Determining the origin of genome aberrations improves the positive predictive value of NIPT for 22q11.2 deletion syndrome

Abstract

Non-invasive prenatal testing (NIPT) has been endorsed by the American College of Medical Genetics and Genomics as the preferred method for screening fetal 22q11.2 deletion syndrome (22q11.2 DS). Maternal genomic aberrations represent a significant source of false positives in NIPT, and there are currently no solutions that effectively address this challenge. We have devised an innovative NIPT bioinformatics pipeline designed to discern the origins of copy number variations (CNVs). Then, we recruited a cohort of 39cases of 22q11.2 DS to validate the effectiveness of our methodology. Follow-up tests including amniocentesis and genome sequencing of maternal leukocytes were conducted. Leveraging a dataset of over 900 CNVs, we developed a new pipeline that classifies CNVs into those of fetal, maternal, and maternal-fetal origin based on NIPT data. The use of our pipeline led to a notable increase in the positive predictive value of NIPT for detecting 22q11.2 DS from 87% (34/39) to 94% (34/36). Furthermore, our approach has the potential to reduce the number of invasive tests by 8% (3/39). Our innovative and reliable bioinformatics pipeline has enabled the accurate differentiation of CNV origin into fetal, maternal, and maternal-fetal categories. Incorporating this pipeline into the analytical workflow could reduce false positives in NIPT results and minimize the need for invasive prenatal diagnoses.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-10446-8.

Introduction

Non-invasive prenatal testing (NIPT) relies on the analysis of placental cell-free DNA (cfDNA) circulating in the peripheral blood of pregnant women¹. Initially, NIPT was primarily employed to screen for fetal trisomies 21, 18, and 13, and its wide clinical adoption can be attributed to its high sensitivity, low false positive rate, and high positive predictive value (PPV)². With advances in the analytics pipelines, NIPT has expanded to include screening of copy number variations (CNVs), with a PPV of 32–47% in different studies³. Notably, the American College of Medical Genetics and Genomics (ACMG) now conditionally recommends NIPT for the screening of fetal 22q11.2 deletion syndrome (22q11.2 DS), also known as DiGeorge syndrome (DGS), based on moderate certainty of evidence, for all pregnant women⁴.

22q11.2 DS is one of the most prevalent microdeletion disorders, with an estimated incidence of 1 in 3000 to 1 in 6000 live births. It is characterized by congenital heart defects, palate abnormalities, immune system dysfunction, and other clinical manifestations⁵. Genomic investigations in a cohort of children with intellectual disability and congenital abnormalities revealed that 20% of 22q11.2 DS cases were inherited from parents⁶. Due to its broad phenotypic variability^7,8, major medical centers with limited experience managing 22q11.2 DS patients may fail to diagnosis it. In some cases, adults have received a diagnosis only following the birth of an affected child⁹.

When a pregnant woman has genome aberrations, such as being a carrier of 22q11.2 DS, it can be challenging to accurately determine via NIPT whether the fetus has inherited these genomic anomalies¹⁰. This challenge arises primarily because cfDNA in the peripheral blood is a mixture of fragmented maternal and fetal genomic DNA which cannot be distinguished from each other. Moreover, maternal cfDNA predominates and accounts for an average proportion of 85–90%^11,12. In such cases, invasive prenatal diagnosis through amniocentesis is routinely recommended, causing anxiety for pregnant women¹³. Previous studies have highlighted the fact that maternal background aberrations significantly contribute to false positive results in NIPT^14–16. There is currently no solution available to address these issues.

In this study, we utilized a dataset comprising more than 900 CNVs obtained from NIPT data to develop an innovative bioinformatics pipeline capable of distinguishing the origin of CNVs. We then used a 22q11.2 DS cohort to validate its performance. Our new pipeline effectively differentiated CNVs into fetal CNVs, maternal CNVs, and maternal-fetal CNVs. This achievement reduced the false positive rate in NIPT and thus minimized the need for invasive prenatal diagnoses.

