WO2017093561A1 - Method for non-invasive prenatal testing - Google Patents

Abstract

The present invention is directed towards a method for non-invasive detection (and diagnosis) of fetal aneuploidy from a biological test sample comprising cell-free DNA obtained from a pregnant woman. More specifically the method for non-invasive diagnosis of fetal aneuploidy allows to stratify the likelihood of false positive and false negative results to predict maternal chromosome anomalies and to robustly detect frequent, pathogenic and recurrent copy number variations using a combination of discriminating factors, which are fetal fraction and z-score/signal intensity. More specifically, the invention is directed towards methods for detecting fetal aneuploidy from a maternal biological test sample, comprising the steps of (a) extracting cell-free DNA from said test sample; (b) analyzing the size distribution of the cell-free DNA within each sample and selecting a set of samples based on the size distribution of the DNA molecules within said samples, (c) performing massive parallel sequencing of the extracted cell-free DNA of step (a) to obtain sequence reads, (d) mapping the obtained sequence reads to the human genome for each sample; (e) calculating the z-score for the signal intensity for the sequence reads of a chromosome or chromosomal region of interest and comparing it to a z-score threshold value; (f) calculating the fetal fraction for the sequence reads of a chromosome or chromosomal region of interest and comparing it to a fetal fraction threshold value; and (g) detecting a fetal aneuploidy in said test sample or not, based on the combined interpretation of steps e) and f).

Description

Method for non-invasive prenatal testing

Field of the invention

The present invention is directed towards a method for non-invasive diagnosis of fetal an- euploidy from a biological test sample comprising cell-free DNA obtained from a pregnant woman.

Technical Background

Cell -free DNA (cfDNA) testing for fetal aneuploidies was broadly implemented for the common trisomies and the sex chromosome anomalies (SCAs). However, the approaches known from the state of the art identify only 75% to 85% of clinically relevant aneuploidies, although accurately.

Large-scale clinical studies and an updated meta-analysis comprehensively described the performance of cell-free DNA (cfDNA) testing for the common trisomies - 21 , 1 8 and 1 3 - and to a lesser extent for the sex chromosome anomalies (SCAs) (see references 1 -4, Gil MM et al, Ultrasound Obstet Gynecol 201 5; 45: 249-266 ; Zhang H et al. Ultrasound Ob- stet Gynecol 201 5; 45: 530-538; McCullough RM, et al. PLoS One 201 4; 9: e1091 73. doi: 10.1 371 /joumal-pone.01091 73 ; Bianchi DW, et al. Obstet Gyn 201 5; 1 25: 375- 382). However, the common trisomies comprise only about 75% of aneuploidies that can be detected by karyotyping in Down syndrome screen positive cases, and 85% including the SCAs (see reference 5 Davis C, et al. Prenat Diagn 2014; 34: 1044- 1048). Corroborating this notion, registry data showed that 1 7% of clinically relevant anomalies are missed if cfDNA testing is limited to the common trisomies and SCAs (see reference 6, Wellesley D, et al. Eur J Hum Genet 201 2; 20: 521 -526). Similarly, a review of 4Ό00 prenatal karyotype results revealed that 24% of the reported anomalies would not have been detected by contemporary cfDNA screening (see reference 7, Lebo RV, et al. J Transl Med 201 5; 1 3: 260. doi 1 0.1 1 86/s1 2967-01 5-0569-y). The main classes of anomalies that go undetected with the current generation of cfDNA screening methods comprise the rare autosomal trisomies, and the structural chromosome anomalies, specifically disease-causing copy number variation (CNV). Both types of aberrations generate positive results in conventional aneuploidy screening used to prescribe cfDNA testing. In fact, first trimester screening for trisomy 21 , 1 8 and 1 3 is also sensitive to a broad range of rare autosomal trisomies and chromosomal mosaicism (see reference 8, T rring N, et al. Prenat Diagn 201 5; 35: 61 2-61 9). Chromosomal microarray (CMA) platforms provide an incremental diagnostic yield of 5% in fetuses with increased nuchal translucency (NT) diagnosed by first trimester ultrasound and a normal karyotype (see reference 9, Grande M, et al. Ultrasound Obstet Gynecol 201 5; doi: 10.1002/uog.14880).

Independent follow-up of pregnancies considered unaffected by such cfDNA-tests indeed confirmed that rare pathogenic aneuploidies such as for example mosaicism for trisomy 22 are ignored (see reference 10, Neufeld- Kaiser WA, et al. BMC Medicine 201 5; 1 3: 1 29. doi 10.1 1 86/s1 291 6-01 -0374-8). Given the significant incidence of mosaic aneuploidy (see reference 1 1 , 1 2, Kalousek DK, et al. Monogr Pathol 1 991 ; 33: 228-256 ; Malvestiti F, et al. Prenat Diagn 201 5; doi: 1 0.1002/pd.4656) and the fact that knowledge on the clinical consequences of rare mosaic trisomies detected in amniocytes has made considerable progress (see reference 13, Wallerstein R, et al. Prenat Diagn 201 5; 35: 1 -7), routine detection of such anomalies in mainstream cfDNA clinical practice should be considered. In addition, the phenotypic impact of rare autosomal trisomies reaches beyond fetal mosaic trisomies; after trisomy rescue uniparental disomy (UPD) can cause fetal pathology even in diploid fetuses when chromosomes undergoing imprinting are concerned (see references 1 2, 1 4, Malvestiti F, et al. Prenat Diagn 201 5; doi: 1 0.1002/pd.4656, Eggermann T, et al. Trends in Molecular Medicine 201 5; 21 : 77-87). Placental trisomy accounts for a low but significant risk of intrauterine growth restriction (IUGR), and an even higher risk at birth for small for gestational age (SGA) infants (see reference 1 5, Robinson WP, et al. Prenat Diagn 2010; 30: 1 -8).

The population history (see reference 1 6, Sudmant PH, et al. Science 201 5. doi: 10.1 1 26/science.aab3761 ) and morbidity maps (see reference 1 7, Cooper GM, et al. Nat Genet 201 1 ; 43: 838-846. Corrected after print 27 August 2014; doi: 1 0.1038/ng.909) of deletion and duplication CNVs have been extensively described. It has been estimated that CNVs >400kb in size account for almost 1 5% of disease burden in children affected by intellectual disability and congenital anomalies (see reference 1 7, Cooper GM, et al. Nat Genet 201 1 ; 43: 838-846. Corrected after print 27 August 201 4; doi: 1 0.1038/ng.909). A meta-analysis confirmed the incremental diagnostic yield of performing MCA analysis for cases with increased NT after first trimester ultrasound (see reference 9, Grande M, et al. Ultrasound Obstet Gynecol 201 5; doi: 1 0.1 002/uog.1 4880). Implementation of routine CNV detection using cfDNA sequencing posed challenges, however. Initial studies suggested that a higher genomic coverage was required to detect CNVs than for numerical chromosome anomalies (see references 1 8, 1 9, Jensen TJ, et al. Clin Chem 201 2; 58: 1 148- 1 1 5 1 ; Srinivasan A, et al. Am J Hum Genet 201 3; 92: 1 67- 1 76). Also, experimental protocols robustly detecting defined CNV size classes that take the fetal fraction into account and use a genomic coverage similar to that of numerical anomalies were not available until recently (see reference 20, Zhao C, et al. Clin Chem 201 5; 61 : 608-61 6). A clinical follow- up study effectively showed that cfDNA-testing can be extended to include reliable detection of deletion CNVs (see reference 21 , Helgeson J, et al. Prenat Diagn 201 ; doi: 1 0.1 002/pd.4640).

The inventors previously described a technical validation study using cfD A screening with low genomic coverage, and robustly detecting a broader array of anomalies comprising the common trisomies, the SCAs, the rare autosomal trisomies (RAT), as well as deletion and duplication CNVs (see reference 22, Guex N, et al. Prenat Diagn 201 3 ; 33: 707-710). Importantly, the algorithms were optimized to allow for uniformly robust detection of numerical anomalies of autosomes sex chromosomes and CNVs, similarly to a more recent report (see reference 24, ersy E, et al. Pub Health Genomics 201 5; doi: 0.1 1 59/000435780).

To sum up: Common trisomies - trisomies 21 , 1 8 and 1 3 - comprise about 75% of all an- euploidies detected by karyotyping in Down syndrome screen positive cases. The common trisomies and sex chromosome abnormalities amount to about 85%.1 7% of clinically relevant anomalies are missed if cfDNA testing is limited to the common trisomies and SCAs. The review of prenatal karyotypes shows that a high percentage of reported anomalies would not have been detected by contemporary cfDNA screening.

There is therefore a demand for improved non-invasive prenatal tests.

The inventors provide for the first time such improved tests covering a broader array of an- euploidies, including the rare autosomal trisomies and the deletion and duplication copy number variations (CNVs). The reliability of the new test method was verified with a more complete and stratified follow-up by amniocentesis. Summary of the Invention

The present invention is directed towards a method for non-invasive detection (and diagnosis) of fetal aneuploidy from a biological test sample comprising cell-free DNA obtained from a pregnant woman. More specifically the method for non-invasive diagnosis of fetal aneuploidy allows to stratify the likelihood of false positive and false negative results to predict maternal chromosome anomalies and to robustly detect frequent, pathogenic and recurrent copy number variations. The methods of invention allow to identify false positives and false negatives using a combination of discriminating factors, which are fetal fraction and z-score/signal intensity.

More specifically, the invention is directed towards methods for detecting fetal aneuploidy from a maternal biological test sample, comprising the steps of (a) extracting cell-free DNA from said test sample; (b) analyzing the size distribution of the eel I -free DNA within each sample and selecting a set of samples based on the size distribution of the DNA molecules within said samples, (c) performing massive parallel sequencing of the extracted cell-free DNA of step (a) to obtain sequence reads, (d) mapping the obtained sequence reads to the human genome for each sample; (e) calculating the z-score for the signal intensity for the sequence reads of a chromosome or chromosomal region of interest and comparing it to a z-score threshold value; (f) calculating the fetal fraction for the sequence reads of a chromosome or chromosomal region of interest and comparing it to a fetal fraction threshold value; and (g) detecting a fetal aneuploidy in said test sample or not, based on the combined interpretation of steps e) and f).

In some embodiments, the step of selecting a set of samples based on the size distribution of the DNA molecules comprises a step of elimination of DNA molecules having a size greater than 200 bp, greater than 180 bp, greater than 1 50 bp, greater than 1 25 bp from the sample to obtain an enriched sample, wherein at least 90 wt%, preferably more than 95 wt% of the DNA molecules have a size of less than 200 bp, less than 1 80 bp, less than 1 50 bp, less than 1 25 bp for example between 1 1 5 and 1 25 bp.

In some embodiments, the step of performing a massively parallel sequencing includes the preparation of a sequencing library followed by sequencing.

In some embodiments, the sequencing step comprises sequencing at least 25 million sequences for each sample. In some embodiments, the chromosome of interest is chromosome 21 , chromosome 16, chromosome 18, chromosome 1 3 or chromosome 1 1 ,

The methods of invention allow to identify false positives and false negatives using a combination of discriminating factors, which are fetal fraction and z-score/signa! intensity. Thus, a signal intensity in step e) close to the z score threshold of >6x 3.3 and <5 indicates a false positive result.

Method according to any preceding claim, wherein a signal intensity in step e) close to the z score threshold of > 6x 3 and <5 indicates a false positive result.

Figure Legends

Figure 1 shows the aneuploidy classes detected: the absolute numbers of the aneuploidy classes are shown, from left to right the common trisomies (trisomy 21 , 18 and 1 3), the SCAs (monosomy X, triple X- and Klinefe!tersyndromes), the rare autosomal trisomies, and the deletion and duplication CNVs (structural anomalies).

Figure 2 shows the individual rare autosomal trisomies detected: the absolute numbers of the individual rare autosomal trisomies are shown, from left to right in decreasing order of frequency found (namely trisomy 7, 8, 22, 1 6, 1 7, 6, 1 5, 4, 3, 22, 20, 1 2, 1 1 , 10, 9).

Figure 3 shows the standardization of detection of Di George syndrome (DGS) duplication: average 1 SD signal intensities of the negative control (left), and of serial dilutions (from left to right 5%, 10%, 1 5%, 20%, 25%) of the CM A validated fetal case of maternal origin with a 3Mb duplication 22q 1 1 .2 (shown as 100% dosage on the far right). The detection thresholds are shown as dashed lines, the lower (orange) line represents the 99% confidence level (z-score +2.575829), the upper (red) line reflects the 99.9 % confidence level (z-score +3.290527).

Figure S 1 Internal positive control. Discrimination of samples is based on positive (red) and negative (pink and cyan) control reference values, with a defined, single genome-wide threshold that separates euploid (blue and pink dots) from aneuploid. The dotted lines correspond to probability thresholds of 1 /10000 (outer bolder dotted line), respectively 1 /1 000 (inner lighter dotted line), i.e. only one sample over 10000, respectively 1 000, statistically exceeds this threshold. The test sample corresponds to the bolder spot (indicated by black arrow), in the middle of each plot. The spots between the dotted lines, on the left and on the right of the test sample, correspond to euploid reference samples. The spots above the dotted lines, on the right of each plot, correspond to aneuploid reference samples. In this example, the experimental value (black target) coincides with the range covered by negative controls and hence is consistent with euploidy. Reference data define the final score comprising a value that needs to exceed the threshold (e.g. 10-3 and 10-4 corresponding to z-scores of +3.3 and + 3.9 in the graph) to call an aneuploidy (trisomy 1 3 in this case).

