Interchangeability of Caenorhabditis elegans DSL proteins and intrinsic signalling activity of their extracellular domains in vivo

During animal development, intercellular communication, mediated by ligand-receptor interactions, plays critical roles in specifying cell fates. Ligands of the DSL family (for Delta, Serrate, lag-2) are important mediators of intercellular communication. DSL proteins are transmembrane proteins with a ‘DSL’ domain and epidermal growth factor (EGF)-like motifs in their extracellular domains, and relatively short intracellular domains (reviewed in Simpson, 1995). Receptors for DSL proteins are members of the lin-12/Notch family. These receptors are transmembrane proteins with multiple EGF-like motifs and three LIN-12/Notch repeat (LNR) motifs in their extracellular domains and six cdc10/SWI6 (also termed ‘ankyrin’) motifs in their intracellular domains (reviewed in Greenwald, 1994; Artavanis-Tsakonas et al., 1995).

In C. elegans, two ligands of the DSL family, lag-2 and apx-1, have been described. LAG-2 and APX-1 are only 25% identical, and have different numbers of epidermal growth factor (EGF)-like motifs. LAG-2 appears to be the ligand for the lin-12-mediated anchor cell/ventral uterine precursor cell (AC/VU) decision in the hermaphrodite, the glp-1-mediated mitosis/meiosis decision in the germline, and lin-12 or glp-1-mediated decisions affecting viability (Lambie and Kimble, 1991; Henderson et al., 1994; Tax et al., 1994; Wilkinson et al., 1994). APX-1 appears to be a maternally provided ligand for a glp-1-mediated decision in the early embryo (Mango et al., 1994; Mello et al., 1994). apx-1 mutants do not have any apparent defects in later embryonic or postembryonic development, although this issue remains open until null mutants of apx-1 are examined.

Genetic studies have not provided any evidence for functional redundancy of lag-2 and apx-1 (Mello et al., 1994). Because lag-2 and apx-1 mutants have different phenotypes, and LAG-2 and APX-1 have a relatively low level of homology, they might be biochemically distinctive and hence not functionally interchangeable. Furthermore, the fact that the DSL family have conserved amino acid sequence motifs raises the issue of the function of different structural domains. In this paper, we address these issues by expressing wild-type and mutant proteins under the control of lag-2 regulatory sequences and examining their ability to replace lag-2 in defined cell fate decisions and to activate LIN-12 and GLP-1 ectopically. We provide evidence that APX-1 and LAG-2 are biochemically interchangeable. We also show that secreted extracellular domains of DSL ligands have intrinsic signalling activity and that the DSL domain alone appears to enhance the activity of the endogenous LAG-2 protein.

Methods for handling and culturing C. elegans have been described by Brenner (1974). The wild-type parent for all strains used was C.

elegans var. Bristol strain N2 (Brenner, 1974). The key mutations used in this study, and their relevant features, were the following:

LGI: smg-1(r861) and unc-54 (r293) (Hodgkin et al., 1989). smg-1 mutations stabilize aberrant mRNAs (Pulak and Anderson, 1993), and unc-54(r293) is suppressed by smg-1(r293).

LGIII: unc-32(e189), dpy-18(e364) (Brenner, 1974), and glp-1(e2141ts) (Priess et al., 1987).

LGV: lag-2(s1486) (Johnsen and Baillie, 1991), lag-2(q411) (Henderson et al., 1994), and unc-46(e177) (Brenner, 1974).

eT1 is a reciprocal translocation between LGIII and LGV whose breakpoint appears to be in unc-36 (Rosenbluth and Baillie, 1981).

pLGX was constructed from p226. p226 contains a 4.0 kb genomic lag-2 fragment capable of complementing lag-2(0) mutations and was derived from JT225 and JT280 as described by Wilkinson et al., (1994). The genomic fragment contains 3.1 kb of 5′ flanking region as well as the lag-2 coding and 3′ untranslated regions. pLGX, a lag-2 expression vector, was constructed by using the polymerase chain reaction to insert a SpeI linker at position +1 of the genomic fragment.

