Larrabee PB, Johnson KL, Lai C, Ordovas J, Cowan JM, Tantravahi U, Bianchi DW. Global Gene Expression Analysis of the Living Human Fetus Using Cell-Free Messenger RNA in Amniotic Fluid. JAMA. 2005;293(7):836-842. doi:10.1001/jama.293.7.836
Author Affiliations: Department of Pediatrics,
Divisions of Newborn Medicine (Dr Larrabee) and Genetics (Drs Johnson, Cowan,
and Bianchi), Tufts–New England Medical Center, Boston, Mass; Nutrition
and Genomics Laboratory, Jean Mayer–United States Department of Agriculture
Human Nutrition Research Center on Aging, Tufts University, Boston (Drs Lai
and Ordovas); and Department of Pathology and Laboratory Medicine, Women and
Infants Hospital, Brown University, Providence, RI (Dr Tantravahi).
Context No molecular biological tests are available to monitor the ongoing development
of human fetuses in vivo.
Objective To determine whether cell-free fetal messenger RNA (mRNA) in amniotic
fluid can be detected using oligonucleotide microarrays to study large-scale
gene expression in living human fetuses, with analysis of sex, gestational
age, and fetal pathology as variables.
Design, Setting, and Patients Four samples of cell-free amniotic fluid were analyzed from pregnant
women between 20 and 32 weeks’ gestation and undergoing amnioreduction
for polyhydramnios associated with twin-twin transfusion syndrome or hydrops
fetalis (cases). The control consisted of 6 pooled amniotic fluid samples
from women at 17 weeks’ gestation and undergoing genetic amniocentesis.
After extraction from the normally discarded fraction of amniotic fluid, RNA
was amplified twice, labeled, and analyzed using gene expression microarrays.
Main Outcome Measure Relative mRNA expression in cell-free samples of amniotic fluid from
fetuses with polyhydramnios at different gestational ages vs cell-free amniotic
fluid from a pooled control.
Results Thirty-six percent of 22 283 probe sets represented on the arrays
were present in the cell-free amniotic fluid, and a median of 20% of all probe
sets differed between cases and the pooled control. Only male samples expressed
1 Y chromosome transcript. The expression of some developmental transcripts,
such as surfactant proteins, mucins, and keratins, changed with gestational
age by up to 64-fold. A water transporter gene transcript was increased up
to 18-fold in both twin-twin transfusion samples. Placental gene transcripts
were not present in any samples.
Conclusions This pilot study demonstrates that cell-free fetal mRNA can be extracted
from amniotic fluid and successfully hybridized to gene expression microarrays.
Preliminary analysis suggests that gene expression changes can be detected
in fetuses of different sexes, gestational age, and disease status. Cell-free
mRNA in amniotic fluid appears to originate from the fetus and not the placenta.
Because of the inherent difficulties with research on human fetuses,
the analysis of fetal gene expression has in large part been limited to the
examination of tissue from human abortuses and the assessment of animal models
for genes and developmental pathways that are conserved across species. Fetal
monitoring in vivo is limited to noninvasive methods such as the measurement
of uterine size or anatomic evaluation by fetal sonography. In addition, genetic
analysis can be performed on amniotic fluid components, including amniocytes,
which typically require time-consuming expansion in vitro before use, and
cell-free proteins in the amniotic fluid, such as α-fetoprotein, which
can serve as biomarkers for genetic anomalies. The cell-free component of
the amniotic fluid is discarded after these analyses and is therefore available
for research and future clinical applications.
Cell-free fetal DNA in the serum and plasma of pregnant women was first
described by Lo et al1 in 1997 after others
demonstrated the presence of circulating tumor-specific DNA sequences in cancer
patients.2- 4 After
it was shown that cell-free fetal DNA is also present in the urine of pregnant
women,5 we hypothesized that amniotic fluid,
as a reservoir of fetal urine, would contain fetal DNA. In a preliminary study,
we demonstrated that much larger quantities of fetal DNA are present in amniotic
fluid than in maternal serum (100- to 200-fold difference).6
Cell-free fetal messenger RNA (mRNA) in maternal plasma was first demonstrated
by Poon et al7 in 2000. In 2002, Tsui et al8 subsequently showed that fetal mRNA in peripheral
blood is unexpectedly and remarkably stable. We then hypothesized that if
large amounts of DNA were present in amniotic fluid, RNA would likely be present
and stable enough to allow for the study of fetal gene expression. Furthermore,
recent technologic advances and information from the Human Genome Project
have made possible the development of gene expression microarrays, which can
be analyzed for the presence and quantity of tens of thousands of gene transcripts
There is no comprehensive way to examine global fetal gene expression
in vivo. The purpose of this study was to determine whether it was possible
to extract fetal RNA from cell-free amniotic fluid supernatant, successfully
hybridize it to oligonucleotide microarrays, and analyze the arrays for the
presence and quantity of thousands of different gene transcripts. Gene expression
profiling of amniotic fluid by using microarrays could provide important information
about the well-being, development, and potential disease status in the living
fetus. This report presents preliminary results of gene expression analyses
in the living human fetus by using a fraction of amniotic fluid that is normally
discarded: the cell-free supernatant.
