The circos plot wraps a whole genome into a ring, so rearrangements that join distant chromosomes can be drawn as lines across the middle.
The genome browser of Part 2 draws one stretch of one chromosome. A rearrangement that joins chromosome 9 to chromosome 22 has its two ends in places such a view never shows together, and so do the blocks of sequence that mouse and human still share after roughly 75 to 90 million years of separate shuffling. Circos, published by Martin Krzywinski and colleagues in 2009, puts every chromosome on one circle, so that any locus can be joined to any other by a line across the middle. The chart it draws is called a circos plot.
Figure 1 is a simulated tumor genome on a real scaffold. The 24 chromosomes are drawn to their GRCh38 lengths with their real G-bands, and the cancer genes sit at their real positions. The mutation counts, copy number, GC content and methylation are random, and so are 28 of the 30 structural variants. The other two are planted at real loci: the BCR::ABL1 translocation t(9;22)(q34;q11), labeled in the figure, and a tandem duplication across MYC on 8q24. Copy number is drawn red for gain and blue for loss.
Figure 1. A simulated tumor genome on the real GRCh38 scaffold. Rings from the outside in: G-banded ideogram with 25-Mb ticks, cancer-gene labels and tiles, mutation count per 3-Mb bin, copy number (log2 ratio), GC content and methylation. The lines across the middle are structural variants, colored by type, each joining its two breakpoints; for a deletion, duplication or inversion the arc between the two feet is the affected segment. Event sizes are exaggerated (2 to 40 Mb) so they show at genome scale. Hover or tap a chromosome to isolate its links, or a line for its breakpoints; with the plot focused, left and right arrows step through chromosomes and up and down through the lines. On a narrow screen only the genes of the planted events are labeled.
The run of blue copy-number dots on the short arm of chromosome 17, under the TP53 label, is a third planted event: one copy of 17p is gone, and one copy of TP53 with it. That is half of the commonest way a tumor loses p53, the first route in Five Views of a Broken Pathway: a mutation disables one copy of TP53, and losing 17p removes the other. In a survey of more than ten thousand tumors, over 91% of those with a TP53 mutation had lost the second working copy too, by a deletion, a second mutation or loss of heterozygosity.
The figure folds together views a clinical genomics team would otherwise read separately: karyotype, copy-number profile, mutation map, methylation and a list of structural variants. What it shows well is which chromosomes the long-range breakpoints join; in this genome chromosomes 3, 4 and 12 each carry 2 of the 16 translocation ends, and no other chromosome carries more than 1.
Without the tracks and links, what is left is the ideogram: a ring of arcs, one per chromosome, sized by length in megabases. A full ideogram also draws cytobands, the alternating dark and light Giemsa-stained bands of cytogenetics. They are landmarks from microscopy and FISH mapping rather than exact base-pair boundaries. Figure 2 draws the GRCh38 bands at about the 850-band level, darker for more strongly stained, with the centromere in red at the boundary between the short p arm and the long q arm.
Figure 2. The bare scaffold: 24 human chromosomes (22 autosomes, X and Y) drawn to GRCh38 lengths, with G-bands and centromeres from the UCSC hg38 cytoband table. Ticks mark 25-Mb intervals, labeled every 50 Mb (every 100 Mb on a narrow screen). Hover or tap a chromosome, or focus the plot and use the arrow keys, to draw it in the center as a karyogram with its arms, centromere and labeled cancer genes.
Inside the ideogram are data tracks. Circos offers more track types than this, among them text labels, connectors and highlights, but five carry most quantitative data:
Figure 1 stacks all five. Untick one and its ring goes, and the rings inside it move outward. By convention, schematic annotations such as gene names go outside, near the ideogram, and quantitative signals inside.
The interior of the disc holds links between positions. A link joins two points, which is what a structural variant has: two breakpoints where the sequence is joined in a way the reference does not have. A ribbon joins two intervals, and the width of each end is the length of its interval. A breakpoint has no length to show, so Figure 1 draws links. Ribbons earn their width when both ends are stretches of sequence, as with the two pieces a translocation swaps (Figure 3) or a block conserved between two genomes (Figure 4).
