In the initial turtle gene publication, the exon-intron map provided by the authors indicates that none of the generated deficiencies removes all turtle transcripts 14. The potentially diffusible isoforms encoded by expressed sequence tag (EST) clones AT02763 and GH15753 near the 5'-end of the gene are still present even in the largest deletion, tutl⁴. With this in mind, we mobilized the tutl^k14703 P-element insertions to generate null mutation of the turtle gene (Figure 1A). A newly generated allele, tutl^ex383, is embryonic lethal and fails to complement the adult lethality of the previously identified turtle mutant allele tutl⁰¹⁰⁸⁵. Southern blotting, PCR, and DNA sequencing indicate that this allele contains an 11,430 base pair (bp) deficiency within the turtle gene that should disrupt all transcripts (Additional file 1). Staining with a variety of cell fate markers indicates that the embryonic lethality of tutl^ex383 is not due to a defect in cell fate induction (Additional file 2).

A Blast search using turtle genomic sequence indicated that the gene encodes five different isoforms. The isoforms encoded by EST clones AT02763 and GH157753 do not contain any hydrophobic membrane-spanning domains, suggesting that they may be diffusible proteins, while the isoforms encoded by the EST clones HL010565, LD28824, and GH08133 have a membrane-spanning domain, indicating that they are membrane bound.

To determine whether the proteins encoded by AT02763 and GH157753 are secreted, we transiently transfected Drosophila Schneider 2 (S2) cells with carboxy-terminal His-tagged versions of these transcripts. Western blots against the His tag indicate that most of the tagged proteins were present in the medium fraction and not in the cell pellet, confirming that they are secreted (Figure 1B).

Using isoform-specific primers and RT-PCR, we detected the expression of all the turtle isoforms in embryos during the start of neuronal axonogenesis (embryonic stages 12 to 13; Figure 1C). However, the expression levels of AT02763 and HL010565 were much lower than those of GH157753, LD28824, and GH08133.

We made anti-sense Dig-labeled RNA in situ hybridization probes against the exons encoding the fifth immunoglobulin domain and the first Fibronectin III domain of turtle. During stages 12 and 13, when neurons begin to project axons, expression was initially restricted to a segmentally repeated cluster of cells surrounding the midline (Figure 1D). Later, expression spread throughout the central nervous system (Figure 1E).

We also generated RNA probes against sequences that are unique to each isoform. Each probe type was labeled with a different fluorescent tag. RNA in situ hybridization with GH15753, LD28824, and GH08133 isoforms in wild-type embryos indicated strong, overlapping nervous system expression during stages 13 to 14 (Figure 1F, arrow). We also observed overlapping expression of LD28824 and GH08133 in certain areas of the gut and salivary glands (Figure 1F, arrowhead and asterisks, respectively). However, RNA in situ hybridization performed against the AT02763 and HL010565 isoforms did not produce a signal above background (data not shown), thus further confirming their low expression levels.

We were unable to detect fluorescence signal of any isoform in early stage 12 embryos. A possible explanation is that fluorescence probes are too short to give a detectable signal in stages when turtle levels are low; indeed, like other cell recognition molecules, turtle transcription might be up-regulated in later embryonic stages, further enhancing in situ detection.

To further clarify the expression pattern of the turtle isoforms within the nervous system, we performed RNA in situ hybridization with GH15753- or LD28824-specific probes on embryos stained with the neuronal marker anti-Elav or non-midline glia marker anti-Repo. Both isoforms are largely expressed in segmentally repeated neurons both near to and distant from the midline, with some minor expression in glia (Figure 1G-J).

In wandering third instar larvae, RNA in situ hybridization of imaginal discs and brains showed robust and partially overlapping expression of GH15753, LD28824, and GH08133. In the eye disc, GH15753, LD28824, and GH08133 are expressed within the morphogenetic furrow (Figure 1K, asterisks). However, in the tip of the eye disc, where the optic stalk is located, we observed a cluster of cells that are only GH15753-positive (Figure 1K, arrow). Those cells were surrounded by a group of cells that express only LD28824 (Figure 1K, arrowhead). In the brain, strong and overlapping expression of GH15753, LD28824, and GH08133 was observed in the optic lobe (Figure 1L).

