Keywords RNA-sequencing - cord blood - preterm birth - transcriptomics
In the United States, preterm birth (PTB) accounts for 9.6% of live births and is
the leading cause of infant morbidity and mortality in nonanomalous infants.[1 ] PTB poses a significant economic burden of up to $26 billion to care for these infants
in the United States. Primary prevention strategies such as antioxidant supplementation
or screening and treatment of maternal infections have failed to reduce or eliminate
spontaneous PTB (sPTB).[2 ] To date, much of the focus in understanding PTB has focused on maternal factors
that incite preterm labor, such as inflammation, infection, and maternal decidual
or endometrial factors,[3 ] and not on possible contributing fetal factors. Little is known about the fetal
contribution to sPTB; however, there are emerging data that show that variants in
the DNA of the fetus, not the mother, may be the trigger for some PTBs.[4 ] Identifying transcriptomic signatures at the fetal molecular level by examining
differentially expressed genes between preterm and term cohorts using RNA-sequencing
(RNA-seq) fills the knowledge gaps in our understanding of the fetal contribution
to pregnancy-specific disorders such as sPTB.
New and emerging high throughput next-generation genomic technologies have led to
the ability to sequence messenger RNA, permitting interrogation of the entire fetal
transcriptome in umbilical cord blood. The transcriptome is the sum total of messenger
RNA expressed in a tissue. Transcriptome analysis captures a snapshot of cellular
activity that reflects the response to genetic, environmental, and epigenetic changes
in a biological system. Knowledge of the transcriptome allows for the quantification
and assessment of genes that may be active in different disease processes and at various
stages of development.[5 ] This technology has been applied to cancer therapeutics and diagnostics[6 ] leading to new insights and therapeutics. While transcriptomics has been studied
in pregnancy conditions such as sPTB, preeclampsia, and obesity among others,[7 ]
[8 ]
[9 ] this field is still emerging in terms of shedding light on the molecular underpinnings
of these complex pregnancy disorders.
The objective of the study was to measure fetal gene expression from umbilical cord
blood at the time of delivery in term and preterm pregnancies to identify differentially
regulated genes related to common PTB pathways, such as inflammation, immune function,
and oxidative stress. Our second objective was to evaluate differences in gene expression
in preterm compared with term fetuses to gain insight into fetal development. We hypothesized
that fetal genes are differentially expressed in common PTB pathways following sPTB
compared to term birth (TB). These findings have the potential to increase our understanding
of the fetal molecular contribution to sPTB, and will lay the foundation to improve
diagnosis, prognosis, and therapeutic strategies in obstetrics and pediatrics.[10 ]
Methods
The study was approved by the University of North Carolina at Chapel Hill Institutional
Review Board. This prospective case–control study included eight women who delivered
via idiopathic sPTB (<34 weeks) and eight women who delivered at term (>37 weeks)
with singleton fetuses who delivered at the University of North Carolina at Chapel
Hill. Preterm labor was defined as the presence of regular uterine contractions and
documented cervical effacement and/or dilatation in patients <37 weeks' gestational
age (GA). The preterm patients were admitted to the antepartum service and presented
with preterm contractions. TB was defined as delivery at greater than 37 weeks gestation
with no labor. Preterm premature rupture of membranes (PPROM) was confirmed by vaginal
“pooling,” and positive “nitrazine” or “ferning” tests.[11 ]
The sPTB and TB cohorts were matched for factors that could affect fetal gene expression
including: maternal age, race, fetal sex, medication exposure except for glucocorticoids
(all women in the sPTB received betamethasone as per American College of Obstetricians
and Gynecologists' guidelines), and mode of delivery (all women had cesarean sections).
