Evaluating the Utility of RNAseq in Prenatal Diagnostics: Expression Profiles of Cultured Chorionic Villus and Amniotic Fluid Samples

ABSTRACT

Objective

While RNAseq has enhanced variant interpretation in postnatal cases, its potential in the prenatal setting remains underexplored. This study investigates the utility of RNAseq in prenatal diagnostics by analyzing the expression profiles of cultured chorionic villus samples (cCVS) and amniotic fluid (cAF) samples.

Methods

We performed RNAseq on 25 prenatal samples (10 cCVS and 15 cAF) and compared their expression profiles with those of postnatal tissues—blood and skin fibroblasts.

Results

To evaluate the clinical relevance of gene expression in these samples, we curated a list of genes associated with fetal‐onset genetic disorders (n = 375). Using this curated list as a reference, our analysis revealed that cAF samples have the highest proportion of highly expressed genes (60%), surpassing cCVS (54%), fibroblasts (54%), and blood (34%). Differential expression analysis identified unique gene expression patterns in cCVS and cAF samples, reflecting their distinct tissue origins. Specifically, cAF samples showed elevated expression of genes involved in fetal kidney development, whereas cCVS samples were enriched for genes related to trophoblast function.

Conclusions

These findings demonstrate that prenatal RNAseq reliably detects clinically relevant gene expressions, offering insights into prenatal conditions and fetal organ development. This underscores the potential of prenatal RNAseq as a valuable tool for supporting genetic variant interpretation.

Key Summary

• What is already known about this topic?

  • ◦Prenatal RNAseq is an emerging NGS methodology in the clinical diagnosis space. Despite its mainstream utilization in research, few studies are investigating prenatal RNAseq’s utility in clinical diagnosis due to many challenges such as the lack of availability of comprehensive normal control databases for fetal tissues.

• What does this study add?

  • ◦Our study investigates the establishment of expression level thresholds in prenatal samples, tissue types that are commonly used in clinical genetic testing. Additionally, it examines whether genes associated with prenatal disorders are accurately represented in these prenatal samples which can help guide clinical decision making for healthcare providers.

Abbreviations

ACMG

American College of medical Genetics

cAF

cultured amniocytes

cCVS

cultured chorionic villi

CMA

chromosome microarray

CNV

copy number variant

ES

exome sequencing

FISH

fluorescence in situ hybridization

GS

whole genome sequencing

GTEx

Genetype‐Tissue Expression

HPO

Human phenotype ontology

INDEL

insertion and deletion

NGS

next generation sequencing

OMIM

Online mendelian inheritance in man

RNAseq

RNA sequencing

SNP

single nucleotide polymorphism

TPM

transcripts per million

VUS

variant of unknown significance

1. Introduction

Implementation of next‐generation sequencing (NGS) assays such as exome and whole genome sequencing (ES and GS) after non‐diagnostic cytogenetic results has considerably increased the incremental diagnostic yield of prenatal diagnostic testing, particularly for monogenic disorders and/or after negative cytogenetic results. While ∼25% of fetal abnormalities are diagnosed prenatally by conventional cytogenetic assays (e.g., fluorescence in situ hybridization‐FISH, karyotyping and chromosomal microarray), ES has provided an incremental diagnostic yield of up to 30% in some studies, particularly in patients with skeletal dysplasia and multiple congenital abnormalities [1, 2, 3, 4, 5, 6]. Similarly, GS has also demonstrated promising diagnostic yields of up to ∼19% in prenatal samples while expanding the detection repertoire to mitochondrial, intronic, copy number variants and balanced structural rearrangements [7, 8, 9, 10]. With the advent of NGS technologies, particularly GS, there is a pressing need for functional assays to promote accurate interpretation of the increasing number and spectrum of variants detected in prenatal samples [8].

High‐throughput transcriptome profiling or RNAseq has recently shown great promise to support variant interpretation in postnatal sample types such as whole blood and skin fibroblasts [11, 12, 13, 14, 15, 16, 17, 18, 19]. While whole blood is a more frequently used sample type for RNAseq due to its ease of access, studies have shown that skin fibroblasts express a larger proportion and higher levels of clinically relevant genes [18, 19]. Detection of abnormal gene expression, mono‐allelic expression or aberrant splicing in these sample types seen by RNAseq increased the molecular diagnostic yield by up to ∼8–36%. This is often achieved by providing evidence to reclassify variants of unknown significance (VUS) detected by ES or GS, particularly variants around the exon‐intron junctions, deep intronic variants, or variants within non‐coding regions with potential to affect gene expression (e.g., promoter regions). Beyond serving as a supporting functional assay, because of its ability to survey the whole transcriptome, RNAseq also has the potential to agnostically detect abnormalities within expression profiles, similar to genome‐wide DNA methylation assays [20].

