The accuracy of chromosomal microarray testing for identification of embryonic mosaicism in human blastocysts

Chromosomal mosaicism has been associated with human morbidity but the phenotypic manifestations are dependent on mosaicism level, the chromosome involved, and tissue distribution. An accurate assessment of the prevalence of mosaicism in the newborn population is uncertain because of ascertainment bias toward abnormal clinical presentations. However, in prenatal specimens large studies indicate that mosaicism is found in 0.20-0.25% of amniotic fluid [1, 2] and 0.8-2% of chorionic villus samples [3–6]. Chromosomal mosaicism is significantly higher in IVF created embryos than in other prenatal specimens. In a recent literature review and meta-analysis of human preimplantation embryo chromosomal mosaicism, van Echten-Arends et al. [7] reported that mosaicism is the most common chromosome constitution in spare IVF embryos and was found in 73% of all embryos. Of these embryos, 67% were at cleavage stage and most of the studies used FISH to determine chromosome constitution as well as spare embryos that were deemed unsuitable for transfer. Using chromosomal microarrays with cleavage stage embryos, Mertzamidou et al. [8] also found 71% of embryos to be mosaic. Most previous studies assessed mosaicism at cleavage stage from spare or arrested embryos. There is currently little data on chromosomal mosaicism derived from progressing day 5/6 hatching blastocysts.

Chromosomal microarrays have replaced FISH and comparative genomic hybridization (CGH) based methods for assessing aneuploidy in developing embryos [9, 10] and aneuploidy has been implicated in implantation failure and pregnancy loss by spontaneous miscarriage following IVF embryo transfer [11, 12]. The relatively low levels of mosaicism reported in prenatal specimens and the decreasing frequency of mosaicism with increased gestational age suggests the presence of selective pressure against embryos with high levels of mosaicism for ongoing pregnancy. Although mosaicism has been reported in embryos using CMAs with TE biopsies [13, 14], the effects of this mosaicism on embryonic development, implantation, and pregnancy outcome is currently unknown and the degree to which a CMA result from a small sample of embryonic cells at day 5/6 accurately reflects embryonic mosaicism levels has not been adequately described. We report here on the limits of mosaicism detection using cell mixing experiments, the accuracy of mosaicism detection from FISH follow-up studies, and estimate the prevalence of mosaicism at the blastocyst stage of development.

Chromosomal microarrays have documented aneuploidy and mosaicism to be common at the blastocyst stage of development [7, 13–16]. While arrays can confidently identify non-mosaic aneuploidy, there is less certainty regarding the technical limits of detection of mosaicism especially when the biopsy cell number is limited and whole genome amplification is required. Because microarray data is dependent on array platform, the quality of DNA and software algorithms, it is important that individual laboratories develop their own protocols for assessing mosaicism. This has been recommended by both the American College of Medical Genetics and the European best practices guidelines for constitutional microarray testing [17, 18]. As a prerequisite for assessing the accuracy of arrays for the identification of embryonic mosaicism, we first determined the limits of detection of mosaicism by performing reconstitution experiments using DNA mixed from normal and aneuploid samples and analyzing CMA results for mosaicism detection using BlueGnome’s 24sure array and data generated by BlueGnome’s BlueFuse Multi software.

