To identify lineages where lifespan has increased, we reconstructed the longevity quotient (LQ), i.e. the ratio of observed to predicted longevity, with squared-change parsimony on a molecular phylogeny of 67 bats [28 (link)] using Mesquite v. 3.6 [30 ]. We obtained longevity and body mass records from AnAge, build 14 [31 (link)]. Given that maximum lifespan is an order statistic that is expected to increase by diminishing amounts as sample size increases [32 (link)] we used only acceptable quality records with medium or large sample sizes to minimize sample size bias [33 (link)]. We predicted bat longevity by least-squares regression, i.e. log10(longevity) = 0.5609 + log10(body mass) × 0.1868, from 804 non-flying placental mammals.
To identify variables that explain variation in bat longevity, we evaluated alternative multivariate models [34 ] that corrected for common ancestry using PGLS, as implemented in CAPER [29 ]. Because LQ correlates with both longevity and body mass, we included log10(body mass) as a variable for potentially predicting log10(longevity). We used the absolute value of the median latitude (hereafter ‘latitude') of the species' range as a proxy for annual temperature and hibernation duration because in rodents hibernation duration increases linearly with mean annual temperature [20 (link)]. We added an interaction between hibernation and latitude to allow for the possibility that latitude may not affect longevity in non-hibernators. Additional variables included cave use (yes/no), diet type (animal/plant material), number of offspring produced per year and log10(breeding aggregation size). We used sexual dimorphism in total body length (TL), as measured by log2 (male-TL/female-TL), to determine if sexual selection on body size contributes to variation in longevity. Trait values were obtained from the literature, museum collections (see the electronic supplementary material), personal observation or personal communication. Because bat longevity records come from either captive or wild animals, we included data source in the models to insure it would not bias results.
We measured the relative importance of the phylogeny in predicting each trait by calculating Pagel's λ for continuous variables and D for binary variables [35 (link)] and then fitted all possible models using PGLS [36 (link)]. We rank-ordered models by the corrected Akaike information criterion (AICc) and calculated Akaike weights to determine model strength. We used models within 4 AICc of the best-fitting model for model averaging and estimated weighted coefficients, confidence intervals and relative importance for each variable [34 ,37 (link)]. Because the interaction between hibernation and latitude had significant influence, to interpret effects of the remaining variables we split the data by hibernator/non-hibernator, and then repeated the analyses described above.
To identify variables that explain variation in bat longevity, we evaluated alternative multivariate models [34 ] that corrected for common ancestry using PGLS, as implemented in CAPER [29 ]. Because LQ correlates with both longevity and body mass, we included log10(body mass) as a variable for potentially predicting log10(longevity). We used the absolute value of the median latitude (hereafter ‘latitude') of the species' range as a proxy for annual temperature and hibernation duration because in rodents hibernation duration increases linearly with mean annual temperature [20 (link)]. We added an interaction between hibernation and latitude to allow for the possibility that latitude may not affect longevity in non-hibernators. Additional variables included cave use (yes/no), diet type (animal/plant material), number of offspring produced per year and log10(breeding aggregation size). We used sexual dimorphism in total body length (TL), as measured by log2 (male-TL/female-TL), to determine if sexual selection on body size contributes to variation in longevity. Trait values were obtained from the literature, museum collections (see the electronic supplementary material), personal observation or personal communication. Because bat longevity records come from either captive or wild animals, we included data source in the models to insure it would not bias results.
We measured the relative importance of the phylogeny in predicting each trait by calculating Pagel's λ for continuous variables and D for binary variables [35 (link)] and then fitted all possible models using PGLS [36 (link)]. We rank-ordered models by the corrected Akaike information criterion (AICc) and calculated Akaike weights to determine model strength. We used models within 4 AICc of the best-fitting model for model averaging and estimated weighted coefficients, confidence intervals and relative importance for each variable [34 ,37 (link)]. Because the interaction between hibernation and latitude had significant influence, to interpret effects of the remaining variables we split the data by hibernator/non-hibernator, and then repeated the analyses described above.
Free full text: Click here
