SpaRCLIn was developed to refine the expression pattern of a Gal4 driver using the basic strategy shown in Figure 1. In common with other existing methodologies, SpaRCLIn uses a recombinase (i.e. Cre) to excise an otherwise ubiquitously expressed construct encoding Gal80, a suppressor of the Gal4 transcription factor (Figure 1A–B). As in the CLIn technique, recombinase expression—and thus the excision of Gal80—occurs only in targeted NBs, rendering the progeny of these NBs permissive to Gal4 activity (Figure 1C). Those progeny that lie within the expression pattern of the Gal4 driver will be competent to drive UAS-reporters and effectors, such as UAS-GFP. In the SpaRCLIn technique, distinct NBEs are used to express components of a bipartite Split Cre molecule in restricted subsets of NBs. In lineages of these NBs that contain mature neurons within the Gal4 expression pattern, Gal4 will be active. This population of neurons can be additionally parsed using a tripartite Split Cre to further restrict the subset of NBs that make active Cre (Figure 1D).

Although most recombinase-based expression systems in Drosophila, such as MARCM (Lee and Luo, 1999), Flp-out Gal80 (Gordon and Scott, 2009), and FINGR (Bohm et al., 2010) have preferentially used the Flp recombinase for Gal80 excision, we selected Cre for use in SpaRCLIn because of its demonstrated ability to retain high activity in a variety of bipartite forms (Hirrlinger et al., 2009; Jullien, 2003; Kawano et al., 2016; Kennedy et al., 2010; Rajaee and Ow, 2017). Although Cre activity has been reported to be toxic in Drosophila when chronically expressed at high levels (Heidmann and Lehner, 2001; Nern et al., 2011), it has previously been used in NBs without apparent adverse effects (Awasaki et al., 2014; Hampel et al., 2011; Ren et al., 2016). Because our system requires use of a tripartite Cre to achieve the most refined targeting it was also desirable to use a method of splitting Cre that would permit reconstitution of the intact molecule to obtain the highest activity levels. Split inteins, which are capable of autocatalytically joining two proteins to which they are fused, are well-suited to this purpose and distinct split inteins have been previously shown to support reconstitution of recombinase activity from complementary Cre fragments fused to them (Ge et al., 2016; Han et al., 2013; Hermann et al., 2014; Wang et al., 2012). Figure 1E shows the primary structure of Cre, indicating the location of the breakpoints (green highlight) at which we introduced split intein moieties into the molecule. These breakpoints separate the amino acid residues in the primary structure that form the DNA-binding sites (blue) and the active site (yellow highlight), thus insuring that none of the fragments retains catalytic activity. Two Split Cre fragments, Cre_AB and Cre_C, were generated by the breakpoint between amino acids P250 and S251 to implement the bipartite Split Cre system (Figure 1F,G), while dividing the Cre_AB fragment at the breakpoint between amino acids D109 and S110 was used to create two further fragments (i.e. Cre_A and Cre_B) which together with Cre_C form the basis of the tripartite Split Cre system (Figure 1H,I). The split intein pairs used to generate these fragments, gp41-1 and NrdJ-1, were chosen based on their trans-splicing efficiency and their lack of cross-reactivity (Carvajal-Vallejos et al., 2012). The latter criterion was critical for avoiding the generation of unproductive fusion products of the Cre fragments.

After confirming the ability of the bi- and tripartite constructs to reconstitute Cre activity when co-expressed in transfected S2 cells (data not shown), we used them to generate transgenic fly lines in which they were expressed in patterns dictated by individual enhancers that exhibited activity in neuroblasts. Most of the NBEs selected for this purpose were taken from the large collection of enhancer fragments with fully defined sequences created by the Rubin lab (Pfeiffer et al., 2008 #43). Most of the NBEs selected were from a previously characterized collection of embryonically active NB enhancers (Manning et al., 2012), with the remainder characterized as indicated in the Materials and Methods and Supplementary file 1. A total of 134 NBEs were used to make two libraries of transgenic fly lines, one expressing the Cre_B fragment under the control of each of the 134 NBEs and the other similarly expressing the Cre_C fragment. These lines thus collectively express Cre_B and Cre_C in a large number of distinct and often overlapping subsets of NBs (Figure 1—figure supplement 1). However, because the 134 enhancers are also typically active in mature neurons, the production of full-length Cre is not necessarily restricted to NBs (Jenett et al., 2012).

To ensure NB-specific reconstitution of Cre activity, we placed the Cre_A and Cre_AB fragments under the control of a compound enhancer formed by fusing individual enhancer elements of the NB-specific genes, deadpan (dpn) and nervous fingers-1 (nerfin-1; see Materials and methods). This synthetic dpn-nerfin-1 enhancer (i.e. DNE) combines the complementary temporal characteristics of both component enhancers, maintaining strong, broad, and specific activity throughout embryonic neurogenesis (Figure 1—figure supplement 2A,B). Use of the DNE thus ensured that full-length, active Cre would be generated only in NBs where expression of the Cre fragments overlapped, and not in fully-differentiated neurons (Figure 1G,I). This enhancer also expresses in most of the NBs that give rise to the Drosophila CNS with the exception of those found in the late-developing optic lobes, and thus guarantees substantial coverage of the mature neurons found within the expression patterns of Gal4 lines.

