Genes required to maintain pluripotency in human embryonic stem (hES) cells are largely unknown, with the exception of OCT-4, a homolog of mouse Oct-4, which is critical for the establishment of the embryonic inner cell mass and the generation of totipotent mouse embryonic stem (mES) cell lines. In the current study, we identified two genes with expression similar to OCT-4, in that they are largely restricted to pluripotent hES cells, premeiotic germ lineage cells, and testicular germ cell tumor cells. Furthermore, we determined that upon hES cell differentiation, their expression is downregulated. The genes we identified in the current study include the human stella-related (STELLAR) gene, which encodes a highly divergent protein (with just 32.1% identity to mouse stella over the 159 amino acid sequence) that maps to human chromosome 12p13. Notably, human STELLAR is located distal to a previously uncharacterized homeobox gene, which is the human homolog of the recently identified murine gene, Nanog, and proximal to the GDF3 locus, whose transcription is restricted to germ cell tumor cells. Our characterization of STELLAR, NANOG, and GDF3 suggests that they may play a similar role in humans as in mice, in spite of their remarkable evolutionary divergence.
Little is known of the determinants of pluripotency in human cells. However, one of the best characterized genes is the POU-domain class-5 transcription factor, POU5f1 (OCT-4). OCT-4 is a classic marker of pluripotency in human embryonic stem (hES) cells; expression has been shown to coordinately decrease with differentiation in vitro into embryoid bodies (EBs) and in vivo with the formation of teratocarcinomas . Additional pluripotency factors have not yet been identified in humans, though several candidate factors that function in murine ES (mES) cells, such as Fgf4, Foxd3, and the recently identified Nanog genes, have been reported [2–5]. Although we might assume that a program similar to that in mES cells is required for maintenance of hES cell pluripotency and self-renewal, there are some obvious differences between the two mammalian systems. For example, it is well known that the critical leukemia inhibitory factor-signaling pathways are required for undifferentiated self-renewal of mES cells under normal physiological conditions , yet they are not sufficient for the same function in hES cells [1, 7, 8]. As a result, novel approaches will be necessary in order to identify the candidate genes associated with the maintenance of hES cell pluripotency.
A common biological link among hES cells, the epiblast, and premeiotic germ cells is pluripotency, or the ability of these cell types to contribute to multiple embryonic lineages [1, 9]. Interestingly, Oct-4 is enriched in each of these three cell populations in mice; with respect to germ cells, expression is reduced as meiotic differentiation is initiated . Thus, we hypothesized that genes with restricted expression patterns similar to OCT-4 also may be critical regulators of pluripotency in humans.
Examination of Unigene EST databases for tissues that express human OCT-4 revealed that it is largely expressed in human germ cell tumors. The elevated expression may reflect an increase in the germ cell (OCT-4+) to somatic cell (OCT-4−) ratio in these tumors. Based on these observations, we sought to identify novel genes that are expressed exclusively in pluripotent cell types by taking advantage of an observation that a consistent structural chromosomal abnormality associated with the formation and/or overproliferation of germ cell tumors in men is the formation of 12p isochromosomes. These isochromosomes may overexpress 12p genes that are associated with excessive growth of undifferentiated germ lineage cells [11, 12]. As a result, we chose to identify novel genes that may be associated with the molecular regulation of pluripotency by focusing on human chromosome 12p. The region of the mouse genome syntenic to human 12p is the distal end of mouse chromosome 6.
A search for loci present on the mouse chromosome 6 identified stella . stella encodes a putative DNA-binding protein that is expressed differentially between nascent germ cells and their somatic neighbors at E7.25 in mice. As development progresses, stella expression is restricted to germ line stem cells. Given that mouse stella passed our criteria of having Oct-4-like expression and mapped to a chromosomal location syntenic to human chromosome 12p, we explored this genomic locus in more detail.
Materials and Methods
The genomic structure of mouse stella was obtained by Basic Local Alignment Sequence Tool (BLAST) analysis of the National Center for Biotechnology Information (NCBI) High Throughput Genomic Sequences (HTGS) database with the stella cDNA sequence (AY082485) and confirmed by analysis of the homologous mouse BAC (RP23117I23) using ENSEMBL (http://www.ensembl.org). The mouse stella locus, 2410075G02Rik (NM_139218), is also known as PGC7 (AB072734) and Dppa3 (NM_139218, G73534, AF490347). The syntenic human stella locus was identified by BLAST analysis of mouse stella in the human ENSEMBL database.