Approval was obtained from Tufts–New England Medical Center and
Women and Infants Hospital institutional review boards to obtain amniotic
fluid supernatant samples for this study. Informed consent was deemed unnecessary
because discarded samples were used anonymously.
In healthy pregnancies, between 10 and 30 mL of amniotic fluid can safely
be removed from the fetal sac, but only about 8 to 15 mL of supernatant remains
after clinical testing, including karyotype analysis and α-fetoprotein
measurement. Preliminary experiments showed that this remaining volume of
amniotic fluid from a normal singleton fetus might not contain a sufficient
quantity of mRNA for microarray analysis. Therefore, 5 large-volume amniotic
fluid samples were obtained from 4 pregnant women undergoing therapeutic amnioreduction
for polyhydramnios. Two of these women had a fetus with hydrops (gestational
ages of 294/7 weeks and 32 weeks), 1 had a fetus with twin-twin transfusion
(TTT) syndrome (gestational age of 20 weeks), and another woman with fetal
TTT underwent amnioreduction at 2 gestational ages (216/7 weeks and 243/7
weeks) and thus provided 2 samples. Cell-free supernatant was obtained by
centrifugation of at least 350g for 10 minutes.
To obtain sufficient RNA from healthy fetuses for comparison, multiple
10-mL samples of frozen, archived amniotic fluid supernatant were combined
to form larger pools. These samples were obtained from pregnant women who
were between 17 and 18 weeks’ gestational age and underwent routine
genetic amniocentesis for advanced maternal age. Six samples from male fetuses
and 6 from female fetuses were selected according to known normal karyotypes
and similar gestational age. These control samples were combined by sex, with
each 60-mL pool representing amniotic fluid supernatant of an average 17-week
fetus. Cell-free supernatant was obtained by centrifugation of 350g for 10 minutes.
After centrifugation, total RNA was extracted from all samples using
the QIAamp Viral RNA Vacuum Protocol for Large Sample Volumes (Qiagen Inc,
Valencia, Calif), with modification. Volumes of viral lysis buffer (AVL) and
ethanol were increased to 20 mL for each 5 mL of amniotic fluid per column,
and 60-mL syringes were attached to the columns to accommodate the large volume
samples. The mRNA was amplified twice and converted to complementary RNA (cRNA)
by in vitro transcription in the presence of biotinylated nucleoside triphosphates
following the GeneChip Eukaryotic Small Sample Target Labeling Technical Note
(Affymetrix Inc, Santa Clara, Calif). Samples were further purified by phenol-chloroform
extraction by using Phase Lock Gels (Eppendorf AG, Hamburg, Germany). To verify
the quantity and quality of biotinylated cRNA, samples were analyzed using
gel electrophoresis and fragmented before hybridization to Affymetrix Test3
oligonucleotide arrays according to the manufacturer’s documentation.
Subsequently, 15 to 75 μg of biotinylated cRNA was hybridized to Affymetrix
U133A arrays, which are composed of 22 283 probe sets and more than 500 000
distinct oligonucleotide features, representing 14 239 of the best-characterized
human genes. Accession numbers herein are from the National Center for Biotechnology
Each array was scanned at 570 nm by a confocal scanner (Agilent, Palo
Alto, Calif) with a resolution of 3 μm per pixel. Pixel intensities were
measured, and expression signals were extracted and analyzed by using Microarray
Suite 5.0 (Affymetrix). All microarrays were scaled to the same target signal
of 50 by using the “all probe sets” scaling option so that the
expression signals from all experiments could be directly compared.