However, in the region where the optic stalk contacts the optic lobe, we noticed a cluster of cells that express only LD28824 (Figure 1L inset, arrow).

In the late embryonic stages, most central nervous system axons are arranged in two longitudinal connectives that are linked in each body segment by two midline-crossing commissures. These structures can be easily visualized using the BP102 antibody (Figure 2A). We noticed that close to 10% of tutl^ex383 embryonic segments have gaps in the longitudinal connectives and commissures (Figure 2B, M), a defect that is typically observed in embryos that lack netrin or the genes for the Netrin receptor Frazzled 91019. A minor enhancement of this phenotype was observed in embryos that are tarns-heterozygous tutl^ex383 over a genomic deficiency Df(2L)ed-dp that removes the tutl gene, thus indicating that this mutant allele is almost a genomic null (Figure 2L, M).

The anti-Fasciclin (Fas)II antibody stains three longitudinal axon fascicles on each side of the midline (Figure 2F). In 60% of homozygous tutl^ex383 embryos, these axons are defasciculated, and some inappropriately project to the midline (Figure 2G). This phenotype is distinct from that of mutations that cause a reduction in midline repulsion, such as robo or slit. In robo mutants, the innermost fascicles remain intact but repeatedly crisscross the midline 21112, while in tutl^ex383 embryos the FasII tracks maintain their relative distance from the midline, but axons emanating from all three fascicles bundle together as they cross the midline (Figure 2G).

To confirm that the observed defects are due to loss of turtle function, we performed complementation tests and transgenic rescue. Embryos trans-heterozygous for tutl^ex383 and tutl⁰¹⁰⁸⁵ have defects in commissures, longitudinal connectives, and FasII-tracks that are similar to those of tutl^ex383 homozygotes. However, these defects occur with a lower frequency and severity than is observed in tutl^ex383 embryos (data not shown). Post-mitotic neuronal expression of any of the different turtle isoforms in a tutl^ex383 background, using either Elav-Gal4 or the midline cell driver singleminded-Gal4 (Sim-Gal4), fully rescues these midline crossing defects (Figure 2C, D, H, I). Furthermore, transgenic expression using the pan-neuronal Elav-Gal4 driver of the diffusible isoform encoded by the EST clone AT02763 on a tutl^ex383 mutant background is sufficient to rescue adult lethality even though this isoform is weakly expressed in the wild type.

Both Elav-Gal4 and Sim-Gal4 are expressed close to or within the midline. If turtle expression near the midline is critical for its axonal midline crossing functions, then its expression in cells that are away from the midline, using Repo-Gal4, should not rescue the mutant phenotype. Indeed, mutant tutl^ex383 embryos that transgenically express turtle in only Repo-Gal4-positive glia do not show any rescue (Figure 2E, J). These results indicate that mutation in the turtle gene is indeed responsible for the midline crossing defects and lethality seen in tutl^ex383 embryos, and that turtle expression close to or within the midline is most important for this function.

We also performed pan-neuronal overexpression of the different isoforms in a wild-type background using a Sca-Gal4 driver. While no obvious phenotype resulted from overexpression of the membrane-bound isoforms, overexpression of the diffusible isoforms produces FasII-tracks that crisscross the midline in a manner reminiscent of robo mutants (Figure 2N), a phenotype that is typically interpreted as an increase in midline attraction 21112202122232425. The same phenotype was also produced when the diffusible isoforms were expressed only in the midline cells and not in the FasII-positive axons using Sim-Gal4 (Figure 2O), which indicates that the secreted turtle isoforms can produce their effects on midline axons via a non-cell autonomous mechanism.