Figure S2 Test for chromosomal aneuploidy for a sample including a determination of the fetal fraction: A validated sample with trisomy 1 3 having a very low signal intensity/z- score (black bold spot, red arrow in right panel) corresponding to the lowest range of signal intensities/z-scores obtained in the technical validation study (red dots) was used as the internal positive controKthe other spots in the plot represent euploid samples (samples between the dotted lines) and aneuploidy samples (samples above the dotted lines, on the right)). The thresholds of 1 /1000 and 1 /10000 are shown as dotted lines (light dotted line: 1 /1000; bolder dotted line: 1 /10000). This sample was run with every sequencing batch, and replication of the trisomy 1 3 was required for the samples in this batch to be reported. The average fetal fraction observed for such samples is between 3% - 5%, in this case The estimated fetal fraction for the tested sample is 4.5% (left-hand panel). Determination of the fetal fraction (left panel): Ten individual determinations are shown as individual spots, and as a whisker-and-box plot, showing the interquartile range. The lower dotted line represents the fetal fraction threshold of 3%, which is equivalent to the lowest-validated chromosomal dosage in a given reference set. Samples with a fetal fraction above the threshold are considered as having a sufficient fetal fraction for a reliable diagnosis. The upper dotted line represents the fetal fraction of 5.8%.

Figure S3 Training/calibration set: the predictor was trained based on positive correlations with Y-sequence tags in male fetuses. This predictor was independently developed but performs similarly from an in the meantime published version (Prenat Diagn 201 5; 35: 810- 81 5). Subsequently, the predictor was trained in an analogous fashion on single-end reads. The results obtained with single-end reads for the training set comprising 480 male samples is shown (correlation coefficient of 0.78 between predicted values on the y-axis and Y-sequence tags on the x-axis). Figure S4 Validation set: the predictor was validated on an independent dataset based on positive correlations with Y-sequence tags in male fetuses. The results obtained with single- end reads for the validation set based on a different, completely non-overlapping set of 436 male samples analyzed in clinical practice are shown (correlation coefficient of 0.64 between predicted values on the y-axis and Y-sequence tags on the x-axis).

Figure S5 The predictor was further validated i) for its ability to predict in a sex-independent manner, since it was trained on and derived from Y sequence tags (left panel) . An independent group of samples not overlapping with the training was selected for analysis, for which complete clinical and follow-up information was available. The predictions shown as box-plots for 214 male and 1 76 female fetuses were not significantly different (t-test P=0.63 ). The predictor was in addition validated using the same data set described in the left panel ii) for its ability to faithfully represent the distribution of % values in comparison with the literature (Zhang H et al. Ultrasound Obstet Gynecol 201 5; 45: 530-538) (right panel; n=390). The median (black dashed lines) and interquartile range (IQR; lighter blue dashed lines) were comparable with reported values (median 1 0.6% with IQR of 8.7- 1 2.7% four our data, and 1 1 .4% with IQR 8.5- 1 4.6%, n=397 for the reference) ( Fetal Diagn Ther 20 2; 3 1 :237-243).

Figure S6 Lower reliable threshold of fetal fractions to call aneuploidies. For a given an- euploidy, a lower reliable threshold of fetal fraction is defined, corresponding to the lowest signal intensities obtained for validated aneuploidies that reflect these values in directly proportional fashion (Am J Obstet Gynecol 201 2; 206:31 9. e 1 -9). Based on experiments, the fetal fractions for trisomy 3 were the lowest among the common trisomies, and thus the lower reliable threshold (range of 3% - 5.8% fetal fraction) was defined with one of the historically lowest signal intensities and fetal fractions obtained for trisomy 1 3. This is illustrated by two samples discordantly euploid and aneuploid having a fetal fraction in the same range of the lowest reliable values. In Fig S6a a "normal/unaffected result" with a fetal fraction of 4.4%, in Fig S6b a trisomy 1 3 sample with a comparably low fetal fraction of 4.5%. The signal intensity (black targets indicated by black and red arrows) allows to correctly classify both samples, one as being "unaffected/normal" and the other as being "affected" by trisomy 1 3. This holds true at a genome-wide scale, no other aneuploidy is called under these conditions. In the left plots in the right plots, the thresholds of 1 /1 000 and 1 /1 0000 are shown as dotted lines (light dotted line: 1 / 1 000; bolder dotted line: 1 / 1 0000) . The spots between the dotted lines represent euploid samples and the spots above the dotted lines on the right aneuploidy samples (samples above the dotted lines, on the right) ).

Figure S7 The same concept is demonstrated on replicates. Shown are independent predictions of the fetal fraction for the internal control sample 3522 with trisomy 1 3 (Fig S7a) that has a very low signal intensity/fetal fraction. The variability of the prediction falls within the lower reliable range (i.e. 3% - 5.8% ) delimited by dashed lines (Fig S7a). This consistently allows to correctly classify all replicates as aneuploid (Fig S7b), the three samples with the highest predicted values of the fetal fraction also obtain the highest aneuploid z-scores.

Figure S8. A set comprising the last 1 06 consecutive cfDNA enrichments performed in 201 5 was selected. Among those there were 67 male pregnancies, for which we compared the native Y-sequence tag number with that after enrichment (Fig S8a). As can be seen the procedure does enrich Y-sequence tags individually (dashed lines connecting native and enriched Y-tags) and for the whole group of samples significantly (interquartile range for the native Y-tags in grey (lower left box), and interquartile range for enriched Y-tags in turquoise (upper right box)). The pFF reflects this enrichment faithfully (Fig S8b), showing again individually (dashed lines connecting native and enriched fetal fractions) and group- wise a significant enrichment (interquartile range of native fetal fractions in grey (lower left box), and interquartile range of enriched fetal fractions in turquoise (upper left box)).

Figure S9. If a deletion/duplication CNV of maternal origin is suspected it is first confirmed by CMA on maternal blood. Once confirmed there is a 50% risk of transmitting it to the fetus, and if its potential pathogenicity is corroborated confirmatory amniocentesis is warranted. If a CNV of fetal origin is suspected, the decision for a workup with amniocentesis is taken based on its pathogenicity score (penetrance, phenotype, overlap with known disease genes, absence of annotation as a frequent polymorphic variant) .

Figure S 1 0. Example for the workup of a maternal CNV is shown. For a sample with an adequate fetal fraction of 1 0.4% a duplication of maternal origin of the Di George syndrome (DGS) region was suspected (Fig S 1 0a and S 1 0b). In a first step the breakpoint were mapped, and an overlap with the DGS-critical region was diagnosed ( Fig S10c). For this reason, it was first confirmed by MCA (Fig S 1 0d), followed by amniocentesis and confirmation that the fetus was affected as well. Detailed Description

The present invention is directed towards a method for non-invasive diagnosis of fetal aneupioidy from a biological test sample comprising cell-free DNA obtained from a pregnant woman by combining the interpretation of signal intensity /z-scores and fetal fraction. More specifically the method for non-invasive diagnosis of fetal aneupioidy allows to stratify the likelihood of false positive and false negative results to predict maternal chromosome anomalies and to robustly detect frequent, pathogenic and recurrent copy number variations.

The methods of invention allow to identify false positives and false negatives using a combination of discriminating factors, which are fetal fraction and z-score/signal intensity. More specifically the method comprises the steps of:

Method for detecting fetal aneupioidy from a maternal biological test sample, comprising a) extracting cell-free DNA from said test sample;

b) performing massive parallel sequencing of the extracted cell-free DNA of step (a) to obtain sequence reads,

c) mapping the obtained sequence reads to the human genome for each sample; d) calculating the z- score for the signal intensity for the sequence reads of a chromosome or chromosomal region of interest and comparing it to a z-score threshold value;

e) calculating the fetal fraction for the sequence reads of a chromosome or chromosomal region of interest and comparing it to a fetal fraction threshold value;

f ) diagnosing a fetal aneupioidy in said test sample or not, based on the combined interpretation of steps d) (calculating the z-score) and e) (calculating the fetal fraction).

More specifically, the method for detecting fetal aneupioidy from a maternal biological test sample, comprises the steps of a) extracting cell-free DNA from said test sample;

b) analyzing the size distribution of the cell -free DNA within each sample and selecting a set of samples based on the size distribution of the DNA molecules within said samples c) performing massive parallel sequencing of the extracted cell-free DNA of step (a) to obtain sequence reads,

d) mapping the obtained sequence reads to the human genome for each sample; e) calculating the z-score for the signal intensity for the sequence reads of a chromosome or chromosomal region of interest and comparing it to a z-score threshold value;

f ) calculating the fetal fraction for the sequence reads of a chromosome or chromosomal region of interest and comparing it to a fetal fraction threshold value;

g) detecting a fetal aneuploidy in said test sample or not, based on the combined interpretation of steps e) (calculating the z-score) and f) (calculating the fetal fraction).

Based on a review of a compound series of 6'388 consecutive singleton pregnancies with nearly complete and stratified follow-up using a test that covers an extended range of anomalies the new method was developed. Moreover, for a subset of consecutive cases the measurement of the fetal fraction was integrated (combined interpretation of signal inten- sity/z-score and fetal fraction). The latter was used i) to predict samples that had a too low fetal fraction as an important source of false negative results, ii) to identify samples with numerical anomalies or CNVs of likely maternal origin, iii) to help stratify the likelihood of true positive versus false positive results, and iv) to establish the detection threshold for recurrent pathogenic CNVs. The new method can be implemented in the clinics on a broader scale.

When testing for autosomal trisomies according to known methods, the test performance data are integrated in post-test risk calculation in the following way: pre-test risk x positive/negative likelihood ratio (based on false positive and false negative rates of the test) = post-test risk. According to the known methods one single likelihood ratio is used, assuming that it is the same for all data. In contrast, the new method of the invention is able to (i) provide modified positive likelihood ratios in 75 % of the false positive results, because those can be safely predicted based on low z-scores yet average fetal fraction and (ii) provide modified negative likelihood ratios for fetal fractions below and above the threshold (3- 5.8% )

When testing for SCAs according to known methods it was known that maternal mosaic monosomy/trisomy X can cause false positive results. The new method of the invention is able to (i) provide a robust prediction of maternal mosaic monosomy/trisomy X based on z-scores and fetal fraction = z-scores manifold too high for fetal fraction, and (ii) stratify the risk for fetal aneuploidy for maternal mosaic aberration carriers = not all need invasive diagnosis for confirmation, since risk increase yet low in absolute terms (low percentage range).

Previous testing for or prediction of CNVs was based on size and fetal fraction. The new method of the invention is able to (i) use negative and positive control cell-free DNA from cases with anomalies of defined size and derive dilution curve and z-scores (see figure) to predict at what fetal fraction, what fetal aberration of defined size is detected against background, and (ii) predict (using the same concept) maternal CNVs, since signal for non-physiological "fetal fractions" > 50% can be predicted from the same dilution curve.

The new methods lead to a 50% increase in diagnostic yield and further clinical benefits. Regarding common trisomies, they allow for a stratification of FPRs and FNRs based on z- scores/fetal fraction.

Regarding SCAs the new methods allow evidence-based aneuploidy detection and reporting in fetus and pregnant women with lower positive predictive values (PPVs) because of CPM.

With respect to RATs the new methods allow for the detection of fetal trisomy mosaicism, and CPM-associated risk for UPD, and adverse pregnancy outcome (IUGR, preeclampsia) with very low PPVs because of CPM

With regard to CNVs the new methods allow for size- and fetal fraction-dependent detection of fetal and maternal genomic disorders.

The following definitions and embodiments apply to the method of the invention (unless specified otherwise):

The method according to the present invention can be carried out on any biological sample containing nucleic acids. The biological sample is for example derived from a bodily fluid, a tissue, or an organ and the biological sample is essentially cell-free. Where the biological sample is derived from a bodily fluid, the bodily fluid can be blood, blood plasma, blood serum, urine, breast milk, saliva, amniotic fluid, cerebrospinal fluid (CSF), mucus, peritoneal fluid, pleural fluid, synovial fluid. Blood, blood plasma or blood serum are preferred, blood plasma being particularly preferred. Where the biological sample is derived from a tissue or an organ, it is for example derived from a diseased tissue or organ, such as a tissue or organ affected by a tumor. The biological sample can be derived from an invasive procedure, such as a biopsy, chorionic villus sampling, amniocentesis, or from a non-invasive procedure such as blood sampling, urine sampling, CSF sampling, breast milk sampling, mucus sampling, or peritoneal fluid sampling.

The biological sample can originate from any pregnant mammalian female, preferably a pregnant woman, to detect aneuploidy of the fetus carried by the pregnant woman. The biological samples can be collected at any time during the pregnancy, but are preferably collected from 7 weeks of pregnancy, for example between 7 weeks and 20 weeks of pregnancy, preferably from 7 to 1 4 weeks of pregnancy, still preferably from 7 to 1 2 or 7 to 1 0 weeks of pregnancy. The biological sample contains nucleic acids, such as DNA and/or RNA originating from the pregnant female and/or the fetus. In one embodiment, the biological sample contains eel I -free nucleic acids, namely eel I -free DNA (cfDNA) and/or RNA (cfRNA), more specifically fetal cell-free DNA and/or RNA and maternal cell-free DNA and/or RNA.

