In the present study we presented a re-analysis of previously published data that further elucidated the relative roles of maternal and fetal genomes on spontaneous PTD. In the previous study that used overlapping data, the most significantly associated genes were COL1A2 and PTGER3 in the maternal and TFPI and PON1 in the fetal analyses. We confirmed in our analyses an association with PON1. However, TFPI, which was found in the previous study, was only significant in our hybrid analysis. It is likely that the original finding was due to population stratification, a factor minimized by the family based analyses we used.
Using our approach, the most consistent significant gene across all analyses was COL5A2, which is involved in the production of type V collagen. The previous analysis only provided minimal evidence for the association with this gene . COL5A1, which also contributes to the production of type V collagen, was also found to be significant in the single locus analysis when maternal effects were included, and several SNPs were significant when examining maternal and fetal alleles separately (Table 5 and 6). Type V collagen plays a critical role in early fibril initiation and in the determination of fibril structure and matrix organization . Defects in type V collagen due to mutations in COL5A1 and COL5A2 are the cause of the classical type (types I and II) of the heritable connective tissue disorder Ehler-Danlos syndrome that confers an increased risk for PTD if the fetus is affected, especially from pPROM [39–41]. It is therefore reasonable to hypothesize that variations in these genes might be involved in the pathophysiology leading to PTD. However, the results must be interpreted with care, as the QQ-plots shows that the observed p-values do not deviate from what would be expected by chance. Few other studies have tested the association between COL5A2 and the risk of spontaneous PTD. A recent study by Romero et al found an association between rs189683203 in fetal DNA and the risk of pPROM with an unadjusted p-value of 0.021 (odds ratio, (OR) = 1.42, CI: 1.06, 1.92) . In another study on the same study population the authors found an association between rs6750027 in maternal DNA and the risk of PTL with an unadjusted p-value of 0.043 (OR = 1.32, CI: 1.01, 1.74) . Another study by Velez et al found significant associations in three fetal SNPs and one maternal SNP in COL5A2 .
PON1 was significantly associated with PTD in both the triad analysis and the hybrid analysis of the fetal genes alone, but not in the combined analysis of maternal and fetal genes. This is most likely because the combined effects of maternal and fetal genes may not reach significance due to reduced power in this type of analysis. None of the maternal SNPs showed an association with spontaneous PTD, while several of the fetal SNPs did. PON1 was found to be significantly associated with PTD in the previously published case-control analysis on the same fetal samples as well, and possible mechanisms of how this gene might contribute to preterm delivery are discussed there .
TFPI which was the strongest fetal association in the case-control study , showed at best weak association in the triad-analysis, but was the most significant gene in the hybrid design. The most significant SNP in the case-control study, rs6434222 in TFPI, also had a significant unadjusted p-value in the triad analysis (p = 0.02).
Overall, the results from the case-parent-triad analysis and the previously performed case-control analysis only overlapped in a few genes. The results from the hybrid analysis lay somewhere in-between the results from the case-parent triad and the case-control study. These differences may indicate that within our study population there was population stratification that could have led to spurious results in the case-control analysis. This is minimized using triads. Although the hybrid analysis has more power than the triad analysis, it may still be affected by population stratification, but to a lesser degree than a case-control design. The case-parent triad analysis is therefore the most reliable in terms of reducing the problem of stratification, and we present these as our most compelling results.
Our study also identified several pathways as associating with PTD. Significant results were found in the Focal Adhesion, Cell Communication and ECM-receptor interaction pathways, all of which include COL5A2 and COL5A1, but none of the other associated genes in our study. The ECM-receptor interaction pathway is involved in tissue and organ morphogenesis associated with the bleeding disorders Bernard-Soulier syndrome and Glanzmann thrombasthenia. The Focal adhesion pathway is involved in cell matrix adhesion and also associated with the bleeding disorder Glanzmann thrombasthenia. For the Notch signaling (hsa04330), Gluthatione Metabolism (hsa00480) and Glyoxylate and dicarboxylate metabolism (hsa00630) pathways only one gene was available for inclusion and the results from these pathways must be interpreted with care. Because this was a candidate gene study, the number of included genes and SNPs in each pathway was limited. Nevertheless, those pathways that provide strong evidence of association can probably be taken as truly being involved in spontaneous PTD.
The major strength of this study was that we used the case-parent triad design and the hybrid design and compared these results to those of the traditional case-control design. Few other candidate gene studies on PTD have been performed using the case-triad design, which offers protection against bias due to population stratification. Additionally, we performed a hybrid analysis, which has increased statistical power over both the case-triad and the case-control designs. These designs also provide separate estimates of fetal and maternal alleles, as well as an overall p-value estimating the combined effect of maternal and fetal alleles. In this way, confounding through maternal alleles, which can affect the intrauterine environment and thus the phenotype of the fetus, can be avoided. Our study was limited in that the small sample size was small compared to modern GWAS level analyses and in that it was based on a limited number of candidate genes. Also, no covariates were included in the hybrid analysis and we were not able to separate spontaneous PTD into pPROM and PTL at the time of analysis. Another weakness of this study is that we did not have an external replication sample to corroborate our findings. The findings should thus be regarded as exploratory, although the prior plausibility of the genes provides increased confidence in our results.