We identified a predicted gene with homology to mouse stella (ENS0000034235) on human chromosome 12p. Full-length human STELLA cDNA was obtained by reverse-transcription polymerase chain reaction (RT-PCR) from testis and ovary cDNA. Genomic structure of the predicted STELLA-related (STELLAR) gene was determined by BLAST analysis in ENSEMBL. Identification of the 3′ end of STELLAR was obtained by 3′ rapid amplification of cDNA ends (RACE) from testis cDNA using the SMART/CDSIII cDNA library construction kit (Clontech Inc.; Palo Alto, CA; http://www.clontech.com/index.shtml). BLAST analysis of the human STELLAR nucleotide sequence in the NCBI database also identified predicted genomic pseudogenes (intronless genes) on human chromosome 12, 14, and X; there were no corresponding mouse pseudogenes identified to any of these syntenic locations in the mouse genome.
Information regarding the hES cell lines HSF-6, HSF-1, and H9 can be obtained at http://stemcells.nih.gov/stemcell. The ES lines, HSF-6 and HSF-1, were obtained from the University of California, San Francisco (UCSF; National Institutes of Health [NIH] code UCO6 and UC01, respectively). H9 (NIH code W-9) was obtained from the University of Wisconsin . Undifferentiated hES cell colonies were cultured on irradiated CF1 mouse embryonic fibroblast feeder cells at 5% CO2 in medium containing knock-out Dulbecco's-modified Eagle's medium (DMEM)-high glucose, 20% KnockOut Serum Replacer, 1 mM glutamine, nonessential amino acids (all from GIBCO BRL; Carlsbad, CA; http://www.invitrogen.com), 0.1 mM β-mercaptoethanol (Sigma; St. Louis, MO; http://www.sigmaaldrich.com), and 4 ng/μl fibroblast growth factor-2 (R&D Systems; Minneapolis, MN; http://www.rndsystems.com). To differentiate hES cells into embryoid bodies (EBs), cells were cultured at 5% CO2 in medium containing knockout DMEM-high glucose, 20% fetal calf serum, 1 mM glutamine, and 0.1 mM β-mercaptoethanol. At days 0, 3, 7, and 14, EBs were collected, centrifuged, and resuspended in 600 ml RLT buffer (Qiagen; Valencia, CA; http://www.qiagen.com).
RNA and cDNA Production
Adult human testis mRNA was from Clontech Inc. Adult female ovary mRNA was from Ambion (Austin, TX; http://www.ambion.com). cDNA was generated from mRNA using 250 ng of random hexamers under standard conditions with murine leukemia virus RT (Promega; Madison, WI; http://www.promega.com). Seminoma samples were obtained from the UCSF Cancer Center Tissue Registry; testicular biopsies and oocytes were from the UCSF Center for Reproductive Health.
All human samples were obtained after Institutional Review Board approval. Total RNA isolated from all tissues and cells, with the exception of oocytes, was extracted via the RNeasy system according to instructions (Qiagen). cDNA was generated via random priming as above. Total RNA from oocytes was extracted using the PicoPure RNA isolation system (Arcturus; Mountain View, CA; http://www.arctur.com) followed by cDNA production with 250 ng/μl random hexamers (Promega) as above. Following extraction from human oocytes, cDNA was concentrated using DNA Clean and Concentrator (Zymo Research; Orange, CA; http://www.zymoresearch.com). cDNA from pooled human fetal ovaries at 20–29 weeks of gestation was obtained from Spring Bioscience (Fremont, AZ; http://www.springbio.com), and the fetal cDNA panel was from Clontech. PCR was performed with 50 ng or 100 ng of first-strand cDNA reaction as specified.