The term "aneuploidy" as used herein refers to a deviation between the structure of the subject chromosome and a normal homologous chromosome. Thus, aneuploidy refers to an imbalance of genetic material caused by a loss or gain of a whole chromosome, or part of a chromosome, whereas the term "euploidy" refers to a normal complement of chromosomes (which in humans is a euploid genome with 46, XX or 46, XY, also referred to as "diploid" or "disomy").

As used herein, the term "aneuploid" with reference to a sample refers to a sample obtained from a euploid mother carrying an aneuploid fetus and can be used with reference to a specific chromosome or chromosomal region of interest, or with reference to the whole genome.

In some embodiments, the aneuploidy is an autosomal aneuploidy in the form of monosomies or trisomies. The term "monosomy" as used herein refers to lack of one chromosome of the normal complement and partial monosomy, which can occur in unbalanced translocations or deletions, in which only a segment of the chromosome is present in a single copy. Examples of monosomy or partial monosomy include Wolf-Hirschhom syndrome, cri du chat syndrome, 5q deletion syndrome, Williams syndrome, Jacobsen syndrome, Angelman syndrome, Prader-Willi syndrome, Miller-Dieker syndrome, Smith-Magenis syndrome, 1 8q deletion syndrome, DiGeorge syndrome. The term "trisomy" as used herein refers to gain of one extra chromosome and partial trisomy, which is gain and/or duplication of a portion of a chromosome. Trisomies include both common trisomies and rare autosomal trisomies (RAT). Examples of trisomy include trisomy 1 , trisomy 2, trisomy 3, trisomy 4, trisomy 5, trisomy 6, trisomy 7, trisomy 8 (War- kany syndrome), trisomy 9, trisomy 10, trisomy 1 1 , trisomy 1 2, trisomy 13 (Patau syndrome), trisomy 14, trisomy 1 5, trisomy 1 6, trisomy 1 7, trisomy 18 (Edwards syndrome), trisomy 1 9, trisomy 20, trisomy 21 (Down syndrome), trisomy 22. Other examples of disorders involving a loss (deletion) of one or several chromosomal regions include 1 p36 deletion syndrome, TAR deletion, 1 q21 .1 deletion, 2q 1 1 .2 deletion, 2q 1 1.2q 1 3 deletion, 2q 1 3 deletion, 2q37 deletion, 3q29 deletion, Wolf-Hirschhorn deletion, Sotos syndrome deletion, 6q 1 6 deletion, Williams syndrome deletion , WBS-distal deletion, 8p23.1 deletion, 9q34 deletion, 10q23 deletion, Potocki-Shaffer syndrome, SHANK2 FGFs deletion, 1 2q 14 deletion syndrome, 1 3q 1 2 deletion, 1 5q 1 1 .2 deletion, Prader- Willi/Angelman syndrome, 1 5q 1 3.3 deletion, 1 5q24 BP0-BP1 deletion, 1 5q24 BP0-BP1 deletion, 1 5q24 BP2-BP3 deletion, 1 5q25.2 deletion, Rubinstein-Taybi syndrome, 1 6p1 3.1 1 deletion, 1 6p1 1 ,2p1 2.1 deletion, 16p 1 2.1 deletion, 16p1 1 .2 distal deletion, 16p1 1 .2 deletion, 1 7p1 3.3 deletion, 1 7p1 3.3 deletion, HNPP, Smith-Magenis syndrome deletion, NF1 deletion syndrome, RCAD (renal cysts and diabetes), 1 7q21 .31 deletion, DiGeorge VCFS deletion, 22q1 1 .2 distal deletion, Phelan-McDermid syndrome. in other embodiments, the aneuploidy is a sex chormosome aneuploidy (SCA). Sex chromosome aneuploidies may include X0 (Turner Syndrome), XYY (XYY syndrome; Jacobs Syndrome), XXX (Triple X Syndrome) and XXY ( Klinefelter^ Syndrome). In certain embodiments, only a portion of cells in an individual are affected by a sex chromosome aneuploidy which may be referred to as a mosaicism (e.g., Turner mosaicism). Other cases may include those where the gene on the Y chromosome triggering embryonic development as a male may either be damaged (leading to an XY female), or copied to the X (leading to an XX male).

Other examples of disorders involving a loss (deletion) of one or several chromosomal regions include 1 p36 deletion syndrome, TAR deletion, 1 q21 .1 deletion, 2q 1 .2 deletion, 2q 1 1 2q 1 3 deletion, 2q 1 3 deletion, 2q37 deletion, 3q29 deletion, Wolf-Hirschhorn deletion, Sotos syndrome deletion, 6q 1 6 deletion, Williams syndrome deletion , WBS- distal deletion, 8p23.1 deletion, 9q34 deletion, 10q23 deletion, Potocki-Shaffer syndrome, SHANK2 FGFs deletion, 1 2q 14 deletion syndrome, 1 3q 1 2 deletion, 1 5q 1 .2 deletion, Prader-Willi/Angelman syndrome, 1 5q 1 3.3 deletion, 1 5q24 BP0-BP1 deletion, 1 5q24 BP0-BP 1 deletion, 1 5q24 BP2-BP3 deletion, 1 5q25.2 deletion, Rubinstein-Taybi syndrome, 1 6p 1 3.1 1 deletion, 1 6p 1 1 ,2p 1 2.1 deletion, 1 6p1 2.1 deletion, 1 6p 1 1 .2 distal deletion, 1 6p 1 1 ,2 deletion, 1 7p1 3.3 deletion, 1 7p1 3.3 deletion, HNPP, Smith-Magenis syndrome deletion, NF 1 deletion syndrome, RCAD (renal cysts and diabetes), 1 7q21 .31 deletion, DiGeorge VCFS deletion, 22q 1 1 .2 distal deletion, Phelan- cDermid syndrome.

Examples of disorders involving the duplication of a chromosomal region include the split hand/split foot syndrome 3, the Beckwith-Wiedemann Syndrome, the Pelizaeus- Merzbacher disease, the Charcot- arie-Tooth disease type 1 A, 1 5q duplication-related autism.

In yet other embodiments, the aneuploidy is a copy number variation (CNV). The copy number variation can be a germline copy number variation or a somatic copy number variation. The copy number variation can be a chromosomal aneuploidy, or a subchromosomai copy number variation. A copy number variation can be a deletion (e.g. micro-deletion), duplication (e.g., a micro-duplication) or insertion (e.g., a micro-insertion). In certain embodiments, the prefix ^"micro^" may represent a segment of nucleic acid less than 5 Mb in length. A copy number variation can include one or more deletions (e.g. micro-deletion), duplications and/or insertions (e.g ., a micro-duplication, micro-insertion) of a segment of a chromosome. In some embodiments, a duplication may or may not comprise an insertion. A copy number variation may be a maternal and /or fetal copy number variation. A copy number variation may be a (maternal and/or fetal) heterozygous copy number variation where the variation (e.g., a duplication or deletion) is present on one allele of a genome. A copy number variation can be a (maternal and/or fetal) homozygous copy number variation where the variation is present on both alleles of a genome.

The term ^"mosaicism" as used herein refers to aneuploidy in some ceils, but not all cells, of an organism. Actual genetic mosaicism can be either somatic or germinal. Somatic mosaicism, mosaicism not affecting the germ cells, Germinal mosaicism, in contrast, includes two or more genetically distinct cell lineages including in the germ cells. Both somatic and germinal mosaicism can be of single genes or of whole chromosomes. Mosaicism can involve both the fetus (true fetal mosaicism or TFM ) and the placental tissues or the placental tissues alone (confined placental mosicism or CPM ) and can be classified according to the distribution of the abnormal cell line: in CPM type I and TFM type IV, only the cytotrophoblast is affected; in CPM type I! and TFM type V.only the stromal villous core (mesenchyme) is involved, and in CPM type III and TFM type VI, both placental tissues are affected by the abnormal cell line. It has been shown that cases where the cytotrophoblast is cytogenetically discrepant from the fetus are sources of false positive and false negative results: CPM type I and III with an abnormal cytotrophoblast and normal amniocytes can be potential sources of false positive, while TFM type V with a normal cytotrophoblast and abnormal amniocytes can be a potential source of false negative results.

As used herein, the determination of the presence or absence of an aneuploidy may be in a quantitative way and/or in a qualitative way with the outcome of the determination being ^"positive^" or "negative" (where the method does not allow the drawing of a conclusion on the presence or absence of an aneuploidy, the determination may result in a qualitative probabilistic appreciation, such as "probable", "highly probable", "low probability", etc... or else "indeterminate").

When predicting or diagnosing abnormality or normality, there are four possible types of outcomes for any given prediction: true positive, true negative, false positive, or false negative. The term "true positive" as used herein refers to a subject correctly diagnosed as having a chromosome abnormality. The term "false positive" as used herein refers to a subject wrongly identified as having a chromosome abnormality. The term "true negative" as used herein refers to a subject correctly identified as not having a chromosome abnormality. The term "false negative" as used herein refers to a subject wrongly identified as not having a chromosome abnormality.

The term "positive predictive value" (or PPV) as used herein, refers to the probability that an individual diagnosed as having a condition actually has the condition. Positive predictive value can be calculated as the number of true positives divided by the sum of the true positives and false positives.

The term "fetal fraction" as used herein is the ratio of the amount of cell-free fetal DNA to the total amount of cell-free DNA (cf DNA) in a biological sample as defined herein, in particular in a blood sample, from a pregnant woman. The total amount of cell-free DNA generally corresponds to cell-free (fetal and maternal) DNA. As used herein, the term "fetal fraction" represents the ratio of the amount of cell-free fetal nucleic acids (fetal cfDNA) to the total amount of cell- free nucleic acids (total cfDNA) in a biological sample, in particular in a blood sample, from a pregnant woman. The fetal fraction of a biological sample containing fetal and maternal cfDNA is determined by detecting quantitatively the presence of the one or more fragments of fetal cfDNA using known methods. For example, if the fetus is a male fetus, the target sequences can be selected from the Y chromosome, such as the SRY gene. Alternatively, the target sequences can be selected from a set of known or presumable sequence variants, such as different alleles or SNPs. For example, the fetal fraction fetal fraction can be inferred from the number of sequence tags mapped to an aneuploid chromosome 1 3, 1 8, 21 or from chromosome X in the case of male fetuses, and the number of sequence tags mapped to all chromosomes (see Fan and Quake, P!oS One, 2010 May 3;5(5):e10439. doi: 10.1 371 /journal. pone.0010439). More specifically, a bioinformatics-based predictor of the fetal fraction based on paired-end sequence tags was used similarly to a detailed published description (Prenat Diagn 201 5; 35: 810-81 5). A shown in Figure S3 it performs similarly based on correlations with Y-sequence tags (Pearson R=0.952, see their corresponding Fig. 2A with R=0.932). Based on this paired-end sequence tag-based pipeline a further developed predictor based on single-end reads was used. Its technical validation using Y-sequence tags and its functional implementation is illustrated in Figures S3 to S9.

In some embodiments , the fetal fraction, i.e. the proportion of cell-free circulating fetal DNA among the total amount of circulating cell-free DNA, is enriched by size selection. This enrichment takes advantage of the fact that the size distribution of cell-free fetal DNA differs from the size distribution of maternal cell free DNA. Generally, the average length of maternal DNA is greater then the average legth in bp of fetal DNA. Thus, selection of smaller MW species, for example less than 200 bp, less than 180 bp, less than 1 50 bp, less than 1 25 bp for example between 1 1 5 and 1 25 bp allows the exclusion of the higher MW species of maternal DNA. Size selection can be carried out by any technique known in the art, e.g. by electrophoresis, in particular by capillary electrophoresis or gel electrophoresis, or by affinity capture, e.g. by purification on beads. The use of magnetic beads is particularly preferred, for example AMPure XP® beads. According to this technology, DNA fragments bind to the magnetic beads, and are then separated from contaminants by application of a magnetic field. The bound DNA is washed with ethanol and is then eluted from the magnetic particles. Enrichment of featl cfDNA can be performed at any time before the sequencing. In one embodiment, the enrichment is carried out concurrently with the extraction of the nucleic acids, for instance when the nucleic acids are extracted by purification on beads. Alternatively, or in addition, the enrichment is performed after the nucleic acid extraction, and before the preparation of the sequencing library, or during the preparation of the sequencing library. or after the preparation of the sequencing library and before the sequencing. Preferably, this step may be performed after extraction of the cell-free D A from the test sample and before massive parallel sequencing, more particularly before library preparation.

The term "threshold value" as used herein in reference to fetal fraction or z-score refers to a number that is calculated using at least one qualifying reference data set (for example one, two, three, four, five or six reference data sets) and serves as a reference value for the fetal fraction and z-score obtained in a test sample. If a threshold is exceeded by results obtained from a test sample, a subject or may not be diagnosed with an aneuploidy. In some embodiments, a threshold may be internal to the biological sample, i.e. it is obtained from the analysis of the same test biological sample as that analyzed in the framework of the method for determining the presence of a biological condition. The threshold is for example obtained from the analysis of further nucleic acid species in the test biological sample (designated as ^"reference nucleic acid species"), namely chromosomes or chromosomal regions which different from the test chromosome(s) or chromosomal region(s), i.e. not affected by the aneuploidy. In some embodiments, a threshold may be external to the biological sample, i.e. they are obtained from one or more reference samples different from the test sample.

