Abstract
Problem:
Current scientific guidelines recommend collecting placental specimens within two hours of delivery for gene expression analysis. However, collecting samples in a narrow time window is a challenge in the dynamic and unpredictable clinical setting, so delays in placental specimen collection are possible. The purpose of our analysis was to investigate temporal changes in placental gene expression by longitudinally sampling placentas over a 24 h period.
Method of Study:
Eight placentas from individuals with uncomplicated, term pregnancies delivered by scheduled cesarean section were collected and sampled following the placental delivery and again at 1, 2, 4, 6, and 24 h post-delivery. At each time point, biopsies of chorionic villous tissue were taken from 3 cotyledons to account for intra-placental heterogeneity. The 3 biopsies from each time point were pooled prior to RNA extraction. Expression of 382 mRNA transcripts was quantified using the NanoString nCounter System. Fold change values were calculated for each time point relative to delivery, and a fold change threshold of 1.25 was used to determine a meaningful change from delivery.
Results:
Based on a fold change threshold of 1.25, 84.3% of transcripts were stable for at least 1 h, 80.2% were stable for at least two hours, and 20.6% of transcripts were stable through the collection at 24 h.
Conclusion:
Our results suggest that for some mRNA transcripts, expression changes as time to sample collection increases. We have developed a Web application to allow investigators to explore transcripts relevant to their research interests and to set appropriate thresholds to aid in determining whether placentas with delayed sample collection can be included in analyses (https://placentaexpression.foundationsofhealth.org/).
Keywords: gene expression, mRNA, placenta, specimen handling
1 |. INTRODUCTION
Many studies have evaluated associations between placental gene expression and adverse birth outcomes in an attempt to identify potential mechanisms. Several characteristics of the placenta pose challenges to standardizing sample collection for gene expression analysis. For example, gene expression patterns vary across placental sites due to the clonal nature of placental development.1 Gene expression has also been shown to vary by sample depth and location, with one study reporting greater expression of hypoxia-related genes along the periphery of the placenta.2 Given this multifaceted variation, pooling samples from multiple sites is recommended when analyzing expression patterns.3 Challenges that are harder to over-come include the unpredictability of delivery time and the occurrence of complications. Both of these factors can result in variation in the timing of sample collection relative to actual delivery, introducing noise and/or bias into gene expression measurements.3 To address this problem, several studies have investigated time-related changes in placental gene expression, aiming to identify optimal windows for specimen processing.
Initial studies indicated that RNA quality, a global measure reflected in RNA integrity number (RIN), may be stable over time. Fajardy et al. (2009) evaluated RIN in 14 placentas sampled immediately following term delivery and then every 24 h for 96 h4 The authors reported that RIN was stable for 48 h for both cesarean and vaginal deliveries. Jobarteh et al. (2014) also evaluated whether delays in sample collection and processing impact RIN.5 Placentas from 300 participants were placed on ice immediately following delivery and transferred to the laboratory. The longest time between delivery and sample collection was 300 min, and there was a significant negative correlation between processing time and RIN, though this study design did not account for variation across placentas.
While RIN captures the overall integrity of an RNA sample, it is uninformative about gene-to-gene variation in rates of degradation.6,7 Also, degradation is not the only potential consequence of delayed specimen processing. Many transcriptional pathways that contribute to adverse pregnancy outcomes, such as inflammation, oxidative stress, and hypoxia, are likely to be up-regulated as specimens await processing.8,9 RIN changes do not capture this process. Recognizing these limitations, several studies have evaluated changes in mRNA expression patterns rather than RIN. Two studies evaluated a small number of transcripts, including TNF-α, COX2, CDH1, CDH11, KISS1, PLAC1, and ID2, and reported that some transcripts were positively correlated with time, some were negatively correlated with time, and some did not change over time.4,9 The best data to date on longitudinal changes in placental mRNA expression come from Wolfe et al. (2014).10 Three placentas from individuals with healthy term births that delivered via scheduled cesarean section were obtained, and samples were collected at 0, 60, and 120 min following delivery. The authors reported that expression levels of ≥94.7% of genes did not change across time based on p-values and recommend collection within two hours of delivery for gene expression studies.