Despite its increasing popularity as a functional assay among postnatal samples, there are very few studies investigating RNAseq in a prenatal setting and fewer investigating its clinical utility in prenatal diagnostics [21]. Reference datasets are also scant for prenatal sample types, rendering the data analysis of prenatal RNAseq challenging. For instance, large RNAseq reference datasets such as GTEx contain only adult tissue‐specific expression profiles, while a few fetal single‐cell RNAseq reference databases are only just emerging [22, 23].

Given the limited availability of RNAseq data from clinically accessible prenatal samples, we investigated the expression profiles of cultured chorionic villus sampling (cCVS) and cultured amniotic fluid (cAF) samples. Specifically, we examined and compared the expression profiles of prenatal samples against postnatal sample types, assessed if genes that have been associated with distinct prenatal phenotypes are reliably and reproducibly detected, and compared prenatal expression profiles with those of tissue‐specific fetal and adult organs. We also aimed to reveal insights into prenatal expression profiles unique to cCVS and cAF sample types that reflect developmental processes which when disrupted during fetal life, could cause or be associated with prenatally detectable diseases.

2. Material and Methods

2.1. Clinical Samples

Excess materials from clinical cytogenetic testing at the Baylor Genetics diagnostic laboratory, including 10 cCVS and 15 cAF samples, were deidentified and included in this research study. Ten days after clinical cytogenetic diagnostic results were released, the cultured residual samples were processed for RNAseq studies. Prenatal clinical samples were referred for diagnostic genetic testing based on abnormal 1^st trimester screening findings as Described in Table S2‐Demographics‐Referral information. Control postnatal samples which included 10 uncultured peripheral blood leukocytes and 10 cultured skin fibroblast samples from pediatric patients were obtained from the Undiagnosed Diseases Network cohort [24]. Postnatal samples were selected to achieve sex and age concordance between the blood and the skin fibroblast groups.

2.2. MCC Contamination Analysis

Maternal cell contamination (MCC) was assessed using an RNA‐seq–based noise profiling approach across a targeted panel of 507 common single‐nucleotide polymorphisms (SNPs) with minor allele frequency (MAF) > 0.3. Read pileups were generated using samtools mpileup, and for each locus the major allele was identified. A noise variant allele frequency (VAF) was calculated as the fraction of non–major‐allele reads divided by the total read depth.

To ensure robustness and minimize false positives, loci with total depth < 30× or evidence of heterozygosity were excluded. Recurrent noisy loci—defined as SNPs present in ≥ 5 samples with a mean noise VAF ≥ 2%—were also excluded to reduce the influence of systematic mapping or sequencing artifacts. The remaining loci therefore represent homozygous‐like sites where minor allele reads are unexpected in the absence of contamination.

For each sample, two metrics were calculated: (1) the fraction of discordant SNPs, defined as homozygous‐like loci with detectable minor allele signal (VAF > 0.1%), and (2) the mean noise VAF across these discordant loci. Samples were classified as positive for MCC only if both criteria were met: ≥ 30% of discordant SNPs showing minor allele signal and a mean noise VAF exceeding 4%. These thresholds were selected conservatively to exceed expected background sequencing and mapping noise and to identify contamination levels likely to be biologically meaningful. A simulated 10% contamination control was included to verify assay sensitivity.

2.3. RNAseq

RNA from cCVS, cAF, whole blood and cultured skin fibroblasts was extracted and processed using a stranded, polyA‐tailed kit (Illumina) prior to multiplexing and 150 bp paired‐end sequencing on NovaSeq 6000 or NovaSeq X, with ∼95 million read depth per sample (Table S2). The sequencing data were processed with the standard GTEx v10 pipeline (https://github.com/broadinstitute/gtex‐pipeline), specifically fastq files were aligned to the reference genome GRCh38 with STAR v2.7.8a_sentieon and SAMtool 1.15.1/HTSlibv1.10.2. Picard v2.23.3 was used to mark duplicates. Gene expressions were quantified by RNA‐SeQC v2.4.2. Isoform‐level quantification was calculated by RSEM v1.3.3. Transcripts were annotated with GENCODE v39. FastQC v0.11.9 was used to generate quality control measurements.

