Siphiwe Dlamini, Oleg N. Reva1
Centre for Bioinformatics and Computational Biology; Department of Biochemistry, Genetics and Microbiology; University of Pretoria; Lynnwood Rd., Pretoria 002, South Africa
A handful of soil contains an astonishingly complex community of microorganisms. Thousands of bacterial species can live side by side, interacting with plants, competing with one another and contributing to processes such as nutrient cycling, soil fertility and plant health.
But there is a problem: most of these microorganisms cannot easily be grown in the laboratory. To study them, researchers increasingly turn to metagenomics: sequencing all the DNA extracted directly from an environmental sample.
The result is rather like emptying thousands of different jigsaw puzzles into one box and then trying to reconstruct the original pictures.
Long reads give us bigger pieces of the puzzle
Traditional short-read sequencing produces millions of small DNA fragments. Although these fragments contain enormous amounts of information, assembling them into individual microbial genomes can be extremely difficult in something as diverse as soil.
Long-read sequencing technologies help enormously because they generate much larger pieces of DNA. Longer reads can span repetitive regions and other difficult parts of microbial genomes, making it possible to reconstruct much more complete genomes directly from environmental samples.
Yet assembly is only part of the challenge.
Once DNA fragments have been assembled into longer sequences called contigs, bioinformatics programs must determine which contigs came from the same microorganism. This process is known as contig binning, and the reconstructed genomes are called metagenome-assembled genomes, or MAGs.
Current binning programs are remarkably useful, but they are not perfect. They usually decide which pieces belong together by examining properties such as DNA sequence composition and how frequently particular sequences occur in a sample. Closely related microorganisms can have very similar properties, meaning that pieces from different genomes may occasionally end up in the same bin.
But DNA carries another kind of information
This is where long-read sequencing offers an intriguing additional opportunity.
DNA is not simply a string of the letters A, T, G and C. Microorganisms can chemically modify particular DNA bases through a process called methylation.
Different bacteria possess different sets of enzymes that create these modifications. As a result, even closely related microorganisms can carry distinctive patterns of methylation across their genomes. These patterns can effectively act as epigenetic fingerprints.
Modern long-read technologies such as PacBio SMRT and Oxford Nanopore sequencing can detect information associated with these DNA modifications while the DNA is being sequenced. In other words, valuable information about a microbial genome may already be present in the sequencing data. We simply need to make better use of it.
That led us to a simple question:
Could we use a microorganism’s methylation fingerprint to determine whether DNA fragments grouped into the same reconstructed genome really belong together?
Adding a second identity check
To explore this idea, we developed a new sort-contigs module for our open-source software, SeqWord Motif Mapper (SWMM; https://github.com/SeqWord/SWMM).
Instead of replacing existing metagenomic binning programs, SWMM provides an additional check.
Conventional programs first group contigs according to characteristics such as sequence composition and abundance. SWMM then examines the DNA methylation patterns of those contigs.
If the pieces assigned to one genome show similar methylation signatures, this provides additional support for keeping them together. If some pieces have distinctly different signatures, SWMM can separate them for further analysis.
In our study, we tested the approach using both artificially constructed mixtures of bacterial genome fragments and real soil metagenomes obtained from the rhizosphere, the soil directly surrounding the roots, of wheat plants grown under controlled experimental conditions. This work was carried out at the University of Pretoria as part of the EU Horizon TRIBIOME project, where we are using advanced genomic and bioinformatic approaches to better understand microbial communities associated with wheat. The study was supported by the EU-HORIZON TRIBIOME project (https://www.tribiome.eu/). The project workflow is summarized in the figure below:
Figure was created using OpenAI ChatGPT Images 2.0.
What did we find?
The results showed just how useful this additional layer of information can be.
Across 236 groups containing multiple contigs, at least 45% contained contigs with significantly different methylation patterns. This suggested potential grouping errors that had not been resolved using sequence composition and sequencing coverage alone.
Tests using artificial mixtures, where we already knew which genome every DNA fragment came from, provided further proof of concept: sorting by methylation signatures enriched the resulting groups with contigs originating from the same genome. The method could even distinguish closely related bacterial lineages when informative methylation patterns were available.
There was another interesting possibility. Abrupt changes in methylation patterns along an individual contig sometimes highlighted potentially chimeric contigs – sequences that may have been incorrectly assembled from DNA originating from different microorganisms. Methylation information may therefore help not only with sorting DNA fragments but also with checking the quality of metagenomic assemblies.
From genome sequence to genome identity
The approach is still developing. Methylation information does not solve every metagenomic assembly problem, and its usefulness depends on factors including sequencing coverage, contig length and the number of detectable methylated sites. Our current implementation can still split fragments belonging to the same genome or occasionally retain fragments from different genomes together.
But the broader idea is promising.
Long-read sequencing allows us to look beyond the DNA sequence itself. The same sequencing experiment can provide another layer of biological information that helps identify where pieces of DNA came from.
For highly complex environments such as soil, where conventional approaches can struggle to distinguish closely related microorganisms, these epigenetic fingerprints could become a valuable additional tool for reconstructing microbial genomes.
Perhaps the most exciting aspect is that we do not necessarily need another experiment to obtain this information. With long-read sequencing, the methylation signal can already be captured alongside the DNA sequence. The challenge now is learning how to use this additional information effectively.
Instead of asking only “What is the DNA sequence?”, metagenomics can increasingly ask another question:
“What biological fingerprint does this DNA carry and can that fingerprint tell us where it belongs?”
More on this topic can be found in the following preprint publication:
Dlamini S, Maseko S, Mokoatsi B, Mehlomakulu N, Reva ON. Methylation-guided refinement of metagenomic contig binning using the SeqWord Motif Mapper sort-contigs module. Available at SSRN 7235437.
TAGS: #Metagenomics #LongReadSequencing #SoilMicrobiome #Bioinformatics #Epigenetics #DNA Methylation #MicrobialGenomics #MAGs #TRIBIOME