Amniotic fluid is a complex biological material that provides a unique window into the developing human. It is sampled for the prenatal assessment of many conditions, most commonly to obtain amniocytes for chromosome analysis. The residual amniotic fluid supernatant, which is usually discarded, is a rich source of fetal cell–free DNA and RNA. Cell-free nucleic acids in amniotic fluid are distinct from those in maternal blood. First, amniotic fluid cell–free nucleic acids are more likely to originate from the fetus itself, in contrast to circulating cell-free fetal nucleic acids, which are predominantly of trophoblast origin.1,2 Second, amniotic fluid cell–free fetal nucleic acids are relatively uncontaminated by maternal nucleic acids, because maternal-fetal nucleic acid trafficking is overwhelmingly unidirectional from fetus to mother.3 Finally, cell-free fetal nucleic acids are 100- to 200-fold more abundant in amniotic fluid compared with maternal plasma, making downstream applications such as global gene expression profiling more feasible.4
The first aim of this study was to identify transcripts that are ubiquitously present in a euploid midtrimester amniotic fluid supernatant data set. We called this gene list the amniotic fluid core transcriptome. Functional analyses were then performed on this gene list to infer the major organ systems that contribute to amniotic fluid cell–free fetal RNA and to determine the pathways of biological significance in the midtrimester fetus. Our second aim was to identify the likely tissue sources of amniotic fluid cell–free fetal RNA by using the GNF Gene Expression Atlas to map organ-specific genes. However, several important fetal organs such as the heart, kidney, intestine, and skin are not represented in the GNF Gene Expression Atlas. Therefore, our third aim was to create gene lists associated with the development of these organs and to ascertain the presence of these genes in amniotic fluid supernatant. The unifying goal of these three aims was to produce clinically relevant information about second trimester gene expression in the living human fetus.
MATERIALS AND METHODS
This was an in silico (computational) investigation using gene expression data previously produced by our group from 12 midtrimester amniotic fluid samples. This sample size is within the range of biological replicates (10–15) recommended to achieve statistical stability and confidence in microarray experiments.5 The Tufts Medical Center institutional review board approved the collection and analysis of the samples. These data are publicly available at www.ncbi.nlm.nih.gov/geo/(GSE33168). Eleven of the 12 samples were originally used as euploid controls in studies focusing on the pathophysiology of trisomies 18 and 21.6,7 In the present study, we reanalyzed the data from these 11 cases together with a previously unpublished sample to better understand gene expression in euploid fetuses. The methods of amniotic fluid processing, RNA extraction, fragmentation, labeling, and hybridization have been previously described.7 In brief, RNA was extracted from 10 mL of amniotic fluid supernatant from 12 women undergoing fetal testing for clinical indications. Eleven of the women had testing for advanced maternal age (median maternal age, 36 years; range 23–40 years). Three of these women had abnormal second-trimester serum screening results. One fetus was a twin (chorionicity unknown). None of the fetuses had a known major ultrasonographically detectable anomaly at the time of amniocentesis. Amniotic fluid supernatant samples from six male and six female euploid fetuses (median gestational age, 18 weeks; range 16–21 weeks) were included. The final amplified cDNA products were hybridized to Affymetrix U133 Plus 2.0 microarrays.
Data were normalized using the three-step function from the affyPLM package in Bioconductor 2.8.18 with ideal-mismatch background-signal adjustment, quantile normalization, and the Tukey biweight summary method.9 This summary method includes a logarithmic transformation. To obtain detection calls consistent with those produced by Affymetrix' 5.0 software, we used the mas5calls function from the Bioconductor affy package.
We defined the amniotic fluid core transcriptome as the list of genes corresponding to those Affymetrix probe sets that were called “present” in all 12 of the amniotic fluid samples. A probe set was called “present” if it had a detection P value of <.04,10 consistent with other published research.11
The web-based software tool Ingenuity Pathway Analysis 9.0 (content version 3210) was used for the biological interpretation of the amniotic fluid core transcriptome gene list. This pathways-analysis tool uses a manually curated repository of biological interactions and functional annotations to identify the most significantly enriched signaling pathways and biological processes represented in a given gene set. The enriched pathways for the categories “Cellular and Molecular Functions,” “Physiological Systems Development and Function,” and “Canonical Pathways” were reported separately. The pathways-analysis software uses the right-tailed Fisher exact test to calculate a P value representing the probability that a biological function not really relevant to the amniotic fluid core transcriptome is reported as relevant. Pathways that contained at least one functional annotation with a P<.01 were considered statistically significant. In addition, we applied a multiple testing correction using the Benjamini-Hochberg approach, which bounds the false discovery rate. Here, we report all biological functions with Fisher P<.01 and their corresponding false discovery rates.
