A genome-wide analysis reveals that the Drosophila transcription factor Lola promotes axon growth in part by suppressing expression of the actin nucleation factor Spire

Transcriptional regulation is central to the development of neural connectivity [1–4]. First, by controlling the identities and differentiation of individual neurons, and of the non-neuronal cells that often provide them with growth and guidance cues, it specifies patterns of axon growth and dendritic ramification. For example, classic homeotic selector genes produce segment-specific patterns of axon projection by controlling transcription of key cell fate and guidance molecules [5, 6]. At a higher level of specificity, the axonal projections of many individual neurons and classes of neurons are determined by the expression of particular combinations of transcription factors. Perhaps the best-characterized example of this is the 'code' of LIM domain transcription factors that controls the dorsal versus ventral projection of motoneurons from the central nervous system (CNS) in both vertebrates and invertebrates [7–9]. Thus, in Drosophila, for instance, the Lim/Islet/Eve-dependent pattern of motor axon projection arises in part from regulated transcription of a critical axon guidance receptor, the Netrin receptor Unc5 [10]. In addition to specifying axon projection by defining cell identity per se, transcriptional regulation is also key to determining more local properties of axon patterning. For example, recent experiments have shown that the attractive midline receptor DCC in addition to controlling cytoplasmic signal transduction and growth cone motility directly, further contributes to midline axon guidance by inducing expression of commissureless mRNA [11].

Beyond simply expressing the right proteins, however, a neuron must express these proteins at the proper relative levels such that it responds appropriately to precise ratios of specific external and internal cues [12–16]. Again, tight regulation of gene expression is essential to make this happen. The Drosophila transcription factor gene lola is an exemplar of this latter role of transcriptional regulation. Thus, for example, lola orchestrates midline crossing by CNS axons by coordinately regulating the expression of both the midline repellant Slit and also its axonal receptor, Robo [17]. The effects of lola on slit and robo expression individually are modest, but in combination, the cumulative effect is profound, and further enhanced by lola-dependent regulation of other components of the pathway (robo2 and robo3; HJ Song, BJ Taylor and EG, unpublished observations). Similarly, detailed analysis of mutant phenotypes suggested that lola is likely to be a 'master regulator' of the patterning of the ISNb motonerve of Drosophila, coordinately regulating expression of the suite of guidance proteins responsible for ISNb motoneuron/muscle interaction [18].

In addition to its well-characterized functions during early embryonic axonogenesis, lola is also involved in construction or function of neural circuits throughout the lifespan of the fly. This includes dendritic morphogenesis in embryonic sensory neurons [19] and both dendritic and axonal development of adult olfactory projection neurons [20]. Moreover, lola has been identified in a number of genetic and genomic screens for genes involved in complex behaviors, including locomotion, alcohol sensitivity, mating and aggression [21–24]. While the role of lola in these processes is less well understood than is its well-studied function in axon patterning, the varied effects of lola underscore its importance for establishing patterns of gene regulation that produce a properly functioning nervous system. Beyond its neuronal functions, lola also plays key roles in several non-neuronal tissues. For example, lola is essential for oogenesis, in part due to its regulation of programmed cell death at different stages of oogenesis [25].

The lola locus encodes a family of more than 20 transcription factor isoforms by alternative splicing [26–28]. All share a BTB dimerization domain at the amino terminus, allowing co-expressed Lola proteins in a cell to form a variety of heterodimeric species in vivo. At least 17 lola isoforms encode Zn fingers in their variant 3' exons. While some isoforms have a single Zn finger, in most isoforms, the Zn fingers are present as a pair, with an amino-terminal CCHC Zn finger that is apt to be a protein interaction domain, followed by a C2H2 Zn finger that is likely to bind nucleic acid [26]. Biochemical experiments reveal that isoform K binds directly to the transcriptional enhancer of the copia retrotransposon, activating its expression in transient transfection assays [29]. A different isoform, isoform T, has a nearly identical sequence as isoform K in its Zn fingers and binds the same DNA sequence, but it suppresses copia expression, competing with isoform K in co-transfection experiments. Consistent with these evidently antagonistic effects of lola isoforms, expression of the endogenous copia is upregulated in the CNS of lola mutants, but downregulated in the gonads of the same individuals.

