Detection of DNA Contamination in Prenatal Samples from Whole Exome Sequencing Data

Abstract

Background

Maternal cell contamination (MCC) in prenatal samples poses a risk for misdiagnosis, and therefore, testing for contamination is necessary during genetic analysis of prenatal specimens. MCC testing is currently performed as a method separate from the diagnostic method. With the increasing application of whole exome sequencing (WES) in prenatal diagnosis, we sought to develop a method to estimate the level of contamination from WES data, aiming to eliminate the need for a separate MCC test.

Methods

To investigate the impact of MCC on the distribution of the variant allele fraction in WES data, contamination was both simulated in silico and artificially induced. Subsequently, a bioinformatic WES contamination method was developed and validated by comparing its performance to that of the gold standard (short tandem repeat [STR]) MCC test, validated for detecting ≥5% contamination. Finally, post-implementation performance was monitored for a 15-month period.

Results

During validation, 270 prenatal samples underwent analysis with both WES and the gold standard test. In 259 samples, the results were concordant (248 not contaminated, 11 contaminated with both tests). In 11 samples, contamination was only detected in WES data (2 of which contained ≥5% contamination with WES, which is above the detection limit of the gold standard test). The data of the post-implementation evaluation on 361 samples, of which 68 were contaminated, were in line with the validation data.

Conclusions

Contamination can reliably be detected in WES data, rendering a separate contamination test unnecessary for the majority of samples.

Introduction

The presence of maternal cells in direct and cultured amniotic fluid (AF) and chorionic villi (CV) samples, as well as in umbilical cord and umbilical cord blood samples, is well documented and represents a potential source of errors in prenatal genetic testing (1–3). Therefore, professional guidelines recommend examining these samples for the presence of maternal contamination simultaneously or in parallel with genetic diagnostic requests (4–6). Moreover, current guidelines recommend analyzing prenatal DNA samples in parallel with maternal DNA samples, and the absence of maternal microsatellite alleles in the fetal DNA excludes the presence of significant contamination with maternal DNA (4, 5). The guidelines of the American College of Medical Genetics and Genomics (ACMG) state that “prenatal samples should be examined in parallel with a maternal sample to rule out error due to maternal cell contamination” (6) and according to the UK guidelines, it is not acceptable to state that significant MCC has been excluded if only the fetal sample has been analyzed with the current MCC tests, without analyzing the maternal sample (5). Therefore, currently maternal cell contamination (MCC) testing is performed as a separate test on both fetal and maternal DNA samples. Even though the tests that are being performed actually test for the presence of contaminating maternal DNA instead of maternal cells, the tests are referred to as MCC tests. The assays that are currently mostly used for MCC testing are based on the analysis of short tandem repeats (STRs).

Whole exome sequencing (WES) is a valuable tool in prenatal diagnosis and is being applied in a steadily increasing number of laboratories (7–9). Not only can single nucleotide variants (SNVs) be analyzed but also other variants, such as copy number variants, absence-of-heterozygosity, and uniparental disomies, can be analyzed simultaneously from the same data, with turnaround times of <3 weeks (10). In the prenatal setting, WES is often performed as a trio-based test, in which both fetal and parental DNA samples are sequenced and analyzed simultaneously, to enable a more informed interpretation of the detected fetal variants. Since MCC represents a potential source of errors (especially for SNV analysis), it is important to exclude MCC levels above a certain threshold [≥5% according to the guidelines (5, 6)]. We hypothesized it would be more time- and cost-efficient to check for the presence of contaminating DNA directly in WES data, without the need for a separate test.

In the present paper, we describe a novel method to detect and estimate the amount of contamination from WES data, based on the distribution of the variant allele fraction (VAF) of SNVs. With this method, any kind of contaminating DNA in the sample, including maternal DNA, can be detected, and, therefore, the method is applicable to both trio- and singleton-sequencing strategies. Additional MCC testing, using a separate test, only remains necessary in cases where contaminating DNA levels above a certain threshold are detected in the WES data, in order to determine the origin of the contaminating DNA.

