Three Drosophila Hox Complex microRNAs Do Not Have Major Effects on Expression of Evolutionarily Conserved Hox Gene Targets during Embryogenesis

The discovery of microRNAs has resulted in a major expansion of the number of molecules known to be involved in gene regulation. Elucidating the functions of animal microRNAs has posed a significant challenge as their target interactions with messenger RNAs do not adhere to simple rules. Of the thousands of known animal microRNAs, relatively few microRNA:messenger RNA regulatory interactions have been biologically validated in an normal organismal context. Here we present evidence that three microRNAs from the Hox complex in Drosophila (miR-10-5p, miR-10-3p, miR-iab-4-5p) do not have significant effects during embryogenesis on the expression of Hox genes that contain high confidence microRNAs target sites in the 3′ untranslated regions of their messenger RNAs. This is significant, in that it suggests that many predicted microRNA-target interactions may not be biologically relevant, or that the outcomes of these interactions may be so subtle that mutants may only show phenotypes in specific contexts, such as in environmental stress conditions, or in combinations with other microRNA mutations.


Introduction
MicroRNAs (miRNAs) are a recently discovered class of biological molecules that have greatly expanded our knowledge concerning post-transcriptional regulation of gene expression. Since the discovery of these small RNA regulatory molecules, many of the proteins involved in their biogenesis, transport, and regulatory functions have been characterized [1]. While much certainly remains to be elucidated concerning the protein components of the miRNA pathway; the larger 'black box' is that of target interactions between miRNAs and messenger RNAs (mRNAs).
The first major discoveries in the field of miRNA research were made through careful analyses of genetic mutations in the nematode C. elegans with obvious phenotypes [2][3][4]. These studies gave rise to the first ideas of how miRNAs interact with their target genes. Subsequent systematic mutational analyses of established miRNA target sites as well as synthetic miRNA targets provided researchers with the first set of target pairing 'rules' for animal miRNAs [5][6][7][8]. These aided in the development of a number of computational algorithms which allow researchers to predict potential mRNA targets for a given miRNA, or conversely, potential miRNAs which target a specific mRNA [9]. However, due to the relative lack of functional data for validated miRNA:mRNA target interactions, these algorithms were largely based on small training sets in combination with the aforementioned target pairing 'rules'. In general, these algorithms produced lists of hundreds or even thousands of targets for a typical miRNA, of which a very limited number have ever been experimentally validated.
The majority of experimental 'validations' of computationally predicted miRNA:mRNA target pairs are performed in cell culture (e.g. luciferase reporter assays) or utilize transgenic over-expression assays, which can indicate the potential for regulation, but do not necessarily indicate that they are biologically relevant interactions. There is a growing body of literature which adds another layer of complication, suggesting that many miRNAs and their predicted mRNA targets are not expressed in the same cells. Instead, they are often expressed in complementary patterns, making in vivo validation of endogenous interactions very difficult [10].
Microarray data, both for miRNAs and putative mRNA targets, have been used to support the validity of the claims of hundreds of targets per miRNA [10][11][12][13]. This has proven to be effective to exclude genes as putative targets, as mRNAs that are co-expressed in a particular tissue with a given miRNA are found to be devoid of predicted target sites for that miRNA [10,[14][15][16]. Unfortunately, when mRNAs levels are altered in miRNA mutant animals or animals with ectopic overexpression of a given miRNA, this does not explicitly mean the mRNAs are directly targeted by the given miRNA. Additionally, quantitative mass spectrometry analysis of protein levels shows that a high proportion of predicted targets, even those which are conserved, are unaffected by changes in miRNA expression [17].
Recent studies have confounded matters even further. Systematic deletions of every miRNA gene in the C.elegans genome resulted in the surprising discovery that the vast majority had no significant phenotypes [18,19]. Experiments using CLIP-seq/ HITS-CLIP techniques have suggested that a large proportion of predicted miRNA-target sites are not bound by Argonaute proteins [20,21].
The general lack of functional data and dearth of knowledge about concurrent expression of specific miRNAs and their putative targets at a cellular level in developing animals, have led to a situation where most of the proposed miRNA:mRNA target interactions found in databases are conjectural. Many of the aforementioned studies revealed that target pairing 'rules' are more complicated than previously assumed and that different miRNAs may have different sets of target pairing 'rules'. Additionally, effective downregulation of a target genes may require additional context dependent sequences external to the target sites [22][23][24][25]. Taken together, these data imply that computational predictions may only be a reliable indicator of the target landscape for a limited number of miRNAs.