The above results, combined with turtle expression near the midline during critical stages of axon development, implicate turtle in increasing axonal midline crossing. All isoforms are fully capable of rescuing the axon midline crossing defects in turtle mutants. However, the diffusible isoforms are the most potent as they are the only isoforms that induce a gain-of-function phenotype when overexpressed on a wild-type background.

Turtle proteins may promote midline attraction by one of three potential mechanisms. They may function as part of the netrin pathway, mediating netrin signaling. Alternatively, they may directly attract axons via their own unique pathway. Finally, they may function as inhibitors of the midline repellent slit pathway, which would then produce an indirect increase in the midline attraction signal.

Null mutations in essential genes in the same pathway do not modify each other's phenotype in double homozygous configurations, while genes that affect the same biological process but are part of different and redundant pathways do in both heterozygous and double homozygous combinations. If turtle directly facilitates netrin function, then reducing turtle levels in embryos with no netrin signaling should produce no additional modifications of those embryos' mutant phenotype, since there is no netrin signaling in the first place to be modified. However, if turtle functions in a parallel pathway that induces midline attraction via a netrin-independent mechanism, then reduction in turtle signaling should produce a synergistic effect that further enhances netrin or frazzled mutant phenotypes. The fly genome contains two netrin genes, netA and netB, and one frazzled gene (fra). The genomic deficiency Df(1)NP-5, which removes both netrin genes and the fra mutant allele fra^GA957, results in embryos that have fragmented commissures and gaps in longitudinal connectives thought to reflect a complete lack of netrin pathway signaling 910 (Figure 3A, D). The removal of even one copy of the turtle gene dramatically enhances the netrin or fra^GA957mutant phenotype (Figure 3B and 3E, respectively), while the complete removal of both turtle and netrin signaling in Df(1)NP-5; tutl^ex383 or fra^GA957, tutl^ex383 double homozygotes produces embryos that typically have only one thin and fragmented commissure along the entire length of the embryo (Figure 3C and 3F, respectively), a phenotype that bears similarities to that of the robo down-regulator comm 112. This synergistic enhancement of turtle mutation in embryos that lack netrin signaling indicates that turtle promotes midline attraction via a netrin-independent mechanism, and not as part of the netrin signaling pathway.

In Drosophila, the removal of the slit gene produces embryos with commissural axons that fail to exit the midline, while the loss of the slit receptor robo produces a less severe midline phenotype. This is due to the presence of two other robo-like slit receptors that can partially compensate for robo loss-of-function 1112021. Double heterozygous slit/+, robo/+ embryos have a sufficient drop in slit signaling to produce minor defects in midline crossing revealed by FasII staining 1112021 (Figure 3G). This phenotype can be modified when other components of the slit pathway are genetically removed 1112202122232425. If the turtle proteins promote midline crossing by functioning as a slitpathway inhibitor, then the reduction in turtle function should suppress the midline crossing defect seen in slit¹/+, robo⁵/+ embryos. We did not detect any modification of this defect in slit¹/+, robo⁵/+, tutl^ex383/+ embryos (Figure 3H, I), which makes it unlikely that turtle proteins functions as an inhibitor of the slit pathway.

The Abelson tyrosine kinase (abl) is a critical point of convergence for both attractive and repulsive midline signals received by growing axons. The integrative capacity of abl is due to its ability to physically link midline guidance cue receptors, such as the netrin receptor fra and the slit receptor robo, to actin cytoskeleton modulators such as ena and trio 21222324. The general effect of increased abl signaling on both netrin and slit pathways is to produce an increase in the ability of axons to cross the midline. Genetic null mutations in the abl gene alone, such as abl¹, produce only minor midline crossing defects. This might be due to the fact the abl protein is maternal loaded by abl¹ heterozygous mothers into their homozygous mutant progeny. However, the minor midline crossing defects observed in those embryos can be genetically modified if other components of the abl pathway are also reduced. We find that removing one copy of abl dominantly enhances the commissural defect seen in tutl^ex383 mutant embryos (Figure 3K). Furthermore, tutl^ex383; abl¹ double homozygous embryos have a commissural axon defect that is similar to that of netA,netB,tutl^ex383 and fra^GA957, tutl^ex383homozygous embryos (Figure 3L).

