Amino acid predictors of HIV-1 coreceptor use in different subtypes and their application to antiretroviral treatment

Abstract

The human immunodeficiency virus type 1 (HIV-1) group M is responsible for the HIV pandemic and has great genetic variability. The main subtypes and circulating recombinant forms belonging to this group differ in their worldwide distribution, their disease progression patterns, as well as in their coreceptor use phenotype distributions. The coreceptor use phenotypes of HIV-1 are based on which coreceptor the virus uses for cell entry, with the two main coreceptors being CCR5 and CXCR4, corresponding accordingly to the R5 and X4 phenotype. The R5 phenotype is found early in the infection while the X4 emerges over time and is associated with disease progression. The main determinant of coreceptor use in HIV-1 is the third variable region (V3) of the HIV-1 envelope glycoprotein 120, which is the docking surface protein that attaches itself to the primary CD4 receptor and a coreceptor during cell entry.

In papers I and II, observations were made regarding the role of the V3 glycan and V3 charge in coreceptor use, based on sequences belonging to different HIV-1 subtypes retrieved from the Los Alamos HIV Sequence database, which contains sequences submitted from infected individuals all over the world. The V3 glycan was shown in paper I to be strongly associated with CCR5 use while a net high charge acquired from different positions in the V3 was shown in paper II to be important for CXCR4 use. As a result, a model adjustable for different subtypes was created, referred to as the glycan- charge model, for distinguishing between the coreceptor use phenotypes based on their biological properties that can be deduced from the V3 amino acid sequence.

CCR5 inhibitors are a class of antiretroviral drugs that target the CCR5 coreceptor, thereby blocking the entry of R5 viruses into cells. However, prior to their administration it is important to verify that a patient does not harbour CXCR4-using variants, which could otherwise be selected for. Biological methods of coreceptor use determination are expensive and time-consuming. Hence, coreceptor use prediction algorithms, which can predict the coreceptor use from HIV-1 V3 sequences, could help to make CCR5 inhibitors more universally accessible, but their prediction accuracy needs to be improved.

Infections with HIV-1 subtype C, which is the dominating subtype worldwide and in sub-Saharan Africa, are usually associated with low CXCR4 use, but several studies have found an increased CXCR4 use among treatment failure patients. To investigate this further, 24 treatment failure patients infected with subtype C in Botswana were in paper III compared with 26 treatment-naïve patients with regard to coreceptor use, which was determined using the coreceptor use prediction algorithm Geno2pheno with a false positive cut-off rate of 10 % as well as the glycan-charge model on population sequences. Increased CXCR4 use was found in the treatment-experienced group, suggesting that treatment with the only CCR5 inhibitor in clinical use to date, maraviroc, would be less suitable in this group, which is of special significance since maraviroc is mainly used as a salvage therapy drug.

Finally, in paper IV all currently available coreceptor use prediction algorithms, including the glycan-charge model algorithm, were evaluated from a CCR5 inhibitor treatment perspective in a uniquely suited testing material, which consisted of V3 sequences of the major HIV-1 subtypes retrieved from the Los Alamos HIV Sequence database. A rigorous scrutiny of the original source articles was performed to verify that the reported coreceptor use was determined biologically. The results showed that learning algorithms were found to perform well in all studied subtypes, along with subtype-specific complex rule algorithms.

In summary, papers I and II elucidated the biological properties of coreceptor use in different subtypes, which could be determined using the V3 amino acid sequence, paper III applied this knowledge to help investigate the increased CXCR4 use in subtype C infected treatment failure patients, while paper IV compared all current coreceptor use prediction algorithms, including the glycan-charge model based on the observations in papers I and II, and applied in paper III, from a CCR5 inhibitor treatment perspective.