Despite the ambiguity involved in computational predictions, there are indications that Hox genes are likely to be an important class of miRNA targets. Evidence for post-transcriptional regulation of Hox gene expression has been accumulating for some time. Discrepancies have been found between transcript and protein levels for the mouse Hox gene Hoxb4 in the posterior neural tube [26], Hoxc6 in the chick hindlimb [27], and the the Sex combs reduced (Scr) ortholog in the first thoracic segment of Porcellio scaber [28]. It is currently unknown whether these discrepancies are due to miRNA regulation or other mechanisms such as localized protein instability.
In contrast to the modest number of documented cases of posttranscriptional Hox regulation, most Hox genes have been predicted to be direct targets of miRNAs in both vertebrates [29] and invertebrates [30]. One example of a Hox miRNA target being verified is the case of Hoxb8 regulation by miR-196, which was shown in mice to cause endonucleolytic cleavage of the mRNA target site [31,32]. This target site is conserved in vertebrate Hoxb8 genes, and suggests that miR-196 has a conserved role in vertebrate axial patterning [33][34][35]. On the basis of partial complementarity between miRNA sequences and 39 untranslated regions (UTR) sequences, Hox transcripts from the Drosophila melanogaster Scr, Antp, Ubx, abd-A and Abd-B genes have been proposed as targets of miRNA regulation [30], and several studies have been published in the last few years testing a number of these interactions in Drosophila [36][37][38][39], as well as analogous interactions in other animals [34,40,41].
The animal Hox complexes seem to be relatively rich in miRNAs, and considering the theory that miRNAs are spawned by neighboring genes which they can then go on to regulate [42], it seems compelling that these Hox cluster miRNAs might regulate nearby Hox genes, which they are often predicted to target. Most vertebrate and arthropod Hox complexes have at least three regions containing miRNA-producing hairpins. Although they reside in comparable positions in their respective Hox complexes, the arthropod miRNAs miR-iab-4 and miR-993 do not exhibit sequence homology to the vertebrate miR-196 and miR-615. On the other hand, miR-10 is a highly conserved miRNA, not only in sequence, but also in its genomic position in the complex between the Hox4 and Hox5 orthologs of most bilaterian animals [43].
In this study we investigated three strongly predicted miRNA:-Hox-target interactions: miR-10-5p:Scr, miR-10-3p:Abd-B, and miR-iab-4-5p:Antp. Using various techniques we show that these predicted interactions do not appear to play detectable roles during embryonic development, and at least considered in isolation, have very subtle effects on Hox protein levels.

Results
Expression of Drosophila pri-mir-10 transcripts miRNAs are initially transcribed as RNA polymerase II primary transcripts which are serially processed to produce mature miRNAs [44]. In order to determine the primary transcript that produces miR-10 (pri-mir-10), 59 and 39 RACE (rapid amplification of cDNA ends) were performed on a cDNA library created from polyadenylated (polyA) RNA from a 0-24 hour embryo collection. The 39 RACE products reveal the polyA cleavage site to be about 700 bp downstream of the hairpin sequence ( Figure 1A). 59 RACE gave multiple products, the longest of which revealed exon-intron boundaries and a putative transcriptional start site about 6.8 kb upstream of the hairpin ( Figure 1A). The sequence just upstream of the 59 end of the RACE products contains properly spaced INR, DPE, and MTE elements [45], and is also well conserved amongst Drosophilids ( Figure S1), which point to this region as a likely pri-mir-10 promoter. We therefore conclude that at least one pri-mir-10 transcript in Drosophila melanogaster spans an approximately 7.5 kb region of the chromosome between Scr and Dfd with a single approximately 5.5 kb intron ( Figure 1A and B), although the existence of secondary transcripts remains a possibility.
In order to determine the expression pattern of the mir-10 gene during embryogenesis, we performed in situ hybridizations using probes antisense to pri-mir-10 sequence and analyzed expression throughout embryogenesis. mir-10 primary transcripts are first expressed during the blastoderm stage of embryogenesis [46]. At this point pri-mir-10 is expressed in a broad band which corresponds to the trunk primordia of the embryo and is reminiscent of gap gene or Hox gene expression patterns (Figure 2A). The expression pattern at this stage is largely complementary to that of hunchback, suggestive of a possible negative regulatory interaction. However, expression of pri-mir-10 was unaltered when examined in embryos mutant for the hunchback gene (data not shown). The mir-10 primary transcript was also detected in yolk cells of blastoderm stage embryos (data not shown). During late blastoderm development and the beginning of gastrulation the expression of pri-mir-10 becomes downregulated in a subset of cells, taking on a striped appearance, which is similar to, but not as refined as that of the pair rule genes (data not shown). These stripes are approximately in register with those of the Hox complex gene fushi tarazu (data not shown). Shortly after the beginning of gastrulation, transcription of pri-mir-10 appears to shut off and then re-initiates during the early stages of germband elongation in a different pattern. As germband elongation proceeds, pri-mir-10 transcription is initiated in the anal pad and large intestine primordia and in ventral neurectoderm of the trunk segments where it is expressed in the developing neuroblasts ( Figure 2B). After germband retraction pri-mir-10 is transcribed in the anal pads and large intestine endoderm as well as a subsection of the central midgut endoderm ( Figure 2C). In late embryogenesis pri-mir-10 staining can be seen in the anal pads, large intestine endoderm, midgut endoderm, and in the ventral nerve cord of the central nervous system (CNS) ( Figure 2E-F). The anteroposterior extent of pri-mir-10 expression in the CNS is reminiscent of its blastoderm expression pattern. No pri-mir-10 transcription is detected in the brain, with the anterior border of expression in the posterior region of the sub-esophageal ganglion, extending posteriorly to near the terminal end of the CNS ( Figure 2E-F). It is possible that this pattern of expression in the CNS may be conserved in all bilaterians as a similar expression pattern is seen in zebrafish embryos [47,48].