Pertinent clinical information about the patients is provided in [Table 1 ]. Term patients were enrolled prior to elective cesarean delivery at term (390/7 to 396/7 weeks). All preterm infants received steroids prior to delivery with the median number
of days from administration of steroids to delivery of 2.18 (range: 1–19 days). Four
of the preterm fetuses were born via sPTB after PPROM with the median number of days
from rupture to delivery of 16 (range: 3–38 days). The phenotype of sPTB was limited
to women who had spontaneous labor with or without PPROM. Patients with abruption,
diabetes, clinical chorioamnionitis, fetal anomalies, and medication exposures were
excluded because these factors can all affect neonatal gene expression. Clinical chorioamnionitis
was defined by fever of ≥38.0°C in addition to at least two other signs of chorioamnionitis,
including uterine fundal tenderness, maternal tachycardia (>100 per minute), fetal
tachycardia (>160 per minute), and purulent or foul-smelling amniotic fluid.
Table 1
Demographic characteristics of the sPTB and TB cohorts
Demographic characteristics
sPTB
TB
Maternal age, mean (SD) [range] (y)
30 (8.3)
[22–44]
35.2 (3.92)
[30–42]
Ethnicity
Caucasian
4
4
African American
4
4
Gestational age at delivery, [range] (wk)
29
[245/7 to 341/7 ]
39
[370/7 to 404/7 ]
Fetal sex (number of males, number of females)
6, 2
6, 2
Cesarean sections
8
8
Power studies of gene expression using DESeq2 have shown that eight cases and controls
are a sufficient number to have adequate power to detect significant differential
gene expression between dichotomous groups (term/preterm).[12 ] It was not possible to match by GA or glucocorticoid exposure as this would be unethical
given the known benefits of glucocorticoids in preterm infants. A post hoc power analysis
with ssizeRNA shows that we have 70% power to detect a twofold difference in gene
expression based on the average number of counts (1,296) and dispersion (0.167) as
estimated from the data, and the Benjamini–Hochberg false discovery rate (FDR) of
5%.[13 ] For the analyses performed by race, there was inadequate power to detect significant
differences but analysis was performed to identify trends and generate hypotheses
to test in future studies.
Umbilical cord blood was collected in RNA PAXgene tubes and stored at −80°C. The PAXgene
blood tubes stabilize intracellular RNA and in processing the tubes, intact cells
are pelleted and RNA extracted. Total RNA was extracted from whole blood using a PAXgene
Blood miRNA kit on the QIAcube. A globin depletion protocol using custom Human Globin
AnyDeplete (NuGEn) probes enriched with fetal globin gene primers was applied to reduce
uninformative reads, enabling more RNA transcripts to be detected. Quality control
was performed for all samples using the LabChip GX system. All samples were handled
in a uniform way by one technician. Universal Plus mRNA-seq library prep kit (NuGen)
was used to generate mRNA-seq libraries from 500-ng umbilical cord blood total RNA.
Cord blood (coding and noncoding) was sequenced on the Illumina HiSeq 4000 PE 2 × 50.
Sequencing reads were analyzed as follows: chastity filtered reads were aligned to
hg38 using single pass STAR[14 ] with default parameters and transcript abundance estimates for each sample were
estimated with Salmon[15 ]
[16 ] to quantify the transcriptome defined by Gencode v22. Gene level counts were summed
across isoforms and the resulting gene count estimates were analyzed with DESeq2[17 ] to test for differences between sPTB and TB adjusted for race. Race was modeled
as an additive covariate, and sPTB versus TB effects were tested using the DESeq2
version 1.4.5 likelihood ratio test comparing the bivariate model of race and birth
to a reduced model of just race. Significant differences in gene expression were defined
by a Benjamini–Hochberg FDR (q -value) <0.1 for the primary outcome (GA).[18 ] Individual cellular fractions by race and GA were estimated by xCell followed by
association testing with a linear model.[19 ] Gene set analysis was performed to identify differentially expressed pathways.[20 ] The Gene Set Analysis in R (GSAR) is an open-source R/Bioconductor software package
for gene set analysis. A gene set is a group of genes that shares pathways, functions,
chromosomal localization, or other features.
