Gene regulatory networks that control the terminally differentiated state of a cell are, by and large, only superficially understood. In a mutant screen aimed at identifying regulators of gene batteries that define the differentiated state of two left/right asymmetric C. elegansgustatory neurons, ASEL and ASER, we have isolated a mutant, fozi-1,with a novel mixed-fate phenotype, characterized by de-repression of ASEL fate in ASER. fozi-1 codes for a protein that functions in the nucleus of ASER to inhibit the expression of the LIM homeobox gene lim-6,neuropeptide-encoding genes and putative chemoreceptors of the GCY gene family. The FOZI-1 protein displays a highly unusual domain architecture, that combines two functionally essential C2H2 zinc-finger domains, which are probably involved in transcriptional regulation, with a formin homology 2(FH2) domain, normally found only in cytosolic regulators of the actin cytoskeleton. We demonstrate that the FH2 domain of FOZI-1 has lost its actin polymerization function but maintains its phylogenetically ancient ability to homodimerize. fozi-1 genetically interacts with several transcription factors and micro RNAs in the context of specific regulatory network motifs. These network motifs endow the system with properties that provide insights into how cells adopt their stable terminally differentiated states.

The differentiated state of a cell is defined by the array of genes (gene battery') expressed by that cell (Davidson,2001). The mechanisms controlling the gene expression profile of a single cell are sometimes surprisingly complex. Feedback, feed-forward and multi-tier parallel pathways are required to attain, stabilize and maintain the complete signature of a cell-specific program of gene expression. Though these types of regulatory networks have been rigorously described in single cell organisms (e.g. Lee et al.,2002; Shen-Orr et al.,2002), a thorough analysis of complex regulatory networks in multicellular organisms has only been described in few isolated examples(Davidson, 2001). In complex metazoan organisms, the functional analysis of gene regulatory networks are often complicated by various factors, including the pleiotropies caused by mutations in regulatory genes and the unavailability of a sufficient number of molecular markers that define cellular fates on a single cell level.

To address the mechanisms controlling single cell-specific gene expression programs, we have characterized the neuronal subclass diversification process executed by the ASE class of gustatory neurons in C. elegans. This neuron class consists of a pair of two bilaterally symmetric neurons, ASE left(ASEL) and ASE right (ASER) (Fig. 1A). ASEL is the primary sodium sensor, whereas ASER is the primary Cl- and K+ sensor(Pierce-Shimomura et al.,2001). The left/right asymmetric (lateral') separation of chemosensory capacities endows the worms with the ability to discriminate between distinct chemosensory cues and thereby widens its chemosensory repertoire. Other than these functional differences, the ASE pair of neurons are bilaterally symmetric by anatomical criteria such as cell body position,axodendritic structure and synaptic connectivity(White et al., 1986).

ASEL and ASER subclass diversification is defined by an array of left/right asymmetrically expressed cell fate markers including specific subsets of guanylyl cyclase receptors, encoded by GCY genes, and FMRFamide-type neuropeptides, encoded by FLP genes, (Fig. 1A) (Johnston et al.,2005; Ortiz et al.,2006; Yu et al.,1997). We have previously shown that a network of micro RNAs(miRNAs) and transcription factors controls ASEL/R cell fate diversification(Fig. 1A)(Chang et al., 2004; Chang et al., 2003; Hobert et al., 1999; Johnston and Hobert, 2003; Johnston et al., 2005; Johnston and Hobert, 2005). A bi-stable feedback loop constituted by these regulatory factors is an essential network motif required for the establishment and stabilization of lateral ASE fate. In this loop, ASEL-specific inducer genes, such as the lsy-6 miRNA and die-1 transcription factor, activate expression of other ASEL-specific inducer and effector genes and repress ASER-specific inducer and terminal genes. By contrast, ASER-specific inducer genes such as the mir-273 miRNA and cog-1 transcription factor control expression of ASER-specific genes and repression of ASEL-specific genes. Genetic experiments indicate that lsy-6 and its upstream activator, the Zn-finger transcription factor lsy-2,provides the input into the loop, whereas the output of the loop is provided by the Zn-finger transcription factor die-1(Johnston et al., 2005; Johnston and Hobert, 2005)(Fig. 1A).

In this paper, we describe the cloning and characterization of fozi-1, a novel protein containing two Zn fingers and a single FH2 domain. This gene plays a role downstream of the bi-stable feedback loop to repress expression of ASEL-specific effector genes in ASER. We anticipate that the complex regulatory network described here will provide a paradigm for subtype fate specification in other cellular contexts.

### Strains, DNA and transgenes

Strains used for mapping, RNAi sensitization and manipulation of ASE development, as well as transgenes used to assess or manipulate ASE fate have all been previously described (Chang et al., 2004; Chang et al.,2003; Hobert et al.,1999; Hodgkin and Doniach,1997; Johnston and Hobert,2003; Johnston et al.,2005; Simmer et al.,2002).

### List of new transgenes

All contain the elt-2::gfp(Fukushige et al., 1998) as injection marker and are named as follows: otEx2179-2182, Ex[fozi-1 genomic; elt-2::gfp]; otEx2190-2193, Ex[fozi-1::gfp;elt-2::gfp]; otEx2512-2515, Ex[gcy-5prom::fozi-1cDNA::gfp;elt-2::gfp]; otEx2523-2526,Ex[gcy-5prom::fozi-1cDNAΔZn-finger::gfp;elt-2::gfp]; otEx2516, 2518, 2520,Ex[gcy-5prom::fozi-1cDNAΔFH2::gfp; elt-2::gfp];and otEx2534, Ex[ceh-36prom::fozi-1cDNA::gfp;elt-2::gfp].

### DNA for transgenic lines

The genomic fozi-1 locus was amplified from N2 genomic DNA using the primers 5′-CACCCCAAGATGGTAGTAATCC and 5′-GAAGAACTGGACAATTCGG.

#### fozi-1::gfp

fozi-1::gfp was generated by PCR fusion(Hobert, 2002). The first PCRs were carried out on genomic DNA with 5′-GGAGTGGACGATGACATTGTG and 5′-AGTCGACCTGCAGGCATGCAAGCTAGGAGACGAGACATTGATGTG and on the gfpsequence in pP95.75 with 5′-CACATCAATGTCTCGTCTCCTAGCTTGCATGCCTGCAGGTCGACT and primer D(Hobert, 2002). The fusion PCR was carried out with the non-nested primer 5′-GGAGTGGACGATGACATTGTG and primer D* (Hobert,2002).

#### gcy-5prom::fozi-1cDNA::gfp and ceh-36prom::fozi-1cDNA::gfp

The fozi-1-coding sequence from start to stop codon was PCR amplified from the EST clone yk288g3 with the primers 5′-TTGGATCCATGATGCTTGCATCATCAGCG and 5′-AATGGCCAAGGAGACGAGACATTGATGTG. The amplicon was cloned into gcy-5prom::gfp and ceh-36prom::gfpvector constructs that will be described elsewhere in detail.

#### gcy-5prom::fozi-1cDNAΔZn-finger::gfp

The construct gcy-5prom::fozi-1cDNA::gfp was mutated by standard mutagenesis using the primers 5′-CTATCCCTGTACATTTCAATATgGATTAGCGGGTGACCCGATCAG and 5′-CTGATCGGGTCACCCGCTAATCcATATTGAAATGTACAGGGATAG.

#### gcy-5prom::fozi-1cDNAΔFH2::gfp

Truncated fozi-1 was amplified from the fozi-1 EST clone yk288g3 using the primers 5′-ttggatccATGATGCTTGCATCATCAGCG and 5′-aatggccaAATTGGCTGAATCGGAATTATAGATGATGACAAGG. The resulting amplicon was subcloned into the gcy-5prom::gfp vector construct.

### Genetics

The genetic screen from which ot61 was isolated has been described before (Johnston et al.,2005). Briefly, animals carrying the otIs3 transgene,which express gcy-7prom::gfp exclusively in ASEL, were mutagenized with EMS. F1 progeny were singled and F2 progeny were scored under a Zeiss SV6 fluorescent dissecting microscope. Mutant mapping was carried out using a combination of single nucleotide polymorphisms from the Hawaiian wild-type isolate CB4856 (Hodgkin and Doniach, 1997) and three-factor mapping with the visible markers dpy-17 and unc-49.