pLGX::apx-1 expresses the full length apx-1 cDNA (Mello et al., 1994) under lag-2 promoter control. pLGX::GFP was constructed using the polymerase chain reaction to create SpeI sites both 5′ of the GFP ATG and 3′ of the GFP TAA. This PCR product was digested with SpeI and ligated to pLGX which had also been digested with SpeI. pLGX::apx-1 was constructed by creating XbaI restriction sites both 5′ of the apx-1 ATG and 3′ of the apx-1 TAA . This apx-1 cDNA PCR product, which is flanked by XbaI sites, was cloned into pLGX digested with SpeI, resulting in pLGX::apx-1. Truncated derivatives of apx-1 and lag-2 expressed in the pLGX vector were constructed using PCR in a similar fashion to that used to construct pLGX::apx-1 with the following modifications: the 3′ and 5′ primers both contained SpeI sites instead of XbaI sites, and the 3′ primers contained an in-frame stop codon. The 3′ stop codon containing primers were at the following positions: for pLGX::apx-1(extra), amino acid 402; for pLGX::apx-1(DSL), amino acid 188; for pLGX::lag-2(extra), amino acid 281. pLGX::apx-1(1/2DSL) was created by placing a stop codon into a unique NruI site at amino acid 159 in pLGX::apx-1. pLGX::apx-1(extra)+GFP is the same as pLGX::apx-1(extra) except the 3′ stop codon is now derived from Green Fluorescent Protein (Chalfie et al., 1994). GFP was inserted, in frame, on the 3′ end of apx-1(extra), thereby creating an apx-1(extra)::gfp gene fusion. pLGX::apx-1(extra)+ TM-lacZ was created in a similar fashion to pLGX::apx-1(extra) + gfp with lacZ inserted with its own stop codon in frame to apx-1(extra) at amino acid 402. The inserted modified lacZ gene was derived from a combination of the plasmids 16.43 and 34.110 (Fire et al., 1990).

Transgenic lines were established by microinjection of plasmid mixtures into the hermaphrodite germline to create extrachromosomal arrays (Fire, 1986; Mello et al., 1991). By accepted convention, ‘Ex’ is used to represent extrachromosomal arrays. pLGX and derivatives were injected at 1 μg/ml into a recipient strain of genotype smg-1(r861); dpy-18(e364)/eT1; lag-2(s1486) unc-46(e177)/eT1 and at 50 μg/ml into a recipient strain of genotype smg-1(r861) unc-54(r293). pRF4, a plasmid containing the cloned dominant rol-6(su1006) gene (Mello et al., 1991) was used as a cotransformation marker and injected at a concentration of 100 μg/ml. F₁ Roller progeny were picked, and F₂ Roller progeny used to establish lines.

To assess rescue of lag-2(0) phenotypes, all Rol progeny from at least three heterozygous smg-1(r861); dpy-18(e364)/eT1; lag-2(s1486) unc-46(e177)/eT1; Ex hermaphrodites were picked individually and scored in the next generation for Dpy-18 Unc-46 Rol-6 phenotypes. Rescued homozygous lines segregated only viable Dpy Unc Rol animals and Lag arrested larvae, indicating that the extra-chromosomal arrays were necessary for viability. At 1 μg/ml, 4/4 independent lines of pLGX::apx-1(+), 2/2 lines of pLGX::apx-1(extra), 2/3 of pLGX::apx-1(extra-GFP), and 2/2 of pLGX::apx-1(extra+TM-BGAL) displayed rescue; transgenic lines could be maintained indefinitely at 25°C. Living hermaphrodites segregating from these strains were scored under Nomarski optics for the number of anchor cells and germline anatomy. At 1 μg/ml, 0/4 lines of pLGX, 0/2 of pLGX::apx-1(DSL), and 0/2 of pLGX::apx-1(1/2 DSL) rescued the lethality of lag-2(s1486). When the concentration of pLGX or pLGX::apx-1(DSL) was increased, rescue was still not obtained: for pLGX::apx-1(DSL), 0/2 lines made at 25 μg/ml rescued lag-2(s1486), and 0/2 lines of pLGX made at 50 μg/ml rescued lag-2(s1486). In addition, 0/1 lines of pLGX::apx-1(DSL) made at 50 μg/ml rescued lag-2(q411).

Rescued hermaphrodites [relevant genotype: lag-2(s1486); Ex ] were also examined for phenotypes associated with ectopic glp-1 or lin-12 activation. Rescued hermaphrodites containing pLGX::apx-1(extra) and pLGX::apx-1(extra-GFP) exhibited a low penetrance (2-20%) tumorous germline.