We identified organ-specific genes in the amniotic fluid core transcriptome using the Novartis Research Foundation Gene Expression Database (http://biogps.gnf.org). This publicly available atlas of the human protein-encoding transcriptome used Affymetrix Human Genome-U133A and GNF1H custom human arrays to map gene expression profiles in 78 normal human tissues.12 We chose this resource because of its coverage of normal adult and fetal tissues, high reproducibility, and good correlation between transcript levels and protein abundance.13 Samples from the brain, liver, and lung were pooled from spontaneously aborted fetuses (15–33 weeks of gestation); the placentas were collected at birth. No gestational age data were available for the fetal thyroid and placental samples.
The GNF Gene Expression Atlas allowed us to assess the gene expression patterns of individual Affymetrix probe sets in the amniotic fluid core transcriptome. Data were downloaded on September 2, 2011. We categorized probe sets as highly organ-specific if they mapped to a single organ with an expression value more than 30 multiples of the median and had no unrelated tissue expression greater than one third of the maximum expression level.
Because several important fetal organs were not represented in the GNF Atlas, we used a combination of manual literature searching and pathways-analysis software to generate lists of genes that we would expect to be expressed in the midtrimester fetal heart, kidney, bladder, skin, intestine, placenta, and amnion. We then examined the amniotic fluid core transcriptome for the presence of these genes as an indirect method of ascertaining potential organ contributions to amniotic fluid cell–free fetal RNA. A full description of the methods used in the literature and pathways-analysis searches can be found in Appendix 1 available online at http://links.lww.com/AOG/A271.
The average present call rate for all 54,675 gene probe sets contained in the data set was 19.8%. The total number of probe IDs that were expressed in 12 of 12 samples was 796 (1.46%). After excluding duplicate genes resulting from multiple probe sets, hypothetical genes, pseudogenes, nonprotein coding genes, and genes with unknown functions, the final number of well-annotated genes in the amniotic fluid core transcriptome was 476 (Appendix 2, http://links.lww.com/AOG/A271). The three most common molecule types within the amniotic fluid core transcriptome were enzymes, ribosomal proteins, and transcription regulators.
Functional analysis of the amniotic fluid core transcriptome identified six physiologic systems development and function pathways, 11 canonical pathways, and 27 cellular and molecular pathways with a Fisher P<.01 (Tables 1, 2, and 3). The most significantly overrepresented physiologic systems were skeletal and muscular system development and function, tissue development, and hematologic system development and function (false discovery rate <0.05). Individual functional annotations within these major categories include growth of muscle and developmental process of cardiac muscle. Biological processes associated with nervous system development and function were also overrepresented. The most significant functional annotation in the nervous system category was neurogenesis of stem cells (P=.03, false discovery rate=0.09). The mammalian target of rapamycin signaling pathway, an important central regulator of cell growth and nutrient sensing, was identified as a key canonical pathway. The most significant cellular and molecular pathway was protein synthesis.
After examining the tissue expression patterns of each of the genes in the amniotic fluid core transcriptome using the GNF Gene expression atlas, we identified 23 highly organ-specific genes associated with the brain, spinal cord, lung, pancreas, liver, tongue, thyroid, placenta, and blood (Table 4). Seven of these genes are highly expressed by the fetal and adult central nervous system (CNS). The plakophilin 4 gene (PKP4) mapped to the adult spinal cord but not the fetal brain. The fetal spinal cord was not represented in the GNF Atlas. We therefore confirmed PKP4 expression in fetal spinal cord in a GEO data set for two second-trimester normal spinal cord samples (GEO accession no. GSE1481).
Fetal lung was represented in amniotic fluid by the surfactant protein genes SFTPB and SFTPC and the bronchial epithelial cell gene stratifin. This is consistent with both midtrimester fetal physiology and prior results from our laboratory.14 Gastrointestinal-specific transcripts for coagulation factor 7 and pancreatic trypsin were also present in the amniotic fluid; these are known to be expressed in the midtrimester human liver and pancreas, respectively.
