8th Annual Retrovirus Conference
8th CROI: Review of HIV Resistance and Resistance Testing
Andrew Zolopa, MD, Stanford University, Feb. 12, 2001
The Virtual Phenotype predicts virologic
Brendan Larder from Virco, UK presented a study that evaluated the predictive capacity of the Virtual Phenotype (vPT). Recall that the vPT is simply a probabilistic estimate of the real phenotype that is derived from a patient's genotype. Virco maintains a database with thousands of clinical isolates in which both the genotype and PT are determined. The vPT is determined by taking a patient's genotype and matching it with genotypes in the Virco database. The phenotypes that correspond to those matching genotypes are summed to give an average PT (average fold change compared to reference strain) for each of the drugs. The group at Virco has previously shown that the vPT estimates the real PT quite well (correlations of 85-90%). The question that remained about the vPT is how well did it predict response to therapy. Dr. Larder evaluated patients from the VIRA 3001 trial in which response to therapy was measured. Using the DAP analysis he showed that the vPT was an independent predictor of response and appeared to be better then the genotype in this particular study. Interestingly, he also showed that the vPT predicted response to a <50 copy level of response better than the actual PT.
The vPT does have some advantages over genotype tests and phenotype tests. Genotypes are complex and difficult for most clinicians to interpret. This technology may allow for an objective and quantifiable method to 'interpret" the genotype in a consistent manner. Secondly, since it is based on a genotype test it is likely (although not certain) to be cheaper and faster than phenotype tests. However, there are several caveats to note, first this study does not prove that the vPT is superior to either genotyping or phenotyping in managing patients. The patients in VIRA 3001 trial where all experiencing failure of a first line PI-based ARV regimen. So these findings can not be generalized to all clinical situations. Additional studies using different cohorts need to be evaluated including, ideally, cohorts from independent groups evaluated in a fashion that is blinded to the investigators at Virco. Secondly, this study shows predictive capacity but not clinical utility of vPT. We need to see if a vPT directed ARV therapy changes results in better outcomes for patients in a prospective trial compared to PT and or GT. None the less, it appears that as the Virco database grows and the matching of submitted genotypes becomes more sophisticated that there is much promise in this technology.
The Virtual Inhibitory Quotient?
There has been much talk recently about the use of Inhibitory Quotients (IQ's) as a better measure of resistance. Dale Kempf of Abbott Laboratories noted in his talk that the IQ concept comes from the microbiology literature and refers to a measure that includes information about a drugs actual or expected concentration (e.g. plasma Cmin) divided by a measure of the pathogens susceptibility to that drug (e.g. IC 50 or IC90). So the inhibitory quotient could be written: IQ = Cmin/ IC50. Theoretically the IQ is a better measure of resistance, because, as we know for most drugs, resistance is relative to drug concentrations. Investigators at Abbott showed last October at the Glasgow meeting that IQ predicts response to LPV/r. Dr. Kempf reported on the predictive capacity of a "Virtual IQ" in predicting response to ritonavir intensification of patients on IDV-based regimens with ongoing viremia (>50 copies). The Virtual IQ uses the virtual PT in place of the actual measured IC50, and the actual Cmin for IDV were measured in this PK study of 37 patients. So the vIQ = Cmin / vPT x IC50 for wildtype HIV corrected for 50% human serum (note: vPT is given as average fold change in IC50). Dr. Kempf's analysis indicated that not only did the vIQ for indinavir predict virologic response over 48 weeks but seem to discriminate response from lack of response better than did a simple genotype score (number of PI-related resistant mutations) or the vPT.
Phenotypic break points for Abcavir
One of the shortcomings of phenotype tests is the lack of "cut-points" or "cut-offs" that define virologic response to a particular drug. Two important cut-points could be defined for each antiretroviral agent. First the point at which the response to the drug becomes attenuated but still has antiviral activity and a second point at which the drug has no meaningful activity. Defining these cut-points would require studies that evaluate the phenotype at baseline and virologic response to each of the drugs currently used in clinical practice. One of the difficulties in establishing these cut-points is that most patients are treated with combinations of drugs, and therefore, teasing out the contribution to response that an individual drug makes is difficult. Randell Lanier of Glaxo et al presented an important study that defines the cut points for abacavir. In collaboration with the investigators at ViroLogic, Dr. Lanier took stored specimens from various studies in which abacavir was added to stable background ARV therapy as single drug intensification. Therefore, any virologic response observed in these studies was related to the abacavir. In evaluating 24 week response, he showed that any abacavir phenotype of less than 4.5 fold was associated with maximal response, and any fold change above 7 fold was associated with "severely" limited or no response. Phenotypes >4.5 <7 -fold were associated with a partial response. Whether these cut-points would be the same with the Virco PT is yet to be determined.
