Noninvasive prenatal diagnosis of Mendelian disorders for consanguineous couples by relative genotype dosage

Noninvasive prenatal diagnosis relies on the presence in maternal blood of circulating cell-free fetal DNA released by apoptotic trophoblast cells. Widely used for aneuploidy screening, it can also be applied to monogenic diseases (NIPD-M) in case of known parental mutations. Due to the confounding effect of maternal DNA, detection of maternal or biparental mutations requires relative haplotype dosage (RHDO), a method relying on the presence of SNPs that are heterozygous in one parent and homozygous in the other. Unavoidably, there is a risk of test failure by lack of such informative SNPs, an event particularly likely for consanguineous couples who often share common haplotypes in regions of identity-by-descent. Here we present a novel approach, relative genotype dosage (RGDO) that bypasses this predicament by directly assessing fetal genotype with SNPs that are heterozygous in both parents (frequent in regions of identity-by-descent). We show that RGDO is as sensitive as RHDO and that it performs well over a large range of fetal fractions and DNA amounts, thereby opening NIPD-M to most consanguineous couples. We also report examples of couples, consanguineous or not, where combining RGDO and RHDO allowed a diagnosis that would not have been possible with only one approach.


| INTRODUCTION
Noninvasive prenatal diagnosis (NIPD), the analysis of fetal circulating cell-free DNA (ccfDNA) in maternal blood, is becoming one of the most frequently prescribed genetic tests in many countries 1 since, contrarily to more invasive prenatal techniques, it does not impart a risk of miscarriage.[4][5][6][7][8][9] A major challenge in analyzing fetal ccfDNA is that it only represents a minor fraction of maternal ccfDNA.The fetal fraction (FF, fetal ccfDNA/total ccfDNA) varies during the course of the pregnancy but is rarely higher than 30%.While detection of a paternal mutation, absent in the mother, is relatively straightforward, determining whether a maternal mutation was transmitted to the fetus is more challenging since it is necessarily present in maternal ccfDNA.An initial solution, relative mutation dosage (RMD), involved measuring the mutant/normal allelic ratio at the mutation site. 10If the fetus inherited the mutation, both mother and fetus are heterozygous, and the allelic ratio should approach 50%.In case of a homozygous wild-type fetus, the allelic ratio diverges from 50% by a value equivalent to FF.While proof-of-principle papers demonstrated the validity of RMD, [11][12][13] a major hindrance toward its routine application is the typically low abundance of ccfDNA, which makes it very difficult to gather a statistically significant number of molecules from a single genomic position.
To overcome this problem, adjacent single nucleotide polymorphisms (SNPs) are included in the computation of a "haplotype ratio," that is used to determine which maternal haplotype was transmitted to the fetus: "high-risk" (with the mutation) or "low-risk" (wild-type). 2This method, relative haplotype dosage (RHDO), is now routinely used in several diagnostic laboratories in Europe with impressive results. 3,8wever, its reliance on SNPs makes it difficult to apply to consanguineous families, as suitable combinations of parental SNPs (one heterozygous parent, the other homozygous) can seldom be found.
Although rare in European countries, consanguinity is frequent in many regions in the world, such as Northern Africa, Turkey or Pakistan, where endogamous unions are customary for a number of cultural, religious and social reasons. 14Unions between first cousins or between uncle and niece, respectively sharing 1/8th and 1/4th of their DNA, result in numerous genomic regions of identity-by-descent (IBD), especially with several loops of consanguinity present over several generations.With the current worldwide population migrations, western laboratories increasingly encounter requests from consanguineous couples where both parents carry a specific autosomal recessive disease.Often, genetic diagnosis was established in an affected child, and the parents are seeking advice for subsequent pregnancies.In such situations, NIPD-M would be indicated but often proves impossible, as genomic regions of IBD do not contain the type of SNPs required for RHDO analysis.