The small proline-rich protein genes SPRR1B and SPRR2B were found in the amniotic fluid core transcriptome. We propose that these transcripts are likely to derive from the fetal skin, although fetal skin is not represented in the GNF Atlas. These small proline-rich proteins are involved with formation of the cell envelope in keratinocytes and are highly expressed in differentiating squamous epithelium, including fetal skin. Postnatally, SPRR genes are known to be highly expressed in the squamous epithelium of the upper gastrointestinal tract, as reflected in the GNF Atlas results (adult tongue, tonsil).15
We created lists of genes associated with developing heart, kidney, bladder, intestine, skin, placenta, and amnion based on our Medline and pathways-analysis software searches. A major caveat to these results is that most of the genes identified in the literature and pathways-analysis software searches were not tissue-specific in their expression pattern and therefore could originate from more than one fetal tissue. The analysis is summarized in Appendices 6–12 (http://links.lww.com/AOG/A271). These gene lists and literature references are an additional source of functional annotation for our ongoing project characterizing gene function in human development (http://dflat.cs.tufts.edu) and represent an additional resource for future studies of gene expression in human fetuses.
In the present study we have defined the amniotic fluid core transcriptome, which is comprised of 476 well-annotated genes consistently expressed in the amniotic fluid supernatant of euploid midtrimester fetuses. The multiple organ-specific transcripts and physiologic systems detected in the amniotic fluid core transcriptome provide strong evidence that amniotic fluid cell–free fetal RNA originates from more than one cell type. Moreover, both the functional analysis and GNF Atlas mapping suggest that amniotic fluid contains genes derived from the fetal nervous system. The presence of these CNS genes in the amniotic fluid was unexpected given the lack of direct physical contact between the fetal brain and the amniotic fluid. We speculate that these transcripts may enter the amniotic fluid directly from CNS regions with relatively thin tissue barriers such as the olfactory mucosa, the tympanic membrane, or cerebral fontanelles. Alternatively, CNS genes may enter the amniotic fluid indirectly through the fetal cardiovascular circulation. Another possibility is that they may be transcripts that were directly released from the CNS into the amniotic fluid before neural tube closure during the first trimester and persisting long-term. Despite our lack of knowledge about their precise route of entry and their in vivo half-life, these fetal brain-specific transcripts may be a novel source of biomarkers of nervous system development that are uniquely accessible from living fetuses.
Evidence that mammalian target of rapamycin signaling is active in midtrimester human fetuses is another significant finding in live human pregnancies. The mammalian target of rapamycin pathway is a central regulator of cell growth, integrating extracellular signals from amino acids, growth factors such as insulin growth factor-1, energy, and stress.16 This pathway is increasingly recognized as an important placental mechanism for influencing fetal growth in response to nutrient availability.17,18 Prior work from our laboratory has documented the presence of the mammalian target of rapamycin gene in midtrimester amniotic fluid using NanoArray polymerase chain reaction.19 The present study confirms that the mammalian target of rapamycin signaling pathway as a whole is significantly overrepresented in the amniotic fluid core transcriptome. This illustrates the ability of amniotic fluid cell–free fetal RNA to contribute translational human data to results from animal and in vitro work. The detection of fetal mammalian target of rapamycin signaling in amniotic fluid may have clinical significance in future studies of abnormal fetal growth.
The presence of lung gene transcripts for the surfactant proteins B and C was an expected finding given established knowledge about fetal lung development,20 the contribution of lung fluid to amniotic fluid volume, and previous research on amniotic fluid cell–free fetal RNA.14 The detection of organ-specific genes from the liver and pancreas and the less tissue-specific mucin genes associated with the intestine are similarly biologically plausible given their developmental expression and direct communication between the fetal gastrointestinal tract and the amniotic cavity.21–24
Surprisingly, we did not find significant enrichment of renal system development and function in the pathways analysis. There were no kidney-specific genes identified using the GNF Atlas. This is despite the substantial contribution of fetal urine to amniotic fluid and the presence of cell-free fetal nucleic acids in maternal urine.25 Although there were some kidney and bladder-associated genes in the amniotic fluid core transcriptome identified from the literature and pathways-analysis software searches, these were all nonorgan-specific genes that are widely expressed in all fetal tissues such as the fibroblast growth factor receptor 1.