Hermaphrodites carrying extrachromosomal arrays made at 50 μg/ml [relevant genotype: lag-2(+); Ex] displayed phenotypes associated with ectopic glp-1 or lin-12 activation, with penetrances ranging from 10 to 80%. All transgenic lines containing pLGX::apx-1(extra), pLGX::apx-1(extra-GFP); pLGX::lag-2(extra), or pLGX::apx-1(DSL) segregated hermaphrodites with a tumorous germline and/or a Multivulva phenotype. For pLGX::apx-1(DSL), these phenotypes were seen in transgenic lines carrying extrachromosomal arrays made at 25 μg/ml and 50 μg/ml, but not at 1 μg/ml. The tumorous germline or Multivulva phenotype was never observed in lines carrying extrachromosomal arrays made from pLGX::apx-1(1/2 DSL), pLGX::apx-1(extra-TM-BGAL), pLGX::apx-1(+), and pLGX. Greater than 50 Rol-6 animals were observed for each line carrying these constructs.

In the experiments described here, we expressed lag-2 or apx-1coding sequences under the control of lag-2 regulatory sequences, using the pLGX lag-2 expression vector (Fig. 1 and Materials and Methods). The pLGX vector is based on a genomic clone that can rescue a lag-2(0) mutant (Tax et al., 1994; Tax, 1994). When the inserted cDNA encodes β-galactosidase or green fluorescent protein (GFP), reporter protein activity is detectable in a spatially and temporally restricted pattern consistent with genetic analysis of lag-2, including in Z1.ppp and Z4.aaa (Wilkinson et al., 1994) and the distal tip cells (Fig. 1). Since the mRNA expressed from such a construct is predicted to contain an abnormal, long 3′ untranslated region, efficient expression is only obtained in the presence of smg-1 mutations, which stabilize mRNAs with long 3′ untranslated regions (Pulak and Anderson, 1993). The smg-1 mutations do not appear to affect hermaphrodite gonadal anatomy or development, the vulval precursor cell lineages, or viability (Hodgkin et al., 1989; Pulak and Anderson, 1993; Wilkinson et al., 1994), and do not suppress the lag-2(s1486) and lag-2(q411) alleles used in this study.

Several independent extrachromosomal arrays were generated for each construct. Transgenic lines were examined in the presence of endogenous lag-2 activity for phenotypes associated with lin-12 and glp-1 activation, and in the absence of lag-2 activity for rescue of individual cell fate decisions requiring lag-2 activity. Details for individual constructs are given in Materials and Methods. However, we note that, for constructs that rescue lag-2(0) mutants, rescue was observed even when constructs were injected at a concentration of 1 μg/ml. At this concentration, extrachromosomal arrays contain a low copy number of construct (Mello et al., 1994) and should not be grossly overexpressed with respect to the normal level of lag-2 gene expression.

Here, we summarize what is known about the roles of lag-2 and lin-12 or glp-1 relevant to the phenotypes we scored.

lag-2 null mutants arrest as larvae with numerous cell fate transformations (Lambie and Kimble, 1991). The Lag phenotype is distinctive and resembles that of lin-12(-) glp-1(-) double mutants (Lambie and Kimble, 1991). Together with the finding that GLP-1 can fully substitute for LIN-12 (Fitzgerald et al., 1993), the available evidence suggests that LAG-2 normally activates LIN-12 and GLP-1 in certain cells so that the cell fate decisions required for viability occur properly.

The lin-12-mediated AC/VU decision involves an interaction between Z1.ppp and Z4.aaa, two initially equivalent cells of the somatic gonad. During wild-type development, Z1.ppp and Z4.aaa interact with each other, so that in each hermaphrodite one becomes the AC while the other becomes a VU (Kimble and Hirsh, 1979; Kimble, 1981; Seydoux and Greenwald, 1989). When lin-12 or lag-2 activity is eliminated, both Z1.ppp and Z4.aaa become ACs (Greenwald et al., 1993; Lambie and Kimble, 1991). Mutations that activate LIN-12 cause both cells to become VUs (Greenwald et al., 1983; Greenwald and Seydoux, 1990; Struhl et al., 1993).