Depending on the alleles present in each parent, SNPs can be divided into 5 types (Table I).Types 1 and 4 form the basis of RHDO analysis, 2 type 3 is only required in case of biparental mutations or when no type-1 SNPs are available, whereas type 2 is dispensable since background noise can also be estimated from invariant positions.
In their original description of RHDO, 2 Lo and coworkers suggested that type-5 SNPs might be used for consanguineous couples but did not provide a practical protocol, nor a specific the mathematical approach.To our knowledge, no analytical method based on type-5 SNPs, nor experimental results, have been published so far.
Couples with several consanguinity loops in the family (e.g., double first-cousins) are likely to share both haplotypes in the region of interest, and thus only type-2 and type-5 SNPs will be available for analysis, thereby preventing the use of RHDO.Here we propose a practical analytical framework (Figure 1) to leverage information provided by type-5 SNPs, so as to achieve diagnosis in consanguineous couples.This implies overcoming 3 major challenges: (1) measuring FF, (2) phasing type-5 SNPs, and (3) determining the fetal genotype.
Precise determination of FF is critical to evaluate the reliability of the test: a moderate deviation of the allelic ratio from 50% might be interpreted as significant when FF is low, but as nonsignificant when FF is high.However, reliable determination of FF requires SNPs with at least one allele unique to the father (type 1 or 3), which do not exist in regions of IBD.A possible solution when FF cannot be determined is using a worst-case scenario: assume high FF in case of a homozygous fetus and low FF when heterozygosity is suspected.This approach, however, considerably reduces sensitivity and is likely to make diagnosis impossible in many, if not most cases (examples provided in Supplementary Table 1).
A better alternative is to measure FF outside of the region of IBD.
For instance, we have successfully used a pan-genomic panel of 355 SNPs, selected for an average heterozygosity close to 50% in the general population, which can be mixed with the gene-specific panel and analyzed with minimal extra cost (see Supplementary File 1).
Statistically, 25% of these SNPs should be type-1 thus, even in highly consanguineous couples who may share up to half their DNA, it is virtually certain that enough SNPs will be available for FF determination.
Another challenge is phasing SNPs, that is, determining which allele belongs to which haplotype.This is best achieved by genotyping a prior conceptus of the couple: healthy or affected child, prenatal sample obtained invasively, fetal tissue from a terminated pregnancy, etc.
For type-5 SNPs, phasing is only possible if the prior conceptus was homozygous, as heterozygous SNPs do not allow to determine which parent contributed which allele.Since all SNPs in the target region are in linkage disequilibrium by design, there is a 50% risk that all are heterozygous, which would prevent phasing altogether.In practice, consanguineous parents generally consult because of a prior affected pregnancy; it is thus likely that this conceptus will be homozygous for the mutation and all adjacent SNPs.Otherwise, one can genotype the grandparents, with the same requirement for homozygosity.
T A B L E I SNP types and usage.When no appropriate relative is available for phasing, a seemingly obvious solution would be to infer parental haplotypes from an ethnically matched reference population.While extremely useful in population genetics, this approach is not reliable enough to be used in a diagnostic setting.Supplementary File 2 shows our attempts at phasing the GCK locus with two dedicated software packages: in both cases phasing errors were occasionally detected, which would have caused analytical failure or, less likely, an incorrect result.Furthermore, currently available reference populations are mostly of European ancestry, which is rarely the case for consanguineous couples.
A better solution is long-read sequencing, which allows determining large haplotype blocks by "walking" along the chromosome, from one heterozygous SNP to another, since long reads connecting two adjacent heterozygous SNPs determine their phase unambiguously.
Supplementary Figure 1 illustrates phasing of a 1 MB region by Nanopore sequencing: 8 haplotype blocks were properly phased, separated by stretches of homozygosity that prevented connecting blocks with bridging reads.Fortunately, it is easier to infer haplotype from a small number of large blocks than from hundreds of individual SNPs and the figure shows that the phase of all blocks could successfully be inferred.