Hierarchical clustering and principal component analysis. Two-dimensional unsupervised hierarchical clustering, dendrogram generation, and
heat-map plotting were performed using the pheatmap R package version 0.7.7. A complete
linkage algorithm with Manhattan distance function was applied to normalized, log2transformed
fold change data for statistically significant differentially expressed RNAs. Principal
component analysis for differentially expressed transcripts was accomplished using
the ‘prcomp()’ function in R version 3.0.3, in which a singular value decomposition
algorithm was applied to a centered and scaled correlation matrix of normalized, log2transformed
fold change RNA-seq data. Three-dimensional scatterplots of the first three principal
components (PC1, PC2, PC3) were generated using SigmaPlot version 12.5 (Systat Software,
San Jose, CA; [Supplementary Fig. 1 ]).
Results
RNA-seq was performed on cord blood collected from TB (n = 8) and sPTB (n = 8) events. The median RNA integrity number (RIN) value of the cord blood samples
was 8.5 (range: 7–10); there was no statistical difference in RIN values between cases
and controls. Fetal sex and maternal race displayed only minor effects on transcript
abundance as fetal sex was associated with only two genes (XIST and TSIX) and race
with none at the specified criteria.
One hundred and forty eight genes were differentially expressed in the setting of
sPTB compared with TB. Two pathways, cell cycle/metabolism and immune/inflammatory
signaling, represent the predominant pathways with significant differential expression
in the preterm versus term cohorts. Cell cycle/metabolism (e.g., electron transport
chain) gene expression was significantly higher (z -score = 18–24), and immune/inflammatory signaling gene expression (cytokines, interleukins,
and T-helper cells) was significantly lower (z -score = −41 to −26) in the sPTB cord blood compared with term ([Figs. 1 ] and [2 ]).
Fig. 1 Heat map of gene expression of electron transport chain gene expression. Gray (top)
represents preterm samples and red represents term samples. Blue color on the heat
map represents low gene expression and yellow color represents high gene expression.
Fig. 2 Heat map of gene expression with regard to T-cell receptor gene expression. Gray
(top) represents preterm samples and red represents term samples. Blue color represents
low gene expression and yellow color represents high gene expression.
Exploratory analysis was performed to investigate interactions of GA with race. The
results yielded 18 genes decreased in preterm African American (AA) infants and increased
in term AA infants (q < 0.1). However, the opposite pattern was seen in non-Hispanic White (NHW) preterm/term
infants ([Fig. 3 ]). These included genes associated with cell signaling, neutrophil activity, and
major histocompatibility complex type 1: DmX-like protein 2 (DMXL2 ), ubiquitin-like modifier ISG15 (ISG15 ), and butyrophilin subfamily 3 member A2 (BTN3A2 ).
Fig. 3 Heat map of gene expression showing significantly different gene expression patterns
in AA and NHW. Gray (top) represents preterm samples and red represents term samples.
Blue color represents low gene expression and yellow color represents high gene expression.
Gene expression data were evaluated using xCell to quantify a variety of cellular
phenotypes. There were no differences between leukocyte/lymphocyte compositions of
the preterm and term fetal cord blood; however, we were underpowered to detect a difference.
Preterm infants had significantly lower macrophage gene expression (p = 0.001) compared with term infants. This cellular composition analysis testing for
race-dependent effects showed that preterm and term AA infants had lower native CD4 T-cell gene expression compared with NHW infants
(p = 0.01; [Fig. 4 ]).
Fig. 4 CD4+ naïve T-cells. Preterm and term AA infants have lower CD4 T-cell gene expression
compared with term infants (p = 0.01).
Half of the preterm placentas and cords showed evidence of chorionitis and funisitis
even though none of the patients had clinical chorioamnionitis. None of the neonates
had sepsis. All of the term infants transitioned to room air without difficulty, had
Apgar scores at 1 and 5 minutes of >8, and did well clinically.