To detect activity of the Split Cre constructs in vivo, we created transgenic flies carrying a reporter construct in which the floxed Gal80 gene, the expression of which is driven by a ubiquitously active Actin 5C promoter, is followed by the gene encoding the red fluorescent protein, tdTomato (Figure 1J). Expression of tdTomato from this construct, which we call Cre80Tom, thus identifies neurons in which Gal80 has been excised. Gal80 excision, identified by the appearance of tdTomato expression, is identifiable as early as embryonic stage eight when the Cre_AB and Cre_C fragments are driven by the DNE and it is widespread in the developing CNS by stage 13 (Figure 1—figure supplement 2C). The onset of Cre activity is sufficiently early to label many of the progeny of a defined NB lineage targetable by the R59E09-Gal4 line, previously identified by Lacin and Truman (2016) (Figure 1—figure supplement 2D). These data suggest that Gal80 excision occurs relatively early.

The bipartite system, using the DNE-Cre_AB and Cre_C fragments expressed under the control of two different neuroblast enhancers (NBE_43H02 and NBE_44F03), also generated expression patterns in neuronal lineages of third instar larvae (Figure 1K and L). The expression patterns include not only the NBs in which Cre activity is reconstituted, but also the progeny of these NBs, since tdTomato expression is activated in all cells born within these lineages after Gal80 is excised. Although the expression patterns differ in the two cases, they share a small number of common NB lineages as is revealed by application of the tripartite Cre system using the NBE_44F03 and NBE_43H02 enhancers to drive Cre_B and Cre_C, respectively, together with DNE-Cre_A (Figure 1M). Expression in this case is limited to approximately three bilateral lineages in the ventral nerve cord (VNC) and two in the brain. These examples illustrate how the bi- and tripartite Split Cre constructs selectively reconstitute Cre activity in NBs targeted by individual NBEs, and demonstrate that the tripartite Split Cre system can be used to restrict Cre activity to only those NBs in which two distinct NBEs are active.

As this example illustrates, the tripartite system generates intersections of two NBE-Cre_C expression patterns by substituting one Cre_C fragment with a Cre_B fragment driven by the identical NBE. To facilitate such substitutions, all NBE-Cre_C insertions were made on Chromosome III and all NBE-Cre_B insertions were made on Chromosome II. To evaluate the reproducibility of expression driven by individual NBEs, we examined NBE-Cre_C∩DNE-Cre_AB >CreTom crosses for all 134 NBEs in multiple preparations (on average four per NBE) and compared these patterns with expression observed in all NBE-Cre_B∩DNE-Cre_A∩ DNE-Cre_C >CreTom crosses. The large size of the patterns and the inability to reliably identify identical neurons and lineages across preparations prevented a systematic analysis, but in general the patterns were similar across preparations for any given cross (Figure 1—figure supplement 3) In most cases, patterns obtained with a given NBE-Cre_B line also resembled the pattern obtained with the NBE-Cre_C line made with the same enhancer (compare panels A and B in Figure 1—figure supplement 3). However, for 24 NBEs overt differences were observed (Figure 1—figure supplement 3C vs 3D). Apart from these NBEs, which have been marked with an asterisk in Supplementary file 1 along with guidance as to their use, the tripartite system represents a reliable intersectional method for restricting Cre activity to subsets of NBs. The progeny of these NBs that are generated after Cre activation will not only express the reporter tdTomato, but will also fail to express the Gal80 transgene, thus permitting Gal4 to function.

The selective disinhibition of Gal4 activity in targeted lineages permits UAS-transgenes to be expressed in cells of those lineages whenever they lie within the expression pattern of a Gal4 driver. This allows targeted lineages to be parsed according to the properties of the mature neurons to which they give rise using cell-type specific Gal4 drivers. Such so-called ‘cell class-lineage intersections’ have been previously performed to identify subsets of neurons generated by Type II NBs of the Drosophila brain, which can be selectively targeted using a Type II-specific enhancer (Ren et al., 2016; Ren et al., 2018; Ren et al., 2017). Among the neurons generated by Type II NBs are several populations of dopaminergic neurons, identified by a Tyrosine Hydroxylase-specific Gal4 driver (TH-Gal4). Dopaminergic neurons are of considerable interest because of their roles in a variety of important neurobiological processes, including learning, sleep, and locomotion (for review see Kasture et al., 2018). The approximately 120–130 dopaminergic neurons in the Drosophila CNS are produced by diverse NBs and numerous reagents have been generated to selectively target them (Aso et al., 2014; Friggi-Grelin et al., 2003; Xie et al., 2018).

