Type II Clusters 
Based on the Type I cluster results, we speculated that genes contained in Type II clusters might be related by function or regulation.
Type II groupings contain a combination of both known and unknown gene members and could provide preliminary clues about the function of unknown genes within a particular cluster by associating their expression with neighboring genes of known and defined function.
Alternatively, co-expression of genes results from common regulation, and Type II associations may suggest shared regulatory mechanisms for clustered genes.
We note, however, that despite the statistical framework with which groupings are assigned, experimental evidence is necessary to confirm functional or regulatory relatedness.
We do not suggest simply assigning either based on cluster membership; rather, cluster associations may provide some preliminary functional or regulatory clues for gene members.
A total of 18 (90%) of 20 Type II clusters (Table 3 and Figure S3) may not have been identified without neighbor clustering: eight (44.4%) of 18 gained additional gene members; the remaining ten comprised genes that demonstrated significant differential expression only after applying GenomeCrawler.
Only two clusters (spy0127-0130 and spy1701-1704) could have been identified without neighbor clustering; however, a number of these genes were initially annotated as hypothetical proteins, so a potential relationship between the gene members may not have been readily apparent.
The upregulated spy0127-0130 cluster is part of a larger genomic region known as FCT (for fibronectin- and collagen-binding proteins and T antigen-encoding loci), which spans spy0123-0136 in the SF370 genome and encodes surface proteins and transcriptional regulators [46].
A search of both the PFAM database [47] (http://pfam.wustl.edu) and sortase database (http://www.doe-mbi.ucla.edu/Services/Sortase) predicted that spy0129 encodes a sortase enzyme, which are transpeptidases that cleave protein substrates at conserved C-terminal motifs (often LPXTG) and then anchor these proteins to the bacterial cell wall [48,49].
Recently, it was reported that the four genes spanning spy0127-0130 encode, and are responsible for, the formation of surface-localized, trypsin-resistant pili that induce protective immunity against a lethal dose of group A streptococci in a mouse model of infection [36].
This same report provided the first experimental evidence supporting the sortase prediction, indicating that the gene product of spy0129 is responsible for the cell-wall sorting of the proteins encoded by both spy0128 (annotated as a Cpa homolog [50]) and spy0130 (annotated as a protein F homolog [14]).
Furthermore, the spy0128-encoded protein is the structural backbone of the pili, and the gene product of spy0130 may be involved in stabilizing the structure [36].
Together with the identification of this cluster by GenomeCrawler, these results prompted us to study this cluster and the contributions of the gene products to pharyngeal cell adherence.
We determined experimentally that cluster spy0127-0130 is an operon, verifying both related function and regulation of the gene members.
Reverse transcription of SF370 RNA, with primer combinations that spanned all four genes, produced cDNA fragments of sizes that could only result from a polycistronic mRNA template (Figure 3).
In silico sequence inspection identified a single putative promoter sequence upstream of spy0127 (see Table S6).
Although GenomeCrawler is not an operon-identifying algorithm, these results show that it could (1) identify this commonly regulated gene cluster and (2) define the cluster boundaries, excluding other proximate genes, such as an additional sortase-encoding gene, spy0135.
