Supplementary Figure 1: Simulation demonstrating that calculation of N_e using a large number of sequence tags provides an accurate high-resolution estimation of N_b.

(a-b) We simulate a population with 5, 50, 500, and 10⁸ tags present in equal frequencies. Then they were passed through a bottleneck that reduced the population to 10¹-10⁶ (bottleneck population size). In case of the 5, 50, and 500 tag simulations, this was followed by a second sampling step (5x10⁵) that fits the number of sequenced barcodes. 10⁸ tags represent the ideal case where (virtually) each bacterium has a unique tag such that after passage through the simulated bottleneck each bacterium is expected to have a distinct barcode. Bottlenecks were simulated by multinomial sampling with replacement and we used equations (1) and (2) from Krimbas & Tsakas⁵ to determine N_b. The results of 1000 independent simulations are shown. To illustrate the relative deviation from the theoretically expected N_b, data are normalized to the simulated bottleneck and the red, dotted line indicates the theoretically expected N_b (shown at 100 %). (a) In this box plot the median (black line), interquartile range (box), and 95 % confidence interval (whiskers) are indicated. (b) The scale is changed so that outliers (black squares) can be visualized. Note that the median of 1000 independent simulations accurately predicts N_b even with only 5 tags; however, the wide distribution of data-points make N_b estimations from this few tags inaccurate or impossible (negative values) with small numbers of experiments (i.e., few animal infections).