Benchmarking of methods for DNA methylome deconvolution.

Defining the number and abundance of different cell types in tissues is important for understanding disease mechanisms as well as for diagnostic and prognostic purposes. Typically, this is achieved by immunohistological analyses, cell sorting, or single-cell RNA-sequencing. Alternatively, cell-specific DNA methylome information can be leveraged to deconvolve cell fractions from a bulk DNA mixture. However, comprehensive benchmarking of deconvolution methods and modalities was not yet performed. Here we evaluate 16 deconvolution algorithms, developed either specifically for DNA methylome data or more generically. We assess the performance of these algorithms, and the effect of normalization methods, while modeling variables that impact deconvolution performance, including cell abundance, cell type similarity, reference panel size, method for methylome profiling (array or sequencing), and technical variation. We observe differences in algorithm performance depending on each these variables, emphasizing the need for tailoring deconvolution analyses. The complexity of the reference, marker selection method, number of marker loci and, for sequencing-based assays, sequencing depth have a marked influence on performance. By developing handles to select the optimal analysis configuration, we provide a valuable source of information for studies aiming to deconvolve array- or sequencing-based methylation data.

Genomic origin, fragmentomics, and transcriptional properties of long cell-free DNA molecules in human plasma.

Recent studies have revealed an unexplored population of long cell-free DNA (cfDNA) molecules in human plasma using long-read sequencing technologies. However, the biological properties of long cfDNA molecules (>500 bp) remain largely unknown. To this end, we have investigated the origins of long cfDNA molecules from different genomic elements. Analysis of plasma cfDNA using long-read sequencing reveals an uneven distribution of long molecules from across the genome. Long cfDNA molecules show overrepresentation in euchromatic regions of the genome, in sharp contrast to short DNA molecules. We observe a stronger relationship between the abundance of long molecules and mRNA gene expression levels, compared with short molecules (Pearson's r = 0.71 vs. -0.14). Moreover, long and short molecules show distinct fragmentation patterns surrounding CpG sites. Leveraging the cleavage preferences surrounding CpG sites, the combined cleavage ratios of long and short molecules can differentiate patients with hepatocellular carcinoma (HCC) from non-HCC subjects (AUC = 0.87). We also investigated knockout mice in which selected nuclease genes had been inactivated in comparison with wild-type mice. The proportion of long molecules originating from transcription start sites are lower in Dffb-deficient mice but higher in Dnase1l3-deficient mice compared with that of wild-type mice. This work thus provides new insights into the biological properties and potential clinical applications of long cfDNA molecules.

Cell-free DNA methylome analysis for early preeclampsia prediction.

Preeclampsia (PE) is a leading cause for peripartal morbidity, especially if developing early in gestation. To enable prophylaxis in the prevention of PE, pregnancies at risk of PE must be identified early-in the first trimester. To identify at-risk pregnancies we profiled methylomes of plasma-derived, cell-free DNA from 498 pregnant women, of whom about one-third developed early-onset PE. We detected DNA methylation differences between control and PE pregnancies that enabled risk stratification at PE diagnosis but also presymptomatically, at around 12 weeks of gestation (range 9-14 weeks). The first-trimester risk prediction model was validated in an external cohort collected from two centers (area under the curve (AUC) = 0.75) and integrated with routinely available maternal risk factors (AUC = 0.85). The combined risk score correctly predicted 72% of patients with early-onset PE at 80% specificity. These preliminary results suggest that cell-free DNA methylation profiling is a promising tool for presymptomatic PE risk assessment, and has the potential to improve treatment and follow-up in the obstetric clinic.

Machine learning-based detection of immune-mediated diseases from genome-wide cell-free DNA sequencing datasets.

