The tidy story
Across the thirteen cores, the archaeal share of the DNA falls where the
nitrogen-fixing rhizobia rise. The correlation is strong — Spearman's
ρ = ,
ranking the cores rather than trusting their raw values — and the
two-group split between legume and non-legume patches is itself significant
(p = ).
Read forward, that is a mechanism: fixed nitrogen reaches the soil, the
archaea have less ammonia to oxidise, and their numbers fall. It is the kind
of result a paper is built around.
So ask the cheap question first. Would a number that strong turn up even if
the archaea were not answering to the legumes at all? Here it would, and the
rest of this page is why.
The result is real, but it is not specific
A correlation is only evidence for the story told about it if it is
specific to the players in that story. So the fixers' correlation
with the archaea is ranked against the same correlation computed for every
other reasonably abundant genus — all
that clear a small
abundance and prevalence floor. If the nitrogen mechanism were doing the
work, the fixers should sit at the top of that ranking.
They do not. The nitrogen-fixers rank
, and
unrelated genera — groups with no known part in nitrogen fixation —
track the archaea at least as closely, past a threshold of
ρ = .
A story that singles out the legumes cannot explain why so many groups that
have nothing to do with them follow the archaea just as faithfully.
Two duller explanations are ruled out before drawing that conclusion. The
first is compositional closure: in a fixed total of reads, a slice as large
and variable as the archaeal share
(% to
% here) pushes almost
everything else the other way by arithmetic alone. Re-run on centred
log-ratio abundances, which are free of that constraint, the association
survives
(ρ = ,
p = ), with
genera still
past the line. Closure did not manufacture it, so the association is
genuine. What fails is its specificity.
What the archaea are actually tracking
That negative result points at something. If dozens of
unrelated genera all follow the archaeal share, the simplest reason is that
the whole community varies along one dominant direction and the archaeal
share is simply a position along it. That is directly checkable. The leading
axis of the ordination — the same projection the home page plots —
accounts for % of the
variation between cores, and the archaeal share tracks it tightly:
ρ = ,
p = .
So the clover correlation and the fescue correlation and a dozen others are
not a dozen separate facts about a dozen plants. They are all the same fact:
the community sorts along a single axis, the plants sit at different places
on it, and the archaea rise and fall with the axis. The
legume signal is real because the legumes really do sit at one end — but
the mechanism the number seemed to name is not established, because the same
number falls out of the community's overall structure without any nitrogen
story at all.
And it is not a simple growth-strategy gradient either
The obvious next question is what that dominant axis is. One
standing candidate is a life-history gradient: soils are often described as
running from slow, resource-poor communities of oligotrophs to fast,
resource-rich communities of copiotrophs, and fast growers tend to carry more
copies of the 16S gene per genome.
If the axis were that gradient, community-weighted copy number should climb
along it.
Each core's community-weighted mean copy number was computed from
,
a published per-taxon database, covering
% of reads. It ranges
from to
copies per genome across the
thirteen cores, so there was real variation for the check to find.
It found very little. Copy number does not differ between legume and non-legume
cores
(p = ),
shows no clean gradient along the dominant axis, and barely tracks the
archaeal share
(ρ = ).
The same test rules out the dull artifact it was first run to catch —
the archaea-to-bacteria ratio is not an effect of copy number — but it
does not name the axis. Whatever organises this community, a plain
copiotroph-to-oligotroph reading is not it.
Why this is worth publishing
The mistake this result guards against is a common one. Small studies that
compare microbial communities
across a handful of conditions — five plants, three treatments, two
soils — routinely read a guild's correlation with one condition as
evidence for a mechanism tied to that condition. This dataset shows how that
can go wrong even when every individual number is sound: the correlation is
strong, it is not an artifact of closure, and it is still not specific,
because the guild is riding a single community axis that many things move
along together.
So the finding is narrower than the tidy version, and it holds up better.
There is a real, dominant axis of community structure in this
pasture; the plants sit at different points on it; and the striking archaeal
numbers are a readout of that axis rather than a demonstrated response to
legume nitrogen. Naming what the axis is — which needs the
measurements set out under ground truth —
is the open question this leaves behind.