Collectively, these studies form the basis for the recommended sampling within two hours of delivery. However, the largest studies are based on overall RNA integrity, which, as noted above, does not provide sufficient information for developing effective guidelines. The only study to evaluate changes in specific genes with repeated sampling used just three placentas and did not continue sample collection beyond two hours. In busy clinical settings when the clinical team is involved in some or all of the sample processing, collection may occur more than two hours after delivery. In particular, specimen processing may be at greater risk of delay when a patient experiences clinical complications at the time of delivery; arguably, these are the cases for which gene expression data may be most critical to understanding pathways to poor clinical outcomes. Thus, there is a need for research that characterizes changes in the expression patterns of specific transcripts across a clinically feasible time period, which was the purpose of our study. We investigated temporal changes in transcription by longitudinally sampling placental villous tissue over a 24 h period.
2 |. METHODS
2.1 |. Study sample
Eight placentas were collected from deliveries at NorthShore University HealthSystem, Evanston Hospital. Placentas were eligible for inclusion if the delivery occurred via scheduled cesarean section at term (≥37 weeks) and if there were no known pregnancy complications. Using placentas from scheduled cesarean sections helps to reduce some of the observed variation in gene expression related to characteristics of labor, such as compression resulting from con-tractions.3,11 This study used de-identified biospecimens. Our IRB ruled that because the de-identified biospecimens cannot be linked to individuals, the study was not considered to be research involving human subjects. Thus, the study was excluded from IRB review and did not require consent.
2.2 |. Sample collection and RNA expression
Following delivery, a biopsy of chorionic villous tissue was sampled from three separate cotyledons. Sampling from multiple cotyledons helps to account for variation in gene expression across placental sites.3 Each biopsy was obtained by making a small incision in the chorionic plate and removing a 0.4 cm3 piece of villous tissue. Biopsies were rinsed in buffer (PBS), stored individually in RNAlater, and placed in a refrigerator at 4°C for 48 h. Following this, samples were frozen and stored at −80°C until analysis, based on the manufacturer’s instructions. Subsequent biopsies were taken one hour, two hours, four hours, six hours, and twenty-four hours following the initial collection at delivery. In between sampling, the placentas were stored in sealed containers and placed in a refrigerator at 4°C.
Prior to RNA extraction, three biopsies from each collection time were pooled and homogenized. First, biopsies were thawed and placed in cryomolds with OCT compound. All three biopsies were placed into one mold ensuring each biopsy was equally represented when the block was cut. The 48 total blocks (8 placentas, 6 collection times per placenta) were kept frozen at −80°C until extraction. On the day of extraction, each block was sectioned into 15 pieces of 25 um sections and placed into an RNase/DNase-free tube. RNA was then extracted using Norgen’s Animal Tissue RNA Purification Kit (Norgen Biotek), according to the manufacturer’s protocol. RNA purity and concentration were measured using a NanoDrop ND-1000 (Thermo Fisher Scientific), and 260/280 ratios were all ≥1.99.
Transcript expression was quantified with the NanoString nCounter System (NanoString Technologies, Inc.) using a custom panel of 382 mRNA probes. We selected mRNA targets based on two criteria: (A) there was sufficient variation in the transcript in a prior study of 86 individuals (encompassing >10% of the global range) and/or (B) the transcript had reported associations with adverse birth outcomes in the literature. Of selected mRNA targets, 70% were variable across samples in the prior study and 65% were associated with birth outcomes in the literature (35% satisfied both criteria). The NanoString nCounter System uses unique color-coded probes that act as a barcode to identify and count individual transcripts without reverse transcription or amplification. The platform performs similar to real-time PCR (R2=0.95) and is more sensitive than microarrays.12,13 The nSolver Analysis Software (NanoString Technologies, Inc.) was used for data processing, normalization, and evaluation of differential expression.14 Raw data were normalized to reference genes identified based on the geNorm algorithm15 as implemented by the nSolver Analysis Software. Coefficients for differential expression were determined based on an algorithm that identifies the optimal model (“Optimal Method” in the nSolver Analysis Software). A negative binomial mixture model is used for low-expression probes, a simplified negative binomial model is used for high-expression probes, and if the models fail to converge, a log-linear model is used.14
2.3 |. Statistical analysis
Differential expression was determined by comparing each collection time with the collection at delivery, adjusting for placenta ID to account for correlation due to repeated sampling. Several fold change thresholds were used to identify a meaningful change in expression. Results for a fold change of 1.25 are reported in the main text, and results based on fold changes of 1.1 and 1.5 can be found in the supplemental material. Patterns were determined similar to the methods used by Functional Heatmap.16 Briefly, change in expression for each transcript at each collection was classified as increased (+), decreased (−), or no change (0) relative to the collection at delivery based on the appropriate fold change threshold. Determinations for the five post-delivery collections were summarized into a string. For example, “+++++” indicates a transcript was increased at 1, 2, 4, 6, and 24 h relative to delivery and “000--” indicates that a transcript did not change from delivery levels for the first three collections and decreased for the last two collections. Expression patterns were used to group transcripts with similar changes over time. Patterns that were present in ≥10 transcripts were grouped for analysis, and patterns present in <10 transcripts were grouped as variable.
