To Örebro University

Next generation sequencing (NGS) has revolutionized precision diagnostics by enabling high-throughput analysis of nucleic acids. This thesis combines technical validations with innovative applications of NGS across five studies.

Papers I–II focus on molecular autopsies, demonstrating that hybridisation-based whole-exome sequencing can be successfully applied to formalin-fixed paraffin-embedded (FFPE) tissue, even in severely fragmented samples. Using matched blood and FFPE samples, our complete workflow for variant detection achieved a sensitivity of 97% and a positive predictive value of 98%. Applied to clinical cases, 23 of 35 FFPE samples were successfully sequenced, and relevant variants were detected in previously unresolved cases of sudden unexplained death. Paper III expanded forensic analysis on blood using the same hybridisation-based NGS-technology.

Papers IV–V explore liquid biopsy for pan-cancer detection. Using enzymatic conversion and targeted methylation sequencing of plasma circulating cell-free DNA (cfDNA), we identified 162 differentially methylated regions (DMRs) and developed a classifier for pan-cancer detection with sensitivity and specificity of 83.8%. Fragmentomics analysis revealed cancer-associated patterns in cfDNA fragment length and end motifs: cancer samples exhibited shorter median fragment lengths and alterations in fragment end motifs. These findings highlight fragmentomics as a promising biomarker for cancer detection.

Together, these studies illustrate the versatility of NGS for precision diagnostics—from post-mortem genetic analysis to minimally invasive cancer screening—and underscore the importance of rigorous validation to bridge research and clinical implementation.

Cancer detection is challenging, especially in patients with diffuse symptoms that overlap with non-malignant conditions. Here we show that plasma protein profiling can identify cancer among patients with non-specific symptoms. Using proximity extension assay-based proteomics of 1463 plasma proteins from 456 patients presenting with non-specific symptoms sampled prior to cancer diagnostic work-up and diagnosis, we identify 29 proteins associated with new cancer diagnoses. We develop a model able to stratify 160 cancer cases and 296 non-cancer cases with an area under the curve of 0.80, maintaining performance (0.82) in an independent replication cohort of 238 patients. The model also distinguishes cancer from autoimmune, inflammatory and infectious diseases. Designed as a triage tool, our model based on a blood test could help prioritize patients at higher cancer risk for rapid and highly sensitive diagnostic modalities such as positron emission tomography-computed tomography. These findings emphasize the potential of blood proteome profiling to support timely diagnosis and transform clinical medicine.

OBJECTIVE: Clinical validation of sperm selection device ZyMōt™ for standard IVF.

METHODS: The pre-clinical validation of ZyMōt™ included several steps. First, split semen preparation compared density gradient centrifugation (DGC) to ZyMōt™ with primary outcome fraction and absolute number of progressive motile sperm. Second, sibling oocytes were fertilized with sperms prepared with DGC and sperms selected by ZyMōt™, primary endpoint fertilization rate, utility rate, embryo development pace quality. After this, DGC was replaced by ZyMōt™, first without centrifugation steps, and then with a five-minute centrifugation step and subsequent media change prior to gamete co-incubation. Endpoint was assessment of key performance indicators against previous results using DGC for standard IVF.

RESULTS: ZyMōt™ resulted in purer sperm selection compared to DGC (fraction progressive motile sperm 97.2±3.1% vs. 83.0±14.1%, p<0.01). Fertilization of sibling oocytes resulted in similar fertilization rates and utility rates, and no differences in embryo development pace or quality. However, after changing sperm selection protocol from DGC to ZyMōt™ for standard IVF for all fresh semen samples with motile sperm, the fertilization rates and utility rates were significantly reduced, and cases of total failure of fertilization increased substantially. Adding five-minute centrifugation and media change after centrifugation to the sperm selection protocol restored fertilization rate, including total failure of fertilization rate, to normal.

CONCLUSIONS: To conclude, the ZyMōt™ sperm selection device is suitable for standard IVF only after inclusion of five minutes centrifugation and subsequent media change prior to gamete co-incubation.

BACKGROUND AND AIMS: Familial hypercholesterolemia (FH) and other disorders with similar features are common genetic disorders that remain underdiagnosed and undertreated, due in part to the cost of screening. The aim of this study was to design and implement a whole gene targeted NGS panel for the molecular diagnosis of FH and statin intolerance with an emphasis on high quality variant calling, including copy number analysis.

METHODS: A whole gene panel for hybridisation-based short read NGS was designed for the dominant FH-genes low density lipoprotein receptor (LDLR), apolipoprotein B (APOB), proproteinconvertas subtilisin/kexin type 9 (PCSK9), apolipoprotein E (APOE) and the recessive FH-genes low density lipoprotein receptor adaptor protein 1 (LDLRAP1), ATP binding cassette subfamily member 5/8 (ABCG5/8) and lipase A, lysosomal acid type (LIPA), as well as solute carrier organic anion transporter family member 1B1 (SLCO1B1), not an FH gene but linked to statin intolerance. Polygenetic risk score markers were also included. The panel was used for screening of a Swedish FH-study population (n = 133).