2.4. Identifying Genes Associated With Prenatally Detectable Phenotype

To identify genes associated with known prenatal phenotypes, we combined gene lists from multiple curated sources, and performed a review of the primary literature. First, we accessed the most recent signed‐off fetal anomalies list from PanelApp (v3.0; accessed October 7, 2022), a repository of curated genotype‐disease associations. Genes are classified as “green” (diagnostic‐level evidence), amber (some evidence), and red (poor or lacking evidence). We considered genes from the green and amber gene sets, for a total of 1538 loci. Second, we searched the Human Phenotype Ontology (HPO) database with the term “Abnormality of prenatal development or birth” (HP:0001197), the highest‐level HPO term that relates to fetal anomalies, for a total of 894 genes. Third, we queried the Online Mendelian Inheritance in Man catalog (OMIM) using the terms “fetal” and “prenatal”. These searches were performed on October 7, 2022 and yielded 501 and 585 genes, respectively. In our review, we used search terms that were variations of the key words such as “prenatal diagnosis” and “exome sequencing,” and then included studies which were of sufficient size (≥ 10 pregnancies), had a non‐diagnostic or negative CMA and/or karyotype, and which had initiated testing based on a prenatal phenotype. Using this search strategy, we identified 66 papers that detail prenatal phenotype‐genotype correlations. Of these, we prioritized studies describing 15 or more unique genotype‐phenotype correlations, and a total of 27 papers were ultimately included (Table S1). The list of studies contributing to establishing a list of genes associated with the prenatal phenotype is provided in Table S4 and combined for a total of 375 genes. We stratified this gene list by affected organ system, including the following categories: Musculoskeletal, Digestive, Urogenital, Craniofacial, Hydrops, Stillbirth, Multisystem, Lymphatic, Inborn Errors of Metabolism, Fetal Akinesia, Neurodevelopmental, Hematologic, Immunologic, Ciliopathy, Overgrowth Disorders, and Secondary/Incidental Findings. Depending on the phenotype associations, a single gene may be present in multiple phenotype categories.

2.5. Analysis of Gene Expression Levels

Hierarchical clustering and differential gene expression analysis (Wald test p‐value) were performed on raw counts using the standard DeSeq2 normalization and analysis pipeline (v.1.38.3/R.4.2.3). Gene rank was generated by multiplying the log₂FC with ‐log₁₀(p‐value) [25]. Gene expression levels were primarily delineated based on transcripts per million (TPM) value, complemented by splice junction counts for genes with low expression. This approach was employed to minimize interference from overlapping genes and background noise, ensuring a more accurate representation of gene expression. Categories were ascertained based on values described in prior publications [13, 14, 15, 19, 26, 27] and defined as high (median TPM > 10), intermediate (median 1 ≤ TPM ≤ 10), low (median 0.1 ≤ TPM < 1), very low (median TPM < 0.1 & total junction reads across the whole gene > 0), not detected (ND) (median TPM < 0.1 & 0 total junction reads across the whole gene). The percentage of genes expressed was represented based on one of the following:

The total number of annotated protein coding genes (n = 19,313), prenatal phenotypes associated genes frm literature cohorts (n = 375), ACMG's v3.1 secondary findings genes (n = 78) [28], disease‐associated repeat expansion genes (n = 27), disease‐associated mitochondrial genes (n = 176, MitoMap Baylor Genetics), imprinted genes (n = 80) [29], neurodevelopment associated methylation genes (n = 50) [30], PanelApp genes (n = 1500), OMIM genes (n = 362), HPO genes (n = 878) (Table S4).

2.6. Differentially Expressed Genes in Human Embryonic or Adult Organs

Differentially expressed genes (i.e., genetic markers) in each human embryonic organ were obtained from a previous study [23]. Primary and secondary genetic markers are listed in Table S8. Highly expressed genes (top 1000) in each human adult organ were obtained from the GTEx Portal on 08/30/2023 (https://www.gtexportal.org/home/; Table S9). For the genetic markers of each embryonic or adult organ, we tested the proportion of the markers that are adequately expressed in prenatal samples (cAF or cCVS) or control samples (blood or fibroblast), defined as TPM > 5 or TPM > 10.

