CpG is rarer than it should be
CpG means a cytosine followed by a guanine along one strand (the p is the phosphate between them). In a random sequence with 40% GC, CpG should appear once in every 25 dinucleotides, but in most of the human genome it is found about five times less often. Methylated cytosines in CpG mutate to thymine over evolutionary time, so CpG has been slowly lost everywhere except in regions that stayed unmethylated.
What a CpG island is
A CpG island is a stretch where CpG is as common as chance predicts. The usual criteria, from Gardiner-Garden and Frommer (1987):
- at least 200 bp long,
- GC content above 50%,
- observed-to-expected CpG ratio above 0.6, where expected = (number of C × number of G) ÷ length.
Stricter criteria (Takai and Jones, 2002: 500 bp, 55% GC, ratio 0.65) exclude Alu repeats, which otherwise pass. About 60% of human genes, including nearly all housekeeping genes, have a CpG island at their promoter.
Why they matter
Promoter CpG islands are normally unmethylated and the gene is open for transcription. When an island becomes methylated the gene is silenced, which is how many tumour suppressors are switched off in cancer and how imprinting and X inactivation work. Finding islands in a sequence is the first step in picking regions for bisulfite sequencing or methylation-specific PCR.
Scanning a sequence
- Paste the sequence into the CpG Island Finder.
- Set the window (200 bp is standard), the minimum GC % and the minimum observed/expected ratio.
- The tool slides the window along the sequence and reports every region that meets the thresholds, with its position, length, GC % and ratio.
Windows that pass and overlap are merged into one island. Check borderline results with the stricter settings before relying on them.
Related tools
DNA Stats gives the GC content and dinucleotide counts of a whole sequence, and DNA Pattern Find lists every CG position.