All of the above results indicate that the turtle proteins function as a midline attractant via a netrin-independent mechanism that does not involve the suppression of the slit pathway. Due to maternal loading of abl proteins, it is difficult to conclusively determine if turtle signaling functions in an abl-independent or -dependent manner. Nevertheless, these results do indicate that turtle pathway signaling is integrated with other guidance cue pathways to ultimately produce appropriate midline crossing behavior in growing axons.

In stage 16 wild-type embryos, the ventral body wall muscles are innervated by an invariant pattern of motor axons coming from ISNb, SNc, and ISNd nerves 326 (Figure 4A). In tutl^ex383 embryos, the ISNb motor axons successfully reach the vicinity of their respective targets. However, the majority of them fail to send one or more of the final axon branches to contact their muscle targets. Half of the hemisegments also lack ISNd nerves (Figure 4B, G). Both tutl⁰¹⁰⁸⁵ and tutl^k14703 mutant embryos exhibit a similar phenotype, albeit with a lower frequency (Figure 4G). Since the initial publication on turtle reported that there were no motor axon pathfinding defects, we conducted a blind test by an independent observer to confirm our results in tutl⁰¹⁰⁸⁵, one of the initially reported alleles 14. The observer was able to correctly identify the mutant embryos. Furthermore, his count of the percentage of defects in ISNb and ISNd nerves was similar to our estimate. We therefore conclude that tutl⁰¹⁰⁸⁵ mutants have a robust motor axon pathfinding defect.

Using complementation testing and transgenic rescue, we verified that, like the midline crossing defects, these motor axon pathfinding defects were caused by mutations in turtle. Embryos trans-heterozygous for tutl^ex383 and tutl⁰¹⁰⁸⁵ have similar embryonic motor axon projection defects to tutl^ex383homozygotes (Figure 4G), but with lower frequency. Furthermore, expression of either of the two diffusible turtle isoforms via a Elav-Gal4 driver in tutl^ex383 homozygotes can rescue these defects (Figure 4C, G). However, the motor axon projection defect is only reduced, rather than fully rescued, by the two membrane-spanning isoforms (Figure 3G), which further points to the strong potency of the diffusible isoforms in mediating turtle's function.

We examined the consequences of neuronal overexpression of the different turtle isoforms in a wild-type background using the pan-neuronal driver Sca-Gal4. This overexpression produced excessive branching and stalling of motor nerves, with some nerves, such as the transverse nerve (TN), innervating muscles that are not their normal targets (Figure 4D). The diffusible isoforms were, once again, more potent in producing axonal abnormalities when overexpressed than the membrane-bound isoforms (Additional file 3).

Even though turtle expression is largely restricted to the nervous system during development, some isoforms are secreted, suggesting that they may function as extracellular signaling molecules rather than as cell autonomous receptors. Therefore, one would predict that ectopic turtle overexpression solely in axonal targets, such as body wall muscle, should produce an alteration in motor axon morphology similar to that observed when turtle is overexpressed in the motor neurons themselves. When the different turtle transgens were expressed in body wall muscles using 24B-Gal4 and G14-Gal4 drivers, we indeed observed motor axon defects similar to those seen using the neuronal driver Sca-Gal4 (Figure 4E, F), even though the axons in this case expressed wild-type levels of turtle.

