The multidrug transporter AcrB in Escherichia coli exists and functions as a homo-trimer. The assembly process of obligate membrane protein oligomers, including AcrB, remains poorly understood. In a previous study, we have shown that individual AcrB subunit is capable of folding independently, suggesting that trimerization of AcrB follows a three-stage pathway in which monomers first fold, and then assemble. Here we destabilized the AcrB trimer through mutating a single Pro (P223) in the protruding loop of AcrB, which drastically reduced the protein activity. We replaced P223 separately with five residues, including Ala, Val, Tyr, Asn, and Gly, and found that AcrB(P223G) was the least active. Detailed characterization of AcrB(P223G) revealed that the protein existed as a well-folded monomer after purification, but formed a trimer in vivo. The function of the mutant could be partly restored through strengthening the stability of the trimer using an inter-subunit disulfide bond. Our results also suggested that the protruding loop is well structured during AcrB assembly with P223 served as a "wedge" close to the tip to stabilize the AcrB trimer structure. When this wedge is disrupted, the stability of the trimer is reduced, accompanied by a decrease of drug efflux activity.