Polymerase Chain Reaction
Polymerase chain reactions contained 3 mM MgCl2; 10 mM each of dATP, dGTP, dCTP, and dTTP; 2 μM primers; and 0.25 U platinum Taq (Invitrogen; Carlsbad, CA; http://www.invitrogen.com). The PCR reaction was initiated by hot start at 95°C for 5 minutes followed by 35 cycles at 95°C for 30 seconds, 60°C for 30 seconds, and 72°C for 30 seconds. Real-time PCR for human STELLAR, GDF3, NANOG, SYCP1, and OCT-4 was performed as above except for addition of 1× SYBR Green (Molecular Probes; Eugene, OR; http://www.probes.com), 1× fluorescein (Bio-Rad; Hercules, CA; http://www.bio-rad.com), and 2% dimethylsulfoxide. All reactions for real-time PCR were performed in the presence of 4.5 mM MgCl2 and analyzed using an iCycler iQ (Bio-Rad), calibrated for use with SYBR green. Equal reaction efficiencies of all five genes were verified, and mean normalized expression was calculated using Relative Expression Software Tool (REST XL) .
All experiments included negative controls with no cDNA, and primers were designed to span exons to distinguish cDNA. Primers were: STELLAR primers: (5′-GTTA CTGGGCGGAGTTCGTA-3′)/(5′-TGAAGTGGCTTGG TGTCTTG-3′) (174 bp); GDF3 primers: (5′-AGACTTAT GCTACGTAAAGGAGCT-3′)/(5′-CTTTGATGGCAG ACAGGTTAAAGTA-3′) (150 bp); NANOG primers: (5′-CAGCTGTGTGTACTCAATGATAGATTT-3′)/(5′-CAACTGGCCGAAGAATAGCAATGGTGT-3′) (142 bp); Deleted in AZoospermia-Like (DAZL) primers: (5′-ATGTTA GGATGGATGAAACTGAGATTA-3′)/(5′-CCATGG AAATTTATCTGTGATTCTACT-3′) (178 bp); GAPDH primers: (5′-ACCACAGTCCATGCCATCAC-3′)/(5′-TCCA CCACCCTGTTGCTGTA-3′) (500 bp); OCT-4 primers: (5′-ACATCAAAGCTCTGCAGAAAGAACT-3′)/(5′-CTGAATACCTTCCCAAATAGAACCC-3′) (133 bp); NCAM-1 primers: (5′-ATGGAAACTCTATTAAAGTGA ACCTGA-3′)/(5′-TAGACCTCATACTCAGCATTCC AGT-3′) (187 bp); and SYCP1 primers: (5′-AAGATTTAC AGATAGCAACAAACACA-3′)/(5′-AATCTTTGCTGT TCTGTTCTCAATAA-3′) (169 bp).
Purified STELLAR (nucleotides 362-444; AY230136) and NANOG (nucleotides, 115-413; AY230262) PCR products were labeled via random priming per the manufacturer's instructions (Roche; Indianapolis, IN; http://www.roche.com) and incorporation of freshly labeled α32P dCTP. Human Northern membranes and hybridization solution were purchased from Clontech for analysis of STELLAR and from RNWAY Laboratories Inc. (Seoul, Korea; http://www.rnway.com) for analysis of NANOG. Membranes were hybridized and washed as previously outlined  and exposed to XOMAT film at −80°C for 10 days. The membranes were stripped then reprobed with labeled GAPDH probe to ensure equal loading of samples.
Pair-wise comparisons were made using analysis of variance, followed by two-tailed t-tests. Significance was assumed at p < 0.05.
Characterization of Syntenic Human STELLAR and Surrounding Genomic Sequences
We localized the mouse genomic locus of stella to BAC RP23-180N22 (house mouse) and RP23-117I23 (C57BL/6J) by BLAST analysis of the NCBI HTGS database. All predicted and known sequences homologous with BAC RP23-117I23 and RP23-180N22 were then identified using ENSEMBL. From this analysis, a RIKEN deposited sequence (2410075G02Rik [NM_139218]) corresponding to the gene PGC7  was also identified. Using synteny maps in the Mouse Genome Informatics and Mouse Genome Resource databases, together with the location of the neighboring MIT marker D6MIT151 as an anchor, we identified the region of the short arm of human chromosome 12p, which was syntenic to the mouse stella locus (mouse chromosome 6 at 59.3cM). Analysis of the human map at this site (BAC RP11-277J24) revealed that the homologous region contained a predicted gene that had no previous experimental evidence for transcription. Analysis of this gene revealed that, like mouse stella (Fig. 1A), the predicted syntenic human gene (which we called STELLAR) contains four exons of length similar to the mouse gene (Fig. 1A). Further sequence analysis of BAC RP11-277J24 revealed a novel locus 72.5 kb distal to STELLAR called FLJ12581 fis (AK022643; Unigene Hs.329296) (Fig. 1B).