We additionally evaluated stability over time based on average change in expression relative to delivery across all placentas. Stability was determined based on leading 0s in the pattern, which indicate no change in average expression relative to delivery. For example, patterns beginning with “0” were considered stable for the one-hour collection, patterns beginning with “00” were considered stable for the collections through two hours, and patterns beginning with “+” or “−” were considered unstable. As the goal of the analysis is to identify subtle changes in expression, results are based on fold change thresholds rather than statistical testing. The latter can be especially misleading in studies with small Ns like ours, because p-values are highly dependent on sample size. However, readers interested in p-values can obtain them from raw data in our Web application (https://placentaexpression.foundationsofhealth.org/). Analyses were conducted using the nSolver Analysis Software (NanoString Technologies, Inc.) and R version 4.0.
3 |. RESULTS
The first placenta sample collection occurred an average of 27 min after the delivery of the neonate (range: 15–36 min). All samples satisfied NanoString’s quality control metrics, which assess binding density, background noise, replicate variability, and optical problems. Using geNorm, eight reference genes were identified based on consistent expression across placentas and collection times. The reference genes, in order of selection, were as follows: CAPZB, BUB3, YTHDF1, POLR2G, GPN1, TIMM23, MTMR14, and TTC4. Descriptive statistics (Table S1) and pairwise variation (Figure S1) of the selected reference genes can be found in the supplemental material. Of the 382 probes evaluated, 18 fell below background levels too frequently and were excluded, leaving 364 for the analysis.
Based on a fold change threshold of 1.25, eight patterns containing at least 10 transcripts were identified (Figure 1). Three of the patterns (90 total transcripts, 25%) reflect a trend of increasing expression relative to delivery, four patterns (129 total transcripts, 35%) reflect a trend of decreasing expression relative to delivery, and one pattern reflects no change in expression relative to delivery (75 transcripts, 21%). 70 transcripts (19%) had expression patterns that were uncommon (<10 transcripts). Patterns are depicted in greater detail in Figure 2. Figures S2 and S3 depict patterns observed when more conservative (1.1) and less conservative (1.5) fold change thresholds were applied.
FIGURE 1.
Expression patterns over time based on a fold change threshold of 1.25. N indicates the number of transcripts per pattern (total n = 364). Patterns with fewer than 10 transcripts are grouped in solid gray
FIGURE 2.
Log fold change plotted over time and grouped by pattern (n = 364 transcripts). In pattern notation, “0” indicates no change relative to delivery, “+” indicates an increase, and “−” indicates a decrease. Patterns determined based on a fold change threshold of 1.25 indicated by dashed gray line (log2 fold change = 0.32). Patterns with fewer than 10 transcripts are grouped as variable
When transcripts were grouped based on stability over time, we identified that 84.3% were stable and 15.7% were unstable at the 1 h collection using a fold change threshold of 1.25 (Figure 3). Further, 80.2% were stable for at least 2 h, 61.2% were stable for at least 4 h, 40.1% were stable for at least 6 h, and 20.6% were stable through the 24 h collection. In contrast, when we set the fold change threshold to 1.1 only 61.8% of transcripts were stable at the 1 h collection, whereas with a fold change threshold of 1.5, 92.9% of transcripts were stable at 1 h (Figure 3). While a majority of transcripts exhibited low variability across placentas within a collection time (coefficient of variation [CV] <10%), some transcripts exhibited greater variability (5% of transcripts had a CV >20%), which may limit interpretation of fold change estimates (Figure S4).
FIGURE 3.
Percentage of transcripts with stable expression at each collection time (n = 364 transcripts). Each bar represents a different fold change threshold. The different shades of blue indicate how long the transcripts were stable for, with the darkest blue indicating the percentage of transcripts that were stable through 24 h and the lightest blue indicating the percentage of transcripts that were stable for only 1 h. The gray portion indicates the percentage of transcripts that were unstable at the first collection following delivery
To enable investigators to apply varying thresholds for a meaningful difference and to explore temporal trends in specific transcripts pertinent to their research interests, we have created an interactive Web application, which can be accessed at https://placentaexpression.foundationsofhealth.org/. Investigators can specify fold change thresholds, and the number of transcripts need to determine a pattern to evaluate trends. Expression patterns for individual transcripts can also be plotted, and all log2 fold change values are available in a table.
