The extracellular metalloprotease AdamTS-A anchors neural lineages in place within and preserves the architecture of the central nervous system

The marriage of form to function is a unifying theme of biology. Yet the genes and pathways that govern tissue architecture remain poorly characterized. The elucidation of these mechanisms is essential for a full understanding of tissue structure, growth and development.

The Drosophila CNS adopts its characteristic shape at the end of embryogenesis and maintains this shape through its massive growth during larval stages. In invertebrates and vertebrates, tissue shape is largely dictated by the basement membrane, a special type of extracellular matrix (ECM) that forms a meshed network of protein polymers that surrounds organs, providing them with structural support and resiliency (Yurchenco, 2011). The fly CNS is covered by a thick basement membrane – the neural lamella – that governs CNS structure via its mechanical properties and interactions with underlying glia (Meyer et al., 2014; Stork et al., 2008). The tremendous growth of the CNS during larval stages places great demands on the neural lamella: it must impose structure on the CNS while continually remodeling itself and its interactions with the underlying glia to allow for its own growth and that of the CNS. These properties, together with the power of Drosophila genetic, make the fly CNS an ideal system in which to dissect the cellular, genetic and molecular mechanisms that govern tissue architecture and homeostasis during growth and development.

From the outside in, the entire Drosophila larval CNS is consecutively enwrapped by the neural lamella, hereafter referred to as the CNS basement membrane, and the membranes of the perineural and subperineural surface glia (Fig. 1); these layers provide structural support to the CNS and insulate it from the hemolymph (Coutinho-Budd and Freeman, 2013). Perineural glia, the outermost cell layer, are small cells that largely arise post-embryonically and divide extensively during larval development to cover the surface of the CNS by the late-third instar stage (Awasaki et al., 2008; Doherty et al., 2009). Perineural glia directly underlie the basement membrane, contribute to the blood-brain barrier (Stork et al., 2008) and help nourish the CNS (Volkenhoff et al., 2015). Sub-perineural glia form an inner-layer of large, polyploid sheet-like cells that associate with each other via septate junctions to form the major component of the fly blood-brain barrier (Bainton et al., 2005; Stork et al., 2008). Internal to these three layers reside thousands of neurons and three other types of glia – cortex, ensheathing and astrocyte-like glia – that enwrap, respectively, the cell bodies, axons and terminal synapses/dendrites of neurons with their cell membranes (Fig. 1) (Awasaki et al., 2008; Doherty et al., 2009). Genetic studies indicate that surface glia control CNS structure: RNAi-mediated depletion of integrin function in surface glia yields an elongated nerve cord, suggesting interactions between surface glia and the basement membrane impose structure on the CNS (Meyer et al., 2014). Few other factors, however, have been shown to act in glia to preserve CNS structure.

The major components of basement membranes include Collagen IV, the heparan-sulfate proteoglycan Perlecan [also called terribly reduced optic lobes (trol) in flies], and Nidogen and Laminin (Fig. 1) (Yurchenco, 2011). These proteins form a multi-layered, meshwork-like covering over most tissues or organs. Tethered to the cell surface largely via interactions between Collagen IV, laminins and cell surface receptors, such as integrins and Dystroglycan, this network provides structure, resiliency and tensile strength to tissues. It also provides a scaffold that helps organize basement membrane assembly and serves as a substrate for cell migration (Yurchenco, 2011). In Drosophila imaginal discs, collagen IV and βPS-integrin appear to promote tissue stiffness or constriction, while perlecan appears to act in an opposite manner to reduce tissue stiffness (Kim et al., 2014; Pastor-Pareja and Xu, 2011). The apparent mutual antagonism between these basement membrane proteins likely underlies the semi-rigid, yet resilient, nature of tissues and organs. Additional unidentified factors almost certainly act with these factors to maintain this dynamic homeostasis, essential for the growth and function of basement membranes during larval development.

Genes of the a disintegrin and metalloproteinase with thrombospondin motifs (AdamTS) family encode a large family of extracellular proteases (Kelwick et al., 2015; Kim and Nishiwaki, 2015; Kuno et al., 1997). Different subfamilies of AdamTS proteins preferentially modify different ECM proteins and regulate diverse biological functions (reviewed by Apte, 2009; Rodriguez-Manzaneque et al., 2015). The Gon-1 subfamily of AdamTS proteins, named after C. elegans gon-1 gene, have been shown to cleave chondroitin sulfate proteoglycans in vertebrates and to promote cell migration in worms and flies (Carlos Rodríguez-Manzaneque et al., 2002; Kelwick et al., 2015; Sandy et al., 2001; Somerville et al., 2003). For example, in C. elegans, gon-1 drives the distal tip cell on its stereotyped ‘U’-shaped migration to form the gonad (Blelloch and Kimble, 1999; Nishiwaki et al., 2000). In flies, the ortholog of gon-1, AdamTS-A, promotes the normal migratory behavior of germ cells, visceral mesoderm, and tracheal cells (Ismat et al., 2013). A role for AdamTS proteins in regulating CNS structure and anchoring cells in place, however, has not been uncovered.

Here, we show that, within Drosophila, AdamTS-A acts in perineural and sub-perineural surface glia to ensure neural lineages remain fixed in place and to maintain the architectural integrity of the CNS by regulating the organization of the CNS basement membrane and opposing the actions of viking/collagen IV and βPS-integrin. As collagen IV and βPS-integrin are known to regulate tissue structure by promoting tissue stiffness (Kim et al., 2014; Pastor-Pareja and Xu, 2011), our work supports a model in which AdamTS-A, a molecular scissors, acts to keep cells in place and maintain the architectural integrity of the CNS during its rapid growth in larval stages by relaxing tissues.