Materials and methods

Samples

A total of 927 anonymized NIPT datasets from the clinical laboratory of the Beijing Genomic Institution (BGI) were retrieved. These data were generated through BGI’s genome-wide NIPT test, the NIFTY^® test. The gestational age of participants ranged from 10 to 30 weeks, with a median of 16 weeks. Maternal ages varied between 17 and 46 years, with a median of 28 years. All pregnancies included in this study were singleton gestations.

In this study, CNVs were classified as being of maternal, maternal-fetal, or fetal origin. A CNV carried by a pregnant woman but not passed to the fetus was defined as maternal. The maternal background could cause a false positive in NIPT if maternal CNVs are misattributed to the fetus. CNVs inherited by the fetus from the pregnant woman were considered maternal-fetal CNVs. In cases where a pregnant woman was not a carrier of a CNV, the maternal background would not interfere the detection of CNV from the fetus, and the CNV was defined as a fetal CNV.

From the 927 anonymized NIPT datasets, we further investigated 39 samples with deletions overlapping the chromosomal region 22q11.2 to test the performance of our bioinformatics pipeline. Chromosomal region 22q11.2 was selected because it is the most common chromosomal deletion and is clinically reported in BGI’s NIFTY^® test. When historical maternal blood samples were available, the maternal peripheral blood leukocytes were subjected to genome sequencing to determine the status of the maternal genome.

Data preparation

Sequencing reads 35 base pairs (bp) long were aligned to the human reference genome (GRCh37/hg19) using the Burrows-Wheeler Aligner (BWA) software¹⁷. To ensure the accuracy of the alignment, mismatched reads and duplicated reads were removed from the dataset. Only unique mapped reads (UR) were retained for following analysis.

Fetal fraction estimation

The fetal fraction (ff) of male fetal samples was determined by calculating the relative coverage of the Y chromosome compared to the whole genome. For female fetal samples, the estimation of fetal fraction was carried out using the pre-trained neural network model FF-QuantSC¹⁸. This model was specifically designed to accurately quantify the fetal fraction in the absence of Y chromosome information.

CNV calling

The reference genome was divided into continuous, non-overlapping genomic windows of 100 kb. Mapped reads were then partitioned among these windows, and the read depth of each window was determined. Because various factors (i.e., blood collection, cell-free DNA extraction, library preparation, PCR, etc.) can cause variations in read depth, GC correction and principal component analysis (PCA) were applied for normalization¹⁹. Subsequently, a hidden Markov model (HMM) was employed to detect CNVs²⁰. For each CNV identified by the HMM, the copy ratio was calculated based on the read depth of the CNV region. To eliminate the effects of sequencing depth, the read depth of each window was divided by the average depth of all windows within the same sample. The copy ratio was defined as the ratio of the average depth in the CNV region of the sample to the average depth of a control set.

CNV sorting

Given that the copy ratio distribution of maternal CNVs and maternal-fetal CNVs differs from the distribution of fetal CNVs, the first step was to differentiate fetal CNVs from maternal CNVs and maternal-fetal CNVs with the 927 anonymized NIPT datasets.

Next, to distinguish maternal CNVs from maternal-fetal CNVs, we employed a linear model²¹. In principle, the copy ratio of maternal CNVs is positively correlated with fetal fraction, whereas the copy ratio of maternal-fetal CNVs is not related to fetal fraction. The copy ratio of the two types of CNVs is equal when the fetal fraction is 0. Consequently, the relationship between copy ratio and fetal fraction can be expressed by the following formula:

and are the constants of the linear model and represents experimental noise. We assumed that the noise was the same in the two linear models and that ε was uncorrelated across different samples, following a normal distribution with a mean of 0 and a constant variance σ². Then, we estimated , and using maximum likelihood estimation. For each CNV, the probabilities of observing the copy ratio associated with the fetal fraction (ff) in a maternal CNV model (P_M) and in a maternal-fetal CNV model (P_MF) were expressed as follows:

The likelihood function (L) was:

By maximizing the likelihood function, we obtained the estimated values of , and σ. Thus, the relationship between copy ratio and fetal fraction of maternal CNVs was:

We classified CNVs based on P_M and P_MF values. If P_M was higher than P_MF, the CNV was considered a maternal CNV; if P_M was lower than P_MF, the CNV was classified as a maternal-fetal CNV. To minimize the risk of misclassification, we established a gray zone. The criterion for the gray zone was that the log fold change of P_M and P_MF was less than 2. CNVs within the gray zone require further investigation.