### Biochemical analysis of the FOZI-1 FH2 domain

The cDNA encoding the FOZI-1 FH2 domain (residues 367-733) was subcloned by standard techniques into pGEX-6P2 and expressed as a GST fusion protein in E. coli strain BL21. FH2FOZI-1 protein was purified by cleavage from the GST moiety (Copeland et al., 2004). Protein purity was assessed by Coomassie Blue staining of samples subjected to SDS-PAGE, concentration was determined by Bradford and by OD280. The isolated FH2 domain (0.4 μM) was added to G-actin (4 μM,5% pyrene-actin) for in vitro actin polymerization assays (Cytoskeleton)according to the supplied protocol. For cross-linking, the purified FH2FOZI-1 domain (20 nM) was incubated at room temperature for 60 minutes in HEK buffer with 30 μM of the crosslinker Bis-maleimidohexane (Pierce) dissolved in DMSO or DMSO alone. A second control sample was boiled for 10 minutes in 2% SDS prior to crosslinking. The crosslinking reactions were quenched with 45 mM DTT and then subjected to SDS-PAGE. The proteins were transferred to PVDF membranes and visualized by immunoblotting with an anti-6xHis monoclonal antibody(Clontech).

### Isolation of a novel ASE cell fate mutant

In order to identify genes essential for the ASEL/R cell fate decision, we performed screens for mutants defective in expression of the ASEL-specific gcy-7prom::gfp reporter gene. We have previously termed mutants that arose from these screening efforts lsy' mutants [forlaterally symmetric' (Johnston and Hobert, 2003)]. The lsy mutants retrieved from past genetic screens fall into three phenotypic categories.

1. Class I (two ASEL') mutants. The ASER neuron loses expression of ASER-specific genes and gains expression of the normally ASEL-specific gene expression profile (examples: cog-1, unc-37 mutants)(Chang et al., 2003).

2. Class II (two ASER') mutants. The ASEL neuron loses the ASEL-specific gene expression profile and displays the normally ASER-specific gene expression profile (examples: lsy-6, die-1, lsy-2, ceh-36, lin-49 mutants)(Chang et al., 2004; Chang et al., 2003; Johnston and Hobert, 2003; Johnston and Hobert,2005).

3. Class III (no ASEL/R') mutants. Complete loss of both asymmetric and symmetric features of the ASE neuron (example: che-1 mutants)(Chang et al., 2003).

One recessive mutant allele identified in the screen, ot61,exhibits a unique phenotype not described previously. All ot61 mutant animals display de-repression of gcy-7prom::gfp in ASER(Fig. 1B). However, although the phenotype is completely penetrant (all animals show the defect), the levels of aberrant gcy-7prom::gfp expression in ASER are variable. Sometimes, the level of aberrant gcy-7prom::gfpexpression in ASER is equivalent to the normal and unaffected gcy-7prom::gfp expression in ASEL, but more often, the aberrant gcy-7prom::gfp expression level in ASER is not as high as the gcy-7prom::gfp expression level in ASEL(Fig. 1B, Table 1). The fozi-1mutant phenotype is therefore variably expressive'. This variable expressivity contrasts the previously described phenotype in the ASER neuron of class I (two ASEL') mutant animals in which the level of aberrant gcy-7prom::gfp expression in ASER always reaches levels indistinguishable from the level in ASEL (i.e. complete expressivity)(Chang et al., 2003). To determine whether this complete penetrance and partial expressivity applies to other ASEL-specific fate markers, we crossed ot61 with animals containing the ASEL-specific lim-6prom::gfp reporter transgene and again observed de-repression of the reporter gene with complete penetrance and variable expressivity (Fig. 1B).

### fozi-1 encodes a protein with an unusual domain composition

Before characterizing the ot61 phenotype in more detail, we first determined the molecular identity of the mutant locus. Given the variable expressivity of the ot61 mutant phenotype, we were particularly interested in determining whether this phenotype is merely explained by a hypomorphic nature of the ot61 allele. We mapped the ot61mutant to a small interval on chromosome III using a combination of SNP and three-factor physical mapping. Transformation rescue narrowed the position of the locus to a single cosmid, K01B6 (Fig. 2A), and RNA interference of a single predicted gene on that cosmid, K01B6.1, recapitulated the defects of ot61 mutants(Table 1). We sequenced K01B6.1 in ot61 mutant animals and found that it contains a nonsense mutation resulting in premature termination of the protein-coding sequence(Fig. 2B). Two additional alleles that delete essential parts of the predicted K01B6.1 gene phenocopy the ot61 mutant allele (Fig. 2B; Table 1). Last,a genomic fragment that contains only the K01B6.1-coding sequences rescues the mutant phenotype (Fig. 3A). We conclude that the loss of K01B6.1 function causes ASE differentiation defects.

The predicted K01B6.1 gene structure was confirmed by analyzing an expressed sequence tag provided by Y. Kohara. The protein encoded by K01B6.1 contains three readily recognizable motifs(Fig. 2B): (1) a Q-rich motif often found in transcription factors (Titz et al., 2006); (2) two C2H2 Zn fingers with a standard CX2CX11-13HX4H spacing normally found in DNA-binding transcription factors (Iuchi,2001) (see Fig. S1 in the supplementary material); (3) a formin homology 2 (FH2) domain, which is surprising as this domain is normally found only in cytosolic actinpolymerizing proteins in which the domain catalyses actin polymerization (Zigmond,2004). Owing to the unusual combination of the formin homology 2 and zinc-finger domains, we have named the gene fozi-1 (pronouncedfozzy-1').

Fig. 1.

fozi-1 mutants display a de-repression of ASEL-specific fate markers in ASER. (A) Summary of previously described regulatory interactions that determine ASEL and ASER fate. Some permissively acting genes(Chang et al., 2003) are not shown for simplicity. (B) In fozi-1(ot61) mutants,ASEL-specific gcy-7prom::gfp (otIs3) and lim-6prom::gfp (otIs114) expression is de-repressed in ASER. See Table 1 and Fig. 4A for quantification.

Fig. 1.

fozi-1 mutants display a de-repression of ASEL-specific fate markers in ASER. (A) Summary of previously described regulatory interactions that determine ASEL and ASER fate. Some permissively acting genes(Chang et al., 2003) are not shown for simplicity. (B) In fozi-1(ot61) mutants,ASEL-specific gcy-7prom::gfp (otIs3) and lim-6prom::gfp (otIs114) expression is de-repressed in ASER. See Table 1 and Fig. 4A for quantification.

fozi-1 is highly conserved in all three available nematode genome sequences (see Fig. S1 in the supplementary material). No other predicted C. elegans protein contains a FH2 domain combined with C2H2 Zn fingers. Database searches revealed only a single nonnematode gene in the zebrafish Danio rerio genome sequence that is predicted to contain a FH2 domain combined with C2H2 Zn fingers (data not shown). However, as this gene is not confirmed by cDNA evidence, we cannot exclude the possibility of a genome assembly or gene prediction error.

### The ASER neuron adopts a mixed fate' in fozi-1 mutants

After cloning fozi-1, two additional alleles of fozi-1,cc607 and tm563, were made available to us. Like ot61mutants, cc607 and tm563 mutants are viable, fertile and display no obvious behavioral or morphological abnormalities. The cc607 allele was retrieved from a screen for mesoderm lineage differentiation defects (J.L., unpublished). cc607 is a putative molecular null allele as it encodes a nonsense mutation causing premature termination before both the Zn fingers and the FH2 domain(Fig. 2B). Using rescue assays,we show below that the Zn fingers are essential for fozi-1 function. The other allele, tm563, kindly provided by a C. elegansknockout consortium, completely deletes the Q-rich domain-containing exon 3 and a large part of exon 4, including the first Zn finger and the first half of the second Zn finger (Fig. 2B). Owing to a premature stop codon introduced by the deletion, tm563 can also be considered a molecular null allele. The availability of these two molecular null alleles allowed us to examine whether the variable phenotypic expressivity of fozi-1(ot61) mutants is merely a reflection of a partial loss of gene function. Crossing the putative null alleles cc607 and tm563 with the ASEL-specific lim-6prom::gfp transgene, we observed gene expression defects essentially identical to those seen in ot61 mutant animals(Table 1). At least in the context of the ASE subclass determination, ot61 may therefore similarly be a null allele of fozi-1.