In an EMS-based forward genetic screen of over 2000 independently mutagenized third chromosomes, we uncovered a mutation, runaway¹, that when homozygous led to the exodus of hundreds to thousands of cells out of the CNS through nerves towards and into peripheral tissues, such as eye and leg imaginal discs (Fig. 2A). These invading lineages, often massive in size, are found in the nerves that connect brain lobes to eye discs and the ventral nerve cord to leg discs (Fig. 2A, asterisks), as well as in nerves that connect the nerve cord to other peripheral tissues (Fig. 2B, arrows). The invading cells are composed of neuroblasts (Fig. 2B, arrowhead; Fig. 2C), the stem cells of the CNS and neurons (Fig. 2A-B), indicating that entire neural lineages exit the CNS. This ‘neural invasion’ phenotype, barely detectable in second instar larvae, becomes fully manifest in third instar larvae (Table 1), concomitant with the most rapid growth phase of the Drosophila CNS (Truman and Bate, 1988; Maurange et al., 2008). Despite the size of the invading neural masses, the proliferative index of neuroblasts in runaway¹ larvae appeared similar to wild type (Table S1), indicating the runaway¹ mutation has little effect on cell proliferation. Thus, when homozygous, the runaway¹ lesion severely distorts CNS structure and leads to a mass exodus of neural lineages out of the CNS through nerves to distant tissues.

During the course of our analysis, we fortuitously uncovered a second non-complementing allele of runaway, runaway², tightly linked to the Sb¹ allele (see Materials and Methods). runaway¹ and runaway² are recessive alleles that fail to complement each other for the neural invasion phenotype and an associated 2- to 3-day delay in larval development. These alleles thus identify the same gene. To uncover the molecular nature of runaway, we used the method of Zhai et al. (2003) to map the runaway¹ developmental delay phenotype to genetic map position 57.4±2.0 (Fig. 3A). Deficiency mapping of this region localized the runaway¹ lesion to a 135 kb genomic region uncovered by Df(3R)Exel6174 and excluded from Df(3R)BSC741 and Df(3R)BSC515 (Fig. 3A). Twenty-four protein-coding genes reside in this region, and both runaway alleles failed to complement two alleles of one of these genes – AdamTS-A – for the neural invasion and developmental delay phenotypes. Sequence analysis revealed no alterations to the AdamTS-A open-reading-frame in either runaway allele, but did uncover a Tirant retrotransposon insert in the 5′ most exon of AdamTS-A in runaway¹ and a Zam retrotransposon insert in nearly the same location in runaway² (Fig. 3B). Using primers that flanked the ∼25 kb first intron of AdamTS-A, qRT-PCR analysis revealed that the levels of this AdamTS-A transcript were reduced to 1.3-3.5% of wild-type levels in the CNS of late-third instar AdamTS-A^rnwy1 larvae (P<0.0005; n=3). We infer that these insertions cause the runaway phenotype by reducing the expression and function of the main isoform of AdamTS-A, AdamTS-A[RB], which is predicted to encode a membrane-anchored extracellular protease of an AdamTS subfamily thought to modify and cleave ECM proteins, notably chondroitin sulfate proteoglycans (Carlos Rodríguez-Manzaneque et al., 2002; Kelwick et al., 2015; Sandy et al., 2001; Somerville et al., 2003). We now refer to the runaway alleles as AdamTS-A^rnwy1 and AdamTS-A^rnwy2. As cellular stress is known to enhance retrotransposon jumping in many eukaryotes (Capy et al., 2000), we speculate that our EMS treatment enhanced retrotransposon jumping by inducing cellular stress, leading to the insert of the Tirant element in AdamTS-A in runaway¹.

Despite the severity of the neural invasion phenotype, genetic tests identify both alleles as hypomorphic (Table 2). Flies homozygous for AdamTS-A^rnwy1 or trans-heterozygous for AdamTS-A^rnwy1 and AdamTS-A^rnwy2 eclosed as uncoordinated, sub-fertile adults at ∼60% the expected frequency of wild type (Table 2). In contrast, flies transheterozygous for the AdamTS-A^KO null allele and Df(3R)88F5, a small deficiency of the region (Ismat et al., 2013), are lethal at the second larval instar (Table 2) and exhibit a CNS reduced in size with rare instances of perturbed structure and neural invasion (Fig. 3Cv, arrowhead). Flies transheterozygous for either AdamTS-A^rnwy1 or AdamTS-A^rnwy2 and AdamTS-A^KO or Df(3R)88F5 rarely pupated and eclosed (Table 2). Larvae of these genotypes that reach the late-third larval instar stage exhibit a severe neural invasion phenotype (Fig. 3Ciii), whereas those that fail to reach this stage exhibit a milder neural invasion phenotype, likely due to an earlier block in development [Fig. 3Civ; compare size of leg discs (dotted lines) between Fig. 3Ciii,Civ]. Thus, the runaway phenotype appears to manifest in conditions of reduced, but not abrogated, AdamTS-A function.