The early detection of tissue and organ damage associated with autoimmune diseases (AID) has been identified as key to improve long-term survival, but non-invasive biomarkers are lacking. Elevated cell-free DNA (cfDNA) levels have been observed in AID and inflammatory bowel disease (IBD), prompting interest to use cfDNA as a potential non-invasive diagnostic and prognostic biomarker. Despite these known disease-related changes in concentration, it remains impossible to identify AID and IBD patients through cfDNA analysis alone. By using unsupervised clustering on large sets of shallow whole-genome sequencing (sWGS) cfDNA data, we uncover AID- and IBD-specific genome-wide patterns in plasma cfDNA in both the obstetric and general AID and IBD populations. We demonstrate that pregnant women with AID and IBD have higher odds of receiving inconclusive non-invasive prenatal screening (NIPS) results. Supervised learning of the genome-wide patterns allows AID prediction with 50% sensitivity at 95% specificity. Importantly, the method has the potential to identify pregnant women with AID during routine NIPS. Since AID pregnancies have an increased risk of severe complications, early recognition or detection of new-onset AID can redirect pregnancy management and limit potential adverse events. This method opens up new avenues for screening, diagnosis and monitoring of AID and IBD.

Pan-Cancer Detection and Typing by Mining Patterns in Large Genome-Wide Cell-Free DNA Sequencing Datasets.

BACKGROUND: Cell-free DNA (cfDNA) analysis holds great promise for non-invasive cancer screening, diagnosis, and monitoring. We hypothesized that mining the patterns of cfDNA shallow whole-genome sequencing datasets from patients with cancer could improve cancer detection. METHODS: By applying unsupervised clustering and supervised machine learning on large cfDNA shallow whole-genome sequencing datasets from healthy individuals (n = 367) and patients with different hematological (n = 238) and solid malignancies (n = 320), we identified cfDNA signatures that enabled cancer detection and typing. RESULTS: Unsupervised clustering revealed cancer type-specific sub-grouping. Classification using a supervised machine learning model yielded accuracies of 96% and 65% in discriminating hematological and solid malignancies from healthy controls, respectively. The accuracy of disease type prediction was 85% and 70% for the hematological and solid cancers, respectively. The potential utility of managing a specific cancer was demonstrated by classifying benign from invasive and borderline adnexal masses with an area under the curve of 0.87 and 0.74, respectively. CONCLUSIONS: This approach provides a generic analytical strategy for non-invasive pan-cancer detection and cancer type prediction.

More of less: Novel multi-ome profiling of single human neurons.

Epigenetic modifications to DNA and chromatin interact to influence gene expression and cellular phenotypes, but defining these omics layers in complex tissues is a daunting task. In this issue of Cell Genomics, Luo et al. describe a novel single-cell multi-omic method, simultaneously profiling transcriptome, DNA methylome, and chromatin accessibility, to shed light on human neurons.

Non-invasive prenatal testing suggesting a maternal malignancy: What do we tell the prospective parents in Belgium?

Cancer is diagnosed in one in 1000 to 1500 pregnancies. Most frequently encountered malignancies during pregnancy are breast cancer, hematological cancer, cervical cancer and malignant melanoma. Maternal cancer is associated with an increased risk of IUGR and preterm labor, especially in patients with systemic disease or those receiving chemotherapy during pregnancy, requiring a high-risk obstetrical follow-up. Fetal aneuploidy screening by non-invasive prenatal testing (NIPT) can lead to the incidental identification of copy number alterations derived from non-fetal cell-free DNA (cfDNA), as seen in certain cases of maternal malignancy. The identification of tumor-derived cfDNA requires further clinical, biochemical, radiographic and histological investigations to confirm the diagnosis. In such cases, reliable risk estimation for fetal trisomy 21, 18 and 13 is impossible. Therefore, invasive testing should be offered when ultrasonographic screening reveals an increased risk for chromosomal anomalies, or when a more accurate test is desired. When the fetal karyotype is normal, long term implications for the fetus refer to the consequences of the maternal disease and treatment during pregnancy. This manuscript addresses parental questions when NIPT suggests a maternal malignancy. Based on current evidence and our own experience, a clinical management scheme in a multidisciplinary setting is proposed.

Comprehensive genome-wide analysis of routine non-invasive test data allows cancer prediction: A single-center retrospective analysis of over 85,000 pregnancies.