2.7. Copy Number Analysis

The read count for each gene from the RNAseq data was calculated using RSEM. Single nucleotide variants from RNAseq data were called using Sentieon. Copy number analysis was performed using RNAseqCNV.

3. Results

A total of 25 prenatal samples were collected for this analysis with abnormal first trimester screening indications, including 10 cCVS and 15 cAF (Figure 1A). cCVS samples were from CVS procedures performed between 11.5 and 13.9 weeks gestation, while cAF samples were from amniocenteses performed across a wider gestational age range, between 16.5 and 32.2 weeks gestation (Table S2). Maternal cell contamination (MCC) analysis showed no evidence of significant contamination, indicating minimal bias in downstream gene expression analyses (Table S2). Diagnostic chromosomal abnormalities were identified in 5 of 25 samples (20%) using fluorescence in situ hybridization (FISH), conventional karyotyping, or chromosomal microarray analysis (Figure 1B). These results were either full trisomies (Trisomy 21 and 13) or Monosomy X. For the remainder 20/25 (80%), the testing results were classified as non‐diagnostic. These included samples without chromosomal abnormalities detected and samples with variants of unknown significance with no diagnostic implications, including one copy number loss (∼0.6 kb), two copy number gains (∼0.039 Mb, 134 Mb) and one with an 11Mb region of absence of heterozygosity (Table S2).

FIGURE 1.

Clinical and expression profile characterization of prenatal and postnatal samples. Clinical cytogenetic diagnostic spectrum of prenatal samples by sample type (A) and cytogenetic diagnosis (B) (n = 25). Principal component analysis (PCA) plot based on the expression profiles of cultured amniocytes (n = 15), cultured CVS (n = 10), fibroblasts (n = 10), and whole blood (n = 10) samples (C). NA, not available; employed only for UDN blood and fibroblast samples with no available diagnoses in the current study. The corresponding distance matrix and hierarchical clustering plots are shown in Figure S1A and B.

Clustering analysis based on RNAseq data of prenatal and postnatal samples demonstrated distinct expression profiles primarily driven by sample types, as shown on the PCA plot (Figure 1C, Figure S1A and B). This result is consistent with previously reported findings [19, 21]. Interestingly, cCVS and fibroblast samples clustered closely together, supporting gene expression similarities between these two sample types. This may be explained by the enrichment of fetal mesenchymal populations in cCVS samples [31, 32], which share strong resemblance to fibroblast cells [33, 34].

Gene expression levels of protein‐coding genes were comparable across sample types, which aligned with previously reported data, particularly for blood and fibroblasts. Overall, the expression levels were high for 34%–44%, intermediate for 19%–25% [19, 26], low for 10%–13%, and markedly low for 10%–13% of protein‐coding genes (Figure 2A–D, Table S3). Only 15%–20% of the genes remained undetected. There was a tendency for more highly expressed genes (44%) in cAF samples, followed by cCVS and fibroblasts (40% and 39% respectively), and blood samples (34%). Interestingly, the combined ratio of highly and lowly expressed genes‐‐ defined as 1 minus the percent of undetectable genes‐‐ is higher in cAF (85%) and cCVS (82%) compared to blood and fibroblasts (both at 80%) (Table S3). This suggests a greater potential in cAF and cCVS for improved clinical utility when RNAseq can be performed at ultra high depth, allowing for enhanced detection of low and very low expressed genes [35]. Consistent with these findings, diagnostic genes identified from ES/GS studies displayed a similar trend in expression levels in cAF and cCVS samples (Table S11).

FIGURE 2.

Percentage of gene expression levels are comparable across sample types. Gene expression levels were defined based on TPM values and categorized as high (median > 10 TPM), intermediate (median < 10 & ≥ 1 TPM), low (median < 1 & ≥ 0.1 TPM), very low (median < 0.1 TPM & > 0 total junction reads), not detected (ND) (median < 0.1 TPM & 0 total junction reads). Number of genes are represented as percentage out of the total number of protein coding genes (n = 19,313; gencodeV39) in cultured amniocytes (n = 15) (A), cultured CVS (n = 10) (B), fibroblasts (n = 10) (C), and blood samples (n = 10) (D).