To better understand the regulation of axon growth and guidance by lola, we performed a microarray study to identify genes whose transcript level was altered in a lola mutant. In addition to the sole known direct target of Lola, the retrotransposon copia, we found changes in expression level of a substantial number of genes that have been implicated previously in well characterized lola-dependent developmental events, including axon patterning, eye development, Notch signaling and programmed cell death. Among the affected genes was an unexpected downstream target, the gene encoding the actin nucleation factor Spire, which was upregulated in the lola mutant. spire is required for both anteroposterior and dorsoventral patterning of the developing oocyte [30], but its zygotic functions have never been characterized. We found that mutation of spire by itself caused defects in a lola-dependent axon patterning event, extension of the ISNb motonerve, and further that reduction of spire expression, by introduction of a heterozygous spire mutation, suppressed the axonal defects of lola in extension of the ISNb motonerve. These data suggest that overexpression of spire makes a significant contribution to the ISNb axonal phenotype of lola.

The transcription factor Lola is required for a variety of axon growth and guidance events in the developing fly embryo [42, 43]. Expression microarray analysis of lola mutant embryos now reveals that, rather than producing large changes in the levels of a restricted number of major-effect downstream targets, Lola appears to exert its influence via the cumulative effects of small, quantitative changes in a broad spectrum of downstream loci. One key Lola target is spire, which encodes an actin nucleation factor [38] that has been studied intensively for its role in regulating cytoskeletal structure in the developing fly oocyte [39, 40]. spire, like lola, is required for development of ISNb motor axons, its level goes up in lola mutants, and reduction of spire dosage suppresses, but does not eliminate, the ISNb mutant phenotype of lola.

Previous analysis of candidate genes implicated in various lola-dependent axon guidance processes identified several whose expression was subtly affected by lola, but none that were dramatically altered. This and other observations led to the proposition that Lola might execute its effects by fine-tuning the expression levels of genes that contribute quantitatively to various guidance decisions, and not simply by turning these genes ON versus OFF [17]. As an unbiased test of this hypothesis, we used expression microarrays to perform a genome-wide comparison of the embryonic transcriptome of wild-type and lola zygotic mutant embryos. We analyzed RNA isolated from animals 10 to 12 hours after egg laying, at a time when a large number of lola-dependent axons are extending. By this analysis, the expression of no single-copy Drosophila gene was altered more than four-fold by lola, and few were altered more than 2.5-fold. It is possible that this is an underestimate due to the compression of expression ratios in microarray experiments, but qRT-PCR results were largely consistent with the array data. It is also possible that expression of some genes may have been altered by a greater factor in just a small subset of expressing cells, but we note that most lola isoforms are themselves expressed very broadly, making this possibility less likely [26]. Finally, we know that some genes can be affected oppositely by different lola isoforms, or in different tissues [29], so it may be that a small net change in expression of a lola target gene hides larger but counteracting changes in different cells. Nonetheless, it remains that a genome-wide analysis failed to identify any single major-effect lola target that would account for the lola axonal phenotypes. It is also true that there is a substantial maternal contribution of Lola to the embryo [42], and this may limit the measured effect of the mutation on downstream targets. We note, however, that it is the zygotic mutant phenotype of lola that we are seeking to explain, and it is therefore the quantitative effect of that zygotic mutant that is the relevant measurement for investigating the phenotype.

Microarray analysis has been widely used to identify genes associated with, or responsible for, many developmental and physiological processes. Typical analyses of expression microarray data emphasize genes whose level is strongly altered by the biological manipulation, often setting numerical cutoffs for change in expression level, together with statistical criteria, to identify true positives. In our experiments, we were compelled to eschew the use of a quantitative cutoff in fold change; for example, a commonly used criterion of a two-fold minimum change would have excluded from analysis all but 26 single-copy genes in the genome. Rather, the nature of the biological process we studied, and the nature of lola, required that we minimize the biological and technical variance to achieve exceptionally tight statistics. In the end, qRT-PCR validation of expression changes from 1.2-fold (genderblind) to 2.5-fold (spire) provided support for 50% of the putative downstream effects of lola. We note that this is likely to be an underestimate of the reliability of the array results since at these small fold-differences we were at or beyond the usual sensitivity of RT-PCR itself, and it is as likely that RT-PCR was reporting false negatives as the microarrays were reporting false positives. Validity of the results was also supported more globally by independent expression profiling of another lola allele. Thus, these data underscore the efficacy of microarray analysis for detecting reliably even quite small changes in expression level.

