Whole-Genome Sequencing

Core Concepts

Sanger (dideoxy) sequencing: reading one fragment at a time

Sanger sequencing, developed by Frederick Sanger in the 1970s, exploits a chemical trick (commonly taught). A DNA polymerase copies a template strand, and the reaction mixture includes normal nucleotides plus a small amount of dideoxynucleotides (ddNTPs). A ddNTP lacks the 3′ hydroxyl group needed for the next nucleotide to attach, so when one is incorporated, chain synthesis stops right there. Because incorporation is random, the reaction produces fragments of every possible length, each ending at a known base.

Labeled ddNTPs (one fluorescent color per base) allow a machine to separate the fragments by size through a gel or capillary and read the color sequence — and therefore the DNA sequence — from shortest to longest. Sanger reads are typically a few hundred to about 1,000 base pairs (commonly taught range). The method is accurate and is still used to confirm important results, but it is far too slow for a whole genome.

Shotgun sequencing: shred, read, and reassemble

Genomes are far longer than any single read, so sequencing requires breaking the DNA into pieces and putting the information back together.

• Whole-genome shotgun: cut the entire genome into random, overlapping fragments, sequence the ends of millions of fragments, and let a computer assemble the reads by matching overlapping ends.
• Hierarchical (clone-by-clone) shotgun: first build a physical map of large cloned pieces — commonly BACs carrying roughly 150,000 base-pair inserts (commonly taught figure) — then shotgun-sequence each clone separately and use the map to order the assembled clones along the chromosome.

The Human Genome Project used the hierarchical approach; the competing private effort led by Craig Venter's company used whole-genome shotgun (commonly taught history). Shotgun sequencing is faster, but assembling repetitive DNA is its greatest challenge.

Next-generation sequencing: millions of fragments in parallel

NGS platforms (Illumina sequencing-by-synthesis is the commonly taught example) attach millions of DNA fragments to a glass surface, amplify each fragment into a local cluster of identical copies, and read all the clusters at once by detecting the fluorescent nucleotides incorporated in each cycle. A single run produces an enormous flood of short reads — commonly ~100–300 base pairs each. Reading paired ends (both ends of the same fragment) preserves information about how far apart the two reads sit in the original DNA, which helps the assembler bridge repeated sequences. This parallelization is what collapsed sequencing costs.

Coverage, assembly, and the problem of repeats

Coverage (depth) is the average number of reads that cover each base — 30× coverage means every base was read about 30 times on average. Higher coverage makes base calls more trustworthy and fills in more of the genome. Assembly proceeds in stages: overlapping reads merge into contigs, and contigs are ordered and oriented into larger scaffolds using paired-end information. Repetitive sequences are the assembler's nightmare: identical repeats make it impossible to know which copy a read came from, so assemblies frequently contain gaps. Even genomes described as "complete" often have unassembled regions (commonly acknowledged in the field).

Annotation: from letters to genes

A raw sequence is just letters. Annotation is the computational step that finds the features: genes (open reading frames, splice sites, promoters), regulatory elements, and repeats. BLAST (Basic Local Alignment Search Tool) compares a new sequence against databases of known genes to infer function from similarity — a gene that matches a known human gene across species probably does something similar. Much of the human genome is noncoding, and interpreting those regions is still an active research frontier.

Common Confusions