Comparison to the expert‐curated gene list associated with prenatally detectable phenotypes (n = 375 genes, Table S4, Figure 3A) identified more highly expressed genes in cAF samples (60%) compared to cCVS (54%) and fibroblasts (54%). The percentage of highly expressed genes from this curated list was the lowest in blood (34%). Among the highly expressed genes in cAF samples were FLNA, COL4A1, and TUBA1, associated with multisystem, musculoskeletal, and central nervous system phenotypes, while in cCVS and fibroblast samples, they included COL3A1, COL1A2, LOX and RPL11, TUBB, FBN1, respectively, associated with stillbirth, musculoskeletal, and hydrops (Table S5, Table S10). In blood samples, highly expressed genes included HBA2, FKBP8, and STAT3. Overall, the trend that RNAseq on cAF samples captures more highly expressed genes was observed across the different categories of gene lists associated with prenatal phenotypes (PanelApp n = 1500, OMIM n = 362, HPO n = 878) (Figure 3A).

FIGURE 3.

Comparison of expression profiles between prenatal (n = 25) and postnatal (n = 20) samples with disease associated gene lists. Number of gene overlaps labeled between prenatal and postnatal samples labeled with expression levels and shown by prenatally detectable phenotype genes (n = 375), PanelApp genes (n = 1500), HPO genes (n = 878), and OMIM genes (n = 362) (A). Number of gene overlap between prenatal and postnatal samples categorized by clinically relevant diagnostic genes: repeat expansion genes (n = 26), mitochondrial genes (n = 176), methylation genes (n = 50), incidental genes (n = 78) and imprinting genes (n = 80) (B). Gene expression levels are defined as in Figure 2A–D. Gene lists are described in Table S4 based on the respective categories. Gene expression levels in prenatal and postnatal sample types are outlined in Table S5 for the prenatal phenotype gene category.

Next, we investigated the ability to detect expression levels of a variety of other clinically relevant diagnostic gene categories in prenatal and postnatal samples. These included genes associated with repeat expansion, nuclear encoded mitochondrial genes, genes associated with methylation and imprinting disorders, and American College of Medical Genetics and Genomics (ACMG) [28] secondary findings genes (Figure 3B). A similar pattern as described above was also observed in this analysis. These results highlight the ability of RNAseq on prenatal samples to adequately capture the expression of genes relevant to prenatal disorders, particularly in cAF samples where they appear to be better represented. These results also suggest that in the context of postnatal RNAseq testing, fibroblasts from a skin biopsy may be more suitable than blood when seeking expression data for genes associated with a prenatal phenotype.

We then conducted differential gene expression analysis between RNAseq data from cAF and cCVS samples and identified unique highly expressed genes in each sample type (Figure 4A, Table S6). Examination of a database of prenatally expressed genes [36] revealed that the top differentially expressed genes in cAF compared to cCVS are all highly and uniquely expressed in fetal kidneys, while the top differentially expressed genes in cCVS versus cAF tend to be highly expressed in trophoblast cells. Pathway analysis also revealed distinct biological pathways associated with these gene sets, including neurodevelopmental pathways in cAF samples and extracellular signaling pathways in cCVS samples (Figure 4B, Table S7). Furthermore, certain genes demonstrated elevated expression levels exclusively in either cAF or cCVS, suggesting that expression of these genes is dependent on the sample type analyzed (Figure 4C and D). The discovery of these unique genes through prenatal RNAseq demonstrates its potential to uncover novel gene variants specific to different sample types associated with prenatal conditions. This analysis also highlights the unique expression differences between sample types and holds potential to guide sample‐type selection for future functional validation solely based on the RNA detection levels independent of phenotype associations.

FIGURE 4.