The reference samples selected to obtain a threshold are selected on the basis of various criteria, including the sample quality. Where the method involves massively parallel sequencing, the reference samples can be selected according to the number of total sequences obtained, and/or the number of UEMs. Exemplary criteria for selecting reference samples for the purpose of deriving a threshold, in the field of NIPT, are described e.g. in WO2014/068075, the content of which is hereby incorporated by reference.

A threshold value is calculated based on total separation of all normal (or euploid) reference values from pathological (or aneuploid) reference values (to give maximum ( 100%) specificity and sensitivity).

The classification of test results into euploid versus aneuploid status was based on the single genome-wide thresholds of < 1 0^"3 and < 1 0 respectively, which corresponds to a z-score of +3.3 (+3.2905 ) and +3.9, respectively. In some embodiments a z-score threshold of >6x 3.3 and <5 is applied, in some embodiments a z-score threshold of > 3 and < 5 is applied. Thus, any signal intensity measured close to threshold (i.e. z-scores >6x 3.3 and <5 or >3 and < 5) is potentially a false positive result (even without measuring and integrating the fetal fraction). This likelihood is highest for trisomy 1 3 followed by trisomy 1 8 and finally trisomy 21 (the natural fetal fraction is proportionally lower for the latter two trisomies).

The comparison of the total amount of cfDNA with a z-score/signal intensity threshold as defined above may be carried out by calculating the z-score of the biological test sample, according to the formula: z-score = ( P_test - mean ( Pref))/(SD(P_ref)), wherein P_test is the test parameter indicative of the number of unique exact sequences mapped to the chromosome or chromosomal region of interest, calculated from the test sample. Mean (P-*) and SD(P_re-) are respectively the mean and the standard deviation of the set of reference parameters indicative of the number of unique exact sequences mapped to the chromosome or chromosomal region of interest, calculated from the set of reference samples.

Various factors may influence the calculation of the z-score. For example, maternal CNV may play a substantial role by increasing the number of relative unique mapped chromosome reads, which results in a higher z-score which in turn may lead to a false positive result. Maternal deletion may have the opposite effect and may lead to a false negative result.

According to the methods of the invention the classification of test results into euploid versus aneuploid status was based on the single genome-wide thresholds of < 1 0 ³ and < 10 \ respectively, which corresponds to a z-score of +3.3 (+3.2905) and +3.9, respectively, and the threshold values of fetal fraction results of between 3% - 5.8%.

To call an aneuploidy, the absolute value of the z-score of a biological sample aneuploid for the chromosome or chromosomal region of interest is above the defined threshold value (and analogously the absolute value of the z-score of a biological sample euploid for the chromosome or chromosomal region of interest is below the defined threshold value), while the fetal fraction results are within the specified range..

Thus, the methods of the invention allows to improve the detection of fetal aneuploidy by improving the discrimination of false negative and false positive results.

The term "cell-free DNA" (or cfDNA; also termed "circulating cell-free DNA" or "circulating DNA^") as used herein refers to a DNA molecule or a set of DNA molecules freely circulating outside of cells in a biological sample, such as blood, blood plasma, blood serum, urine, saliva, and /or extracellular fluid. Without being limited by theory, cell-free DNA is assumed to be a product of cell apoptosis, cell necrosis and/or cell breakdown. Cell free DNA can be in the form of nucleic acid fragments having a length which can vary across a large spectrum. Cell-free DNA extraction may be perfomed according to known protocols, in some embodiments via a protocol of phenol-chloroform extraction. This extraction protocol typically comprises: (i) mixing said biological sample with a composition comprising chloroform and phenol; (ii) extracting the aqueous phase from said mixture; (iii) precipitating cell-free DNA from said aqueous phase; and (iv) optionally collecting cell-free DNA.

According to the present invention, the term "pheno!/chloroform" refers to a mixture of phenol and chloroform, i.e. to a composition comprising phenol and chloroform. Said composition is preferably an aqueous solution and preferably also comprises isoamyl alcohol. The pH of the composition is preferably from 7 to 9, still preferably from 7.8 to 8.2. A preferred composition is a 25:24: 1 mixture of phenol:chloroform:isoamyl alcohol at a pH from 7.8 to 8.2. The composition may comprise one or more additives, such as one or more antioxidants and/or stabilizers. In a specific embodiment, the extraction method comprises a step of pre-treating the biological sample with one or more proteases, such as proteinase K. The extraction of the aqueous phase may comprise centrifuging the biological sample mixed with chloroform and phenol, and collecting the aqueous phase. The centrifugation provides a separation of the mixed biological sample into a lower organic phase, comprising mainly phenol, proteins or protein debris, and an upper aqueous phase comprising nucleic acids. In an embodiment, the precipitation of cell-free DNA from the aqueous phase comprises the steps of: mixing at least one precipitation agent with the aqueous phase; centrifuging said mixed aqueous phase; and collecting the centrifugation pellet. The precipitation agent is preferably selected from glycogen, a lower alcohol such as isopropanol or ethanol, or mixtures thereof. The centrifugation pellet containing DNA can then be washed one or more time, for example with ethanol and/or ether. Finally, DNA can be resuspended in a suspension buffer, for example a Tris buffer.

In other embodiments, extraction is carried out using an automated protocol. In some embodiments, the automated extraction is followed a standard protocol and is done on 1 mL plasma on a QIAGEN QIAsymphony instrument using the kit DSP Virus/Pathogen.

The determination of the total amount of cfDNA (i.e. fetal and maternal DNA) is done by counting the sequences aligning uniquely, without mismatch, to a reference genome, or a portion thereof. This determination may include various molecular biology procedures, for example PCR, such as digital PCR, quantitative PCR (qPCR), reverse transcription PCR (RT- PCR) reverse transcription quantitative PGR ( RT-qPCR), nested PCR, multiplex PGR, methyl- ation- specific PGR, allele-specific PCR, emulsion PCR, hybridization, such as southern blotting, hybridization on a nucleic acid-array, e.g. a DNA-array, sequencing, chromatography, such as affinity chromatography, gel permeation chromatography, liquid chromatography, including high performance liquid chromatography (HPLC), or any combination of these techniques.

The determination includes a first step of extracting DNA from a biological sample. In some embodiments, the extraction of DNA from the biological sample may be carried out on e.g. a solid support such as beads, e.g. by magnetic beads, or in a liquid medium, such as an alcohol-containing and/or chloroform-containing solution, e.g. a phenol and chloroform- containing solution. The extraction may involve the use of an extraction column.

The determination includes a second step of sequencing using massively parallel sequencing (which includes preparing a suequencing library on which sequencing is performed) In some embodiments, the step of sequencing may be carried out on the whole genome, whereas in other embodiments, it concerns only a portion of the genome. Although various methods of sequencing can be contemplated, high throughput methods are preferred, such as massively parallel sequencing (MPS), also referred to as next- generation sequencing (NGS) or second-generation sequencing (Buermans et al., 2014).

The massively parallel sequencing technologies generally comprise two main steps: the preparation of a set of templates (sequencing library), on which the actual sequencing will be carried out, and the actual step of sequencing, which generally comprises the detection of the addition or release of nucleotides on the template. The preparation of the sequencing library can take place immediately after the extraction of the nucleic acids or it can take place after a prior processing (e.g. enrichment based on size selection) of the extracted nucleic acid. The preparation of the sequencing library can include one or more amplification steps, a ligation with one or more sequencing adaptors, and/or barcoding the nucleic acid molecules. A typical workflow of the sequencing library preparation includes a step of ligation of one or more adaptor sequences, optionally linked to one or more barcode sequences, to the nucleic acid molecules inside the sample, followed by an amplification of the adaptor/bar- code-ligated nucleic acid molecules. Sequencing adaptors are short nucleotide sequences which are commonly used in modern sequencing technologies. The adaptors are used for anchoring the nucleic acid molecules to be sequenced to a solid surface, for example in a flow cell. These adaptors are thus designed so as to hybridize to target oligonucleotides tethered to the solid surface. The ligation of adaptors is preferably performed by repairing the ends of the nucleic acid molecules, i.e. suppressing or filling out the overhangs of the extracted nucleic acid molecules, for example through the action of one or more exonucleases and/or polymerases, thus yielding blunt-ended nucleic acid molecules. An overhang of one or more ^'A^' bases may then be optionally added at the 3^' end of the blunt-ended nucleic acid molecules. The adaptors containing an overhang of one or more T bases at their 3" end, are then added and are ligated to the overhang of one or more "A' bases at the 3^'end of the nucleic acid molecules. Adaptors can also be blunt ligated. The nucleic acid molecules within the sample can also be barcoded. Barcoding refers to the ligation of a sample-specific tag to the nucleic acid molecules of a sample. Barcoding allows the sequencing of several samples in a single sequencing run, which saves time and resources.

In a specific embodiment, samples are selected wherein at least 80 wt%, still preferably at least 90 wt%, preferably at least 95 wt%, still preferably at least 97wt% of the DNA molecules inside the sample have a size of less than 200 bp, less than 180 bp, less than 1 50 bp, less than 1 25 bp for example between 1 1 5 and 125 bp.

The nucleic acid molecules within the sample can also be subjected to one or more amplification cycles, for example by PCR. The amplification can be performed by bridge amplification, a technique which uses templates attached on a solid support. It can also be performed by emulsion PCR. The amplification is preferably carried out after the ligation of an adaptor sequence to the nucleic acid molecules. The PCR amplification preferably uses primers against the adaptor sequence, thus enriching the library into adaptor-ligated fragments. In some cases, amplification is not necessary, in particular in sequencing platforms with a detection limit sufficient to detect a single nucleic acid fragment.

The sequencing is then carried out on the sequencing library, generally by detecting the addition/release of nucleotides on the sequencing templates. Several technologies exist for the detection, such as visible/UV light detection, fluorescence detection, or pH detection, i.e. detection of the addition or release of a proton (H⁺) in the reaction medium.

Anyone of the NGS platforms that are available commercially, are currently under development, or will be developed in the future, can be used in the present invention The term ^"next-generation sequencing^" (NGS) as used herein refers to a high-throughput sequencing method in which hundreds of thousands of sequencing processes are made parallel. Next-generation sequencing methods are useful for obtaining several millions of sequences in a single run. These methods include single-molecule real-time sequencing, ion semiconductor sequencing, pyrosequencing, sequencing by synthesis, sequencing by ligation.

Prior to the step of massively parallel sequencing, an optional step of pre-sequencing (i.e. a small-scale sequencing) may be performed according to known methods.

The step of massively parallel sequencing or NGS may be performed using any of the various massively parallel sequencing technologies and platforms. These platforms include e.g . the GS FLX Titanium XL+ (Roche), GS Junior System ( Roche), 454 ( Roche), Ion Torrent (Life Technologies), Proton (Life technologies), Abi/solid (Life technologies), HiSeq2000/2500 (lllumina), MiSeq (lliumina), NextSeqSOO (lllumina), HiSeq X Ten (lllumina), RSII ( Pacific biosciences) or Heliscope (Helicos). Preferred platforms include e.g. "sequencing by synthesis" (or SBS) methods where the iterative addition of specific nucleotides in a template dependent, polymerase mediated extension reaction are identified and used to provide the underlying sequence of the template nucleic acid. These SBS processes are generally divided into ( 1 ) short read sequencing technologies, e.g., employed in lliumina HiSeq, MiSeq, and NextSeq sequencing systems, as well as the Ion Torrent Proton and PGM systems (available from Thermo Fisher), and ( 2 ) long read sequencing technologies such as single molecule, real time, or SM RT( R) sequencing systems available from Pacific Biosciences.

The term ^"read^" refers to a sequence read from a portion of a nucleic acid sample. It may be stored in a memory device and processed as appropriate to determine whether it matches a reference sequence or meets other criteria. A read may be obtained directly from a sequencing apparatus or indirectly from stored sequence information concerning the sample. In some cases, a read is a.DNA sequence of sufficient length as defined herein that can be used to identify a larger sequence or region, e.g. that can be aligned and specifically assigned to a chromosome or genomic region or gene.

The sequencing workflow in a "sequencing-by-synthesis^" system can be summarized in 3 phases: First, the preparation of the DNA library: this step has already been described and, as mentioned above, it can be carried out at an early phase of the whole process of selecting euploid appropriate reference samples, or of the diagnosis process. It is for example performed immediately after DNA extraction, or immediately after size selection of the extracted cell-free DNA. During this phase, DNA molecules are ligated with adaptors at both ends. In addition, they contain primer sites that are used to amplify the library by PC and to sequence it.

Second, the cluster generation: during this phase, DNA molecules are hybridized to oligonucleotide probes tethered on a solid surface inside a flow cell. Each DNA molecule is amplified by solid-phase bridge-amplification, forming a cluster of molecules with identical sequences.

