One locus, a thousand wild plants, ordered by the climate each came from.
Every plant here was collected somewhere wild — a Spanish hillside, a
Swedish verge, a Kyrgyz mountainside — then sequenced, and grown side by
side under one set of conditions. Coldframe lines those plants up by the
climate of the place each came from, takes one gene at a time, and asks
whether the DNA carries any trace of it. It is a browser for
hypotheses, not a detector of adaptation: what it shows is where an
allele sits and what travels with it, which is a reason to ask a
question, not an answer to one.
How to read this
Pick a gene. Nine of them, each with a note on what
it does. They cover cold tolerance, light sensing, and the flowering
decision both of those feed into. There is a brief delay the first
time a gene is selected due to data retrieval.
Pick something to order the plants by. Ten measures
of the places they came from — temperature, sunlight, growing season,
elevation. The list is sorted by which one produces the strongest
gradient at the gene you are looking at, so the top entry is usually
the interesting one.
Choose what puts a site first. Every site the
statistic can score is shown — 39 at CO, 178 at
CBF, 340 at CMT2 — so this changes the order and
the measure, not what exists. Climate gradient scores each
site by how strongly it tracks the measure above; expression
effect by how strongly it tracks how much of the gene is made.
The two disagree sharply: at FRI the strongest expression
effect is only the 86th strongest climate gradient. The slider
narrows the set if the full width is too dense.
Read the panel downward. It opens with
every variable site in the region at its true position,
each tick as tall as the environmental difference between the plants
carrying it and everyone else. Dimmed ticks are the ones too rare to
correlate — still there, still clickable. A dashed bracket marks the
slice the rest of the panel is showing. Below that, every scorable
site gets a bar; then the plants grouped into 32 bands running
coldest to warmest, shaded by how common the alternate allele is in
each band; then one thin row per plant. Warm means the alternate
allele gets commoner as the measure rises, cool the reverse. Wide
regions scroll sideways.
Click anything, then use the arrow keys. Click any
column, or any tick in the all-sites strip: the map jumps to that
site and colours every plant by what it carries there, and the chart
beneath splits expression by genotype. Left and right then walk the
cursor across sites; up and down walk it through the plants or the
climate bands — whichever you clicked last, coldest at the top
either way. A card on each plot reports the cell at its crossing:
this band at this site, this plant at this site. The two strips at
the top still answer to the mouse.
Rare does not mean unimportant. Most variation in a
region is carried by a handful of plants, and a correlation over a
handful is noise — so 85% of sites here get no correlation. That is a
limit of the statistic, not a verdict on the site, and rare often
means young, which is where recent adaptation lives. Those
sites are measured a different way instead: how far the environment
of the carriers sits from everyone else's, in standard deviations,
which works down to a single plant. The counts beside the gene name
give the whole funnel — drawn, testable, variable. One thing does
stay hidden: the environmental measures are a chosen ten out of
AraCLIM's 212, so "strongest gradient" always means strongest among
those ten.
Trust the solid bars, not the pale ones. Each site
draws two. Pale is the raw correlation; solid is the same correlation
measured within ancestry groups. Related plants tend to grow
in similar places, so a lot of any raw gradient is family resemblance
rather than adaptation. The solid bar is what survives that test — at
PHYB, almost nothing does.