The CNS is composed primarily of neurons and glia. To determine whether the genetic function of AdamTS-A is required in neurons or glia, or both, to regulate CNS architecture and keeps cells in place, we used the GAL4/UAS system to deplete AdamTS-A function via RNAi in either all neurons or all glia of otherwise wild-type larvae (Brand and Perrimon, 1993) (see Materials and Methods). When AdamTS-A function was depleted specifically in neurons (elav>AdamTS-A^RNAi; repoGAL80), no larvae exhibited a neural invasion phenotype (n=34; Fig. 4Aiii). In contrast, when AdamTS-A function was depleted specifically in glia (repo>AdamTS-A^RNAi), 48 out of 49 larvae exhibited a strong neural invasion phenotype (Fig. 4Aii). Thus, the genetic function of AdamTS-A is required in glia to maintain CNS structure and ensure neural lineages stay put.

To see whether AdamTS-A expression is sufficient to rescue the AdamTS-A^rnwy phenotype, we drove its expression in all neurons or all glia of otherwise AdamTS-A^rnwy mutant larvae. Here, we used UAS-linked transgenes for AdamTS-A[RB] as well as AdamTS-A[RE] (Ismat et al., 2013), which encodes the smallest AdamTS-A isoform; the encoded proteins are essentially identical, except that AdamTS-PE lacks a membrane anchor. Both transgenes behaved identically in the assays. Expression of AdamTS-A in neurons failed to rescue the neural invasion phenotype of AdamTS-A^rnwy mutant larvae (n≥13/13; Fig. 4Biii) and had no discernible effect on CNS morphology of wild-type larvae (n≥20/20; Fig. 4Ciii). In contrast, expression of AdamTS-A in glia fully rescued the AdamTS-A^rnwy neural invasion phenotype and replaced it with one marked by an elongated nerve cord and flattened brain lobes (n≥15/15; Fig. 4Bii), a phenotype identical to that observed upon glial overexpression of AdamTS-A in wild-type larvae (n≥30/30; Fig. 4Cii). Thus, AdamTS-A expression in glia is sufficient to rescue the AdamTS-A^rnwy phenotype and the ability of AdamTS-A overexpression in glia to elongate the nerve cord and deflate brain lobes suggests AdamTS-A controls CNS structure by relaxing the tissue. We note that the level of AdamTS-A expression is a key determinant of phenotype: low level expression of AdamTS-A in glia, as occurs when AdamTS-A is expressed under the control of the neural-specific elavGAL4 driver in the absence of repoGAL80 (Berger et al., 2007), is sufficient to restore AdamTS-A^rnwy mutant nerve cords to wild-type morphology (Fig. 4Biv).

To determine whether the proteolytic activity of AdamTS-A is essential for its rescuing activity, we repeated the above experiments with protease-dead versions of AdamTS-A – AdamTS-A^E591A for AdamTS-A[RB] and AdamTS-A^E439A for AdamTS-A[RE], both of which behaved identically in the assays. In the zinc-binding metalloproteases of which AdamTS-A is a member, this E-to-A mutation (underlined) in the consensus catalytic protease site (HEXXHXXG/N/SXXHD) reduces enzymatic activity ∼1900-fold without altering protein structure or stability (Crabbe et al., 1994). As observed above, expression of the protease-dead versions of AdamTS-A in neurons failed to rescue the AdamTS-A^rnwy neural invasion phenotype and had no gross effect on CNS morphology in wild-type larvae (not shown). In contrast, expression of the transgenes in glia of wild-type larvae led to an elongated nerve cord (n≥11/11; Fig. 4Civ), a phenotype similar to but less severe than that observed upon overexpression of wild-type AdamTS-A in glia (compare Fig. 4Cii with Civ). Expression of the transgenes in glia of AdamTS-A^rnwy larvae partially rescued the neural invasion phenotype: ∼45% of larvae exhibited a modest neural invasion phenotype and an elongated nerve cord (n=14/31; Fig. 4Bvi), and ∼55% of larvae lacked a neural invasion phenotype and exhibited an elongated nerve cord (n=17/31; Fig. 4Bv). The observation that protease-dead forms of AdamTS-A provide significant, yet partial, rescuing activity suggests AdamTS-A preserves CNS structure and keeps cells in place through protease-dependent and -independent mechanisms. Owing to our inability to generate high-quality AdamTS-A antibodies, we could not determine whether the wild-type and protease-dead AdamTS-A transgenes drove protein expression at similar levels. Thus, it remains formally possible that some of the phenotypic differences observed between expression of the wild-type and protease-dead AdamTS-A transgenes may arise due to differences in protein expression.

At least five different subtypes of glia reside within the CNS (Coutinho-Budd and Freeman, 2013). To pinpoint whether AdamTS-A acts within all or a defined subset of these glia, we generated a transcriptional reporter of the AdamTS-A[RB] transcript from the Mi{MIC}[09007] insert (AdamTS-A^GAL4; Fig. 2B; see Materials and Methods) and used it to drive a nuclear-targeted GFP transgene in AdamTS-A-expressing cells (Fig. 5). Within the CNS, this reporter specifically labeled all perineural and sub-perineural glia (Fig. 5Ai; Fig. S1), which completely and consecutively enwrap the entire surface of the CNS, but did not appear to label cortex or astrocyte-like glia, which are found internally in the CNS, or neurons (Fig. 5Aii-iv; Fig. S1). Owing to the sparse nature of ensheathing glia and their co-mingling with perineural and sub-perineural glia, we cannot exclude that some ensheathing glia also express AdamTS-A. Within perineural and subperineural glia, AdamTS^GAL4 activates gene expression during the second larval instar stage, coincident with when the AdamTS-A neural invasion phenotype first starts to manifest (not shown; Table 1). Identical results were obtained from a second AdamTS-A GAL4 transcriptional reporter generated from the Mi{MIC}[14156] insert (Fig. 3B; Fig. S2). And, RNA in situ analysis of AdamTS-A transcript confirmed that these transcriptional reporters faithfully recapitulated endogenous AdamTS-A expression (Fig. S2).

