Genome Assembly
In 1995, the first genome was published, the virus Haemophilus influenzae. The first draft human assembly was published in 2001. 2022 saw the publication of the first complete, telomere to telomere human genome assembly. But while the tools and methods for achieving highly contiguous genome assemblies have greatly progressed in recent years, even telomere to telomere length chromosome-scale scaffolds for many species, genome assembly is still hard. This is due to biological complexity, species specific genome characteristics, and technological constraints.
Evolution is messy and a genome represents the four billion years of evolution from the beginning of life to modern organisms. Because of this, genomes themselves are complex quagmires of nucleic acid strings. On paper, assembling a genome is as simple as overlapping sequences of ATGC, but the reality is that the genome contains a huge diversity of sequences with a wide range of functions from the repetitive telomeric sequences at the end of eukaryotic chromosomes to the introns of coding genes. Adding to this complexity are the genomic parasites called transposons and their fossilized remains which are scattered throughout the genome. An additional layer of complexity is that of ploidy. Humans have a rather boring genome which contains two sets of 23 chromosomes, each set inherited from one parent and hypothetically identical. Meaning, that if a genome assembly is to accurately capture the genetics of an individual, then there should be two sets of each chromosome-scale scaffold, aka diploid. Achieving this is known as genome assembly phasing.
Biology is never so straightforward though and many organisms, plants being a good example, have multiple sets. For example, the cultivated strawberry is octoploid. In some strange cases such as the hybrid cultivated Duran wheat, the offspring inherited both diploid chromosome sets. The Both parents were 2n with a total of 14 chromosome. Duran wheat is 28 with 2n from one parent and 2n from the other. On the other end of the ploidy spectrum is an ant with only one set of chromosomes and on the extremely high end is the fern Ophioglossum reticulatum with a total chromosome count of 1440. Heterozygosity can also be a significant challenge in genome assembly. Humans have low levels of heterozygosity, meaning variation between chromosome is relatively low. However, in some species such as sea cucumbers or fruit trees heterozygosity is high. This makes it is difficult for genome assemblers to figure out if slightly different sequences are alleles or duplications in a sequence. A Pacbio summary offers a good insight into these challenges.
In addition to the strange ploidy levels, wide range in number of chromosomes, and heterozygosity there is also large variation in genome size and gene structure. The largest estimated genome size is that of the amoeba Polychaos dubium, with a size of 670gb. The c-value for this was determined in 1968 and modern techniques have not been applied. The smallest genome goes to the parasite Encephalitozoon intestinalis with a genome size of 2.25mb. A good overview of genome size and the c-value enigma can be found on Wikipedia. Gene structure is also widely different between major organismal groups, for example plants and echinoderms tend to have introns that span a few hundred to a few thousand bases whereas introns in mammals tend to significantly larger, often in the tens to hundreds of thousands of bases. Another complication is that plants are more resilient to duplication events, meaning they tend to have a higher number of duplicated regions in the genome. Duplication events can create repetitive sequences which increase the difficulty of genome assembly.
The previously mentioned characteristics of a genome can create difficulties in assembly. For example, while the cost of genome sequencing has significantly decreased in recent years, particularly large genomes can be prohibitively expensive. The largest genome sequenced as of writing this (January 2024) is the European Mistletoe which has a haploid size of 94gb. With current costs, this likely represent hundreds of thousands of dollars of sequencing. Ploidy also represent a challenge as until recently, most genome assemblers output a merged haploid assembly. With improvements in Hi-C scaffolding and optical genome mapping, new genome assemblers take in account ploidy, but most are only designed for diploid genomes.
A telomere-to-telomere genome assembly can be achieved using 50X coverage short read, 30X coverage long read, 10X coverage ultra-long read, and 30X coverage short read Hi-C. Although this is going to be highly dependent on the characteristics of genome. Long reads are assembled into contigs. Hi-C is used to scaffold and orient the contigs into long chromosome-scale sequences. Ultra-long reads are used to fix assembly errors and fill in gaps in the scaffolds, and short reads are used to further polish the assembly, removing any erroneous indels and correcting base errors.
Draft assemblies can be acquired using long reads or short reads. In the case of long reads, these assemblies often contain several thousand long contigs which capture most the genes in the genome but are missing structural and repetitive content such as telomeres, centromeres, transposons, and other repetitive elements. Short read assemblies are of a much lower quality, often containing more than 100,000 contigs and having an N50 less than 10kb, meaning that if you’re lucky the majority of genes will be present, albeit possibly fragmented.
Short Read Assemblies
Illumina short read sequencing is cheap, relatively speaking. For example, 110X coverage of a 900mb sea cucumber genome cost about $1200 in 2019 and it is set to get a lot cheaper with the Illumina patents starting to expire. While short read data won’t get you chromosome length scaffolds, what it will “hopefully” get you is access to most of the gene space in a genome. With some caveats of course. You likely won’t be seeing a complete Titin gene and gene counts will likely be inaccurate as a result of gene fragmentation. For example, in a short read sea cucumber genome, the telomerase gene was split into two separate genes. However, if you’re interested in shorter genes such as Mortalin or highly conserved genes such as Survivin then you’ll probably be just fine with a short read assembly. If you are interested studying gene regulation or repetitive elements, you will want chromosome length scaffolds and will need at least long read sequencing.
Genome assembly using short reads is unfortunately extremely computationally expensive relative to long read genome assembly. However, it has had two decades of algorithm refinement and current tools are designed to make the most out of the data and resources available.