We originally named this novel human locus Neighbor Of STELLAR (NOSTEL); however, recent work has identified this novel locus as the human homolog of the mouse nanog gene . The proximal end of BAC RP11-277J24 hybridizes with a second BAC called RP11-44J21, which contains the growth and differentiation factor 3 (Gdf3) locus (Fig. 1B). By forming a contig between the two BACs, it can be determined that human STELLAR is 15.6 kb distal to GDF3 (Fig. 1B). In the mouse, Gdf3, stella, and nanog are represented in BAC RP23-117I23. The current ENSEMBL order is shown in Figure 1B; however, as more sequence information becomes available, the distances between STELLAR, NANOG, and GDF3 in the mouse genome will potentially change. Currently, stella is placed 71.4 kb distal to Gdf3 and 53.7 kb proximal to nanog. In the human genome, NANOG is annotated as a four-exon gene comparable to the mouse chromosome 6 homolog annotated in the ENSEMBL database (Fig. 1C).
To obtain the full-length cDNA sequence of human STELLAR and NANOG, we performed PCR amplification of human testis and hES cell (HSF-6) cDNA and compared the sequences of the cloned fragments with their respective annotated genomic loci. The accession numbers of cloned STELLAR and NANOG obtained in the current analysis are AY230136 and AY230262, respectively.
STELLAR and NANOG Expression in Adult and Embryonic Tissues
Northern blot analysis was used to examine expression of NANOG and STELLAR in adult human tissues (Fig. 2A and Fig. 2B). Expression of GDF3 in adult human tissues was previously reported . Expression of both STELLAR and NANOG in all human adult tissues examined was extremely low (Fig. 2A and 2B). In both cases, a minimum of 10 days of exposure was required in order to observe the weak signals. STELLAR probe hybridized to a 1.1-kb band in ovary, testis, and thymus consistent with the length of the cDNA transcript identified from human testis (Fig. 2A). In addition, a larger 4.4-kb STELLAR band of unknown origin was also detected in testis and ovary; 3′ RACE failed to identify this STELLAR transcript, although it may represent an alternatively spliced variant or novel 5′ structure not identified in the current analysis. NANOG was detected as a faint 2.2-kb band exclusively in the testis (Fig. 2B). Given this extremely low expression level, we used a more sensitive RT-PCR approach to examine expression of these three loci, and OCT-4, in human fetal tissues (Fig. 2C). We found that OCT-4, STELLAR, and NANOG were expressed only in fetal ovary. All other fetal tissues had basal levels of transcription comparable to OCT-4 and lacked detectable expression. By comparison, GDF3 was enriched in fetal ovaries; however, expression was also detected in fetal kidney, lung, skeletal muscle, and thymus.
Given that GDF3 was previously cloned from human teratocarcinoma libraries  and human chromosome 12p is an apparent hotspot for chromosomal abnormalities associated with teratocarcinoma, we used real-time PCR analysis to compare expression of STELLAR and NANOG in normal human testis with that in two independently isolated testicular germ cell tumors (seminomas) (Fig. 3). We determined that transcription of all three genes was greater in germ cell tumors than in normal adult testis. In particular, STELLAR was greater by approximately seven- and fourfold, GDF3 was greater by 12- and fourfold, and NANOG was greater by four- and threefold, respectively, in the two seminomas (Fig. 3). Furthermore, OCT-4, which in silico can be identified largely in germ cell tumor libraries, was greater by approximately ninefold compared with normal testis.
Amino Acid Identity of Human STELLAR
Translation of human STELLAR revealed a highly basic protein (pI = 8.3) of 159 amino acids, with a predicted mass of 17.9 kD (Fig. 4), similar to mouse stella (150 amino acids; 18 kD) . Analysis of the human STELLAR peptide sequence using the pSORT prediction program suggests that human STELLAR, like mouse STELLA, has no known functional motifs. Furthermore, amino acid analysis predicted that, like mouse STELLA, the human syntenic STELLAR locus also encodes a nuclear protein. However, given these similarities, the degree of amino acid identity between the mouse and human proteins is remarkably low, with just 32.1% identity over the protein sequences. Comparisons of NANOG and GDF3 proteins with their respective mouse proteins have been described [5, 17] and also demonstrate low identity between these homologs (Table 1).