The absence of gene transcripts specific to kidney highlights one of the major limitations of our study. The reliance on the presence of “organ-specific genes” to identify putative organ sources is subject to an inherent bias in the GNF Atlas as a result of the range of tissues sampled and the variable number of organ-specific genes in each organ. The variability in tissue specificity has been quantified by Dezso et al who determined that fetal brain has a relatively high number of organ-specific genes, whereas the fetal kidney has relatively few.26 This variation may explain both the high numbers of fetal brain transcripts and the absence of fetal kidney-specific genes found in the amniotic fluid core transcriptome gene list. Similarly, annotation biases may be present in the pathways-analysis software database that may underrepresent renal development. Presently, there are no definitive methods to determine the origins of cell-free gene transcripts within amniotic fluid. Despite these limitations, we considered the tissue expression patterns of the universally expressed genes and the functional analyses as valuable approaches to understanding the biology of amniotic fluid cell–free fetal RNA.
Another limitation of our study is the lack of clinical follow-up on the women who contributed the amniotic fluid supernatant samples. The most common indication for amniocentesis was advanced maternal age. Although all of the fetuses were euploid, we were not able to collect long-term obstetric outcome data because our samples were anonymized. Our results should therefore be interpreted with caution, because they may not be completely representative of uncomplicated pregnancies.
Clinically, there are numerous potential applications for amniotic fluid cell–free fetal RNA in fetal medicine. Functional genomic analysis is a powerful tool for understanding both normal physiology and disease. Prior studies of human development have mainly relied on postnatal tissue specimens27,28 or extraembryonic prenatal samples.29 Amniotic fluid supernatant has the unique property of being a pure fetal sample that can provide gene expression information on live fetuses without posing unacceptable risks to the pregnancy. If reproducible gene expression profiles can be demonstrated for specific phenotypes, then these could be used to learn more about the pathophysiology of fetal diseases and to identify biomarkers. This approach has already been successfully used to study fetuses with trisomies 18 and 21 and has generated candidate therapies for translational research.6,7 Furthermore, the presence of fetal brain-specific transcripts suggests novel approaches to the study of developmental disorders that involve the CNS.
In conclusion, this study provides a detailed examination of amniotic fluid cell–free fetal RNA from euploid midtrimester pregnancies and suggests future potential applications for fetal gene expression studies. Our broad approach, using multiple publicly available resources, enabled us to discover significant biological processes represented in the amniotic fluid core transciptome, determine putative organ sources of amniotic fluid cell–free fetal RNA, and identify genes of specific biological interest that represent potential fetal biomarkers. This work develops the concept of amniotic fluid as a “summary fluid” derived from multiple organs and lays the foundation for future studies of gene expression in abnormal fetal development.
1. Orozco AF, Bischoff FZ, Horne C, Popek E, Simpson JL, Lewis DE. Hypoxia-induced membrane-bound apoptotic DNA particles: potential mechanism of fetal DNA in maternal plasma. Ann N Y Acad Sci 2006;1075:57–62.
2. Maron JL, Bianchi DW. Prenatal diagnosis using cell-free nucleic acids in maternal body fluids: a decade of progress. Am J Med Genet C Semin Med Genet 2007;145C:5–17.
3. Sekizawa A, Yokokawa K, Sugito Y, Iwasaki M, Yukimoto Y, Ichizuka K, et al.. Evaluation of bidirectional transfer of plasma DNA through placenta. Hum Genet 2003;113:307–10.
4. Bianchi DW, LeShane ES, Cowan JM. Large amounts of cell-free fetal DNA are present in amniotic fluid. Clin Chem 2001;47:1867–9.
5. Pavlidis P, Li Q, Noble WS. The effect of replication on gene expression microarray experiments. Bioinformatics 2003;19:1620–7.
6. Koide K, Slonim DK, Johnson KL, Tantravahi U, Cowan JM, Bianchi DW. Transcriptomic analysis of cell-free fetal RNA suggests a specific molecular phenotype in trisomy 18. Hum Genet 2011;129:295–305.
7. Slonim DK, Koide K, Johnson KL, Tantravahi U, Cowan JM, Jarrah Z, et al.. Functional genomic analysis of amniotic fluid cell-free mRNA suggests that oxidative stress is significant in Down syndrome fetuses. Proc Natl Acad Sci U S A 2009;106:9425–9.
8. Gentleman RC, Carey VJ, Bates DM, Bolstad B, Dettling M, Dudoit S, et al.. Bioconductor: open software development for computational biology and bioinformatics. Genome Biol 2004;5:R80.
9. Bolstad BM. Low level analysis of high-density oligonucleotide array data: background, normalization and summarization. Berkeley (CA): University of California; 2004.
