DNA was extracted from placental tissue from 125 live births (83 normal birth weight, 42 low birth weight) at Baystate Medical Center (Springfield, MA) in a case-control study of the genetic predisposition to fetal growth restriction. All subjects were Caucasian, both Hispanic and non-Hispanic. Classification of newborns as fetal growth restricted (or IUGR, intrauterine growth retarded) followed the definition of the American College of Obstetricians and Gynecologists as those newborns below the 10^th percentile of size for gestational age 20. Our cut-off of 2,500 g at term (>37 weeks gestation) corresponds to the lowest 7.5 percentile of all US Caucasian deliveries, or the lowest 10^th percentile of female and lowest 6.5 percentile of male deliveries 21. Stringent inclusion/exclusion criteria (Table 1) were employed by a placental pathologist (TKB) to reduce nongenetic contributors to birth weight variation. Ethical approval to conduct this study was obtained from the Human Subjects Institutional Review Board of the University of Massachusetts.

The region from -624 (relative to the start of transcription; GenBank accession J03071) to +1,726 (197 nucleotides after the termination codon) of GH1 was amplified in two overlapping fragments: -624 to +541 (GHN-1F 5' AGGGACCTGGGGGAGCCCCAGCAA 3', GHN-1R 5' TCACCCCTTCCTGCCACCCCTGAT 3') and +450 to +1,726 (GHN-2F 5' CCATCGTCTGCACCAGCTGGCCTT 3'; GHN-2R 5' GCCCTACAGGTTGTCTTCCCAACT 3'). Approximately 50 ng of DNA was used as a template in a polymerase chain reaction with 30 cycles of 95°C (1 minute), 62°C (1 minute), and 72°C (2 minutes 30 seconds). PCR products were purified from agarose using a QIAquick PCR Purification kit (Qiagen) and sequenced directly with BigDye v2.0 chemistry (Applied Biosystems) and run on either an ABI Prism 377 or 3100 automated DNA sequencer. PCR products were sequenced with the PCR primers and additional internal primers: 5' AAGCACAGCCAATAGATTG 3', -459 to -441; 5' GCACAAGCCCGTCAGTGGCC 3', -108 to -89; 5' GGATTTTAGGGGCGCTTACC 3', +71 to +90; 5' CATCTCCCTGCTGCTCATC 3', +931 to+949; 5' GCGCTTGGGYACTGTTCCCT 3', +1280 to +1299.

Sequence traces were aligned and assembled into contigs by the program Polyphred 22. Contigs were viewed in either the program Consed 23 or Sequencher (Gene Codes Corp.) and polymorphisms confirmed visually. Twenty-six polymorphic sites were identified (Table 2), including all of the sites identified by Horan et al. 18 in 154 British military recruits with the exception of site -339 which had a minor allele (deletion of G) frequency of 3.6% in their study.

The five genes of the human growth hormone locus exhibit high sequence similarity, and the paralogous regions corresponding to the portion of GH1 sequenced in this study (-624 to +1,726) were multiply aligned. Nucleotide positions (Table 2) were designated as invariant if all five genes had the same nucleotide or the four paralogs of GH1 were identical and matched the major allele at that site in GH1. This categorization explicitly assumes that only the minor alleles in GH1 are the product of gene conversion and that minor alleles not observed in paralogs of GH1 are the result of unique mutations. The proportion of GH1 polymorphisms at invariant versus variant sites was compared by Fisher's Exact Test to determine if there was an over-representation of polymorphic sites among variant sites.

Logistic regression, using FGR status as the outcome, was performed on gestational age and genotypes at sites with a minor allele frequency above 5% (Table 3). Sites -301 and -308 (relative to the start of transcription) are in complete linkage disequilibrium and the minor C alleles at sites -1 and +3 are in complete linkage disequilibrium. Therefore, sites -301 and +3 were excluded from logistic regression to avoid multicollinearity. Based on the results from logistic regression, separate ANOVA (Table 4) was performed on gestational age and the AA versus AC genotypes at site -1 within low birth weight and within normal birth weight subjects. Empirical p values for the F statistic for the genotypic effect, corrected for multiple comparisons, were determined by 2,000 random permutations of the genotypic data 24. All statistical analyses were performed using the Stata program (Stata Corp., College Station, TX).