4 |. DISCUSSION
To our knowledge, this is the largest study to evaluate changes in the expression of specific placental mRNAs over time and the most extensive in terms of sampling. Our results suggest that degradation is not uniform across all transcripts. While the most common pattern reflected a decrease in expression over time (35.4%), a substantial proportion of transcripts exhibited an increase in expression over time (24.7%). When categorized based on stability at each collection time relative to delivery, our results suggest that 84.3% of transcripts were stable for at least one hour and 61.3% were stable for at least six hours. Only 20.6% of transcripts were stable across the 24 h period.
Our results are consistent with previous (smaller) studies describing changes in expression over time. Similar to Avila et al. (2010), we observed decreasing trends for PLAC1 and CDH1. KISS2 exhibited a variable trend (+00--), but declined in the last two samples, consistent with prior work. Wolfe et al. (2014) collected samples from three third-trimester placentas across two hours and reported that 94% of transcripts were stable. In contrast, we observed that 80.2% of transcripts were stable for two hours. Our results likely differ due to the use of a different metric to define stability (fold change threshold vs. p-value) and the selection of transcripts that are highly variable and/or associated with birth outcomes for the analysis. Given the selection of highly variable transcripts, our results may underestimate stability relative to all transcripts expressed by the placenta.
Our results also build on prior work evaluating degradation by quantifying trends in specific transcript expression over time. While correction for RNA integrity using RIN is a common method for adjusting for sample quality in analyses, this approach does not fully solve the problems associated with delayed sample processing, because as we show above, some transcripts decline while others increase.7,17 One proposed framework to account for differences in sampling involves using an experimental dataset of measurements over time to identify the transcript features most susceptible to change.7 This information can then be used to correct for time-related variation. Thus, our data may be useful as a reference for investigators who wish to apply this framework in their own studies of placental mRNA expression using the nCounter platform.
Strengths of this analysis include the use of biologic replicates and the collection of samples across an extended period of time. We also collected samples from three different cotyledons at each time point to account for intra-placental heterogeneity. Limiting the analysis to placentas from term, scheduled cesarean sections also reduces variation attributable to gestational age and method of delivery. We were also able to collect the first sample an average of 27 min after delivery of the neonate (time elapsed between delivery of the placenta and initial specimen collection is likely shorter). Another strength of our analysis is the use of multiple fold change thresholds to determine stability. We have also made the data publicly available to allow investigators to examine the patterns of specific transcripts of interest and to apply their own criteria for stability.
The limitations of this analysis include the lack of clinical data. Due to the use of de-identified biospecimens, which is not classified as human subjects’ research and does not require consent, we were unable to obtain data on delivery or neonatal characteristics. Some characteristics, such as clinical history, type of anesthesia, or fetal sex, may affect gene expression. However, even if these characteristics modify expression in general, it is not apparent they would also modify changes in expression over time, which is the focus of this analysis. Similarly, our analysis was restricted to term, scheduled C-sections, and results may not generalize to other delivery conditions. Estimates of stability based on fold change may also be misleading for transcripts with high variability across placentas. Additional limitations of the analysis include the selection of clinically relevant and/or highly variable transcripts, which may not be representative of all transcripts expressed by the placenta, and the use of the nCounter platform, which may not generalize to other platforms. With that said, nCounter assay directly measures mRNA, without a reverse transcription step. Those features are likely to make its results more sensitive and replicable than transcriptome-wide methods that use sequencing.12,13
Our results suggest that while the expression of some transcripts is stable for 24 h, the expression of other transcripts changes as time to sample collection increases. We hope that our findings can serve as a resource for investigators, as timing of delivery is often unpredictable and collection of placental tissue is dependent on availability of research or clinical staff. While our results are consistent with prior work indicating that collection within two hours is optimal, our results also suggest that many transcripts are stable over a longer period of time. Depending on the specific transcripts and biologic pathways of interest, it may not be necessary to exclude samples collected beyond two hours of delivery. Future work should investigate whether clinical and neonatal characteristics, such as method of delivery, pregnancy complications, and fetal sex, impact changes in expression over time.
Supplementary Material
ACKNOWLEDGEMENTS
We would like to thank MaryAnn Regner (NorthShore University HealthSystem) for processing the placental specimens and Miles Pfefferle for assistance with the Web application.