Results

Overview of the bioinformatics pipeline

The process of determining the origin of CNVs included four main steps: data preparation, fetal fraction estimation, CNV calling, and CNV sorting (Fig. 1). In the CNV sorting phase, fetal CNV was first differentiated from maternal CNV and maternal-fetal CNV based on the density curve. Subsequently, the separation of maternal CNVs from maternal-fetal CNVs was performed. To minimize the risk of misclassification, a “gray zone” was established. In cases where a CNV was classified as a fetal CNV or maternal-fetal CNV or fell within the gray zone, it was recommended to consider invasive prenatal diagnosis. Conversely, if a maternal CNV was detected, invasive prenatal diagnosis was deemed unnecessary. More details are shown in “Materials and methods’’ section.

Fig. 1.

Workflow for differentiating the source of copy number variations in NIPT. CNV, copy number variation; BWA, Burrows-Wheeler Alignment; PCA, principal component analysis; HMM, Hidden Markov Mode; ff, fetal fraction; P_M, the probability of observing the copy ratio associated with the fetal fraction in maternal CNV mode; P_MF, the probability of observing the copy ratio in maternal-fetal CNV mode.

Differentiating fetal CNVs from maternal background

Given that the fraction of maternal and fetal cfDNA varies in the peripheral blood of a pregnant woman, the distribution of copy ratios between maternal and fetal CNVs are theoretically different. We plotted a density curve of the copy ratios of 927 deletions. They were randomly distributed across chromosomes 1 to 22, except for chromosome 19. This was because chromosome 19 was excluded from our NIPT analysis because of unreliable results caused by its complex genomic characteristics^22,23. The distribution CNV sizes across chromosomes is listed in Supplemental Fig. 1.

The density curve of the copy ratio showed two distinct peaks, representing CNVs of maternal origin and fetal origin (Fig. 2). The lowest density of copy ratio was 0.762 and was chosen as the threshold. A copy ratio greater than or equal to 0.762 was considered to indicate fetal CNV, while a copy ratio less than 0.762 was considered to be either a maternal CNV or maternal-fetal CNV.

Fig. 2.

Copy ratio distribution of 927 copy number variations from NIPT data. The plot shows a histogram of the observed copy ratio and fitted density (solid line). The vertical dashed line represents the threshold value (0.762) used for fetal CNVs.

Differentiating maternal CNVs and maternal-fetal CNVs

The copy ratios of 201 of the 927 CNVs was less than 0.762. These were considered maternal CNVs or maternal-fetal CNVs. To further differentiate them, they were analyzed using a linear regression model. The association between the copy ratio and the fetal fraction was estimated as follows: maternal CNV , and maternal-fetal CNVs  (Fig. 3).

Fig. 3.

Distribution of 201 maternal and maternal-fetal CNVs. represents the fitted equation for maternal CNVs. represents the fitted equation for maternal-fetal CNVs. Gray zone indicates that CNVs in this zone need further investigation. The vertical dashed line represents the fetal fraction threshold value (0.035) for quality control in non-invasive prenatal testing. CNV, copy number variation.

To minimize the risk of misclassification, a gray zone was established. CNVs located in the gray zone require further investigation. The other two ranges were called maternal CNV zone (above the gray zone) and maternal-fetal CNV zone (below the gray zone) (Fig. 3). As a result, 79 out of the 201 CNVs were classified as maternal CNVs, 94 were maternal-fetal CNVs, and 28 were in the gray zone.

