The chicken β-globin insulator element conveys chromatin boundary activity but not imprinting at the mouse Igf2/H19 domain

In placental mammals a small subset of genes is expressed from only one of the two available alleles, according to parental origin – the imprinted genes (Bartolomei and Tilghman, 1997; Ferguson-Smith and Surani, 2001) (see http://www.mgu.har.mrc.ac.uk/imprinting/imprinting.html). As the two alleles of imprinted genes exhibit differential epigenetic states in the same cell, and these states are faithfully maintained throughout cell division, they constitute a unique resource for the understanding of epigenetic processes of gene regulation during development.

Much evidence has accumulated that the methylation of cytosine residues in CpGs of DNA is an integral part of the system of genomic imprinting. Most, if not all, imprinted genes are associated with differentially methylated regions (DMRs) (Mann et al., 2000) and disruption of this differential methylation, usually through loss of hypermethylation at one allele, results in loss of monoallelic expression (Caspary et al., 1998; Howell et al., 2001; Li et al., 1993; Tucker et al., 1996). Primary DMRs are those DMRs inherited from the gametes, and their methylation may constitute the epigenetic mark that transmits imprinting information from gamete to embryo (Nabetani et al., 1997; Shibata et al., 1997; Shibata et al., 1998; Tremblay et al., 1997; Tremblay et al., 1995). A stronger case for this can be made for sperm in comparison to egg DNA, because, being packaged by protamines, there is no opportunity for the transmission of epigenetic information by a ‘histone code’ (Jenuwein and Allis, 2001). Aside from DNA methylation, sperm DNA may be relatively epigenetically naïve until it enters the ooplasm.

To understand the mechanisms of genomic imprinting, much work has focused on the imprinted genes insulin-like growth factor 2 (Igf2), an embryonic mitogen, and H19, which produces an untranslated RNA with no known function. In mouse, these genes are located ∼80 kb apart on distal chromosome 7 and are coordinately expressed in tissues of mesoderm and definitive endoderm origin. This coordination is due to the sharing of a suite of enhancers located downstream of H19 (Bartolomei et al., 1993; Leighton et al., 1995b; Zemel et al., 1992). Monoallelic expression of the two genes is regulated by an intervening imprinting control element (ICR) located –2 kb to –4.4 kb relative to the transcription start site (RTSS) of H19 (Leighton et al., 1995a; Ripoche et al., 1997; Thorvaldsen et al., 1998). On the maternal chromosome, the ICR acts as a chromatin insulator through the binding of the protein CTCF and prevents access of the Igf2 promoter to the enhancers (Bell and Felsenfeld, 2000; Bell et al., 1999; Hark et al., 2000; Kanduri et al., 2000b; Srivastava et al., 2000; Szabó et al., 2000). On the paternal chromosome, the ICR lacks insulator activity and the paternal Igf2 promoter now accesses the enhancers (Fig. 1A). The idea that methylation inhibits insulator activity of the ICR (Schmidt et al., 1999) has considerable support from a number of studies: (1) loss of paternal ICR methylation leads to loss of Igf2 activity in cis or the gain of insulator function (Li et al., 1993); (2) no protein binding is present at the four CTCF sites of the hypermethylated paternal ICR, as shown by in vivo footprinting (Szabó et al., 2000), and no CTCF protein is associated with the paternal ICR, as shown by chromatin immunoprecipitation (Kanduri et al., 2000b); (3) pre-methylation of the ICR leads to the loss of CTCF binding in electrophoretic mobility shift assays – methylation of only the single CpG in the consensus binding sequence CCGCnnGGnGGCAG is sufficient for inhibition (Bell and Felsenfeld, 2000; Hark et al., 2000; Kanduri et al., 2000b); (4) gain of maternal ICR methylation leads to gain of Igf2 activity in cis, or the loss of insulator function (Reed et al., 2001); and (5) methylation of the ICR in vitro leads to loss of insulator activity in an episome system (Holmgren et al., 2001).

The hypermethylated paternal ICR, while lacking insulator activity, acts to inactivate the H19 promoter in cis during early development (Srivastava et al., 2000; Thorvaldsen et al., 1998). This element becomes hypermethylated and packaged into a closed chromatin structure (Bartolomei et al., 1993; Ferguson-Smith et al., 1993; Szabó et al., 2000). As this epigenetic information is inherited from sperm, it is likely that this function is dependent on the methylation in the ICR.