Expression of mature miRNAs from pri-mir-10
After a pre-miRNA is cleaved by Dicer, one strand of the resulting ,22 nt dsRNA is packaged into an RNA Induced Silencing Complex (RISC). The strand with a lower 59 binding energy is typically packaged more often into RISC as a mature miRNA [49]. This type of analysis predicts that there is no biochemical reason why Drosophila miR-10-5p (previously called miR-10) should be preferentially packaged into RISC more often than miR-10-3p (previously called miR-10*), and in fact suggests that miR-10-3p should be packaged into RISC preferentially over miR-10-5p ( [49] and data not shown).
Both miR-10-5p [50,51] and miR-10-3p [50] were cloned in early efforts to determine the miRNA complement of the Drosophila genome. However, the data from these studies was not enough to statistically determine the relative abundance of either strand for most miRNA duplexes. Due to the presence of miR-10-5p sequences in most other animals the possibility of a second miRNA from this locus was largely ignored. Although the sequence from the 59 arm of the mir-10 hairpin is well conserved in bilaterians, the sequence from the 39 arm is not particularly well conserved. The conservation that does exist among disparate animal clades ( Figure S2) may be due mostly to the necessity for complementary bases which allow formation of a hairpin capable of being processed by the miRNA machinery. However, while the 39 hairpin sequence is not well conserved in deuterostomes ( Figure  S3B), if only arthropods are examined, both miR-10-5p and miR-10-3p sequences are found to be well conserved ( Figure S3A).
Additionally, although the putative seed sequence of the potential deuterostome miR-10-3p is conserved, free energy calculations suggest that it is unlikely to be packaged into RISC at appreciable levels (data not shown). (B) In germband extended embryos pri-mir-10 transcription begins in the anal pad primordia (app) and hindgut primordia (hgp) as well as ventral neurectoderm (vne). (C) In stage 14 embryos pri-mir-10 staining is seen in the developing anal pads (ap), hindgut endoderm (hge) of the large intestine, a small patch of midgut endoderm (mge), and continues to be expressed in the developing nerve cord (out of view). (E,F) In late stage embryos the domains of expression are maintained and CNS staining has an anterior boundary near the sub-esophageal ganglion and extends to near the posterior end of the ventral nerve cord. Starting around stage 13 both mir-10-5p (D) and miR-10-3p (G) mature miRNAs are detected by LNA in situ in the same patterns as for pri-mir-10 (compare to C and F respectively). (H) Detection of miR-10-3p shows mature miRNA is localized in the cytoplasm. (I) Northern blots of total RNA purified from Drosophila embryos at various developmental stages using 32 P end labeled antisense oligo probes against miR-10-5p and miR-10-3p. Time course shows both miR-10-5p and miR-10-3p are produced at very low levels in blastoderm embryos and during early germ band extension (0-4 hrs) and progressively increase in abundance throughout embryogenesis. miR-10-3p is found at much higher levels throughout embryogenesis when compared to miR-10-5p. doi:10.1371/journal.pone.0031365.g002 In order to test the abundance and timing of expression of the products of the mir-10 duplex, Northern blot analysis was performed on RNA from multiple embryonic stages using antisense probes complementary to each arm of the hairpin. Both miR-10-5p and miR-10-3p were found to be present in embryos, with levels increasing throughout embryogenesis, and with miR-10-3p present at significantly higher levels at all stages ( Figure 2I). These data agree well with recent large scale miRNA sequencing efforts which profile expression levels of all mature products of a miRNA gene [52]. In addition to Northern analysis of miRNAs, mature miRNAs can be detected by in situ hybridization by using modified oligonucleotides composed of Locked Nucleic Acid (LNA) nucleotides [47]. Both miR-10-5p ( Figure 2D) and miR-10-3p ( Figure 2G) can be detected in late embryonic stages utilizing this technique, and the expression pattern of each is equivalent to pri-mir-10 (compare Figure 2D to 2C and Figure 2G to 2E). Additionally, staining for the mature miRNAs is in the cytoplasm ( Figure 2H), unlike primary transcript staining, which is found exclusively in the nucleus.