Table Table 1.. Germ cell and ES cell restricted loci at human chromosome 12p
Percentage homology (nt)
Percentage identity (peptide)
Human STELLAR and NANOG Are Enriched in Pluripotent Cell Populations
Given the expression of the STELLAR, NANOG, and GDF3 genes, albeit at low levels in normal adult human testis and ovary, we sought to further characterize their expression in somatic cells and/or germ cells of the gonads. We compared expression of human STELLAR, NANOG, and GDF3 in adult ovary, testis, and fetal ovary at 20–29 weeks of gestation. We also further defined expression in tissues with and without germ cells (Fig. 5A–5CFigure 5.). Our results show that all three genes and the positive control, DAZL, are expressed in the adult ovary; however, expression of NANOG in adult ovary was much lower than that of STELLAR or GDF3 (Fig. 5A). Similarly, GDF3 expression in the adult testis was lower than that of STELLAR and NANOG. RT-PCR on mRNA extracted from tissues of men who lack all germ cells (and were diagnosed with Sertoli cell only syndrome [SCOS]) was used to determine if STELLAR, GDF3, and NANOG are germ cell specific in the human testis (Fig. 5B).
Our results indicate that, in the absence of DAZL (in biopsies that lack germ cells), STELLAR, GDF3, and NANOG are also absent (Fig. 5B). Finally, we performed RT-PCR on ovulated, unfertilized, mature oocytes from women (Fig. 5C). We found that STELLAR was expressed in isolated oocytes together with the positive control, DAZL [18, 19], whereas, the somatic marker, neural cell adhesion molecule-1 (NCAM-1), was not expressed. Notably, unlike STELLAR, expression of both GDF3 and NANOG were not detected in ovulated oocytes.
Human STELLAR, NANOG, and GDF3 Expression Decreases in hES Cells with Differentiation
Given the similarities of expression of the STELLAR, GDF3, and NANOG genes to those of OCT-4 in germ cells and germ cell tumors, we compared the relative levels of STELLAR, GDF3, and NANOG in three undifferentiated hES cell lines (HSF-6, HSF-1, and H9; Fig. 6). We found that all three genes were expressed in undifferentiated ES cells. NANOG was expressed at significantly higher levels than STELLAR and GDF3 in HSF-1 (p = 0.023), H9 (p = 0.004), and HSF-6 (p = 0.002); there was no statistically significant difference in expression between STELLAR and GDF3 in any of the three cell lines. In particular, NANOG was expressed at approximately 10-fold higher levels than GDF3 and threefold higher levels than STELLAR in the hES cell line HSF-1 (Fig. 6B). Similar expression was observed in the H9 and HSF-6 cell lines.
In order to ensure that the hES cells in the current study were not expressing differentiated germ cell markers, we examined expression of the SYnaptonemal Complex 1 Protein (SYCP1), which is expressed only during meiosis I; it was not expressed at detectable levels (Fig. 6A). Finally, we examined expression of STELLAR, GDF3, and NANOG in differentiating hES cells in suspension culture at days 0, 3, 7, and 14 of EB formation (Fig. 7). Expression of OCT-4, NANOG, GDF3, and STELLAR were all decreased as differentiation progressed. In particular, OCT-4 was significantly decreased by day 7 (p = 0.003), NANOG was significantly decreased by day 14 (p = 0.02), STELLAR was significantly decreased by day 3 (p = 0.005), and GDF3 was significantly decreased by day 14 (p = 0.009). Although there was an initial apparent rise in GDF3 expression on day 3 compared with day 0, this was not statistically significant. In contrast, the somatic lineage markers NCAM-1 and α fetoprotein (AFP) were significantly (p = 0.009 and p < 0.001, respectively) increased with differentiation (Fig. 7).
The restricted expression pattern of Oct-4 during mouse embryo development, together with its critical role in maintaining mES cell self-renewal and pluripotency, prompted us to search for other loci in humans that share similar expression patterns to mouse Oct-4. Since an in silico search for OCT-4 in human cDNA databases suggested that, in humans, transcripts with similar expression to OCT-4 would be restricted to human germ cell tumors, we sought to identify additional loci with restricted human germ cell tumor expression.