Comparison analyses were performed with the Wilcoxon signed rank test
via the Microarray Suite 5.0 software between each of the TTT or hydrops cases
and the pooled male control. Data from the TTT3 sample at 216/7 weeks and
the pooled female sample were not used because of noisy data (see “Results”).
Data were copied into Excel files (Microsoft, Redmond, Wash) and sorted for
probe sets called “present” in either the case or control. Data
sets for each case were then narrowed to transcripts that were increased or
decreased relative to the pooled male control by a 2-fold or greater difference.
The 2 remaining TTT data sets were then compared with each other, as were
the 2 hydrops data sets, to detect genes consistently increased or decreased
in both cases, with the same disease compared with the pooled control. Finally,
expression levels of selected genes of interest, such as Y chromosome genes,
surfactant, mucin, keratin, aquaporin, and placental genes, were reviewed
in all cases relative to the pooled male control.
Table 1 shows the volumes of amniotic
fluid used for extraction, amount of RNA eluted, and quantities of biotinylated
cRNA available for the microarrays after amplification. Five of the 7 samples
hybridized well to the arrays, as measured by scale factors within 3-fold
of one another, as recommended by the manufacturer. Therefore, data are presented
only for these 5 samples (TTT1, TTT2, hydrops 1, hydrops 2, and the pooled
male control). The other 2 samples (TTT3 and the pooled female control) had
low signals, as shown by higher scale factors, and were therefore not included
in this analysis. The average background level of the images (median, 55.37
units; range, 49.64-61.50) was highly similar across all the arrays (typical
values range from 20-100). Noise, a measurement that reflects sample quality
and electrical noise of the scanner, was also comparable across the arrays
(median, 2.21 units; range, 2.06-2.37).
For the 5 analyzed samples, a median of 36% (range, 11%-44%) of the
probe sets represented on the microarrays was detected as “present,”
62% (range, 54%-88%) was not detectable (ie, “absent”), and 2%
(range, 1%-2%) was “marginal.” There was evidence of low-level
false or cross-hybridization according to the presence of randomly distributed
probe sets; these results were not statistically significant and were therefore
not included for analysis. Within individual samples, there was some variation
in the 3′/5′ ratios of the internal control genes (glyceraldehyde-3-phosphate
dehydrogenase and actin) that are used to assess RNA sample and assay quality.
When these control genes were compared across all samples, certain control
genes consistently had a normal (ie, less than 3) 3′/5′ ratio
in every sample, whereas other control genes always had a high 3′/5′
ratio (10 to 100).
Of the 22 283 probe sets present on the microarray, a median of
20% (range, 15%-29%) had significant differences in their levels of expression
between the cases and the pooled male control. Table 2 and Table 3 show a
selection of genes with the most statistically significant different levels
of expression (larger than 4-fold) in both TTT fetuses and both hydrops fetuses,
respectively, compared with the pooled control. One Y chromosome probe set
(accession NM_001008) was present in all 4 samples from male fetuses (TTT2,
hydrops 1, hydrops 2, and the pooled male control) but not in the sample from
the female fetus (TTT1).
To determine whether genes involved in fetal development showed differential
expression with increasing gestational age of the fetal samples, specific
developmental gene families were investigated and compared with the pooled
male control. Statherin (accession NM_003154), a gene involved in saliva secretion
and ossification, was up to 28 times more concentrated in the older fetuses
compared with the 17-week pooled control. Surfactant genes (Table 4) increased from only 3 transcripts present in the pooled
control to all 9 present in hydrops 1 (29 weeks’ gestational age). Only
a few transcripts in the mucin gene family (Table
5) were present in any sample, but certain transcripts such as tracheobronchial/gastric
mucin and salivary mucin were increased up to 56-fold in the older fetuses
compared with the 17-week pooled control. Most keratin genes decreased with
increasing gestational age (Table 6);
several transcripts had up to a 4-fold decrease compared with the control.
Other genes were reviewed in the context of fetal pathology or maternal-fetal
trafficking of cell-free nucleic acids. One transcript for aquaporin 1, a
water transporter, was elevated up to 18-fold (Table 7) in both TTT fetuses but not the hydropic fetuses compared
with the pooled control. Aquaporin 3 expression varied minimally in the cases
compared with the pooled control, with 2 of 3 transcripts not detected in
any fetus. Placenta-specific transcripts, including corticotropin-releasing
hormone, chorionic somatomammotropin hormone 1 (placental lactogen), and the β
subunit of chorionic gonadotropin, were not present in any of the samples.