Identification of putative targets
Due to the conserved location of the mir-10 gene in animal Hox complexes ( Figure S4), and the potential for miRNAs to be spawned from, as well as to target nearby genes [42], we were interested in the possibility that miR-10-5p and miR-10-3p might be regulators of Hox gene expression. Simple sequence motif searches of the 39UTR sequences of Drosophila Hox genes resulted in identification of potential target sites of varying strength.
The best putative target site for miR-10-5p was found in the 39UTR of the Hox gene Sex combs reduced (Scr). Although the 59 seed sequence pairing of miR-10-5p with this Scr site is not as extensive as the 'prototypical' mRNA pairing-a continuous helix of base pairs 3-7, as opposed to the 'ideal' seed sequence of miRNA nucleotides 2-8-there is potential for a long continuous helix formed between Scr and the 39 end of the miR-10-5p (bases 11-22) ( Figure 1D and Figure S5), suggesting formation of an energetically stable duplex with miR-10-5p primed RISC. While not a top hit in most computational predictions due to the shortened seed match, Scr has been previously suggested as a target for miR-10-5p [30]. Analysis of the putative Scr 39UTR sequences from other insects indicates that the predicted interacting bases are extremely well conserved in nearly all species examined, despite the fact that the neighboring sequences have diverged ( Figure S5), suggesting specific selection since the evolutionary split between the major insect groups more than 300 million years ago. In fact, the putative target sites from most of the non-Drosophilid insects display even higher complementarity to the seed region, which could provide added strength to the potential interaction. We also used the RNAHybrid program [53], and our analysis identified this Scr sequence as one of the strongest putative targets for miR-10-5p within all Drosophila 39UTRs (data not shown). Although miRNA target prediction algorithms have had varying success at prediction of true positive miRNA-mRNA interactions [54], its status as the top hit in our predictions, along with the conservation in arthropods, suggests that Scr is one of the best putative targets of miR-10-5p in Drosophila.
We also identified a putative target site for miR-10-3p in the 39UTR of the Hox gene Abdominal-B (Abd-B) ( Figure 1E). Similar to the putative target site for miR-10-5p in the Scr 39UTR, this sequence appears as one of our top hits among all Drosophila 39UTRs using RNAHybrid, contains potential long continuous helix pairings with the miRNA, and is well conserved in the putative 39UTR sequences of most insect Abd-B orthologs ( Figure 1E and data not shown). This site is also conserved within a conserved block sequence surrounded by non-conserved 39 UTR sequence among fly Abd-B orthologs ( Figure S6). Additionally, among our top predicted miR-10-3p targets, this is the only site with significant evolutionary conservation (data not shown). This target site for miR-10-3p in the Abd-B mRNA was also predicted in a previous study [55].
Functional analysis of miR-10-5p miRNAs may function in a number of ways to downregulate their target genes. Most initial studies indicated that miRNA targeted genes were downregulated by inhibition of protein production without affecting the levels of cytoplasmic mRNA [56], suggesting one could look for discrepancies between mRNA and protein levels for a given gene in order to identify regions of potential miRNA regulation. While there have been no published examples of post-transcriptional regulation of Scr during Drosophila embryogenesis, there is apparent post-transcriptional regulation of the Porcelio scaber Scr ortholog in the maxilliped segment during embryogenesis [28]. This led us to reexamine the expression pattern of Scr transcripts and protein more closely.
The expression of mRNA and protein products from the Scr gene overlap, for the most part, in all tissues during embryogenesis. However, in germband extended embryos, there is field of cells in the ventral first thoracic segment which transcribes Scr and contains low levels of Scr cytoplasmic mRNA, but that never exhibits any detectable SCR protein accumulation ( Figure S7). During this stage, pri-miR-10 is not actively transcribed in all of these cells, but it seemed possible that miR-10-5p produced from early transcripts might still be present in these cells to inhibit translation. However, LNA antisense oligo staining did not detect any mature miR-10-5p in these cells at this stage (data not shown). Furthermore, in situ analysis using a probe flanking the 39 of the miR-10 hairpin region suggests that pri-miR-10 transcripts that include a polyadenylation signal are not made in the ectoderm of early blastoderm embryos (data not shown). Taken together these data suggest that, although Scr appears to be post-transcriptionally regulated in these cells, it is not through the activity of miR-10-5p.