We examined the effect of the fozi-1(cc607)-null allele on several additional markers that define the ASEL fate, namely the two GCY genes gcy-6 and gcy-7 and the neuropeptide-encoding flp-4gene. We found that all examined ASEL fate markers are affected in a similar manner in that they become partly de-repressed in a completely penetrant manner (Fig. 4A). The incomplete and variably expressive de-repression of ASEL fate markers in the ASER neuron of fozi-1-null mutants suggests that additional, fozi-1-independent mechanism(s) must exist to prevent complete de-repression of the ASEL fate markers.

In contrast to all previously defined ASE cell fate mutants, the de-repression of ASEL terminal fate markers in the ASER neuron of fozi-1 mutants is not accompanied by loss of ASER fate markers. Expression of the ASER-specific gcy-4, gcy-5 and hen-1 genes in fozi-1 null mutants is indistinguishable from wild-type expression(Fig. 4B). Taken together, the ASEL neuron appears completely unaffected in fozi-1 mutants in that ASEL markers are expressed and ASER fate markers are not de-repressed. By contrast, ASEL fate markers are de-repressed in ASER, but ASER fate markers remain unaffected. The ASER neuron therefore displays a novel mixed'phenotype in fozi-1 mutants (Fig. 4C). Extending our previous mutant classification, we term this phenotype a class IV' phenotype.

### fozi-1 acts in a left/right asymmetric manner in ASER, but not ASEL

The phenotypic analysis of fozi-1 demonstrates that ASER, but not ASEL, is affected in fozi-1 mutants. As fozi-1 appears to encode a gene regulatory factor, the gene expression defects in ASER are most easily explained by fozi-1 being expressed and acting in ASER. We tested this prediction by a variety of means, including gene expression pattern analysis, cell-specific rescue and mis-expression approaches.

In order to investigate fozi-1 expression, we generated a construct, fozi-1::gfp, in which the genomic fozi-1 locus is tagged with gfp. This construct efficiently rescues the fozi-1 mutant phenotype (Fig. 3A). Transgenic, adult animals display gfp expression in the nucleus of the AWC and ASE neuron class(Fig. 3B and data not shown). No other expression was observed in head ganglia. Three out of four lines displayed a bias in expression to ASER; by contrast, expression in the AWCL/R neurons is bilaterally symmetric (Fig. 3B,C). To corroborate that fozi-1 indeed functions in ASER, we generated a gcy-5prom::fozi-1 cDNA::gfp construct in which the gcy-5 promoter drives expression of fozi-1fused to gfp specifically in postmitotic ASER neurons. The gfp moiety confirmed that fozi-1 was expressed and localized to the nucleus of transgenic animals. Four out of four gcy-5prom::fozi-1 cDNA::gfp transgenic lines rescued the lim-6prom::gfp de-repression defects in ASER observed in fozi-1(cc607)-null mutant animals. Taken together, expression and rescue experiments with the heterologous promoter demonstrate that fozi-1 acts specifically in ASER to repress ASEL cell fate.

Transgenic lines that contain multiple copies of genomic fozi-1DNA show not only a rescue of the ASER defects of fozi-1 mutants(Fig. 3A, right bar) but also show a partial repression of lim-6 expression in ASEL(Fig. 3A, left bar). This ectopic activity of fozi-1 correlates with the degree of ASEL versus ASER bias of fozi-1::gfp expression(Fig. 3C). Transgenic line 3,which shows the most bias to ASER rescues the ASER defect but has little impact on ASEL fate determination, while transgenic lines 1 and 4, which show a larger degree of fozi-1::gfp in both ASER and ASEL(Fig. 3C), show suppression of the ASEL fate marker lim-6 in ASEL(Fig. 3A). These observations suggest that fozi-1 may be sufficient to repress ASEL fate if overexpressed in ASEL. To corroborate this notion, we generated a ceh-36prom::fozi-1 cDNA::gfp construct in which the ceh-36 promoter drives equal expression of gfp-tagged fozi-1 in both ASEL and ASER. Introduction of this construct into the fozi-1(cc607) mutant background showed equivalent repression of lim-6prom::gfp expression in both ASEL and ASER(Fig. 3A). fozi-1 is therefore sufficient to repress lim-6 when ectopically expressed in ASEL. Taken together, we conclude that fozi-1 expression is biased to ASER and that it acts autonomously to repress the expression of lim-6and other ASEL-specific genes in ASER.

Last, as two transcription factors in the ASEL/R cell fate regulatory network are repressed via 3′UTR dependent mechanisms (cog-1,die-1), we tested whether the 3′UTR of fozi-1 contains cis- regulatory information that may contribute to the left/right asymmetric function of fozi-1. Using a sensor gene approach that revealed the 3′UTR-dependent regulation of the cog-1 and die-1 genes (Chang et al.,2004; Johnston and Hobert,2003), we found this not to be the case (data not shown).

### Terminal differentiation genes are also controlled by pathways that act in parallel to fozi-1 and lim-6

The results presented so far can be summarized as shown schematically in Fig. 4E. fozi-1 is expressed in ASER and is required to repress ASEL-specific features in ASER. The incomplete nature of de-repression of the ASEL markers suggests one of two scenarios. In scenario 1, fozi-1 acts together with an unknown repressor X to repress ASEL-specific features in ASER and only loss of both fozi-1 and X causes complete de-repression of ASEL fate in ASER. Alternatively, in scenario 2, de-repression of ASEL markers in ASER of fozi-1 is incomplete as an activator Y is missing in ASER(Fig. 4E).

The LIM homeobox gene lim-6 is expressed in ASEL and its loss causes a mixed phenotype that is superficially the mirror image of the fozi-1 loss-of-function phenotype. The ASER neuron is unaffected, but the ASEL neuron displays a mixed phenotype' characterized by a failure to repress ASER features in ASEL (Hobert et al., 1999; Johnston et al.,2005; Ortiz et al.,2006). Like in fozi-1 mutants, the defects in lim-6-null mutants are also variably expressive(Fig. 4D), thereby suggesting the existence of lim-6- independent means to repress ASER fate(Fig. 4E). Consistent with the existence of a parallel pathway, repression of ASER fate markers induced by ectopic expression of the lsy-6 miRNA in ASER does not absolutely require lim-6 (Fig. 4D). We have previously shown that lim-6 is also required to positively regulate the expression of ASEL-specific markers, namely the two FMRFamide-encoding genes flp-4 and flp-20(Johnston et al., 2005). lim-6-null mutant animals show only a partially penetrant loss of flp-4 and flp-20 expression, again indicating the presence of a parallel pathway (indicated in by a factor Z in Fig. 4E). Taken together, these observations suggest that both lim-6 and fozi-1 require parallel pathways to exert their function in ASEL and ASER, respectively.

### fozi-1 acts downstream of the bi-stable feedback loop

We tested how fozi-1 gene function relates to the function of components of the bi-stable feedback loop that controls ASEL and ASER fate(Fig. 1A). We found that disruption of this bi-stable loop causes a symmetrization' of the normally left/right asymmetric expression of fozi-1(Fig. 5A). Specifically, in animals in which the ASEL-inducers die-1 or the lsy-6 miRNA are mutated, fozi-1 expression becomes de-repressed in ASEL. In animals lacking the ASER-inducer cog-1, the bias of fozi-1expression to ASER is also lost. These results are consistent with our previous observations that the bi-stable feedback loop controls all features of the ASEL/R fate decision (Johnston et al., 2005).

We corroborated the downstream role of fozi-1 using two genetic epistasis tests. In animals in which the lsy-6 miRNA is deleted, ASEL displays the complete ASER gene expression profile including the repression of lim-6 expression (Johnston and Hobert, 2003). Repression of lim-6 in lsy-6(ot71) null mutant animals requires fozi-1 activity as fozi-1(ot61); lsy-6(ot71) double mutants display a de-repression of lim-6prom::gfp in ASEL and ASER(Fig. 5B). The same genetic interaction is observed using die-1 mutants. In die-1(ot26)mutant animals, lim-6 expression is repressed in ASEL(Chang et al., 2004). This repression requires fozi-1 as die-1(ot26); fozi-1(ot61)double mutants display a de-repression of lim-6prom::gfpin ASEL and ASER (Fig. 5B). Together, these data demonstrate that fozi-1 acts downstream of die-1, the output regulator of the bi-stable feedback loop(Fig. 5C).

Fig. 2.

fozi-1 encodes a protein with two Zn fingers and a formin homology 2 domain. (A) Mapping of the ot61 mutation.(B) Structure of the predicted FOZI-1 protein.

Fig. 2.

fozi-1 encodes a protein with two Zn fingers and a formin homology 2 domain. (A) Mapping of the ot61 mutation.(B) Structure of the predicted FOZI-1 protein.

