Microarray analysis grouped strains by phylogenetic divisions and serotype. However, a subcluster of Division I, D1b, grouped more closely to D2 than to D1a. This subcluster consisted of serotype 1/2b strains from human sporadic and environmental sources. This grouping was consistent even when data were analyzed using three different cluster algorithms (UPGMA, neighbor joining, and Ward's minimum variance), using different intensity range scores (i.e., with <0.15 scored as 1), or simply using normalized intensity data to produce the dendrogram (Ward's minimum variance). Two probes (probes 55 and 891) identified by DFA as differentiating D1b from D1a were sequenced and found to be serovar specific (differentiated serovar 1 from serovar 4) (Table ). Therefore, it likely that a combination of probe differences makes D1b appear more similar to D2 than to D1a.
D1 strains were separated into four main subclusters, with D1a containing three subclusters and D1b consisting of a single 1/2b subcluster. One of the subclusters within D1a included 15 of the 17 serotype 4b strains associated with epidemics (Fig. ). DFA was used to identify three probes that are most useful in defining this subcluster, and further analysis of these probes is under way.
Strains epidemiologically linked to particular epidemics were included in the microarray analysis to determine whether microarray subtyping did indeed group these strains together. Isolates obtained from patients and implicated foods from the 1981 Halifax epidemic (F495E and F496E), the 1994 Illinois epidemic (B507E and B508B), and the 1998 multistate epidemic (F470E, F581E, and F584E) grouped according to epidemic (Fig. ). Two of the three strains associated with the 1988 to 1990 United Kingdom epidemic also grouped together. Investigation of the later outbreak identified pÂté as the likely source of an observed upsurge in listeriosis cases; however, no samples of pÂté eaten by patients with listeriosis were available for subtyping (19
). Interestingly, the two strains from this outbreak that did cluster together were both obtained from patients, whereas strain F497E, a strain also associated with this epidemic but in a separate cluster, was a food isolate.
Strain A503E, a serotype 1/2a isolate that caused a multistate deli meat-associated epidemic in 2000, clustered with three other 1/2a strains (Fig. ). Two of these strains are particularly interesting, because one (A501N) was isolated from the same food-processing plant in 1988 as A503E and another (A502S) was from a human sporadic case associated with A501N (17
The resolutions of four different subtyping methods were compared using a subset of strains (Fig. ; Table ). Microarray analysis and PFGE subtyping showed the highest resolution, MLST had moderate subtyping resolution, and ribotyping had the lowest resolution. The microarray analysis subtyping resolution was similar to that of PFGE with two enzymes, the current gold standard for molecular subtyping of L. monocytogenes
). Nevertheless, occasionally the two techniques placed strains in different groups (Fig. ; Table ). This is not surprising, because the two techniques sample the genome differently.
Microarray analysis and subsequent DFA processing of data resulted in the identification of 22 subtype-specific probes. Thirteen of these probes were further analyzed by PCR and sequence analysis (Table ). Sequence analysis indicated that the microarray hybridization was capable of detecting approximately 10% sequence divergence between strains. These data agree with the microarray sensitivity threshold reported previously (9
), although microarray sensitivity is obviously dependent on hybridization conditions, sequence content, and signal analysis.
The 22 probes identified as important for division and subtype definition included seven probes with sequence similarity to cell wall-associated proteins (probes 119, 205, 265, 321, 553, 657, 891, and 951). Three of these were serovar or serotype specific (Table ). Five probes had sequence similarity to proteins important for survival in the environment or host (probes 57, 837, 1133, 1229, and 1263), and four probes were similar in sequence to virulence-associated proteins (probes 55, 875, 887, and 1117).
In conclusion, these data indicate that microarray analysis has a resolution similar to that of PFGE and better than those of MLST with housekeeping genes and ribotyping. Microarray analysis accurately clustered epidemiologically linked strains. Most epidemic-related strains formed a monophyletic cluster within Division I. Additionally, microarray analysis allowed identification of 22 probes that simultaneously distinguish divisions, serotypes, and subtypes.