| Principle of RHDO
Traditional RHDO analysis relies on the allelic ratio of type-4 SNPs to determine the inherited maternal haplotype.Furthermore, in case of a  Point p is of particular interest: it corresponds to the minimum number of DNA molecules required to achieve a diagnosis of heterozygosity and can be calculated from FF and the chosen likelihood threshold (Equation 8).Note that, due to experimental variations, the observed cumulative allelic ratio might slightly deviate from 0.5, thus more than p molecules may be required to achieve significance.An adjacent, but not identical point, p1, is the minimum molecular count required to achieve a diagnosis of homozygosity (Equation 9).Here also, experimental variations may cause the allelic ratio to deviate from the expected value, which may result in a statistically significant result with less than p1 molecules (as with the 4th SNPs in Figure 2B).
Although this might be genuine, for instance if FF was underestimated, we recommend to only validate results obtained with more than p1 molecules.

| Panel design and sequencing
For each gene, 100 SNPs were selected throughout a $

| Mathematical analysis
RGDO analysis was performed in Excel: type-5 SNPs were sampled in either direction from the mutation (not included in the count) and cumulative allelic ratio H1/(H1 + H2) was calculated by tallying DNA molecules for this SNP and all preceding SNPs, then plotted as a function of the number of DNA molecules tallied.
Threshold curves were built according to the following equations (rearranged from Lo et al. 2 ): where LR is the selected likelihood ratio, N is the number of DNA molecules, FF is fetal fraction and g = (1 + FF)/(1 À FF).
Once the possibility of a crossing-over was ruled out, a global likelihood ratio was calculated using the following algorithm: where r is the allelic ratio calculated with all type-5 SNPs in the region.
The minimum number of molecules for a diagnosis of heterozygosity is: The minimum number of molecules for a diagnosis of homozygosity is: Note that these two equations can easily be rearranged to calculate the minimal FF required to achieve diagnosis from a given number of molecules.
Detailed methods are available in the Supplementary Information.This couple had 2 daughters, both carrying the maternal mutation, and donated two plasma samples during the first pregnancy, at 12 and 30 weeks amenorrhea, and one during the second pregnancy, at 28 weeks.The respective FFs were measured at 8.1%, 17.6%, and 13.9% and classical RHDO analysis correctly predicted transmission of the maternal mutation for the three samples, with likelihood ratios of 1.7 Â 10 8 , 9.9 Â 10 22 , and 1.4 Â 10 7 , respectively. 15We also determined that, while both girls inherited the same maternal haplotype, they each received a different haplotype from their father (data not shown).

| Examples of RGDO analyses
Although this couple is not consanguineous, genotyping the family at the GCK locus revealed a 200 kb region of identity-by-state, containing 32 type-5 SNPs.While RGDO analysis was not strictly required in this case, we deemed interesting to compare its outcome with that of RHDO.Panels A and B demonstrate that, in the region of identity, fetus 1 is heterozygous while fetus 2 is homozygous for haplotype 1, with global likelihood ratios of 2.1 Â 10 6 , 10 Â 10 22 , and 1.9 Â 10 39 , respectively, for the three samples.The presence of a stretch of biallelic identity precludes the use of RHDO but enables RDGO, and conversely its absence enables RHDO.
Since the parents had only one haplotype in common on the telomeric side of the gene, we were able to apply standard RHDO analysis to this region (not shown) and confirmed that both parents had transmitted haplotype 1 to the fetus, effectively ruling out a crossing-over.
These examples demonstrate that RHDO and RGDO can be combined to reinforce each other and rescue situations in which using only one approach would be inconclusive.

| Investigation of critical parameters
Two parameters are known to be critical to the success of RHDO analyses: the FF and the number of DNA molecules analyzed, which in turn depends on the initial amount of DNA and the number of informative SNPs.We deemed important to verify that the same holds true for RGDO and that the effect of these parameters on the likelihood ratio fits the theoretical model (Figure 2D).
