sensitivity due solely to swapping intact 39 UTRs. Transgenes containing three other endogenous 39 UTRs (K10, drs, and His2AvD; see Table 1) were also insensitive to NMD, supporting the conclusion that the SV40 39 UTR harbors a critical targeting element. Although we have not identified the specific sequence or structural characteristic (e.g., length [17,40]) within the SV40 39 UTR responsible for targeting by NMD machinery, it does not require the small, naturally occurring intron within the UTR, nor does it require any specific sequences in the translated region of the targeted mRNA. We conclude that a 39 UTR can provide a critical signal for regulation by the NMD pathway in Drosophila, as has been observed in yeast [7] and humans [17].
Our analysis also indicates that the SV40 39 UTR has a second, positive effect on transgene expression in Drosophila, similar to one noted in insect cell culture [41]. In wild-type Drosophila, this positive effect is partially offset by destruction of the RNA by the NMD pathway, resulting in an intermediate level of reporter expression (Figure 7). However, in NMD pathway mutants, the positive effect of the SV40 39 UTR is unmasked, resulting in enhanced reporter expression and the photoshop phenotype. This contrasts with results for transgenes carrying the 39 UTR from hsp70, which does not strongly enhance transgene expression and is not targeted by the NMD pathway, resulting in a lower and photoshopmutation-insensitive level of reporter expression.
Because expression of UAS-GFP and other reporters is affected by mutations in homologs of all three NMD genes on the chromosome we screened, the assay can likely be used to identify and characterize additional NMD pathway genes on other chromosomes. The assay has two important features. First, because it can be carried out in single cells in genetically mosaic animals, a requirement of candidate NMD genes for organismal viability can be bypassed. Second, the assay is very sensitive to perturbations in the NMD pathway. For example, loss of Smg1 activity leads to only a modest increase in stability of PTC-containing transcripts but a readily detectable enhancement of GFP reporter expression. Together these features suggest that the transgenic assay system can be used to test requirements in vivo of candidate NMD genes and drugs that influence pathway activity, which could be useful in modulating expression of human disease genes carrying PTCs [42,43].
The photoshop mutations were obtained by mutagenesis of an isogenized y w FRT 19A chromosome [22] with 25 mM ethane methyl sulfonate overnight [47] in a tracheal mutant screen (to be described elsewhere). The mutations used were on this chromosome unless otherwise noted. To generate homozygous mutant clones, 2-to 6-h-old embryos were collected at 25 8C from a cross of y w * FRT 19A /FM7 females to gal80 FRT 19A , hsFLP122/Y; btl-GAL4, UAS-GFP males. After a 45-min heat shock at 38 8C to induce FLP expression, embryos were returned to 25 8C to continue development. L3 larvae of genotype y w * FRT 19A /gal80 FRT 19A , hsFLP122; btl-GAL4, UAS-GFP/þwere identified by GFP mosaicism within the tracheal system and scored for the photoshop phenotype.
The original Smg1 32AP chromosome (designated 32AP ) carried lethal mutations not associated with the photoshop phenotype. Lethals were removed by crossing 32AP /y w FRT 19A females to w/Y; btl-GAL4, UAS-GFP males and identifying L3 larvae with enhanced GFP throughout their tracheal systems. These Smg1 32AP /Y; btl-GAL4, UAS-GFP/þ larvae developed into viable adult males. We also found a viable wing morphology mutation on 32AP that is allelic to wavy. Existing wavy alleles do not show a photoshop phenotype, and the wing phenotype is separable from the photoshop phenotype, so wavy does not seem to contribute to the photoshop phenotype.
The Upf2 29AA chromosome carries a linked lethal mutation. When recombined away from Upf2 29AA , the mutation had no effect on tracheal development or reporter expression. However, we have not obtained a recombinant containing Upf2 29AA without the extraneous mutation.
For complementation tests of Upf2, we used genomic rescue transgenes located on the autosomes to generate males of genotype y w Upf2 14J v g f FRT 19A /Y; Pfw þ , Upf2 þ g/þ and crossed these to y w * FRT 19A /FM7c females, where the asterisk indicates the tested mutation. Absence of Bar þ , white-eyed female progeny indicated failure to complement. For complementation tests of Upf1, we used the Y-linked duplication Dp(1;Y)BSC1, y þ , which covers the Upf1 locus. Males of genotype Upf1 13D /Dp(1;Y)BSC1, y þ were crossed to Upf1 26A / FM7c females, and the absence of female Bar þ progeny indicated a failure to complement.