To our knowledge, this is the first in vivo study of global gene expression
in the living human fetus by oligonucleotide microarray analysis of fetal
mRNA isolated from cell-free amniotic fluid. Cell-free fetal RNA was successfully
extracted from this typically discarded component of amniotic fluid, amplified,
labeled, and hybridized to oligonucleotide microarrays. Our analyses revealed
important information about the presence and level of gene expression in living
human fetuses. In addition, these preliminary data appear to show that observed
gene expression patterns correlated with known variables (sex, gestational
age, and disease status) between the cases and control. Although validating
our results with real-time quantitative reverse-transcriptase polymerase chain
reaction would have been optimal, it was impossible because of limited sample
template. However, the presence of 1 Y chromosome transcript in all 4 male
samples but not the female sample provided physiologic validation of the data.
Next, expression differences were evaluated in gene families known to
be important in fetal development to look for changes with gestational age.
Significant differences were observed in several genes expressed in lung,
intestine, and skin epithelial cells, which are all in contact with amniotic
fluid. For example, it is well known that the type and quantity of surfactant
genes expressed in human fetal lungs increase during development.11 Messenger RNA for surfactant proteins B and C is
detectable as early as 13 weeks, and by 24 weeks, the levels are 50% and 15%,
respectively, of adult levels. Surfactant protein D mRNA is first detectable
in the second trimester, with expression increasing throughout fetal and postnatal
development. Surfactant protein A expression begins only after about 30 weeks
and reaches maximum near term. The findings of the current study are consistent
with the published data. All of the fetuses older than 24 weeks produced increased
amounts of surfactant proteins B and C compared with the 17-week control,
and surfactant proteins D and A were observed only after 29 weeks. The fact
that the 20-week fetus and the 32-week fetus produced fewer surfactant transcripts
than some of the more immature fetuses may be due to differing hybridization
efficiencies but may also be explained by their severe disease status.
Many differences with gestational age were observed in expression of
proteins produced by epithelial cells and assumed to be in contact with the
amniotic fluid. Mucins are filamentous glycoproteins present at the interface
of epithelia and extracellular environments in the gastrointestinal tract,
lungs, or urogenital tract.12 As fetuses mature,
they produce mucin in increasing amounts to protect their epithelia in preparation
for life outside the womb. This is consistent with the current study, in which
the level of several mucin transcripts increased with advancing gestational
age. The kidney, intestine, and skin also produce keratins. In the skin, keratin
proteins are first produced in the intermediate layer during the 11th week
of human fetal development. During the fifth month of gestation, this layer
develops into definitive layers of keratinocytes, and as cells progress from
the basal layer of stem cells to the outer horny layer, they stop production
of keratins, which are then bundled and cross-linked. By arrival of the top
layer, metabolic activity of the cells has ceased, with the scalelike terminally
differentiated keratinocytes forming the horny protective layer.13 The
reason for the observed gestational age–related decrease in keratin
expression could be that as the fetal skin matures, fewer keratin-producing
cells are in direct contact with the amniotic fluid and may not release their
mRNA into the cell-free fraction. Rather, the layer of hard, cross-linked
keratin itself may protect the buried keratin-producing cells from releasing
mRNA into amniotic fluid.
