Thirty-five strains representative of the three Y. pestis biovars (11 Antiqua isolates, 12 Medievalis isolates, and 12 Orientalis isolates) isolated from 1947 to 1996 from various host species in 13 countries are presented in Table 1. Nineteen of these isolates have been previously characterized by Achtman et al. (8). Nucleic acid was extracted as previously described (9), and species identification was confirmed for all the strains by partial sequencing of the rpob gene (10).

We analyzed the complete genome sequences of Y. pestis strain CO92, biovar Orientalis (GenBank accession no. NC-003143) (4) and Y. pestis strain KIM, biovar Medievalis (GenBank accession no. NC-004088) (5), which were obtained from the Kyoto Encyclopedia of Genes and Genomes (KEGG) database (11). We used the Primer3 program (http://frodo.wi.mit.edu/cgi-bin/primer3/primer3_www.cgi) to determine the primer sequences specific for the genomic segments of interest (12). The primers flanked intergenic sequences of Y. pestis CO92 that exhibited large sequence differences with the homologous Y. pestis KIM strain sequences. We generated a list of Y. pestis CO92 intergenic sequences of 50 to 300 bp and carried out BLASTN searches to identify the homologous intergenic sequences in Y. pestis KIM strain by using the Y. pestis CO92 genes flanking the intergenic sequences as queries (13). When both genes flanking the intergenic sequence exhibited best-matches with the BLAST score >120 bits, we estimated the length of the corresponding intergenic sequence in the Y. pestis KIM strain. We then aligned the homologous intergenic sequences (<300 bp) and selected eight pairs of sequences with insertion or deletion divergence between the two Y. pestis biovars to locate the primers (Table 2). Two microliters of DNA extracted as previously described (9) were amplified in a 50-µL mixture containing 10 pmol of each primer; 200 mmol/L (each) dATP, dCTP, dGTP, dTTP (Invitrogen, Cergy-Pontoise, France); 1.5 U Taq DNA polymerase (Invitrogen); and 2.5 µL of a 50-mM solution of MgCl₂ in 1 x Taq buffer. Each polymerase chain reaction (PCR) was performed in a T3 thermocycler (Biométra, Archamps, France) under the following conditions: an initial 5 min of denaturation at 95°C was followed by 39 cycles of denaturation for 30 s at 94°C, annealing for 30 s at 60°C, and extension for 1 min at 72°C. The amplification was completed by holding the reaction mixture for 5 min at 72°C to allow complete extension of PCR products. These products were purified by using the Multiscreen PCR plate (Millipore Corp., Bedford, MA), as described by the manufacturer. Sequencing reactions were carried out with a DNA sequencing kit (Big Dye Terminator Cycle Sequencing V2.0; PE Biosystem, Courtaboeuf, France), as described by the manufacturer. Sequencing products were purified and underwent electrophoresis with the 3100 Genetic Analyzer (Applied Biosystems) and aligned by using the multisequence alignment CLUSTALX version 1.8 (13).

For phylogenetic analyses, DNA sequences were aligned by using the CLUSTALW software, version 1.81 (13). Because variations among spacer sequences were only due to deletions (see below), every deletion was considered a unique molecular event that resulted in a unique mismatch, regardless of its length. For any position in the spacer sequence, identity of nucleotide was coded "1"; a mismatch was coded "0"; and a pairwise binary matrix was constructed with the MEGA 2.1 software package (14). Distance matrices were determined by using p-distance analysis and were used to infer dendrograms by the unweighted pair group method with arithmetic mean (UPGMA), maximum parsimony, and neighbor-joining, available in the MEGA 2.1 software package (14). We also used the maximum likelihood method within the PHYLIP software package (15).

The Anthropology Laboratory of Bordeaux University 1 studied a first series of 60 skeletons discovered during excavations in 1989 in Sens, France. The skeletons were buried in four adjoining mass graves, dated by radiocarbon to be from the 5th to 6th century A.D (16). The second series came from a cemetery in Dreux, France, where nine mass graves were identified during excavations in 1990 (Figure A1). Each grave contained 2–22 skeletons. Ceramic fragments, found in the burial site, dated the graves from the 12th to the 14th century A.D (17). Archeological data (burial patterns) and anthropologic data (absence of bone fractures, indications of sex and age of persons) supported the hypothesis that the two grave sites contained remains of persons who died during an epidemic. No historical written records for the Sens and Dreux grave sites exist, but comparisons with demographic models suggest that the graves resulted from the Justinian plague (6th–8th century) and the Black Death, respectively. The Saint-Côme and Saint-Damien site in Montpellier had been used as a church cemetery outside the city walls during the 9th to 17th centuries A.D (18). Four graves have been dated as having been dug in the 13th and late 14th centuries, on the basis of their position on top of a 13th-century remblai (a small hill created by burying bodies) and the fact that they were behind a wall that dated from the second half of the 14th century. Dating of the different parts of this site was based on historical data, stratigraphy, the study of 7,059 ceramic fragments, and ¹⁴C dating.