Mapping of photoshop mutations. Identification of SNPs, construction of the SNP map of the X chromosome, and details of their use in mapping X chromosome mutations will be described elsewhere. Briefly, the location of the lethality associated with a photoshop mutation was mapped by crossing y w * FRT 19A /sc cv ct v g f FRT 19A females (where the asterisk indicates the lethal mutation) to FM7c/Y males and scoring the viable male progeny for the visible markers to determine the lethal interval. To refine the map position, Figure 7. Model for Drosophila NMD Pathway Action on SV40 39 UTR In wild type (WT), SV40 39 UTR enhances transgene expression (arrow). However, effect is partially offset by transcript degradation by NMD machinery, giving an intermediate level of expression (þ þ). In photoshop (NMD) mutants, transcript degradation is abrogated, resulting in strong enhancement by SV40 39 UTR (thick arrow) and high level of expression (þ þþ ). The hsp70 39 UTR does not enhance expression (or alternatively promotes degradation) but is also not a target of the NMD pathway, giving a low level of expression (þ) insensitive to photoshop mutants. doi:10.1371/journal.pgen.0020180.g007 we collected male progeny in which a recombination event had occurred within the mapped interval and scored them for SNPs. For each mutation we typically scored 300-400 males for SNPs. For Upf1 26A and Upf1 13D , which map between v and g, we crossed the y w * FRT 19A /sc cv ct v g f FRT 19A females to males of genotype Df(1)64c18, g 1 sd 1 /Dp(1;2;Y)w þ to distinguish y sc þ w cv þ ct þ v þ g f recombinants, which we could not otherwise identify because of epistasis of w over v and g. We also crossed the y w * FRT 19A /sc cv ct v g f FRT 19A females to sc cv ct v g f FRT 19A /Y males to identify recombinant females, which were then tested for the photoshop phenotype in genetic mosaics to confirm that the lethality and photoshop phenotype mapped to the same interval. For the viable mutation 32AP, we followed a similar strategy as for the lethals, except we scored recombinant males for the presence or absence of 32AP by testing the enhancement of GFP in btl-GAL4, UAS-GFP transgenic animals.
Transgene construction. For the Upf2 genomic rescue construct, Drosophila BAC 24A2 [48], which contains Upf2, was transformed into Escherichia coli strain EL250, which harbors heat-shock-inducible homologous recombination machinery [49]. We then cloned a 200-bp fragment located upstream, and a 300-bp fragment located downstream, of Upf2 coding sequence and UTRs based on the cDNA RE04053 (rather than the canonical Upf2 cDNA SD07232, which appears to be defective as it lacks a conserved portion of Upf2 coding sequence) tandemly into the Drosophila transformation vector pCaS-peR4 [50] with a unique NotI site between the fragments to give pMM#200. pMM#200 was linearized with NotI and transformed into EL250[24A2], in which the recombination machinery had been induced. Transformants were plated onto LB plates containing carbenicillin to select for gap repair of pMM#200, which can occur by homologous recombination with the BAC and result in transfer of Upf2 into pMM#200. The resultant plasmids were analyzed by restriction digestion, and one with the expected pattern (pMM#201) was used to establish transgenic lines on the second and third chromosomes by P-element transformation. Six lines were tested and all six rescued hemizygous male Upf2 14J mutants to viability. The degree of rescue varied based on insertion site, but for the strongest lines (Pfw þ , Upf2 þ g11A and 24) rescued animals appeared indistinguishable from Upf2 þ animals and were readily maintained as stocks of genotype Upf2 25G ; Pfw þ , Upf2 þ g11A or 24/þ or Upf2 14J ; Pfw þ , Upf2 þ g11A or 24/þ.
For GFP reporter constructs, coding sequence of eGFP (Clontech; http://www.clontech.com) was amplified by PCR using KpnI linker primers Kpn5GFP and Kpn3GFP (see Table S3 for primer sequences). The product was digested with KpnI and cloned into the KpnI site of pUAST to generate pUAST-eGFP. A PCR-based strategy was used to insert a 61-bp intron derived from gene CG3585 between nucleotides 330 and 331 of the eGFP coding sequence to make eGFPþI constructs.