In contrast to the partially expressive effects of fozi-1 on lim-6 expression, the effects of disruption of bi-stable loop components such as die-1 or cog-1 on lim-6expression are completely expressive. For example, levels of aberrant lim-6 expression in ASER in cog-1 mutants are similar to normal levels of lim-6 expression in ASEL(Chang et al., 2003). The fozi-1 parallel pathway that we evoked above(Fig. 4E), therefore, depends genetically on the activity of the bi-stable loop whose output regulator is the die-1 gene. The most parsimonious way to present these regulatory interactions is shown in Fig. 5C with an arrow from die-1 to lim-6 that is parallel to the fozi-1-mediated regulation of lim-6. In other words, lim-6 expression genetically requires both the presence of die-1 and the absence of fozi-1.

Loss of die-1 also causes a completely expressive de-repression of ASER fate in ASEL (Chang et al.,2004), which contrasts the partially expressive de-repression of ASER fate in the ASEL neuron of lim-6-null mutants. The lim-6-parallel pathway that we evoked above(Fig. 4E), therefore also depends on the loop output regulator die-1. The most parsimonious explanation of the genetic interactions is that the factor that acts in parallel to lim-6 (factor Z in Fig. 4E) is die-1itself (Fig. 5C).

A loss of the lim-6 LIM homeobox gene, which acts downstream of the bi-stable feedback loop (Fig. 1A) causes a partially penetrant defect in maintaining the left/right asymmetric expression of loop components(Johnston et al., 2005)(broken line in Fig. 1A). As lim-6 is de-repressed in ASER of fozi-1 mutants, we asked whether a similar partially penetrant defect can be observed in fozi-1 mutants, and we indeed find this to be the case(Fig. 5D,E). Moreover, as would be expected from a partial disruption of activity of bi-stable loop components in lim-6-null mutants, asymmetric fozi-1 expression is also partially affected in lim-6 mutants(Fig. 5A).

Taken together, left/right asymmetric fozi-1 expression is controlled by components of the bi-stable feedback loop and asymmetric fozi-1 augments the maintenance of the asymmetric expression of loop components, probably through the regulation of lim-6 expression.

### The Zn fingers but not the FH2 domain are essential for fozi-1 function

All factors previously known to play roles in the lateral ASE cell fate decision are bona fide transcription factors or miRNAs. Therefore, we were surprised to have identified with fozi-1 a gene that, on the one hand, contains conventional signatures of a transcription factor (Zn fingers and Q-rich domain), but on the other hand contains a FH2 domain, which has been characterized as a cytoplasmic actinnucleation domain(Zigmond, 2004). This prompted us to learn more about the FH2 domain of FOZI-1 (henceforth termed FH2FOZI-1).

Fig. 3.

Rescue, expression, site of action and domain requirements of fozi-1. (A) Rescue of the fozi-1 mutant phenotype. The constructs do not contain the first, non-coding exon of the fozi-1 gene, which is located >7 kb upstream of the ATG-containing second exon (Fig. 2A). The right column indicates rescue of the fozi-1 defect, i.e. suppression of aberrant lim-6 expression in ASER, and the left column indicates suppression of normal lim-6 expression in ASEL fate, caused by ectopic expression of fozi-1. All constructs show similar expression levels and exclusive localization to the nucleus. (B) A representative fozi-1::gfp-expressing animal, showing gfp expression in ASER but not ASEL, and in the two olfactory neurons AWCL and AWCR. Broken lines approximately indicate the head of the worm. The insets show that fozi-1::gfp predominantly localizes to the nucleus. The red cyto'marker is dsRed2 protein, expressed under control of the ceh-36promoter (otIs151 transgene). (C) Three out of four fozi-1::gfp-expressing transgenic, wild-type lines display varying levels of asymmetric gfp expression in ASER. Circles indicate absent,ASEL alone, ASEL and ASER, and ASER alone, respectively.

Fig. 3.

Rescue, expression, site of action and domain requirements of fozi-1. (A) Rescue of the fozi-1 mutant phenotype. The constructs do not contain the first, non-coding exon of the fozi-1 gene, which is located >7 kb upstream of the ATG-containing second exon (Fig. 2A). The right column indicates rescue of the fozi-1 defect, i.e. suppression of aberrant lim-6 expression in ASER, and the left column indicates suppression of normal lim-6 expression in ASEL fate, caused by ectopic expression of fozi-1. All constructs show similar expression levels and exclusive localization to the nucleus. (B) A representative fozi-1::gfp-expressing animal, showing gfp expression in ASER but not ASEL, and in the two olfactory neurons AWCL and AWCR. Broken lines approximately indicate the head of the worm. The insets show that fozi-1::gfp predominantly localizes to the nucleus. The red cyto'marker is dsRed2 protein, expressed under control of the ceh-36promoter (otIs151 transgene). (C) Three out of four fozi-1::gfp-expressing transgenic, wild-type lines display varying levels of asymmetric gfp expression in ASER. Circles indicate absent,ASEL alone, ASEL and ASER, and ASER alone, respectively.

Fig. 4.

Analysis of ASEL and ASER fate in fozi-1 and lim-6mutant animals. (A) In fozi-1(cc607)-null mutant animals,ASEL-specific lim-6prom::gfp (otIs114),flp-4prom::gfp (otIs178), gcy-7 prom::gfp (otIs3)and gcy-6 prom::gfp (otIs162) are de-repressed in ASER.0=0' indicates no expression, L>0' indicates exclusive expression in ASEL, L>R' indicates expression in ASEL is stronger than in ASER, L=R'indicates equal expression in ASEL and ASER, L<R' indicates expression in ASER is stronger than in ASEL, and 0<R' indicates exclusive expression in ASER. (B) Analysis of ASER fate markers in fozi-1(cc607) null mutants. Reporter arrays used were ntIs1 (gcy-5), otEx2409 (gcy-4)and otEx1274 (hen-1ASER). (C) Summary of the fozi-1 mutant phenotype and comparison with previously described ASE fate mutants. Blue circles indicate the ASEL-specific gene expression battery,red circles indicate the ASER-specific gene expression battery. (D) lim-6 is not sufficient to repress ASER cell fate. (E) Summary of genetic interaction data. The incomplete expressivity of fozi-1and lim-6 null alleles argues for the existence of parallel pathways(indicated by factor X, Y, Z). The two different scenarios make different predictions about the site of action of the parallel pathways. In scenario 1,it is active in ASER; in scenario 2, it is active in ASEL.

Fig. 4.

Analysis of ASEL and ASER fate in fozi-1 and lim-6mutant animals. (A) In fozi-1(cc607)-null mutant animals,ASEL-specific lim-6prom::gfp (otIs114),flp-4prom::gfp (otIs178), gcy-7 prom::gfp (otIs3)and gcy-6 prom::gfp (otIs162) are de-repressed in ASER.0=0' indicates no expression, L>0' indicates exclusive expression in ASEL, L>R' indicates expression in ASEL is stronger than in ASER, L=R'indicates equal expression in ASEL and ASER, L<R' indicates expression in ASER is stronger than in ASEL, and 0<R' indicates exclusive expression in ASER. (B) Analysis of ASER fate markers in fozi-1(cc607) null mutants. Reporter arrays used were ntIs1 (gcy-5), otEx2409 (gcy-4)and otEx1274 (hen-1ASER). (C) Summary of the fozi-1 mutant phenotype and comparison with previously described ASE fate mutants. Blue circles indicate the ASEL-specific gene expression battery,red circles indicate the ASER-specific gene expression battery. (D) lim-6 is not sufficient to repress ASER cell fate. (E) Summary of genetic interaction data. The incomplete expressivity of fozi-1and lim-6 null alleles argues for the existence of parallel pathways(indicated by factor X, Y, Z). The two different scenarios make different predictions about the site of action of the parallel pathways. In scenario 1,it is active in ASER; in scenario 2, it is active in ASEL.