Next, aquaporin genes, a family of water transporters, were reviewed
because the large differences in aquaporin 1 levels between fetuses with TTT
and the control were highly significant. In fetal sheep hearts, aquaporin
1 expression has been shown to be developmentally regulated with fetal age
and also in response to chronic anemia, a condition associated with interstitial
fluid generation and hydrops.14 In human fetal
membranes, there is evidence that aquaporin 1 is present on the apical aspect
of the chorionic plate amnion but aquaporin 3 is not active.15 It
has been postulated that aquaporin 1 may play a role in water movement from
the amniotic cavity across the placenta into the fetal circulation. Our findings
support the presence of aquaporin 1 and the relative lack of aquaporin 3 in
amniotic fluid. The significant increase of aquaporin 1 in TTT patients suggests
that it might play a role in the polyhydramnios associated with TTT but not
Several placenta-specific genes, including corticotropin-releasing hormone,
chorionic somatomammotropin hormone 1, and chorionic gonadotropin, were also
examined. The presence of these genes in the plasma of pregnant women has
been presented as proof of fetal-maternal trafficking of cell-free RNA.16,17 The absence of these placental transcripts
in amniotic fluid suggests that fetal-maternal transplacental trafficking
of nucleic acids is primarily unidirectional toward the mother. Previous work
has shown that cell-free fetal DNA in maternal plasma is significantly more
concentrated relative to cell-free maternal DNA in the fetal plasma.18
This study demonstrated that for the majority of samples, cell-free
RNA from amniotic fluid successfully hybridized to microarrays. However, some
samples hybridized less well, possibly because the RNA was degraded, which
could occur from delays before sample processing or introduction of a freeze/thaw
cycle. However, freezing and thawing did not appear to be detrimental to the
male control, which hybridized well despite its composition of archived amniotic
fluid samples that had been stored at –80°C. Additionally, it has
been demonstrated that a single freeze/thaw cycle produces no significant
effect on the cell-free RNA concentration in plasma or serum.8 It
is possible that cell-free RNA is inherently degraded and therefore has different
properties than RNA extracted from whole cells. RNA is labile, so it is surprising
that any cell-free RNA in amniotic fluid survives until extraction. There
is evidence that circulating RNA in plasma is associated with stabilizing
particles.19 In this study, certain internal
control genes had normal 3′/5′ ratios in every sample, whereas
ratios of certain other genes were always high. This discrepancy suggests
a pattern of preservation of specific RNA transcripts, which could be related
to alteration and packaging of mRNA during apoptosis. Housekeeping genes vary
significantly in their expression patterns between various tissues and organisms,20 and these patterns in cell-free RNA in amniotic fluid
The low levels of nonsignificant false positives observed in this study
could be due to cross-hybridization of the short oligonucleotide probes on
the arrays with different mRNAs that have short sequences in common. However,
the algorithms take this into account and have largely eliminated this source
of error.21 Additionally, the cases and control
had different genetic backgrounds and other variables (gestational age and
disease status) because amniocentesis is not generally performed on healthy
fetuses older than 19 to 20 weeks. Further, small amounts of maternal contamination
could possibly confound the results. Therefore, the data on diversity of fetal
gene expression must be interpreted with caution at this stage of investigation,
and further study is necessary.
For this pilot study, large-volume samples were used to demonstrate
feasibility. However, it appears that amniotic fluid samples from healthy
fetuses contain a higher concentration of cell-free mRNA than amniotic fluid
samples from fetuses with polyhydramnios (Table
1). Our technical improvements are directed toward improving extraction
of mRNA from routinely collected amniotic fluid samples (<30 mL) so that
individual fetuses may be studied.
In summary, this study demonstrates that cell-free fetal mRNA can be
isolated from amniotic fluid and successfully detected using oligonucleotide
microarrays. Preliminary gene expression analyses appear to show gene expression
patterns that vary among fetuses of different sexes, gestational age, and
disease state. The entire study was conducted by using a portion of amniotic
fluid that is typically discarded and thus is readily available for use. The
intriguing gene expression differences observed suggest that this technology
could facilitate the advancement of human developmental research, as well
as the cultivation of new biomarkers for assessment of the living fetus.
Corresponding Author: Diana W. Bianchi,
MD, 750 Washington St, NEMC #394, Tufts–New England Medical Center,
Boston, MA 02111 (firstname.lastname@example.org).
Author Contributions: Dr Bianchi had full access
to all of the data in the study and takes responsibility for the integrity
of the data and the accuracy of the data analysis.
Study concept and design: Lai, Bianchi.
Acquisition of data: Larrabee, Johnson, Ordovas,
Analysis and interpretation of data: Larrabee,
Johnson, Lai, Ordovas, Bianchi.
Drafting of the manuscript: Larrabee, Johnson,
Critical revision of the manuscript for important
intellectual content: Larrabee, Johnson, Lai, Ordovas, Cowan, Tantravahi,
Statistical analysis: Larrabee, Johnson.
Obtained funding: Bianchi.
Administrative, technical, or material support:
Lai, Ordovas, Cowan, Tantravahi.
Study supervision: Johnson, Bianchi.
Financial Disclosures: Drs Larrabee, Johnson,
and Bianchi have filed for patents for the methodology described in this article.
No other authors reported financial disclosures.
Funding/Support: This study was supported by
grant NIH HD42053 from the National Institutes of Health (Dr Bianchi).
Role of the Sponsor: The NIH did not participate
in the design and conduct of the study, data management and analysis, or manuscript
preparation, including review and authorization for submission.