Extended Data Figure 6: Illustration of differences between RMP and QMP methodology.

Two samples, each containing four genera, are analysed (numbers are illustrative). Genus abundance distributions in sample A and B are markedly distinct, with the microbial load in sample B more than double that of the load in sample A. Genus ‘purple’ carries two copies of the 16S rRNA gene. (1) DNA extraction and library preparation. Neither RMP nor QMP correct for biases introduced by DNA extraction, primer specificity, PCR amplification, or other library preparation steps. The resulting sequencing depth is independent of microbial load. (2) By rarefying to an even number of reads per sample, RMP assumes similar genus abundance distributions in samples A and B: sample A is therefore sequenced far more intensively than sample B. The resulting profiles therefore poorly reflect the genus distribution in the original samples. Given the multiple copies of the 16S rRNA gene, the relative abundance of ‘purple’ is overestimated. (3) The first step of QMP corrects for 16S rRNA copy number variation. In the resulting copy-corrected profile (CCP), each read corresponds with a single bacterium sequenced. (4) By dividing the CCP reads total (R) by the microbial loads (X), sampling depth is estimated for each sample. For sample A and B, sampling depth is [R]_A divided by [X]_A and [R]_B divided by [X]_B, respectively. The sampling depth for B is the lowest (3.33%) of the two; sample A is rarefied to the same level. This implies that ‘orange’ is no longer detected. As ‘orange’ was equally abundant in A and B, the fact that it is included in sample A RMP can be considered an artefact of uneven sampling intensity. The resulting rarefied genus abundances are proportional with sample microbial loads and can be extrapolated to generate QMPs expressed as number of cells per gram.