To investigate the dependency of RGDO on FF, we analyzed a consanguineous couple (double first cousins) with a healthy daughter and a couple of first cousins with five children and a history of two distinct autosomal recessive disorders.We genotyped the parents at several loci and found blocks of IDB in the CFTR and SOS1 regions in family 1 and in the FBN1 and GLDN regions in family 2. For family 1, we constructed synthetic samples by mixing various amounts of circulating ccfDNA from mother and daughter to simulate various FFs.For family 2, we selected two of the children, previously tested in our clinic and respectively homozygous and heterozygous at the loci of interest, and constructed a range of synthetic samples for each, using genomic DNA since ccfDNA was not available.Figure 4 demonstrates an exponential relationship (note the log scale) between FF and RGDO likelihood at all four loci.As a result, the diagnostic threshold of 1200:1 was often not reached with 5% and 6.7% FF, whereas diagnosis was possible with higher FFs (lower FFs are further illustrated in Supplementary Figure 5).In our experiments, the critical range of FF to reach threshold was 3%-14%.This is somewhat higher than what we observed with RHDO analyses, for which the threshold was usually around 3%-5% (unpublished observations).
The second critical parameter affecting likelihood is the number of DNA molecules analyzed (to avoid a confounding effect of this factor in Figure 4, the number of molecules was set to the lowest value in the series).To obtain a more precise appraisal of the effect of molecular counts, we analyzed the results of the 20% FF samples after subsampling, that is, adding a software filter to only include in the analysis a fraction of the available DNA molecules (see detailed methods in Supplementary Information).Figure 5A demonstrates an exponential relationship between number of analyzed molecules and global likelihood ratio, concordant with the theoretical model.Thus, both FF and molecular counts strongly influence the significance of RGDO analyses, implying that one might compensate for the other in an exponential manner (see mathematical model in Figure 2D).
The number of available SNPs is the main factor affecting molecular counts.Figure 5B demonstrates this fact on 2 samples in which a total of 21 type-5 SNPs were available at the GLDN locus.We randomly generated subsets comprising fewer SNPs and performed 10 000 RGDO tests for each number of SNPs, from 1 through 21.
The fraction of tests failing to reach the 1200:1 likelihood threshold was then plotted against the number of SNPs.As expected, FF plays a major role in the outcome and it can be seen that 9 SNPs sufficed to ensure success at high FF (16.7% and 20%), whereas lower FF (10% and 13.3%) required 17 to 21 SNPs.With an FF of 6.7%, even 21 SNPs proved insufficient to reach the required threshold (computations show that around 27 SNPs would be necessary).
It is worth emphasizing that no incorrect diagnosis was observed among the tests reaching significance.Supplementary Figure 4 shows that the fraction of erroneous results remains negligible down to five SNPs and that none of the erroneous results obtained with fewer SNPs reached a likelihood of 1200:1.Thus, while low FF or insufficient molecular counts can lead to test failure, they would not result in incorrect diagnosis.

| DISCUSSION
Here we propose an analytical framework for NIPD-M in consanguineous couples, or any couple with a stretch of biallelic IBD in the locus of interest preventing traditional RHDO.Our method is based on the analysis of type-5 SNPs, an approach initially suggested, but not implemented, by Lo et al. 2 We suggest strategies to overcome the main hurdles in this approach: determining FF and phasing the available SNPs.We then introduce RGDO, a mathematical approach similar to RHDO but relying on type-5 SNPs, the only informative SNP type when the parents share both haplotypes.Whereas RHDO independently determines the parental haplotypes transmitted to the fetus, RGDO aims at directly determining fetal genotype.This is possible because only 2 haplotypes are present at this locus, instead of the usual 4, allowing the determination of fetal genotype from the (im)balance between these 2 haplotypes.We show that RGDO provides diagnostic results with likelihoods similar to that of RHDO.As with RHDO, global likelihood in RGDO strongly depends on FF and on molecular counts, in accordance with the theoretical model, and diagnosis may be impossible if both these factors are suboptimal.