Known genes whose expression was altered in the lola mutant shed light on many lola-dependent processes. Previous experiments had led to the notion that lola likely co-regulates a suite of interacting genes that are important for particular axon guidance decisions, and indeed, we find expression of a number of well-characterized guidance receptors to be altered in lola. frazzled, which was identified as a downstream target of lola, is on its own known to be required for three lola-dependent axonal processes: ISNb development in the periphery, and both longitudinal and commissural axon extension in the CNS [44]. Among other factors downstream of lola are midline fasciclin (longitudinal and commissural axons) [45], fasciclin 3 and capricious (ISNb) [46, 47] and neural lazarillo (thought to be involved in both longitudinal and commissural axon guidance) [48]. Also identified were genes for a number of ligands, receptors and receptor-modifying proteins not previously associated with lola-dependent processes, such as sugarless, dallylike, wnt4a and PVF-1. It now becomes interesting to investigate the potential role of these genes in axon patterning, and in migration and orientation of sensory neurons. Aside from cell surface and extracellular proteins, expression of genes encoding a number of intracellular signaling proteins was found to be altered, including prospero (which in hypomorphic alleles produces a phenotype very similar to that of lola (EG, unpublished observations)), as well as moesin, Rac2, and a calmodulin-dependent protein kinase (CAKI). An unexpected cluster of downstream effects comprised genes for proteins modulating microtubule structure and function, including katanin, stathmin, NudC and KLP-59C. lola also interacts genetically with the axon patterning function and other activities of the receptor Notch [49, 50] (EG, unpublished observations), and we find a cluster of affected genes that modulate Notch action, including sca, Nak, Dap-160 and O-fut1. In addition to these known genes, Gene Ontology analysis [37] identifies a large number of lola-dependent loci that have not yet been characterized in the fly, but whose annotations cluster with lola-dependent genes of known function. This provides a substantial list of excellent candidates for additional contributors to lola-dependent processes. Unfortunately, the large number of Lola isoforms, and their heteromeric combinations, makes it impossible for us to extract Lola binding site consensus sequences from these candidates using standard computational approaches. Extensive molecular experiments will be necessary in the future to identify response elements for individual heteromeric forms of Lola.

lola has several characterized functions outside of axon patterning. For example, it affects cell fates in the eye [50], and indeed, there is a substantial group of eye patterning genes included in the list of lola-affected loci (sickle, charlatan, asense, rap, roughex, Lobe, target of eyeless and fat facets). Additionally, consistent with the role of lola in controlling programmed cell death during oogenesis [25], we find grim, scylla, charybde, bunched and Nedd2-like caspase among the downstream effects. It should be noted that since our microarray analysis was performed only with mid-stage embryos, we cannot distinguish whether the effects of lola on these postembryonic processes are mediated by the same downstream targets that we see affected during embryogenesis. Seeing that these genes can be modulated by lola at one stage of the lifecycle, however, makes them more attractive candidates for analysis at other stages. Finally, in addition to genes affecting known lola-dependent processes, the set of genes altered in lola mutants identifies clusters associated with new processes that would be worth investigating for a role of lola. These include aging, oxidative stress, hormonal regulation of development, tracheal development and maintenance, cell polarity and olfactory learning, among others.