Prenatal samples show distinct differentially expressed genes. Volcano plot contrasting highly expressed genes with log₂FC > 5 and p‐value < 10⁻³² in cultured amniocytes (n = 15) compared to cultured CVS samples (n = 10) (A). Comprehensive differential gene expression analysis results are described in Table S6. Tables highlighting the top 10 gProfiler biological process pathways scored on significance of p‐value shown in cultured amniocytes (B) and in cultured CVS samples (C). gProfiler analysis gene lists are described in Table S7 for both cAF and cCVS. Differential gene expression was conducted with DeSeq2. Gene ranking was generated from log₂FC x ‐log₁₀(p‐adj) and the top 5% of genes were used for ranked analysis. Barplots showing expression levels of selected genes in cultured amniocytes (D) and cultured CVS samples (D). Normalized expression values are displayed on the y‐axis. The top differentially expressed genes in cAF are predominantly and uniquely expressed in fetal kidneys. The top differentially expressed genes in cultured CVS are generally highly expressed in trophoblasts.

We obtained gene expression profiles from fetal single‐cell RNAseq data, which we used as a reference for comparison with our samples. We then assessed the percentage of highly expressed genes overlap of fetal organ‐specific genes using prenatal and postnatal clinically non‐accessible tissues (Figure 5). In general, cAF samples demonstrated a larger proportion of detectable gene overlap (∼60–85%), specifically for fetal brain, cerebellum, kidney, lung, stomach, intestine, liver, eye, adrenal gland, and pancreas (Figure 5). cCVS demonstrated the highest level of gene overlap with placentas, and showed similar but slightly lower proportion of gene overlap compared with cAF for fetal heart and muscle. Blood samples demonstrated ∼95% gene overlap with fetal spleen likely due to the involvement of the spleen in hemoglobin synthesis and immune cell development during fetal life but overall, a trend toward lower overlap. These findings emphasize the ability of RNAseq in prenatal samples to detect genes relevant to organ development during both fetal stages, which ultimately may have clinical implications in disease diagnosis. Extending the comparative analysis into adult tissues based on gene expression data from GTEx yielded similar results (Figure S3).

FIGURE 5.

Comparison of gene expression profiles between prenatal (n = 25) and postnatal (n = 20) samples and clinically non‐accessible fetal single‐cell dataset. Differentially expressed genes in each human embryonic organ were obtained from Cao J et al., 2020 and used in the comparison of adequately expressed gene markers (TPM > 5) of prenatal samples (n = 21) or postnatal samples (n = 20). Bar plots represent the proportion of gene overlap.

4. Discussion

Between 90% and 92% of genes associated with prenatal disorders are detectable in prenatal samples at high, intermediate, and low expression levels. This high detection rate is promising since the majority of genes from the curated prenatal phenotypes‐associated gene list are relevant to prenatal phenotypes detectable by ultrasound and thus ultimately to prenatal diagnosis. Obstetricians/gynecologists, prenatal geneticists and genetic counselors who offer NGS based diagnostic tests in the prenatal diagnostic work‐up of pregnancies complicated by fetal structural anomalies will highly benefit from the clinical implementation of prenatal RNAseq, especially when encountering non‐coding VUSs that could benefit from RNA functional validation. Notably, because all samples in our study were obtained from clinically indicated pregnancies, they may not represent a fully normal prenatal transcriptome; nevertheless, since the data are intended to aid clinical interpretation of abnormal prenatal samples undergoing ES/GS, they remain valuable despite these limitations. A smaller percentage of other genes, encompassing those associated with repeat expansion, and mitochondrial, imprinting, and methylation disorders, were also detected. We should specifically acknowledge the detection of expressed ACMG secondary findings genes in 70% of cCVS and 77% of cAF. ACMG guidelines on secondary findings do not currently recommend that they be reported when detected during prenatal ES or GS [28]. Therefore, it is important to recognize the potential of identifying these secondary findings with prenatal RNAseq and testing laboratories and physicians should develop reporting standards and strategies to manage downstream implications of their detection.