RESULTS: The panel sequencing resulted in high coverage and confident variant calling of included genes. Known causal variants were found in common dominant FH-genes in 43 % of the cohort. Copy number variants were found in LDLR in 10 individuals and a whole gene deletion of SLCO1B1 in one individual. In addition, coding variants in recessive genes and rare non-coding intronic and untranslated region variants were found in a large proportion of the study individuals highlighting the need for extended gene panels.

CONCLUSIONS: This new tool can be used for a comprehensive high-quality molecular genetic analysis according to guidelines for the diagnosis and treatment of FH.

PURPOSE: Map the nuclear error phenotypes in the two-cell embryo after assisted reproduction using time lapse images and the effect on good quality blastocyst formation.

METHODS: Retrospective cohort study using time lapse images, categorizing 2331 two-cell embryos from 392 patient couples and 504 ART cycles categorizing each embryo as mononucleated, multinucleated, micronucleated, binucleated, split nucleation or mixed error. Correlating nuclear error phenotype with good quality blastocyst formation rate (BFR) using contingency tables and unadjusted odds ratio.

RESULTS: An overall nuclear error rate of 47.1% was observed in two-cell embryos. The most frequent error was multi-nucleation (14.2%) followed by mixed error (11%), micro-nucleation (8.6%), bi-nucleation (7.4%) and split nucleation (5.8%). Blastocyst formation rate (BFR) was reduced in embryos with nuclear errors, 46.2% for embryos with one cell affected, 27.6% for embryos with both cells affected, compared to 58.6% for mononucleated cells, p < 0.001 for both. Binucleated embryos were as likely as mononucleated embryos to become clinically useful blastocysts (56.8% vs 58.6%, n.s., unadjusted OR 0.94), whereas all the other phenotypes were less likely to develop into good quality blastocysts. The worst outcome was noted for embryos with split nucleation, with just 12.4% BFR, OR 0.12 (0-08-0.21), p < 0.001.

CONCLUSION: Nuclear errors are common at the two-cell stage. Overall, presence of nuclear errors reduces the likelihood of becoming good quality blastocysts. Both the number of affected cells and the different nuclear error phenotypes have impact on blastocyst formation rate, except binucleated embryos.

We investigated differences in circulating cell-free DNA (cfDNA) methylation between patients with cancer and those presenting with severe, nonspecific symptoms. Plasma cfDNA from 229 patients was analyzed, of whom 37 were diagnosed with a wide spectrum of cancer types within 12 months. Samples underwent enzymatic conversion, library preparation, and enrichment using the NEBNext workflow and Twist pan-cancer methylation panel, followed by sequencing. Methylation analysis was performed with nf-core/methylseq. Differentially methylated regions (DMRs) were identified with DMRichR. Machine learning with cross-validation was used to classify cancer and controls. The classifier was applied to an external validation set of 144 controls previously unseen by the model. Cancer samples showed higher overall CpG methylation than controls (1.82% vs. 1.34%, p < 0.001). A total of 162 DMRs were detected, 95.7% being hypermethylated in cancer. Machine learning identified 20 key DMRs for classification between cancer and controls. The final model achieved an AUC of 0.88 (83.8% sensitivity, 83.8% specificity), while mean cross-validation performance reached an AUC of 0.73 (57.1% sensitivity, 77.5% specificity). The specificity of the classifier on unseen control samples was 79.2%. Distinct methylation differences and DMR-based classification support cfDNA methylation as a robust biomarker for cancer detection in patients with confounding conditions.

Sudden unexpected death (SUD) is an unexpected event that in many cases are caused by diseases with an underlying genetic background. Forensic molecular autopsy is an approach that has gained wide-spread attention, in part explained by the rapid progress of DNA sequencing techniques. The approach leverages genetic data in combination with medical autopsy findings in post-mortem samples to explore a potential underlying genetic cause of death. Traditional forensic approaches to molecular autopsy focus on a small panel of genes, say <200 genes, with strong association to heart conditions whereas clinical genetics tend to capture entire exomes while subsequently selecting targeted panels bioinformatically. The drop in price and the increased throughput has promoted wider exome sequencing as a viable method to discover genetic variants. We explore a targeted gene panel consisting of 2422 genes, selected based on their broad association to sudden unexplained death. A hybridization capture approach from Twist Bioscience based on double stranded DNA probes was used to target exons of the included genes. We selected and sequenced a total of 98 post-mortem samples from historical forensic autopsy cases where the cause of death could not be unambiguously determined based on medical findings and that had a previous negative molecular autopsy. In the current study, we focus on the performance of the hybridization capture technology on a 2422 gene panel and explore metrics related to sequencing success using a mid-end NextSeq 550 as well as a MiSeq FGx platform. With the latter we demonstrate that our sequence data benefits from 2×300 bp sequencing increasing coverage, in particular, for difficult regions where shadow coverage, i.e. regions outside the probes, are utilized. The results further illustrate a highly uniform capture across the panel of genes (mean fold80=1.5), in turn minimizing excessive sequencing costs to reach sufficient coverage, i.e. 20X. We outline a stepwise procedure to select genes associated with SUD through virtual bioinformatical panels extracting tier of genes with increasing strength of association to SUD. We propose some prioritization strategies to filter variants with highest potential and show that the number of high priority genetic variant requiring manual inspections is few (0-3 for all tiers of genes) when all filters are applied.

