Data analysis
Since data showed large year-to-year variation, we analyzed each year separately (PerManova: F(1.313)=4.45, P=0.009). We first tested the dispersion of chemical profiles of all trees between sampling years (2014, 2015), and then between months within a year using NMDS ordination with Bray-Curtis distance (metaMDS function, packagevegan v2.5-6). We used PERMANOVA to verify whether chemical profiles changed over time with block set as strata (adonis2function, vegan package v2.5-6 in R v4.0.3). An overview of our approach for each question is detailed below. Due to multiple permutations within our analyses, we summarized which treatments were compared to evaluate each research question in Table 1.
1. Do treatments influence NSCs of ponderosa pines? Since increment cores were collected from the same trees over multiple times, we used repeated measures analysis. We first checked the assumption of sphericity using Mauchly’s test by testing whether starch, total sugars, and total NSCs change with month as a function of treatment (anova_test function, package rstatix v0.6.0). Results showed the data met sphericity assumptions and the correlation structure was avoided in final models. Therefore, linear mixed models were used to test the interaction between treatments and months with blocks and trees as nested random effects to identify whether treatments affected NSCs (function lmer , package lme4 v1.1-23). Residuals of NSC analyses were normally distributed and had homogeneous variances. We reported Type III tests using Satterthwaite’s method and pairwise comparison p-values are Tukey adjusted (package emmeans v1.5.3).
Does trenching affect NSCs?We compared starch, total sugars and total NSCs between trenched-control (n=8) and untrenched-control (n=8) trees using the same analysis as above.
Do bark beetle attacks affect NSCs? We compared starch, total sugars and total NSCs between untrenched-attacked (n=8) and untrenched-control (n=8) trees using the same analysis as above.
Do microbial inoculations affect NSCs? We compared starch, total sugars and total NSCs between untrenched-inoculated (n=8) and untrenched-control (n=8) trees using the same analysis as above.
2. Is drought-biotic stressor interaction significant for NSCs and defense metabolites? In this and the research questions 3-5 below, we only focused on the phloem monoterpenes as there is currently no evidence that sesquiterpenes affect bark beetle biology.
We tested whether trenching impacted the effect of biotic stress (either bark beetle attacks or inoculation) on NSCs and terpenes. For each carbohydrate and defense compound, we ran separate models for bark beetle attack vs. control and inoculation vs. control treatments. Each repeated measures model included a full 3-way interaction with month, trenching and either beetle attack or inoculation treatment. Defense concentrations were log-transformed to meet assumptions of normality and homogeneity of variance but least squared means in original scale are presented.
3. Do changes in NSCs result in changes in defense metabolites?Using the same statistical approach as in the first question above, we tested whether the interaction of the NSCs, month and treatments affected concentrations of the total diterpenes or total monoterpenes with a repeated measures ANCOVA. Both monoterpenes and diterpenes were natural log-transformed to meet normality and homogeneity of variance assumptions. We removed the 3-way interaction from the final analysis because it was not significant and instead used a simpler model with fixed effects of NSCs, treatment, month, treatment by NSCs, and treatment by month with the same repeated measures model structure as above. Where there were significant interactions between treatment and NSCs, we tested whether slopes were different using Tukey’s adjusted p-value (function lstrends in package lsmeans v2.27-2).
4. Do concentrations of NSCs and terpenes vary between dying and live trees? We used the same type of analyses as described under the first question above but only used the data from 2014 since six out of eight trees attacked by bark beetles died that year. We compared six dying trees with the two live trees in the trenched-attacked treatment for concentrations of starch, total sugars, total NSCs, total diterpenes, and total monoterpenes. We conducted overall pairwise comparisons between treatments with Tukey’s adjusted p-values and then separately for each month with Sidak adjusted p-values (emmeansv1.5.3).
5. Can interactions between NSCs and terpenes explain tree mortality? We tested the correlation between average NSCs (starch, total sugars, or total NSCs) and terpenes (total diterpenes or monoterpenes) for the six dying trees in the trenched-attacked treatment in a Pearson correlation analysis. We also conducted the same correlation tests for live trees in the remaining six treatments including those that survived in the trenched-attacked treatment. We only used data from 2014 as all six dying trees in the trenched-attacked treatment died in 2014.