Results
In general, the topology of the obtained phylogeny was coincident with
the known phylogenetic relationship of these species (Fig 1a). Also, the
phylogenetic signal of the Crabtree effect (treated as categorical
trait) was high and significant, indicating that species resemblance
among lineages, in this trait is high (the observed number of
transitions is significantly less than what was expected by chance, Fig
1b). This is also evident after examining a heatmap including the whole
set of traits, which qualitatively suggests a high degree of resemblance
between species. This was true for physiological traits (Gly, RQ and
EthY), and contrasts with growth rate (DW), which shows a rather random
pattern of variation (Fig 2). Blomberg’s K, an index for phylogenetic
signal in continuous traits, was large and significant for fermentative
traits (Gly: K=0.82; prand = 0.010; RQ: K=1.58;
prand = 0.001; EthY: 1.60; prand =
0.001), but it was small and non-significant for DW (K=0.25;
prand = 0.959). Thus, phylogenetic signal analysis
suggests that fermentative performance shows high levels of ecological
specialization in yeasts.
The detection of evolutionary shifts using the OU-lasso method revealed
that a model where we allowed for a maximum of three shifts
(k0 =3, BIC weights all above 50% of variance
explained by the model, Table 2) better explained the results compared
to a model of random walk evolution (i.e., a Brownian motion model), an
OU model with no shifts (k0 =0) and a OU fixed
model with three shifts (Table 2; 56.5%, 85.5%, 97.1% and 60.9% of
the BIC weights, for DW, Gly, RQ and EthY respectively). In particular,
one shift was detected for DW whereas two shifts were detected for
physiological traits (Table 3), all of which are visualized in Fig 3.
Given that the algorithm associates disproportionate trait values with
nodes in the phylogeny, the identified shift for DW was located at tip
with particularly large growth rate (Kluyveromyces nonfenmertans ,
Fig 3a), which is probably due to the fact that the growth conditions
were optimal for this species.
The other identified shifts coincide with internal nodes and the WGD. In
particular, glycerol production had one shift located at the WGD and
another for a single species that is characteristic by its high
production glycerol (Naumovozyma castelli ). For EthY, there was a
shift also in the WGD and another shift with negative trait values
(i.e., values below the mean) involving the clade of Eremothecium
– Kluyveromyces (Fig 3d), which are lactose assimilating yeasts
(Nurcholis et al.).
According to Table 3, both Gly and EthY showed the largest alpha
parameter, which putatively is indicating the strength of selection
“pulling” to the optimum. Also, the sigma-squared parameter, which is
a measure of the Brownian Motion effect, is maximum for DW (suggesting
random factors explaining diversification) and lowest for EthY (Table
3). These patterns of differentiation involving WGD+ and WGD- species
can be visualized in the phenograms (Fig 4), which shows clear contrasts
between both groups, excepting DW (Fig 4a). A clear differentiation
between WGD+ and WGD- species in proxies of fermentative performance can
be observed, starting about 75 MYA(Fig 4b-d). This conclusion should be
considered just approximate, in the absence of fossil records for proper
calibration.
Interestingly, the analysis did not detect significant effects of either
the loss of respiratory complex I or the URA1 horizontal transfer as
important factors shaping phenotypic variation (see Fig 1a), which would
be a support of the idea that the WGD was the single most important
factor explaining the evolution of physiological traits in this dataset.