To test whether AdamTS-A function is required in either perineural or sub-perineural glial cells, or both, to preserve CNS architecture, we depleted AdamTS-A function in either cell type of both cell types via RNAi. Depletion of AdamTS-A function in either perineural glia (PG-GAL4) or sub-perineural glia (SPG-GAL4) failed to elicit a neural invasion phenotype (n>0/15 larvae per genotype; Fig. 4Di,Dii), but depletion of AdamTS-A function in both glial subtypes yielded a mild neural invasion phenotype ∼90% of the time (n=21/23; Fig. 4Diii). We infer that AdamTS-A acts in both perineural and subperineural glia to anchor neural lineages and preserve CNS structure. Consistent with this model, AdamTS-A expression in subperineural glia, but not cortex glia, was sufficient to rescue the AdamTS-A neural invasion phenotype (Fig. 4Div,Dv), although AdamTS-A expression in perineural glia resulted in embryonic or early larval lethality.

In AdamTS-A^rnwy larvae, perineural and sub-perineural glia form, survive and completely enwrap the CNS, including the invading neural masses that are minimally composed of neurons, neuroblasts and cortex glia (Fig. 5B; Fig. S3). All other glial subtypes also form, survive and differentiate (Fig. S3). Thus, the neural lineages and cortex glia that inappropriately exit the CNS in AdamTS-A^rnwy mutant larvae typically do not break through any glial cell layers in the CNS or the basement membrane. Rather, they tunnel through the nerves out of the CNS toward peripheral tissues, fully encased by the surface glia the entire time (Fig. 5B; Fig. S3). As surface glia appear to be the only cells that express AdamTS-A in the CNS, reduction of AdamTS-A function in these two cell types may trigger, apparently in a non-autonomous manner, the underlying neural lineages and cortex glia to exit the CNS.

Basement membranes regulate tissue structure and cell migration, the two key processes disrupted in the CNS upon reduction of AdamTS-A function. In flies, the CNS basement membrane directly overlies the surface glia, and mammalian orthologs of AdamTS-A are known to associate with and cleave basement membrane proteins (Apte, 2009; Sandy et al., 2001). Thus, we first asked whether reduction in the function of three key basement membrane constituents – laminins, Perlecan and Viking (one of two type IV collagens in flies) – and a main cellular receptor for laminins and Collagen IV, βPS-Integrin (myospheroid), yielded phenotypes similar to those observed for AdamTS-A. Reduction of laminin A or laminin B function during larval stages did not grossly disrupt CNS morphology (not shown), but reduction of function in perlecan [originally identified as terribly reduced optic lobes (trol)] led to CNS phenotypes similar to those of AdamTS-A^rnwy mutant larvae (Fig. 6A-B): the presence of ectopic neural masses in the optic stalk and protrusions of varied size from the brain and nerve cord. These phenotypes were highly penetrant in larvae depleted for trol function via RNAi (n=21/22) and in trol^G0271 hypomorphic mutant larvae (n=20/21) (Fig. 6A,B), where larvae reached the third-larval instar stage, but were less penetrant in larvae mutant for stronger trol alleles, such as trol⁸ (n=8/22) and trol^SD (n=1/5), which typically arrest in the second-larval instar stage (Fig. S4; not shown).

To explore the functional relationship between AdamTS-A and perlecan, we asked whether overexpression of each gene modified the loss of function phenotype of the other. We found that AdamTS-A overexpression in glia in trol^G0271 (n=25/25) or trol^SD (n=10/10) mutant larvae gave rise to an elongated nerve cord (Fig. 6D; Fig. S4). Thus, perlecan is not required for AdamTS-A to exerts its effects on CNS structure. AdamTS-A overexpression, however, did not suppress perlecan-dependent growth defects, as trol^SD mutant larvae remained arrested in the second larval-instar stage even when AdamTS-A was overexpressed (Fig. S4). Similarly, perlecan overexpression in the fat body, a center of basement membrane protein synthesis (Pastor-Pareja and Xu, 2011), suppressed the AdamTS-A neural invasion phenotype (Fig. 6C, Table 3), even though perlecan overexpression on its own had no gross effect on CNS structure. Thus, perlecan can modify CNS structure when AdamTS-A function is limiting. We infer that AdamTS-A and perlecan act in parallel to regulate CNS structure and anchor cells in place.

In contrast to perlecan, knockdown of viking function or glial-specific knockdown of βPS-integrin function yielded a phenotype similar to that observed upon overexpression of AdamTS-A in glia (Fig. 6H) – an elongated ventral nerve cord (Fig. 6F,G). The similarity of the phenotypes suggests AdamTS-A acts opposite viking and βPS-integrin to regulate CNS structure. If viking and β-integrin oppose AdamTS-A function during CNS development, reduction of their function should suppress the AdamTS-A^rnwy neural invasion phenotype. In support of this model, a small deficiency, Df(2L)BSC172, that removes viking and the other type IV collagen gene in Drosophila dominantly suppressed the AdamTS-A^rnwy neural invasion phenotype (Table 3). More impressively, knockdown of viking function or glial-specific knockdown of β-integrin function strongly suppressed the AdamTS-A^rnwy neural invasion phenotype (Table 3, Fig. 6I-K). Confirming the specificity of these interactions, mutations in laminin A, laminin B or three other ECM or cell adhesion molecules failed to reproducibly suppress the AdamTS-A neural invasion phenotype (Table 3), and RNAi-mediated depletion of laminin A, laminin B, ras or raf function also failed to suppress the AdamTS-A neural invasion phenotype (Table 3). We infer that the neural invasion phenotype observed in AdamTS-A^rnwy larvae requires collagen IV and βPS-integrin function, and that AdamTS-A normally opposes the action of viking/collagen IV and βPS-integrin to preserve CNS structure and anchor neural lineages in place.