In wild-type Drosophila adult eyes the retinal axons are arranged in a regular array, with photoreceptor R7 axon terminations forming a solid line across the M6 layer of the medulla 27 (Figure 4H). In the weak mutant allele tutl^k14703, approximately 20% of the homozygotes reach the adult stage; these adults have an optic chiasm that is disorganized and gaps in the R7 termination line (Figure 4I). Expression of the diffusible isoform using the pan-neuronal driver Elav-Gal4 in tutl^ex383/tutl^k14703 rescues most aspects of this defect (Figure 4J). Overexpression of the different turtle isoforms in a wild-type background using the retina-specific GMR-Gal4 driver produces abnormalities that are strikingly similar to defects seen in turtle-overexpressing motor axons, in that many retinal axons stall before reaching their targets and some sprout extra axonal processes (Figure 4K, arrow and arrowhead, respectively). The diffusible isoforms were, again, more potent in producing these axonal abnormalities when overexpressed than the membrane-bound isoforms.

As noted above, turtle is expressed in both the eye disc and the contact zone between the optic lobe and optic stalk in third instar larvae. If turtle functions as a signaling molecule that guides retinal axons to their optic lobe targets, the critical domain of expression should be the optic lobe rather than the eye disc. To test this, flies with eyes composed of turtle mutant cells and brains expressing wild-type levels of turtle were generated using the eyeless-GAL4 UAS-FLP (EGUF)/hid mosaic technique 28. Although the resulting eyes were smaller than wild-type, they had normal optic chiasma and R7 projections (Figure 4L), indicating that the source of retinal axon defects in tutl^k14703 mutants is reduced turtle levels in the optic lobe, rather than in the retinal axons. These results indicate that, in retinal axons, turtle does not function as part of a guidance receptor complex, nor as a homophilic or hetrophilic cell adhesion molecule that is needed in both retinal axons and their brain targets.

The fly and vertebrate central nervous systems share a common two-layered organization when viewed under histological staining (Figure 5A). The fly cortex is exclusively composed of neuronal cell bodies, while the deep layer, or neuropil, is largely composed of axons and neurites (Figure 5A). Overexpression of the diffusible turtle isoforms in a wild-type background produces neuronal structures that do not conform to these histological boundaries. Diffusible turtle isoform overexpressing retinal axons occasionally invade the cortex (Figure 4K, asterisks), and, conversely, turtle overexpressing neuronal cell bodies invade the normally cell-free neuropil layer (Figure 5D, E).

Based on the results obtained from both tissue-specific transgenic expression and EGUF/hid mutant eye analysis, we conclude that some turtle isoforms can function as extracellular signaling molecules that act via a non-cell autonomous mechanism to promote axonal branching and invasive behavior.

Two P-element insertions in the turtle locus, P {ry^+t7.2 = PZ} l(2)01085 and P {w^+mc = lacW} l(2)k14703, were obtained from the Berkeley Drosophila Genome Project, and their flanking genome sequences were determined by the inverse PCR primer method

A ClaI restriction map was made of the 30 kbp genomic interval where the turtle gene is located. A Southern blot of ClaI digests using genomic DNA of wild-type, and tutl^l(2)k14703, and tutl^lex383 mutants was performed, and the blot was sequentially probed with two different radio-labeled probes that span different portions of the turtle genomic interval. The first probe, which spanned the 3'-end of the transcript, should hybridize to a 6-kb band in wild-type DNA and two different bands in the tutl^l(2)k14703/CyO DNA: one will be a 6-kb band corresponding to the wild-type copy of turtle in the CyO balancer, and the other will be an 8-kb band corresponding to the turtle mutant chromosome (since the P-element contains a ClaI site 7 kb away from its 5' end). When using the 3'-end probe, a clear ClaI polymorphism in tutl^ex383 was detected and a new band of approximately 13 kb appeared. This band was not present in either wild-type or tutl^l(2)k14703 genomic DNA, thus demonstrating that a deficiency was induced in the turtle locus (Figure 1A). The second probe flanked the 5'-end of the turtle gene. This probe should produce bands of 1.5, 3, 7, and 8 kb in both wild-type and tutl^l(2)k14703. This probe did not produce any ClaI polymorphism in the tutl^ex383 DNA digest (Figure 1A). We concluded from the previous results that the deficiency did indeed occur in the turtle gene in tutl^ex383 and is restricted to the region between the two probes.