Third, the ^"sequencing-by-synthesis^" phase. A mixture of the four nucleotides, each containing a fluorescently-labeled terminator, is introduced into the flow-cell. The fluores- cently-labeled terminator is imaged as each dNTP is incorporated into the growing DNA strand, and then cleaved to allow incorporation of the next base. Since all four reversible terminator-bound dNTPs are present during each sequencing cycle, natural competition minimizes incorporation bias. Base calls are made directly from intensity signal measurements during each cycle.

Alternatively, the massively parallel sequencing platform may for instance consist in a semiconductor-based next generation sequence technology.

In a specific embodiment, the massively parallel sequencing step consists in sequencing at least 1 0 millions, preferably at least 20 millions, still preferably at least 30 million sequences per sample.

The sequences obtained by sequencing can be mapped to a reference sequence, or in other terms aligned with a reference sequence. ^"Indexing" is also used as a synonym of ^"mapping", in the present disclosure. The alignment of the sequences can be carried out using any standard alignment software, for example as described in Chiu et al., PNAS, vol 1 05, 20458, 2008 or Fan et al. , PNAS, vol. 1 05, 1 6266, 2008. T he reference sequence is preferably a reference sequence of the human genome, such as the sequences established by the N BCI (http://www.ncbi.nlm.nih.gov/assembly/2758) or the UCSC (http://hgdown- load.cse.ucsc.edU/downloads.html#human). The reference sequence is preferably the Genome Reference Consortium GRCh37 released in February 2009, also referred to as hg 1 9; or the Genome Reference Consortium GRCh38 released in December 201 3, also referred to as hg38. Mapping can be done over the whole genome, or over only a portion of the genome.

Generally speaking, a partial sequence of a genome, such as the human genome used in score calculation is obtained by masking predefined regions of the genome. The regions to be masked can be chosen on the basis of a number of different parameters, including: a lower quality of sequencing of a region (these regions are also known as ^"non-well annotated regions"); the occurrence of a high number of repeats within a region; the duplication of a region within the genome; a region with a complex architecture. The masked regions are thus preferably selected among the non-well-annotated regions of the genome, the high copy repeat regions of the genome, the duplicated regions of the genome, or the regions with a complex architecture.

A region with a lower quality of sequencing or a "non-well annotated" region is for instance a region with scaffold N50 of less than 46,395,641 and/or a contig N50 of less than 38,508,932, and/or with total assembly gap length of more than 239,845, 127/3, 1 37, 144,693, and/or with a genome coverage of at least 90%, preferably at least 95% (Yandell et al., 201 ). Examples of non-well annotated regions are sub- telomeric regions and pericentromeric regions.

The sequences are then counted from the mapping step. This is generally done by counting certain categories of sequences. Among the sequences which can be counted from the mapping are the Unique Exact Sequences (UES) or Unique Exact Matches (UEM), which are sequences which are uniquely mapped to the reference sequence with no mismatch. By ^'unique' or 'uniquely', it is meant that the sequence has a single match on the reference sequence, i.e. it does not match to different genome regions.

In a specific embodiment of the methods of the present invention at least 6 million, preferably at least 8 million, still preferably at least 10 million, or at least 1 2 million or at least 14 million or at least 1 5 millions unique exact sequences per sample are obtained in the mapping step. Alternatively or in addition, a mean number of at least 1 2 million, preferably at least 1 5 million, still preferably at least 20 million unique exact sequences per sample is obtained in the mapping step. In a specific embodiment, the method comprises selecting samples with a total number of at least 1 0 million, preferably at least 20 million, still preferably at least 30 million sequences per sample. Alternatively or in addition, the method comprises selecting samples with at least 6 million, preferably at least 8 million, still preferably at least 1 0 million, or at least 1 2 million or at least 1 4 million or at least 1 5 million unique exact sequences, most preferably between 1 0 million to 1 2.5 million unique exact sequences.

Other categories of sequences can be counted/highlighted from the mapping steps, such as unique sequences with at most one, or two or three mismatches, or aligned exact sequences (unique or not unique), or aligned sequences with at most one, or two or three mismatch, or aligned sequences with at most 1 %, 2%, 5%, 10%, 1 5%, 20% mismatched nucleotides.

The total amount of cell free maternal and fetal DNA can be represented by a ^'score^' calculated for a given chromosome or chromosomal region, i.e. a parameter indicative of the count of sequences mapped to a chromosome or chromosomal region.

In one embodiment, the score is calculated from the UEMs. In another embodiment, the score is calculated from the unique sequences with at most one, or two, or three mismatches, in another embodiment, the counted sequences are the aligned exact sequences (unique and not unique), and /or the aligned sequences with at most one, or two or three mismatch, or the aligned sequences with at most 1 %, 2 %, 5%, 1 0%, 1 5%, 20% mismatched nucleotides. The score can also be calculated from any combination of these sequence categories. The score can be calculated over a whole genome sequence, or over a partial sequence of a genome or, in other terms a sequence from which some regions have been masked. Calculating the score only over a carefully selected portion of a genome is a way to increase the degree of statistical confidence of the diagnosis method. Generally speaking, the partial sequence of the genome used in score calculation is obtained by masking predefined regions of the human genome, in the same way at that explained previously to obtain a partial sequence of the genome. A number of parameters can be considered for defining the regions to be masked, including a lower quality of sequencing of a region (also defined, in other terms as a non-well annotated region), the occurrence of a high number of repeat within a region, the duplication of a region within the human genome, a region with a complex architecture. The masked regions are thus preferably selected among the non-well-annotated regions of the human genome, the high copy repeat regions of the human genome, the duplicated regions of the human genome or the regions with a complex architecture.

The score for each chromosome can be calculated by dividing each chromosome into bins of a predefined length, for example 50 kb bins. The division can be carried out on a whole human genome sequence or on a partial human genome sequence. The number of unique exact sequences (UES) mapped to a given bin is then counted, thus yielding a LIES count for each bin.

In a specific embodiment, the count of UES for each bin is bias-corrected, i.e. it is corrected to take into account the bias related to the sequencing process. A known bias is caused by the variation in GC distribution across the genome.

The median UES count value for all bins over a chromosome or chromosomal region of interest is then calculated. This value is representative of the count of UESs across the chromosome or chromosomal region, and is referred as the sequence tag density of a chromosome or chromosomal region. This median value can be calculated by using non- weighted UES counts, or by weighting each UES count with a bias-correction factor, as indicated above. In another embodiment, other values than the median value are selected for representing the UES count across a chromosome: for instance the sum of the UES counts for all bins within a chromosome.

Finally, the sequence tag density of the chromosome or chromosomal region of interest can be normalized to the median sequence tag density for all chromosomes. Alternatively, it can be normalized to the median sequence tag density for all autosomes. Still alternatively, it can be normalized to the median sequence tag density for a predefined set of chromosomes. As used herein "set of chromosomes^'" refers to any combination of chromosomes selected from chromosome 1 to chromosome 22 and chromosome X and Y. Still alternatively, it can be normalized to the median sequence tag density for a predefined set of chromosomal regions. Still alternatively, it can be normalized to the sum of sequence tag densities for all chromosomes, or for all autosomes, or for a predefined set of chromosomes, or for a predefined set of chromosomal regions.

The normaiized sequence tag density of a chromosome or chromosomal region can be used as a parameter indicative of the number of unique exact sequences mapped to a chromosome or chromosomal region of interest for a given sample. This parameter can however be represented by other values: the sequence tag density of a chromosome or chromosomal region of interest, the number of UESs mapped to said chromosome or chromosomal region of interest; the number of UESs mapped to said chromosome or chromosomal region of interest normalized by th total number of UES for the sample; the number of UESs mapped to said chromosome or chromosomal region of interest normalized by the total number of UES mapped to a predefined set of chromosomes or chromosomal regions.

Preferably, the chromosome of interest is chromosome 21 and/or the fetal aneuploidy is trisomy 21 . Alternatively, the chromosome of interest is chromosome 1 8 and/or the fetal aneuploidy is trisomy 1 8. Alternatively, the chromosome of interest is chromosome 1 3 and /or the fetal aneuploidy is trisomy 1 3. Alternatively, the chromosome of interest is chromosome 22 and/or the fetal aneuploidy is trisomy 22. Alternatively, the chromosome of interest is chromosome 4 and/or the fetal aneuploidy is Wolf-Hirschhom syndrome.

Alternatively, the chromosomal region of interest is a portion of chromosome 4 comprising the deleted region in Wolf-Hirschhom syndrome. Alternatively, the chromosome of interest is chromosome 5 and/or the fetal aneuploidy is cri du chat syndrome. Alternatively, the chromosomal region of interest is a portion of chromosome 5 comprising the deleted and/or duplicated region in cri du chat syndrome and/or the fetal aneuploidy is cri du chat syndrome. Alternatively, the chromosome of interest is chromosome 1 9. Alternatively, the chromosome of interest is chromosome 1 . Any combination of the aforementioned chromosomes or chromosomal region can also be chosen as a specific embodiment.

More preferably, the chromosome of interest is chromosome 21 , chromosome 1 8, or chromosome 1 3, still preferably, the chromosome of interest is chromosome 21 or chromosome 1 8.

Examples

Maj^fiailLand Methods

Study Design: The consecutive cases consisted of two non-overlapping data sets. The first data set aimed at a complete follow-up of newborns based on the predicted birth of all pregnancies included, the second at evaluating the effects of integrating the routine measurement of the fetal fraction . The principal readout was the accurate test classification of singleton pregnancies into euploid and aneuploid status, and the outcome of the follow-up of pregnancies with invasive testing/amniocentesis for aneuploidies considered pathogenic, likely pathogenic or of unknown clinical significance (US). The clinical outcome was ascertained at birth for euploid results considered not pathogenic, likely not pathogenic, and for US. The follow-up was stratified: the proportion of cases followed-up with amniocentesis performed by internal or external cytogenetics laboratories was raised with increasing uncertainty about their pathogenic status. The pregnancy outcome was monitored i) by online registries for verified aneuploidies and for birth outcomes, both based on voluntary information provided by physicians, ii) by an inquiry on the birth outcome comprising two 250 random samples in the two main linguistic regions, and iii) by proximity with addressing physicians resulting from the small country size. The classification of test results is described in the Supplementary Materials and Methods Information.

Classification of test results: The classification of test results into euploid versus aneuploid status for numerical and sex chromosome anomalies was based on a single genome-wide threshold, a reference value < 1 0 ^:i, which corresponds to a z-score of +3.3 (+3.2905) (supplementary Figure SI ) based on perfect discrimination of positive and negative control values (see reference 22, Guex N, et al. Prenat Diagn 201 3; 33: 707-710). Results with reference values < 10^"3 were considered pathogenic, ii) values identical to 10^"3 were considered likely pathogenic, iii) values < 10^"2 and > 1 0^"3 were considered of US, iv) values > 10 ² were considered benign or likely benign. A historical trisomy 13 sample with very low signal intensity retrospectively corresponding to a fetal fraction of 3% - 5% (supplementary Figure S2) was used as positive internal control. Results of individual batches were only reported if this trisomy was correctly called. After routine measurements of the fetal fraction the lower reliable threshold for reporting results, i.e. fetal fractions between 3% - 5.8%, was defined experimentally (supplementary Figures S3-S7).

Copy number variants were detected based on variation of the signal intensity in experimental samples compared to independent reference values; for defined genomic regions positive controls were available (see reference 22, Guex N, et al. Prenat Diagn 201 3; 33: 707-710). The signal intensity was used to predict whether candidate CNVs are of purely fetal, or maternal ± fetal origin (supplementary Figures S9-S1 0). if a maternal origin is suspected CMA testing is performed on a maternal blood sample. If confirmed, the evaluation of pathogenicity and penetrance determined the further follow-up (see reference 1 7). If likely fetal, the breakpoints are mapped (supplementary Figure S1 0), and three criteria used to decide whether amniocentesis is warranted, i) overlap with a known genomic disorder of significant pathogenicity and penetrance (see reference 27, Coppinger J, et al. Prenat Diagn 2009; 29: 1 1 56-1 66), ii) overlap with pathogenic OM I M genes of significant penetrance, iii) absence of annotation as polymorphic CNV in databases. These criteria combined with the determination whether the CNV is inherited or occurred de novo by parental analysis, and whether one or more than one CNV are observed allow CNV classification into benign, likely benign, US, likely pathogenic, pathogenic (see reference 28,29, Hehir-Kwa JY, et al. Clin Genet 201 3: 84: 41 5-421 ; Vulto-van Silfhout AT, et al. Hum Mutat 201 3; 34: 1 679- 1 687).

After routine measurements of the fetal fraction the analytical sensitivity/specificity for defined size classes and specific genomic CNV regions was defined experimentally. Briefly, in the example shown cfDNA of an affected fetus having a 3 Mb duplication of the Di George syndrome (DGS) region 22q 1 1 .2 of maternal origin was diluted with cfD A of an unaffected case to determine the limits of detection of CNVs of defined size (CM A validated case represents dosage of 100%). 1st cfDNA was spiked with a cfDNA sample displaying no warnings in the incriminated region. In order to define the CNV detection levels, dilutions were made to reflect the physiological range of fetal fractions, namely 5%, 10%, 1 5%, 20% and 25%. Based on these serial dilutions the analytical sensitivity for CNVs of defined size class and breakpoints were determined. This was replicated with a second, independent fetal duplication of maternal origin.