The opposing actions of AdamTS-A and viking/collagen IV led us to examine the effect of reducing or increasing AdamTS-A function on the distribution of Viking, as well as Perlecan and Laminin in the CNS basement membrane using antibodies specific to Perlecan and Laminin (Fessler et al., 1987; Friedrich et al., 2000) and a Viking-GFP protein trap (Morin et al., 2001). In wild type, Viking-GFP, Perlecan and Laminin display a smooth, uniform laminar-like localization along the surface of the brain lobes and ventral nerve cord (Fig. 7A,C; Fig. S5). In AdamTS-A^rnwy mutant larvae, these proteins accumulate on the surface of the CNS, but are also found in irregularly shaped puncta or aggregates at or near the surface of perineural glia (Fig. 7B,D; Fig. S5). Western blot assays confirmed increased Viking-GFP levels in AdamTS-A^rnwy larvae relative to wild type (Fig. 7H). Of note, we observe these puncta of ECM material in all AdamTS-A^rnwy backgrounds, including those in which we deplete viking and βPS-integrin function. This phenotype is then independent of the AdamTS-A neural invasion phenotype, but appears tightly linked to reduction of AdamTS-A function.

In support of the model that AdamTS-A inhibits the accumulation of Viking-GFP/Collagen IV in the CNS basement membrane, overexpression of AdamTS-A triggered a strong decrease in Viking-GFP accumulation in the CNS basement membrane (Fig. 7E-G). This phenotype was accompanied by increased Perlecan levels but not Laminin levels (Fig. 7E-G). Western blot assays confirmed decreased Viking-GFP levels in the CNS basement membrane upon AdamTS-A overexpression (Fig. 7H). We infer that AdamTS-A opposes tissue stiffness at least in part by opposing Viking/Collagen IV accumulation in the CNS basement membrane.

Our work, coupled with the molecular nature of AdamTS-A and the tissue organizing role of the basement membrane, indicate that AdamTS-A acts in surface glia to preserve CNS structure and keep neural lineages in place by opposing the actions of collagen IV/viking and βPS-integrin. In this role, our work suggests AdamTS-A acts in parallel to perlecan, which yields a similar CNS phenotype to AdamTS-A, and has been shown to function opposite collagen IV/viking to inhibit tissue stiffness in imaginal discs (Pastor-Pareja and Xu, 2011). These findings point to a finely tuned balance between the actions of AdamTS-A and perlecan, and those of collagen IV and βPS-integrin, being crucial for homeostatic control of CNS structure. On one side of the balance, collagen IV/viking and βPS-integrin function to promote tissue or matrix stiffness; on the other side, AdamTS-A and perlecan function to oppose the actions of collagen IV/viking and βPS-integrin, and thus inhibit tissue or matrix stiffness. Imbalance in either direction triggers gross defects in CNS structure: flattened brain lobes and an elongated nerve cord when the balance tilts towards AdamTS-A and perlecan, and gross structural defects and neural invasion when the balance tilts toward collagen IV/viking and βPS-integrin. Thus, opposing, yet balanced, molecular forces appear to help bestow on the CNS basement membrane its unique combination of stiffness, compliance and resiliency that is essential for preservation of CNS structure during its rapid growth in larval stages.

Reduction of AdamTS-A function triggers the mass exodus of scores of neuronal lineages and cortex glia out of the CNS through nerves towards and onto peripheral tissues, such as eye and leg imaginal discs. Throughout their journey, the invading cells appear fully encased by surface glia and the basement membrane and thus likely tunnel through nerves toward these peripheral tissues. As AdamTS-A appears to be expressed and function in surface glia, but not in neurons or cortex glia – the cells that actually invade the nerves – our data raise the intriguing possibility that AdamTS-A acts non-autonomously to drive the exodus of neural lineages from the CNS.

How might loss of AdamTS-A in surface glia trigger the apparent migration of neuronal lineages and cortex glia out of the CNS? We find that collagen IV/Viking and βS-integrin function are required for the neural invasion phenotype observed in AdamTS-A^rnwy mutant larvae and that AdamTS-A negatively regulates Collagen IV/Viking levels on the CNS basement membrane. In addition and as noted, viking and βPS-integrin have been shown to promote tissue stiffness or tension in Drosophila (Kim et al., 2014; Pastor-Pareja and Xu, 2011), and work from vertebrate systems reveals that increased tissue or matrix stiffness can induce cell migration and cancer metastasis (Kai et al., 2016; Lo et al., 2000), likely at least in part by activating Rho-Actin signaling and the SRF pathway (Medjkane et al., 2009). Thus, reduction of AdamTS-A function in surface glia may allow for increased collagen IV/viking and βPS-integrin function and a resulting increase in tissue or matrix stiffness, which could trigger Rho-Actin signaling and the SRF pathway to drive cell migration. Alternatively, reduction of AdamTS-A function, and its apparent ability to relax tissue stiffness, may inhibit the ability of the CNS basement membrane to breathe and expand as the CNS grows rapidly during the third larval instar stage, leading to a compressive stress that causes groups of neural lineages to become passively squeezed out of the CNS through nerves. Clearly, future work that tests the effect of AdamTS-A on matrix stiffness and the role of Rho-Actin signaling and the SRF pathway in promoting the neural invasion phenotype will help clarify the molecular basis through which AdamTS-A anchors cells in place in the Drosophila CNS.

