Objective
We sought to develop a novel biochemical assay and algorithm for the prenatal evaluation of risk for fetal trisomy 21 (T21) and trisomy 18 (T18) using cell-free DNA obtained from maternal blood.
Study Design
We assayed cell-free DNA from a training set and a blinded validation set of pregnant women, comprising 250 disomy, 72 T21, and 16 T18 pregnancies. We used digital analysis of selected regions in combination with a novel algorithm, fetal-fraction optimized risk of trisomy evaluation (FORTE), to determine trisomy risk for each subject.
Results
In all, 163/171 subjects in the training set passed quality control criteria. Using a Z statistic, 35/35 T21 cases and 7/7 T18 cases had Z statistic >3 and 120/121 disomic cases had Z statistic <3. FORTE produced an individualized trisomy risk score for each subject, and correctly discriminated all T21 and T18 cases from disomic cases. All 167 subjects in the blinded validation set passed quality control and FORTE performance matched that observed in the training set correctly discriminating 36/36 T21 cases and 8/8 T18 cases from 123/123 disomic cases.
Conclusion
Digital analysis of selected regions and FORTE enable accurate, scalable noninvasive fetal aneuploidy detection.
The American Congress of Obstetricians and Gynecologists (ACOG) recommends that pregnant women be offered noninvasive screening for fetal chromosomal abnormalities. However, existing screening methods exhibit detection rates in the range of 90-95% and false-positive rates in the range of 3-5%. Thus, ACOG also recommends that patients categorized by screening as high risk for fetal aneuploidy be offered invasive testing such as amniocentesis or chorionic villus sampling. Although these invasive procedures are highly accurate, they are expensive and entail a risk of miscarriage. To address these limitations, several groups have pursued methods for noninvasive fetal aneuploidy detection.
For Editors’ Commentary, see Contents
See related editorial, page 269
Click Supplementary Content under the title of this article in the online Table of Contents
Initial efforts, which were focused on isolation and analysis of circulating fetal cells, turned out to be challenging. The realization that fetal nucleic acids are present in maternal blood spawned efforts to analyze cell-free DNA (cfDNA) for fetal conditions. In the last few years, massively parallel shotgun sequencing (MPSS) has been used to quantify precisely cfDNA fragments for fetal trisomy detection. Several groups have recently used this approach to identify fetal trisomy 21 (T21), and with less success, trisomy 18 (T18) and trisomy 13.
The chromosomal dosage resulting from fetal aneuploidy is directly related to the fraction of fetal cfDNA. For example, a cfDNA sample containing 4% DNA from a T21 fetus should exhibit a 2% increase in the proportion of reads from chromosome 21 (chr21) as compared to a normal fetus. Distinguishing these 2 scenarios with high confidence requires a large number (>93,000) of chr21 observations. Because MPSS is indiscriminate with respect to chromosomal origin, and because chr21 represents ∼1.5% of the human genome, ∼6.3 million uniquely mapped reads are required to ensure sufficient chr21 counts. Given typical MPSS mapping yields of ∼25%, this translates to 25 million raw sequencing reads per sample. This requirement constrains the throughput, cost efficiency, and clinical utility of MPSS for aneuploidy detection. For example, a recently launched product that detects T21 via MPSS has a list price of approximately $2700 per test.
Selective sequencing of relevant chromosomes can address these constraints. We recently described a novel assay, digital analysis of selected regions (DANSR), which enables highly multiplexed sequencing of selected loci from specific chromosomes of interest ( Appendix ). We used DANSR to evaluate loci on chromosome 18 (chr18) and chr21 in a set of subject samples whose aneuploidy status was known at the time of analysis and demonstrated accurate aneuploidy detection.
In this study, we extend DANSR to assay simultaneously polymorphic and nonpolymorphic loci in a single reaction, enabling estimation of chromosome proportion and fetal fraction. We describe a novel analysis algorithm, the fetal-fraction optimized risk of trisomy evaluation (FORTE), which uses this information to compute the likelihood of fetal trisomy in each subject. We demonstrate the power of this approach in a blinded set of 167 pregnant women, including 36 T21 and 8 T18 pregnancies.
Materials and Methods
Subjects
Subjects were prospectively enrolled upon providing informed consent, under protocols approved by institutional review boards. Subjects were required to be at least 18 years of age, to be at least 10 weeks’ gestational age, and to have singleton pregnancies. A subset of enrolled subjects, consisting of 250 women with disomic pregnancies, 72 with T21 pregnancies, and 16 with T18 pregnancies, was selected for inclusion in this study. The subjects were randomized into a training set consisting of 127 disomic pregnancies, 36 T21 pregnancies, and 8 T18 pregnancies, and a validation set consisting of 123 disomic pregnancies, 36 T21 pregnancies, and 8 T18 pregnancies. The trisomy status of each pregnancy was confirmed by invasive testing (fluorescent in situ hybridization and/or karyotype analysis). The trisomy status of the training set was known at the time of analysis; in the validation set, the trisomy status was kept blinded until after FORTE analysis.