We next compared the expressed genes in the prenatal samples to organ‐specific marker gene lists, to determine if these organ‐specific patterns are recognizable and well represented, as this will benefit the interpretation of consequences on organ development of variants in prenatal genes detected during prenatal sequencing. We found that cAF, cCVS, and fibroblasts are overall superior for detecting expressed genes than blood, with the exception of the spleen [37] and placenta [23] (Tables S8 and S9). Our comparative analysis of gene expression levels using curated disease genes with prenatally detectable phenotypes did not show significant enrichment of cAF or cCVS genes (Table S5). This may reflect a lack of statistical power due to the small number of defined disease genes with prenatal onset. Differential gene expression analysis highlighted that the cAF transcriptome most resembles that of fetal kidney cells, suggesting that cAF includes a considerable proportion of cells shed from fetal renal epithelium or contains precursor cells that can develop into fetal kidney cells. As expected, the top differentially expressed genes in cCVS are attributed to trophoblast cells. Comparing gene lists from clinically accessible NGS panels linked to developmental disorders affecting kidney [38] and cardiac mesenchyme [39] revealed increased gene overlap with prenatal samples compared with postnatal samples, particularly in cAF and kidney disorders (Figure S2A and B). Analysis against fetal organ‐specific single‐cell RNAseq data confirmed observations from the differential gene expression analysis. Additionally, fetal organs including the stomach, intestine, and lung were highly represented by gene expression in cAF, consistent with the notion that cAF contains cells shed from the epithelium of the lung and digestive tract. The strong representation of fetal brain gene expression in cAF cannot be explained by direct shedding of cells, but may suggest that amniotic fluid stem cells or mesenchymal stem cells in the cAF expand and acquire features pushing them toward the neural lineage [40]. Therefore, our findings suggest that cAF provides a broader detectable gene coverage, making it potentially a preferred sample type for RNAseq. However, the choice between performing RNAseq on cCVS or cAF remains complex, and further functional validation is required to assess the significance of these findings. Clinical teams must also consider additional factors, such as gestational timing, the availability and feasibility of follow‐up testing, the risk of maternal cell contamination, laboratory culture success rates, turnaround time, and the complexity of interpreting mosaic findings, among many other factors.

While amniotic fluid cell free RNA has been also been explored in the diagnostic space [40, 41], our study employed RNAseq on cultured, rather than direct, specimens of cCVS and cAF, as this approach will more closely mimic clinical practice (Figure 6). Furthermore, no additional sample volume was required during clinical sampling, as the cultures established by our clinical laboratory provided sufficient residual material for RNA sequencing following routine genetic diagnostic testing. Based on our results, RNAseq shows promise as a complementary tool alongside exome or genome sequencing, to maximize clinical utility. In instances where invasive sampling yields insufficient material for comprehensive analysis via chromosomal microarray, exome or genome sequencing, and RNAseq, the DNA‐based tests should be prioritized. Our results show that RNAseq may yield clinically useful results even after additional culturing. The practice of cell culturing may, in theory, reduce variability introduced during sampling by homogenizing the cell population and controlling the environmental conditions. However, extended culture time can introduce selection and/or cultural artifacts bias through clonal selection and gene expression drift with increasing passage number, thereby reducing cellular diversity and potentially affecting the interpretation of cell‐type specific expressed genes, X‐linked genes, or mosaicism. In addition, the small sample size (n = 25) and the wide range of gestational age among the cAF sample may have introduced further variability within the gene expression profiles. It is crucial to recognize the potential variation in cell type composition from various direct CVS and AF samples to fully understand the advantages and limitations of prenatal RNAseq when comparing direct versus cultured specimens.

FIGURE 6.

Proposed workflow diagram for clinical implementation in diagnostic lab. DNA based assays (e.g. CMA, ES) would be performed from an established prenatal culture. RNAwould be obtained from the same established culture. Prenatal cells could be frozen and stored at −80C pending DNA based assay results.

Additionally, while serving as a supporting functional assay for variant interpretation, our preliminary findings suggest that RNAseq could detect full chromosome copy number variants through read‐depth analysis (Figure S4). However, it faces challenges in identifying smaller copy number gains or losses. Although DNA‐based analysis remains the gold standard for diagnostic cytogenetic findings, our results suggest that RNAseq can offer confirmatory or supplementary identity verification to parallel DNA testing. The cCVS and cAF gene expression profiles reported in our study may provide a useful foundation for clinical laboratories to build SNP/INDEL‐based sample identity verification assays from RNAseq data. These assays can be used to verify the identity of RNAseq in conjunction with fetal exome or GS data.