Discussion
Our findings that immune and inflammatory gene expressions were higher in term compared
with preterm infants provide novel insights into fetal development and fetal gene
expression in sPTB compared with TB. While other authors studied PTB by focusing on
gene expression changes in maternal whole blood,[8 ] we chose to focus on umbilical cord blood samples as they are representative of
the in utero environment of a neonate shortly before birth. While other investigators have examined
gene expression via microarray in amniotic fluid supernatant,[21 ] a few have performed whole transcriptome analyses using RNA-seq from cord blood,[10 ] and none to date have done this on preterm infants following sPTB. Our findings
suggest that despite the inflammatory milieu so commonly seen in sPTB, preterm infants
born due to sPTB had lower gene expression of immune/inflammatory signaling function
compared with term infants. This finding alone is not surprising as the immune system
is immature in preterm infants and not even fully mature at the time of TB as the
adaptive immune system must develop specificity and memory, which does not occur until
the early childhood years.[22 ]
[23 ] We found fetal genes associated with inflammation were down-regulated in the infants
born preterm compared with term despite previously published clinical and experimental
data showing a measurable fetal inflammatory response following preterm labor and
other inflammatory processes.[24 ]
In fact, our group found that even prior to the onset of symptoms of sPTB, there is
evidence of a fetal inflammatory response. In a previous study, we extracted cell-free
fetal (cff) RNA from amniotic fluid supernatant and used a customized nanostring panel
containing genes related to common PTB pathways (oxidative stress and inflammation)
and compared gene expression in fetuses who delivered preterm and those who delivered
at term. Compared with fetuses who ultimately delivered at term, fetuses who delivered
preterm had significantly increased gene expression of β-2 microglobulin in amniotic
fluid, a marker of the fetal inflammatory response.[25 ]
The question then arises regarding why other data including data from our laboratory
show evidence of a fetal inflammatory response and this cohort does not. We hypothesize
several possible reasons. Previous studies including our own measured gene expression
via cff RNA in amniotic fluid;[21 ]
[25 ] this study uses RNA extracted from cord blood. Gene expression in two different
biofluids and from two different forms of RNA (cff RNA vs. RNA) can differ. The gene
expression from RNA in amniotic fluid supernatant is a reflection of gene expression
in various fetal secretions including from urine, saliva, and tracheal secretions
compared with gene expression found in umbilical cord blood. Prior studies using amniotic
fluid showed that the regulation of gene expression varies extensively among fetal
tissues with skeletal and muscular system development and function, tissue development,
and hematological system development and function having higher gene expression compared
with other organ systems.[26 ] In our study using umbilical cord blood, we also noted differential gene expression
in genes related to tissue development and hematological system development. Previous
studies using amniotic fluid supernatant made use of customized nanostring panels
or microarrays that had limited gene coverage. Our data using RNA-seq suggest that
preterm infants in utero have lower gene expression specific to immune/inflammatory response at the time of
sPTB. The question then arises whether this results in an inability to mount an immune
response or contributes to PTB, which cannot be ascertained by our study.
There are multiple strengths to using RNA-seq from umbilical cord blood compared with
using cff RNA from amniotic fluid supernatant to understand fetal disease processes.
These include (1) less degradation of RNA compared with amniotic fluid supernatant
and higher quality of cellular RNA compared with cff RNA; (2) ability to identify
alternative splicing; and (3) identification of specific biological pathways from
comprehensive interrogation of the genome. Given the ability to interrogate the whole
transcriptome robustly with RNA-seq of umbilical cord blood, RNA-seq methods have
the potential to identify dysregulated genes specific to certain pregnancy conditions
as described in the current study with more sensitive detection of genes, transcripts,
and differential expression compared with previously published methods using cff RNA
and microarrays.
This study also sheds light on in utero developmental factors. Compared with term, preterm infants have higher cell cycle
and metabolism gene expression, suggesting a metabolic focus on growth and development.