Fig. 5.

fozi-1 acts downstream of the bi-stable feedback loop to control ASER differentiation. (A) Asymmetric fozi-1::gfpexpression (otEx2192; line #3 in Fig. 3C) is disrupted in die-1(ot26), lsy-6(ot71) and cog-1(sy607) null mutants animals, and partially affected in lim-6-null mutants. The L=R'category indicates equal expression in ASEL and ASER, including de-repression of gfp expression in ASEL (yielding strong gfp expression in both ASEL and ASER; die-1 and lsy-6 phenotype) and reduction of gfp expression in ASER (yielding equally low expression in ASEL and ASER; cog-1 phenotype). (B) Removing fozi-1reverts the loss of lim-6 expression (otIs114) observed in lsy-6(ot71) or die-1(ot26) mutant animals. (C)Summary of genetic interactions, pooling data from A and B, combined with the data from Fig. 4. The completely penetrant and expressive phenotype observed in die-1mutants argues that the pathways parallel to fozi-1 and lim-6 are under die-1 control. The simplest explanation is that die-1 itself acts in parallel to fozi-1 and lim-6 to control expression of target genes. (D) fozi-1(cc607) null mutant animals display weakly penetrant defects in lsy-6 expression, assayed with lsy-6prom::gfp(otIs162) and cog-1 expression, assayed with cog-1::gfp(syIs63). (E) Summary of the feedback data. The lim-6-dependent feedback to lsy-6 (or die-1; for simplicity, the arrow only points to lsy-6), described in Johnston et al.(Johnston et al., 2005), is represented by a broken arrow to indicate that the effect is partially penetrant and only required to maintain the asymmetric expression of loop components.

Fig. 5.

fozi-1 acts downstream of the bi-stable feedback loop to control ASER differentiation. (A) Asymmetric fozi-1::gfpexpression (otEx2192; line #3 in Fig. 3C) is disrupted in die-1(ot26), lsy-6(ot71) and cog-1(sy607) null mutants animals, and partially affected in lim-6-null mutants. The L=R'category indicates equal expression in ASEL and ASER, including de-repression of gfp expression in ASEL (yielding strong gfp expression in both ASEL and ASER; die-1 and lsy-6 phenotype) and reduction of gfp expression in ASER (yielding equally low expression in ASEL and ASER; cog-1 phenotype). (B) Removing fozi-1reverts the loss of lim-6 expression (otIs114) observed in lsy-6(ot71) or die-1(ot26) mutant animals. (C)Summary of genetic interactions, pooling data from A and B, combined with the data from Fig. 4. The completely penetrant and expressive phenotype observed in die-1mutants argues that the pathways parallel to fozi-1 and lim-6 are under die-1 control. The simplest explanation is that die-1 itself acts in parallel to fozi-1 and lim-6 to control expression of target genes. (D) fozi-1(cc607) null mutant animals display weakly penetrant defects in lsy-6 expression, assayed with lsy-6prom::gfp(otIs162) and cog-1 expression, assayed with cog-1::gfp(syIs63). (E) Summary of the feedback data. The lim-6-dependent feedback to lsy-6 (or die-1; for simplicity, the arrow only points to lsy-6), described in Johnston et al.(Johnston et al., 2005), is represented by a broken arrow to indicate that the effect is partially penetrant and only required to maintain the asymmetric expression of loop components.

Primary sequence comparison of FH2FOZI-1 and other FH2 domains suggests that FH2FOZI-1 is clearly related to FH2 domains(Fig. 6A). Secondary structure predictions of FH2FOZI-1 also agree well with known FH2 domain structures. Homology modeling of FH2FOZI-1 reveals that patterns of conservation in the hydrophobic cores of the lasso and post regions, two structural motifs involved in FH2 domain dimerization(Otomo et al., 2005; Xu et al., 2004), also allow a potential dimeric interaction of FH2FOZI-1 domain(Fig. 6A,B). However, the similarity of FH2FOZI-1 with other FH2 domains is poor in those regions that contact actin. In particular, two highly conserved amino acids,equivalent to Ile1431 and Lys1601 in the Bni1 FH2 domain, which have been shown to be essential for actin nucleation activity(Otomo et al., 2005; Xu et al., 2004), are altered in FH2FOZI-1 (Fig. 6A). In addition, the linker between the lasso and knob regions,which is also important for actin nucleation function, is severely shortened in the FH2FOZI-1, thereby likely reducing its flexibility. The FH2 domains of all other nematode FH2 domain proteins (six besides FOZI-1; http://smart.embl-heidelberg.de/)look like standard FH2 domains in terms of conservation of individual key residues and length of linker regions (data not shown). Taken together, our sequence analysis makes two predictions about FOZI-1, namely that FOZI-1 may still homodimerize via its FH2 domain but may have no role in controlling actin polymerization as it lacks actin-binding surfaces and as the ring structure made by FH2 domain dimers (Fig. 6B) may not be large enough to accommodate actin.

We tested these predictions using in vitro and in vivo assays. FH2 domains affect actin polymerization by either inducing polymerization of actin monomers or capping F-actin barbed ends(Kovar et al., 2003; Pruyne et al., 2002). We tested the ability of purified FH2FOZI-1 to induce actin polymerization in vitro using the pyrene actin assay. In contrast to the control FH2 domain of the mouse FH2 protein Dia1, FH2FOZI-1 is unable to induce actin polymerization (Fig. 6C). Similarly, FH2FOZI-1 is unable to cap barbed ends(see Fig. S2A in the supplementary material) and does not bind actin filaments in sedimentation assays (data not shown). These experiments confirm the prediction that FH2FOZI-1 has lost its role in regulating actin polymerization. The prediction that FH2FOZI-1 still maintains its ability to dimerize is confirmed by our observation that purified FH2FOZI-1 elutes from a Sephadex 200 gel filtration column as a single peak of ∼100 kDa, consistent with the formation of a FH2FOZI-1 dimer (see Fig. S2B in the supplementary material). In separate crosslinking experiments, the formation of FH2FOZI-1oligomers could also be detected (Fig. 6D).

Fig. 6.

The FH2 domain of FOZI-1 is unusual and does not affect actin dynamics. (A) Alignment of FH2FOZI-1 with other FH2 domains, calculated with T-coffee version 2.03(Notredame et al., 2000). FH2FOZI-1 shows 9% to 20% sequence identity to other sequences in the alignment (49 FH2 domains from Pfam; only four shown here), which is in the range of sequence identities between the remaining FH2 domain sequences in the alignment. The sequence identities of FH2FOZI-1 and FH2 domains of known structure, yeast FH2Bni-1 [PDB-id:1y64;(Otomo et al., 2005)] and mouse FH2Dia1 [PDB-id:1v9d;(Shimada et al., 2004)], are 16% and 14%, respectively. The profile-profile alignment program hmap(Tang et al., 2003) assigns a very good E-value (1.4e-26) for alignments between FH2FOZI-1 and yeast FH2Bni1. Secondary structure prediction for FH2FOZI-1 also agrees well with the secondary structure found in the known structures (colored bar above alignment for FH2Bni1). Neither Ile1431 nor Lys1601, which are crucial for the actin nucleation activity of the FH2 domains are conserved in FH2FOZI-1 (red and black arrow). The colored bar at the bottom shows sequence conservation as calculated previously (Valdar,2002) (red, high degree of conservation; blue, low degree of conservation). (B) Dimeric structure of the FH2 domain of yeast Bni1(Otomo et al., 2005).(C) FH2FOZI-1 does not stimulate actin polymerization. The FH2 domain of mouse Dia1 serves as a positive control. (D)FH2FOZI-1 multimerizes. Isolated FH2FOZI-1 (predicted to be 42 kDa) migrates as a single band (lane 1, -BMH). Treatment of FH2FOZI-1 with the crosslinking reagent Bis-maleimidohexane (BMH) produces discrete slower migrating bands(lane 3, +BMH). The putative FH2 multimer is not present if the protein sample is denatured prior to crosslinking (lane 2, +SDS+BMH). FH2FOZI-1was detected by immunoblotting.

Fig. 6.