RNA is dynamic, and RNA expression changes based on temporal and spatial information. As a result, tissue and time‐specific databases will be necessary for accurate analysis in the future (similar to whole DNA methylation analysis), which may prove a challenge. However, we anticipate that in the prenatal context where there are only three primary sample types (i.e., amniotic fluid, chorionic villi and cord blood) and where the sampling period is usually temporally constrained, it will be possible to generate RNAseq reference datasets to facilitate prenatal gene‐expression analysis. Lastly, in addition to the tissue‐ and time‐specific databases used as gestational age‐specific reference ranges, further considerations will be required for the clinical implementation of RNAseq in diagnostic laboratories. These may include but are not limited to the standardization of culture methods, clinical validation procedures, analytical and bioinformatic pipelines, and interpretation and reporting guidelines, particularly considering expectations of prenatal turnaround times.

5. Conclusions

Our study highlights the potential of prenatal RNAseq to enhance diagnostics by offering detailed gene expression profiles from cultured amniotic fluid and chorionic villus samples. In addition, prenatal RNAseq can reveal unique patterns linked to fetal development and disorders, providing a broader view than traditional molecular and cytogenetic methods. This advancement can improve the accuracy of prenatal diagnoses and support the interpretation of genetic variants, offering valuable insights into fetal conditions and development. However, prenatal RNAseq remains investigational and is not yet clinically validated for routine prenatal diagnostics. Further studies are needed to establish its clinical utility before it can be implemented in clinical practice.

Funding

This project is supported by National Human Genome Research Institute (NHGRI) grant R35HG011311. R.Z. and C.M.P. are supported by grant T32GM007526.

Ethics Statement

The Baylor College of Medicine Institutional Review Boards approved the study under protocol H‐51130, with a waiver of consent to conduct the proposed research in a deidentified manner.

Conflicts of Interest

Baylor College of Medicine (BCM) and Miraca Holdings Inc. have formed a joint venture with shared ownership and governance of Baylor Genetics (BG), which performs genetic testing and derives revenue. P.L. is an employee of BCM and derives support through a professional service agreement with BG. S.T. is an employee of BG.

Supporting information

Figure S1: Distance matrix of hierarchical clustering of all sample types (n = 45) (A). Heatmap of the top 50 genes ordered based on normalized expression value (B).

Figure S2: Comparison of expression profiles between prenatal (n = 25) and postnatal (n = 20) samples with gene lists from clinically available NGS panels. Number of gene overlap labeled between prenatal and postnatal samples labeled with expression levels shown by NGS panel associated with kidney disorders (segmental glomerulosclerosis) (n = 39) (A) and cardiac mesenchyme disorders (inherited cardiomyopathies) (n = 44) (B).

Figure S3: Comparison of gene expression profiles between prenatal (n = 21) and postnatal (n = 20) samples and GTEx dataset. Highly expressed genes (top 1000) in each human adult organ were obtained from the GTEx (https://www.gtexportal.org/home/) and used in the comparison of adequately expressed gene markers (TPM > 5) of prenatal samples (n = 21) or postnatal samples (n = 20). Bar plots represent the proportion of gene overlap.

Figure S4: Copy number analysis based on RNASeq data. The copy number of each chromosomes was analyzed based on the normalize expression matrix and minor allele frequency using the RNAseqCNV package in R. A copy number deletion or duplication is defined by a distorted minor allele frequency profile and corresponding expression changes.

Table S1: Identified literature review utilized to create the prenatal phenotypes genes.

Table S2: Prenatal samples demographics (n = 25) (tab 1); RNAseq read depth of samples used in analysis (n = 45) (tab 2); MCC contamination analysis (tab 3).

Table S3: Gene expression values (TPM) per sample type.

Table S4: Prenatally detectable phenotypes and gene associations.

Table S5: Gene expression levels of prenatally phenotype associated genes.

Table S6: Differential gene expression analysis comparing AF and CVS samples.

Table S7: Pathway analysis using differentially expressed genes for AF (tab1) and CVS (tab 2) samples.

Table S8: Primary and secondary markers for fetal tissues. Primary and secondary markers for fetal tissues were obtained from Cao et al.²³.

Table S9: Top 1000 expressed genes in each Gtex tissue. Top 1000 expressed genes in each tissue were obtained from the Gtex database (https://www.gtexportal.org/home/).

Table S10: Summary of highly expressed genes associated with prenatally detectable phenotypes.

Table S11: Simulated analysis evaluating gene expression levels in cAF and cCVS samples of published genes from ES/GS studies associated with prenatal diagnoses.

Data Availability Statement

Raw data used in this research are available upon request from the authors.