Our work raises the possibility that AdamTS-A preserves CNS structure and anchors cells in place through proteolytic-dependent and -independent activities, as protease-dead versions of AdamTS-A appear to provide partial, yet significant, rescuing activity. In flies, current substrates of AdamTS-A remain largely unknown. Our work, however, identifies Collagen IV/Viking as a potential substrate: Viking levels in the CNS are increased in AdamTS-A^rnwy loss-of-function mutants and decreased upon AdamTS-A overexpression in glia. Clearly, it will be important to see whether Viking is a substrate of AdamTS-A and to identify additional AdamTS-A substrates and interacting proteins. In this context, we note that the most closely related mammalian homologs of AdamTS-A have been found to cleave chondroitin-sulfate proteoglycans (CSPGs) rather than Collagen IV (Somerville et al., 2003), ECM molecules known to repress axon guidance and regeneration and neural cell migration (Burnside and Bradbury, 2014; Zimmer et al., 2010). In flies, AdamTS-A may then also act through CSPGS to govern CNS structure and cell movement. Clearly, directed proteomic and biochemical studies are required to identify AdamTS-A substrates and interacting proteins and to clarify the precise molecular basis through which AdamTS-A preserves CNS structures and keeps neural lineages rooted in place.

Why does reduction, but not elimination, of AdamTS-A or perlecan function disrupt CNS structure and trigger neural invasion? Elimination of AdamTS-A or perlecan activity blocks larval development in the first or second larval instar stage, yielding a small, but largely morphologically normal, CNS (Datta, 1995; Datta and Kankel, 1992; this paper). Prior work indicates that perlecan function is required to reactivate neuroblast proliferation and promote CNS growth during larval stages (Datta, 1995). The small size of the CNS of larvae homozygous for null alleles of AdamTS-A suggest it too is required for larval neuroblast reactivation and CNS growth during larval stages. In contrast, reduction of AdamTS-A or perlecan function allows for continued neurogenesis and CNS growth during larval life and also triggers the neural invasion phenotype and the gross distortion of CNS structure. CNS growth may then be a pre-requisite for these phenotypes to manifest, as only in the presence of CNS growth might deficits in basement membrane function become manifest. For example, if AdamTS-A is required for the basement membrane to breathe and expand to keep pace with the growth of the CNS, reduction of AdamTS-A function might specifically inhibit basement membrane growth. Such a result might increase tissue tension on the growing CNS, perhaps indirectly triggering the mass exodus of neural lineages out of the CNS and the associated mass distortion of CNS structure.

Superficially, the AdamTS-A^rnwy phenotype resembles the phenomenon of perineural invasion, a major mode of metastasis, particularly in head and neck cancers, in which cancer cells migrate along or through nerves to distant sites (Liebig et al., 2009). This similarity hints that reduction of AdamTS function in mammalian glial cells may, in some contexts, contribute to perineural invasion and other modes of cancer spread. In support of this idea, like surface glia in flies, the glia limitans in mammals, a thin layer of astrocytic cellular processes overlain by the parencyhmal basement membrane, serves as the outermost layer of the CNS, fully enwrapping the CNS and spinal cord (Liu et al., 2013). In addition, AdamTS-1, AdamTS-4, AdamTS-5 and AdamTS-9, the mammalian genes most closely related to AdamTS-A, are expressed in astrocytes and glia, but not in neurons, in the mammalian CNS (Demircan et al., 2013; Tauchi et al., 2012). And, in humans reduction in the expression of AdamTS-9 has been associated with breast, pancreatic, colorectal and gastric cancer (Porter et al., 2004; Zhang et al., 2010), and specifically with cancer metastasis in nasopharyngeal cancer, a head and neck cancer (Lung et al., 2008). Clearly, it will be critical to determine whether AdamTS gene function is relevant to perineural invasion in humans and whether from flies to humans the AdamTS genes play conserved roles in the outermost cells of the CNS to preserve CNS structure and anchor neural lineages in place.

A standard autosomal recessive forward genetic screen was carried out using 30 µm EMS to mutagenize a P{w+}FRT2A P{neo+}FRT82B isogenic chromosome. Homozygous mutant late-third instar larvae were manually dissected and scored for defects in CNS morphology.

The UAS-linked AdamTS-A[RB] wild-type transgene was generated by amplifying region 905-5971 of the AdamTS-A cDNA, GH16393 (accession number AY094716). During PCR amplification, the 21-nucleotide sequence for Syn21, a translational enhancer effective in Drosophila (Pfeiffer et al., 2012), was added to the 5′ primer immediately upstream of the ATG start codon. The resulting PCR product was then cloned directionally into the KpnI and XbaI sites of pJFRC28 (Pfeiffer et al., 2012) using Gibson Assembly Cloning Kit (New England Biolabs). Germline integrants of the UAS-AdamTS-A[RB] construct were then generated via phi-C31-mediated cassette exchange into the P{CaryP}attp2 site in the third chromosome (Rainbow Transgenic Flies). The UAS-linked AdamTS-A[RB] (AdamTS-A^E591A) protease-dead transgene was generated in an essentially identical manner and inserted into the same third chromosomal P{CaryP}attp2 site.

