Rarefaction
Rarefaction of microbial data to normalize for varying library size can lead to data loss and may be detrimental to interpretation of results (McMurdie and Holmes 2014). To ensure that patterns observed in non-rarefied data are not due to bias in library size, we rarefied all libraries to 7,000 reads. We then conducted alpha and beta diversity analyses, described below. Results from rarefied tests did not qualitatively differ from the non-rarefied data (Figure S1). In the main text, we present and discuss the results of non-rarefied data.