Genotypic-phenotypic correlations performed during early drug development
Drug resistance mutations have traditionally been identified during the pre-clinical and initial clinical evaluation of new antiretroviral drugs. Such evaluations involve culturing a wild-type HIV isolate in the presence of increasing concentrations of the drug being studied, and subsequently identifying mutations that allow the virus to continue to replicate. Site-directed mutagenesis experiments are then conducted to confirm that mutations arising during passage experiments will confer drug resistance when introduced directly into wild-type virus. Likewise, viruses cultured from patients during phase I or dose-finding studies of a new drug are sequenced, tested for resistance, and confirmed by site-directed mutagenesis.

Drug resistance mutations identified by this process acquire widespread acceptance as the predominant mutations responsible for resistance to the drug under evaluation, and are referred to as "canonical" resistance mutations. While the process of characterizing resistance mutations by susceptibility testing and site-directed mutagenesis is the most rigorous means of demonstrating that a particular mutation confers drug resistance, there are several limitations to this approach. First, these evaluations identify mutations that are sufficient to cause resistance to the drug under evaluation, but do not prove that they are in fact necessary for the development of drug resistance. Second, such data are generally derived from laboratory isolates that typically contain only one or two drug resistance mutations and not the more complicated patterns of mutations observed in clinical isolates from patients receiving combination drug therapy.

Genotypic-phenotypic correlations performed on clinical HIV-1 isolates
Clinical isolates are highly variable. For example, the protease and RT sequences of clinical isolates generally differ from the consensus B reference sequence at many positions. A portion of these differences may be canonical mutations, but many are uncharacterized. Some of these uncharacterized mutations may be polymorphisms—which commonly occur in isolates from untreated patients—while others may be associated with drug resistance.

The advantage of studying clinical isolates—as opposed to laboratory isolates—is that gene sequences reflect those encountered by physicians in clinical practice. However, the complexity of the sequences prevents direct correlation between resistance phenotypes and individual mutations, and also rules out site-directed mutagenesis experiments, which would prove too costly and time consuming to perform for each possible combination of mutations in every isolate.

Despite these difficulties, the systematic study of genotypic and phenotypic data derived from clinical isolates obtained from large numbers of patients can often reveal previously unsuspected patterns of mutations. Comparing these mutation patterns will help elucidate how known and uncharacterized drug resistance mutations interact with one another in vivo to contribute to drug resistance. The potential importance of genotypic-phenotypic correlations on clinical isolates is underscored by the fact that at least one company (Virco, Mechelen, Belgium) has created a proprietary database linking genotypic data to phenotypic drug susceptibility results [3].

Correlations between HIV-1 genotype and treatment history
Correlations between certain drug treatments and their corresponding mutations in HIV strains isolated from patients receiving the treatment provide additional knowledge about drug resistance. First, such correlations show which mutations and which mutation patterns occur in patients (rather than just in vitro). Second, such correlations are essential for elucidating the genetic mechanisms of resistance to drugs that are difficult to test in vitro. This is the case for the nucleoside RT inhibitors ddI, ddC, and d4T. For example, the canonical d4T-resistance mutation is V75T. However, this mutation is extremely rare clinically. In contrast, patients receiving d4T commonly exhibit the classical AZT-resistance mutations. This has provided a strong clue to the mechanism of d4T resistance even though phenotypic assays using that drug have difficulty detecting drug resistance.

Correlation between genotype and clinical outcome
Evidence suggesting the clinical utility of drug resistance testing has come from both retrospective and prospective intervention-based studies. The retrospective studies have been invaluable in allowing physicians to infer how patients with a particular mutation will respond to a new anti-HIV treatment regimen. Recent examples include two studies from Stanford University Medical Center that have examined the genotypic correlates of response to salvage therapy with ritonavir/saquinavir- and efavirenz-containing regimens [4,5].

Clinical outcomes may also be used to standardize data generated by phenotypic drug resistance assays. For example, clinical data shows that M184V generally causes about two-fold resistance to abacavir (ABC). However, Virco uses a susceptibility cutoff of 4-fold and will therefore report most isolates with this mutation alone as being susceptible to ABC. ViroLogic uses a susceptibility cutoff of 2.5-fold and will therefore report such isolates as being partly resistant to ABC. Such discrepancies in reporting make it difficult for physicians to determine whether patients harboring isolates with M184V will respond well to abacavir treatment [6]. In this case, the clinical data should then be used to complement the phenotypic data or to help determine the cutoffs.

In conclusion, there is much more to know about drug resistance than the canonical mutations identified during pre-clinical testing of a new drug. This added knowledge is extremely clinically important. However, it is too much to ask for physicians to keep track of the details in this area. It is not, however, unreasonable for reference labs doing genotypic and phenotypic assays to keep track of this information and to incorporate it into the interpretation of the genotypes that they run. As part of my research, I have been attempting to represent HIV drug resistance knowledge in a queryable online database [7]. The database currently contains the first three types of correlations. The schema of the database is being expanded to include the fourth type of correlation.