Funding information
This work was supported by The National Institute on Minority Health and Health Disparities (award number R01MD011749) and The Eunice Kennedy Shriver National Institute of Child Health & Human Development (award number F32HD100076)
Footnotes
CONFLIC T OF INTEREST
None declared.
DATA AVAILABILITY STATEMENT
The data that support the findings of this study are openly available at: https://placentaexpression.foundationsofhealth.org/
SUPPORTING INFORMATION
Additional supporting information may be found online in the Supporting Information section.
REFERENCES
- 1.Penaherrera MS, Jiang R, Avila L, Yuen RK, Brown CJ, Robinson WP. Patterns of placental development evaluated by X chromosome in-activation profiling provide a basis to evaluate the origin of epigenetic variation. Hum Reprod. 2012;27(6):1745–1753. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Wyatt SM, Kraus FT, Roh CR, Elchalal U, Nelson DM, Sadovsky Y. The correlation between sampling site and gene expression in the term human placenta. Placenta. 2005;26(5):372–379. [DOI] [PubMed] [Google Scholar]
- 3.Burton GJ, Sebire NJ, Myatt L, et al. Optimising sample collection for placental research. Placenta. 2014;35(1):9–22. [DOI] [PubMed] [Google Scholar]
- 4.Fajardy I, Moitrot E, Vambergue A, Vandersippe-Millot M, Deruelle P, Rousseaux J. Time course analysis of RNA stability in human placenta. BMC Mol Biol. 2009;10:21. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Jobarteh ML, Moore SE, Kennedy C, Gambling L, McArdle HJ. The effect of delay in collection and processing on RNA integrity in human placenta: experiences from rural Africa. Placenta. 2014;35(1):72–74. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Reiman M, Laan M, Rull K, Sober S. Effects of RNA integrity on transcript quantification by total RNA sequencing of clinically collected human placental samples. FASEB J. 2017;31(8):3298–3308. [DOI] [PubMed] [Google Scholar]
- 7.Jaffe AE, Tao R, Norris AL, et al. qSVA framework for RNA quality correction in differential expression analysis. Proc Natl Acad Sci U S A. 2017;114(27):7130–7135. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Wu G, Mannam AP, Wu J, et al. Hypoxia induces myocyte-dependent COX-2 regulation in endothelial cells: role of VEGF. Am J Physiol Heart Circ Physiol. 2003;285(6):H2420–H2429. [DOI] [PubMed] [Google Scholar]
- 9.Avila L, Yuen RK, Diego-Alvarez D, Penaherrera MS, Jiang R, Robinson WP. Evaluating DNA methylation and gene expression variability in the human term placenta. Placenta. 2010;31(12):1070–1077. [DOI] [PubMed] [Google Scholar]
- 10.Wolfe LM, Thiagarajan RD, Boscolo F, et al. Banking placental tissue: an optimized collection procedure for genome-wide analysis of nucleic acids. Placenta. 2014;35(8):645–654. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Cindrova-Davies T, Yung HW, Johns J, et al. Oxidative stress, gene expression, and protein changes induced in the human placenta during labor. Am J Pathol. 2007;171(4):1168–1179. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Geiss GK, Bumgarner RE, Birditt B, et al. Direct multiplexed measurement of gene expression with color-coded probe pairs. Nat Biotechnol. 2008;26(3):317–325. [DOI] [PubMed] [Google Scholar]
- 13.Malkov VA, Serikawa KA, Balantac N, et al. Multiplexed measurements of gene signatures in different analytes using the nanostring nCounter assay system. BMC Res Notes. 2009;2:80. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.nCounter Advanced Analysis 2.0 User Manual. In: Seattle, Washington: NanoString Technologies, Inc; 2018: https://www.nanostring.com/download_file/view/1169/3843 [Google Scholar]
- 15.Vandesompele J, De Preter K, Pattyn F, et al. Accurate normalization of real-time quantitative RT-PCR data by geometric averaging of multiple internal control genes. Genome Biol. 2002;3(7):RESEARCH0034. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Williams JR, Yang R, Clifford JL, et al. Functional heatmap: an automated and interactive pattern recognition tool to integrate time with multi-omics assays. BMC Bioinformatics. 2019;20(1):81. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Gallego Romero I, Pai AA, Tung J, Gilad Y. RNA-seq: impact of RNA degradation on transcript quantification. BMC Biol. 2014;12:42. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.