No other regions exhibiting discrepancies between Scr transcripts and protein were found during Drosophila embryogenesis. The other possibility was that miR-10-5p might be acting through Scr mRNA degradation as has been shown for a number of miRNA targets [57,58]. If this were the case, then transcript pattern and/or levels would be altered in embryos ectopically expressing or lacking the mir-10 gene. Unfortunately, available chromosomal deletions which lack mir-10 sequences also delete Scr, precluding the possibility of ascertaining the change in Scr expression in these mutant embryos.
In order to test whether Scr expression is altered in embryos that ectopically express mir-10, transgenic flies were constructed that allow expression of pri-mir-10 under the control of a variety of spatially and temporally specific GAL4 drivers. The pri-mir-10 transgene can produce significant amounts of mature miRNA, as visualized by antisense LNA oligo staining in embryos containing both the UAS-mir-10 transgene as well as a prd-GAL4 driver ( Figure S8A).
Ectopic expression of pri-mir-10 by GAL4 drivers that persistently produce pri-mir-10 in all embryonic cells results in death during early larval development and minor cuticle phenotypes in the anterior of the early larvae. The transgenic larvae die typically during the first instar and occasionally do not hatch, but none survive to adulthood. These larvae have malformations of their anterior cuticle which appear as cuticular breaks near the mouthhooks (compare Figure S8C-D with Figure  S8B). Although these cuticle phenotypes are reproducible, they do not obviously correlate with known embryonic loss of function Scr mutant phenotypes, which include reduction of the first thoracic beard structure, and labial sense organ abnormalities [59,60].
Scr and pri-mir-10 are expressed in largely complementary patterns in both early embryonic ectoderm ( Figure 3A-C) and in late stage CNS ( Figure 3H-J). This correlation is consistent with the hypothesis that miR-10-5p is negatively regulating Scr mRNA levels, but ectopic expression of pri-mir-10 did not provide any support for this hypothesis. The engrailed-GAL4 driver expresses in a small number of cells which overlap the Scr expression pattern. When pri-mir-10 is expressed under the control of engrailed-GAL4, these cells have no obvious decrease in SCR protein levels when compared to wildtype controls (compare Figure 4C-D with Figure 4A-B). Additionally, embryos ectopically expressing primir-10 from a variety of other GAL4 drivers did not exhibit detectable changes in either pattern or levels of Scr transcript or protein accumulation (data not shown).

Functional analysis of miR-10-3p
In order to characterize a possible interaction between miR-10-3p and Abd-B, we used in situ hybridization to determine their relative patterns of expression. During embryogenesis pri-miR-10 and Abd-B are expressed in mostly non-overlapping regions. While Abd-B is expressed in the majority of the posterior ectoderm, it is completely excluded from the developing anal pads, where pri-miR-10 and miR-10-3p are strongly expressed ( Figure 2G-H; Figure 3D-F and H-J). Abd-B is also expressed throughout later embryogenesis in the visceral mesoderm surrounding the large intestine of the hindgut, whereas miR-10-3p is expressed only in the endodermal cells of this region ( Figure 3G). This complementarity is suggestive of either unidirectional or mutual negative regulatory interactions. However, embryos which are mutant for Abd-B gene products do not show altered expression of pri-mir-10 (data not shown). Conversely, embryos which are deleted for the mir-10 gene (Df(3R)CP1), and thus lacking miR-10-3p [46], do not exhibit ectopic expression of Abd-B mRNA or protein in either the large intestine endoderm or in the anal pads, suggesting that miR-10-3p does not regulate Abd-B in these tissues (data not shown).
Although their transcription is largely complementary in most tissues, in the ventral nerve cord of later stage embryos, pri-mir-10 and Abd-B overlap from the posterior compartment of segment A5 through A8/9 (parasegments 10-14) ( Figure 3D-F and H-J). The Abd-B gene produces multiple transcript variants which code for two different proteins. Of these, the shorter ABD-B r-type protein is produced in the CNS only in the most posterior regions (A8/9), whereas the longer m-type protein is expressed in more anterior regions of the CNS (A5-A8) [61]. Additionally, Abd-B may utilize either of two polyA signals to terminate transcripts [62]. The longer Abd-B 39UTR contains the putative miR-10-3p target site, while the shorter 39UTR lacks the target site. Most tissues only produce Abd-B transcripts with the shorter 39UTR, while the longer 39UTR is only found on Abd-B transcripts in the CNS ( [63], and data not shown). Although it is not known whether the longer UTR belongs exclusively to transcripts that encode either ABD-B m-type or r-type proteins, the longer UTR has been found on transcripts that encode ABD-B-r [62]. Since the domain of pri-mir-10 transcription extends into the posterior of the CNS, but appears to decrease toward the extreme posterior ( Figure 3H, J), we hypothesized that miR-10-3p might be repressing r-type ABD-B protein in more anterior segments of the CNS. To test this possibility, Abd-B D14 (an mutant that eliminates only ABD-B mtype protein) was recombined with Df(3R)CP1. This strain  Figure 4L). This suggests that the longer 39UTR staining detected in A5-A7 are found on transcripts encoding ABD-B m-type protein.