Samples and cfDNA aneuphidy screening: Standard EDTA blood tubes (Becton Dickinson, Sarstedt) were exclusively used because of regulatory requirements (CE marking). Exclusion criteria: transportation time >48h, total DNA concentrations >4ng/ul, and visible hemolysis (degree defined by photographic references). CfDNA extracted from one ml of plasma was analyzed by shotgun sequencing on lllumina sequencers (HighSeq 2000) (see reference 22, Guex N, et al. Prenat Diagn 2013; 33: 707-710) with a minimal genomic coverage of 0.01 56 fold (> 10x10⁶ unique exact 50nt sequence tags).

Data sets and patients: The test report obtained after expert data interpretation by board certified laboratory geneticists according to standard operating procedures including clinical information were entered into the clinical database (Glims), and used to retrieve consecutive data sets. Two consecutive data sets were retrieved from Glims, one for the period of beginning of March 201 3 to the end of August 2014, and one after integration of the routine measurement of the fetal fraction from the beginning of September 2014 to the end of May 2015. Overall 6'388 samples were addressed. The first data set included 4'545 pregnancies, 4'497 singleton and 48 twin pregnancies. Similar proportions of samples came from the Swiss-German (53.4%) and Swiss-French (46.5%) parts of Switzerland. Ethnic origins represented age-related sections of the Swiss population (http://www.bfs. admin. ch): 2 /3rd Swiss, ethnically german, french, and Italian speaking, and 1 /3rd foreigners (Europe, Balkans and Turkey 85.1 %, Africa 4.3%, America 4.04%, Asia 6.34%, Oceania 0.21 %). The gestational age was available for 6Ί 00 (95.5%) pregnancies. The average ( 1 3.1 9 weeks 2.36 standard deviation) and median values ( 1 3.0 weeks) were similar. Minimal and maximal values were 5 and 32 weeks of pregnancy

The results for annotated singleton pregnancies according to the thresholds described were used as the basis for the statistics, after one additional review by an independent expert geneticist. The second data set included Γ843 samples not overlapping with the first data set. The bioinformatics-based measurement of the fetal fraction was integrated into clinical routine after validation on over 1 '000 samples (supplementary Figures S3-S7). The historical sample with one of the lowest fetal fractions affected by trisomy 13 was used to define the lower reliable threshold of fetal fractions consistent with accurate calling of numerical anomalies (supplementary Figures S6-S7). Samples with fetal fractions <3% were excluded and a second blood draw asked. Samples considered pathogenic or likely pathogenic (concordant values < 10³) were reclassified as potential mosaic anomalies or suspected vanishing twin if the signal intensity was not proportional to the fetal fraction.

From the 6'388 samples that were received, 6Ί 92 (96.9% ) were included and received a primary call. For 1 95 (3.05%) a second blood draw was asked, for the following reasons, 53 because of DNA concentrations higher than specifications (>4ng/ul), 30 because of hemolysis, 30 because of a combination of these two factors, and 82 for other, unrelated reasons. Latter mostly comprised samples with a fetal fraction below specification (3 %) that made up less than one percent of all samples included ( 1 % of values being <3.2%). All but two received a secondary call requiring a redraw.

Twenty-eight percent of pregnancies ( 1 '81 4/6'388) had no known risk (pregnancies ( 1 of these two '605/6'388) belonged to defined risk groups, being mostly factors, and 82 for other, unrela'399/6'388) had an advanced maternal age (s, being mostly '871 or 44. 94% were addressed because of advanced maternal age as the sole indication; 14.4% (91 8/6'388) had increased risks (sed because of advanced maternal age as the sole indication; 14.4% (91 8/6attening ( 14/6388), 1 .28% (82/6'388) had ultrasound abnormalities, 2.39% had a personal or family history of chromosomal anomalies ( 1 53/6'388), and finally 5.7% (365/6388) had increased by not further specified risks.

The cases were representative of the major European populations. Their gestational age profile of 1 3 weeks at cfDNA screening was compatible with accurate DS detection, with a spontaneous miscarriage risk having declined, and with timely confirmation by amniocentesis (see reference 24, Mersy E, et al. Pub Health Genomics 201 5; doi: 10.1 1 59/000435780). The fact that a second blood draw was required for 3% of the samples was mostly explained by the fact that DNA-preserving blood tubes were not allowed until recently (because not CE-certified, a requirement of the national legislation). This led to a higher total DNA yield and a lower fetal fraction for samples with more than 24h transportation time, presumably resulting from an increased cell lysis (data not shown). The clinical reasons for cfDNA testing and the risk profile were similar to a larger US study (see reference 3, McCullough RM, et al. PLoS One 2014; 9: e1091 73. doi:10.1 371 /jour- nal-pone.01 091 73).

The incidence of individual anomalies that was found was comparable to those of reference studies addressing equivalent risk groups. The global incidence of the common trisomies in our data set was 1 .86% ( 1 19/6388), which is similar to the 2.1 7% reported (see reference 3, McCullough RM, et al. PLoS One 2014; 9: e1 091 73. doi: 10.1 371 /journal- pone.01091 73). The detection rate of SCAs in our series was 0.83%, and 1 .1 % in a reference study; most of the individual SCA subtypes were also equally distributed, namely MX accounted for 71 % of all SCAs in our versus 72% in the reference study, and TXS was detected in 1 8.9% in our versus 1 8.6% in the control study. However, the inventors found KS more frequently, in 9.4% versus 5.9% for the reference study, and 47.XYY less frequently, namely < 1 /6388 versus 2.9%. Only two of their twelve KS and of their six 47.XYY cases were verified, two KS and one 47,XXY were true positives, the other 47.XYY turned out to be a KS, indicating either suboptimal discrimination of KS and 47.XYY, and/or overestima- tion of the 47.XYY incidence by them, non-detection or sub-optimai detection by us. For the rare autosomal trisomies making up 0.78% in our cohort no direct comparison based on cfDNA screening currently exists. If cfDNA depends on a pathological cell line present in the trophoblast, then one can estimate based on chorionic villous sampling (CVS) data an incidence of 0.55% for detectable autosomal trisomies (see reference 25, Grati FR, et al. Eur J Hum Genet 2006; 14: 282-288), which is close to what was found. Further in support of the notion that both data sets can be interrelated, trisomy 7 is the most prevalent trisomy in both series (see reference 25, Grati FR, et al. Eur J Hum Genet 2006; 14: 282-288). The incidence of recurrent pathogenic duplication and deletion CNVs has been estimated to be 0.3% (see reference 26, Grati FR, et al. Prenat Diagn 201 5; 35: 801 - 809), which is again in the same range of our findings (0.56%).

Example 1 : (a) Detection of aneuip!oidy classes

Overall, 258/6^*388 (4.04% ) samples were considered pathogenic or likely pathogenic (Figure 1 ), comprising 1 1 9 common trisomies ( 1 .86% ), 53 SCAs (0.83%), 50 rare autosomal trisomies (0.78%), and 36 CNVs (0.56%). Conversely, 6Ί 30 samples (95.96%) were considered benign or likely benign.

The 1 1 9 common trisomies were invariably classified as pathogenic or likely pathogenic, and comprised 84 trisomy 21 ( 70.6% ), 1 9 trisomy 1 8 ( 1 6%), and 1 6 trisomy 1 3 samples ( 1 3.4% ). Fifteen additional samples were rated of US (0.23% ). Discriminating euploid versus aneuploid was simple for trisomy 21 , since 75/84 (89.3 %) z-scores were at least one and a half times higher than the threshold. All but three results (81 /84; 97.6%) were considered pathogenic - these three were rated likely pathogenic - and three were classified as being of US. Trisomy 1 8 scores were also generally high ( > 1 .5x threshold z-score) for 1 6/ 1 9 (84.2%) , with 1 8/ 1 9 rated pathogenic and one likely pathogenic, but a higher number, eight overall, were considered of US. Twelve out of sixteen trisomy 1 3 results (75%) obtained high scores ( > 1 .5x threshold z-score), 1 5/ 1 6 were scored pathogenic, one likely pathogenic, and four were considered of US.

(b) Follow-up validation

The follow-up with amniocentesis was 47% for the pathogenic/likely pathogenic group (57/ 1 21 ), and 80% for the US group ( 1 2/ 1 5 ). The work-up of the newborn babies who tested "benign/likely benign" did not yield indications for discrepant results. The US samples without follow-up belonged all to the first data set and were born healthy (composite fol- lowup of US samples 1 5/ 1 5 or 1 00% ). For trisomy 2 1 there were four false positive (FP), and no false negative ( FN ) result, for trisomy 1 8 two FN and no FP result, and for trisomy 1 3 four FP and no FN (Table 1 ). Thus, the estimates for the false positive rates (FPRs; Table 1 ) are 0.063 % for trisomy 21 , <0.001 % for trisomy 1 8, and 0.062 % for trisomy 1 3. The detection rates (DRs; Table 1 ) are > 99.99% for trisomy 2 1 and trisomy 1 3 (84/84 and 1 6/1 6, respectively), and in the order of 90% for trisomy 1 8 ( 1 7/ 1 9). Three out of three US classified samples for chromosome 2 1 , and four out of four US samples for chromosome 1 3 turned out to be normal diploid; four of out the eight US for chromosome 1 8 turned out to be normal diploid, and two were FN .

Table 1

Detection Rate [CRl Posit.* Predictive False PCHIHW Rate j Negative Predictive Sensitivity Value (PfV] [FPR] j VaJue [NPVj

j l -Specificity !

[9596 confidence irtwveij ¹ ( S¾ otA eit emtennQ f95¾co ¾fkfoi cintefv¾j¾ { [95% confidence irvtgfval]

Tffeowf »^{' ' '} _. / *»-99* - lOO}_. 95.45* [88,12 - 98.53J 0.063»* [0.17 - 0.02 J 99.99% [99.92 - 1O0J Trhofny 18 90.47%· [61.17 - 98.13) >99.99* [79.07 - 100] 0001% 0.07 - 0.00] 99.96* [99.87 - 99 99] Trisomy 13 99.99* [75.92 - 100] 80.0* [55.73 - 95.6] 0.062» [0.17-0.02] >99.99% [99.92 - 100]

•one of two false negatives would be avoided with the current test conditions

t*three of the four fa!se positives eeukt be predicted and potentially avoided under the current test conditions

The reasons for the two FN trisomy 1 8 results were in one case a low fetal fraction retroactively determined to be < 3% (in the first data set before introduction of routine fetal fraction measurements), and a bona fide true fetal mosaicism of type 5 (TFM5 ) for the other, since the fetal fraction was average ( 1 0%; determined retroactively, also in the first data set), yet the signal intensity was between that of diploid reference values and the threshold.

Example 2: Detection of FP vs. CPM

Low signal intensity in the presence of adequate, average fetal fractions was the best predictor of FP results. Overall there were eight FPs: four for T21 , four for T1 3, and none for T1 8. The key difference between true positives (TPs) and false positives (FPs) is that TPs (n = 1 0) had significantly higher z-scores than the FPs (n = 7 ) yet similar fetal fractions (z- scores and fetal fractions for TPs/ FPs were 1 5.6 ± 3.7/4.97 ± 0.79 ( f-test P < 0.0001 ) and 1 1 .6 ± 1 .91 %/ 1 0.2 ± 1 .75% ( P = 0.1 623 )). One FP for T21 was indistinguishable from TPs based solely on the z-score of 29.8 and the fetal fraction of 25%.

Three out of four FP trisomy 21 results, and three out of four FP trisomy 1 3 results had intermediate z-scores between the threshold of +3.3 and +4.9, yet fetal fractions of 7.5%, 9.2%, 9.4%, 1 1 .1 %, 1 1 .2 %, 1 2.6%, suggesting placental mosaicism (confined placental mosaicism of type 1 or 3; CPM 1 /3). One each of FP trisomy 1 3 and 21 results were considered to represent candidate high-grade CPM /3, because the signal intensity and the fetal fraction were concordantly high and very high (z-scores +6.8, fetal fraction 9.8% and 25% ), yet amniocentesis showed diploidy.

If low-grade placental confined mosaicism (CPM ) is the principal reason for FPs, then the trisomy z-scores will increase only marginally, even for maximal fetal fractions of 50%. For example, for five FPs, the fetal fraction was enriched from 1 0.28 ± 1 .56% to 51 .08 + 8.2% ( P < 0.0001 ) and the corresponding z-scores moved from 5.1 1 ± 0.93 to 7.74 ± 2.97 ( P = 0.0964). For one single FP with an initial z-score of 29.8, the enrichment from 25% to 56.7% moved the z-score to 54.25. This indicates that low z-scores as a proxy for low- grade CPMs are a more frequent cause of FPs than high z-scores as proxy for high-level mosaicism, provided that the fetal fractions are comparable and average for both, in accordance with predictions based on CVS data ( alvestiti F, et al. Prenat Diagn 201 5; doi: 1 0.1 002/pd.4656; Grati FR, et al. Eur J Hum Genet 2006; 14: 282-288; Grati FR, et al. Prenat Diagn 201 5; 35: 801 - 809; Grati FR, et al. Genet Med 2014; 1 6: 620-624).