A glp-1-mediated interaction between a somatic gonadal cell, the distal tip cell, and the germline promotes germline mitosis (and/or inhibits meiosis) (Kimble and White, 1981). The adult hermaphrodite gonad is a two-armed structure, with a distal tip cell at the end of each arm. Normally, each arm contains mitotic cells in the most distal region of the germline, followed consecutively by a region of meiosis, a region of maturing oocytes and a region of sperm (Hirsh et al., 1976; Kimble and Hirsh, 1979). When lag-2 or glp-1 activity is decreased, all germ cells enter meiosis (Austin and Kimble, 1987; Priess et al., 1987; Lambie and Kimble, 1991; Crittenden et al., 1994). A mutation that causes GLP-1 to be active independent of ligand causes germ cells to continue to divide mitotically (L. Wilson Berry and T. Schedl, personal communication).

Genetic analysis had indicated that LAG-2 can interact with both LIN-12 and GLP-1 (Lambie and Kimble, 1991; Tax et al., 1994), but apx-1 was only known to interact with glp-1 (Mango et al., 1994; Mello et al., 1994). LAG-2 and APX-1 are only 25% identical, and have different numbers of epidermal growth factor (EGF)-like motifs. By our criteria (Fig. 2 legend), there are two EGF-like motifs in LAG-2 and four in APX-1. We placed apx-1 under the control of lag-2 regulatory sequences and showed that expression of APX-1 in a lag-2(0) mutant can rescue the defects causing lethality, as well as defects in the AC/VU decision and germline development. Thus, APX-1 can fully substitute for LAG-2 (Figs 2,3).

In addition, C. Mello and J. Priess identified another DSL gene, arg-1, which encodes a DSL ligand of unknown function. ARG-1 has two EGF-like motifs and is only 25% identical to LAG-2. They kindly made arg-1 available to us, and we found that ARG-1 expressed under the control of lag-2 regulatory sequences can also substitute for LAG-2 (data not shown).

We completely replaced the transmembrane and intracellular domains of LAG-2 by a synthetic transmembrane domain and a modified β-galactosidase protein. The resulting chimeric protein can rescue the defects causing lethality, and defects in the AC/VU decision and germline development, and therefore is able to substitute fully for endogenous lag-2 activity (Fig. 3). This result implies that the transmembrane and intracellular domains are not necessary for LAG-2 activity (see also Henderson et al., 1994).

In order to test the dependence of the activity of the extracellular domain on membrane association, we completely deleted the transmembrane and intracellular domains of LAG-2 and APX-1. We found that just the extracellular domains of LAG-2 or APX-1 were able to rescue the defects causing lethality, and defects in the AC/VU decision and germline development of lag-2(0) mutants (Fig. 3). These observations suggest that the extracellular domains of DSL ligands have intrinsic signalling activity.

To demonstrate that these truncated proteins are secreted, we tagged the APX-1 extracellular domain with the Green Fluorescent Protein (Chalfie et al., 1994). The resulting protein is fully functional and is found in coelomocytes, suggesting it is secreted into the pseudocoelom (Fig. 4). Non-secreted forms of GFP expressed under the control of lag-2 regulatory sequences, as in pLGX::GFP (Fig. 1), are not found in coelomocytes (K. F., unpublished observations). Furthermore, we also see evidence that secreted extracellular domains can activate LIN-12 in the VPCs, causing a characteristic Multivulva phenotype (Fig. 4). This Multivulva phenotype implies that the secreted extra-cellular domains of DSL proteins can diffuse to and function at an ectopic location, since lag-2 does not appear to be expressed in the vulval precursor cells (K. F. and I. G., unpublished observations), and does not appear to function in the vulval precursor cells (Tax et al., 1994; S. Kaech and S. Kim, personal communication). A Multivulva phenotype was not observed in hermaphrodites containing transgenes encoding transmembrane DSL ligands.

While cell fate decisions often occur normally in animals expressing secreted LAG-2 or APX-1 extracellular domains, some hermaphrodites display germline abnormalities suggestive of GLP-1 ectopic activation or hyper-activation (Fig. 5). As described above, each gonadal arm of wild-type adult hermaphrodites shows an orderly progression of germline maturation, with mitotic cells limited to the most distal region of the germline. In contrast, in hermaphrodites expressing secreted LAG-2 or APX-1 extracellular domains, there is often mitosis in the proximal region of the germline (the ‘proximal mitosis’ phenotype; Seydoux et al., 1990) or mitosis throughout the germline (the ‘tumorous germline’ phenotype; Francis et al., 1995 and L. W. Berry and T. Schedl, personal communication).

