EPIC-seq, (short for Epigenetic Expression Inference by Cell-free DNA Sequencing), is a high-throughput method that specifically targets gene promoters using cell-free DNA (cfDNA) sequencing. By employing non-invasive techniques such as blood sampling, it infers the expression levels of targeted genes. It consists of both wet and dry lab stages.
EPIC-seq involves deep sequencing of the transcription start sites (TSS). It hypothesizes that with deep sequencing of these TSSs, usage of fragmentomic features, chromatin fragmentation patterns or properties, can allow high-resolution analyses, as opposed to its alternatives.
The method has been shown effective for gene-level expression inference, molecular subtyping of diffuse large B cell lymphoma (DLBCL), histological classification of nonsmall-cell lung cancer (NSCLC), evaluation of results of immunotherapy agents, and assessment of the genes' prognostic importance. EPIC-seq uses machine learning to deduce the RNA expression of the genes and proposes two new metrics: promoter fragmentation entropy (PFE), an adjusted Shannon Index for entropy, and nucleosome-depleted region (NDR) score, the depth of sequencing in NDR regions. PFE showed superior performance compared to earlier metrics for fragmentomic features.
Additionally, EPIC-seq has been mentioned as a possible solution for detecting tissue damage and esophagus cancer using methylation profiles of cfDNAs, profiling of donor liver molecular networks, and inflammatory bowel disease (IBD) detection.
Background
Historical Usage of cfDNA and fragmentomic features cfDNA, cell death-related and chromatin fragmented DNA molecules contained in blood plasma, has been used to detect transplant tissue rejection, prenatal fetal aneuploidy testing, tumour profiling, and early cancer detection in previous research. Nevertheless, prevalent liquid biopsy methods for cfDNA profiling depend on detecting germline or somatic genetic variations, which may be absent even in high disease burden-bearing patients and cancers with high tumour mutation rates.
Historically, the usage of fragmentomic features of cfDNA samples was shown to be another method to approach the problems mentioned. They demonstrated the capability to inform about the originated tissue classification of cfDNA molecules, which can help segregate tumour-related somatic mutations. However, current methods that use fragmentomic features, such as shallow whole genome sequencing (WGS) on cfDNA, do not fully cover all the tissues' effects and provide low sequencing depth and breadth to infer low-level, for example, gene level, properties. Hence, these methods require a high tumour burden from the patients.
Circulating Tumor DNA profiling Circulating tumour DNA (ctDNA) molecules are tumour-derived cell-free DNA (cfDNA) circulating in the bloodstream and are not associated with cells. CtDNA primarily arises from chromatin fragmentation accompanying tumour cell death and can be extracted by liquid biopsy. CtDNA analysis has been implemented for noninvasive identification of tumour genetic characteristics and early recognition of various cancer forms. The majority of current ctDNA analysis depends on genetic differences in germline or somatic cells to diagnose diseases and detect tumour cells at an early stage. While looking at genetic variations of ctDNA can be beneficial, not all ctDNAs contain genetic mutations. EPIC-seq unitized epigenetic features of ctDNA to inform tissue-of-origin of these unmutated molecules, which is helpful for cancer classification.
Fragmentomic Features for Tissue-of-origin classification The majority of circulating cfDNA molecules are fragments linked to nucleosomes, so they represent unique chromatin arrangements found in the nuclear genomes of the cells they originate from. In particular, open chromatin areas j, whereas genomic regions linked to nucleosomal complexes are often shielded from endonuclease activity. Several studies have identified specific chromatin fragmentomic characteristics that aid in informing tissue origins through cfDNA profiling. These features include:
Reduced sequencing coverage depth Disruption of nucleosome positioning near transcription start sites (TSSs) Length of cfDNA fragments
Principles of EPIC-seq
Currently, the majority of circulating tumour DNA (ctDNA) fragmentomic techniques lack the ability to achieve gene-level resolution and are effective mainly in inferring expression at elevated ctDNA levels. Consequently, they are primarily applicable to patients with notably advanced tumour burdens typically seen in late-stage cancer.
To address this limitation, EPIC-seq employs hybrid capture-based targeted deep sequencing of regions flanking transcription start sites (TSS) in cfDNA. This approach allows for the acquisition of ctDNA fragmentation features crucial for predicting gene expressions, such as Promoter Fragmentation Entropy (PFE) and Nucleosome Depleted Region (NDR). These key fragmentomic features possess the capability to capture associations at the gene level with expression levels throughout the genome, enabling the construction of a predictive model for transcriptional output. This would allows for the high-resolution monitoring of cfDNA fragmentation and gene-level analysis.
Promoter Fragmentation entropy Epic-seq hypothesizes that cfDNA fragments originating from active promoters, which are less shielded by nucleosomes and thus more susceptible to endonuclease cleavage, will display more erratic cleavage patterns compared to fragments from inactive promoters, which are better protected by nucleosomes. PFE is a variation of the Shannon Index, which is a quantitative measure for estimating diversity. In the context of Epic-seq, PFE calculates the diversity of cfDNA fragment lengths where both ends of the fragment are situated within the 2 kb flanking region of each gene's TSS. The higher the PFE of a gene's TSS, the more likely the gene is highly expressed.