The FH2 domain of FOZI-1 is unusual and does not affect actin dynamics. (A) Alignment of FH2FOZI-1 with other FH2 domains, calculated with T-coffee version 2.03(Notredame et al., 2000). FH2FOZI-1 shows 9% to 20% sequence identity to other sequences in the alignment (49 FH2 domains from Pfam; only four shown here), which is in the range of sequence identities between the remaining FH2 domain sequences in the alignment. The sequence identities of FH2FOZI-1 and FH2 domains of known structure, yeast FH2Bni-1 [PDB-id:1y64;(Otomo et al., 2005)] and mouse FH2Dia1 [PDB-id:1v9d;(Shimada et al., 2004)], are 16% and 14%, respectively. The profile-profile alignment program hmap(Tang et al., 2003) assigns a very good E-value (1.4e-26) for alignments between FH2FOZI-1 and yeast FH2Bni1. Secondary structure prediction for FH2FOZI-1 also agrees well with the secondary structure found in the known structures (colored bar above alignment for FH2Bni1). Neither Ile1431 nor Lys1601, which are crucial for the actin nucleation activity of the FH2 domains are conserved in FH2FOZI-1 (red and black arrow). The colored bar at the bottom shows sequence conservation as calculated previously (Valdar,2002) (red, high degree of conservation; blue, low degree of conservation). (B) Dimeric structure of the FH2 domain of yeast Bni1(Otomo et al., 2005).(C) FH2FOZI-1 does not stimulate actin polymerization. The FH2 domain of mouse Dia1 serves as a positive control. (D)FH2FOZI-1 multimerizes. Isolated FH2FOZI-1 (predicted to be 42 kDa) migrates as a single band (lane 1, -BMH). Treatment of FH2FOZI-1 with the crosslinking reagent Bis-maleimidohexane (BMH) produces discrete slower migrating bands(lane 3, +BMH). The putative FH2 multimer is not present if the protein sample is denatured prior to crosslinking (lane 2, +SDS+BMH). FH2FOZI-1was detected by immunoblotting.

Fig. 7.

Summary of the gene regulatory architecture in the ASE neurons.(A) Summary of regulatory interaction in ASEL and ASER. Broken line indicates partially penetrant feedback interaction (see Fig. 5E). See D for deconvolution of individual regulatory interactions. Several permissively acting factors, i.e. factors expressed in both ASEL and ASER(Chang et al., 2003) are not shown here for simplicity. Such factors could, for example, activate the ASER-expressed GCY genes in the absence of the ASEL repressors die-1and lim-6. (B) Network motifs. A FFL motif occurs when one gene (Gene A) controls a second gene (Gene B) and together these factors are required to regulate a target gene (Gene T). The addition of other factors(e.g. factor C) transforms the FFL motif to a bi-parallel motif'(Milo et al., 2002), which(analagous to FFL motifs) one could also envision to work as a persistence detector. (C) die-1 and fozi-1 may control target genes through a FFL motif. For a target to be activated, it requires both the presence of die-1 and the absence of fozi-1. See D for identity of target genes. All arrows shown in this figure represent genetic interactions and do not necessarily imply direct physical interactions. Therefore, the identification of additional factors may alter network architecture. For example, die-1 may not only repress fozi-1but also an additional factor, repressor Z', which together with fozi-1 may repress ASEL-specific GCY genes. Such a repressor Z would transform the network motif from a FFL motif to a bi-parallel motif' (B).(D) Deconvoluted regulatory motifs extracted from A. Owing to their differential behavior upon loss of upstream regulators, ASE terminal differentiation genes can be placed into three distinct categories, all of which controlled by the basic FFL motif architecture shown in B. Target Gene Category 1: ASEL-specific expression of the GCY genes gcy-6, gcy-7,gcy-14 and gcy-20 does not require lim-6, but depends on the loop output regulator die-1 and the downstream regulator fozi-1. As a complete elimination of fozi-1 activity only results in partially expressive de-repression of the ASEL-specific GCY genes,an additional factor must be involved in repressing these GCY genes. This factor could be an unknown repressor that cooperates with fozi-1 or,alternatively, the failure to completely activate ASEL-specific GCY genes in ASER may be due to the lack of an activator in ASER(Fig. 4E). The loop output regulator die-1 is the best available candidate for this activator as die-1 is predominantly expressed in ASEL and die-1 mutation leads to a completely penetrant and expressive effect on ASEL-specific gene expression. As die-1 also regulates fozi-1, the genetic interaction therefore may define a FFL motif. This motif is the most parsimonious illustration of the genetic observation that ASEL-specific genes depend on two different factors: the presence of die-1 and the absence of fozi-1. Target Gene Categories 2 and 3: Regulation of genes in this category is distinguishable from control of Category 1 genes by the distinct role of the LIM homeobox gene lim-6. die-1 represses fozi-1 expression in ASEL; in the absence of fozi-1, lim-6is expressed. Together, lim-6 and die-1 (or a die-1-dependent pathway) activate ASEL-specific FLP genes and repress ASER-specific GCY genes in ASEL. This motif architecture is also a FFL motif,but now with an additional tier of regulation. Similar to the case of Category 1 target genes, the argument for this network architecture is revealed through the completely penetrant effect of disruption of the components of the feedback loop (including die-1) on all downstream genes (lim-6,fozi-1 and terminal target genes), and the incompletely penetrant and expressive effect of lim-6 on the terminal target genes. This incomplete penetrance and expressivity implies the need for another regulatory factor, for which die-1 is at present the best candidate, given its completely penetrant and expressive effect on the terminal target genes. An additional potential feed-forward motif in the interaction of these factors is suggested by the incomplete penetrance of fozi-1 on lim-6expression. As lim-6 is affected by die-1 in a completely penetrant manner, lim-6 is controlled by a potential feed-forward loop, receiving inputs from die-1 and fozi-1. This renders lim-6 under the same control as the above mentioned ASEL-specific GCY genes, which are also controlled by a combination of die-1 and fozi-1.

Fig. 7.

Summary of the gene regulatory architecture in the ASE neurons.(A) Summary of regulatory interaction in ASEL and ASER. Broken line indicates partially penetrant feedback interaction (see Fig. 5E). See D for deconvolution of individual regulatory interactions. Several permissively acting factors, i.e. factors expressed in both ASEL and ASER(Chang et al., 2003) are not shown here for simplicity. Such factors could, for example, activate the ASER-expressed GCY genes in the absence of the ASEL repressors die-1and lim-6. (B) Network motifs. A FFL motif occurs when one gene (Gene A) controls a second gene (Gene B) and together these factors are required to regulate a target gene (Gene T). The addition of other factors(e.g. factor C) transforms the FFL motif to a bi-parallel motif'(Milo et al., 2002), which(analagous to FFL motifs) one could also envision to work as a persistence detector. (C) die-1 and fozi-1 may control target genes through a FFL motif. For a target to be activated, it requires both the presence of die-1 and the absence of fozi-1. See D for identity of target genes. All arrows shown in this figure represent genetic interactions and do not necessarily imply direct physical interactions. Therefore, the identification of additional factors may alter network architecture. For example, die-1 may not only repress fozi-1but also an additional factor, repressor Z', which together with fozi-1 may repress ASEL-specific GCY genes. Such a repressor Z would transform the network motif from a FFL motif to a bi-parallel motif' (B).(D) Deconvoluted regulatory motifs extracted from A. Owing to their differential behavior upon loss of upstream regulators, ASE terminal differentiation genes can be placed into three distinct categories, all of which controlled by the basic FFL motif architecture shown in B. Target Gene Category 1: ASEL-specific expression of the GCY genes gcy-6, gcy-7,gcy-14 and gcy-20 does not require lim-6, but depends on the loop output regulator die-1 and the downstream regulator fozi-1. As a complete elimination of fozi-1 activity only results in partially expressive de-repression of the ASEL-specific GCY genes,an additional factor must be involved in repressing these GCY genes. This factor could be an unknown repressor that cooperates with fozi-1 or,alternatively, the failure to completely activate ASEL-specific GCY genes in ASER may be due to the lack of an activator in ASER(Fig. 4E). The loop output regulator die-1 is the best available candidate for this activator as die-1 is predominantly expressed in ASEL and die-1 mutation leads to a completely penetrant and expressive effect on ASEL-specific gene expression. As die-1 also regulates fozi-1, the genetic interaction therefore may define a FFL motif. This motif is the most parsimonious illustration of the genetic observation that ASEL-specific genes depend on two different factors: the presence of die-1 and the absence of fozi-1. Target Gene Categories 2 and 3: Regulation of genes in this category is distinguishable from control of Category 1 genes by the distinct role of the LIM homeobox gene lim-6. die-1 represses fozi-1 expression in ASEL; in the absence of fozi-1, lim-6is expressed. Together, lim-6 and die-1 (or a die-1-dependent pathway) activate ASEL-specific FLP genes and repress ASER-specific GCY genes in ASEL. This motif architecture is also a FFL motif,but now with an additional tier of regulation. Similar to the case of Category 1 target genes, the argument for this network architecture is revealed through the completely penetrant effect of disruption of the components of the feedback loop (including die-1) on all downstream genes (lim-6,fozi-1 and terminal target genes), and the incompletely penetrant and expressive effect of lim-6 on the terminal target genes. This incomplete penetrance and expressivity implies the need for another regulatory factor, for which die-1 is at present the best candidate, given its completely penetrant and expressive effect on the terminal target genes. An additional potential feed-forward motif in the interaction of these factors is suggested by the incomplete penetrance of fozi-1 on lim-6expression. As lim-6 is affected by die-1 in a completely penetrant manner, lim-6 is controlled by a potential feed-forward loop, receiving inputs from die-1 and fozi-1. This renders lim-6 under the same control as the above mentioned ASEL-specific GCY genes, which are also controlled by a combination of die-1 and fozi-1.