Short-read Genome Assembly
Long Read Assemblies
With the advent of “long-read” sequencing technologies offered by Pacbio and Oxford Nanopore Technologies, the genome sequencing landscape dramatically changed. It became possible to acquire chromosome long scaffolds with repetitive elements mostly resolved. While certain regions of chromosome are still difficult, such as telomeres and centromeres, much of the genome is now available for assembly. A great summary of how long reads have changed genomics can be found here.
While the current state of genetics leaves most researchers interested in the “gene space” of the genome, it has becomes increasingly clear that the regulation of these genes and consequently the expression and phenotype are controlled by numerous factors including cis-regulatory sequences, methylation, and the 3-dimensional folding of chromatin. In order to be able to capture a complete picture of gene regulation and how it contributes to development and phenotype, a chromosome-scale genome assembly is needed. The bare minimum necessary to achieve this is copious amounts of long read sequence data.
The caveat of current long read sequencing technologies is the error rate per a base, often denoted as a Phred Score. Early versions of the long read technology had high rates of errors ranging from an error every 10 to 100 bases. Current versions for Pacbio HiFI are expected to have an error every 1,000 bases and for the most accurate ONT chemistry and nanopores it is five to ten errors every 1,000 bases. Both technologies have strengths and weakness. For example, ONT sequencing can produce what are called Ultra-long reads ranging in length from 100kb to several megabases long. This is especially useful for assembling extremely repetitive genomic regions such as telomeres and centromeres. However, this comes at a cost to the “throughput” as there is a large decrease in the number of reads sequenced. Normal read length for Pacbio HiFi is limited to 15kb-20kb, whereas the range for a standard ONT run is typically 10kb-100kb. Both sequencing platforms have other caveats, a deeper review on this can be found here. That being said, the more long reads you have, the closer and easier it will be to get a chromosome-scale gapless genome assembly.
Assembling Contigs using Long Read Data
Scaffolding using Hi-C and Optical Genome mapping
Scaffolds consist of contigs which have been assigned a position and orientation relative to other contigs and “glued” together using “N”s. These are called “gaps”. If the distance between the contigs is known, then the gaps will be filled to capture that length. However, in some cases the size is unknown, and the gap size is set to an arbitrary number such as 500 for scaffolding tools such as Salsa2 and 3D-DNA.
Additional information is often needed in order to generate chromosome length scaffolds. While there are a number of different methods, the most commonly used approach is using Hi-C data and statistical methods to predict location and orientation of contigs. However, relying solely on this approach, especially for complicated genome, may increase the likelihood of scaffolding errors and manual curation of the chromosome-scale scaffolds is often performed to correct inversions and misjoins. This is critical to perform prior to any attempts at polishing or gap-filling as inversions and misjoins may prevent successful completion of those steps. Additional
Physical approaches such as Optical Genome Mapping (OGM) offer a way to further refine and assign scaffolds to chromosomes as Hi-C scaffold genome assemblies often contain errors that require additional data and manual curation. In OGM, DNA is fixed to a surface and restriction enzymes nick the strand at known motifs. The distance between nicks is then used to identify contigs and orient them relative to each other. The average molecule length for OGM is ~225kb which is sufficient to span most repetitive elements. The cost of OGM is significantly less than ultra-long read sequencing and should be considered as a complement to it for achieving a truly telomere-to-telomere genome assembly.
Scaffolding Long Read Contigs using Hi-C and Optical Genome Mapping
Assembly Phasing
As noted above, most organisms have multiple sets of each chromosome that are “almost” identical. Due to cost, computations, and data limitations genome assemblies were historically haploid. These assemblies often represented a combination of haplotypes. However, this has limitations when trying to understand the genetics of a individual or population. More recently developed genome assembly tools now attempt to separate out the haplotypes and create what is termed either a “haplotype-resolved” or “phased” genome assembly that represents all haplotypes present in a genome.
While genome phasing has been around for sometime as a concept, it was largely relegated to high profile projects. With the advent of Hi-C and long read sequencing, genome phasing has become standard practice for high-quality genomes. However, standards for publication and file type along with tools for analyzing phasing data is still relatively immature. However, having this data can be particularly useful when looking at agricultural crops that are not true-to-type. Meaning the phenotype we consume is specific to one individual that has been cloned. Understanding how the alleles contribute to the phenotype can be important to breeding and crop refinement.
There are several methods for achieving haplotype-resolved genome assemblies and which method to use is largely dependent on the organism’s genome complexity. Plants can have some truly wild genomes that require specific knowledge of karyotype and parental data. As noted earlier on this webpage, microorganisms can also have some extraordinarily complex genomes as well and which genome phasing would likely proof fascinating.
Phasing Genome Assemblies
Organelle Genome
Organelles, such as the powerhouse of the cell - mitochondria, are often forgotten about in the rush to assembly the nuclear genome. They shouldn’t be. Genes found in both mitochondria and chloroplast are critical for cell and organismal survival in additional to being the subject of numerous kinds of research such as evolutionary and physiology studies. Most genome assemblers do also assemble the organelles, but care should be taken to manually check how complete the organelle assembly is. If it is incomplete when compared to closely related species or if expected gene content is missing, then those organelle contigs should be removed from the assembly fasta file and the organelle reassembled using tools specifically for that such as GetOrganelle.
Manually Checking and Reassembling Mitochondria and Chloroplasts.