Replication of analyses in Gomes et al. (2011)
Presented in Table 1 are the results reported by Gomes et al. (2011) and
our re-analysis – both using two phylogenetic multiple regression
models. Consistent with Gomes et al. (2011), we found that both mass and
TL significantly predicted lifespan. However, the remaining results are
all substantially discordant. While Gomes et al. (2011) found no
association between telomerase activity and lifespan, we observed a
significantly positive association. In predicting body mass, Gomes et
al. (2011) found that telomerase activity was a significant predictor,
but TL was not. We found the opposite: telomerase activity did not
predict mass, but longer TL predicted decreased body mass. Using current
estimates for mass and lifespan from the AnAge database (Table S9) did
not qualitatively alter the findings described above except the
association between telomerase activity and lifespan was attenuated
(β=0.0036, p=0.0594, Table S2).
Gomes et al. (2011) reported a series of bivariate plots (their Fig. 4)
for key variables in their analyses along with p-values. These p-values
are identical to those presented in their Fig. 2 (our Table 1), strongly
suggesting that these significance values were from phylogenetic
multiple regressions, not phylogenetic bivariate regressions. A casual
reader is likely to interpret these p-values as bivariate associations.
Here, we present phylogenetic linear bivariate regression analyses of
TL, body mass, maximum lifespan, and telomerase activity to examine the
sensitivity of the multiple regression results and more clearly define
these bivariate associations.
The bivariate results (Fig. 1) are qualitatively consistent with our
multiple regression results, except lifespan did not show any
association with telomerase activity (Fig. 1d). Based on these bivariate
results, a 1% increase in body mass predicts a 0.08% decrease in TL
(Fig. 1a), while a 1% increase in lifespan predicts a 0.43% reduction
in TL (Fig. 1b). The co-evolution of TL and body mass is illustrated in
the phylogeny in Fig. 2.
In an effort to further reconstruct the reasons for our discrepant
results with those of Gomes et al. and explore the sensitivity of our
analyses to modeling strategies, we experimented with alternative
transformations of the telomerase activity measurements (see Supporting
Information). Briefly, we find that in multiple regression models, as
telomerase activity is first log-transformed (Table S3) and then binary
transformed (Table S4), TL becomes less and less predictive of either
mass or lifespan and the correlation between TL and telomerase activity
increases (Table S5). When the bivariate association of mass predicting
binary telomerase activity (0/1) is modelled in a phylogenetic logistic
regression (see Supporting Information), complete telomerase repression
is significantly associated with higher body mass with an inflection
point at 0.8 kg (Fig. S1).