For meiotic mapping, (i) P{w[+]}BG00690, (ii) P{w[+]}Alh[BG02270], (iii) P{w[+]}cher[BG02734] and (iv) P{w[+]}BG01881 strains were used.

For complementation crosses and deficiency mapping, TM3 Sb¹, TM6C Sb¹, Gla Sb¹, Sb^spi, Sb²⁵⁶³, Sb^sbd-2, Sb^63b, Df(3R)BSC750, Df(3R)ED5664, Df(3R)BSC741, Df(3R)Exel6174, Df(3R)BSC515, Df(3R)BSC569, Df(3R)88F5 (Ismat et al., 2013), AdamTS-A^K0 (Ismat et al., 2013) and AdamTS-A^MB04384 strains were used.

To restrict UAS-linked transgene expression specifically to glia, we used the repoGAL4 driver line. To restrict UAS-linked transgene expression specifically to neurons, we used a fly strain that carried both the elavGAL4 driver line, which drives high level gene expression in neurons and lower level expression in glia, with repoGAL80, which blocks GAL4-mediated gene activation in glia. To drive UAS-linked transgenes ubiquitously, we used the tubP-GAL4 driver line.

In the GAL4/UAS experiments that drove UAS-linked transgenes for AdamTS-A, we used two wild-type and two protease-dead AdamTS-A transgenes: in each case, one transgene encoded for the AdamTS-A[RB] transcript, which is specifically disrupted in the AdamTS-A^rnwy alleles and is predicted to produce a protein that contains an N-terminally, membrane-anchored protein, and the other transgene encoded for the AdamTS-A[RE] transcript (Ismat et al., 2013), which is predicted to produce a protein that lacks a membrane domain. The two wild-type transgenes and the two protease-dead transgenes behaved essentially identically in the noted assays. The n values provided for each experiment are the minimum number of samples assayed for each transgene.

The GAL4 lines used the study were: (i) repoGAL4 UAS-CD8-GFP/TM6 Tb, (ii) repoGAL4 UAS-CD8-GFP AdamTS-A^rnwy2/TM6B Tb, (iii) repoGAL80/CyO-Dfd-GFP; elavGALl4 (Awasaki et al., 2008; Lin and Goodman, 1994), (iv) repoGAL80/CyO-Dfd-GFP; elavGAL4 AdamTS-A^rnwy2/TM6B Tb, (iv) NP577GAL4 (specific for cortex glia; Hayashi et al., 2002), (v) NP577GAL4; AdamTS-A^rnwy1/TM6B Tb, (vi) RL82-GAL4 (specific for subperineural glia; Sepp and Auld, 1999), (vii) RL82-GAL4; AdamTS-A^rnwy1/TM6B Tb, (viii) SPG-GAL4 (specific for sub-perineural glia; Silies et al., 2007), (ix) SPG-GAL4; AdamTS-A^rnwy1/TM6B Tb, (x) NP6293-GAL4 (specific for perineural glia; Hayashi et al., 2002), (xi) NP6293-GAL4; AdamTS-A^rnwy1/TM6B Tb, (xii) mZ0709-GAL4 (specific for ensheathing glia; Ito et al., 1995), (xiii) mZ0709-GAL4 AdamTS-A^rnwy2/TM6B Tb, (xiv) ALRM-GAL4 (specific for astrocyte-like glia; Doherty et al., 2009), (xv) ALRM-GAL4; AdamTS-A^rnwy1/TM6B Tb, (xvi) tubP-GAL4/TM6 Tb (Lee and Luo, 1999) and (xvii) ppl-GAL4 (Colombani et al., 2003).

UAS-linked transgenes used in the study were: (i) UAS-AdamTS-A[RB] (this paper) and UAS-AdamTS-A[RE] (Ismat et al., 2013), (ii) UAS-AdamTS-A^E591A (Transcript RB; this paper) and UAS-AdamTS-A^E439A (Transcript RE; Ismat et al., 2013), (iii) UAS-AdamTS-A^RNAi (TRiP.HMJ21762); UAS-DCR2/TM6B Tb, (iv) UAS-AdamTS-A^RNAi (VDRC.KK.110157), (v) UAS-AdamTS-A^WT; Df(3R)88F5/TM6B Tb, (vi) UAS-AdamTS-A^E439A; Df(3R)88F5/TM6B Tb, (vii) UAS-NLS-GFP (BSC#4775), (viii) UAS-Ras^DN (UAS-Ras^N17; BSC#4846); AdamTS-A^rnwy1/TM6B Tb, (ix) UAS-Raf^DN (de Celis, 1997); AdamTS-A^rnwy1/TM6B Tb, (x) UAS-mys^RNAi (TRiP.JF02819) AdamTS-A^rnwy1/TM6B Tb, (xi) UAS-mys^RNAi (TRiP.HMS00043) AdamTS-A^rnwy1/TM6B Tb, (xii) UAS-trol^RNAi (TRiP.JF03376), (xiii) UAS-perlecan[RG] (Cho et al., 2012) and (xiv) UAS-viking^RNAi (TRiP.HMC02400).

