Abstract Co-amoxiclav resistance in E. coli is a clinically important phenotype associated with increased mortality. The class A beta-lactamase bla TEM-1 is often carried by co-amoxiclav-resistant pathogens, but exhibits high phenotypic heterogeneity, making genotype-phenotype predictions challenging. We present a curated dataset of n =377 E. coli isolates representing all 8 known phylogroups, where the only acquired beta-lactamase is bla TEM-1 . For all isolates, we generate hybrid assemblies and co-amoxiclav MICs, and for a subset ( n =67/377), bla TEM-1 qPCR expression data. First, we test whether certain E. coli lineages are intrinsically better or worse at expressing bla TEM-1 , for example, due to lineage differences in regulatory systems, which are challenging to directly quantify. Using genotypic features of the isolates ( bla TEM-1 promoter variants and copy number), we develop a hierarchical Bayesian model for bla TEM-1 expression that controls for phylogeny. We establish that bla TEM-1 expression intrinsically varies across the phylogeny, with some lineages (e.g. phylogroups B1 and C, ST12) better at expression than others (e.g. phylogroups E and F, ST372). Next, we test whether phylogenetic variation in expression influences the resistance of the isolates. With a second model, we use genotypic features ( bla TEM-1 promoter variants, copy number, duplications; ampC promoter variants; efflux pump AcrF presence) to predict isolate MIC, again controlling for phylogeny. Lastly, we use a third model to demonstrate that the phylogenetic influence on bla TEM-1 expression causally drives the variation in co-amoxiclav MIC. This underscores the importance of incorporating phylogeny into genotype-phenotype predictions, and the study of resistance more generally.