This is the kind of information clinicians require for each of the ARV drugs in use today. With these clinically defined cut -offs the advantage of phenotyping over genotyping may be fully realized as it would be much more difficult to define such cut-offs based on mutation patterns.
So where do we stand in regards to resistance testing? At present genotypes and phenotypes are commercially available. The virtual phenotype shows promise as a clinically useful tool and should be evaluated further. Measures of IQ may eventually prove more useful but require that PK measurements be made as part of clinical practice. This would likely mean timed blood specimens, which adds to the cost and complexity of patient management but should be studied further. However, the IQ measurement is only useful if one can manipulate the PK parameter of the drug. This can be done for many of the currently available PI's with RTV boosting but may be less applicable for the other available ARV's. And finally, the full potential of phenotyping will only be realized when clinically defined cut-offs are defined for all the drugs used in practice.
Update on resistance testing: Virologic
impact in prospective studies
There are now more than a half dozen randomized controlled studies evaluating the impact of resistance testing on virologic outcomes compared to "standard of care" (i.e. no resistance testing). Most of these studies have shown a modest (0.5 log) benefit in short-term virologic response in the groups receiving the resistance tests (either genotype or phenotype). To these studies we now add the ARGENTA study from Italy (an interim analysis of this study was presented by De Luca at the Worlds AIDS conference in Durban). In this study, patients with ongoing viremia while on ARV were randomized to receive genotype (TruGene, VGI) versus standard of care. Both groups received expert advice on the new ARV regimen. 174 patients were randomized. Baseline CD4 was roughly 260 and baseline viral loads were 4.2 log. Approximately half the patients were on their first ARV regimens with a quarter on their second and a quarter on their third or later ARV regimen. At baseline, the genotype group had slightly more evidence of drug resistance (reverse transcriptase mutations at codon 215, protease inhibitor mutations at codons 46, 82, 82 and 90). Despite these baseline differences, the genotype group had a statistically significant better response at month 3, as measured by % <500 copies, 27% versus 11% in the SOC arm (p<0.05, using intent-to-treat analysis). At month 3 genotyping was made available to the SOC arm for those who failed to achieve < 1 log decline or viral loads >500 copies. Adherence, as measured by self-report, appeared to play a role in response rates. The group with good adherence and genotyping had the best response at month 3 (33% <500 copies) followed by the SOC with good adherence (26%), followed by the genotype group with poorer adherence (18%) and finally the SOC with poorer adherence (7%). In multivariate models, genotyping, adherence, having a history of prior viral load <500 copies and being on 1st or 2nd ARV were independent predictors of response.
We also have the final results of the HAVANA trial that was first presented as a late breaker at ICAAC this past September. Investigators from Barcelona, have shown that genotype (TruGene, VGI) as well as expert advice improve virologic response at 24 weeks in this factorial designed study in which patients could receive genotyping or no genotyping with or without expert advice.
The evolving story of nucleoside resistance
At this conference we saw several new studies that evaluated the problem of cross-resistance between AZT and D4T. Nancy Shulman and colleagues from Stanford evaluated genotype predictors of response to D4T in patients from the ACTG 302 study. This was a study conducted in the early 1990's in which AZT experienced patients were treated with d4T monotherapy. Dr. Shulman genotyped (ABI) stored samples at baseline (prior to D4T) and determined that response to D4T was unaffected by the presence of the K70R mutation (often the first mutation to arise to AZT therapy) but that any additional AZT-related mutations severely limited response to D4T.
It is difficult to demonstrate resistance to D4T with currently available phenotype assays. In a series of elegant experiments, Dan Kuritzkes and colleagues at the University of Colorado in collaboration with investigators from ViroLogic compared the enzymatic activity of wild type reverse transcriptase with that of RT with multiple AZT-related mutations (41/67/70/215/219). The activity of the two RT enzymes was compared in the presence and absence of AZT and D4T (both in the active triphosphorylated form). The enzymatic activity of the wildtype RT was inhibited by both AZT and D4T and the enzymatic activity of the mutated RT was maintained in the presence of either drug to a similar degree.
Cohen in collaboration with GlaxoSmithKline investigators evaluated the impact of prior AZT compared to prior D4T use on virologic response in a post hoc analysis of the VIRA 3001 trial. Dr. Cohen demonstrated that prior D4T and AZT use resulted in a similar degree of resistance as measured by baseline genotype. Prior D4T appeared to have a more negative impact on subsequent response to ARV than prior AZT therapy even after controlling for important potential confounding variables. However, a formal multivariate analysis was not presented nor where differences in adherence evaluated. The investigators appropriately point out that clinicians need to think strategically in terms of nucleoside sequencing given the problem of cross-resistance within this class.