The germline-specific processes that determine differential ICR methylation, which could be viewed as equivalent to the imprinting mechanisms, are undefined. In addition, it is not known whether these processes are entirely separate from the later somatic ICR functions of chromatin insulation and H19 promoter silencing, or if they are in some way connected. What is apparent is that the acquisition of differential ICR methylation is an autonomous function, at least when the ICR is moved to another location within the Igf2/H19 region (Srivastava et al., 2000). Furthermore, the property is retained, although inconsistently, in randomly integrated multiple copy H19 transgenes (Brenton et al., 1999; Elson and Bartolomei, 1997; Stadnick et al., 1999).

To examine these questions, a dissection of the ICR sequences through gene targeting should identify those that are important for mediating its various properties. In this study, we describe the results of a related approach in which the ICR was substituted for two copies of the chicken β-globin insulator element (ChβGI) in mice. This introduced sequence, (ChβGI)₂, resembles the ICR in size, and in containing CTCF-binding sites and CpGs, but otherwise lacks homology. We hoped to shed light on the answers to two questions: (1) can the (ChβGI)₂ effectively substitute for the ICR as a chromatin insulator; and (2) if so, could differential CTCF binding in the two germlines be a sufficient explanation for the establishment of differential methylation and imprinting in the context of the Igf2/H19 domain? This latter possibility should not be considered unlikely on the basis that other CTCF insulator sites in the genome are not associated with paternal methylation in somatic cells. As their methylation state in sperm has not been examined, it is possible that it is acquired in the male germline, but then lost during fertilization as part of general demethylation of the paternal genome at this stage (Oswald et al., 2000).

We show that on maternal and paternal inheritance, the (ChβGI)₂ functioned as a strong chromatin insulator, suggesting that the ICR is not uniquely specialized for chromatin insulation at the Igf2/H19 domain. In addition, on maternal and paternal inheritance, the (ChβGI)₂ was hypomethylated, i.e. the property of differential methylation acquisition or imprinting was lost. This indicates that differential ICR methylation is not mediated through differential CTCF binding.

A standard replacement vector for targeting in embryonic stem (ES) cells was designed to enable substitution of the ICR. The ICR as defined here is delineated by two BglII sites at –2.0 kb and –4.4 kb RTSS of H19. The 3′ arm was a 11.6 kb BglII genomic fragment from a lambda library of strain 129/SvJ (Stratagene, La Jolla, CA) extending from –2.0 kb to +9.6 kb RTSS of H19. The 5′ arm was a 1.1 kb fragment extending from –4.4 kb to –5.5 kb RTSS of H19 and was made by PCR using primers 5′-AAAAATCGATCCTTCCTGGGTAATACCTCTGAGCCTG-3′ (upper) and 5′-AAAAATCGATTCTTCTTCCAGAAACAAGTTAGGCATG-3′ (lower) (GenBank Accession Number, AF049091). The ClaI site near the end of both primers allowed for subcloning. At the end of the 5′ arm the RNAPolII promoter-diphtheria toxin, a chain-SV40 polyA (dptx) negative selection cassette (Lexicon Genetics, The Woodlands, TX), was cloned; between the two arms, the Pgk1 promoter-neomycin-bovine polyA (neo)-positive selection cassette flanked by loxP sites (floxed) was cloned for later excision using Cre recombinase. This base-targeting vector contained a unique BglII site just 3′ of the neo cassette and between the two arms of H19 sequence – this represented the site of deletion of the ICR. At this BglII site the 2.4 kb (ChβGI)₂, as excised from plasmid pJC13-1 (Chung et al., 1993), was inserted using EcoRI and BamHI. Upon homologous recombination, followed by excision of the floxed neo cassette with Cre, the ICR was substituted for the (ChβGI)₂, together with a single loxP site flanked by a small amount of polylinker sequence directly adjacent to the 5′ end of the (ChβGI)₂ (Fig. 2).