Other studies confirmed differential gene expression related to GA with term fetuses
showing enrichment of salivary gland, tracheal, and renal transcripts as compared
with brain and embryonic neural cells in the second trimester.[27 ] Functional analysis of genes upregulated at term in amniotic fluid supernatant revealed
pathways that were highly specific for postnatal adaptation such as immune function,
digestion, respiration, carbohydrate metabolism, and adipogenesis.[21 ] Similar to the current study using cord blood, inflammation, a key process involved
in normal labor, has been also shown to increase in studies using term amniotic fluid.[21 ]
There are other intriguing findings from this study to highlight. We did not identify
any increase in gene expression related to the neonatal hypothalamic pituitary adrenal
(HPA) axis gene expression in the sPTB cohort. Corticotropin-releasing hormone (CRH),
a hypothalamic neuropeptide that regulates HPA axis activity, is integral to the regulation
of labor. There is a premature surge of CRH levels in the amniotic fluid, placenta,
and fetal membranes among pregnancies of women who deliver preterm. Fetal HPA axis activation is a potential therapeutic target pathway for sPTB prevention.
However, we found no alteration in gene expression in HPA gene expression in preterm
versus term neonates. All of the preterm infants received betamethasone prior to delivery,
which has also been shown to suppress fetal HPA axis function.[28 ]
[29 ]
[30 ] We also found racial/ethnic differences in gene expression related to immune-mediated
cell signaling. Racial disparities exist in neonatal sepsis, with infants of AA women
having higher rates of neonatal sepsis at term and preterm compared with infants of
NHW mothers.[31 ]
[32 ] Cell composition analysis showed that white blood cell gene expression also differed
in both term and preterm AA infants compared with NHW infants. These findings warrant
further investigation and we hypothesize that epigenetic mechanisms may be responsible
for this discrepant pattern.
There are limitations of the current study. The small sample size allowed us enough
power to detect differential gene expression between dichotomous groups but we were
not powered to detect differences by race or fetal sex. Despite this, we showed evidence
of differential gene expression by race. However, it is possible that this difference
between races is a false-positive result (despite the use of statistical methods to
minimize the chance of FDRs) due to normal variations in the human population. A larger
sample size with validation of results by reverse transcription-polymerase chain reaction
(RT-qPCR) is needed to further study this finding. Because we recruited patients in
real clinical situations, fetuses born were exposed to antenatal corticosteroids which
can change fetal gene expression.[33 ] However, the data on fetal gene expression changes with antenatal corticosteroids
are limited to murine and in vivo studies with minimal data on humans.[33 ] It is also possible that chorioamnionitis was present despite no clinical signs
of chorioamnionitis. Half of the preterm placentas showed evidence of chorionitis
and funisitis. However, this is neither a sensitive nor specific screening test for
chorioamnionitis. To the best of our ability, we limited variability in the cohorts
which could alter gene expression. However, a limitation inherent in the study design
was differences in GAs. This may contribute to some of the changes that were observed
in this study, which may be caused by different developmental stages. However, for
practical and ethical reasons, relative matching was considered to be the most reasonable
alternative for our study. Other authors have overcome this limitation by comparing
the human placental transcriptome to the macaque placental transcriptome at equivalent
GAs because macaques are closely related to humans.[34 ] Another limitation was that we included women who had premature rupture of membranes
(PROM) prior to onset of labor. Half of the preterm cohort had PROM and the other
half had spontaneous labor. Because we enrolled patients who were in spontaneous labor
with or without PROM with 50% of the sPTB cohort delivering after PROM, there was
heterogeneity in the cohort. Further work to study differences may be accomplished
with use of noninvasively obtained cff RNA in a larger cohort of patients.
In summary, RNA-seq of umbilical cord blood leukocytes can identify novel insights
in neonates born via sPTB and TB. PTB prevention remains elusive because the mechanisms
driving spontaneous preterm labor are poorly understood. Identification of differences
in gene expression between preterm and term fetuses will lay the groundwork for future
prospective studies to delineate mechanisms leading to sPTB, ultimately allowing development
of novel prevention strategies. For example, if decreased immune function in preterm
fetuses is a contributing factor to sPTB, novel strategies can be developed to bolster
the fetal immune system. These novel pharmacologic interventions can be tested in
animal models. This is the first of many steps necessary to better understand the
mechanism of sPTB and ultimately lead to effective preventive strategies.
