Deletion of morpholino binding sites (DeMOBS) to assess specificity of morphant phenotypes

Two complimentary approaches are widely used to study gene function in zebrafish: induction of genetic mutations, usually using targeted nucleases such as CRISPR/Cas9, and suppression of gene expression, typically using Morpholino oligomers. Neither method is perfect. Morpholinos (MOs) sometimes produce off-target or toxicity-related effects that can be mistaken for true phenotypes. Conversely, genetic mutants can be subject to compensation, or may fail to yield a null phenotype due to leakiness (e.g. use of cryptic splice sites or downstream AUGs). When discrepancy between mutant and morpholino-induced (morphant) phenotypes is observed, experimental validation of such phenotypes becomes very labor intensive. We have developed a simple genetic method to differentiate between genuine morphant phenotypes and those produced due to off-target effects. We speculated that indels within 5′ untranslated regions would be unlikely to have a significant negative effect on gene expression. Mutations induced within a MO target site would result in a Morpholino-refractive allele thus suppressing true MO phenotypes whilst non-specific phenotypes would remain. We tested this hypothesis on one gene with an exclusively zygotic function, tbx5a, and one gene with strong maternal effect, ctnnb2. We found that indels within the Morpholino binding site are indeed able to suppress both zygotic and maternal morphant phenotypes. We also observed that the ability of such indels to suppress morpholino phenotypes does depend on the size and the location of the deletion. Nonetheless, mutating the morpholino binding sites in both maternal and zygotic genes can ascertain the specificity of morphant phenotypes.

It is not uncommon for knockdown-and knockout-based approaches to yield different results [15][16][17][18] . Such discrepancies have also been observed in zebrafish [19][20][21][22][23] . Sometimes they can be attributed to built-in shortcomings of each approach. Mutant alleles may exhibit leakiness: small indels do not always result in complete lossof-function due to a variety of phenomena including splicing artifacts and translation initiation at downstream AUGs leading to production of a functional protein and masking the null phenotype [24][25][26] . A further complication in the analysis of mutant phenotypes arises from the fact that for some genes, maternally-deposited mRNAs (and proteins) partly mask mutant phenotypes necessitating the use of maternal-zygotic mutants 27,28 . Additionally, some frameshift and nonsense mutants induce transcriptional compensation by closely related genes 22,29,30 . Deleting the whole coding sequence appears to be the best way to eliminate these possibilities. However, regulatory complexity of vertebrate genomes raises the possibility that the observed phenotype may be caused by deletion of intron-residing cis-regulatory elements for other genes (for an example, see 31,32 ).
With the notable exception of short upstream reading frames 33 , 5′ UTRs appear to be sparse in significant regulatory features. We speculated that indel mutations within 5′ UTRs are unlikely to significantly impair the expression of the downstream gene 34 . Indels introduced within a morpholino target site should reduce, if not entirely abolish, MO binding making the "mutant mRNA" partly or completely refractive to morpholino activity. We further hypothesized that since few genes are haploinsufficient, heterozygosity for such MO-refractive mutations would be sufficient to suppress specific morpholino phenotypes. Using tbx5a and ctnnb2 as test loci, we demonstrate that deletions can be readily generated and used to test the specificity of both zygotic and maternal morphant phenotypes.
Since tbx5a mRNA is not contributed maternally, outcross of a parent heterozygous for a potentially MO-refractive mutation would produce a clutch of embryos where half would be genotypically wild type and susceptible to the MO, while the other half would be refractive to the MO. Susceptible and refractive embryos, present within a single clutch, would serve as controls for each other, eliminating experimental bias by excluding variables such as active MO concentration, injection volume or timing of the injection.
Two S. pyogenes PAM (protospacer adjacent motif) sites are present within the 5′ UTR sequence targeted by tbx5a-MO4 39 (Fig. 1B Table 2) required for transcription initiation by the T7 RNA polymerase. Guide RNAs were injected along with nCas9n mRNA as previously described 34,40 . PCR fragments amplified from lysates of 20 pooled 3 day post fertilization (dpf) injected embryos were analyzed for sgRNA efficiency by TIDE 41 and Synthego ICE 42 . Both analyses showed that tbx5adeMO2 sgRNA (~ 30% by TIDE, ~ 11% by ICE) was more efficient than tbx5adeMO1 sgRNA (~ 17% by TIDE, ~ 7% by ICE) (Supplementary Fig. 1). We raised embryos injected with tbx5adeMO2 sgRNA and nCas9n and screened three F0 fish for germline transmission of indels using the T7 endonuclease assay (data not shown). One founder produced a high percentage of progeny with indels, and one F1 family was raised. Four out of seven genotyped adult F1s were found to be heterozygous for indels: two for a (− 3) deletion and two for a (− 7) deletion by Poly Peak Parser 43 analysis. F1s heterozygous for (− 3) and (− 7) deletions were incrossed. All embryos were phenotypically normal, indicating that these deletions do not significantly impair the expression of tbx5a. Sequence of the (− 3) and (− 7) deletions was confirmed on homozygous F2s (Fig. 1C).