Materials and Methods

Study design

In short, we hypothesized that contaminating DNA causes a shift in the VAF of the WES data and that this shift can be used to quantify the amount of foreign DNA in the sample. To investigate the effect on the VAF by contaminating DNA, WES reads from fetuses were in silico randomly replaced with those of their respective mothers. The distribution of the VAF was compared to that of a reference data set, which was created using WES data from 50 DNA samples without MCC, isolated from EDTA blood. Lower and upper limits were selected for 2 statistics that most accurately reflected the true amount of MCC, i.e., (a) the average VAF of homozygous variants and (b) the ratio of heterozygous variants with a VAF >75% and the total number of heterozygous variants. Validation of the WES contamination method was performed on samples with known amounts of contamination, as determined by STR analysis, which is considered the gold standard test for MCC. Subsequently, the WES contamination method was implemented as our standard contamination test for all WES samples. Post-implementation performance was monitored during a 15-month period.

Gold standard MCC Testing by STR Analysis

Gold standard MCC testing was performed by STR analysis using the multiplex AmpFLSTR™ Identifiler™ PCR Amplification Kit (ThermoFisher). This STR multiplex assay amplifies 15 tetranucleotide repeat loci and an amelogenin gender-determining marker. The test was performed according to procedures recommended by the manufacturer. In our laboratory, this assay has a lower detection limit of 5% MCC, although sometimes the presence of lower amounts of MCC can be detected. Data were analyzed using Genemarker software (v.2.6.7; SoftGenetics) by trained technicians. The data used for this study were generated as a part of routine diagnostic procedures, and the interpretation was only re-evaluated in specific cases, as described in the results.

Samples

DNA from peripheral EDTA blood was isolated using standard procedures. Fetal DNA was isolated from uncultured CV, AF cells, or umbilical cord. In case insufficient material was sampled or direct DNA isolation failed, DNA was isolated from cultured cells. CV were first dissociated, and only the mesenchymal core cells were used for DNA isolation. Only when insufficient material was sampled was DNA isolated from undigested total villi.

WES

WES was performed as described previously (10). In short, library preparation for all samples was done automatically using a Twist Bioscience Human Core Extended Kit (Twist Bioscience). Sequencing was performed on a NovaSeq6000 (Illumina) and variants were annotated using the diagnostic pipeline version 2.3.0.

In silico MCC Analysis from WES Data

Using WES data from 5 fetuses and their respective mothers, MCC was simulated by randomly replacing 1%, 5%, or 10% of the fetal WES reads with that of their respective mothers. These composed BAM files were reannotated with the pipeline. From the resulting SNVs, 2 groups of variants were selected: (a) homozygous (VAF ≥97%) exonic variants and (b) heterozygous (VAF <97%) exonic variants, with at least 50 reads and a Genome Analysis ToolKit quality score of at least 500 in order to reduce noise. Data from one fetus and its mother were used to create a set of data with 0% to 20% MCC, with intermediate steps of 1%. A reference data set was created using WES data from 50 DNA samples isolated from EDTA blood, ensuring the absence of MCC in the reference pool, and the distribution of the VAF of the contaminated samples was compared to that of the reference data set. Lower and upper limits were selected for the 2 indicators that best described the true amount of contamination: (a) the average VAF of homozygous variants and (b) the ratio of heterozygous variants with a VAF >75% and the total number of heterozygous variants.

Validation of MCC Analysis from WES Data

The limits, as determined from the in silico analyses, were incorporated into our bioinformatic pipeline, in order to generate an additional output file for each sample containing the estimated amount of contamination. Between 0% and 20%, the level of contamination was quantified with intermediate steps of 1%, whereas above 20% contamination this could not be further quantified and the level was denoted as ≥21%.

For validation, the performance of the WES contamination method was evaluated with:

• Seven samples with known levels of MCC, either because MCC was detected in the past using the gold standard test or by mixing known amounts of fetal DNA with that of the respective mothers.

• Comparing the MCC levels as determined by the gold standard test and the newly developed WES method of all the prenatal samples that were sent to our lab for diagnostic WES during a six-month-period. As the lower detection limit of the gold standard is 5% MCC, a level of <5% was considered negative and therefore, for this validation, the gold standard test results were categorized into <5% and ≥5% contamination. The results of the WES contamination method, of which the analysis is not subject to visual inspection, were categorized into 0%, <5% and ≥5% contamination.

Post-implementation evaluation

After validation, all samples that were sent to our laboratory for diagnostic WES were tested for the presence of MCC with the newly developed contamination method (n = 631). In case contamination was detected, a separate gold standard test was performed. After 15 months (from March 2021 until May 2022), the data were evaluated.

Results

In silico MCC analysis from WES data

A reference data set was created using WES data from 50 noncontaminated control samples. WES data from 5 fetuses and their respective mothers was used to simulate 1%, 5%, and 10% contamination. In the absence of contamination, the VAF in the WES data followed a bimodal distribution with peaks around 50% and 100%, representing heterozygous and homozygous variants, respectively. However, in the presence of contamination the distribution of the VAF shifted as described below.