These germline abnormalities appear to be caused by GLP-1 activation, since glp-1(e2141ts) hermaphrodites expressing the APX-1 extracellular domains may have tumorous germlines at the permissive temperature but, after shifting to the restrictive temperature, the germ cells enter meiosis (Fig. 6). Furthermore, the proximal mitosis and tumorous germline phenotypes caused by expression of the APX-1 extracellular domain occur when the activity of the endogenous lag-2 gene is eliminated, indicating that these phenotypes reflect the intrinsic activity of the secreted ligand (Fig. 5). Abnormal germline development was never observed in animals carrying multiple copies of transgenes encoding intact APX-1 or LAG-2 proteins, or the APX-TM-βGAL protein, suggesting that the secreted extracellular domain or DSL domain may be more intrinsically active, more stable, or able to diffuse further than transmembrane protein ligands.

The DSL domain is found only in DSL proteins and its role is not known. To examine this question, we expressed a secreted DSL domain of LAG-2 or APX-1. The secreted DSL domain is not sufficient to rescue the lethality of a lag-2(0) mutant. However, it is able to cause the proximal germline mitosis or tumorous germline phenotype in the presence of lag-2(+) activity. Thus, a secreted DSL domain alone may enhance endogenous LAG-2 signalling. Alternatively, a secreted DSL domain may have intrinsic signalling activity that is insufficient to rescue the lethality of lag-2(0) but is sufficient to activate GLP-1 in the germline.

We have shown that two different C. elegans DSL proteins, APX-1 and ARG-1, can substitute for LAG-2 when expressed under control of lag-2 regulatory sequences. All phenotypes known to be associated with eliminating lag-2 activity, including defects leading to lethality and defects in the AC/VU decision and germline development, were rescued by APX-1 or ARG-1. The fact that APX-1 and LAG-2 appear to be interchangeable despite their considerable differences in amino acid sequence implies that they share conserved structural features. A similar observation has been made in Drosophila, where expression of Serrate under control of a heat shock promoter is able to suppress neuroblast segregation in the absence of Delta (Gu et al., 1995). The interchangeability of APX-1 and LAG-2 implies that differential gene expression rather than different biochemical activity underlies the distinct mutant phenotypes of lag-2 and apx-1.

We have shown that the transmembrane domain and the intracellular domain of LAG-2 can be completely replaced by a synthetic transmembrane domain and β-galactosidase without affecting the ability of the protein to mediate cell fate decisions. This result implies that the transmembrane and intracellular domains are not necessary for LAG-2 activity. Furthermore, when the transmembrane domain and intracellular domain of APX-1 is deleted, the resulting secreted protein can replace endogenous lag-2 activity in all cell fate decisions examined. These results imply that the extracellular domains of LAG-2 and APX-1 have intrinsic signalling activity.

The extracellular domains of DSL proteins have two conserved features: EGF-like motifs and the DSL domain. EGF-like motifs are found in the extracellular domains of many proteins. In DSL proteins, the EGF-like motifs are thought to bind to the EGF-like motifs of LIN-12/Notch proteins and are thought to mediate the ligand-receptor interaction (Rebay et al., 1991, 1993; Lieber et al., 1992, 1993; Rao et al., 1995). The DSL domain thus far appears to be unique to DSL proteins (Tax et al., 1994). We have found that a secreted DSL domain from LAG-2 or APX-1 can cause ectopic activation of GLP-1 in the germline in the presence of endogenous lag-2 activity, but is unable to rescue the lethality of a lag-2(0) mutant, even at a high concentration of injected transgene. One possibility is that the secreted DSL domain has a low level of intrinsic activity that is insufficient to rescue the lethality of lag-2(0) but is sufficient to activate GLP-1 ectopically in the germline. However, rescue of defects causing lethality seems to require a lower level of lag-2 activity than rescue of defects in the AC/VU decision or germline development (Lambie and Kimble, 1991; K.F., unpublished observations). We therefore favor the possibility that the secreted DSL domain enhances the activity of the intact LAG-2 protein. The relevance of this observation for normal ligand activity is not clear, but one possibility is that the DSL domain mediates oligomerization of DSL proteins to form an active ligand.