As FF is known to increase in the course of pregnancy, 16 a strategy to maximize it could be to delay testing, which might be particularly appropriate when the goal is to optimize pregnancy conditions (such as with GCK diabetes) or to prepare for a challenging delivery (e.g. with hemophilia).When therapeutic interruption is considered, though, it is preferable to perform prenatal testing as early as possible.
While testing before 12 weeks is likely to yield an FF inferior to 10%, it may be worth trying anyway since, in case of failure, it is always possible to repeat testing later.
Molecular counts depend on several factors: the initial amount of DNA in the reaction, the efficiency of the library system and the number of available SNPs.It is important to emphasize that a reliable way of tallying DNA molecules is fundamental and cannot be substituted with counting sequence reads (even after removing duplicates).In our view, it is imperative that the library system features molecular barcodes, a method to independently tag each DNA molecule in the sample, prior to amplification and sequencing. 17e initial amount of DNA is limited by the specifications of the library system and quite often by the low amounts of ccfDNA available.We generally obtain 10 to 30 ng DNA from a 10 mL blood sample, which theoretically corresponds to 2700 to 8100 copies of the genome.However, only a minority of these can be sequenced, due to the poor efficiency of most library systems: between 25% and 50% in our experience. 18Furthermore, software quality control routines generally discard a significant portion of molecules.Efficiency can be improved by increasing sequencing depth, since more supporting reads allow retaining more molecular barcodes (Supplementary Fig- ure 2).There is an upper limit around 10 000 reads, as enrichment reaches a plateau and excessively high sequencing depth may even result in "phantom molecules" originating from recurrent sequencing errors in the molecular barcode itself.
When sufficient ccfDNA is available, repeating the experiment may allow to accumulate enough molecules to reach threshold.For instance, the sample in Figure 3D had low FF and our initial attempt at RGDO failed to reach significance (Supplementary Figure 3).We thus prepared another library from the same sample and combining the molecular counts from both tests provided enough molecules to cross the 1200:1 likelihood threshold.
Given the above limitations, the best way to increase molecular counts is to test more SNPs.One caveat is that selecting multiple SNPs in close vicinity may result in all of them being type-2 SNPs if they fall within a stretch of homozygosity, which would make RGDO (and RHDO) impossible.This is unlikely in consanguineous couples with a deleterious recessive mutation, as the homozygous parent would be affected, but it may happen in other situations.Expanding the panel to encompass more distant SNPs decreases the likelihood of such a problem but increases the risk of a crossing-over within the region examined.As a compromise, we routinely design panels of 100 SNPs, spread over 1 Mb, although the number of SNPs could be increased if cost is not an issue.One advantage of RGDO is that, in highly consanguineous couples, 50% of the SNPs are expected to be of type 5, whereas with RHDO one only expects 25% informative SNPs.
Like RHDO, RGDO does not return results regarding the presence or absence of a mutation but indicates whether the fetus inherited the high-risk or low-risk haplotypes.An advantage of this approach is that these analyses can be applied to mutations undetectable in the short ccfDNA fragments (e.g., triplet repeat expansions, large deletions, inversions, etc.) or to loci where there is no mutation (for method validation).Like other laboratories offering RHDO, 3,4 we selected as diagnostic threshold a likelihood ratio of 1200:1.In the case of RGDO, this means that the most likely genotype is at least 1200 times more likely than the second most likely.This threshold was initially suggested to compensate for multiple testing in genome-wide analyses 2 and could probably be reduced when focusing on a single 1-Mbase locus, which would decrease failure rate.Yet, we show here that a threshold of 1200 virtually eliminates the risk of erroneous results.No laboratory test is 100% accurate but, in the case of highly consanguineous couples, a residual risk of 1/1200 may well be inferior to the risk of a yet undiscovered biparental mutation in a different gene.