Homozygous variants

In the presence of contamination, the number of variants with an allele fraction equal to 100% decreased, while the number of variants with a fraction less than 100% increased (online Supplemental Fig. 1).

The distribution of the VAF of the homozygous variants was compared to that of the reference data set using the Kolmogorov–Smirnov (K-S) test. Although the K-S test discriminated between samples with and without contamination, it could not be used to estimate the amount of contamination, as samples with different levels of contamination all showed a P value of 0 (online Supplemental Fig. 2).

Data from one fetus and its respective mother were used to create a set of data with 0% to 20% contamination, with incremental steps of 1% contamination. The average VAF of homozygous variants decreased until 13% contamination, after which the average VAF increased again. In this data set, this resulted in an overlap in the average VAF between samples with 5% to 20% contamination. The K-S P value was only significant for samples with 1% to 15% contamination, but even when combining these 2 statistics, a distinction between samples with 11% to 15% contamination still could not be made. Furthermore, extrapolation of the data suggested that it would not be possible to distinguish between samples with very high amounts of contamination or no contamination at all (Fig. 1). As only low-level MCC can be quantified using the VAF distribution of homozygous variants, we studied whether the distribution of heterozygous variants could be used too, either as standalone or combined with the VAF distribution of the homozygous variants.

Heterozygous variants

In the presence of contamination, the number of variants with a VAF of 50% decreased, and a second peak of variants emerged around approximately 90% VAF (online Supplemental Fig. 3), representing variants that would have been called as homozygous in the absence of MCC. In order to quantify this additional peak, the fraction of heterozygous variants with a VAF >75% (second peak) of the total number of heterozygous variants was calculated. Below 10% contamination, the fraction of heterozygous variants with a VAF >75% could not discriminate very well between different amounts of contamination. However, at 10% or more contamination, the fraction of heterozygous variants with a VAF >75% steadily increased from approximately 1.5% upwards with increasing amounts of contamination (Fig. 2).

Lower and upper limits and cutoff value

In summary, the average VAF of the homozygous variants (indicator 1) is informative for low-level MCC (0% to 10%), while the fraction of heterozygous variants with a VAF >75% (indicator 2) is informative for higher level MCC (>10%). Therefore, 1.5% was selected as a cutoff value for indicator 2, above which indicator 2 was used to determine the amount of contamination. The lower and upper limits for the 2 indicators (online Supplemental Table 1) were incorporated into our WES pipeline, for the generation of an additional output file for each sample containing the values for indicators 1 and 2 and the estimated amount of contamination, ranging from 0% to 20% and ≥21%.

Validation

For validation, the performance of the WES contamination method was evaluated during a 6-month period, using routine diagnostic samples. During this period, for all samples contamination was measured with both the gold standard test and the WES contamination method (see Supplemental Table 2). In total, 270 prenatal samples were included, 263 of which were sent to our laboratory for routine diagnostic WES: 248 samples were not contaminated according to both tests, whereas 11 samples were. In the other 11 samples, contamination was detected with the WES method only, in 9 of which the contamination with WES was estimated to be <5%, and 5% in 2. However, re-evaluation of the STR data did reveal minor contamination (approximately 5% to 8%) in the latter 2 samples. Contamination occurred in all sample types, but significantly more often in CV samples (data not shown). In general, the gold standard test resulted in a slightly higher MCC estimation compared to the WES contamination method, but overall they corresponded well (r² = 0.64) (Fig. 3A).

Post-implementation evaluation

In March 2021, the WES contamination method was implemented in our routine diagnostic flow with a separate gold standard test performed in all cases when WES contamination was detected, regardless of the level of contamination. Data generated between March 2021 and May 2022 were evaluated. Contamination was detected with WES in 68 out of 631 samples (10.8%). For 65 of these samples, there was (sufficient) DNA available from both the fetus and the mother for subsequent MCC testing with the gold standard test (Table 1 and Fig. 3B).