The vector was linearized at the end of the long arm using NotI, then introduced into W9.5 ES cells of strain 129S1/SvImJ (Szabó and Mann, 1994) by electroporation. Probe a, 0.5 kb, was made by PCR using primers 5′-GGTGCCATCAAGCTACTACAC-3′ (upper) and 5′-CTGGATAGGACATGGGCACAG-3′ (lower) (sequences from GenBank Accession Number, AC013548). Probe b was an EcoRI-BamHI fragment as indicated (Fig. 2). Seven clones were obtained from 144 screened. Excision of the neo cassette was achieved by mating male chimeras with Cre-deleter females. The latter transgenic line, a mixture of 129S1 and Swiss in this study, was made by the insertion of a cre expression cassette into the X-linked Hprt gene by gene targeting. When males carrying a floxed sequence in a reporter transgenic line (Soriano, 1999) are mated to hemizygous cre/0 females, excision of the sequence occurs regardless of cre inheritance and without mosaicism (J. M., unpublished). Positive mice were identified by PCR: a 0.24 kb product spanning the one remaining loxP site after neo cassette excision was amplified using primers 5′-GCCCACCAGCTGCTAGCCATC-3′ (upper) and 5′-CCTAGAGAATTCGAGGGACCTAATAAC-3′ (lower). Two clones were used to make chimeras, both giving germline transmission. Mice derived from both clones had the same phenotype, and mice derived from one clone were used in this study. Full designation of the targeted mutation described here is H19.ICR^{tm1(ChβGI)MdCoh}.

To produce all fetuses analyzed, one set of parents were females and males carrying the (ChβGI)₂ and were born from chimeras mated to cre/0 females. These mice therefore lacked the neo cassette and were heterozygous for the mutation. Females were not typed for the presence of X-linked cre, while males were typed as negative for cre. The other set of parents, unless stated otherwise, were homozygous for the Mus musculus castaneus form of distal chromosome 7, as derived from strain CAST/Ei (CS). More specifically, these mice were of strain FVB/NJ.CS(N3)-distalChr.7^CS/CS. The use of this cross allowed for allele-specific analysis of expression and methylation. Hereafter, heterozygous fetuses maternally and paternally inheriting the (ChβGI)₂ are designated –(M)/+ and +/–(P), respectively.

Northern blots were performed as described (McLaughlin et al., 1996; Szabó and Mann, 1994). Allele-specific expression was determined using RT-PCR single nucleotide primer extension (SNuPE) assays as described (Szabó and Mann, 1995), except that a different H19 SNuPE primer was used (5′-TGAATGTATACAGCGAGTGTG-3′), incorporating the radionucleotide C for the B6-type allele and T for the CS-type allele. Each assay relies on a single known sequence difference between allelic RNAs, as provided by CS mice. Assays were conducted on the same samples used in northern blots.

Southern blots for methylation analysis were performed as described (Szabó and Mann, 1994). Female mice of the transgenic line TgOG2 were mated to +/–(P) males, and from the resulting 18.5 days post coitum (dpc) female and male fetuses of small size – all presumptive +/–(P) fetuses – the gonads were removed, trypsinized and triturated, and cells positive for enhanced green fluorescent protein (EGFP⁺) were sorted using a MoFlo flow cytometer (Cytomation, Fort Collins, CO). TgOG2 mice, of mixed C57BL/6J and CBA/CaJ strain background are homozygous for a transgene in which the egfp reporter (Clontech, Palo Alto, CA) is driven by the Pou5f1 promoter. This line expresses EGFP specifically in the germline (J. M., unpublished). Two other transgenic lines made with the same construct have been reported (Anderson et al., 1999; Yoshimizu et al., 1999). Bisulfite sequencing was performed as described (Reed et al., 2001), except that bisulfite conversion was carried out at 60°C for 6 hours. In each PCR reaction, the DNA from 2000 cells was amplified using primers 5′-CAACCCCCCCCCAAAACCCCCAAAAATA-3′ (upper) and 5′-GGGTTGTTGGTTGATGATTTTGTATATAGT-3′ (lower). Two PCRs were performed on DNA from each of two conversion reactions, the products subcloned and individual clones sequenced.

The nucleotide structure of the 2.4 kb ICR in respect to CTCF binding sites, CpGs and GpCs is compared with the 2.4 kb (ChβGI)₂ (Fig. 1B). For the latter, a high density of CpGs is present in the vicinity of the CTCF-binding sites (Chung et al., 1997); therefore, there is potential for methylation of these regions. Comparison of the two 2.4 kb sequences gave 38% identity using ‘Gap’ global alignment (Seqweb Version 1.2, GCG Wisconsin Package version 10.1; Accelrys, San Diego, CA). They therefore lack homology – comparison of the (ChβGI)₂ with three other 2.4 kb regions selected at random from a 30 kb region downstream of H19 (sequences from GenBank Accession Number, AF049091) gave identities of 37-40%.