To test the possibility that miR-10-3p is regulating Abd-B expression in the CNS more generally, transheterozygote Df(3R)CP1: Df(3R)LIN embryos deficient for mir-10 were analyzed for alterations in Abd-B transcript and protein levels in segments A5-A8/9 of the CNS. No consistent differences were seen in the expression levels or anterior extent of expression of either Abd-B transcripts (data not shown) or protein ( Figure 4K) when compared to wildtype embryos ( Figure 4I), although in A5-6, occasional mutant embryos showed slightly higher levels of ABD-B protein staining. Ectopic expression of pri-mir-10 in the CNS either through ubiquitous overexpression ( Figure 4J) or in a subset of cells ( Figure 4G, H) did not result in obviously reduced levels of ABD-B protein compared to wildtype (Compare to Figure 4E, F, and I).

bithorax complex miRNAs and ANTP regulation
In Drosophila, the bithorax complex (BX-C) is a large genomic region that is responsible for patterning much of the abdominal body region [64]. Although early genetic studies of the BX-C found that it contained only the protein-coding genes Ubx, adb-A, and Abd-B [65,66], large portions of the intergenic regions are also transcribed, giving rise to numerous long non-coding RNAs [67]. While the exact function of these RNAs is unclear, it is believed that early transcription through these regions is necessary to activate cryptic Hox enhancers, and primes the complex for later epigenetic regulation [68][69][70].
These miRNAs have been shown several times to be present throughout embryogenesis via cloning and Northern analyses [37][38][39], and have all been predicted to regulate one or more of the Hox genes. A number of recent studies have presented evidence supporting several of these predicted interactions: miR-iab-4-5p:Ubx [36,39], miR-iab-8-5p:Ubx [37][38][39], and miR-iab-8-5p:abd-A [38,39]. The best genetic evidence for a biologically relevant functional interaction is for miR-iab-8-5p as a negative regulator of UBX protein levels in a subset of CNS cells in the 8th abdominal segment [37].
In addition to these interactions, it has also been predicted that miR-iab-4-5p or miR-iab-8-5p might also bind to the 39UTR of the Hox gene Antp [38,39]. Through our own analysis we identified a strong putative binding site for miR-iab-4-5p in the 39UTR of Antp. This site was not found in previous predictions due to its lack of canonical seed pairing. However, this site does have prototypical miRNA/mRNA 59 and 39 helical pairing, and similar putative target sequences can also be found downstream of Antp orthologs from many other insects ( Figure 1F). Although the conservation of this sequence is not as strong as the predicted miR-10-5p target in Scr orthologs ( Figure S5), the Antp sequences that would pair with miR-iab-4-5p are completely conserved amongst Drosophila species ( Figure S9).

Functional analysis of miR-iab-4-5p and Antp
To investigate a possible interaction between Antp and miR-iab-4-5p, we performed in situ hybridizations on Drosophila embryos with probes against the pri-mir-iab-4 (iab-4) and Antp transcripts ( Figure 1C, and Figure 5A-D). During early embryogenesis Antp and iab-4 are expressed in broad, partially overlapping patterns throughout the future trunk and abdomen ( Figure 5A). However, it is unlikely that co-expression at this stage is informative, as the Antp transcription unit is quite long and does not have time to produce full-length mRNAs during the short nuclear division cycles of the blastoderm [71]. During germband extension and retraction these patterns refine, with the major domain of coexpression corresponding to the CNS of abdominal segments 3 through 7 ( Figure 5B-D). It is also interesting to note that the transcription pattern of Antp seems 'noisier' than that of other Hox genes, as low numbers of transcripts are often observed in the ectoderm of the abdomen in variable patterns ( Figure 5C, arrows). As these never seem to correlate with detectable ANTP protein levels (data not shown), it is possible that one of the functions of the iab-4 miRNAs might be to repress translation of these apparently spurious Antp mRNAs. Unfortunately, the rare and stochastic nature of these transcripts does not allow any firm conclusions to be drawn concerning a role for miR-iab-4-5p in the suppression of ectopic protein production.
