American Society of Human Genetics Annual Meeting (ASHG) 2024

Learn how the 6-base genome—A, C, T, G, 5 5mC, and 5hmC— can be used to identify stage 1 colorectal cancer with unprecedented sensitivity and provides improved predictive power throughout disease progression.

📢 bioRXiv preprint alert! Distinguishing 5mC and 5hmC, enabled through the 6-base genome, increases stage I detection of colorectal cancer in liquid biopsy cfDNA samples. Read more

Liquid biopsy for profiling of cell free DNA (cfDNA) in blood holds huge promise to transform how we experience and manage cancer by early detection and identification of residual disease and subtype. While early work in liquid biopsy focused on the identification of actionable somatic variations at specific loci, the past decade has seen an expansion into epigenetic features, notably methylation. 5-methylcytosine (5-mC) profiles of cancer are differential from non-cancer at many more loci and so provide a stronger signal. Moreover, recent research has suggested that 5-hydroxymethylcytosine (5-hmC) profiles in cfDNA can be a marker for early cancer. However, a standard blood draw yields an average of only 10ng of cfDNA, presenting the dilemma of how to use limited sample to obtain maximum information.

We will present the application of a technology which sequences at single base resolution the complete genetic sequence of input DNA fragments integrated with the modification status (unmodified, 5-mC or 5-hmC) for each CpG from low nanogram input quantities of cell-free DNA (cfDNA). Using this technology, we generated whole genome 6-base data (A, C, G, T, 5-mC, and 5-hmC) on cfDNA extracted from plasma of healthy volunteers and patients with colorectal cancer at stages I-IV. We demonstrate how the technology can be used to extract as complete information as possible from a cfDNA sample.

We compare differential 5-mC and 5-hmC regions, identify copy number variation and clinically relevant germline and somatic single nucleotide polymorphisms, and multiple fragmentomic features including size, end motif and nucleosome position, across different stages of colorectal cancer. We demonstrate that the AUC for detection of Stage I CRC from healthy volunteers is 0.69 or 0.72 when training a classifier using only 5mC or 5hmC features, respectively, and that this increases to 0.93 when training a classifier using the combination of 5mC and 5hmC features. We highlight individual CRC-relevant features where the resolution of 5mC and 5hmC distinguishes between healthy, Stage I and Stage IV samples, and that these differences would be invisible when just using 5mC information. We propose that the ability to measure 5-mC and 5-hmC at high accuracy and single base resolution, alongside genomic and fragmentomic profiles, from a limited quantity of DNA will enable greater insight into the ctDNA in plasma and enable the development of more accurate liquid biopsy based disease detection.

A body of evidence indicates that epigenetic modifications in DNA comprises a fundamental pathway by which genes can be silenced or activated, determining cell fate and function. In mammalian genomes, the fifth carbon of cytosine is one major target for epigenetic modification. 5-methylcytosine (5mC) is associated with gene silencing and patterns of this modification are altered in diseases such as cancer. In contrast, 5-hydroxymethylcytosine (5hmC) has only recently been suggested to play a role in gene regulation and is generally considered a marker of active genes and enhancers. Currently, reading these two modifications remains elusive at the single cell level, making the function and relationship between these modifications difficult to precisely ascertain. Traditionally, a large number of cells are combined and analysed as a bulk, which produces average methylome profiles. This can hide cryptic heterogeneity such as distinct subpopulations with distinct methylation patterns.

Here we present a single nuclei workflow to simultaneously determine 5mC and 5hmC at the single molecule level. Our method utilises a transposase to simultaneously tag and fragment DNA with a hairpin sequence containing a molecular barcode unique to the cell of origin. After tagmentation, we use an armoury of enzymes, including a high-fidelity maintenance methyltransferase together with a methylcytosine dioxygenase, a helicase and a deaminase to unequivocally read 5mC and 5hmC in addition to the four canonical genetic bases from the same molecule.

Applying this method, we determined the 5mC and 5hmC patterns in the ES cells line (ES-E14TG2a) cultured under two conditions (Serum+LIF+2i and Serum+LIF) at high sensitivity (>95% for both modification) and specificity (>98%). The method successfully recovers distinct populations that have distinctive (hydroxy)methylome trajectories at regulatory regions. In addition, we demonstrate consistency between pseudo-bulk and bulk measurements for these different cell types. Furthermore, we extend this method to cells isolated from complex tissues such as mouse cortex, where we identify multiple cell populations based on their respective (hydroxy)methylome patterns.

In summary, we demonstrate simultaneous, phased reading of all four genetic bases and the two most biologically relevant cytosine modifications at the single cell level. This provides a more complete picture of the information stored in cellular genomes and will have applications throughout biology and medicine.

A key challenge in genomics is the generation and integration of data across different modalities. This typically requires several assays to be performed, which is costly and time-consuming, and the subsequent integration of these data across assays is technically challenging. Here, we leverage an assay that enables sequencing the complete genetic sequence and the DNA modifications, 5-methylocytosine (5mC) and 5-hydroxy-methylocytosine (5hmC), from low nanogram amounts of DNA, to provide 6-base genomic data.

Given 5mC and 5hmC play key roles in gene regulation and chromatin organisation, we aimed to explore how these multimodal data could further elucidate key biological processes and yield novel insight. To this end, we trained and evaluated a series of machine-learning models to predict gene expression, chromatin accessibility, and enhancer state from 6-base sequence data.

We use 6-base data from a mouse embryonic stem cell-line, ES-E14TG2A, alongside publicly available polyA RNA-seq, ATAC-seq, TT-seq, and histone modification data of the same cell line as training data and evaluate the performance of the models on held-out test chromosomes. We show that these models can generate highly accurate predictions of gene expression (polyA RNA-seq prediction: R2=0.75, Spearman’s 𝜌;=0.86; TT-seq prediction: R2=0.85, Spearman’s 𝜌; =0.91) and chromatin accessibility (ATAC-seq prediction: R2=0.83, Spearman’s 𝜌;=0.93). Importantly, we found stronger performance of our models for predicting TT-seq over polyA RNA-seq signal, suggesting that 6-base data offers a powerful window into nascent transcriptional activity. As well as being able to predict continuous expression and chromatin accessibility metrics, we show 6-base data predicts enhancer state (Active, Primed, Repressed), defined by histone modifications, with 91% accuracy. In all models the addition of resolved 5hmC signal over undifferentiated 5mc and 5hmC improved predictive performance.

We have shown that the combination of resolved methylation and genomic data combined with machine-learning can generate accurate inference of other data modalities which play key roles in gene regulation. Thus, there is a compounding effect whereby 6-base genomic assays not only yield direct data, but also the foundations for multiple other inferred modalities. Looking ahead, these approaches can enable novel insights into core biological processes and accelerate the speed of iteration for experimental projects, where one can yield multifaceted insights and even conduct pilot experiments in silico using predictive models.