However, AcrBP223G was different from a monomeric AcrB mutant, AcrBDloop, which we created previously. AcrBDloop was completely functionless while AcrBP223G still had a very low level of activity, which implied that at least a small portion of the protein should still exist as trimers in vivo. Subsequently, we successfully trapped trimers formed by mutant containing P223G mutation using an inter-subunit disulfide bond, C225–C777. The stability of AcrBP223G trimer was apparently much weaker than that of the WT AcrB as it dissociated upon detergent extraction and purification. It is intriguing to speculate how the PLX-4720 interaction between the protruding loop and the corresponding tunnel in the neighboring subunit is established. We probed the inter-subunit interface between neighboring subunits in AcrB using the online server of ProtorP. When AcrB trimerizes, the loop-and-tunnel interaction between neighboring subunits contributes approximately 1,600 A˚ 2 of decreased accessible surface area, which is 46.9% of the overall inter-subunit interface. Studies have shown that protein-protein interactions with interfaces larger than,1000 A˚ 2 are likely to undergo conformational changes upon binding. There are three possible scenarios: the loop adopts its final structure first while the tunnel retains a certain degree of flexibility and folds around the loop; the tunnel adopts its final structure first while the loop retains a certain degree of flexibility and folds once it settles inside the tunnel; or both loop and tunnel are flexible and induce each other to fold into the final conformation. To investigate the flexibility of the loop during trimerization, we created a reporter Cys-pair in the loop, C216–C234. We found that this pair of Cys formed disulfide bond in AcrB, and the introduction of the P223G mutation had no effect on the formation of the bond, indicating that the P223G mutation did not affect the conformation of the loop. In addition, the formation of disulfide bond between C216–C234, which greatly restricted the flexibility of the loop, had no effect on the drug efflux activity when introduced into the fully functional Cys-less AcrB background. Assuming disulfide bond forms when the subunit acquires its tertiary structure, prior to trimerization, these results would suggest that the flexibility of the loop structure is not critical for trimerization. This assumption is reasonable, as studies have shown that intra-molecular disulfide bond in proteins formed on the time scale of sec to min, comparable to the time it takes to translate a polypeptide chain with the size of AcrB . In many cases the formation of disulfide bond is actually coupled with protein folding. These results suggested that the loop remained rigid during trimerization. Another line of evidence that support the “loop-first” mechanism is the observation that the activity of AcrBP223Y is comparable or slightly higher than that of AcrBP223G. If the tunnel forms first, a loop with such a large side chain at the tip would have trouble penetrating through the tunnel.