2.9 | Acclimation analyses
To evaluate acclimation of photosynthetic parameters we compared
posterior distributions of TOpt of P400,
P800, VCMax, and JMax,
and of the activation energies of treatment and control plants. Stomatal
acclimation was assessed by comparing TOpt of
P400 and P270 (and P800and P505), and by analyzing changes in the
Ci/ Ca ratio. We further compared the
relationship between gs and VPD between treatments (see
below). Acclimation of respiration was determined with the set
temperature method and the homeostasis method (Atkin, Bruhn & Tjoelker
2005; Slot & Kitajima 2015): The set temperature method compares
respiration (R) of control and treatment leaves at a set temperature
(here, 30°C):
\(\text{Acclim}_{\text{SetTemp}}=\frac{R_{\text{Control\ }}at\ 30C}{R_{\text{Treatment}}at\ 30C}\)Eqn.
AcclimSetTemp > 1.0 indicates thermal
acclimation. The homeostasis method determines the degree of
homeostasis:
\(\text{Acclim}_{\text{Homeo}}=\frac{R_{\text{\ Control}}\text{\ at\ }T_{\text{Control}}}{R_{\text{Treatment}}\text{\ at\ }T_{\text{Treatment}}}\)Eqn. 1
When AcclimHomeo ≈ 1.0 respiration rates are homeostatic
across conditions and respiration has fully acclimated; when
AcclimHomeo < 1.0, acclimation, if any, is
imperfect.
2.10 | Curve fitting and statistical analyses
We fit Eqns 2, 3 and 6 using a Bayesian framework with the MCMC sampler
Stan using the R libraries ’rstan’ (Stan Development team 2018) and
‘brms’ (Bürkner 2018) with R version 3.5, which facilitated a more
thorough exploration of the uncertainty of the parameter estimates. The
full models used for Eqns. 2, 3 and 6 contained terms for all
treatments; control parameters were estimated as the basis, with
parameters for treatment plants, plants transferred from control to
treatment conditions, and from treatment to control conditions being
estimated as deviations from the controls (see Table S1). The code used
to fit these models and to generate Figs. 1−3 and 7 is available at
https://github.com/sw-rifai/Tabebuia_rosea_thermal_co2_acclim.
Informed priors (Table S2) were used to constrain kOpt,
POpt, Ha, and TOpt.
These priors were based on literature (e.g. Medlyn et al., 2002; Slot &
Winter 2017b), selected to be realistic (i.e. only positive values), and
were refined to avoid multimodal posterior fits. Attempts at estimation
of the Hd parameter produced bi-modal posterior
distributions of other model parameters (e.g. TOpt) and
so Hd was fixed at 200 kJ mol–1following the example of Medlyn et al. (2002). Each model was fit using
four chains with 2000 iterations during warmup, and a subsequent 4000
iterations during sampling. ΔS was calculated from TOptand Ha estimates for each iteration with Eqn 5. Models
were checked to ensure convergence (R̂~ 1), posterior
distributions were unimodal, and that posterior predictive checks could
approximate the distribution of the response variable. Comparisons of
plant-level parameters between control, treatment, and transfer effect
plants were made by comparing the distribution of credible intervals
between groups.
Treatment effects on the temperature response of the
JMax to VCMax ratio, and the VPD
response of stomatal conductance were determined with ANCOVA, with
temperature and VPD as the respective covariate. To visualize treatment
effects on the VPD response of stomatal conductance and the temperature
responses of stomatal conductance and the initial A-Cislopes used to assess Rubisco activase status, temperature responses
were fitted with generalized additive models using cubic regression
splines fit with restricted maximum likelihood in the ‘mgcv’ package
(Wood 2017) for R, and 95% confidence intervals were approximated by
plotting curves ± 2 standard errors.