SHaploseek is a sequencing-only, high-resolution method for comprehensive preimplantation genetic testing

Abstract

Recent advances in genomic technologies expand the scope and efficiency of preimplantation genetic testing (PGT). We previously developed Haploseek, a clinically-validated, variant-agnostic comprehensive PGT solution. Haploseek is based on microarray genotyping of the embryo's parents and relatives, combined with low-pass sequencing of the embryos. Here, to increase throughput and versatility, we aimed to develop a sequencing-only implementation of Haploseek. Accordingly, we developed SHaploseek, a universal PGT method to determine genome-wide haplotypes of each embryo based on low-pass (≤ 5x) sequencing of the parents and relative(s) along with ultra-low-pass (0.2-0.4x) sequencing of the embryos. We used SHaploseek to analyze five single lymphoblast cells and 31 embryos. We validated the genome-wide haplotype predictions against either bulk DNA, Haploseek, or, at focal genomic sites, PCR-based PGT results. SHaploseek achieved > 99% concordance with bulk DNA in two families from which single cells were derived from grown-up children. In embryos from 12 PGT families, all of SHaploseek's focal site haplotype predictions were concordant with clinical PCR-based PGT results. Genome-wide, there was > 99% median concordance between Haploseek and SHaploseek's haplotype predictions. Concordance remained high at all assayed sequencing depths ≥ 2x, as well as with only 1ng of parental DNA input. In subtelomeric regions, significantly more haplotype predictions were high-confidence in SHaploseek compared to Haploseek. In summary, SHaploseek constitutes a single-platform, accurate, and cost-effective comprehensive PGT solution.

Figure 1

Validation of SHaploseek by sequencing single cells from members of two families. We inferred the haplotypes of grown-up children based on ultra-low-pass genome sequencing of single lymphoblast cells from the children and low-pass sequencing of the parents and a reference child. The “unsampled” sequencing depths of the parent and reference child are given in Table 1. Other sequencing depths were obtained by down-sampling. Haploseek data is also presented for reference. The accuracy of inference of both maternal and paternal transmitted haplotypes is reported for each child (legend). (a) For Family 1, we plot the proportion of the 200,484 genome-wide array SNVs with high-confidence haplotype prediction (marginal probability > 0.99 or < 0.01) in non-ROC (regions of consanguinity) sites (y-axis). Results are shown for various values of the sequencing depth of the parents and the reference child, as well as for Haploseek (x-axis). (b) The phasing accuracy (based on ‘ground-truth’ haplotypes inferred from bulk DNA) at non-ROC SNVs with high-confidence prediction. (c, d) Same as (a) and (b), for Family 2.

Figure 2

Validation of SHaploseek with clinical PGT embryo biopsies. We evaluated the accuracy of SHaploseek in 31 embryo biopsies from 12 PGT families. (a) In the four families with a child reference individual, we show the proportion of the 200,484 array SNVs with high-confidence haplotype prediction in non-ROC sites in both Haploseek and SHaploseek (box plots). We show results for both the unsampled sequencing data for the parents and reference child (see Table 1 for the sequencing depth of each individual), as well as for lower depths obtained by down-sampling. For each sequencing depth, the box plot represents 20 data points (two for each of the ten embryos), each showing the result of genome-wide prediction of either the maternal or paternal haplotype of one embryo. (b) Box plots for the haplotype phasing accuracy (measured as the concordance with Haploseek) at SNVs with high-confidence prediction in both Haploseek and SHaploseek, for different sequencing depth categories. (c, d) Same as (a) and (b), respectively, for the eight families with grandparental reference individuals. For each sequencing depth, the box plot represents 24 data points, one for each of the 21 embryos, except the three embryos from Family 31 (Table 1; Table S1) who contributed two data points each, because grandparents from both parents were sequenced.

Figure 3

Validation of SHaploseek with clinical PGT embryo biopsies after parental resequencing from low input DNA. We resequenced parent and reference genomes (child or embryo grandparent) at depth 1.5–5.2 × from 1ng of input DNA and predicted genome-wide haplotypes with SHaploseek for 14 embryos from five families (Table 1; Table S1). (a) Box plots of the proportion of array SNVs with high-confidence haplotype calls in non-ROC sites in both Haploseek and SHaploseek. The data points include 12 haplotypes for six embryos from ‘child’ families and 11 haplotypes from eight embryos from ‘grandparent’ families. (b) Box plots of the concordance between SHaploseek and Haploseek haplotype calls at high-confidence sites.

Figure 4

SHaploseek generates higher confidence subtelomeric haplotype predictions than Haploseek. (a) Box plots of the number of subtelomeric SNVs with assigned haplotype predictions. The subtelomeric regions are defined as being within 5Mb distance of an autosomal or chrX telomere or acrocentric centromere. The ‘resequenced’ SHaploseek data set includes 1038 subtelomeric regions, one on each side of each chromosome, from 23 maternal or paternal haplotypes for the 14 embryos in Fig. 3 (after properly accounting for ChrX requiring only maternal haplotyping). The remaining categories each include 2440 subtelomeric regions from 54 maternal or paternal haplotypes for the 36 embryos (or single cells) in Figs. 1 and 2. (b) Box plots of the proportion of subtelomeric SNVs that were sequenced in the embryo and had high-confidence haplotype prediction in SHaploseek/Haploseek, over the same set of embryos and haplotypes described in (a). We used the Wilcoxon signed-rank test to compare Haploseek with each SHaploseek dataset (p values on top of the plot). (c) Results of Haploseek and SHaploseek paternal haplotype prediction for the PKD1 gene-flanking subtelomeric 4.2 Mb portion of chr16p in embryo 113f.-2 of Family 47 (Table S1). For haplotype phasing of the embryo father’s pathogenic variant in PKD1, we collected DNA from the paternal grandmother of the embryo (with the same pathogenic PKD1 variant as the father). In the original Haploseek analysis, we sequenced the 113f.-2 embryo biopsy to depth 0.4 × and genotyped the parents and the paternal grandmother on arrays. For SHaploseek, we sequenced the parents and grandmother at depths 4.8–4.9x, and down-sampled to 4x, 2x, and 1x. We also resequenced these individuals at depths 1.5–3.1 × based on low input DNA (Table 1; see legend on the left of the plot). The paternal haplotypes are depicted in “marginal” and “prediction” plots for each analysis (Materials and Methods). The “marginal”, or posterior, probability indicates the degree of confidence with which the HMM is reporting the haplotype prediction, where a probability of 1 corresponds to certain transmission from the paternal grandmother, and a probability of 0 corresponds to the paternal grandfather. Marginal probabilities < 0.01 or > 0.99 are considered high-confidence. The marginal probabilities are plotted as light blue dots at SNV sites that were also successfully sequenced in the embryo. SNVs within green shaded rectangles have high-confidence marginal probabilities (here < 0.01). The “prediction” plots indicate the HMM’s binary haplotype predictions (the Viterbi paths), where light blue shaded segments indicate that the embryo haplotype around a given SNV matches that of the paternal grandfather (carrying the wild type nonpathogenic PKD1 allele). The approximate location of the PKD1 gene (chr16:2,138,711–2,185,899; hg19) is marked by a dashed vertical line. Note that the high-confidence region in Haploseek does not encompass the PKD1 gene and is much smaller than the high-confidence regions of SHaploseek. The wild type (paternal grandfather) PKD1-flanking haplotype in embryo 113f.-2 was confirmed by PCR-based PGT-M (Table S1).