We next tested the requirement of the FH2 domain and Zn fingers for fozi-1 function in vivo. To this end, we made use of the rescue assay described above (Fig. 3A), in which a gfp tagged fozi-1 cDNA is expressed under control of an ASER-specific promoter. Multiple transgenic lines were generated that express constructs in which individual domains of fozi-1 were deleted. All constructs showed, as assessed by their gfp tag,comparable expression levels in the nucleus of ASER (data not shown). We found that deleting the Zn fingers completely eliminates rescuing ability of the fozi-1 cDNA, demonstrating that the Zn fingers are essential for fozi-1 function in ASER (Fig. 3A). As FOZI-1 primarily localizes to the nucleus(Fig. 3B), requires its Zn fingers for function and is required for the repression of specific gene expression programs, we conclude that FOZI-1 probably is a transcriptional regulatory factor.

Examining the functional relevance of the FH2 domain, we found that deletion of the FH2 domain diminished but did not completely eliminate the ability of the fozi-1::gfp construct to rescue the fozi-1mutant phenotype (Fig. 3A). The FH2 domain of FOZI-1 therefore does not appear to be absolutely crucial for protein function, at least in the context of ASER differentiation.

In summary, FOZI-1 is unlikely to have a role in actin polymerization because: (1) there is no evidence for the existence of filamentous actin in the nucleus where FOZI-1 localizes; (2) detailed sequence analysis reveals that the FOZI-1 FH2 domain does not contain key features required for actin polymerization; (3) in vitro assays fail to reveal actin polymerization activity of the FOZI-1 FH2 domain; and (4) functional in vivo experiments indicate that the FH2 domain is not essential for gene function in the context of lateral ASE subclass specification. The only point that, at first sight,may argue for the importance of the FH2 domain is the ot61 allele,which prematurely terminates the FH2 domain. However, this mutation does not only disrupt the FH2 domain, but de-stabilizes the complete protein, as evidenced by a complete loss of anti-FOZI-1 antibody staining in ot61mutant animals (J.L., unpublished).

Few genes are known to function in a left/right asymmetric manner in bilaterally symmetric neuronal structures. Our genetic analysis establishes that fozi-1 acts in a left/right asymmetric manner in the ASE gustatory neuron class to control the correct execution of a single neuron-specific, left/right asymmetric gene expression program (summarized in Fig. 7). Below, we first discuss specific structural features of the fozi-1 gene and then describe how our analysis of fozi-1 has revealed novel aspects of ASEL/R fate determination.

### The domain composition of FOZI-1 provides an example of domain rearrangement

The combination of FH2 and C2H2 Zn-finger domains in a single protein appears to be unique to the nematode lineage. The fozi-1 gene probably originated by gene arrangement that joined two initially independent genes. What is the selective advantage of maintaining such a gene? The ancient function of the FH2 domain is to control actin polymerization given that FH2 domains from yeast to mammals harbor such an activity(Zigmond, 2004). However, our sequence and biochemical analysis suggests that the FH2 domain of FOZI-1 has lost this function. Nevertheless, the FH2 domain of FOZI-1 does harbor some function as its deletion compromises the ability to rescue the fozi-1mutant phenotype. Our analysis of the FH2 domain of FOZI-1 suggests that the domain has retained its ability to homodimerize, another ancient property of the domain that has been reported for all other characterized FH2 domain proteins. Transcription factors often dimerize, which increases their DNA-binding surface and enables cooperative DNA binding(Harrison, 1991). It is conceivable that the FOZI-1 FH2 domain serves the similar purpose of doubling the DNA contact surface of FOZI-1 through homodimerization. This may also explain why the FOZI-1 protein may still function, albeit only partially, upon deletion of the FH2 domain. In our in vivo rescue assays, transgenic array-induced overexpression of dimerization-defective FOZI-1 protein may increase the cellular FOZI-1 protein concentration enough to alleviate the need for dimerization-induced recruitment of proteins to DNA. Future identification of FOZI-1 DNA target sites and biochemical assays will test the validity of this hypothesis.

### fozi-1 and lim-6 mutants define a novel, mixed'ASE state

Previously, we have defined two states for ASEL/R gene expression profiles:hybrid and stable (Johnston et al.,2005). The hybrid state occurs in both ASEL and ASER during embryonic and early larval stages in which most ASE markers (including the ASEL-inducer lsy-6 and the ASER-inducer cog-1) are expressed in both ASEL and ASER. Dependent on the activity of the bi-stable feedback loop, this hybrid state of gene expression subsequently becomes restricted to either the ASEL or ASER stable state. Genetic ablation of feedback loop components causes both ASE cells to take on the complete ASEL cell fate(class I' mutants including cog-1) or the complete ASER cell fate(class II' mutants including lsy-6, die-1, lsy-2).

fozi-1 null mutant animals display an unusual phenotype. Rather than exhibiting a complete switch of cell fate in ASEL or ASER, fozi-1 mutants display a mixed' fate phenotype in ASER. Whereas ASEL displays its ASEL-specific gene expression profile, ASER expresses ASEL terminal markers while maintaining expression of ASER terminal markers. In an almost mirror image of fozi-1-null mutant animals, lim-6-null mutant animals adopt an essentially normal ASER state in ASER but fail to repress specific ASER terminal markers in ASEL, again causing a mixed' state, this time in ASEL. Taken together, both lim-6(expressed in ASEL) and fozi-1 (expressed in ASER) are regulatory intermediaries that transduce an output from the bi-stable feedback loop. These factors thereby enable the progression of differentiation states from a hybrid precursor state to a terminally differentiated end state.

### ASE subclass determination involves a complex regulatory architecture composed of several network motifs

Systematic analyses of gene regulatory networks in unicellular organisms have revealed that transcription factors interact in numerous combinations of simple network motifs (Lee et al.,2002; Shen-Orr et al.,2002). Here and in our previous work, we have extended the concept of defined network motifs to cell fate determination in the nervous system of metazoan organisms. We have shown (1) that a bi-stable feedback loop motif is a key decision point in the ASEL/R fate determination process(Johnston et al., 2005) and(2) that multiple distinct network motifs can be intertwined into a multi-tier regulatory architecture (this paper). The usefulness in considering these motifs in the context of ASEL/R cell fate determination lies in the well-defined properties of network motifs that can be mathematically modeled(Fig. 7B) and which provide clear predictions about the underlying logic of ASEL/R fate determination.

Specifically, the analysis of the fozi-1 gene and its relationship with other regulatory network components suggests novel regulatory motifs that act in conjunction with the previously described bi-stable feedback loop. Emanating from the die-1 Zn finger transcription factor, and including the fozi-1 Zn finger factor, all these motifs appear to be variants of the feed-forward loop (FFL) network motif(Fig. 7B,C), a motif commonly found in transcription factor networks(Lee et al., 2002; Shen-Orr et al., 2002). A defining feature of FFL motifs is that they provide a persistence detector that measures whether a gene regulatory input persists long enough before target genes are activated (Fig. 7B) (Mangan and Alon,2003; Mangan et al.,2003). In the cases described here, we infer the existence of FFL network motifs based on the variable expressivity of null mutant phenotypes(see legend to Fig. 7D for detailed explanations). With the possible exception of hen-1, all left/right asymmetrically expressed genes may be regulated through FFL motifs. In each of these FFL motif configurations, die-1 and fozi-1jointly regulate a target gene (Fig. 7C); expression of the target requires both the presence of die-1 and the absence of fozi-1. The target gene is either a terminal differentiation marker (gcy-7 etc.) or it is another regulatory factor, lim-6, which then activates FLP genes and represses ASER-specific GCY genes (Fig. 7D).