Validation of the bioinformatics pipeline

To assess the precision of our novel bioinformatics pipeline in classifying the origin of CNVs, we selected 39 CNVs that overlapped with the 22q11.2 DS region for validation. These CNVs had been previously identified in clinical NIPT tests but had not been classified according to origin. The classic 2.5 Mb 22q11.2 DS region was covered by 18 of the CNVs, while 20 were within a range of 1.4 to 2.0 Mb, closely aligning with the critical 1.5 Mb 22q11.2 DS region, and 1 CNV spanned 1 Mb encompassing TBX1, the crucial gene associated with 22q11.2 DS (Supplementary Fig. 2). Subsequent follow-up through amniocentesis revealed that 34 of these CNVs were true positives (affected fetuses) while 5 were false positives, giving a positive predictive value (PPV) of 87% (34/39) (Supplementary Table 1).

With our new pipeline, three cases were categorized as maternal 22q11.2 DS, five as maternal-fetal 22q11.2 DS, and 30 as fetal 22q11.2 DS, with one case in the gray zone (Fig. 4 and Supplementary Table 2). Through amniocentesis and genome sequencing of maternal leukocytes, all three maternal 22q11.2 DS cases and the five maternal-fetal cases were confirmed. Of the 30 fetal 22q11.2 DS cases, 28 were validated as true positives, and two were false positives, resulting in a PPV of 93% (28/30).

Fig. 4.

The validation of the origin of 39 22q11.2 deletion syndromes cases. “+” indicates that the 22q11.2 deletion syndrome was detected from fetal and/or maternal genomic DNA, while “-” indicates it was undetected. A question mark indicates that maternal genomic DNA was unavailable. One case located in gray zone was confirmed to be of maternal-fetal origin.

Overall, of the 36 cases that necessitated amniocentesis (including fetal, maternal-fetal, and gray zone cases), 34 were verified as true positives, yielding a PPV of 94% (34/36) (Supplementary Table 2). Importantly, invasive prenatal diagnosis was unnecessary for women with maternal 22q11.2 DS, potentially reducing the number of invasive tests by 8% (3/39) with the implementation of our new NIPT bioinformatics pipeline.

Discussion

Principal findings

In this study, we developed a novel bioinformatics pipeline for the precise analysis of NIPT, with the aim of effectively distinguishing the origin of genomic aberrations. To our knowledge, this is the first research work reporting the successful achievement of discerning the origin of genome aberrations in NIPT. Our findings highlight the efficacy of our approach in decreasing the false positive rate resulting from maternal genome profiles in NIPT, thereby minimizing the use of unnecessary invasive prenatal diagnostic procedures. Importantly, our bioinformatics pipeline can be integrated into clinical settings without additional sequencing expenses.

Results in the context of what is known

In recent years, the use of NIPT has expanded from the detection of chromosomal trisomies to detecting CNVs in the general population^24,25. Maternal genome aberrations have emerged as an important source of false positive results in NIPT¹⁵. This was attributed to the fact that rare autosomal CNVs had a prevalence of 10.5% in the unselected general population²⁶. Moreover, an investigation of the inheritance pattern of 66 pathogenic CNVs from 653 fetal probands found that nearly three-quarters of the CNVs showed variable percentages of inherited events, ranging from 5.6 to 100%⁶. In the context of prenatal diagnosis, it was observed that approximately half of pathogenic CNVs are inherited from parents^27–29. These data highlight the importance of maternal genome analysis in identifying potential fetal genomic variations via NIPT.

Clinical implications

There are three major benefits to classifying the origin of genomic aberrations using our new pipeline. First, it would reduce disparities and discrimination in the selection of prenatal screening tests. Typically, NIPT is not advised for women who already have a chromosomal abnormality themselves. In China, the technical guidelines for cfDNA-based prenatal screening stipulate that NIPT is not suitable for women with chromosomal abnormalities. This guidance is rooted in the potential for false positive outcomes from maternal genomic profiles³⁰. With the use of our new methods to classify the origin of CNVs by NIPT, those women would also be eligible to choose NIPT as a prenatal screening test.