We focused our search on novel loci on human chromosome 12p, which is a known chromosomal hotspot for structural chromosomal changes associated with formation of germ cell tumors. We identified the human STELLAR gene by synteny mapping to the mouse stella locus on mouse chromosome 6 and characterized the neighboring human genes, NANOG and GDF3, with regard to expression in adult and fetal tissues, as well as expression in pluripotent hES cells. We determined that human STELLAR, NANOG, and GDF3 are enriched in organs containing pluripotent cells. In particular, we identified NANOG and STELLAR expression in fetal ovary with no detectable expression in any other fetal somatic tissue examined. By comparison, GDF3 was enriched in fetal ovary, with lower levels of expression in fetal kidney, lung, skeletal muscle, and thymus. In adult human tissues, we determined that STELLAR, NANOG, and GDF3 were localized to the ovary and testis, and expression of all three genes was upregulated in testicular germ cell tumor samples. In ovulated unfertilized oocytes, we found comparable levels of STELLAR and the germ cell-specific gene, DAZL, and no detectable expression of NANOG or GDF3. Analysis of three independently derived lines of undifferentiated pluripotent hES cells revealed that STELLAR, GDF3, and NANOG were consistently expressed in all three undifferentiated lines, and that expression decreased with differentiation.
This study using human tissues illustrates both the similarities and differences in gene expression of GDF3, STELLAR, and NANOG compared with previous reports from mice. In mice, stella localizes to all cells during the early stages of preimplantation development and is downregulated from E3.5 prior to inner cell mass formation . Expression of stella resumes in the germ cells of the allantoic bud, where it is then found exclusively in migrating germ cells and gonadal primordial germ cells [13, 16]. In the adult ovary, mouse stella is then found in oocytes throughout folliculogenesis, as well as in ovulated unfertilized eggs . In mice, nanog localizes to all stages of preimplantation embryo development, from fertilized eggs to the inner cell mass of blastocyst [5, 20]. Like the current study in humans, mouse nanog is not found in unfertilized eggs. In addition, like mouse stella, mouse nanog is also expressed in primordial germ cells of the fetal gonad or genital ridge , and in the current study, we detected high levels of both NANOG and STELLAR in human fetal gonads. The difference between previously reported expression of mouse nanog and our findings is that we detected NANOG in adult testis and ovary in humans, whereas, in mice, there is no apparent expression in any adult tissue except embryonal carcinoma (EC) cells derived from testicular germ cell tumors .
Furthermore, in the current study, we detected human STELLAR in the adult testis and ovary, whereas, in mice, stella is found in mouse ovary only. These differences could be attributed to more sensitive PCR assays used here or to specific differences in expression. It is well documented that GDF3 is differentially expressed between mice and humans [17, 21]. In particular, in mice, Gdf3 is found in multiple adult somatic tissues, whereas, in humans, GDF3 has only been identified in EC cells and is downregulated with differentiation . The relative lack of amino acid identity at the 12p locus between mouse and humans is also interesting, particularly in light of the well-known phenomena that genes associated with sexual reproduction are more divergent than nonreproductive genes .
Although we currently do not know the functional roles of STELLAR, GDF3, and NANOG in gamete formation or differentiation, it is interesting that these three genes, which are germ cell enriched, share the property of amino acid divergence. OCT-4 has a higher amino acid identity (76%) to its mouse ortholog; however, this is still lower than the majority of human genes (50% of human genes have less than 10% amino acid divergence) . As a consequence, it will be interesting to compare and contrast the functions of these divergent proteins in both mice and humans. The use of mouse and human ES cells may provide a useful model with which to perform this comparative functional analysis. The molecular machinery necessary for establishing the germ cell lineage in humans is almost completely unknown, and the signals necessary for inhibiting a somatic cell fate or promoting a human pluripotent germ cell have yet to be discovered. Our results suggest that genes whose expression is enriched in germ cells or human germ cell tumors and that largely lack expression in somatic tissues could provide a source of novel genes to explore central issues of stem cell pluripotency and self-renewal and germ cell differentiation.
We thank the UCSF Fertility Clinic, in particular, Dr. Shehua Shen, for obtaining the human oocyte samples. We also thank Professor Robert N. Taylor, M.D., Ph.D., and Eugene Y. Xu for helpful comments on the manuscript. This work was supported by grants from the Lance Armstrong Foundation (P.J.T.), the Sandler Foundation, and the National Institute of Child Health and Human Development (Grant R01-HD37095) (R.A.R.P.).