The intron in the SV40 39 UTR of pUAST was deleted by a PCRbased strategy to give pUASTDI. To replace the SV40 39 UTR with the hsp70 39 UTR, primers XbaHsp70 and Hsp70Stu were used to amplify the hsp70 39 UTR from pGATB [27]. The PCR product was digested with XbaI and StuI and cloned between the XbaI and StuI restriction sites of pUAST-eGFP (to make pUAST.h-eGFP) or pUAST-eGFPþI (to make pUAST.h-eGFPþI).
Constructs made using PCR were sequenced to confirm that mutations had not been introduced. Constructs were transformed into Drosophila using standard microinjection techniques using the D2-3 helper plasmid as the transposase source.
RNA analysis and quantitation. L3 larvae of the appropriate genotype were identified by enhancement of GFP for mutant alleles or by sexing using gonad morphology for wild-type controls. Total larval RNA was prepared using Trizol reagent (Invitrogen; http:// www.invitrogen.com), and genomic DNA contamination was eliminated with DNAse (DNA-free, Ambion; http://www.ambion.com). RNA concentration was determined spectrophotometrically and normalized before reverse transcription with MuMLV reverse transcriptase (Retroscript kit, Ambion). Real-time quantitative PCR was done with a thermocycler (iCycler) and real-time PCR mix (iQ SYBR Green Supermix; Bio-Rad Laboratories; http://www.bio-rad.com). Primers used in the qRT-PCR experiments were designed to amplify 60-to 100-bp fragments within single exons; amplification of Drosophila genomic DNA containing a UAS-GFP transgene showed that primers gave a linear amplification response at concentrations ranging over four orders magnitude. Control reactions performed on RNA without reverse transcription or with primers against nontranscribed regions of genomic DNA gave negligible signals compared to experimental reactions. Experimental reactions were carried out in duplicate (except EK161155 was done once) on two RNA samples that were derived independently from the RNA used for microarray analysis. Results were normalized to results with rp18LA transcript, a gene that is not developmentally regulated [51].
For analysis of Adh RNA levels, cDNA was prepared from adult males of the appropriate genotype. Adh transcripts were amplified with primer AdhL and the primer AdhR (for agarose gel analysis and sequencing) or AdhR2Fam (for capillary electrophoresis). PCR products were sequenced directly or digested with PvuII to distinguish the Adh þ and Adh n4 alleles. Capillary electrophoresis was performed on an ABI 3730x1 (Applied Biosystems; http://www. appliedbiosystems.com) and analyzed using GeneMapper v3.0 software (Applied Biosystems).
Microarray analysis. RNA was isolated from Upf2 25G /Y; btl-GAL4, UAS-GFP/þ and y w FRT 19A /Y; btl-GAL4, UAS-GFP/þ L3 larvae using Trizol as described above. cDNA labeled with Cy3 or Cy5 was prepared from each RNA sample and hybridized to microarrays containing ;14,000 gene probes [52,53]. Hybridizations were performed with two independently isolated and labeled RNA samples. Analysis was carried out using the Stanford Microarray Database (http://genome-www5.stanford.edu). During analysis we noted that the Upf2 25G mutants were delayed in development. To avoid confounding effects of changes in gene expression that result from developmental regulation rather than more direct effects of Upf2 loss of function, we used the available wild-type developmental gene expression time course [51] to filter out genes whose transcription changed more than 25% from their maximal value during hours 72-96 of larval development. The wild-type dataset includes ;33% of genes, and we used just this subset for our analysis. Furthermore, the wild-type dataset is for mixed sex populations, while our microarray was performed only on males. To compensate for sex differences, we also excluded from analysis genes that differed between males and females by more than 50% based on data from male and female larvae (E. Johnson and M. A. K., unpublished data). Our analysis of genes regulated by NMD is therefore conservative, covering only non-developmentally regulated and non-sex-regulated genes whose expression was affected by a hypomorphic Upf2 allele, and thus provides only a lower estimate of genes regulated by NMD.
Author contributions. MMM and MAK conceived the experiments. MMM performed the experiments. MMM and MAK analyzed the data and wrote the paper.
Funding. MMM was supported by a Helen Hay Whitney Foundation postdoctoral fellowship. MAK is an investigator of the Howard Hughes Medical Institute.
Competing interests. The authors have declared that no competing interests exist.