Substituting the (ChβGI)₂ for the ICR had little or no effect on the normal allele specific expression pattern of Igf2 and H19. Both genes were expressed monoallelically in the liver and kidney of 18.5 dpc –(M)/+ fetuses. Indeed, the low amount of paternal H19 RNA in the liver of +/+ fetuses was not present in –(M)/+ fetuses (Fig. 3A). Thus, the maternal (ChβGI)₂ successfully substituted for the function of the maternal ICR as a chromatin insulator. Insulator activity is dependent on the lack of methylation. Therefore, as expected, the maternal (ChβGI)₂ was hypomethylated in somatic tissues of perinatal –(M)/+ mice (see Fig. 4A). In addition, CpGs within and adjacent the CTCF binding sites of the (ChβGI)₂ in germ cells of 18.5 dpc female +/–(P) fetuses were hypomethylated (Fig. 4B).

Northern blots showed that the total amount of Igf2 RNA in the liver and kidney of –(M)/+ fetuses was ∼1.3 times that in +/+ fetuses (Fig. 3C). This is suggestive of some degree of hyperactivation of the paternal Igf2 allele in trans to the (ChβGI)₂, as RT-PCR SNuPE assays conducted on the same samples showed that all Igf2 RNA was derived from this allele (Fig. 3A). In any event, any increase in total Igf2 RNA did not result in an increase in weight of –(M)/+ fetuses (Table 1).

18.5 dpc +/–(P) fetuses were 50% or 61% of the weight of +/+ siblings, depending on the mouse strain used, and were of normal proportions and viable (Fig. 3B; Table 1). In addition, small size persisted into adulthood (data not shown). This phenotype was similar to that seen for paternal inheritance of a targeted null mutation of Igf2 – perinatal mutants was 62% of the weight of wild-type siblings, of normal proportions and viable (Baker et al., 1993). This effect indicated that the paternal (ChβGI)₂ was inhibiting activity of the paternal Igf2 promoter. As expected, the amount of Igf2 mRNA in perinatal and midgestation +/–(P) embryos was found to be very low (Fig. 3C,D). Thus, the paternally inherited (ChβGI)₂, rather than reiterating the behavior of the paternal ICR, functioned as a chromatin insulator, as does the maternal ICR. Consistent with insulator activity of the paternal (ChβGI)₂ was its hypomethylated state in somatic tissues of +/–(P) mice (Fig. 4A). In addition, the (ChβGI)₂ was hypomethylated in male germ cells of perinatal +/–(P) mice – only 11% of CpGs analyzed were methylated (Fig. 4B). In comparison, the paternal ICR is hypermethylated in male germ cells of perinatal mice (Davis et al., 1999; Ueda et al., 2000).

In the liver and kidney of +/–(P) fetuses, H19 was biallelically expressed (Fig. 3A), demonstrating that the paternal (ChβGI)₂ was unable to induce post-fertilization inactivation of the H19 promoter in cis – as does the paternal ICR. Interestingly, in both tissues, the normally silent paternal H19 allele had been activated to a level greater than that of the maternal allele, accounting for ∼75% of the total H19 RNA (Fig. 3A). This may represent a level of hyperactivation of the paternal H19 promoter at the perinatal stage: total H19 RNA measured in liver and kidney was 2.8 and 2.1 times the normal level, respectively (Fig. 3C); thus, paternal H19 RNA was 2.1 and 1.6 times the normal maternal level, respectively. However it was noted that in 11.5 dpc +/–(P) embryos, total H19 RNA was only 1.7 times that of normal levels (Fig. 3D).

In this study, we substituted the Igf2/H19 ICR for the (ChβGI)₂ by gene targeting and examined the effects in mice. The former is an element with properties of chromatin insulation, promoter silencing and imprinting (differential methylation), while the latter is a duplication of a chromatin insulator element, which, in its native state, lies at the 5′ end the chicken β-globin locus. The ICR and (ChβGI)₂ are of the same size, possess CTCF insulator binding sites and CpGs, but otherwise lack sequence homology. The aim of this study was to determine if the (ChβGI)₂ could duplicate one or more of the three properties of the ICR. The results should shed light on the mechanisms behind these properties – in particular, the potential role of CTCF binding (see Fig. 5).