As a first test of the SpaRCLIn system, we therefore asked whether it could restrict expression of the TH-Gal4 driver (Figure 2A) to small numbers of distinct dopaminergic neurons based on their different lineages of origin. Using a small subset of the NBE-Cre_C lines in combination with DNE-Cre_AB, we examined the expression patterns produced by intersection with TH-Gal4. The expression patterns produced by these intersections were noticeably reduced compared with the full pattern of the TH-Gal4 driver, but they typically still contained 10’s of dopaminergic neurons distributed broadly across the neuraxis (Figure 2B–C). In cases where the expression patterns produced by the bipartite crosses shared a neuron (Figure 2B–C, arrows), combining the relevant NBEs using the tripartite system succeeded in isolating these neurons from most others in the two original crosses (Figure 2D). In general, restricting NB expression using the tripartite system—by pairing the NBE-Cre_C constructs with NBE-Cre_B constructs made with different enhancers—produced significantly reduced expression patterns, sometimes consisting of one to two cells or bilateral cell pairs (Figure 2D–H).

The expression patterns from 14 NBE-Cre_C∩NBE-Cre_B intersections—produced by combining 15 distinct NBEs—were analyzed in detail to quantify both the average number of dopaminergic neurons and the stereotypy of expression for each intersection (Figure 2I). We found that the average number of labeled neurons per preparation did not exceed 8.5 (±3.8, n = 16) for any intersection and was less than 4.3 (±2.3, n = 17) for two-thirds of them. This sparseness of expression suggests that the NBEs tested do not overlap extensively in their NB expression patterns. Stereotypy of expression was also generally present despite considerable variability. Only in one extreme case, did there appear to be a complete absence of stereotypy, with all CNS preparations that had expression displaying a distinct pattern (Figure 2—figure supplement 1). For all other intersections, at least one principal neuron was found that was shared by multiple preparations, based on cell position and morphology (Figure 2I, black bars). For over half of the intersections, this principal common neuron was shared by 50% or more of preparations. In most cases, other neurons were also found, though preparations containing only such neurons typically occurred at lower frequency (Figure 2I, gray bars). Consistent with this variability of expression, neurons that recurred across preparations were not necessarily found in the same combinations (Figure 2—figure supplement 2).

The sparseness of labeling combined with the variability of expression likely accounts for why half of the intersections yielded at least one preparation without any expression. Interestingly, four of the seven intersections that yielded preparations devoid of expression shared an enhancer (R14E10), suggesting that particular enhancers may strongly influence the extent of labeling. Variability of labeling also appeared to be enhancer-dependent in that use of the same enhancer (i.e. R17A10) to drive both Cre_B and Cre_C components did not necessarily reduce stochasticity. Indeed, although all preparations that had expression shared a common identifiable neuron in this case (Figure 2I), their expression in other neurons varied considerably. A possible source of this variability of expression is weak NBE activity that results in lowered expression of Cre components and consequently more sporadic reconstitution of Cre activity. More work will be required to examine this hypothesis. Regardless, our results demonstrate SpaRCLIn’s ability to substantially restrict expression of a Gal4 driver with sufficient stereotypy in single neurons to be useful for the neuronal manipulations employed in neural circuit mapping.

To examine SpaRCLIn’s efficacy for circuit mapping, we used it to identify neural substrates of proboscis extension (PE), a motor pattern normally elicited by gustatory stimuli, but also by the hormone Bursicon in newly eclosed flies (Peabody et al., 2009). Robust PE can be readily induced even in older flies using a driver (rk^pan-Gal4) that selectively expresses in Bursicon-responsive neurons (Video 1, Figure 3A,B; Diao and White, 2012). Expressing the heat-sensitive ion channel UAS-dTrpA1 under the control of this driver, we performed an initial (‘Step 1’) screen of the Cre_C library using the bipartite SpaRCLIn system (Figure 3—figure supplement 1A). In this screen, crosses were conducted between each NBE-Cre_C line and a line that combined all other components, including Cre80Tom, DNE-Cre_AB, rk^pan-Gal4, and UAS-dTrpA1. To facilitate visualization of neurons within the resulting expression pattern without requiring additional genomic insertions, we used a dual expression construct (Cre80Tom- GFP) that contained actin^Gal80^myr-tdTomato and a 10XUAS-mCD8GFP reporter (Figure 3—figure supplement 2). Progeny were videorecorded in small chambers on a temperature-controlled plate and assayed for heat-induced PE. Interestingly, several different PE phenotypes were apparent, but only those that involved full extension of the proboscis could be reliably scored under our assay conditions and we therefore focused on the latter. Applying this criterion, we identified 23 NBE-Cre_C∩DNE-Cre_AB intersections for which UAS-dTrpA1 activation reliably induced robust PE in greater than 50% of the progeny. The expression patterns resulting from these Cre_AB∩C∩rk^pan-Gal4 (i.e. Step 1) intersections, examined using a UAS-GFP reporter, were clearly restricted relative to rk^pan-Gal4 expression (Figure 3C–D), but they were insufficiently sparse to readily identify the neurons—or population of neurons—responsible for inducing the PE motor pattern.