### Connecting gene regulatory network motifs

Apart from being the starting point for several presumptive feed-forward loops, the die-1 Zn-finger transcription factor is the output regulator of a bi-stable, double-negative feedback loop, which contains additional transcription factors and miRNAs, and was the first network motif identified in ASE fate specification (Fig. 7A). die-1 therefore represents a crucial `node' in connecting the bi-stable feedback loop to the feed-forward loops described here. The theoretical behaviors of feedback and feed-forward motifs allow us to speculate on how the ASEL/R fate decision may occur. The bistable feedback loop may allow amplification of an initial, transient input into the system. Such an input could be an intrinsic, lineage-derived cue or an externally provided signal. After the reception of this input in a left/right asymmetric manner, the feedback loop may increase and/or stabilize die-1 levels in ASEL. As the feed-forward loops that emanate from die-1 may act as a persistence detector, target genes will become activated only once the feedback loop has ensured that die-1 levels are persistently above a specific threshold level. The combined feedback and feed-forward regulatory motifs finely tune the system, amplifying a crucial input and assessing the stabilization and continuity of this amplification. We anticipate that the overall logic of the ASEL/R cell fate choice may apply to other cell fate decisions that have not yet been examined in extensive genetic detail.

We thank S. Mitani (Tokyo Women's Medical University School of Medicine)for the tm563 allele, Y. Kohara for fozi-1 EST clones, Q. Chen for expert injection assistance and P. Sengupta, U. Alon, C. Desplan and Hobert laboratory members for discussion and comments on the manuscript. This work was funded by an NSF predoctoral fellowship (R.J.J.) and by NIH R01 NS050266-01 (O.H.). B.H. and O.H. are Investigators of the Howard Hughes Medical Institute. J.C. is supported by a grant from CIHR.

Chang, S., Johnston, R. J., Jr and Hobert, O.(
2003
). A transcriptional regulatory cascade that controls left/right asymmetry in chemosensory neurons of C. elegans.
Genes Dev.
17
,
2123
-2137.
Chang, S., Johnston, R. J., Frokjaer-Jensen, C., Lockery, S. and Hobert, O. (
2004
). MicroRNAs act sequentially and asymmetrically to control chemosensory laterality in the nematode.
Nature
430
,
785
-789.
Copeland, J. W., Copeland, S. J. and Treisman, R.(
2004
). Homooligomerization is essential for F-actin assembly by the formin family FH2 domain.
J. Biol. Chem.
279
,
50250
-50256.
Davidson, E. H. (
2001
).
Genomic Regulatory Systems
Fukushige, T., Hawkins, M. G. and McGhee, J. D.(
1998
). The GATA-factor elt-2 is essential for formation of the Caenorhabditis elegans intestine.
Dev. Biol.
198
,
286
-302.
Harrison, S. C. (
1991
). A structural taxonomy of DNA-binding domains.
Nature
353
,
715
-719.
Hobert, O. (
2002
). PCR fusion-based approach to create reporter gene constructs for expression analysis in transgenic C. elegans.
Biotechniques
32
,
728
-730.
Hobert, O., Tessmar, K. and Ruvkun, G. (
1999
). The Caenorhabditis elegans lim-6 LIM homeobox gene regulates neurite outgrowth and function of particular GABAergic neurons.
Development
126
,
1547
-1562.
Hodgkin, J. and Doniach, T. (
1997
). Natural variation and copulatory plug formation in Caenorhabditis elegans.
Genetics
146
,
149
-164.
Iuchi, S. (
2001
). Three classes of C2H2 zinc finger proteins.
Cell. Mol. Life Sci.
58
,
625
-635.
Johnston, R. J. and Hobert, O. (
2003
). A microRNA controlling left/right neuronal asymmetry in Caenorhabditis elegans.
Nature
426
,
845
-849.
Johnston, R. J., Jr and Hobert, O. (
2005
). A novel C. elegans zinc finger transcription factor, lsy-2, required for the cell type-specific expression of the lsy-6 microRNA.
Development
132
,
5451
-5460.
Johnston, R. J., Jr, Chang, S., Etchberger, J. F., Ortiz, C. O. and Hobert, O. (
2005
). MicroRNAs acting in a double-negative feedback loop to control a neuronal cell fate decision.
102
,
12449
-12454.
Kovar, D. R., Kuhn, J. R., Tichy, A. L. and Pollard, T. D.(
2003
). The fission yeast cytokinesis formin Cdc12p is a barbed end actin filament capping protein gated by profilin.
J. Cell Biol.
161
,
875
-887.
Lee, T. I., Rinaldi, N. J., Robert, F., Odom, D. T., Bar-Joseph,Z., Gerber, G. K., Hannett, N. M., Harbison, C. T., Thompson, C. M., Simon, I. et al. (
2002
). Transcriptional regulatory networks in Saccharomyces cerevisiae.
Science
298
,
799
-804.
Mangan, S. and Alon, U. (
2003
). Structure and function of the feed-forward loop network motif.
100
,
11980
-11985.
Mangan, S., Zaslaver, A. and Alon, U. (
2003
). The coherent feedforward loop serves as a sign-sensitive delay element in transcription networks.
J. Mol. Biol.
334
,
197
-204.
Milo, R., Shen-Orr, S., Itzkovitz, S., Kashtan, N., Chklovskii,D. and Alon, U. (
2002
). Network motifs: simple building blocks of complex networks.
Science
298
,
824
-827.
Notredame, C., Higgins, D. G. and Heringa, J.(
2000
). T-Coffee: A novel method for fast and accurate multiple sequence alignment.
J. Mol. Biol.
302
,
205
-217.
Ortiz, C. O., Etchberger, J. F., Posy, S. L.,Frøkjær-Jensen, C., Lockery, S., Honig, B. and Hobert, O.(
2006
). Searching for neuronal left/right asymmetry: Genome wide analysis of nematode receptor-type guanylyl cyclases.
Genetics
173
,
131
-149.
Otomo, T., Tomchick, D. R., Otomo, C., Panchal, S. C., Machius,M. and Rosen, M. K. (
2005
). Structural basis of actin filament nucleation and processive capping by a formin homology 2 domain.
Nature
433
,
488
-494.
Pierce-Shimomura, J. T., Faumont, S., Gaston, M. R., Pearson, B. J. and Lockery, S. R. (
2001
). The homeobox gene lim-6 is required for distinct chemosensory representations in C. elegans.
Nature
410
,
694
-698.
Pruyne, D., Evangelista, M., Yang, C., Bi, E., Zigmond, S.,Bretscher, A. and Boone, C. (
2002
). Role of formins in actin assembly: nucleation and barbed-end association.
Science
297
,
612
-615.
Shen-Orr, S. S., Milo, R., Mangan, S. and Alon, U.(
2002
). Network motifs in the transcriptional regulation network of Escherichia coli.
Nat. Genet.
31
,
64
-68.
Shimada, A., Nyitrai, M., Vetter, I. R., Kuhlmann, D., Bugyi,B., Narumiya, S., Geeves, M. A. and Wittinghofer, A. (
2004
). The core FH2 domain of diaphanous-related formins is an elongated actin binding protein that inhibits polymerization.
Mol. Cell
13
,
511
-522.
Simmer, F., Tijsterman, M., Parrish, S., Koushika, S., Nonet,M., Fire, A., Ahringer, J. and Plasterk, R. (
2002
). Loss of the putative RNA-directed RNA polymerase RRF-3 makes C. elegans hypersensitive to RNAi.
Curr. Biol.
12
,
1317
.
Tang, C. L., Xie, L., Koh, I. Y., Posy, S., Alexov, E. and Honig, B. (
2003
). On the role of structural information in remote homology detection and sequence alignment: new methods using hybrid sequence profiles.
J. Mol. Biol.
334
,
1043
-1062.
Titz, B., Thomas, S., Rajagopala, S. V., Chiba, T., Ito, T. and Uetz, P. (
2006
). Transcriptional activators in yeast.
Nucleic Acids Res.
34
,
955
-967.
Valdar, W. S. (
2002
). Scoring residue conservation.
Proteins
48
,
227
-241.
White, J. G., Southgate, E., Thomson, J. N. and Brenner, S.(
1986
). The structure of the nervous system of the nematode Caenorhabditis elegans.
Philos. Trans. R. Soc. Lond. B Biol. Sci.
314
,
1
-340.
Xu, Y., Moseley, J. B., Sagot, I., Poy, F., Pellman, D., Goode,B. L. and Eck, M. J. (
2004
). Crystal structures of a Formin Homology-
2
domain reveal a tethered dimer architecture.
Cell
116
,711-723.
Yu, S., Avery, L., Baude, E. and Garbers, D. L.(
1997
). Guanylyl cyclase expression in specific sensory neurons:a new family of chemosensory receptors.