In 61/65 samples, <5% contamination was reported with the WES method, 48 samples of which also showed <5% contamination with the gold standard test and 13 showed ≥5% MCC with the gold standard test (average 7.2% [5% to 13%]). One sample (of the 65) contained 9% contamination according to the WES method but was originally reported not to contain MCC with the gold standard test: however, re-evaluation of the gold standard test data, for the purpose of the current study, revealed approximately 10% MCC, emphasizing the risk of mis-interpretation when routine data analysis relies on visual inspection, as is the case for the gold standard test. Three of the 65 samples were predicted to contain ≥5% MCC according to both tests. Of note, in 1 of those 3 samples, the WES contamination method predicted 20% contamination, while with the gold standard test 80% to 100% MCC was reported. This AF sample, mostly maternal in origin, was obtained in a pregnancy with oligohydramnios, and was thus most probably not AF in nature. In fact, for this sample, the estimation with WES can be regarded as an accurate estimate of the amount of contamination with foreign DNA, as the maternal sample can also be considered to be the main sample, being contaminated with 20% fetal DNA, which is in line with the 20% contamination predicted by WES. This particular sample underscores the need for further MCC testing with a separate test, in case a significant amount (≥5%) of contamination is suspected, to study the origin of the foreign DNA.

Discussion

In the present paper, we discuss the development and performance of an in-house developed WES contamination method to detect contamination bioinformatically directly in (prenatal) samples that undergo WES examination. Our method is based on the distribution of the VAF in WES data, which shifts in the presence of contaminating DNA. Our method combines 2 different indicators that reflect the extent to which the distribution of the VAF is shifted in contaminated samples, compared to a normal VAF distribution in noncontaminated samples. The level of contamination estimated with WES was compared to the level estimated with the gold standard STR analysis, which in our hands has a lower detection limit of 5%. During the evaluation period, the WES method never failed to detect contamination ≥5%, as determined with STR analysis (albeit the level of contamination as estimated with WES was lower), and thus is at least as sensitive as the currently used gold standard test. Furthermore, the WES contamination method never predicted ≥5% contamination that could not be confirmed with the STR test, indicating comparable specificity. In general, the WES contamination method tended to underestimate the amount of MCC compared to the gold standard test. STR analysis, however, is a semiquantitative assay, and it is therefore uncertain which of both methods most accurately reflects the true amount of contamination. Moreover, as shown during post-implementation, visual inspection of peak patterns can be prone to human error, and therefore an automated bioinformatic test clearly has an added advantage.

With respect to other descriptions of MCC detection by next-generation sequencing, Gabriel et al. mentioned the possibility of detecting MCC in WES data by looking at the presence of non-inherited maternal variants and at the allele fraction of heterozygous variants, but additional details were not described (11). Nabieva et al. presented a method that can calculate and correct for the MCC fraction based on the genotyping data from both parents. They elegantly estimated the amount of contamination from positions in the genome where one parent is homozygous for the reference allele and the other parent is homozygous for the alternative allele (12). In contrast to the method presented by Nabieva et al. (12) and Yoon et al. (13), our WES contamination method does not require parental DNA, which means it can be used for all prenatal WES analysis (both trio and single) and even for postnatal samples. This is a big advantage of our method, since parental DNA, especially paternal, is not always available in the prenatal setting. The WES contamination method only detects the presence of contaminating DNA and not its origin. In our laboratory, we now perform additional MCC testing by STR analysis only in cases where ≥5% contamination is detected. This is not to further quantify the level of MCC but to determine whether the contamination is maternal or foreign in origin. If the level of MCC as determined by WES is acceptable for the type of analysis that is requested (which is usually the case), the diagnostic report is issued without delay and the result of the additional testing is only used for laboratory quality issues. Only in cases where WES analysis shows an MCC level of ≥21% is the gold standard test performed to determine the maternal/fetal DNA ratio. As shown during the post-implementation evaluation, using a ≥5% cutoff level, using this strategy a sequential MCC test is only necessary in 0.6% (4/631) of the prenatal samples.

With the WES method, contamination with DNA from an unrelated person can also be detected. Recently, we WES-analyzed 13 postnatal DNA samples isolated from EDTA blood, in which known amounts of contamination were present and quantified with the gold standard test. Contamination was indeed detected in all samples, although the amount of contamination was slightly overestimated compared to the gold standard test (data not shown). This can be explained by the fact that contaminating DNA from unrelated persons elicits a greater shift in the VAF than contaminating DNA from related persons, and the cutoff values we set for the 2 contamination indicators were based on the presence of maternal DNA. Nevertheless, for postnatal samples, the WES contamination method can be regarded as a useful additional quality control (QC) check.