Taking advantage of SpaRCLIn’s ability to further restrict expression, we used the tripartite system to carry out a second (‘Step 2’) screen in which the 23 identified NBE-Cre_C components were combined pairwise with NBE-Cre_B components made using the same 23 enhancers (Figure 3—figure supplement 1B). The latter were selected from the NBE-Cre_B library and crosses were made that combined distinct NBE-Cre_B and NBE-Cre_C components with DNE-Cre_A, rk^pan-Gal4, and Cre80Tom-GFP. These Step two crosses resulted in Cre_A∩B∩C ∩ rk^pan-Gal4 intersections that were assayed for PE as before. Approximately 70 intersections were tested before screening was discontinued because 11 intersections had already yielded PE phenotypes in greater than 50% of flies. The phenotype observed was typically less sustained than that produced by activation of the full rk^pan-Gal4 expression pattern in that activation typically caused rhythmic, rather than tonic, extension of the proboscis, which after prolonged heating often transitioned to lifting of the rostrum rather than full extension (Video 2; Figure 3E).

The rk^pan-Gal4 expression patterns in flies exhibiting this phenotype were substantially reduced for many of the intersections tested and they consistently included particular neurons in the subesophageal zone (SEZ) that were characterized by somata near the saddle, broad arbors along the superior gnathal ganglion (GNG), and axons that extended medially before turning, with one branch coursing down each side of the midline and then turning laterally along the medial-inferior edges of the GNG (Video 3). Two closely apposed neurons of this type were observed, sometimes as bilateral pairs (Figure 3F), and sometimes on only one side (Figure 3G). These neurons, which we call the PE^rk neurons, were notably prominent in the 16H11-Cre_B∩44F09-Cre_C intersection, where they constituted the entire expression pattern of 16 animals (n = 78 total), all of which exhibited PE induction upon heating. Indeed, all 36 animals from this intersection that tested positive for the PE phenotype and were successfully dissected showed expression in the PE^rk neurons, while none of the animals (n = 38) that tested negative had such expression (Figure 3I). Most of the latter, in fact, had little to no expression. Similar results were obtained with a second intersection (44F09-Cre_B∩10G07-Cre_C). All 19 animals that exhibited induced PE in this intersection had expression in the PE^rk neurons, and in three animals these were the only neurons present. A third intersection that yielded the PE phenotype in all animals likewise showed consistent expression in the PE^rk neurons, but the correlation between the PE phenotype and expression in these neurons was somewhat less readily established because of expression in other neurons (5.6 ± 1.8; n = 14 preparations; Figure 3—figure supplement 3).

The above examples demonstrate that SpaRCLIn can be used to rationally parse the expression patterns of Gal4 drivers using the workflow shown in Figure 3—figure supplement 1. One challenge to using this system, however, is the large number of transgenes required to implement it. This is especially true for Step two screening with the tripartite system. To mitigate this burden, we have created several reagents that will facilitate use of the system. In addition to the Cre80Tom-GFP construct described above, we have developed other dicistronic constructs to facilitate manipulating neuronal activity in SpaRCLIn screens (see Key Resources Table). These include constructs and fly lines for Cre80-Kir2.1 and Cre80-dTrpA1 (Figure 3—figure supplement 2). In addition, we have developed an alternate Step one strategy that may avert the need for Step two screening in favorable cases.

The alternate strategy uses a transiently expressed DNE-Cre_AB designed to be active only during early stages of neurogenesis. This construct, which we call ‘FRTerminator,’ is self-excising in that it is flanked by Flp Recombination Target (FRT) sites and encodes a Flp recombinase gene that is co-expressed with Cre_AB (Figure 4A). Upon expression under control of the DNE enhancer, this construct will remove the Cre_AB gene and thus limit its expression to early (embryonic) neuroblasts (Figure 4B). Cre_AB will thus be available to reconstitute Cre activity only with complimentary Cre_C fragments that are also expressed at this time. Cre_Cs whose expression is driven by NBEs that become active only after the elimination of Cre_AB from neuroblasts, will not lead to the generation of Gal4-competent neurons. Expression patterns resulting from the combination of FRTerminator with NBE-Cre_Cs will thus, in general, be reduced relative to those produced by DNE-Cre_AB (Figure 4C,D).