In our experiments we attempted to identify and characterize the approximate lowest levels of mosaicism that could be confirmed from follow-up studies as biologically real. We interpreted the presence of whole chromosome mosaicism when a clear visual deflection of all hybridization signals above or below the log₂ 0 ratio value was discernible for any chromosome. For chromosome gains, mosaicism as low as between 25-37% can be observed which corresponds to a log₂ ratio of approximately +0.13 and for chromosome losses, mosaicism as low as between 37-50% can be observed which corresponds to a log₂ ratio of approximately -0.15 (see Additional files 1, 2, 3 and 4). There was some variability around these cut-off levels, as expected, because BAC arrays have less reproducibility and more batch-to-batch variation when compared to oligonucleotide constructed arrays [19]. Variation in array lots and in whole genome DNA amplification with small samples can also influence log₂ ratios. In addition, there are also likely to be minor chromosome specific log₂ ratio differences between different individual chromosome trisomies and monosomies due to variation in the number of probes for each chromosome and their unique hybridization properties. The predicted log₂ ratio for a chromosome gain is +0.58 [log₂ (3/2)] and that for a chromosome loss is -1.0 [log₂ (1/2)]. It might be expected, then, that deflection of hybridization signals below the log₂ ratio 0 value would be more prominent for chromosome losses compared to gains and that mosaicism for losses would be relatively easier to identify than for gains. Others have demonstrated that copy number variation (CNV) duplications are obscured by lower levels of maternal cell contamination than for similarly sized CNV deletions [20]. In patient samples displaying both a chromosome gain(s) and loss(es), we usually find the deviation from the log₂ 0 line is greater for chromosome losses than for chromosome gains (unpublished observations). In these situations, losses and gains were determined from the same pool of amplified DNA. However, in our reconstruction experiments where we did not detect lower levels of mosaicism for chromosome losses compared to chromosome gains, these results are likely to be an experimental artifact due to the combining of different pools of amplified DNA products where there could have been minor differences in the quality of the DNA prior to whole genome amplification. The results we obtained for chromosome gains are consistent with those of Mamas et al. [21] who used the same array platform and software but analyzed DNA extracted from mixed populations of aneuploid and normal cells to simulate mosaicism in PGD specimens. They noted a shift in the log₂ ratio when mosaicism reached 25% and confidently interpreted the presence of mosaicism when it was present at the 50% level for chromosome gains. They did not present data for chromosome losses. Northrop et al. [14] combined cells, in graded proportions, from normal male and female cell lines to simulate mosaic blastocyst biopsies and then analyzed these mixed cell populations using a 262 K SNP microarray. Their results indicated that identification of monosomy X mosaicism (male cells) was detectable when > 25% of cells in the population were male. Scott et al. [22] mixed DNA samples extracted from cytogenetically normal and abnormal sources, without whole genome amplification, to determine the limits of mosaicism detection using a 44 K oligonucleotide array. Mosaicism as low as 10% for both gains and losses of whole chromosomes was detectable when data from dye-reversed replicates were combined but this detection limit rose to 20-30% mosaicism in the absence of dye-reversed replicates. While high resolution arrays may detect lower levels of mosaicism, the primary purpose of array testing in the PGD setting is to identify whole chromosome aneuploidy without the additional uncertainty associated with the interpretation of the clinical significance of small copy number variants. Therefore, with this clinical objective in mind and with the technical limitations imposed by small sample sizes and the requirement for whole genome amplification, embryo mosaicism of less than approximately 25% will not likely be detectable.

After establishing threshold levels for mosaicism identification, these criteria were applied to classify embryos as euploid, non-mosaic aneuploid, or mosaic aneuploid. Follow-up FISH studies of our abnormal embryos were evaluated to assess the diagnostic accuracy of CMA testing. In a series of 47 embryos with follow-up data, 26 embryos were considered to be non-mosaic by CMA. While FISH confirmed whole chromosome aneuploidy in all of the embryos (100%), 13 embryos in this group (50%) were mosaic by FISH for at least one chromosome. Among the possible reasons for this observation is unequal distribution of mosaic cells between the inner cell mass (ICM) and TE or the effect of sample size. Recent studies of cells derived from TE and ICM and analyzed by CMAs by Johnson et al. [23] and Northrup et al. [14] do not support the preferential segregation of aneuploid cells to the TE. These studies are also consistent with an earlier report by Evsikov and Verlinsky using FISH [24]. Because more embryonic cells are available for reanalysis by FISH compared to the limited number of cells in a TE biopsy, it is likely that some embryos diagnosed as non-mosaic aneuploid by CMA will display mosaicism by FISH due to sampling error. This observation has been reported by Fragouli et al. [13] when they compared CGH data and FISH reanalysis results. Therefore, we reported an embryo as confirmed abnormal if at least one chromosome in an embryo identified by CMA was also identified as abnormal (non-mosaic or mosaic) by FISH. These criteria are similar to that utilized by Gutierrez-Mateo et al. [25] in their study of embryo reanalysis by FISH. Another consideration regarding mosaic embryos is that embryos can be aneuploid-aneuploid or aneuploid-euploid. While the clinical significance of low levels of mosaicism is uncertain, most likely aneuploid-aneuploid mosaic embryos (no euploid cells) will not lead to viable live births. In our study all mosaic embryos were aneuploid-euploid.