To clarify the extent of the deficiency in tutl^ex383, we designed two series of PCR primers that spanned the genomic interval around the P-element insertion. These were called primer series A and B. In the A series, we designed six primers that were 2 kbp apart. The first primer, A1, was located approximately 16 kb away from the 5'-end of the P-element insertion while the last primer, A6, was located approximately 6 kb from that site. Series B had only two primers, both located after the 3'-end of the P-element insertion; primer B1 was located 1 kbp away from the site of P-element insertion, while primer B2 was 4 kbp away from that site. A PCR reaction was done on genomic DNA isolated from tutl^ex383 with one primer of the A series and one primer of the B series (Additional file 1B). Only 1 minute of elongation time per cycle was given to the PCR reaction. Hence, amplification occurred only if the left and right primers were brought within 1 kb of each other as a result of the deficiency. Only the combination A4/B2 gave a product, which was approximately 900 bp long (Additional file 1C). Since those two primers are about 14 kbp away from each other in the wild type, their ability to produce a PCR product demonstrated a deficiency had removed 12 kbp of genomic DNA within the turtle locus in tutl^ex383. We then subcloned the PCR product and sequenced it. The sequencing result indicates the presence of an 11,430 bp deficiency in tutl^ex383 (Additional file 1D). Such a deficiency is more than enough to effectively remove all the different isoforms of turtle, thereby producing a true null allele of this gene.

The PCR primer series A and B were designed based on the genomic sequence of the P1 Clone DS00830, and have the following sequences: A1, TAC TTT TGC GGA CAC TCT CTC TGT CC; A2, GCT TTG CTG CCT GAC AAA TAG CAA GG; A3, AGG GAA ARC GAA GCC AAA CCG AAA GG; A4, ATA AGA ATC GGA TGC GAA CCG TCA GC; A5, GCC ATG CGA GTT TTA CTG TTC TAG GC; A6, ATC CAG AGT AAC AGA GTC TAC GAG CC; A7, AAT CTG GCT TCA TAG CGC CTT GCA GC; B1, CCA GTG GAG TAT GGC AAT GAG AAT GG; B2, TTT TTC TGG GTG TGA GTT TGC AGG GG.

Total RNA was extracted from 1 mg stage 12/13 wild-type embryos using a Qiagen RNA Extraction Kit (Quigen #74104 Valencia, CA, USA) and the amount of RNA was measured with a Nanodrop spectrophotometer. Different samples were normalized to 1 μg total RNA, and DNA contamination was removed using Dnase1 AM treatment (Sigma #048K6043 St. Louis, MO, USA). After Dnase1 deactivation, reverse transcription was carried using a ISCUP™ cDNA synthesis kit (BIO-RAD #170-889 Hercules, CA, UAS1). Isoform specific primers with a T7 promoter in the 3'-end of the reverse primer were designed: TATTCCAGAAGACG and TAATACGACTCACTATAGGGAGAGCAAT for AT02763; CAAAGTCCTTCGTCAAACGC and TAATACGACTCACTATAGGGAGATTTCA for GH15753; TCAATTGCCAGGCAGATGGC and TAATACGACTCACTATAGGGAGTCACC for HL01565; GCCAACTCGGAGAAGTCG and TAATACGACTCACTATAGGGAGATTATG for LD28224; and GTGCAATTACCTGCCGTTTCG and TAATACGACTCACTATAGGGAGAGGAGA for GH08133.

The PCR reaction was then cleaned by phenol:chloroform extraction, and the DNA was ethanol precipitated and used as a template for in situ probes and hybridization of Drosophila embryos as described by Tear et al. 33. The different probes were tagged with either fluoro-isothiocyanate (FITC), biotin, or Dig.