Much evidence has accumulated that the maternal ICR mediates silence of the Igf2 allele in cis through chromatin insulator activity: deletion of part or all of the maternal ICR leads to expression of the maternal Igf2 allele, and therefore increased concentration of Igf2 RNA and large mice (Leighton et al., 1995a; Srivastava et al., 2000; Thorvaldsen et al., 1998). In addition, the maternal ICR binds the vertebrate insulator protein CTCF in vivo (Kanduri et al., 2000b; Szabó et al., 2000), and the ICR displays insulator activity in enhancer blocking assays in cell culture and transgenic mice (Bell and Felsenfeld, 2000; Hark et al., 2000; Kanduri et al., 2000a; Srivastava et al., 2000). In the present study, when the ICR was replaced by the (ChβGI)₂, the normal state of maternal Igf2 inactivity was retained. As the ChβGI has previously been characterized as an insulator, this finding lends even further support for the idea that maternal Igf2 inactivity is brought about through insulator activity in cis.

The insulator activity of the maternal (ChβGI)₂ was at least as strong as that of the maternal ICR. This was seen in the complete inactivity of maternal Igf2 in cis in the tissues examined, and in the normal size of –(M)/+ fetuses. This suggests that the ICR is not uniquely specialized for chromatin insulation at the Igf2/H19 domain, and that other insulators, even from a different class of vertebrate (Aves), can perform equally well in this role. This result is consistent with a high degree of conservation of function of CTCF-mediated chromatin insulation, in keeping with the high level of conservation in vertebrates of the insulator binding site core and CTCF amino acid sequence (Ohlsson et al., 2001). Furthermore, the present results, in which full insulation was obtained with two ChβGI CTCF-binding sites, are consistent with previous studies that suggest only two CTCF sites are required for effective insulation at the Igf2/H19 domain: deletion of two of the four CTCF-binding sites (deletion a, Fig. 1C) had no affect on insulator activity (Drewell et al., 2000), while deletion of three (deletion b, Fig. 1C) or all (deletion c, Fig. 1C) of the four CTCF sites eliminated much, if not all, insulator activity (Srivastava et al., 2000; Thorvaldsen et al., 1998). Thus, although the CTCF insulator binding site core of the ChβGI is slightly different from those in the ICR, and differs, at least in vitro, in the binding pattern of zinc fingers relative to the ICR sites (Ohlsson et al., 2001), it is apparently equally effective in mediating insulation.

The paternal ICR is methylated, does not bind CTCF and lacks insulator activity. By contrast, the paternal (ChβGI)₂ was hypomethylated and also functioned as a chromatin insulator. This was seen in the substantial reduction of total Igf2 activity in embryos paternally inheriting the (ChβGI)₂. Indeed, Igf2 expression was so low in +/–(P) 18.5 dpc fetuses that they were as small as Igf2 null mutants. This result is consistent with the idea that paternally inherited methylation inactivates potential insulator function of the paternal ICR.

The paternal ICR, while possessing no insulator activity, establishes silence of the H19 promoter in cis during early embryogenesis. The paternal (ChβGI)₂ clearly lacked this property. Indeed, the paternal H19 promoter was even more active than the maternal in the liver and kidney of –(M)/+ 18.5 dpc fetuses, as shown by the analysis of allele-specific expression, and this may represent some degree of hyperactivation of the paternal H19 promoter. The ChβGI, in its native state, is a center for hyperacetylation, although it is not clear if this property is associated with insulator function or CTCF binding (Litt et al., 2001). If the ChβGI retained such a property in transgenic mice, then it might further potentiate the activity of nearby promoters in cis. Nevertheless, it is noted that the maternal (ChβGI)₂ had no detectable effect on H19 expression: in tissues of –(M)/+ and +/+ fetuses, the amounts of H19 RNA were the same, and in both cases all H19 RNA was derived from the maternal allele.