Second, the application of our bioinformatics pipeline can significantly reduce the number of unnecessary invasive tests, alleviating the anxiety experienced by pregnant women¹³. As a first-line screening test, NIPT would reduce unnecessary invasive tests by 39.6%³¹ and decrease puncture-related abortion by 90.8%³². However, since NIPT has not been able to differentiate the origin of genomic aberrations, unnecessary invasive tests are still undertaken. Implementing our novel NIPT bioinformatics pipeline could avert at least 8% of unnecessary invasive prenatal diagnoses when using NIPT for screening 22q11.2 DS, a practice now recommended by the American College of Medical Genetics and Genomics for all pregnant women⁴. While for other microdeletions, confirmatory invasive diagnostic procedures remain clinically mandated when maternal CNVs are detected.

Third, our approach enhances the performance of NIPT by increasing its PPV. Through effective differentiation of the origin of genomic aberrations, our newly developed pipeline increases the PPV from 87 to 94% for 22q11.2 DS while maintaining the same detection sensitivity. Notably, the PPV surpasses the wide range reported in other studies, which varies from 18 to 93%^33–37.

Strengths and limitations

Nonetheless, there are some limitations to this study. First, our method has not been validated for microduplications due to the infrequent reporting of microduplication-related disorders in clinical NIPT, resulting in a scarcity of samples for analysis. Second, we excluded the analysis of genomic abnormalities of sex chromosomes since the differentiation between fetal CNV and maternal-fetal CNV is closely associated with fetal fraction, and sex chromosome aberrations can impact the estimation of fetal fraction. Third, the absence of a control dataset for comparative analysis restricts the generalizability of our CNV ratio calculations. Future studies should incorporate control cohorts to validate these findings.

Conclusions

In conclusion, we developed an innovative and reliable bioinformatics pipeline capable of classifying CNVs in NIPT as being of fetal, maternal, or maternal-fetal origin. The integration of this method into the analytical workflow reduces the influence of maternal genomic aberrations and enhances the PPV of NIPT for screening 22q11.2 DS.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

NIPT

Non-invasive prenatal testing

cfDNA

cell-free DNA

22q11.2 DS

22q11.2 deletion syndrome

DGS

DiGeorge syndrome

CNVs

copy number variations

PPV

positive predictive value

Author contributions

J. Xiang, and Z. Peng conceived and designed the study. J. Xiang, J. Peng, X. Sun, H. Zhang, J. Zhang, S. Zhou, and J. He conducted the statistical analyses. J. Shen, H. Li, L. Hu, S. Zhou, S. Xu, Y. Yang, and J. He collected the data. J. Xiang, X. Sun, and H. Zhang conducted the experiments. J. Peng and J. Li conducted the bioinformatic pipelines. J. Xiang, and X. Sun wrote and revised the manuscript. J. Xiang and Z. Peng supervised the project. All authors read and approved the final manuscript.

Funding

Supported by the National Key R&D Project of China (2023YFC3402504 to Z.P.).

Data availability

Data will be made available to qualified researchers on request. For data requests, please contact the corresponding author, Zhiyu Peng, at pengzhiyu@bgi.com.

Declarations

Competing interests

J. Xiang, J. Peng, X. Sun, H. Zhang, J. Shen, S. Xu, Y. Yang and Z. Peng are employed by the BGI Genomics. The remaining authors declare no competing interests. All authors completed the author disclosure form.

Patents

J. Xiang, J. Peng, X. Sun, J. Shen, and Z. Peng have filed patent applications (No. 202211727322.7 and PCT/CN2023/123292) related to this work.

Ethical approval

This study was carried out with the approval of BGI’s Institutional Review Board (BGI-IRB 22169). Informed consent was obtained from all participants as required by the IRB. All experimental procedures were strictly followed in accordance with the ethical guidelines of the Declaration of Helsinki and applicable national regulations.