10. Affymetrix. Affymetrix technical note: new statistical algorithms for monitoring gene expression on GeneChip probe arrays. Santa Clara (CA): Affymetrix; 2001.
11. Li Y, Zhou X, St John MA, Wong DT. RNA profiling of cell-free saliva using microarray technology. J Dent Res 2004;83:199–203.
12. Su AI, Wiltshire T, Batalov S, Lapp H, Ching KA, Block D, et al.. A gene atlas of the mouse and human protein-encoding transcriptomes. Proc Natl Acad Sci U S A 2004;101:6062–7.
13. Kislinger T, Cox B, Kannan A, Chung C, Hu P, Ignatchenko A, et al.. Global survey of organ and organelle protein expression in mouse: combined proteomic and transcriptomic profiling. Cell 2006;125:173–86.
14. Larrabee PB, Johnson KL, Lai C, Ordovas J, Cowan JM, Tantravahi U, et al.. Global gene expression analysis of the living human fetus using cell-free messenger RNA in amniotic fluid. JAMA 2005;293:836–42.
15. Lee CH, Marekov LN, Kim S, Brahim JS, Park MH, Steinert PM. Small proline-rich protein 1 is the major component of the cell envelope of normal human oral keratinocytes. FEBS Lett 2000;477:268–72.
16. Wullschleger S, Loewith R, Hall MN. TOR signaling in growth and metabolism. Cell 2006;124:471–84.
17. Wen H, Abbasi S, Kellems R, Xia Y. mTOR: a placental growth signaling sensor. Placenta 2005;26(suppl):S63–9.
18. Roos S, Powell TL, Jansson T. Placental mTOR links maternal nutrient availability to fetal growth. Biochem Soc Trans 2009;37:295–8.
19. Massingham LJ, Johnson KL, Bianchi DW, Pei S, Peter I, Cowan JM, et al.. Proof of concept study to assess fetal gene expression in amniotic fluid by nanoarray PCR. J Mol Diagn 2011;13:565–70.
20. Liley HG, White RT, Warr RG, Benson BJ, Hawgood S, Ballard PL. Regulation of messenger RNAs for the hydrophobic surfactant proteins in human lung. J Clin Invest 1989;83:1191–7.
21. Hassan HJ, Leonardi A, Chelucci C, Mattia G, Macioce G, Guerriero R, et al.. Blood coagulation factors in human embryonic-fetal development: preferential expression of the FVII/tissue factor pathway. Blood 1990;76:1158–64.
22. Sarkar SA, Kobberup S, Wong R, Lopez AD, Quayum N, Still T, et al.. Global gene expression profiling and histochemical analysis of the developing human fetal pancreas. Diabetologia 2008;51:285–97.
23. Buisine MP, Devisme L, Savidge TC, Gespach C, Gosselin B, Porchet N, et al.. Mucin gene expression in human embryonic and fetal intestine. Gut 1998;43:519–24.
24. Buisine MP, Devisme L, Maunoury V, Deschodt E, Gosselin B, Copin MC, et al.. Developmental mucin gene expression in the gastroduodenal tract and accessory digestive glands. I. Stomach. A relationship to gastric carcinoma. J Histochem Cytochem 2000;48:1657–66.
25. Shekhtman EM, Anne K, Melkonyan HS, Robbins DJ, Warsof SL, Umansky SR. Optimization of transrenal DNA analysis: detection of fetal DNA in maternal urine. Clin Chem 2009;55:723–9.
26. Dezso Z, Nikolsky Y, Sviridov E, Shi W, Serebriyskaya T, Dosymbekov D, et al.. A comprehensive functional analysis of tissue specificity of human gene expression. BMC Biol 2008;6:49.
27. Kho AT, Bhattacharya S, Tantisira KG, Carey VJ, Gaedigk R, Leeder JS, et al.. Transcriptomic analysis of human lung development. Am J Respir Crit Care Med 2010;181:54–63.
28. Madsen-Bouterse SA, Romero R, Tarca AL, Kusanovic JP, Espinoza J, Kim CJ, et al.. The transcriptome of the fetal inflammatory response syndrome. Am J Reprod Immunol 2010;63:73–92.
29. Farina A, Morano D, Arcelli D, De Sanctis P, Sekizawa A, Purwosunu Y, et al.. Gene expression in chorionic villous samples at 11 weeks of gestation in women who develop preeclampsia later in pregnancy: implications for screening. Prenat Diagn 2009;29:1038–44.