One of the most robust putative downstream effects we identified for lola was downregulation of the actin nucleation factor Spire. This was immediately striking since spire is known to be a critical regulator of the oocyte cytoskeleton during Drosophila oogenesis. spire is required for both anteroposterior and dorsoventral patterning of the developing oocyte [30]. By modulating actin structure, Spire restrains bundling of oocyte microtubules, thereby blocking cytoplasmic streaming in the oocyte until critical antero-posterior and dorso-ventral polarity cues become stably bound to cortical anchoring sites or initiate irreversible signaling cascades [39, 40]. At the biochemical level, Spire nucleates actin filaments by bringing together actin monomers to assemble a filament nucleus [38], and it may then transfer this nucleus to the associated formin, Cappuccino, which stimulates filament growth [51]. While the developmental function of spire has been studied most thoroughly in the oocyte, strong mutations in this gene are largely lethal, with only small numbers of escapers surviving to adulthood, and this suggested the existence of as yet uncharacterized zygotic functions of spire. Moreover, a mouse ortholog of spire is expressed in the developing and adult brain [41]. Here we found that spire is required for a well-characterized lola-dependent neuronal process, extension of the ISNb motonerve. ISNb was an ideal candidate for the sort of function we had previously hypothesized for lola, since it is known to depend on the summed, quantitative effects of a large collection of regulators. We therefore exploited ISNb to examine more carefully the potential interaction of lola and spire, and found that genetic reduction of spire suppressed the ISNb mutant phenotype of lola, consistent with the upregulation of spire in a lola mutant making a significant contribution to ISNb axon stalling in lola. By itself, expression analysis cannot distinguish whether spire is a direct target of Lola or whether the upregulation of spire message is a downstream consequence of other changes set in motion by lola. Further biochemical studies of the DNA binding properties of Lola isoforms will be necessary to assess this. Finally, lola has other phenotypes that are not suppressed by reduction of spire. These may reflect, for example, roles of lola-dependent guidance molecules that are themselves spire -independent, or the action of Spire-independent aspects of growth cone signaling.

Efforts to mimic the lola ISNb phenotype by overexpression of spire were not successful. There are several possible reasons for this. First, there are thought to be at least eight Spire protein isoforms, based on cDNA and expressed sequence tag data (Flybase), and it may be that a particular combination of isoforms, or a specific ratio of expression levels of different isoforms, is necessary to give the ISNb stalling phenotype. Alternatively, it may be that this phenotype is produced only when spire upregulation occurs in the context of some other downstream effect(s) of lola. Additional experiments will be necessary to discriminate among these models.

Superficially, it seems remarkable that complete loss of spire causes stalling of ISNb axons, yet the upregulation of spire that occurs in a lola mutant also contributes to ISNb stalling. Evidently, excessive nucleation of actin filaments from spire overexpression is as detrimental to growth cone motility as is the failure of actin nucleation from absence of the protein. We and others have observed similar non-linearity in the effects of a number of signaling and cytoskeletal regulatory proteins in other axon guidance paradigms, and it appears to be a common feature of the relationship of signaling to morphogenesis. Thus, for example, even though Abl tyrosine kinase pathway signaling appears to be essential for most axon growth [52], extension of longitudinal pioneer axons of the fly CNS requires suppression of Abl signaling to achieve the proper balance in the steps of actin dynamics [53]. Similarly, for the Rac GTPases, expression of dominant negative and dominant constitutive forms of the protein are equally effective for inhibiting axon motility, but in one case from excessive stabilization of actin filaments and in the other from insufficient stabilization [54]. Spire now provides another example of this generalization, and underscores the need for signaling networks to evoke a balance in the steps in actin dynamics, thus optimizing throughput through the mechanical cycle of growth cone motility [55].

lola mutants have profound effects on axon patterning, even though systematic molecular analysis reveals only subtle modulation of downstream target gene expression. This observation highlights the exquisite sensitivity of motility and guidance to the balance among cell signaling networks, and thus also to the gene expression mechanisms that set the boundaries of that balance.

We have used a genome-wide analysis to identify the suite of genes whose expression is altered in embryos lacking the Drosophila transcription factor Lola. Gene Ontology analysis sheds light on the regulation of several characterized lola functions, including axon guidance, synapse formation, eye development and oogenesis, by revealing the lola-dependence of genes known to be involved in those processes, and also by identifying a large number of previously uncharacterized lola-dependent genes that are likely to contribute to these processes. Additionally, these results identify novel processes that are likely to be regulated by lola. Regarding axon patterning, our analysis reveals that Lola suppresses expression of the actin nucleation factor Spire, and this is crucial for its ability to promote growth of motor axons in vivo. These data underscore the critical importance of ensuring the correct levels of actin regulatory proteins in a cell to promote motility effectively.