Additional lines used in the study were: (i) trol^G0271/FM7-GFP, (ii) trol⁸/FM7-GFP, (iii) trol^SD/FM7-GFP, (iv) Laminin A^MI06740 AdamTS-A^rnwy1/TM6B Tb, (v) Df(2L)BSC172/CyO dfd-GFP; AdamTS-A^rnwy1/TM6B Tb, (vi) viking^G00454 (viking-GFP), (vii) viking^G07914 (viking-GFP); AdamTS-A^rnwy1/TM6B Tb, (viii) shg^R69/CyO dfd-GFP; AdamTS-A^rnwy1/TM6B Tb, (ix) Dg^B48/CyO dfd-GFP; AdamTS-A^rnwy1/TM6B Tb and (x) sdc^K10215/CyO dfd-GFP; AdamTS-A^rnwy1/TM6B Tb.

We identified a second allele of runaway carried in the TM3 Sb¹, TM6C Sb¹ and Gla Sb¹ chromosomes, but not in chromosomes that harbored Sb^spi, Sb²⁵⁶³, Sb^sbd-2 or Sb^63b. runaway² is likely l(3)89Aa (Nelson and Szauter, 1992) and was independently identified as an AdamTS-A allele by Ismat et al. (2013).

AdamTS-A^GAL4 lines were generated from AdamTS-A^MI09007 and AdamTS-AMI¹⁴¹⁵⁶ (Nagarkar-Jaiswal et al., 2015) using the TROJAN-GAL4 system (Diao et al., 2015), following the cross scheme as detailed by Diao et al. (2015), except that TM6 Tb was used in place of TM3 Sb due to the presence of the AdamTS-A^rnwy2 allele in the TM3 Sb chromosome.

Flies of appropriate genotypes, e.g. AdamTS-A^rnwy1/TM6B Tb × Df(3R)88F5/TM6B Tb, were allowed to lay eggs for 12 h in vials. Vials were checked daily for Tb and non-Tb pupae and adults. Expected number of non-Tb pupae and adults was calculated as one-half observed number of Tb pupae or adults.

DNA encoding amino acids 78-431 of Deadpan was cloned into pET29a (Novagen) for protein expression and purification. Protein-specific antibody responses were mounted in guinea pigs (Pocono Rabbit Farm and Laboratory, PA, USA) and the resulting sera was used at a 1:500-1:1000 dilution. The resulting anti-sera specifically recognize neuroblasts in the embryonic and larval CNS.

The following antibodies and molecular probes were used in this study: rabbit anti-GFP (1:1000, Torrey Pines Biolabs), mouse monoclonal antibody 9F8A9 (ELAV; 1:20) and rat monoclonal antibody 7E8A10 (ELAV; 1:100) (O'Neill et al., 1994), mouse monoclonal antibody 8D12 (REPO; 1:100) (Alfonso and Jones, 2002), rabbit anti-Laminin (1:1000) (Fessler et al., 1987), rabbit anti-Perlecan (1:1000) (Friedrich et al., 2000), goat anti-HRP-Cy5 (1:1000; Jackson Labs) and DAPI (1:1000; Calbiochem). Gene expression analyses were performed as described by Patel (1994), using the noted antibodies and molecular probes. All imaging was performed on a Zeiss LSM-700 Confocal Microscope and Zen software.

When protein expression levels were compared between genotypes (e.g. for Viking-GFP, Perlecan and Laminin), brain-imaginal disc complexes were dissected at the same time, fixed and stained in the same tube, mounted on the same slide and imaged sequentially using identical parameters on a Zeiss LSM 700 Confocal microscope.

Genomic DNA was obtained from P{w+}FRT2A P{neo+}FRT82B larvae or larvae homozygous or hemizygous mutant, respectively, for AdamTS-A^rnwy1 or AdamTS-A^rnwy2 and provided to GENEWIZ or GTAC (Washington University) for Sanger or next-generation sequencing.

The following protocol was carried out in triplicate for each genotype. Approximately 80-100 CNS-imaginal disc complexes from wild-type (P{w+}FRT2A P{neo+}FRT82B) or AdamTS-A^rnwy1 were dissected on ice and then immediately homogenized in TRIzol reagent (Life Technologies). RNA was extracted following standard protocols, treated with DNase I, and then quantified. For each qRT PCR reaction, 500 ng of total RNA was reverse transcribed using random primers (Promega) and the SMARTScribe Reverse Transcriptase (Clontech). qRT-PCR was then performed on rp49 as an internal control and the AdamTS-A[RB] transcript. The AdamTS-A primers flanked the ∼25 kb first intron of the largest and main AdamTS-A[RB] transcript. Two technical replicates and no-RT controls were performed for each RNA sample for each genotype. rp49 primers were 5′-CCGCTTCAAGGGACAGTATCTG and 3′-ATCTCGCCGCAGTAAACGC; and AdamTS-A primers were 5′-CTGCGGTATTCGCTGAAAGGAC and 3′-CGATGCAGCAGGCATACAGG.

We thank Drs Swathi Arur, Heather Broihier and Andrea Page-McCaw for critical comments on the manuscript. We thank Drs Deborah Andrews, Afshan Ismat, Marc Freeman, Stefan Baumgartner, Steve Johnson, Andrea Page-McCaw, Tim Schedl and Craig Micchelli for fly lines and reagents. We thank the Iowa Developmental Studies Hybridoma Bank for antibodies, and Kathy Matthews, Kevin Cook and the Bloomington Stock Center for countless fly lines.