Example 3: (a) Detection of SCAs

Fifty three cases (0.83 %; 53/6'388) of SCAs were found (Figure 1 ), 38 (71 % ) with monosomy X ( MX), 1 0 ( 1 8.9% ) with triple X syndrome (TXS), and 5 (9.4%) with Klinefelter syndrome (KS). Inversely, 99.1 7% of samples had a physiological sex chromosomal dosage. Out of the 38 monosomy X cases, 32 were classified as pathogenic, and 6 as likely pathogenic. Four out of those 32 considered pathogenic had z-scores largely superior to those obtained for the highest fetal fractions, suggesting a maternal MX mosaicism. The four candidate maternal MX cases had fetal fractions and z-scores of 8.8% ± 4.25 and -30.7 ± 1 1 .1 1 . Compared to the confirmed exclusively fetal reference MX cases (n = 1 5; z-scores -9.28 ± 3.56), the values for the candidate maternal MX mosaicism cases are expected to occur for less than one in a million exclusively fetal cases with 45, X (z-scores <-26.7). All ten TXS, and all 5 KS were considered pathogenic, one TXS had a z-score largely exceeding those of the highest fetal fractions pointing at a maternal TXS. The z-scores for the validated fetal reference 47, XXX cases were 1 2.1 ± 2.29 (n = 5 ) and the the candidate maternal 47,XXX case had a z-score of 75.3; such values are expected for less than one in a trillion cases with purely fetal 47, XXX (z-scores > 28.4).

(b) Follow-up validation The follow-up with amniocentesis was 60.5% (23/38) for MX, 30% (3/ 1 0) for KS, and 20% ( 1 /5) for TXS. The inquiry of suspected newborns did not reveal clinically evident cases present at birth, and particularly not the presence of classical Turner syndromes (TS). Specifically, three cases were followed-up at birth, two FP results, including one with a normal postnatal karyotype 46, XX, and one of US who was clinically unsuspicious (total follow-up 26/39, 67%). There were no known FN and 17 FP results for MX, and no known FN and FP results for both TXS and KS (Table S1 ).

Three FP results were likely attributable to maternal MX, which was however not adressed directly with an analysis performed on maternal blood. Low signal intensity concomitantly with adequate, average fetal fractions was - similarly to the common trisomy group - a reliable predictor of FP results. Four out of a total of six MX results with low signal intensity (belonging to the likely pathogenic group) were followed-up with amniocentesis, and all four were FP, yet the fetal fractions available for three of them were high (8.7%, 9.8%, 1 6.1 %).

The higher prevalence of FP results for MX compared to the other SCA and common trisomies is expected, since the CPM 1 /3 rate is much higher, in the order of 60% for MX and mosaicism exceeding 50%.30 This fully accounts for the positive predictive value (PPV) of 55% observed for MX (Table S1 ). The FPR for MX is 0.26%, the FPRs for the other SCAs is for both KS and TXS <0.01 %. The DRs for the SCAs were nominally >99.9%, specifically 21 /21 for MX (8/8 directly validated), 10/ 10 KS and 5/5 for TXS. This excludes only mosaic aneuploidies that are clinically detectable at birth.

Table S1 : cf DNA test performance for the sex chromosome anomalies (SCAs), monosomy X, Klinefelter- and Triple X-syndromes The detection rates (DRs) and false positive rates (FPRs), as well as positive and negative predictive values (PPV; NPV) expressed in percentages - including the 95% confidence intervals in parentheses - are given. Detection Rate [DR] False Positive Rate [FPRJ Positive Predictive Value Negative Predictive Value

Aneuploldy Sensitivity 1 - Specificity w [NPVj

[n]

195% CI] [95¼ CI] {95% Ci] [95% αι

Monosomy X [38] >99,99S* [80.75 - 100] 0.27%^*** [0.44 - 0.17] 55.26%" [38.4? - 71.00] >93.99¾ (99.92 - 100]

Klinefelter [10] >99.99%* [61.1? - 100] <0.001% [0.07 - 0.00] >93.99¾ [65.54 - 100] >99.S9% (99.92 - 100]

Trisomy X [5] >99.99%* [46.29 - 100J <o,mi% [o.o? - o.oo] >9S.99¾ [46.29 - 100] >99. % [99.92 - 100]

"predicted to be caused chiefly by confined placental mosaicism

*3/i7 F Ps predictiveiy due to maternal mosaic MX

'DRs >99.99¾ and FPRs <0.001¾ in technical validation study²

Abbrev. CI = confidence interval

Example 4: (a) Detection Rare Autosomal Trisomies

In the group of rare autosomal trisomies - autosomal trisomies that are not common trisomies - 50 (50/6388, 0.78%) cases were found (Figure 2). individually, there were 1 6 cases with trisomy 7 (32%), 8 cases with trisomy 8 ( 1 6%), 4 cases with trisomy 1 6 (8%), 3 cases each with trisomy 22, trisomy 1 7 and trisomy 6 (6% each), 2 cases each with trisomy 1 5, trisomy 4, trisomy 3 and trisomy 2 (4% each), and one case each with trisomy 20, trisomy 1 2, trisomy 1 1 , trisomy 10, and trisomy 9 (2% each) (Figure 2).

All trisomy cases potentially associated with uniparental disomy (UPD) were considered to be in principle pathogenic or likely pathogenic, since uniparental disomy (UPD) can be symptomatic even in diploid fetuses after trisomy rescue (trisomy 6, 7, 14, 1 5, 1 6)(see reference 1 4, Eggermann T, et al. Trends in Molecular Medicine 201 5; 21 : 77-87). All other trisomies can potentially occur as full trisomies or mosaic trisomies in affected fetuses (see reference 6, Wellesley D, et al. Eur J Hum Genet 201 2; 20: 521 -526), although this is rare except for trisomy 22 and trisomy 1 6, and were accordingly also rated in principle pathogenic or likely pathogenic. Overall, thirty five out of fifty (70%) trisomies were considered pathogenic, and 1 5 (30%) likely pathogenic (based on signal intensity as outlined previously).

(b) Follow-up validation

The follow-up with amniocentesis was 38% for the group as a whole ( 1 9/50), and focused on critical groups, for instance 100% (3/3) for trisomy 22, 50% (2/4) for trisomy 1 6, and 37.5% (6/1 6) for trisomy 7, which included routine molecular UPD analysis in addition to karyotyping. Four fetal aneup!oidies were confirmed, all three trisomy 22 cases were fetal. as well one case of trisomy 1 2. For all remainder cases amniocentesis revealed normal diploid results, and in the cases with potential UPD no single fetal UPD was identified. This resulted in a nominal FPR of 0.71 %, although based on biological plausibility virtually all of these cases are due to CP 1 /3 (Table S2) (see reference 25, Grati FR, et al. Eur J Hum Genet 2006; 1 4: 282-288). Accordingly, the PPV was low (8%) likely due to placental mosaicism. Globally, the best predictor of true positive results - and by extension of FP - was the natural prevalence of individual aneuploidies.

Table S2: cfDNA test performance for the rare autosomal trisomies: The detection rates (DRs) and false positive rates (FPRs), as well as positive and negative predictive values (PPV; NPV) expressed in percentages - including the 95% confidence intervals in parentheses - are given

Detection Rate [DRj False Positive Rate [FPR] Positive Predictive Value Negative Predictive Value

Aneuploidy Sensitivity 1 -Specificity [PPV] [NPVJ

W [95% CI] {95% CI] [95% CI] 195% a j

Rare

autosomal

>99.99%* (39.57 - 100] Q.71%- (0.96 - 0.53] 8%* 12.59 - 20.11] >S9.99% {99.92 - 100J trisomies

[50]

''predicted to be caused by confined placental mosaicism iCP )

Thighly accurate detection of fetal trisomy mosaicism., and CPM trisomies Abbrev. C» = confidence interval

Example 4: (a) Detection Copy Number Variation (CNV)

Thirtysix candidate CNVs were identified, 25 in the first and 1 1 in the second data set (0.56%) (Table 2). All were rated pathogenic or likely pathogenic without clear cut distinction because of the difficulty of integrating multiparametric factors, such as for instance the implicated critical regions, size, penetrance, and inheritance. Copy number variation in two genomic regions known to undergo recurrent rearrangements and to overlap with wellk- nown genomic disorders were more frequently found, namely two deletions and two duplications of the DGS region 22q1 1 .2 ( 1 1 %), and two deletions and two duplications of the 1 6p1 3.1 1 critical region ( 1 1 % ), (Table 2B) (see reference 1 7, Cooper GM, et al. Nat Genet 201 1 ; 43: 838-846. Corrected after print 27 August 201 4; doi: 1 0.1038/ng.909). Other recurrent CNVs were identified, two duplications of the 1 5q 1 1 .2- and one duplication of the 17p12-region (Table 2B). Non recurrent CNVs not estimated to overlap with known genomic disorders are listed in Table 2A.

Table 2A: CNVs that do or do not overlap with known genomic disorders. Non recurrent CNVs are listed according to individual chromosome (HSA) and chromosome band impli^¬ cated. Duplications lines 1-2, 4-10, 12, 14-18, 20-22, 24-25, deletions lines 3, 11, 13, 19, 23.

A CNVs not overlapping with genomic disorders

1^s' data set (n) 2^nd data set (n)

11q12-13

2

1

1

Table 2B CNVs that do or do not overlap with known genomic disorders. Recurrent CNVs are listed according to individual chromosome (HSA) and chromosome band implicated. Duplications lines 1-2,4-5,7, deletions lines 3, 6. B CNVs overlapping with genomic disorders

1^5t data set (n) 2^nd data set (n)

HSA15:sJjj£l 5q1 1.2-15q13 1

HSA15:djjpJ 5q1 1.2-15q14 1

HSA16:d^t16p13.12-16p12.2/p12.3 1 1

2

1

HSA22:dup 22q11.2 1 1

(b) Follow-up validation

For 1 1 /36 (30.5% ) follow-up data with CMA and/or FISH analysis after amniocentesis and parental analyses were available. For the recurrent CNVs with defined pathogenicity the follow-up was near complete, specifically it was available for 3/4 CNVs of the 22q 1 1 .2 region, and 3/4 CNVs of the 1 6p1 3.1 1 , as well as for the duplication 1 7p1 2 (PMP22). Two duplications and one deletion 22q 1 1 .2, all of maternal origin, were confirmed, as well as one duplication and one deletion 1 6p 1 3.1 1 , again of maternal origin. Likewise, the duplication 1 7p1 2 of maternal origin was confirmed. In the non-recurrent CNV class, a deletion 9q31 of maternal origin was confirmed. Overall three candidate CNVs were not confirmed, one of the deletions 1 6p 1 3.1 1 , and in the non-recurrent CNV class, a deletion 5q 14.3q23.2, and a deletion 1 0q22-q33, totaling three FP results. During the follow-up of pregnancies in the first data set, prior to fetal fraction measurements, a 1 0 Mb terminal deletion 4p had not been called (one FN result).

Based on this experience, the analytical detection rates ( DR) were defined experimentally for recurrent CNVs of defined size, and as a function of the fetal fraction (Figure 3 ). These experiments indicate a robust detection of CNVs of the size class of 3 Mb for the most frequently implicated region 22q 1 1 .2 and the average range of fetal fractions of 1 0% (see also discussion) . Accordingly, it is recommended to limit the clinical use to detection of recurrent CNVs of defined size and penetrance (reference 1 7, Cooper GM, et al. Nat Genet 201 1 ; 43: 838-846. Corrected after print 27 August 201 4; doi: 1 0.1 038/ng.909) for which the DR can be defined with the use of positive controls and depending on the fetal fraction. In conclusion, the follow-up of a series of 6'388 consecutive clinical cases was reviewed using cfDNA screening with a low genomic coverage and detecting a broad range of aneu- p!oidy classes, namely the common trisomies, the SCA, the rare autosomal trisomies, as well as deletion and duplication CNVs.

The clinical fo!low-up with amniocentesis of results considered pathogenic, likely pathogenic or of US was between 30% - 80% for the four aneuploidy categories (mean 42.5% ± 10.8%). It was generally close to complete - between 75% - 80% - for results considered of US and for complex results suggesting that the mother also carried the aneuploidy. The follow-up of normal results - classified as benign or likely benign - was ensured by an online birth registry alimented on a voluntary basis = y gynecologists (n30), by an inquiry of randomly selected samples (n=500; 8.6% of all normal results) and by proximity to addressing physicians. Independently, these follow-up figures typically are higher than those reported in the literature, especially for the group with uncertain pathogenicity.

In comparison with the most recently updated meta-analysis (reference 1 , Gil MM et al, Ultrasound Obstet Gynecol 201 5; 45: 249-266) both very high DRs combined with very low FPRs for the common trisomies are reported. Specifically, overall only two false negative results were recorded, both for trisomy 18; one was caused by a too low fetal fraction and would have been avoided under current conditions, the other had an adequate fetal fraction and cannot be prevented based on current understanding. Out of overall eight FP results six had adequate fetal fractions (average 10.1 % ± 1 .8%) and z-scores close to threshold (between 3.3 and 4.9). Possible explanations include placental mosaicism (CPM 1 /3), vanishing twin and technical factors. The inventors did not find any evidence of maternal CNVs as potential confounding factor, as put forward by one study (reference 2, Zhang H et al. Ultrasound Obstet Gynecol 201 5; 45: 530-538). Perhaps the initial validation studies used to set the thresholds contained truly positive samples with !owgrade, placentally confined mo- sacism diagnosed based solely on CVS data. This would imply that currently thresholds are set too low. Future work will clarify this issue. Concerning technical reasons the inventors are not aware of factors capable of generating such artefacts that however cannot be excluded upfront either. The proportion of FPs explained by a demising twin is hard to access, since there is no systematic measurement of number of heartbeats in early pregnancy. The most important conclusions from this part are, firstly, that the likelihood of FPs can be predicted based on z-scores and fetal fraction, which makes them preventable, and secondly. that if confirmed it is probably advisable in the future to increase the threshold to call aneu- ploidies.