To determine whether the FRTerminator might therefore expedite parsing of Gal4 expression using the SpaRCLIn system, we repeated selected crosses from the rk^pan-Gal4 Step one screen described above. We focused on the 23 NBE-Cre_C lines that yielded flies with PE phenotypes, combining each with the FRTerminator, rk^pan-Gal4 and Cre80-GFP. Progeny were tested for PE upon dTrpA1 activation. We found that three NBE-Cre_C lines (44F09, 57B09, and 14E10) produced progeny with PE phenotypes at frequencies ranging from 9–17%. Although these frequencies were considerably lower than those obtained using DNE-Cre_AB, the resulting expression patterns were substantially sparser compared with those of progeny from DNE-Cre_AB crosses (Figure 4E,F). All animals examined that had PE phenotypes also included in their expression patterns the PE^rk neurons (n = 40). In contrast, only one of the animals examined that lacked the phenotype had these neurons (n = 39). A strong correlation between PE and the presence of the PE^rk neurons was thus observed, again permitting the conclusion that these neurons are substrates for the behavioral phenotype. We conclude that FRTerminator-based Step one screens may serve as a useful shortcut to serial Step one and Step two screens for restricting Gal4 expression and identifying functionally important neuronal subsets.

Our use of SpaRCLIn to identify the RK-expressing neurons that trigger robust proboscis extension demonstrates SpaRCLIn’s ability to systematically parse a neuronal group and identify the functionally relevant subset. Just over 200 crosses—134 crosses for the Step 1 screen of NBE-Cre_C lines and 70 NBE-Cre_B∩C Step two crosses—were required to identify two pairs of command-like neurons capable of inducing PE upon activation (i.e. the PE^rk neurons). Importantly, we discontinued the Step two screen after testing 70 of the 253 possible intersections because of evident redundancy of the command-like neurons. The latter were prominent in the expression patterns of numerous independent Step two intersections and were readily correlated with PE induction in three that produced particularly reduced expression patterns. In the intersection with the sparsest expression, the two pairs of PE-inducing neurons often comprised the entire observable pattern in flies that had the PE phenotype, illustrating the extreme reduction in expression achievable with SpaRCLIn. The demonstration that the PE^rk neurons can be isolated in single crosses using the FRTerminator indicates that this reduction in expression can be attained without the labor of Step two screening. However, the lower frequency of the PE phenotype in FRTerminator crosses in our example also suggests that FRTerminator-based screens may require testing more animals for each intersection than a standard Step one screen in order to reliably identify positives.

Activation of the PE^rk neurons elicits rhythmic proboscis extension, rather than the tonic PE elicited by activation of all rk^pan-Gal4 neurons. This suggests that additional RK-expressing neurons—perhaps lacking command capability—modulate the effects of activating the PE^rk neurons. Based on their induction of rhythmic extension and their apparent lack of a projection to the proboscis muscles, we conjecture that the PE^rk neurons identified here are not motor neurons, the activation of which results in tonic and often partial PE (Gordon and Scott, 2009; Schwarz et al., 2017). Similarly, the anatomy of the PE^rk neurons differs from that of other identified neurons that can drive PE when activated, including second-order projection neurons (Kain and Dahanukar, 2015), modulatory neurons (Marella et al., 2012), and a local SEZ interneuron called the Fdg-neuron (Flood et al., 2013). Like the Fdg-neuron, however, the neurons identified here seem to function in a premotor capacity, perhaps as part of the central pattern generator for PE that regulates fly feeding (Itskov et al., 2014). Further work will be required to determine the precise role of the PE^rk neurons in the feeding circuitry and their relationship to other identified neurons involved in PE.

It also remains to be determined whether activation of both PE^rk neurons is required to induce the PE phenotype. Indeed, from the standpoint of the efficacy of the SpaRCLIn system it is important to ask why SpaRCLIn failed to separate these two pairs of neurons. The similarity of the two PE^rk neurons in both soma position and projection pattern is consistent with their being part of the same lineage. Such neurons will necessarily be more difficult to parse using SpaRCLIn, which can separate neurons within the same lineage only based on their birth order. What would be required to do so is having two NBEs that are active in the same lineage but at different times so that they separate earlier- from later-born neurons. Such NBEs, by generating Cre only in older neuroblasts, will generate sublineages of Gal4-competent neurons. Although many of the NBE’s used to make our Cre_B and Cre_C libraries clearly generate such sublineages—based on the patterns shown in Figure 1—figure supplement 1—it is doubtful that they cover more than a fraction of all temporal windows of neurogenesis in all neuronal lineages. A method for systematically isolating sublineages of later born neurons using SpaRCLIn may become possible if neuroblast-specific enhancers can be found that are selectively active at later stages of neurogenesis. These could then be used in lieu of the DNE used here. Candidates for such enhancers are those that determine expression of the so-called ‘temporal transcription factors’ that regulate the progressive divisions of many neuroblasts (Doe, 2017).