A potential disadvantage of the WES contamination method is that a supposed prenatal sample of (almost) 100% maternal origin will not cause a shift in the distribution of the VAF, and therefore will not be detected with the WES method. Fortunately, this is a very rare event. Moreover, laboratories that receive prenatal samples should have visual inspections in place to determine whether AF or CV are potentially contaminated, although contamination is not always visible (like in the heavily contaminated prenatal sample in our post-implementation cohort). Whenever possible, any visible contamination should be removed, and this information should be shared with the DNA diagnostic laboratory to prepare them for potentially maternally contaminated DNA. Furthermore, very high levels of maternal contamination can still be detected with WES data analysis, either because of a nonmatching gender (in case of a male fetus), and/or because of a lack of paternal calls when performing trio analysis, which is standard practice in many laboratories for prenatal analyses.

Another potential disadvantage of the WES method is that with the selected cutoff values, the upper detection limit of the method is 20%. Samples containing ≥21% contamination are all predicted to contain “≥21%” contamination, without further quantification. For clinical decision-making this is irrelevant, as whenever a sample contains >20% contamination, it cannot be used reliably for any subsequent analysis, and another sample has to be collected and analyzed. In case knowledge about the maternal/fetal DNA ratio is required, a separate MCC test is necessary, but this is always performed if ≥5% contamination is detected with WES.

An important advantage of the WES contamination method is that, for the majority of samples, there is no need for a separate MCC test, thus reducing laboratory process complexity and resources. Moreover, avoiding a separate MCC test also reduces the amount of fetal DNA necessary.

The disadvantage of not performing separate MCC testing prior to initiating the WES process might be that there is a risk of performing WES in vain, if from the WES data it becomes clear that a prenatal sample is contaminated with such a high level of maternal DNA that it is not suitable for the analysis that is requested. This, however, rarely happens and, as discussed above, will probably be noted during visual inspection of prenatal samples. Moreover, this risk is outweighed by the fact that a separate MCC test does not have to be performed anymore for vast majority of samples. Exceptions to this are fetal samples from consanguineous parents, since it is not known how MCC affects the performance of the WES contamination method in such cases. Therefore, we have made it standard practice to perform the STR-based contamination test in cases where consanguinity is indicated on the application form.

The WES contamination method is now part of our routine WES analysis pipeline, for both prenatal and postnatal samples, and the estimated percentage of contamination is depicted in the QC data for all samples. As there is no need for separate MCC testing for the majority of samples, our method contributes to decreased laboratory process complexity and reduced costs. In contrast to other MCC tests, simultaneous analysis of maternal DNA is not necessary for samples with <5% of contaminating DNA. Future revisions of MCC testing guidelines should thus consider inclusion of this type of MCC testing. An additional benefit of the method is that it provides an additional QC check for all types of contamination that might have arisen during sample collection, DNA isolation, or library preparation.

Supplemental Material

Supplemental material is available at Clinical Chemistry online.

Nonstandard Abbreviations

MCC, maternal cell contamination; WES, whole exome sequencing; STR, short tandem repeat; AF, amniotic fluid; CV, chorionic villi; SNV, single nucleotide variant; VAF, variant allele fraction.

Author Contributions

The corresponding author takes full responsibility that all authors on this publication have met the following required criteria of eligibility for authorship: (a) significant contributions to the conception and design, acquisition of data, or analysis and interpretation of data; (b) drafting or revising the article for intellectual content; (c) final approval of the published article; and (d) agreement to be accountable for all aspects of the article thus ensuring that questions related to the accuracy or integrity of any part of the article are appropriately investigated and resolved. Nobody who qualifies for authorship has been omitted from the list.

Sanne Smeekens (Formal analysis-Equal, Methodology-Equal, Writing—original draft-Lead), Raoul Timmermans (Data curation-Equal, Formal analysis-Equal, Methodology-Equal, Visualization-Lead), Dineke Westra (Conceptualization-Equal, Writing—review & editing-Equal), Christian Gilissen (Conceptualization-Equal, Supervision-Equal, Writing—review & editing-Equal), and Brigitte H.W. Faas (Conceptualization-Equal, Methodology-Equal, Supervision-Equal, Writing—review & editing-Equal)

Authors’ Disclosures or Potential Conflicts of Interest

No authors declared any potential conflicts of interest.

Role of Sponsor

No sponsor was declared.

Acknowledgments

We thank all the clinical laboratory geneticists, bioinformaticians, technicians, and the Genome Sequencing facility of the Radboud university medical center, Department of Human Genetics, Nijmegen, the Netherlands and Maastricht University Medical Centre, Department of Clinical Genetics, Maastricht, the Netherlands.

References

Author notes