An efficient noninvasive technique for prenatal testing is particularly important for consanguineous couples, who suffer a high reproductive risk due to shared mutations in recessive genes.It is thus particularly unfortunate that they often cannot be offered NIPD-M, as they lack the informative SNPs required for the current approaches.
Here we suggest a framework to overcome this limitation and propose a novel analytical approach, RGDO, which we show can be successfully applied in situations when the classical method fails.
Although there always remains a risk of analytical failure, it is not higher than for nonconsanguineous couples.We trust that our approach will prove useful to alleviate the considerable reproductive burden of consanguineous couples.

1
Workflow for practical application of NIPD-M.SETUP Parental haplotypes (colored bars) are determined by genotyping a first-degree relative, ideally a prior conceptus from this couple (see text for alternatives).NIPD-M is performed on ccfDNA extracted from maternal plasma.Depending on the presence/absence of a stretch of biallelic IBD, particularly likely in consanguineous couples, different approaches are used, which rely on different types of SNPs (colored boxes).[Colour figure can be viewed at wileyonlinelibrary.com] biparental mutation, type-3 SNPs must be used to determine the paternal haplotype by detecting paternal-specific alleles.Neither can be achieved with type-5 SNPs: since the paternal contribution is unknown, one cannot determine the allelic ratios corresponding to each maternal haplotype and it is impossible to determine the paternal haplotype in the absence of alleles unique to the father.However, this exercise may not be necessary in case of high consanguinity, since there are only two haplotypes shared by both parents instead of the usual four.We thus propose to bypass determination of parental haplotypes and to directly determine fetal genotype with type-5 SNPs, an approach we named "Relative Genotype DOsage" (RGDO).With traditional RHDO analysis, type 4-alpha and 4-beta SNPs must be analyzed separately, as the paternal allele influences the allelic ratio in different ways: the ratio tends toward either 0.5 or 0.5+(FF/2) for alpha SNPs and toward either 0.5-(FF/2) or 0.5 for beta SNPs.This distinction is not necessary with RGDO since all type-5 SNPs are of the same subtype and can be analyzed together to directly infer the 3 possible fetal genotypes: homozygous for haplotype 1, heterozygous, or homozygous for haplotype 2.

Figure 2
Figure2shows a graphical representation of RGDO, in which an increasing number of SNPs are analyzed, moving away from the mutation in either direction, and calculating the resulting allelic ratio from all DNA molecules tallied so far (Figure2A,B).The purpose of this representation is to detect possible meiotic recombination events near the mutation that might cause incorrect diagnosis (Figure1C, simulated crossing-over).Once crossing-over has been excluded, all DNA molecules can be combined into the computation of a global likelihood ratio (Equations 5-7).This graphical representation (Figure2B) is equivalent to a superposition of the two RHDO graphs pertaining to alpha and beta SNPs, and results in 3 diagnostic zones, determined by the relative representation of the two parental haplotypes, and corresponding to the 3 possible fetal genotypes.The horizontal lines correspond to the expected allelic ratios: 0.5 for a heterozygous fetus, 0.5+(FF/2) if homozygous for haplotype 1 (q1) and 0.5-(FF/2) if homozygous for haplotype 2 (q0).The four curves represent diagnostic thresholds for a given likelihood and are calculated from FF and the number of DNA molecules analyzed (Equations 1-4).They delimit three zones of uncertainty: zone A when the father transmitted haplotype 1, zone B for paternal haplotype 2 and zone C where the number of molecules analyzed is insufficient to reach the likelihood threshold.