We collected 10 teeth from three skeletons in Sens, 4 teeth from three skeletons in Dreux, and 5 teeth from two skeletons in Montpellier. The teeth were washed thoroughly with sterile phosphate-buffered saline and fractured longitudinally. Powdery remnants of dental pulp were scraped into sterile tubes for DNA extraction, as previously described (19). Seventeen teeth collected from contemporary dental patients in Marseille without any evidence of plague were used as negative control teeth ("negative teeth").

MST was applied to DNA extracted from 19 teeth from the remains of eight persons in one Justinian era burial and two graves from the time of the second pandemic, as described above. We instituted precautions to avoid bacteriologic and molecular contamination of ancient material (Figure A2). Two successive experiments were conducted. Dental pulp was recovered in a Y. pestis–free building A by two operators (operators 1 and 2), who had never worked with Y. pestis, its DNA, or amplicons. Teeth were processed individually by cleaning and opening the tooth and recovering the dental pulp into a sterile tube, which was closed securely. Each tooth was processed separately in chronologic order from Justinian to Black Death material, and negative control teeth were processed at the end. New sterile forceps were used for every tooth. Dental pulps were then transported into another laboratory, B1, 200 m away from building A, and to building C, 600 m away from building A for DNA extraction. It was performed by operator 3, who also had never worked with Y. pestis. DNA extraction for each experiment was performed with new reagents, ordered directly from the distributors by operators 1 and 3, who followed the protocol for good practices for ancient DNA manipulation (20). Extracted DNA was submitted to a second laboratory in building B2/D for suicide PCRs (18), which were conducted with one negative control for every three specimens. Finally, amplicons were transported to building D for cloning and sequencing, which was carried out by operators 4 to 6, who had never worked with Y. pestis. PCR products were cloned in PGEM-T Easy Vector (Promega, Charbonnières, France), as described by the manufacturer. Six clones were cultivated in LB medium (USB, Cleveland, OH) overnight, and plasmid purification was performed by using the Promega system. Six clones were sequenced from every amplicon by consensus primers. The sequences were compared with sequences available in GenBank by using the BLAST software version 2.2.8 to ensure accurate species identity. The sequences were further compared by using the BLAST software version 2.2.8 with our local Y. pestis spacer sequence database to ensure proper biovar identity.

All the isolates were firmly identified as belonging to Y. pestis species on the basis of their phenotypic profile and partial analysis of 16S rRNA gene and rpoB gene sequences. A total of eight intergenic spacers located throughout Y. pestis genome (Table 1) yielded 2–9 alleles per spacer, resulting in a total of 23 molecular events potentially corresponding to >8 x 10⁶ combinations (2²³ molecular events). The 387-bp spacer YP1 exhibited a 183-bp deletion specific for the Medievalis isolates. For spacer YP3, we observed five alleles: Orientalis isolates 1513 and 695 exhibited a complete, 340-bp sequence, and the other Orientalis isolates exhibited a 16-bp deletion; Medievalis isolates featured a 48-bp deletion and a mutation G → A at position 135 of the spacer; Antiqua isolates exhibited a 32-bp deletion except for isolate 611, which had a 16-bp deletion and mutations C₁₆₄ → T, C₁₆₆ → T and A₁₉₁ → G. Spacer YP3 thus differentiated every biovar. As for spacer YP4, Medievalis isolates were characterized by a 36-bp deletion, whereas Antiqua and Orientalis isolates exhibited a complete spacer. For spacer YP5, Medievalis isolates exhibited a full-length 292-bp spacer, whereas Antiqua and Orientalis isolates exhibited a 32-bp deletion at position 101 of the spacer. For spacer YP7, a total of nine alleles were found; Antiqua isolates were classified within seven alleles, Medievalis isolates within four alleles, and Orientalis isolates within five alleles. Alleles two and three were found only among Orientalis isolates. A full-length sequence of 330 bp was found for strain 549; the other isolates exhibited deletions. At position 126 in the spacer, strains 643 and 695 exhibited a 56-bp deletion; at position 133, strains 304, 1092, 507 and 571 exhibited a 49-bp deletion; at position 140, strains 611, 685, and 537 exhibited a 42-bp deletion; at position 142, strains 616, 548, 565, 518, 520, 560, 561, and 670 exhibited a 35-bp deletion; at position 149, strains 519, 542, 566, 564, and 1513 exhibited a 28-bp deletion; at position 155, strains 552, 613, 772, and 989 exhibited a 21-bp deletion; at position 162, strains 550, 677, and 1594 exhibited a 14-bp deletion; and at position 169 in the spacer, strains 544, 553, and 545 exhibited a 7-bp deletion. As for spacer YP8, Antiqua and Medievalis isolates exhibited a full-length 236-bp spacer, whereas Orientalis isolates had an 18-bp deletion at position 36 of the spacer. As for spacer YP9, Antiqua and Medievalis isolates exhibited a full-length 292-bp sequence; Orientalis isolates had an 18-bp deletion located at position 123 of the spacer. As for YP10 spacer, Y. pestis CO92 strain had a unique sequence of 369-bp; all the other isolates exhibited an 18-bp deletion located at position 177 of the spacer. Sequences herein determined were deposited in GenBank (Table A1).