Among the 125 subjects sequenced from -624 to +1,726 of the GH1 gene, 26 polymorphic sites were identified (Table 2). These included all but one of the 15 sites characterized by Horan et al. (2003). In the region of overlap between the two studies, we failed to detect variation at site -339, where there is a minor allele deletion of a single nucleotide with frequency 3.6% in British army recruits, and identified two additional variants at sites -360 and -352 with minor allele frequencies 2.8% and 1.2%, respectively. Therefore, the discrepancies between the two studies in the identification of SNPs can most likely be ascribed to sampling error. Outside the region surveyed by Horan et al. (2003), we detected 10 additional polymorphisms. All of these had minor allele frequencies ≤ 3.2%, except an intron four polymorphism at site +1169 with a minor allele frequency of 33.1%. In general, polymorphisms in the promoter of GH1 tend to be more densely clustered and exhibit higher minor allele frequencies than in the transcribed region.

An alignment of the region of GH1 sequenced in this study with the paralogous sequences of placental growth hormone and chorionic somatomammotropins indicated that there are 1,979 invariant sites and 293 variant sites (excluding 78 sites that could not be unambiguously aligned), as defined in the methods. Of the invariant sites, 5 are polymorphic in GH1, while 21 of the variant sites are polymorphic. A comparison of the proportion of polymorphic sites at invariant and variant sites by Fisher's Exact Test is highly significant (P <<0.001). This result indicates that there is a strong bias for polymorphisms in GH1 to occur at the minority of sites that exhibit sequence divergence among the paralogous genes. The high correspondence between the sequence of minor alleles in GH1 and nucleotides present in paralogs of GH1 supports previous assertions that the unusually high polymorphism of the GH1 gene is driven by gene conversion 1618.

A proportion of this bias may be explained by selective constraint, in that sites that are polymorphic within GH1 may be under less selective constraint and thus more free to exhibit sequence divergence among paralogs. However, to entirely explain the bias towards polymorphism at sites of divergence, one must assume that about 1,804 (91%) of the invariant sites are selectively constrained and not free to vary. Given that a substantial proportion (814 bp, ~36%) of the sequence surveyed in this study is composed of introns, this assumption seems unreasonable.

Logistic regression (Table 3) was performed on gestational age and SNPs with minor allele frequencies greater than 5%, excluding sites -301 and +3 because they are in complete linkage disequilibrium with other sites, to identify associations with fetal growth restriction. Gestational age was significant because the FGR subjects exhibit a slightly younger average estimate of gestational age (38.1 vs. 39.1 weeks, t = 4.9, p < 0.0001; gestational ages rounded to nearest week). However, even accounting for the effect of gestational age, the C allele at site -1 was significantly associated with FGR in the combined regression. However, this allele did not retain significance in a regression on only gestational age (p < 0.001) and the A/C polymorphism at site -1 (p = 0.242).

Although the A/C polymorphism at site -1 was not significant in a reduced model, we decided to investigate this site for three reasons. First, the C allele at site -1 is in complete linkage disequilibrium with C at site +3. Second, both of these sites are located at the start of transcription, making them good candidates for affecting the level of transcription of GH1. Third, the only paralog that shares C at these sites is CS-5, a pseudogene, consistent with the possibility that C at sites -1 and +3 is disadvantageous. The C allele at both sites exhibits a much higher frequency in low birth weight (6%) versus normal birth weight (0.4%) subjects. Restricting examination only within the normal or low birth weight subjects, the AC genotype is associated with an average reduction of birth weight of 152 g and 97 g in normal and low birth weight subjects, respectively. However, this difference does not achieve significant in an ANOVA on the A/C polymorphism and gestational age (Table 4).