Michael Miller from Gilead Sciences presented genotypic and phenotypic analysis of patients on the "902" study in which ARV experienced patients on stable background regimens were randomized to three doses of tenofovir (TDF) with a placebo control. We have seen from previous meetings that the addition of tenofovir resulted in about a -0.6 log reduction in viral load that was sustained over 48 weeks. Dr. Miller showed that resistance to nucleoside RTI at baseline was common in this study and that this resistance had no major impact on response to TDF intensification in the 300 mg per day arm. Interestingly, the presence of the 184V mutation which had been shown to result in hypersensitivity to adefovir was not seen with tenofovir, on the other hand, the TDF was not as negatively impacted by the other nucleoside-associated mutations. In evaluating TDF phenotype at baseline in the 300 mg subgroup, no clear break point for response could be defined in this study. Although isolates with a baseline phenotype of >2.5 FC compared to reference strain were likely to have a less than 0.5 log viral load response with the addition of tenofovir.
It has been observed for some time that mutations associated with resistance to foscarnet seem to "resensitize" AZT-resistant viruses to AZT. The mechanisms for this interaction were not fully elucidated. Dr. Peter Meyer from University of Miami presented a study that nicely demonstrated the mechanisms by which these mutations associated with foscarnet resistance (88G/S, 89K, 161L, and 117T) exert this effect. The foscarnet -associated mutations inhibited the removal of AZT-MP from blocked primer template complexes. Therefore, these mutations allow AZT to remain on the primer-template thereby inhibiting further polymerase activity. The therapeutic implications of these studies are yet to be defined.
Resistance with LPV/r and APV
The primary mutation patterns associated with resistance to LPV/r were still to be defined. The investigators at Abbott presented additional information on resistance to LPV/r from their study (M98-863) in ARV-naïve patients. In this large (N=653) randomized study the 48 week efficacy of LPV/r based ARV was compared to a NFV based regimen. Of the subjects randomized to the LPV/r regimen 58 experienced virologic failure as measured by a viral load >400 copies. In 37 of these subjects genotypic results were available and there was no major PI-associated resistance identified. Resistance to 3TC was identified in 15/37 (41%). In contrast, 32% of isolates from patients failing NFV had genotypic NFV resistance, and 86% had 3TC resistance. Viruses with median 44-fold LPV/r resistance were only 6-fold resistant to amprenavir. Two patients who had not previously taken saquinavir remained sensitive to SQV. In a related in vitro analysis, phenotypic cross-resistance patterns between LPV/r and other PI's was evaluated. The investigators from Abbott showed that resistance to LPV/r was most closely correlated to resistance to ritonavir and indinavir and less so to amprenavir and saquinavir. Whether or not APV or SQV can be used to successfully control viral replication after resistance to LPV/r develops is yet to be determined.
Investigators from GlaxoSmithKline demonstrated that the resistance pattern to amprenavir appears to depend on the concentration of the drug used. It appeared that higher doses resulting in higher Cmins resulted in the 50V mutation - the signature mutation for this drug, while lower concentrations resulted in mutations at codon 54 with or without other protease mutations. Whether PK boasting of APV would result in the selection of 50V mutation pathway which has not been seen with other PI's and what the clinical implications of the 50V on response to subsequent PI-based regimens remains to be clarified.
Quality Assurance Issues with Genotyping
Concerns about the reliability of genotypes were raised in 1999 when Rob Shuurman and colleagues presented the results of the ENVA testing program. Recall that in ENVA 2 there was great variation in the ability of labs to detect mutations in mixtures and that many labs were unable to detect mutations when present in 50% or less of the mixtures. There were two reports from quality assurance programs on sequencing at this years conference. Investigators from Virco sent sequences from clinic isolates to 4 "independent" labs. In general, there was very high level of concordance of results both at the nucleotide level and at the amino acid level. In fact most of the counted discrepancies at the amino acid level were detected as mixtures by the labs and the mixture reported contained the mutation detected by the reference lab. In a second study, German investigators sent out clinical isolates from 3 HIV-infected patients to 24 labs and received reports back from 20 labs with 22 different sets of results. The labs used different genotypic techniques (10 ABI, 5 VGI, and others). In general, the labs were all able to detect major drug resistant mutations. Although 2 out of 22 missed a K103N, 1 each missed a 215Y, and 151M. Interestingly, L90M was present in one sample as a minority mixture (<15%) and this was missed in 16 of 21 reports but 5 labs were able to detect.
These two genotyping quality assurance studies seem to indicate that at least with the laboratories involved in these comparisons, the genotype results are reasonably comparable. Although we must await the results of the ENVA 3 evaluation for a more definitive comparison, it appears that genotyping labs are getting more consistent results as experience grows with these technologies.
A commercial lab using VGI's TruGene platform with a modified extraction procedure demonstrated reliable sequence results with repeated testing on specimens with viral loads as low as 116. This extends the work first presented at last year's CROI by Rob Llyod et al. using the TruGene system to get ultrasensitive genotypes. It appears that this result is reproducible in a commercial laboratory setting.