If there is an equal distribution of cells within the embryo, a large biopsy will inherently better reflect the total embryonic population and will be less susceptible to sample bias compared to a small biopsy. Of the 13 embryos in the CMA non-mosaic category which by FISH follow-up were identified as mosaic, 10/13 (78%) of these embryos had three or less cells in the biopsy. Among 21 embryos identified as mosaic by CMA, 20 (95%) were confirmed as abnormal by FISH. Interestingly, one embryo in this group was non-mosaic by FISH. The single case with a discordant result had 7% of nuclei with an aneuploid FISH signal which was slightly below our threshold for scoring this embryo as abnormal. In this study we only attempted to confirm mosaicism by FISH for chromosomes identified by CMA to be abnormal. While it is possible that this approach underestimated the number of mosaic chromosomes present in embryos, this is not likely because in our preclinical validation studies we established high analytical sensitivity using CMA. No FISH follow-up studies were performed on embryos identified by CMA to be euploid because these embryos were generally not available for further study and because our preclinical validation experiments demonstrated high analytical specificity.

CMA testing of TE biopsies from unselected blastocyst populations demonstrates high levels of chromosomally abnormal embryos. In a recent randomized clinical trial, Yang et al. [16] reported aneuploidy in 56.6% of blastocysts from young, good prognosis patients. With aCGH(CMA) or CGH, Fragouli et al. [13] reported 42.3% blastocyst euploidy, 30% non-mosaic aneuploidy, and 32.4% mosaic aneuploidy. Using the mosaicism thresholds described above, we analyzed 551 day 5/6 embryos by CMA. Of these, 52.5% were diagnosed to be euploid and 47.5% to be chromosomally abnormal. In the abnormal embryo population, 40.5% were non-mosaic aneuploid, 31.6% mosaic only, and 27.9% of embryos displayed non-mosaic and mosaic aneuploidy together. However, based on our FISH follow-up studies that showed that about one half of CMA diagnosed non-mosaic embryos were, in fact, mosaic because of skewed sampling, then the size of the mosaic class of embryos would be expected to be larger. When the same CMA data were analyzed by age group, younger women (25-34 years old), as expected, had the highest percent of euploid embryos (59.8%) and older women (40-42 years old) had the lowest percent of euploid embryos (24%). The ratio of non-mosaic aneuploid embryos to euploid embryos was 0.24 in the youngest age group and 1.67 in the oldest age group. This observation is consistent with a maternal age effect for meiotic non-disjunction. The ratio of mosaic only embryos to euploid embryos was relatively constant among the different age groups: 0.26, 0.35, and 0.33 for the 25-34, 35-39, and 40-42 year old patients, respectively. While the number of embryos available for study for the 40-42 year old group was small and thus insufficient for statistical comparison, there does not appear to be striking differences between these groups for the mosaic only category. If larger studies confirm this observation, then mitotic non-disjunction leading to mosaicism would appear to be a random event unrelated to maternal age. With demonstrated maternal age effects on meiotic non-disjunction as well as egg cohort size, the practical significance of a steady rate of mitotic non-disjunction across age groups is that it can lead to a less complicated estimation of the number of normal blastocysts that can be expected given the number of fertilized eggs available and a women’s maternal age.

This study confirms that mosaicism is common in day 5/6 blastocysts and that CMA testing can identify a significant proportion of these embryos. FISH follow-up studies confirm that mosaicism suggested by CMA profiles is not likely to be due to technical artifacts of the arrays and thus the likelihood that normal embryos are misidentified and not transferred is reduced. Some abnormal mosaic embryos, however, may not be identified because the TE biopsy may not be representative of the embryo and it may yield only euploid cells. It is important, then, that laboratories perform pre-clinical mixing experiments to validate the performance characteristics of their array platform and to establish detection thresholds for mosaicism in order to reduce the chances of transferring abnormal embryos. Preliminary studies in our laboratory using next generation sequencing suggest that this approach can also be sensitive in the identification of blastocyst mosaicism.

In this study we report the data from reconstitution experiments using DNA mixed from normal and aneuploidy specimens to define the limits of detection of mosaicism using CMA as well as the accuracy of aneuploidy and mosaic aneuploidy identification in blastocyst TE samples. Our findings underscore the challenges in the identification of embryos with low levels of chromosomal mosaicism and the importance of obtaining sufficient cells in the TE specimen to minimize the effects of skewed biopsy sampling.