Water quality effects on transcriptome
A linear mixed model was used to assess percentages of transcriptional
variance accounted for by selected environmental parameters. The linear
mixed model was conducted using the R package variancePartition(Hoffman & Schadt, 2016) v.1.17.6 and included lake as a random effect
(categorical) as well as dissolved oxygen, conductivity, pH, alkalinity,
chlorophyll a, suspended phosphorus, suspended carbon, and total
dissolved nitrogen as fixed effects (continuous). Gene expression, or
the transcriptome-wide count data, was the response variable of
interest. Not all environmental parameters were included in the model
due to collinearity of certain variables. For example, total dissolved
nitrogen, NO3-N, and NH3-N are
positively correlated which may produce misleading results and
overestimated the contribution of these variables; thus, covariates were
dropped. variancePartition fit a linear mixed model that jointly
considered the contribution of all variables on the expression of each
gene in the normalized transcriptome-wide gene count matrix. Using a
multiple regression model, variancePartition assessed the effect
of each individual variable on gene expression while correcting for all
other variables included in the model (see variancePartitiondocumentation for further statistical details).