To determine the effective dose of Tbx5a-MO4, we injected 2, 4, 8 and 12 ng of the morpholino into one-cell zebrafish embryos (Fig. 1D). The lowest dose of the MO resulted in > 90% of embryos displaying pectoral fin defects. In contrast, a much higher 8 ng dose of the MO was needed to elicit severe cardiac defects (> 90% edema). Notably, in humans suffering from Holt-Oram syndrome caused by mutations in Tbx5, forelimb defects are also more penetrant and severe than cardiac defects [44][45][46] .
At the high 8 ng MO dose, all MO-injected embryos from the (− 3) outcrosses displayed identical morphant phenotypes (Fig. 1E). In contrast, approximately 50% of embryos from the (− 7) outcross were completely rescued from the cardiac edema phenotype, but not the pectoral fin defect (Fig. 1E). Embryos from the (− 7) outcross were grouped by phenotype and subsequently genotyped. Genotyping revealed that 14/15 (93%) of individuals with cardiac edema and loss of pectoral fins were wild-type and 13/16 (81%) of individuals with no cardiac edema were heterozygous for the (− 7) allele (P = 3.0E−05) ( Supplementary Fig. 2). These results indicate rescue Introduction of a MO-refractive mutation into mRNA has the potential to exacerbate off-target effects. In heterozygous embryos, only 50% of target mRNA can be bound by the MO, thus increasing the concentration of morpholino available for off-target binding. The fact that we did not observe any new phenotypes in heterozygous embryos further supports the observation that Tbx5a-MO4 is highly specific.
Dose-, phenotype-and deletion size-dependent rescue of morphant phenotypes prompted us to hypothesize that (− 3) and (− 7) deletions were insufficient to make mRNAs entirely refractive to the morpholino. We cloned wild type, (− 3) and (− 7) target sites into the pT3TS in vitro transcription vector 47 ahead of eGFP coding sequence. In vitro transcribed mRNAs were injected into embryos along with pT3TS:mRFP mRNA as a control not affected by the MO. Half of the mRNA-injected embryos were then injected with 8 ng of Tbx5a-MO4. Embryos were scored for RFP and GFP fluorescence and photographed at 1 dpf (Fig. 1F). We found that Tbx5a-MO4 was able to almost entirely block translation of mRNAs containing wild type and (− 3) target sites. Translation of mRNA containing the (− 7) target site was also reduced significantly (Fig. 1F). These findings Our data therefore indicate that a 3′ stretch of identity as short 15-16 nucleotides may be sufficient for a MO to impede translation at a high dose, and suggests that using MOs shorter than the current 25 nucleotide standard may lead to higher specificity.

Suppression of the β-catenin morphant phenotype by maternal contribution of (− 4) binding
site mutant ctnnb2 mRNA. We speculated that for maternally contributed genes, a female heterozygous for a MO-refractive allele would produce embryos which would be refractive to the maternal-zygotic phenotype.
To test this hypothesis, we selected beta-catenin genes coding for an essential component of the Wnt signaling pathway. Co-injection of translation-blocking MOs targeting the duplicated ctnnb1 and ctnnb2 mRNAs results in complete loss of ventral cell fates and a phenotype named ciuffo (Supplementary Table 1) 48 . Partial sequencing of ctnnb2 loci in the TLF genetic background revealed presence of a single nucleotide polymorphism within the ctnnb2-MO1 binding site. Co-injection of ctnnb1-MO2 and ctnnb2-MO1 into TLF embryos at concentrations described previously 48 resulted in nearly 100% penetrant ciuffo phenotype (data not shown), indicating that the polymorphism alone does not appreciably reduce morpholino activity. Two sgRNAs targeting PAM sequences within the ctnnb2-MO1 binding site were designed and tested ( Fig. 2A,  Supplementary Table 2). Only ctnnb2deMO1 had detectable activity by both TIDE (~ 17%) and Synthego ICE (~ 5%) analysis ( Supplementary Fig. 3). Three adult F0 fish injected with ctnnb2deMO1 sgRNA and nCas9n mRNA were tested for germline transmission of indels, leading to establishment of one F1 family. Fourteen F1 fish were tested for loss of Bpu10I restriction enzyme site, and seven were found to be heterozygous for indels. PCR fragments from five F1s were sequenced, and four were found to be heterozygous for a (− 4) deletion (Fig. 2B). A male heterozygous for the (− 4) deletion was outcrossed to establish an F2 family.