The main region of co-expression between ANTP protein and iab-4 is in the CNS of abdominal segments 3-7 of later stage embryos ( Figure 5E-F). Looking closely at the expression patterns, we see that in regions of overlapping expression, most cells with high levels of iab-4 expression have low levels of ANTP protein, which is at least consistent with potential miR-iab-4 regulation of ANTP levels ( Figure 5E-F).
To determine whether significant regulation occurs in vivo in the region of overlapping expression, we took advantage of a Drosophila strain in which the mir-iab-4/8 hairpin has been precisely deleted [37]. Conveniently, as the rest of the primary transcript is still produced as normal in these mutants, we can still stain for the iab-4 precursor RNA to pinpoint areas that might now contain higher ANTP levels. However, no obvious reproducible changes in ANTP protein levels were seen in the co-expressed regions throughout embryogenesis (compare Figure 5G-I to Figure 5J-L). Occasionally, mir-iab-4/8 embryos would be observed in which certain subsets of neuroblasts displayed higher ANTP levels in areas of iab-4 co-expression (data not shown). However, the rarity of these events prevented rigorous quantification of this phenomenon.
Finally, to determine whether the function of miR-iab-4-5p was to buffer ANTP levels against environmental fluctuations [72][73][74], we subjected developing wildtype and miR-iab-4/8 mutant embryos to several rounds of temperature shifts (see methods). Again, we saw no reproducible changes in ANTP protein levels in miR-iab-4/8 mutants versus wild type, even under these stressful developmental conditions (data not shown).

Discussion
A growing body of circumstantial evidence and computational predictions have consistently pointed to Drosophila Hox genes as a significant class of miRNA targets. Our results do not support the idea that Drosophila Hox complex miRNAs play a significant embryonic patterning role through the regulation of their predicted targets in Hox protein-coding messenger RNAs. This is consistent with previous loss-of-function experiments on the role of the mir-iab-4/8 genes, which indicated that regulation of a Hox target gene (Ubx) by miR-iab-8-5p led to only subtle changes in the UBX protein expression pattern in embryos, and no detectable changes in axial body patterning [37]. In C. elegans, the study of mir-57, which is likely to be a derived ortholog of mir-10, indicates that loss-of-function mutants have partially penetrant posterior patterning defects that may in part be due to the derepression of nob-1b, a C. elegans ortholog of Abd-B [41]. It is possible in vertebrates, that Hox complex miRNAs play a greater role in body patterning, as their functional knockdown using antagonistic complementary nucleic acids have led to dramatic changes in Hox protein-coding gene expression and body patterning defects [34,35,48]. If the predicted Hox miRNA:Hox RNA target interactions in Drosophila are functional in vivo, why are they so difficult to observe? Both the miRNAs and their target sites have been conserved over vast evolutionary time in insects, so they must be under strong selective pressure.
Based on the current evidence, we believe that the evolutionarily conserved function of Hox miRNAs in bilaterians is most likely to subtly modulate Hox protein levels in sub-regions of the CNS. In addition to a few examples in Drosophila based on loss-offunction miRNA mutants [37,63], there is compelling evolutionary data to support this theory. In addition to being conserved in both sequence and genomic position, mir-10 appears to have a conserved expression pattern in the CNS of Drosophila and zebrafish [47,48]. Members of the vertebrate specific Hox miRNA family, mir-196, are also expressed in the CNS of zebrafish [35,47] and chicken [75], and mir-iab-4 and mir-iab-8 (while not orthologous by sequence, but orthologous by chromosomal position in the Hox complex) also are expressed in the Drosophila CNS ( [36][37][38][39], and Figure 5). Additionally mir-993 is also expressed in the Drosophila CNS from around the subesophageal ganglion to the posterior abdominal segments (data not shown).
When considering miRNA regulation of Hox genes in the CNS, an added layer of complexity must also be considered. Ubx, Antp, abd-A, and Abd-B all appear to produce both short and long forms of their 39UTRs in various sub-regions of their expression patterns [63]. This could result in a variable mixture of miRNA sensitive and insensitive Hox transcripts, which might be utilized by cells to modulate protein levels in the presence of static miRNA expression. Alternative UTR usage with and without miRNA target sites may be a mechanism utilized by many tissues and cells in general [16]. While very intriguing, this further complicates efforts to characterize endogenous miRNA:target interactions, as down-regulation may be masked by translation from miRNA insensitive transcripts.