FOKSTUEN
ET AL.F I G U R E 2 Graphical representation of RGDO results.(A) Cumulative haplotype ratio calculated by adding up allele counts from SNPs (diamonds) encountered while moving away from the mutation (not tallied) in either direction.Arbitrary values are shown for simplicity.(B) Graphical representation used to detect crossing-over events.For each SNP (diamonds) the x-coordinate is the number of molecules counted for this SNP and all preceding SNPs starting from the mutation (position 0) in either direction (only one is shown).The y-coordinate is the haplotype ratio H1/(H1+H2).Expected values are 0.5 for a heterozygous fetus, q1 = 0.5+(FF/2) for a homozygous H1 genotype and q0 = 0.5-(FF/2) for homozygous H2.The 4 curves define diagnostic thresholds for a given likelihood ratio (here 1200:1) and FF (here 20%).See text for details.(C) A crossing-over event (dashed line) was simulated by inverting allele counts for the last 3 SNPs, resulting in different genotypes on the left (homozygous H1) and right side (heterozygous) of the mutation.As the exact position of the crossing-over relative to the mutation cannot be determined (gray area), this situation would yield an inconclusive result.(D) Minimum theoretical conditions to achieve a likelihood of 1200, depending on fetal fraction and molecular counts.Minimal values for a diagnosis of homozygosity (point p1 in panel B) are marginally higher than for heterozygosity (point p).[Colour figure can be viewed at wileyonlinelibrary.com]F I G U R E 3 Examples of RGDO analyses.Fetal genotype determination and crossing-over exclusion by RGDO in 7 pregnancies from 6 consanguineous and nonconsanguineous couples.See text for details.[Colour figure can be viewed at wileyonlinelibrary.com]

Figure 3
Figure 3 shows examples of fetal genotype determination by RGDO in various loci for couples, consanguineous or not, who consulted our clinic with a history of dominant or recessive disease.Of special interest are panels A and B, which illustrate a family with a maternal p.Gly246Arg mutation in the GCK gene causing dominant monogenic diabetes, which requires different management during pregnancy These results are concordant with our RHDO analyses and were confirmed by genotyping the children after birth.They also demonstrate that RGDO can occasionally be used to reinforce RHDO in nonconsanguineous couples.F I G U R E 4 Likelihood dependency on fetal fraction.RGDO analyses in two consanguineous couple with stretches of IBD in the CFTR and SOS1 regions, respectively containing 11 and 31 type-5 SNPs, and in the FBN1 and GLDN regions, with 14 and 21 type-5 SNPs.Synthetic samples with various FF were built by mixing DNA from a child (circles: homozygous, diamonds: heterozygous) with that of the mother.The number of molecules was normalized to that of the lowest sample in the series.Global RGDO likelihood is plotted against FF.[Colour figure can be viewed at wileyonlinelibrary.com]F I G U R E 5 Likelihood dependency on molecular counts.(A) The samples with a 20% FF in Figure 4 were re-analyzed after bioinformatic downsampling to randomly reduce the number of available DNA molecules.RGDO likelihood is plotted against molecular counts.(B) Data from Figure 4 were reanalyzed by building randomly selected subsets of the available type-5 SNPs.For each number of SNPs, 10 000 subsets were tested and the fraction of tests failing to reach a likelihood of 1200 is plotted against the number of SNPs in the subset.[Colour figure can be viewed at wileyonlinelibrary.com]Panel F illustrates the converse situation, for a couple of first cousins both carrying a heterozygous p.Arg126* mutation in the AGTR1 gene, who lost their first child to autosomal recessive renal tubular dysgenesis.They requested invasive testing for their next three pregnancies and donated plasma for parallel NIPD-M testing during the third.Genotyping the parents and fetus #1 revealed a 700 kb region of identity-by-descent including the AGTR1 gene and containing 12 type-5 SNPs.RGDO analysis correctly predicted that fetus #3 was homozygous for parental haplotype 1, in this case wildtype AGTR1, with a likelihood ratio of 4.6 Â 10 5 .But because the region of biallelic IDB ends shortly after the AGTR gene there were only 2 type-5 SNPs on the telomeric side of the mutation, and we could not exclude a crossing-over between the mutation and the first informative SNP on the centromeric side.Based on the recombination rate in the area, we estimated the probability of such an event as 0.18%, implying a risk of 1/555 for an incorrect diagnosis.