To avoid experimental bias, we performed a blind experiment to test for suppression of ciuffo phenotype. From a single F2 family, we identified two adult females heterozygous for the deletion and two female wild type siblings. Fish were coded A, B, C and D, and embryos obtained from outcrosses were injected with a mixture of ctnnb1-MO2 and ctnnb2-MO1. MO injection lead to high penetrance (89% and 100%) of ciuffo phenotypes in www.nature.com/scientificreports/ the progeny of wild type fish, while presence of one (− 4) allele in heterozygotes almost completely suppressed the ciuffo phenotype (4% and 13%, Fig. 2C). Milder phenotypes were still observed in a subset of embryos ( Fig. 2D-H), which could reflect zygotic requirement for ctnnb1 and/or ctnnb2 49,50 , or off-target effects due to a high cumulative dose of the two MOs. Nonetheless, we observed that females heterozygous for a (− 4) deletion in combination with a serendipitous single nucleotide polymorphism produce embryos which are nearly 100% suppressed for the ciuffo phenotype. These results indicate that our method can be used to ascertain morphant phenotypes of genes with strong maternal contribution of mRNA. Our data clearly demonstrates that indels within MO binding sites can be readily generated and used to test the specificity of both zygotic and maternal morphant phenotypes. The ability to induce deletions within MO binding sites using CRISPR/Cas9 relies on the presence of a PAM site within the MO binding site, preferably close to the 5′ end of the target site. While the mutagenesis method employed by us is likely feasible for the majority of MOs targeting 5′ UTRs, there will inevitably be a subset where PAM sites will be absent or located closer to the middle or the 3′ of the target site. In such cases, oligonucleotide-mediated repair of double strand breaks can be used to engineer desired mutations 34,40,[51][52][53][54][55][56] .
Inconsistency between knock-out and knock-down phenotypes poses a significant challenge for researchers studying gene function. This inconsistency can be attributed to morpholino off-target effects, incomplete loss-of-function in the knockout, genetic (transcriptional) compensation, or a combination of these factors. For scenarios where the observed knockdown phenotypes are more severe than those seen in genetic mutants, our method offers a fairly quick and cost-effective way to test if the morphant phenotypes are indeed specific.
We selected tbx5a and ctnnb2 for this proof-of-principle study due high degree of confidence we had in respective mutant and morphant phenotypes. These two genes can be taken as models for a broad spectrum of zebrafish genes: tbx5a has an exclusively zygotic function while ctnnb2 mRNA is maternally contributed. Furthermore, ctnnb2 is functionally duplicated necessitating simultaneous knockdown of ctnnb1 in order to observe a phenotype. The ability to rescue both tbx5a and ctnnb2 morphants offers a level of confidence that this method can be used for a large subset of other genes that are not as well-characterized.

Materials and methods
CRISPR/Cas9 mutagenesis. Guide RNAs were produced as previously described 34,40 using DR274 57 as the template and diluted to ~ 60 ng/μL. Immediately prior to injection, 8 uL of diluted sgRNA was mixed with 2 μL aliquot of 150 ng/μL nCas9n mRNA 58 to the final volume of 10 μL.
Plasmid construction and mRNA synthesis. Details of plasmid construction are available upon request. eGFP-containing pT3TS 47 vectors [pCMC23 (wt MO binding site), pCMC24 (− 3) and pCMC25 (− 7)] were linearized using XbaI restriction enzyme. Template for the synthesis of mRFP mRNA was amplified by PCR using M13F/M13R primer pair on pT3TS:mRFP (pDB935). Templates were transcribed using T3 mMessage mMachine kit and mRNAs were purified using Qiagen RNeasy MinElute kit. mRNAs were diluted so that the standard 3 nL injection volume would contain 50 ng of tbx5a-eGFP mRNA and 100 ng of mRFP mRNA.
Microinjection. Microinjection volumes were calibrated to 3 nL as previously described, and all microinjections were performed into the yolks of 1-cell zebrafish embryos as described 59 . Ethics approval. All