As for SCAs, an initial technical validation study demonstrated that those and the rare autosomal trisomies could in principle be detected with an accuracy similar to that of the common trisomies (see references 1 ,22, Gil MM et al. Ultrasound Obstet Gynecol 201 5; 45: 249-266 ; Guex N, et al. Prenat Diagn 201 3; 33: 707-710). In accordance with the prediction based on CVS data (see references 25, 30, Grati FR, et al. Eur J Hum Genet 2006; 14: 282-288; Grati FR, et al. Genet Med 2014; 1 6: 620-624) however, placental mosaicism was responsible for a significant proportion of FPs likely accounting for 37% of all MX results. Vanishing twins again constitute a mystifying cause that could not be safely dissociated from placental mosaicism, since i) no systematic recording of the number of heartbeats in early pregnancy is done, and ii) during the first trimester screening period the embryonic sac of the demising twin is not regularly detectable by standard operators, and iii) the inventors did not manage to systematically analyze term placentas to prove mosaicism.

Interestingly, three out of the 1 7 FPs making up 7.9% of all MX results were probably due to a maternal mosaic MX. Although these three cases were not worked up, the first sample detected since termination of the current study - that was indistinguishable from the three based on the z-scores and fetal fraction - was indeed shown to correspond to a 14% maternal mosaic MX (mos 45,X[7]/46,XX[43]). In conclusion for this part, maternal aneu- ploidy is predictable based on z-scores and fetal fraction, and thus represents a preventable source of FPs.

Based on the current data series - although more relevant experience will be required - it is suggested that detection of rare autosomal trisomies is clinically useful for at least three reasons, i) trisomy 22 appears to be a sizeable cause of truly fetal aneupioidy that was reliably detected, ii) although UPD cases were not detected in the present study this appears to be a mare matter of numbers, and iii), even without UPD placental mosaicism carries a small but definite risk for IUGR and even more for SGA infants at birth.

The inventors initially designed CNV detection to be genome-wide in principle. The experience with this data set clearly revealed the limitations of such an approach: for most if not all non-recurrent CNVs solid and reliable data on allele frequencies, segregation, implicated genes and pathways and phenotypic consequences are lacking, which makes clinical implementation in a prenatal setting problematic. Forthis reason the inventors emphasis is put on the diagnostic workup on recurrent CNVs. This experience clearly showed that this subclass of CNVs can be robustly detected in the context of a routine clinical program. Specifically, spike-in experiments ( Figure 3 ) using positive control cfDNA allowed to define the sensitivity of detection for defined CNV size classes dependent on the fetal fraction. Similarly to what the inventors described for the SCAs the signal intensity could also be used to faithfully predict if a CNV is likely of maternal origin. Consequently, such predictions can be implemented to prevent false positive CNVs calling.

The following recommendations can be made based on the present study: i) cfDNA screening should be extended to include both detection of rare autosomal trisomies and deletion/duplication CNVs; ii) integrated interpretation of the fetal fraction and z-scores should be used to stratify the likelihood of false negative results, and false positive results caused by placental mosaicism and maternal aneuploidy/CNVs; iii) CNV detection should be based on experimental validation of defined CNV size-classes integrating the fetal fraction, and shouid be limited to a list of well characterized genomic disorders; the latter can be expanded with increasing experience and knowledge. Extensions of the present study could comprise targeted high-coverage sequencing to closer investigate UPD and the potential recessive unmasking associated with it, as well as the search for frequent single-gene disorders in a more remote future.

REFERENCES

1 Gil MM, Quezada MS, Revello R, et al. Analysis of cell-free DNA in maternal blood in screening for fetal aneuploidies. Ultrasound Obstet Gynecol 201 5; 45: 249-266.

2 Zhang H, Gao Y, Jiang F, et al. Non-invasive prenatal testing for trisomies 21 , 1 8, and 1 3: clinical experience from 1 46958 pregnancies. Ultrasound Obstet Gynecol 201 5; 45: 530- 538.

3 McCullough RM, Almasri EA, Guan Y, et al. Non-invasive prenatal chromosomal aneu- ploidy testing - clinical experience; 1 00,000 clinical samples. PLoS One 201 4; 9: e1 091 73. doi: 10.1 371 /journal-pone.01 091 73

4 Bianchi DW, Parsa S, Bhatt S, et al. Fetal sex chromosome testing by maternal plasma DANN sequencing. Obstet Gyn 201 5; 1 25: 375-382. 5 Davis C, Cuckle H, Yaron Y. Screeing for Down syndrome - incidental diagnosis of other aneuploidies. Prenat Diagn 2014; 34: 1 044- 1 048.

6 Wellesiey D, Dolk H, Boyd PA, et al. Rare chromosome abnormalities, prevalence and prenatal diagnosis rates from population-based congenital anomaly registers in Europe. Eur J Hum Genet 201 2; 20: 521 -526.

7 Lebo RV, Novak RW, Wolfe K, et al. Discordant circulating fetal DNA and subsequent cytogenetics reveal false negative, placental mosaic, and fetal mosaic cfDNA genotypes. J Transl Med 201 5; 1 3: 260. doi 1 0.1 1 86/s 1 2967-01 5-0569-y.

8 T0rring N, Petersen OB, Becher N, et al. First trimester screening for other trisomies than trisomy 2 1 , 1 8 and 1 3. Prenat Diagn 201 5; 35: 61 2-61 9

9 Grande M, Jansen FAR, Blumenfeld YJ, et al. Genomic microarray in fetuses with increased nuchal translucency and normal karyotype - a systematic review and meta-analysis. Ultrasound Obstet Gynecol 201 5; doi: 10.1 002/uog. 1 4880

1 0 Neufeld-Kaiser WA, Cheng EY, Liu YJ, et al. Positive predictive value of non-invasive prenatal screening for fetal chromosome disorders using cell-free DNA in maternal serum; independent clinical experience of a tertiary referral center. BMC Medicine 201 5; 1 3: 1 29. doi 1 0.1 1 86/S1 291 6-01 5-0374-8.

1 1 Kalousek DK, Pantzar T, Tsai M, et al. Pathology of abortion: chromosomal and genetic correlations. Monogr Pathol 1 991 ; 33 : 228-256.

1 2 Malvestiti F, Agrati C, Grimi B, et al. Interpreting mosaicism in chorionic villi: results of a monocentric series of 1 001 mosaics in chorionic villi with foliow-up by amniocentesis. Prenat Diagn 201 5; doi: 1 0.1 002/pd.4656.

1 3 Wallerstein R, Misra S, Dugar RB, et al. Current knowledge of prenatal diagnosis of mosaic autosomal trisomy in amniocytes: karyotype/phenotype correlations. Prenat Diagn 201 5; 35: 1 -7.

1 4 Eggermann T, Soeliner L, Buiting K, et al. Mosaicim and uniparental disomy in prenatal diagnosis. Trends in Molecular Medicine 201 5; 21 : 77-87. 1 5 Robinson WP, Penaherrera MS, Jiang R, et al. Assessing the role of placental trisomy in preeclampsia and intrauterine growth restriction. Prenat Diagn 2010; 30: 1 -8.

1 6 Sudmant PH, Mallick S, Nelson BJ, et al. Global diversity, population stratification, and selection of human copy number variation. Science 201 5. doi: 10.1 1 26/science.aab3761 .

1 7 Cooper GM, Coe BP, Girirajan S, et al. A copy number variation morbidity map of developmental delay. Nat Genet 201 1 ; 43: 838-846. Corrected after print 27 August 2014; doi: 10.1 038/ng.909.

1 8 Jensen TJ, Dzakula Z, Deciu C, et al. Detection of microdeletion 22q1 1 .2 in a fetus by nextgeneration sequencing of maternal plasma. Clin Chem 2012; 58: 1 148- 1 1 51 .

1 9 Srinivasan A, Bianchi DW, Huang H, et al. Noninvasive detection of fetal subchromoso- mal abnormalities via deep sequencing of maternal plasma. Am J Hum Genet 201 3; 92: 1 67- 1 76.

20 Zhao C, Tynan J, Ehrich M, et al. Detection of fetal subchromosomal abnormalities by sequencing circulating cell-free DNAfrom maternal plasma. Clin Chem 201 5; 61 : 608-61 6.

21 Helgeson J, Wardrop J, Almasri E, et al. Clinical outcome of subchromosomal events detected by whole-genome noninvasive prenatal testing. Prenat Diagn 201 5; doi: 10.1 002/pd.4640.

22 Guex N, Syngelaki A, Deluen C, et al. A robust second-generation genome-wide test for fetal aneuploidy based on shotgun cell-free DNA in maternal blood. Prenat Diagn 201 3; 33: 707-71 0.

23 Bayindir B, Dehaspe L, Brison N, et al. Noninvasive prenatal testing using a novel analysis pipeline to screen for all autosomal fetal aneuploidies improves pregnancy outcome. Eur J Hum Genet 201 5; doi: 1 0.1038/ejhg.2014.282.

24 Mersy E, de Die-Smulders CEM, Coumans ABC, et al. Advantages and disadvantages of different implementation strategies of non-invasive prenatal testing in Down syndrome screening programs. Pub Health Genomics 201 5; doi: 10.1 1 59/000435780. 25 Grati FR, Grimi B, Frasoli G, et al. Confirmation of mosaicism and uniparental disomy inamniocytes, after detection of mosaic chromosome abnormalities in chorionic villi. Eur J Hum Genet 2006; 14: 282-288.

26 Grati FR, Gomes DM, Ferreira JCPB, et al. Prevalence of recurrent pathogenic microdele- tions and microduplications in over 9500 pregnancies. Prenat Diagn 201 5; 35: 801 - 809.

27 Coppinger J, Alliman S, Lamb AN, et al. Whole-genome microarray analysis in prenatal specimens identifies clinically significant chromosome alterations without increase in results of unclear significance compared to targeted microarray. Prenat Diagn 2009; 29: 1 1 56- 1 1 66.

28 Hehir-Kwa JY, Pfundt R, Veltman JA, et al. Pathogenic or not? Assessing the clinical relevance of copy number variants. Clin Genet 201 3: 84: 41 5-421 .

29 Vulto-van Silfhout AT, Hehir-Kwa JY, von Bon BWM , et al. Clinical significance of de novo and inherited copy-number variation. Hum Mutat 201 3; 34: 1 679- 1 687.

30 Grati FR, Malvestiti F, Ferreira JCPB, et al. Fetoplacental mosaicism: implications for false- positive and false-negative noninvasive prenatal screening results. Genet Med 2014; 1 6: 620-624.

31 Snyder MW, Simmons LE, Kitzman JO, et al. Copy-number variation and false positive prenatal aneuploidy screening results. N Engl J Med 201 5; 372: 1 639- 1 645.

Claims

Claims

1 . Method for detecting fetal aneuploidy from a maternal biological test sample, comprising

h) extracting cell-free DNA from said test sample;

i) analyzing the size distribution of the cell-free DNA within each sample and selecting a set of samples based on the size distribution of the DNA molecules within said samples

j) performing massive parallel sequencing of the extracted cell-free DNA of step (a) to obtain sequence reads,

k) mapping the obtained sequence reads to the human genome for each sample;

I) calculating the z-score for the signal intensity for the sequence reads of a chromosome or chromosomal region of interest and comparing it to a z-score threshold value;

m) calculating the fetal fraction for the sequence reads of a chromosome or chromosomal region of interest and comparing it to a fetal fraction threshold value;

n) detecting a fetal aneuploidy in said test sample or not, based on the combined interpretation of steps e) and f ).

2. Method according to claim 1 , wherein the step of selecting a set of samples based on the size distribution of the DNA molecules comprises a step of elimination of DNA molecules having a size greater than 200 bp, greater than 1 80 bp, greater than 1 50 bp, greater than 1 25 bp from the sample to obtain an enriched sample, wherein at least 90 wt%, preferably more than 95 wt% of the DNA molecules have a size of less than 200 bp, less than 1 80 bp, less than 1 50 bp, less than 1 25 bp for example between 1 1 5 and 1 25 bp.

3. Method according to claims 1 or 2, wherein the step of performing a massively parallel sequencing includes the preparation of a sequencing library followed by sequencing.

4. Method according to any preceding claim, wherein the sequencing step comprises sequencing at least 25 million sequences for each sample. Method according to any preceding claim, wherein the chromosome of interest is chromosome 2 1 , chromosome 1 6, chromosome 1 8, chromosome 1 3 or chromosome 1 1 .

Method according to any preceding claim, wherein a signal intensity in step e) close to the z score threshold of >6x 3.3 and < 5 indicates a false positive result.

Method according to any preceding claim, wherein a signal intensity in step e) close o the z score threshold of > 6x 3 and < 5 indicates a false positive result.