Nonsense-mediated mRNA decay (NMD) is an RNA surveillance system that down-regulates mRNAs containing early stop codons in all eukaryotes examined 
. NMD functions to clear the cell of transcripts containing potentially harmful nonsense mutations 
. In addition to this role in surveillance of mutations, NMD affects the expression of numerous non-mutant endogenous targets 
. These natural targets include many mRNAs that are the products of alternative splicing; one study reported that 45% of alternatively spliced human genes have at least one isoform that may be degraded by NMD 
In some of these cases, alternative splicing and NMD act together to regulate gene expression, providing an additional layer of post-transcriptional regulation. By altering the abundance and activity of splicing factors, the cell can differentially splice a pre-mRNA into a productive mRNA that encodes a protein or into an unproductive mRNA with an early stop codon that makes the mRNA a target for NMD. Unproductive splicing is used in the regulation and autoregulation of numerous genes 
including mammalian splicing factors, spliceosome components 
and the spermidine/spermine N1-acetyltransferase (SSAT) gene 
Alternative splicing is prevalent in the fruit fly Drosophila
. At least 46% of detected genes show differential expression of alternative regions during development 
. In flies, alternative splicing plays an important role in many processes including sex determination, neuronal wiring, and eye development 
. Although NMD is active in Drosophila
, our understanding of its impact on the fly transcriptome is limited. A study of the effect of NMD on gene expression in Drosophila
showed that levels of 14% of detected genes increased at least 1.5-fold after a key NMD factor, UPF1, was depleted 
. This analysis used gene expression microarrays that assess total mRNA from a gene, and thus it could not measure the levels of distinct alternative splice forms. Natural NMD targets produced by alternative splicing in Drosophila
have not been assayed previously.
The NMD machinery of Drosophila
, as in all eukaryotes studied, requires the core set of UPF proteins, UPF1, UPF2, and UPF3 
. As in mammals, it also involves SMG1, SMG5, and SMG6 (but, unlike mammals, not SMG7), which are involved in the phosphorylation and dephosphorylation of UPF1 
. Although the core NMD machinery is essentially the same in human and Drosophila
, the mechanism by which premature termination codons are recognized is different in the two organisms. In both cases, the nonsense codon seems to be recognized as premature based on its position relative to proteins associated with the transcript, downstream of the stop codon. In human, the primary downstream markers are exon junction complexes deposited during splicing 
. Exon junction complexes are not required for NMD in Drosophila 
. A recent study indicates that, instead, some early stop codons are recognized based on their distance from the poly-A tail, mediated by the binding of cytoplasmic poly-A binding protein (PABPC1) 
. This study provided valuable data about the NMD mechanism based on manipulation of a single reporter construct. Studies of a wider range of NMD targets are necessary before a general rule can be inferred.
Splicing-sensitive microarrays have been used successfully to assay alternative splicing on a global scale (reviewed in 
). This method has been applied in fly to assess global splicing changes when splicing factors are inhibited or overexpressed and to measure sexually dimorphic splicing 
. Microarrays have also been used to measure the effect of NMD on the levels of alternatively spliced mRNAs in human, mouse, worm, and yeast 
. However, most techniques used to analyze these microarrays only measure the change in probes specific to individual alternative splice junctions or alternative exons. One method, successfully used to assay alternative splicing in human, measures changes in exon inclusion events 
, but has yet to be extended to more general splicing events. None of these methods provide isoform-level fold-changes, limiting their ability to find NMD targets. In this work, we have developed a new algorithm that makes it possible to obtain isoform-level measurements for all categories of alternative splicing and alternative processing events. We use a generative non-linear regression model to deconvolve individual probe measurements into estimates of overall isoform-level fold-changes and relative proportions of isoforms.
Our goals in this project were two-fold: first, to determine the effect of NMD on alternatively spliced mRNAs in the Drosophila transcriptome, and second, to identify features of these transcripts that might cause them to be targets of NMD. To assess the effect of NMD, we have inhibited NMD in Drosophila cells and measured changes in expression on a custom splicing-sensitive microarray. After measuring junction and exon splicing changes and then estimating isoform-level fold-changes, we identified NMD targets using a hierarchy of stringent criteria that eliminate many secondary effects and potential artifacts, at the cost of substantially reduced sensitivity to legitimate NMD targets. Using this conservative approach, we have found a high-confidence set of 45 genes where NMD decreases the level of one isoform without impacting the levels of other isoforms. We found that the reading frames of NMD–target mRNAs were often misannotated in sequence databases. After identifying the correct reading frames, we found that the NMD–target mRNAs differed significantly from the nontarget isoforms, with shorter coding regions and longer 3′ untranslated regions (UTRs). Our results show that alternative splicing and NMD affect a diverse set of genes in fly including genes involved in translation and mitosis, suggesting that regulation of unproductive splicing might play important roles in Drosophila.