Mab 1D4, 24B10, anti-Eng, anti-Slit, anti-Eve, and BP102 antibodies were obtained from the Developmental Studies Hybridoma Bank and were used in a 1:5 dilution, while anti-mouse β-gal was purchased from Promega and used in a 1:10,000 dilution. Iowa city, Iowa, USA

turtle mutant chromosomes were balanced over a CyO, Elav-lacZ balancer. Homozygous mutant embryos were identified by the absence of anti-mouse β-gal staining. To quantify nerve defects in a given mutant genotype, eight to ten mutant embryos were fixed, dissected, and stained with anti-FasII. The detection of anti-FasII was achieved using an anti-mouse horseradish peroxidase-conjugated secondary antibody, detected by DAB/1:10 dilution 30% H₂O₂ staining. Only nerves from abdominal hemisegments A2 to A7 were scored.

For immunocytochemistry of adult heads and retinal axons, 5-day-old fly heads were dissected (proboscis and air sacs removed) and fixed for 2 hours in ice with 4% paraformaldehyde in NaCl-free 5× phosphate-buffered saline. They were then washed extensively with phosphate-buffered saline and cryosectioned. The sections were fixed again for 20 minutes in 4% paraformaldehyde, followed by another round of washing. Immunostaining with or without DAPI was then performed.

Wild-type and tutl⁰¹⁰⁸⁵ mutant embryos were collected, fixed simultaneously, and processed under the same conditions. They were dissected and stained with FasII. The embryos were mounted on slides, with each slide containing only one embryo. Samples were given random identification numbers. An independent observer was then asked to examine the slides and attempt to determine each embryo's genotype; the observer's evaluations were compared with the true genotypes of the embryos.

For fluorescence detection, the specimen was imaged under a 20 × 1.4 numerical aperture lens on a Zeiss LSM 510 META confocal microscope. The DAPI channel (Excitation: λ 720 nm; Emission Band Pass 365 to 480 nm) and FITC channel (Excitation: λ 488 nm; Emission Band Pass 500 to 550 nm) were acquired. The acquired Z-stack was flattened as a maximum intensity projection and exported to a TIF file. Cropping and adjustment of brightness and contrast were performed in Adobe Photoshop.

Schneider 2 (S2) cells were maintained at room temperature in a growth medium composed of Schneider medium with 10% fetal calf serum and penicillin/streptomycin (all reagents were from Gibco-Invitrogen Carlsbad, CA, USA). The cells were transfected using the Cell Infecting Kit (Invitrogen Carlsbad, CA, USA), as described by the manufacturer, and were maintained at room temperature in a growth medium containing 400 μg/ml Zeocin.

Both conditioned medium and S2 cells were collected 1 week post-transfection. Conditioned medium (15 μl) and the S2 cell pellet were solubilized using SDS-polyacrylamide sample buffer. The samples were then run on 10% SDS-PAGE and transferred onto a nitrocellulose filter using standard procedures.

cDNA clones of turtle were obtained from Research Genetics-Invitrogen Carlsbad, CA, USA. To subclone the His-tagged GH01573 and AT02763 transcripts into the PIZT insect expression vector (Invitrogen), we performed a standard PCR reaction using high-fidelity Pfu taq. Plasmid DNA of GH15753 and AT02763 (100 ng) was used as a template. Two primers were designed for each EST clone; the 5'-primers contained an EcoRI site before the initiation codon, while the 3'-end primers contained an in-frame His-tag sequence followed by a stop codon and an XbaI restriction site. The resulting PCR product was digested with EcoRI/XbaI overnight, and then ligated to a linearized EcoRI/XbaI digested PIZT vector. The ligation product was transfected into DH10B electrocompetent Escherichia coli, and successful transformants were selected based upon the Zeocin resistance conferred by the PIZT vector. Plasmids were isolated and sequenced by the Keck facility. Yale University, New Haven, CT, USA Only transformants that had the fully correct sequence of GH01573-His were used for S2 Drosophila cell line transfection.

Codon sequences GH15753, AT02763, HL010565, and LD28824 were subcloned into a PUAST vector and were used to transform w¹¹¹⁸ flies using standard transgenic procedures.