DANSR assay
We designed DANSR assays against loci in the human genome as previously described. To assess chromosome proportion, we designed assays against 576 nonpolymorphic loci on each of chr18 and chr21, where each assay consisted of 3 locus-specific oligonucleotides: a left oligo with a 5′ universal amplification tail, a 5′ phosphorylated middle oligo, and a 5′ phosphorylated right oligo with a 3′ universal amplification tail. To assess fetal fraction, we designed assays against a set of 192 single nucleotide polymorphism (SNP)-containing loci on chromosomes 1-12, where 2 middle oligos, differing by 1 base, were used to query each SNP. SNPs were optimized for minor allele frequency in the HapMap 3 dataset ( http://hapmap.ncbi.nlm.nih.gov.easyaccess1.lib.cuhk.edu.hk/ . Accessed Feb. 14, 2012). Oligonucleotides were synthesized by Integrated DNA Technologies (Coralville, IA) and pooled together to create a single multiplexed DANSR assay pool.
DANSR product was generated from each subject sample as previously described ( Figure 1 ). Briefly, 8 mL of blood per subject was collected into a cfDNA tube (Streck, Omaha, NE) and stored at room temperature for up to 3 days. Plasma was isolated from blood via double centrifugation and stored at −20°C for up to a year. cfDNA was isolated from plasma using viral nucleic acid DNA purification beads (Dynal, Grand Island, NY), biotinylated, immobilized on MyOne C1 streptavidin beads (Dynal), and annealed with the multiplexed DANSR oligonucleotide pool. Appropriately hybridized oligonucleotides were catenated with Taq ligase, eluted from the cfDNA, and amplified using universal polymerase chain reaction primers. Polymerase chain reaction product from 96 independent samples was pooled and used as a template for cluster amplification on a single lane of a TruSeq v2 SR flow slide (Illumina, San Diego, CA). The slide was processed on an Illumina HiSeq 2000 to produce a 56-base locus-specific sequence and a 7-base sample tag sequence from an average of 1.18 million clusters/sample. Locus-specific reads were compared to expected locus sequences. An average of 1.15 million (97%) reads had <3 mismatches with expected locus sequences, resulting in an average of 854 reads/locus/sample.
Analysis of nonpolymorphic loci for chromosome proportion
Sequence counts were normalized by systematically removing sample and assay biases. Sequence counts follow a log normal distribution, so biases were estimated using median polish on log-transformed counts. A chr21 proportion metric was then computed for each sample as the mean of counts for selected chr21 loci divided by the sum of the mean of counts for selected chr21 loci and the mean of counts for all 576 chr18 loci. A chr18 proportion metric was similarly calculated for each sample. A standard Z test of proportions was used to compute Z statistics
where p j is the observed proportion for a given chromosome of interest in a given sample j , p 0 is the expected proportion for the given test chromosome calculated as the median p j , and n j is the denominator of the proportion metric.
Z statistic standardization was performed using iterative censoring on each lane of 96 samples. At each iteration, the samples falling outside of 3 median absolute deviations were removed. After 10 iterations, mean and SD were calculated using only the uncensored samples. All samples were then standardized against this mean and SD. The Kolmogorov-Smirnov test and Shapiro-Wilk test were used to establish the normality of the uncensored samples’ Z statistics.
Locus selection using training samples
Sequence count data from the training samples were first normalized as described above and previously. These samples were subsequently analyzed to select 384 of the 576 loci on chr21 and chr18 best able to discriminate T21 and T18 from normal samples. The 384 loci on each chromosome exhibiting the greatest residual difference between normal and trisomy samples were identified using Z statistics derived from individual loci for the test chromosome and all 576 loci for the comparison chromosome.
Analysis of polymorphic loci for fetal fraction
Informative polymorphic loci were defined as loci where fetal alleles differ from maternal alleles. Because DANSR exhibits allele specificities >99%, informative loci were readily identified when the fetal allele proportion of a locus was measured to be between 1-20%. A maximum likelihood estimate using the binomial distribution was employed to determine the most likely fetal fraction based upon measurements from several informative loci. The results correlate well (R 2 >0.99) with the weighted average approach presented by Chu and colleagues.
Aneuploidy detection using FORTE
The FORTE algorithm estimates the risk of aneuploidy using an odds ratio comparing a model assuming a disomic fetal chromosome and a model assuming a trisomic fetal chromosome. Let xj=pj−p0
x j = p j − p 0
be the difference of the observed proportion pj
p j
for sample j
j
and the estimated reference proportion p0
p 0
. FORTE computes:
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