Another scenario is that the major function of the Hox miRNAs, many of which have expression patterns that are complementary to their predicted target Hox mRNAs, function largely to suppress aberrant Hox protein expression outside of their normal functional domains. Most 'validations' of miRNA:-Hox interactions have only demonstrated that misexpression of Drosophila Hox miRNAs in regions where they are not normally found can result in homeotic transformations [36,38,39]. Bender [37] showed that while the function of miR-iab-4-5p and -3p appear dispensable for normal development, only the loss of miRiab-8-5p led to male sterility, possibly due to the expansion of UBX protein expression into some cells in the most posterior segments where it is not normally expressed. Evidence has indicated that many miRNAs and their most strongly predicted targets are often expressed in mutually exclusive patterns [10], which is consistent with what we observed for miR-10-5p:Scr and miR-10-3p:Abd-B, limiting the potential for regulation of Hox protein expression during embryonic development.
Although we were unable to detect evidence for canalizing activity of the Hox miRNAs we tested, our results and others are consistent with the hypothesis that one major role of miRNAs is to canalize expression of their target genes in order to achieve precise reproducible spatial limits and levels of expression in order to maintain a reproducible phenotypic output in spite of environmental fluctuations [72,74]. This attractive, yet difficult to demonstrate hypothesis has some experimental support [41,73]. Our own attempts at inducing aberrant ANTP protein accumulation in iab-4/8 mutants by subjecting embryos to temperature fluctuations were unsuccessful, although recapitulating all the stresses encountered by a developing Drosophila embryo in the wild is difficult to exhaustively test in a laboratory context. Even if canalization was the sole function of many miRNAs, it could account for the observed conservation of both miRNAs and their target sites throughout evolution. It is also possible that in the lineage leading to Drosophila, Hox miRNA functions have become less important in early developmental patterning than in other insect lineages, although important enough to largely conserve their target site sequences in Hox protein coding mRNAs. miRNA genes have been residents of animal Hox clusters for more than 500 million years, and it is conceivable that the gain or loss of Hox miRNA target sites, or changes in spatiotemporal expression of Hox-targeting miRNAs could have played an important role in the evolution of Hox gene function during animal evolution.

Materials and Methods
Cuticle preparations were performed as in [76]. Northern blots were performed as in [77]. Standard in situ hybridizations and immunohistochemistry were performed as in ( [46]; http://biology. ucsd.edu/,davek), except when simultaneously detecting RNA and protein, acetone was substituted for Proteinase K for permeabilization, to preserve epitope integrity [78,79]. LNA in situ hybridizations were performed as in [80], modified for Drosophila with embryo preparation as in [46], using 59 DIG labeled miRCURY LNA TM TM probes from Exiqon.

Antibodies
Mouse primary antibodies against Hox proteins were obtained as concentrates from the Developmental Studies Hybridoma Bank and used at 1:300 dilutions: SCR (6H4.1), ABD-B (1A2E9), and ANTP (4C3). The EN antibody was a gift from Patrick O'Farrell (1:300 dilution). The SCR antibody used in Figure S7 for simultaneous detection with in situ probes was a gift from Debbie Andrews (rabbit polyclonal anti-SCR; 1:100 dilution) Primary antibodies against haptens were obtained from Roche, and used at 1:800 dilutions: mouse anti-BIO; sheep anti-DIG; rabbit anti-DNP; guinea pig anti-FITC; Alkaline Phosphatase (AP) conjugated sheep anti-DIG; and Horseradish Peroxidase (HRP) conjugated sheep anti-DIG. Alexa Fluor labeled secondary antibodies were obtained from Molecular Probes/Life Technologies were used at 1:400 dilutions: donkey antisheep Alexa647; donkey anti-mouse Alexa488; donkey anti-rabbit Alexa555; and goat anti-guinea-pig Alexa594.
To visualize ANTP protein levels in DmiR-iab-4/8/TM3,Ubx-LacZ embryos, the detection scheme was as follows: iab-4-DNP.rabbit anti-DNP.donkey anti-rabbit Alexa555; mouse anti-ANTP.donkey anti-mouse Alexa488; LacZ-FITC.guinea pig anti-FITC.goat anti-guinea-pig Alexa594. LacZ staining was used to differentiate heterozygous and homozygous miRNA deletion embryos.  Figure S8 Mature miRNA is produced from transgenes and results in aberrant phenotypes. When expressed from a prd-GAL4 driver (prd.pri-mir-10), cytoplasmic miR-10-5p is detectable in the prd pattern at high levels in a germband extended embryo (A). When expressed from an Actin-GAL4 (Act5C.primir-10) (C) or da-GAL4 driver (da.pri-mir-10) (D) malformations are noticeable in the head cuticle (arrows) when compared to wild type (B). Two small black dots (sensory organs) can be seen in the dorsal anterior head in some cuticular preparations, but not others. This is not because of an actual sensory organ duplication, but due to differential flattening of the cuticles during their preparation, so that in some cuticles the sensory organs from both sides of the anterior dorsal head can be seen.