Although stochasticity is not an uncommon feature of many expression systems (Bohm et al., 2010; Tastekin and Louis, 2017), the variability of expression generated by SpaRCLIn was notable. Even for intersections that reliably produce very similar expression across animals, it is not common to get exactly the same pattern twice. The infidelity of expression may derive, at least in part, from intrinsic stochasticity of NBE activity. However, our results indicate that individual NBEs drive expression in broadly reproducible patterns. Other factors that may contribute to expression variability include the strength and/or temporal properties of NBE activity. If a Cre component is only weakly expressed, or expressed late during neurogenesis, a limited amount of active, full-length Cre may be produced and excision of Gal80 may be sporadic in the expressing neuroblasts. Also, the success of Cre reconstitution may vary, and may be particularly low when the religation of three fragments is required. Although the two split inteins used in the SpaRCLIn system were chosen based on their favorable reaction kinetics as determined in vitro, the speed, efficiency, and variability with which they react in different types of cells remain unknown. Finally, because the efficacy of Cre-mediated excision depends on the distance between the loxP sites flanking the excised fragment, it may prove possible to increase the efficacy of excision—and thus reduce variability of expression—using strategies that decrease the distance between the loxP sites flanking the fragment to be excised.

While further work will be required to identify the sources of variable expression within the system, the observed stochasticity is not necessarily a disadvantage for circuit-mapping applications, as illustrated by the examples presented here. By providing partially ‘randomized’ expression patterns, SpaRCLIn permits causative relationships to be inferred between groups of manipulated neurons and the effects produced by their manipulation (Jazayeri and Afraz, 2017). Such randomization has been commonly exploited in so-called ‘Flp-out’ methods that rely on stochastically induced recombinase activity to remove an FRT-flanked gene or transcription stop cassette (Flood et al., 2013; Gordon and Scott, 2009; Kain and Dahanukar, 2015). This logic is naturally implemented in SpaRCLIn, but because randomness of expression is considerably more constrained than that observed in systems that rely on strictly stochastic methods, and because the size of the expression patterns is typically small, correlations can be readily established.

One consequence of SpaRCLIn’s stochasticity that must be considered in circuit mapping applications, however, is the lowered frequency of bilateral labeling. Most neurons occur as members of bilateral pairs and we observed numerous instances in which SpaRCLIn-derived expression patterns contained only a single member of each pair in a given preparation—presumably due to the variable success of Gal80 excision in both NBs giving rise to the pair. The reduced bilateral representation of neurons may likewise reduce the frequency of phenotypes observed as a consequence of a particular manipulation if, for example, both neurons in a pair must be affected to produce a phenotype. This is often the case for suppression of function, where both neurons in the pair must be inhibited. It is therefore possible that SpaRCLIn will be most effective in applications that involve neuronal activation where unilateral manipulations are often sufficient to generate an effect as they are for proboscis extension.

The ability of SpaRCLIn to isolate a given set of neurons of interest in a Gal4 pattern depends critically on the extent to which the various Split Cre components are expressed in the neuroblast lineages of the fly. This will be determined both by the breadth of NB expression of the DNE enhancer used here to delimit Cre activity and by the collective coverage of NB expression provided by the NBEs represented in the libraries of Cre_B and Cre_C lines. Our analysis of 3^rd instar larval expression in DNE∩NBE intersections (Figure 1—figure supplement 1 and data not shown) indicates that many, if not most, NB lineages of the ventral nerve cord and central brain are likely represented within the libraries. Indeed, many lineages are clearly represented multiple times in that different intersections repeatedly isolated the same neurons (e.g. the PE^rk neurons) for both the rk^pan-Gal4 and TH-Gal4 drivers. It is less clear, however, that all members of each lineage are represented as not all NBE’s are active during early NB divisions. This is evident from the restriction in NB expression observed when the FRTerminator construct is used, since this construct acts by eliminating lineages or sublineages in which Cre activity is initiated sometime after neurogenesis has begun. It is also clear that the DNE does not express efficiently in NB lineages in the optic lobe (data not shown). To extend the capability of the system to include these lineages will require either the development of a more general neuroblast-specific enhancer or augmenting the system to include an enhancer that specifically targets optic lobe NBs.

The effectiveness of SpaRCLIn also depends critically on the success of Cre reconstitution by the system, which is effected by two pairs of split inteins (Shah and Muir, 2011; Shah and Muir, 2014). These trans-splicing protein fragments function naturally in protein religation and are an emerging technology for use in transgenic animals (Hermann et al., 2014; Wang et al., 2018; Wang et al., 2012). Their advantages are that they lend themselves readily to intersectional methods, are genetically encoded, and in numerous cases display rapid reaction kinetics and low cross-reactivity. A disadvantage, on which some recent progress has been made (Stevens et al., 2017), is that most split inteins require specific flanking amino acid residues in the proteins to which they are fused, in particular a cysteine or serine residue immediately downstream of the N-intein. We were able to create self-ligating split Cre fragments capable of reconstituting full-length, active Cre enzyme in Drosophila NBs by choosing breakpoints in the Cre sequence preceded by a serine residue—the native condition of the NrdJ-1 and gp41-1 split inteins used here (Carvajal-Vallejos et al., 2012). Orthogonal (i.e. non-interacting) split inteins thus represent attractive tools for reconstituting the function of multiply split proteins, a methodology that should be applicable in other model organisms.