Among the 35 studied Y. pestis strains, we identified three main phylogenetic clusters, each of which included only strains from a single biovar, i.e., Orientalis, Medievalis, or Antiqua, as observed in the unrooted dendrogram (Figure 1). The same topology was obtained when data were analyzed by parsimony, neighbor-joining, and maximum likelihood analyses (Figure A3). Within the Medievalis cluster, strains 519 and 616, and strains 557 and 564 were grouped by pairs; no other subgroup was identified. Within the Orientalis cluster, three groups were identified; one group included strains 989, 613, 772, and 1513; a second group was made up of three pairs of strains, i.e., 304 and CO92, 507 and 571, and 685 and 1537; the third group diverged before the differentiation of the other two groups and contained strains 695 and 643. Within the Antiqua cluster, two groups were identified; one group included strains 553, 544, 545, 550, and 677, with the last two clustered; the second group comprised strains 552, 542, and 566; strains 548 and 549 did not group with either of these two groups but diverged before the differentiation of the other Antiqua strains. The Antiqua strain 611 exhibited a unique phylogenetic position. This strain differed from all other studied Y. pestis strains and appeared to have diverged before the separation of the three main clusters.

In the 46 PCR experiments we performed on ancient tooth samples, we obtained 10 Y. pestis sequences (Figure 2); no sequences were found in the 51 PCR experiments with control teeth (p < 10^–4). The teeth from 7 of the 8 persons' remains yielded 10 specific sequences, 3 persons were positive for two molecular targets, but none of the teeth of 17 persons used as negative controls yielded specific sequences (p < 10^–4). YP1 PCR yielded an amplicon in one of six tested persons, YP8 PCR yielded an amplicon with identical sequence in six of six tested persons, and YP3 PCR yielded an amplicon in three of seven tested persons. When compared with GenBank database (Table A2), theYP1 sequence obtained in skeleton 5 yielded complete similarity with the homologous region in Y. pestis C092 strain over 390 positions and 99% sequence similarity with the homologous region in Y. pestis KIM strain over 174 positions; the YP8 sequence obtained in the six persons yielded 99% sequence similarity with homologous region in Y. pestis C092 strain over 178 positions and complete sequence identity with homologous region in Y. pestis KIM strain over 116 positions; the YP3 sequence obtained in skeletons 4 and 8 yielded complete sequence identity with that of homologous region in Y. pestis strain CO92 over 364 positions and 98% sequence similarity with homologous region in Y. pestis KIM strain over 206 positions; the YP3 sequence obtained in human remain 2 yielded a 98% similarity with homologous region in Y. pestis CO92 strain over 283 positions and a 95% similarity with homologous region in Y. pestis KIM strain over 162 positions. For each one of these 10 amplicons, further matches dropped to <90% sequence similarity over short sequences of 10 nt to 50 nt. When blasted to our local Y. pestis spacer sequence database, the YP1 spacer sequence obtained in human skeleton 5 was identical to that of the Orientalis and Antiqua reference sequences (Table A3), whereas smaller BLAST scores were obtained for the Medievalis reference sequences. Regarding the six YP8 spacer sequences obtained in skeletons 1–6, the 12 first best scores were obtained with Orientalis reference sequences. For the YP3 spacer sequence obtained in skeletons 4 and 8, the 10 first best scores were obtained with Orientalis reference sequences. For the YP3 spacer sequence obtained in skeleton 2, the 10 first best scores were obtained with Orientalis reference sequences, and this amplicon exhibited two specific nucleotide substitutions (Figure 2; Table A2). These mutations were consistently obtained in six clones.