DNA from dental pulp was extracted and converted into sequencing libraries as previously described3 (link). Potential sequencing artefacts resulting from deaminated nucleotides were eliminated by treatment of the DNA extracts with uracil-DNA-glycosylase and endonuclease VIII. DNA extracts were subsequently converted into sequencing libraries and amplified to incorporate unique sequence tags on both ends of the molecule. Two Agilent DNA capture arrays were designed for capture of the full Y. pestis chromosome (4.6 megabases), and the pCD1 (70 kb) and pMT1 (100 kb) plasmids using the modern Y. pestis strain CO92 (accession numbers NC_003143, NC_003131, NC_003134) for bait design with 3 bp tiling density. Serial array capture was performed over two copies of each array using the enriched fraction from the first round of capture as a template for a second round. The resulting products were amplified and pooled in equimolar amounts. All templates were sequenced for 76 cycles from both ends on the Illumina GAII platform, and reads merged into single fragments were included in subsequent analyses only if forward and reverse sequences overlapped by a minimum of 11 bp. Reads were mapped against the CO92 genome using the software BWA, and molecules with the same start and end coordinates were removed with the rmdup program in the samtools suite. Reference-guided sequence assembly was performed using Velvet version 1.1.03, with mapped and unmapped reads supplied in separate channels. Single-nucleotide differences were determined at a minimum of fivefold coverage and base frequency of at least 95% for both a pooled data set for all individuals and one in which all individuals were treated separately. A median network was constructed on these base calls using SplitsTree4. Phylogenetic trees were constructed using parsimony, neighbour-joining (MEGA 4.1) and Bayesian methods, and coalescence dates were determined in BEAST using both a strict and a relaxed molecular clock (Supplementary Fig. 9).