Although sophisticated methods for neuronal targeting have been a hallmark of neurobiological studies on the fly, and single cell manipulations are being leveraged in a growing number of cases to elucidate Drosophila brain circuits, targeting every cell in the fly CNS remains an aspirational goal. Recent progress towards this goal has been made using the Split Gal4 system (Dionne et al., 2018; Tirian et al., 2017), and innovative methods continue to be developed using emerging tools (Garcia-Marques et al., 2019). An advantage of SpaRCLIn is that it represents a relatively small set of stand-alone reagents for high-specificity neuronal targeting that can be used with the many existing components of the Gal4-UAS system. Importantly, SpaRCLIn also represents an open resource that can readily be augmented by end-users. As methods improve for rationally identifying NB lineages based on gene expression and enhancer activity, the existing SpaRCLIn libraries can be supplemented with lines that together permit the selective targeting of an increasing number of neuroblast lineages. The NB-specific enhancers recently identified by Lacin and Truman (2016) and used here to characterize our split Cre components (Figure 1—figure supplement 2D) provide good examples of reagents that can be used to improve the SpaRCLIn libraries. By combining these libraries with an optimized set of Gal4 drivers that express in distinct subsets of brain cells (distinguished, for example, by transcription factor expression), one can imagine having a set of 3 libraries that in combination can selectively target most neurons in CNS.

Vinegar flies of the species Drosophila melanogaster were used in this study. Unless otherwise noted, all flies were grown on BDSC Cornmeal Food and maintained at 25°C in a constant 12 hr light–dark cycle. Both male and female progeny of the genotypes indicated in Supplementary file 2 were used in this study. Previously described fly stocks and their sources are listed in the Key Resources Table. Fly lines generated for this study were made using the DNA constructs described below. Injection of these constructs to produce transgenic flies was carried out by Rainbow Transgenic Flies, Inc (Camarillo, CA). All transgene insertions except the insertion of the DNE-Gal4 were mediated by ΦC31 integrase and placed in the defined attP landing sites indicated in Key Resources Table. Flies made with the DNE-Gal4 were generated by p-element mediated transgenesis. Transgenic flies of the NBE-Cre_B library have transgene insertions on the 2^nd chromosome at attP40, while all flies in the NBE-Cre_C library have insertions on the 3^rd chromosome at either VK00033 or VK00027.

All oligonucleotide and gBlock synthesis was carried out by Integrated DNA Technologies, Inc (Coralville, Iowa), and all final constructs were verified by sequencing (Eurofins Scientific, Louisville, KY or Macrogen Corp, Rockville MD). For routine molecular biology, the following reagents were used according to the manufacturers’ supplied protocols: PCR amplification: Q5 High-Fidelity 2X Master Mix #M0492S (New England Biolabs, Ipswich, MA); DNA ligation: Quick Ligation Kit #M2200L (New England Biolabs, Ipswich, MA); Cloning: Gateway LR Clonase II Enzyme mix #11791100 (Thermofisher Scientific, Waltham, MA), and In-Fusion HD Cloning Plus #638911(Takara Bio USA, Inc, Mountain View, CA). gBlocks were used to generate most of the final and intermediate constructs described below, including the DNA fragments encoding the NrdJ-1 and gp41-1 split inteins and the Cre fragments described in the manuscript. DNA sequences of the split inteins were back-translated from the published protein sequences (Carvajal-Vallejos et al., 2012) and all sequences were codon biased for Drosophila. Sequences of all gBlock fragments and PCR primers are listed in Supplementary file 3. The following reagents, which were used to make several constructs as indicated below, are all described in Pfeiffer et al. (2010): pBPGal80Uw-5, pBPLexA::P65, 10XUAS-IVS-myr::tdTomato, 10XUAS-mCD8::GFP, and pBPGAL80Uw-6.

The indicated Cre80Tom constructs were made stepwise using the described procedures.

All split Cre constructs were made by Gateway cloning (LR reaction). Two sets of destination vectors with split Cre components were made: one for use with entry clones containing promoters, and another for entry clones containing enhancers. The 134 NBE entry clones were combined with the latter to make the expression clones used to generate the Cre_B and Cre_C libraries.

