The outer layer of the cell envelope of Gram negative bacteria is the outer membrane, and the outer leaflet of that membrane is composed largely of lipopolysaccharide (LPS). The O antigen is the name given to the outer glycan portion of the LPS - there is a nice diagram at the Wikipedia article I have linked to which makes this clear.
Because this structure is exposed at the bacterial surface it is a target for the immune response, and the serotyping of many bacterial species incorporates this O serotype information.
Table 1 of the review that you link to indicates that PAI SHI-O carries genes gtrA,B,V (originally derived from a bacteriophage - not unusual for a pathogenicity island). These genes encode enzymes which modify the O-antigen via a glucosylation reaction. Presumably this helps the bacteria adapt to an immune response via a shift in serotype. It may be, for example, that an invading population of Shigella includes variants which do or do not express these genes so that a subpopulation can arise which is not recognised by an existing immune response. This is what is meant in Table 1 of the article by "evasion of host immune response". It is a common strategy - examples in Salmonella (flagellar phase variation) and Helicobacter (induced transformation) spring to mind.