The secreted extracellular domains of LAG-2 and APX-1 can promote normal cell fate decisions. However, some hermaphrodites expressing these mutant proteins displayed phenotypes associated with ectopic activation of LIN-12 or GLP-1, even in the absence of endogenous lag-2 activity. Greater stability or mobility of secreted extracellular domains may underlie this apparent increase in intrinsic activity.

Ligand activity of DSL proteins may be regulated at the level of ligand stability. All DSL proteins contain PEST protein turnover sequences (Rogers et al., 1986) in their intracellular domains. An indication that protein stability influences APX-1 activity is that some hermaphrodites expressing a mutant protein containing a partial deletion of the PEST sequence of APX-1 have a tumorous germline phenotype (K. F., unpublished observations). In contrast, ectopic activity is not observed when a modified β-galactosidase that is highly unstable in C. elegans (Way and Chalfie, 1989; Fire et al., 1990; Wilkinson et al., 1994) replaces the entire intracellular domain of APX-1 (this work) or part of the intracellular domain of LAG-2 (Henderson et al., 1994; K. F., unpublished observations).

The prevention of inappropriate cell-cell interactions appears to be important during development (e.g., Henry and Grainger, 1987; Seydoux et al., 1990; Waring and Kenyon, 1990). Membrane-association of DSL proteins, which potentially mediate many different cell-cell interactions, may be one mechanism to prevent inappropriate interactions. Our observation that the secreted extracellular domain of LAG-2 or APX-1 appears to activate LIN-12 in the vulval precursor cells, even though LAG-2 does not appear to be the normal ligand for LIN-12 during vulval precursor cell specification (Tax et al., 1994; S. Kaech and S. Kim, personal communication; K. F., unpublished observations), suggests that inappropriate cell-cell interaction may occur when a DSL protein is able to dissociate from the membrane.

If DSL proteins normally remain associated with the membrane, it implies that cell-cell contact is necessary for ligand-receptor interactions. In lin-12, glp-1 and Notch-mediated decisions, signalling and receiving cells are in contact. Furthermore, if cell contact between Z1.ppp and Z4.aaa is disrupted in the forming gonad primordium, both cells differentiate as anchor cells (Hedgecock et al., 1987). Our observation that the distal tip cells appear to have processes that in some cases can be seen to extend as far as the zone of germline mitosis (see Fig. 1) may be relevant to this issue. The distal tip cell influences the decision (mitosis/meiosis) of germline nuclei that are at least twenty nuclear diameters away from the distal end of the gonad (Kimble and White, 1981). The distal tip cell has been thought to contact only the distal end of the germline syncytium, so that its influence was mediated by a diffusible factor (Kimble and White, 1981), such as an intracellular second messenger produced upon GLP-1 activation (Yochem and Greenwald, 1989; Henderson et al., 1994). However, our observation that distal tip cells extend long processes suggests that direct contact between distal tip cells and membranes associated with germline nuclei throughout the mitotic zone may be important.

The interchangeability of C. elegans DSL proteins and of Drosophila DSL proteins suggests that the multiple DSL proteins found in vertebrate systems may also be interchangeable. In addition, our observation that DSL proteins are active, and indeed may be hyperactive, when the transmembrane and intracellular domains are removed will be useful for investigating cell-cell interactions in mammals that are mediated by DSL proteins and LIN-12/Notch proteins (e.g., Chitnis et al., 1995; Lindsell et al., 1995). Secreted extracellular domains will be useful reagents for tissue culture and biochemical studies. Furthermore, since activated forms of LIN-12/Notch family members have been associated with cancer in mammals (Ellisen et al., 1991; Robbins et al., 1992), it is conceivable that truncated forms of DSL proteins could also function as oncogenes.

We thank Craig Mello and Jim Priess, and Frans Tax and Jim Thomas, for their generosity with unpublished as well as published clones and information; David Baillie, Sam Henderson and Judith Kimble for lag-2 alleles; Andy Fire and Marty Chalfie for reporter genes; Tim Schedl for helpful discussion; Richard Ruiz and Denise Brousseau for valuable assistance; and Mike Beck, Tim Doyle, Jane Albert Hubbard, Gary Struhl and Simon Tuck for critical comments on the manuscript. This work was supported by National Institutes of Health grant GM37602. I. G. is an Associate Investigator of the Howard Hughes Medical Institute.