To make the Cre_A (HJP-176) destination vectors for use with promoter entry clones, a KpnI-IVS-NheI fragment made from annealed oligonucleotides, a NheI-gBlock012-AgeI gBlocks fragment and an AgeI-PmeI-WPRE-HindIII PCR fragment (amplified from pBPGAL80Uw-6 using PrimerS472 and PrimerS473) were placed between the NheI and HindIII restriction sites of the pBPGw vector (Addgene Plasmid #17574 Pfeiffer et al., 2008). Other split Cre destination vectors (i.e. HJP177 ~HJP180; see the Key Resources Table) were made by replacing the NheI-Cre_A-gp41-1^N-AgeI fragment in Cre_A (HJP-176) with fragments consisting of: NheI-gBlock010-SphI + SphI-gBlock011-AgeI (HJP177), NheI-gBlock008-BsaI+BsaI-gBlock009-AgeI (HJP178), NheI-gBlock007-AgeI (HJP179), or NheI-gBlock010-SphI+SphI-gBlock015-AgeI (HJP180). To create a set of destination vectors for use with enhancer entry clones (‘the U-series’), an FseI-DSCP-KpnI synthetic core promoter (Pfeiffer et al., 2008) was made from annealed oligos and inserted between the FseI and KpnI restriction sites of each of the destination vectors made for use with promoter entry clones. This produced constructs HJP194 ~196, HJP-207 and HJP-208 (See Key Resources Table).

Prior to the production of transgenic fly lines, the functionality of all Cre constructs was validated in cultured S2 cells by placing the constructs under the control of the Actin5C promoter and testing in appropriate combinations for expression and activity using a floxed reporter construct.

This construct (HJP-473) was made as follows: an AvrII and PmeI flanked DNA fragment (including partial nerfin-1 enhancer, FRT and Syn21-flipase-T2A-Cre_A-gp41-1^N-AgeI-FRT) were synthesized (Epoch Life Science, Inc, Missouri City, TX) and put between the AvrII and PmeI restriction sites of DNE-CreA-gp41-1^N. The resulting construct can be used in place of DNE-Cre_A in Step 2 SpaRCLIn screens. It was tested, but its use is not described in this manuscript. This construct was used as an intermediary to make the final FRTerminator construct by inserting gBlock-043 (part of the CreAB sequence and Nrdj-1N) into its SbfI and AgeI restriction sites using the In-Fusion HD cloning technique.

Two constructs were used to pre-screen candidate enhancers driving Gal4 expression. These included CreStop (HJP225) and UAS-Cre_C (HJP266). The CreStop construct was made using a NgoMIV-loxP-hsp70 terminator-MluI gBlock to replace the loxP-Gal80 in Cre80Tom (HJP223) by In-Fusion HD cloning. UAS-Cre_C was made by cloning a NotI-NrdJ-1^C-Cre_C-XbaI PCR fragment (Primer116 and Primer117; Cre_C as template) between the NotI and XbaI sites of pJFRC1-10XUAS-mCD8::GFP using the In-Fusion HD cloning technique.

Embryos were prepared and immunostained as described by Lécuyer et al. (2008). Excised nervous system whole mounts were prepared from wandering third-instar larvae or adults after dissection into PBS and fixation in 4% paraformaldehyde in PBS for 20–30 min. Immunostaining was done with the antibodies listed in the Key Resources Table at the indicated dilutions. For confocal imaging, all tissues were attached to poly-L-lysine coated cover glass and mounted in Vectashield (Vector Laboratories, Burlingame, CA) prior to imaging with a Nikon C-2 confocal microscope. Z-series were acquired in 0.85 μm increments using a 20X objective using 488 nm, 543 nm or 633 nm laser emission lines for fluorophore excitation. The images shown are maximal projections of volume rendered z-stacks of confocal sections taken through the entire nervous system. NB expression of Gal4 driven by the DNE enhancer was examined in embryonic fillets by in situ hybridizations as previously described (Ross et al., 2015).

Flies assayed for proboscis extension were raised at 25°C until the white prepupa stage and then transferred to 18°C until the time of testing. For neuronal activation using dTrpA1, the chambers were placed on the surface of the Echotherm Chilling/Heating Dry Bath IC25 (Torrey Pines Scientific, Inc, Carlsbad, CA) at 31°C. For the Step 1 SpaRCLIn screen, approximately a dozen adult flies (3–10 d old) of each genotype were placed in glass TriKinetics tubes (3 mm inner diameter; TriKinetics Inc, Waltham, MA) and videorecorded at 31°C for 3 min using a Sony NEX-VG10 videocamera. Proboscis extension behavior was analyzed from these recordings. If two or more flies exhibited robust, full-length extension, the cross was scored as positive. For the Step two tripartite SpaRCLIn screen, two flies at a time (one male and one female) were videorecorded together in glass minichambers (0.3 cm diameter X 0.7 cm length) for 3 min at 18°C followed by 3 min at 31°C. Flies were subjected to these temperature transitions twice and proboscis extension behavior was analyzed following the recording. The criteria for positive proboscis extension was three or more bouts of full proboscis extension in both tests. For the FRTerminator behavior experiments flies were subjected to only one test. Flies used to make the videos included in the manuscript were back-mounted on a 200 uL pipette tip with 5-Minute-Rapid-Curing, General Purpose Adhesive Epoxy (ITW polymers Adhesive, Danvers, MA) and placed just above the heating plate, which was adjusted to apply temperature changes.