Background/Objectives: Familial hypercholesterolemia (FH) is a common genetic disorder which is primarily caused by pathogenic variants in the LDLR, APOB, and PCSK9 genes. Approximately 10% of pathogenic variants in LDLR may be CNVs. Here, we combine NGS, MLPA, and Optical Genome Mapping (OGM) to investigate CNVs in LDLR.

Methods: A NGS panel was designed for whole gene sequencing (8 genes) of 100 FH patients using Twist technology and Illumina platform. CNVs were detected using CNVexpo, and an in-house pipeline for base-resolved normalized coverage. Identified CNVs were validated using MLPA and OGM. Bionano Services Lab performed the OGM procedure. Purified gDNA was labeled using Direct Label and Stain DNA Labeling Kit. Saphyr chip was run aiming for 100X coverage. De novo assembly and Variant Annotation pipelines were executed on Bionano Solve v3.7. Bionano Access v1.7 was used for CNV reporting and visualization.

Results: In five out of 100 samples NGS and MLPA data showed heterozygous deletions in LDLR. Three deletions, affecting different exons, was analyzed and confirmed using OGM. In two samples, OGM better defined the breakpoints as well as the size of the event, which expanded far beyond the gene of interest. In one sample, an additional CNV of SLCO1B1, a pharmaco-gene, important for transport of statins used for FH treatment was identified.

Conclusion: CNVs in FH genes in FH patients could be detected using targeted NGS, which was further confirmed by MLPA and characterized using OGM.

We developed an efficient approach to diagnostic copy number analysis of targeted gene panel or whole exome sequence (WES) data. Here we present CNV-Z as a new tool for detection of copy number variants (CNVs). Deletions and duplications of chromosomal regions are widely implicated in both genomic evolution and genetic disorders. However, calling CNVs from targeted or exome sequence data is challenging. In most cases, the copy number of a chromosomal region is estimated as the depth of reads mapping to a certain bin or sliding window divided by the expected number of reads derived from a set of reference samples. This approach will inevitably miss smaller CNVs on an irregular basis, and quite frequently results in a disturbing number of false positive CNVs. We developed an alternative approach to detect CNVs based on deviation from expected read depth per position, instead of region. Cautiously used, the cohort of samples in the same run will do as a reference. With appropriate filtering, given high quality DNA and a set of suitable reference samples, CNV-Z detects CNVs ranging in length from one nucleotide to an entire chromosome, with few false positives. Performance is proved by benchmarking using both in-house targeted gene panel NGS data and a publicly available NGS dataset, both sets with multiplex ligation-dependent amplification probe (MLPA) validated CNVs. The outcome shows that CNV-Z detects single- and multi-exonic CNVs with high specificity and sensitivity using different kind of NGS data. On gene level, CNV-Z shows both excellent sensitivity and specificity. Compared to competing CNV callers, CNV-Z shows higher specificity and positive predictive value for detecting exonic CNVs.

Forensic molecular autopsies have emerged as a tool for medical examiners to establish the cause of death. It is particularly useful in sudden unexplained deaths where the cause of death cannot be determined with a regular medical autopsy. We provide the first study of exome data from formalin-fixed paraffin-embedded samples (FFPE) paired with data from high-quality blood samples in forensic applications. The approach allows exploration of the potential to use FFPE samples for molecular autopsies and identify variants in extensive exome data. We leverage the high uniformity of the hybridization capture approach provided by Twist Bioscience to target the complete exome and sequence the libraries on a NextSeq 550. Our findings suggest that exome sequencing is feasible for 24 out of a total of 35 included FFPE samples. When successful, the coverage across the exome is comparatively high (> 90% covered to 20X) and uniform (fold80 below 1.5). Detailed variant comparisons for matched FFPE and blood samples show high concordance with few false variants (positive predictive value of 0.98 and a sensitivity of 0.97) with no distinct FFPE artefacts. Ultimately, we apply carefully constructed forensic gene panels in a stepwise manner to find genetic variants associated with the clinical phenotype and with relevance to the sudden unexplained death.