776 The Journal of Clinical Investigation| April 2001| Volume 107| Number 7 nized with heat-killed bacteria. The exact structure of many of the carbohydrate antigens is not known and may be difficult to discern, because epitopes found on helical polysaccharides may not be linear (6). Almost all human infections (> 95%) are caused by Salmonella that are in groups (defined by O antigen structures) A, B, C, D, and E. Interestingly, groups A, B, and D have the same common trisaccharide backbone structure of their LPS: D mannose α1→ 4 L rhamnose β1→ 3 D galactose α1→ 2. The only difference between the three LPS structures is the di-deoxyhexose that is attached to the 3 position of the mannose. In group A the di-deoxyhexose is paratose, in B it is abequose, and in D it is tyvelose. In group E Salmonella, the backbone structure is made up of the same trisaccharide, but there is no di-deoxyhexose. The group C LPS structure is made up of a linear polymer of four mannose residues and one N-acetyl glucosamine residue. Although a few serotypes of Salmonella have an outer capsular layer (notably S. typhi), in most serotypes LPS forms the layer around the bacterium that protects it from the environment. LPS interacts with both antibodies and with complement. Rough Salmonella mutants, which lack this protective layer, activate complement by the alternative pathway, leading to killing of the bacteria by the membrane attack complex of complement. Not surprisingly, rough mutants are avirulent. Naturally occurring infections are caused by Salmonella with a complete LPS, and they are resistant to killing by complement (7). However, activated complement protein C3 covalently binds to the terminal sugars on the LPS and acts as an opsonin. In the case of groups A, B, and D Salmonella, these terminal sugars are di-deoxyhexoses, which therefore modulate the interaction with complement (C3)(8). Changing the structure of the di-deoxyhexose from tyvelose to abequose increases the amount of C3 that binds to the bacteria, which in turn increases phagocytosis by macrophages (9). This change also affects the virulence of these bacteria in mice (10), which is due largely to differences in phagocytosis and killing by polymorphonuclear leukocytes (J. Fierer, unpublished observations). It is clearly not true that group B salmonellae are more virulent in humans, as S. typhi is a group D Salmonella. Human complement is much more active than mouse complement, which may explain why C3 activation and binding is a limiting step in mice but not humans. Much of the antibody response to Salmonella infection is directed against the LPS, and especially against the dideoxyhexoses, if they are present. The terminal sugar fits into the antigen binding site of the antibody, where hydrogen bonds form between aromatic amino acids and the OH groups on the di-deoxyhexose (11). Although the relative importance of antibodies and T lymphocytes in suppressing Salmonella infections is controversial (12), there is ample evidence that antibodies protect both humans and mice against Salmonella (13, 14). Furthermore, in most experimental systems, immunity to Salmonella is O antigen–specific (15, 16). Thus, it might benefit the bacteria to change the structure of their LPS, which is indeed observed, although not during the course of a single infection. O-antigen variation between isolates can occur because of lysogenic conversion by bacteriophages or mutations in chromosomal genes. Most of the variants are O-acetylated or glucosylated sugars, which can change the LPS structure enough to create new antigens. For instance, OafA is a chromosomal enzyme that O-acetylates …