BACKGROUND: Implausible false positive results in non-invasive prenatal testing (NIPT) have been occasionally associated with the detection of occult maternal malignancies. Hence, there is a need for approaches allowing accurate prediction of whether the NIPT result is pointing to an underlying malignancy, as well as for organized programs ensuring efficient downstream clinical management of these cases. METHODS: Using a data set of 88,294 NIPT performed at University Hospital Leuven (Belgium) between November 2013 and March 2020, we retrospectively evaluated the positive predictive value (PPV) of our NIPT approach for cancer detection. In this approach, whole-genome cell-free DNA (cfDNA) data from NIPT were scrutinized for the presence of (sub)chromosomal copy number alterations (CNAs) predictive for a malignancy, using an unbiased NIPT analysis pipeline coined GIPSeq. For suspected cases, the presence of a maternal cancer was evaluated via subsequent multidisciplinary clinical follow-up examinations. The cancer-specificity of the identified CNAs in cfDNA was assessed through genetic analyses of a tumor biopsy. FINDINGS: Fifteen women without a cancer history were identified with a GIPSeq result suggestive of a malignant process. Their cfDNA profiles showed either genome-wide aberrations or a single trisomy 8. Upon clinical examinations, a solid or hematological cancer was identified in 4 and 7 cases, respectively. Three women were identified as having a clonal mosaicism. For one case no underlying condition was found. These numbers add to a PPV of 73%. Based on this experience, we presented a multidisciplinary care path for efficient clinical management of these cases. INTERPRETATION: The presented approach for analysing NIPT results has a high PPV, yet unknown sensitivity, for detecting asymptomatic malignancies upon routine NIPT. Given the complexity of diagnosing a pregnant woman with cancer, clinical follow-up should occur in a well-designed multidisciplinary setting, such as via the care model that we presented here. FUNDING: This work was supported by Research Foundation Flanders and KU Leuven funding.

Breast Cancer Detection and Treatment Monitoring Using a Noninvasive Prenatal Testing Platform: Utility in Pregnant and Nonpregnant Populations.

BACKGROUND: Numerous publications have reported the incidental detection of occult malignancies upon routine noninvasive prenatal testing (NIPT). However, these studies were not designed to evaluate the NIPT performance for cancer detection. METHODS: We investigated the sensitivity of a genome-wide NIPT pipeline, called GIPSeq, for detecting cancer-specific copy number alterations (CNAs) in plasma tumor DNA (ctDNA) of patients with breast cancer. To assess whether a pregnancy itself, with fetal cell-free DNA (cfDNA) in the maternal circulation, might influence the detection of ctDNA, results were compared in pregnant (n = 25) and nonpregnant (n = 25) cancer patients. Furthermore, the ability of GIPSeq to monitor treatment response was assessed. RESULTS: Overall GIPSeq sensitivity for detecting cancer-specific CNAs in plasma cfDNA was 26%. Fifteen percent of detected cases were asymptomatic at the time of blood sampling. GIPSeq sensitivity mainly depended on the tumor stage. Also, triple negative breast cancers (TNBC) were more frequently identified compared to hormone-positive or HER2-enriched tumors. This might be due to the presence of high-level gains and losses of cfDNA or high ctDNA loads in plasma of TNBC. Although higher GIPSeq sensitivity was noted in pregnant (36%) than in nonpregnant women (16%), the limited sample size prohibits a definite conclusion. Finally, GIPSeq profiling of cfDNA during therapy allowed monitoring of early treatment response. CONCLUSIONS: The results underscore the potential of NIPT-based tests, analyzing CNAs in plasma cfDNA in a genome-wide and unbiased fashion for breast cancer detection, cancer subtyping and treatment monitoring in a pregnant and nonpregnant target population.

Noninvasive prenatal diagnosis by genome-wide haplotyping of cell-free plasma DNA.

PURPOSE: Whereas noninvasive prenatal screening for aneuploidies is widely implemented, there is an increasing need for universal approaches for noninvasive prenatal screening for monogenic diseases. Here, we present a cost-effective, generic cell-free fetal DNA (cffDNA) haplotyping approach to scan the fetal genome for the presence of inherited monogenic diseases. METHODS: Families participating in the preimplantation genetic testing for monogenic disorders (PGT-M) program were recruited for this study. Two hundred fifty thousand single-nucleotide polymorphisms (SNPs) captured from maternal plasma DNA along with genomic DNA from family members were massively parallel sequenced. Parental genotypes were phased via an available genotype from a close relative, and the fetal genome-wide haplotype and copy number were determined using cffDNA haplotyping analysis based on estimation and segmentation of fetal allele presence in the maternal plasma. RESULTS: In all families tested, mutational profiles from cffDNA haplotyping are consistent with embryo biopsy profiles. Genome-wide fetal haplotypes are on average 97% concordant with the newborn haplotypes and embryo haplotypes. CONCLUSION: We demonstrate that genome-wide targeted capture and sequencing of polymorphic SNPs from maternal plasma cell-free DNA (cfDNA) allows haplotyping and copy-number profiling of the fetal genome during pregnancy. The method enables the accurate reconstruction of the fetal haplotypes and can be easily implemented in clinical practice.