The ICR becomes differentially methylated during female and male germ cell development (acquisition or establishment of the imprint); these epigenetic states are inherited and retained by the maternal and paternal copies during embryonic development (maintenance of the imprint) and further epigenetic change in cis is induced (development of the imprint). The (ChβGI)₂ was unable to duplicate this function of the ICR. It was hypomethylated in female and male germ cells – at the stage by which differential ICR methylation has taken place – and remained as such on the maternal and paternal chromosomes during embryonic development. Furthermore, the expression patterns of Igf2 and H19 on the maternal and paternal mutant chromosomes were similar. Thus, imprinting at the Igf2/H19 domain was lost.

These results clearly demonstrate that it is possible to separate the two functions of CTCF-mediated chromatin insulation and the establishment of differential methylation in the germline, i.e., these are not inseparable functions at the Igf2/H19 domain. Therefore, any potential for differential CTCF binding is not the primary difference between the two germlines, which determines their differential acquisition of ICR methylation. This conclusion assumes the ChβGI is not in some unknown way resistant to methylation. Furthermore, this conclusion does not preclude the possibility that differential CTCF binding may be the maintenance mechanism for differential methylation of the ICR in somatic cells, i.e. binding to the maternal ICR may be required for excluding methyltransferases from the region. It has been noted that only 42 bp of the ChβGI, which includes the CTCF site, retains insulator activity in enhancer blocking assays (Bell et al., 1999). Thus, as CTCF binding is inhibited by methylation, this 42 bp region must remain hypomethylated over the course of the assay. This result, together with the observation that the ChβGI possesses a very high CpG density and C+G content around the CTCF site, as seen in CpG islands, raises the possibility that CTCF binding itself may inhibit methylation of the surrounding region. CpG islands are coincident with promoters and their hypomethylated state might result from the exclusion of methyltransferases by bound transcription factors (Brandeis et al., 1994; Macleod et al., 1994).

What other conclusions can be drawn from these results regarding the mechanism of acquisition of differential ICR methylation? As previously pointed out by Jones and Takai (Jones and Takai, 2001), ‘two not-necessarily-conflicting models could account for the varying patterns (of de novo methylation), (i) an exclusion of access to methylation sites by proteins bound to specific DNA or (ii) a methylation-targeting mechanism steered by sequence-specific binding proteins’. Accordingly, differential methylation of the ICR could be set up by a female germ cell-specific protein that binds to the ICR and excludes methyltransferases, or alternatively by a male germ cell-specific protein that binds to the ICR and attracts methyltransferases. In addition, either event could be downstream of CTCF binding – in which case, mutagenesis of the binding sites would be expected to affect the process. Unfortunately, the results of the present ICR substitution do not shed light on either of these two possibilities: the lack of methylation of the (ChβGI)₂ in both germlines and in somatic cells could be accounted for by the possibility that its sequence is not attractive to methyltransferases or cannot bind proteins that can attract methyltransferases.

A third possible mechanism for the establishment of differential ICR methylation is that there a methyltransferase specific to the male germline exists that is attracted to the ICR sequence. A role for the recently discovered DNA methyltransferases IIIA and IIIB in imprinting has been postulated (Okano et al., 1999), and an investigation of their expression in female and male germ cell development would be of interest. Again, such a mechanism could be downstream of CTCF binding. It is of interest that in ICR deletion experiments, when two CTCF sites were left intact, differential methylation was still acquired (Drewell et al., 2000), yet when one was left, differential methylation in somatic cells was not observed (Thorvaldsen et al., 1998). Thus, it is conceivable that grouped CTCF-binding events could invoke a chromatin structure that is attractive to a male germ cell-specific methyltransferase. This possibility is consistent with the arrangement of the CTCF binding sites into two pairs in mouse and rat, and into two triplets in humans, where spacing between any two sites in a pair or triplet is no greater than 0.45 kb (Stadnick et al., 1999). In the present study, the required chromatin structure may not have been initiated in the (ChβGI)₂ because the two CTCF sites were too far apart, or because of sequence.

Further mutagenesis of the ICR in mice should lead to an identification of the sequences that mediate the acquisition of differential germline methylation and shed light on how this is maintained through somatic cell division.

We thank Hans Schöler for the Pou5f1-egfp construct, Adam Bell and Gary Felsenfeld for plasmid pJC13-1, and Lucy Brown, Claudio Spalla and Jim Bolen for flow cytometry. This work was supported by NIH grant 2R01GM48103-04 and NSF grant BIR-9220534.