Nucleosome Depleted region Actively expressed genes have open chromatin at their TSS region, they are less shielded by nucleosomes and, therefore, more susceptible to endonuclease cleavage. Consequently, the depth of cfDNA originating from the TSS of active genes tends to be shallower compared to that of inactive genes. NDR quantifies the normalized depth within each 2-kilobase window surrounding each TSS. The lower the NDR of a gene TSS site, the more likely the gene is highly expressed.
Methods
Wet Lab workflow
1. Collection and Processing of plasma Peripheral blood samples were obtained and processed to isolate plasma following standard protocols. Upon centrifugation, plasma specimens were preserved at −80 °C, awaiting the extraction of ctDNA. The extraction of cfDNA from plasma volumes ranging from 2 to 16 ml was carried out using established laboratory procedures. Following isolation, the concentration of cfDNA was determined using fluorescence-based quantification methods.
2. Sequencing Library preparation A typical amount of 32 ng of cfDNA was utilized for library preparation. DNA input was adjusted to mitigate the effects of high molecular-weight DNA contamination. The library preparation process encompassed end repair, A-tailing, and adapter ligation, which also incorporated molecular barcodes into each read. These procedures were conducted according to ligation based library preparation standardized protocols, with overnight ligation performed at 4 °C. Following this, shotgun cfDNA libraries underwent hybrid capture targeting specific genomic regions, as detailed below.
3. Custom Capture Panels sequencing Custom capture panels tailored to specific cancer types or personalized selectors were utilized in EPIC-seq. The capture panels targeted transcription start site regions of genes of interest. Enrichment for EPIC-seq was performed following established laboratory protocols. Subsequently, hybridization captures were pooled, and the pooled samples underwent sequencing using short read sequencing.
Dry Lab workflow Since EPIC-seq contains certain computational parts after the wet-lab portion for further processing, the following steps are summarized based on the developers' steps provided in the original paper.
4. Demultiplexing and Error correction If multiplexed paired-end sequencing is used, then demultiplexing needs to be done to sort reads for different samples to different files. After the demultiplexing, error correction and read pair elimination based on unique identifier and barcode matching of pairs can be done. Developers adapt the demultiplexing and error correction steps from the CAPP-seq demultiplexing pipeline.
5. Outer Sequence Removal and trimming For the preservation of shorter fragment reads, barcode removal and adapter trimming need to be done. After read preprocessing, the alignment of reads to the human genome reference should be performed. Original EPIC-seq used hg19 but for better results, an updated version of human genome reference can be used. One should be careful about their aligner's options since some aligners can interfere with the inclusion of shorter reads paired with longer ones. For the deduplication, attached molecular customized barcodes should be exploited. These barcodes include endogenous and exogenous unique molecular identifiers (UMIs) and are handy for distinguishing Polymerase Chain Reaction (PCR) duplicates from real duplicates and hence for PCR duplicate cleansing. This portion is especially important for oncologic applications since the low mutation abundance can be suppressed by PCR duplicates.
6. Read Normalization and quality control If the data for different samples are going to be contrasted with each other, one can perform downsampling on the reads to achieve comparability. The reported sequencing coverage depth for reasonable analysis results was reported as bigger than 500 folds, thus any sample whose mean sequencing depth does not exceed the number can be dropped for more accurate outcomes. Also, EPIC-seq uses estimated expected cfDNA fragment length density of 140–185, based on chromatosomal length. The samples that have outlier fragment length density can be dropped for higher correlation results. As the last quality control step, mapping quality should be considered. A looser threshold can be dictated on EPIC-seq reads, compared to WGS, due to the TSS selection criteria imposed during design phases making the reads more unique for EPIC-seq.
Fragmentomic Feature Analysis
7. Shannon's entropy For the measurement of the diversity of fragmentomic features, the PFE metric, derived from Shannon's Index of entropy, is developed. The default number of 201 bins of lengths 100 to 300 are used for density estimation by the maximum likelihood method. The probability of having a fragment with size , () is computed by the division of the number of fragments with size by the total number of fragments. Shannon's entropy is calculated with the formula: .
8. Dirichlet-Multionomial model Next, as a cleansing against different sequencing depths from different runs and other factors that can hinder the fragment length distribution sanity, Bayesian normalization via the Dirichlet-multinomial model should be done. Per every sample, based on the fragment lengths observed in that sample, a multinomial maximum likelihood estimation-based fragment length distribution is generated. Two intervals of 250 base pair length are used, located between -1000th base pair and -750th base pair, and between 750th base pair and 1000th base pair locations to the centre of TSS. This is done due to the prevention of the impact of gene expression on the generated distribution, as the selected intervals are relatively far away from TSS. Then, the fragment length densities from that distribution are sampled for each 201-fragment size and used as a parameter for Dirichlet distribution generation.