Fetal sex determination in twin pregnancies using non-invasive prenatal testing.

Non-invasive prenatal testing (NIPT) is accurate for fetal sex determination in singleton pregnancies, but its accuracy is not well established in twin pregnancies. Here, we present an accurate sex prediction model to discriminate fetal sex in both dichorionic diamniotic (DCDA) and monochorionic diamniotic/monochorionic monoamniotic (MCDA/MCMA) twin pregnancies. A retrospective analysis was performed using a total of 198 twin pregnancies with documented sex. The prediction was based on a multinomial logistic regression using the normalized frequency of X and Y chromosomes, and fetal fraction estimation. A second-step regression analysis was applied when one or both twins were predicted to be male. The model determines fetal sex with 100% sensitivity and specificity when both twins are female, and with 98% sensitivity and 95% specificity when a male is present. Since sex determination can be clinically important, implementing fetal sex determination in twins will improve overall twin pregnancies management.

De Novo Assembly of Two Swedish Genomes Reveals Missing Segments from the Human GRCh38 Reference and Improves Variant Calling of Population-Scale Sequencing Data.

The current human reference sequence (GRCh38) is a foundation for large-scale sequencing projects. However, recent studies have suggested that GRCh38 may be incomplete and give a suboptimal representation of specific population groups. Here, we performed a de novo assembly of two Swedish genomes that revealed over 10 Mb of sequences absent from the human GRCh38 reference in each individual. Around 6 Mb of these novel sequences (NS) are shared with a Chinese personal genome. The NS are highly repetitive, have an elevated GC-content, and are primarily located in centromeric or telomeric regions. Up to 1 Mb of NS can be assigned to chromosome Y, and large segments are also missing from GRCh38 at chromosomes 14, 17, and 21. Inclusion of NS into the GRCh38 reference radically improves the alignment and variant calling from short-read whole-genome sequencing data at several genomic loci. A re-analysis of a Swedish population-scale sequencing project yields > 75,000 putative novel single nucleotide variants (SNVs) and removes > 10,000 false positive SNV calls per individual, some of which are located in protein coding regions. Our results highlight that the GRCh38 reference is not yet complete and demonstrate that personal genome assemblies from local populations can improve the analysis of short-read whole-genome sequencing data.

SweGen: a whole-genome data resource of genetic variability in a cross-section of the Swedish population.

Here we describe the SweGen data set, a comprehensive map of genetic variation in the Swedish population. These data represent a basic resource for clinical genetics laboratories as well as for sequencing-based association studies by providing information on genetic variant frequencies in a cohort that is well matched to national patient cohorts. To select samples for this study, we first examined the genetic structure of the Swedish population using high-density SNP-array data from a nation-wide cohort of over 10 000 Swedish-born individuals included in the Swedish Twin Registry. A total of 1000 individuals, reflecting a cross-section of the population and capturing the main genetic structure, were selected for whole-genome sequencing. Analysis pipelines were developed for automated alignment, variant calling and quality control of the sequencing data. This resulted in a genome-wide collection of aggregated variant frequencies in the Swedish population that we have made available to the scientific community through the website https://swefreq.nbis.se. A total of 29.2 million single-nucleotide variants and 3.8 million indels were detected in the 1000 samples, with 9.9 million of these variants not present in current databases. Each sample contributed with an average of 7199 individual-specific variants. In addition, an average of 8645 larger structural variants (SVs) were detected per individual, and we demonstrate that the population frequencies of these SVs can be used for efficient filtering analyses. Finally, our results show that the genetic diversity within Sweden is substantial compared with the diversity among continental European populations, underscoring the relevance of establishing a local reference data set.