In 1960 Peter Nowell and David Hungerford found an unusually small chromosome in the blood cells of people with chronic myeloid leukemia (CML). In 1973 Janet Rowley, using the new banding stains, confirmed that this Philadelphia chromosome was chromosome 22 missing the end of its long arm, and found that about as much extra material sat on the end of the long arm of chromosome 9. She proposed that the missing piece had moved, not been lost: a translocation rather than a deletion. That the exchange was reciprocal was shown directly only in 1982, when Annelies de Klein and colleagues found that ABL1 moves from chromosome 9 onto the Philadelphia chromosome, where it was later found joined to BCR. The BCR::ABL1 fusion makes a tyrosine kinase that is always on. Imatinib, designed to inhibit that kinase, helped turn CML into a manageable chronic disease for many patients.
Figure 3 keeps only the two chromosomes involved. Chromosome 22 gets 77° of the circle instead of the 5.4° it has in Figure 1, so both breakpoints have room for their labels, and the ribbon can show the two swapped pieces at their real lengths.
Figure 3. The t(9;22) translocation on chromosomes 9 and 22 alone (GRCh38 lengths and bands). The ribbon's ends are the two segments that change places: the end of 9q from ABL1 onward (7.6 Mb) moves to chromosome 22, and the end of 22q from BCR onward (27.6 Mb) moves to chromosome 9. The shortened chromosome 22, carrying BCR::ABL1, is the Philadelphia chromosome. Breakpoints are placed at the gene midpoints; the real ones fall within the genes.
Circos was first built for comparative genomics. Its ribbons can join conserved blocks between two genomes as well as breakpoints within one. Figure 4 takes the 160 blocks of at least 3 Mb that the human genome shares with mouse, from the UCSC alignment net between the human (GRCh38) and mouse (GRCm39) assemblies. Together they span 82% of human chromosomes 1 to 22 and X and land on all 20 mouse chromosomes. It draws them twice: as ribbons on a circle, human on the right and mouse on the left, and as a dot plot, human along the bottom and mouse down the side, each block a short line from its start to its end.
Read both halves of the circle from top to bottom. A ribbon lies flat when its block runs the same way in both genomes as drawn, and twists when the mouse copy runs backwards; in the dot plot the same block slopes down to the right or up to the right. Two ribbons cross exactly when the two halves, as drawn, put their blocks in opposite orders, which in the dot plot means one block sits above and to the right of the other. The first half of human chromosome 4 is the simple case: five blocks in order along mouse chromosome 5, drawn as five ribbons that do not cross one another and as a short staircase in the dot plot. Human chromosome 3, by contrast, is split across six mouse chromosomes, while the 5 blocks of the X all stay on mouse X, as Susumu Ohno predicted in 1967.
Figure 4. Human-mouse synteny from the UCSC hg38 netMm39 alignment net: every top-level block spanning at least 3 Mb of human sequence, colored by human chromosome (six colors, repeating in order). Each ribbon end, and each line in the dot plot, runs from the block's start to its end; a dot marks position 0 of each chromosome, so a flipped mouse chromosome has its dot at the bottom. Blocks span gaps filled by other sequence. Crossings are counted between ribbon midlines. Labels that would collide are left out; hover or tap a block in either panel, or focus the circle and use the arrow keys.
Not every crossing is evolution. The order of the mouse chromosomes and the direction each runs are drawing choices. Drawn the way most circos plots are, with both species running clockwise in numeric order, 7,529 of the 12,720 pairs of ribbons cross, and because the mouse half then runs from bottom to top, the blocks that keep their order are the ones drawn crossing. Running mouse from top to bottom brings the count to 5,191, and sorting the mouse chromosomes by where their blocks sit in human to 3,965. Turning a chromosome end over end changes only the crossings among its own blocks, and moving it changes only those with other chromosomes, so the page can search every order exactly: with human in numeric order, no layout of the mouse half has fewer than 3,605 crossings. Only 177 of them join two blocks on the same mouse chromosome, where the order inside a chromosome really differs between the species. The rest come from interleaving: mouse chromosome 1 takes blocks from human chromosomes 1, 2, 5, 6, 8 and 18, while human chromosome 1 sends blocks to mouse 4, 3, 1, 8 and 13.
A circle gives every chromosome an equal chance to connect to every other, and that is what it is good for: seeing which chromosomes are joined and how many links each carries, as in Figure 1, when there are few enough links to follow. It is poor for order and orientation along two genomes. Figure 4's dot plot shows the same blocks with nothing to untangle, and the best possible circle still has 3,605 crossings. It is poor for exact coordinates and values, too, since arc lengths and ring heights are hard to compare. For those, a genome browser or a linear genome-wide copy-number plot reads more precisely.