The State of Resistance Testing
A central tenet of evolutionary theory is that species undergo genetic change over time. Given a particular environment, some of these genetic variants are more "fit" (that is, they are able to reproduce more effectively) than others, and they become dominant. Other variants are fit enough to survive and coexist with the dominant population. Still others are so unfit that they are unable to reproduce, or they reproduce poorly, and eventually die out. The same, of course, is true with HIV; however, genetic variation is seen on a much greater scale than it is with many other species. Unlike most organisms, including other DNA-based viruses, HIV lacks the ability to check for and correct mutations while it replicates. Because of this, each person's viral population consists of many quasispecies (genetically distinct viruses that have evolved from an initial, or wild-type, virus). Furthermore, HIV replicates rapidly -- in an untreated person, up to ten billion virions can be produced each day. Since there is an average of one change in a nucleotide per replication, every possible mutation resistant to a single drug can be generated daily, although double mutations are much less common and triple mutations are fairly rare.
Generally speaking, mutations, even drug resistant mutations, tend to be less fit. However, fitness is relative to the environment. The goal of combination therapy is to create a hostile environment in which HIV is incapable of replicating at all. While we have not reached this ambitious goal, drug therapy can suppress viral replication to the point where HIV is undetectable by currently available assays. Yet, combination therapy does not always make the environment as hostile as it needs to be. In such cases, drugs may effectively prevent the replication of wild-type virus yet create an environment in which mutant viruses are actually more fit than they would otherwise be. These drug-resistant strains of HIV can replicate in the presence of drugs and viral load starts to rise.
What should be done? That standard answer, e.g., from the International AIDS Society-USA Resistance Testing Consensus Panel, has been to switch to an entirely new combination of drugs, if possible (Hirsch et al., JAMA, 279(24) 1984-91, June 24, 1998). (Word has it, however, that the Panel is revising its recommendations based on more recent studies of resistance testing. The new standards are expected to be published early in 2000.) And this makes sense if you don't know which drug or drugs the virus has become resistant to. But genotypic and phenotypic resistance tests hold out the promise of more educated choices. Ideally, these tests would allow people to remain on drugs that are effective, eliminate drugs that are ineffective, preserve more options for future regimens and even improve virologic response while lowering the cost of therapy.
On November 2 and 3, 1999, the Food and Drug Administration (FDA) held an advisory committee meeting on drug resistance testing. The committee, chaired by Scott Hammer, MD, heard state-of-the-art presentations on resistance testing from a variety of experts, including Drs. Douglas Richman, John Mellors, Susan Little and Richard D'Aquila. While the emphasis of the meeting was the potential role of resistance testing in antiretroviral drug development, both the committee and speakers also frequently, and understandably, discussed aspects that were relevant to clinical management.
Genotypic and Phenotypic Assays
Since Treatment Issues last covered resistance testing (see June 1998 and October 1998), the assays themselves have not changed significantly. Genotypic tests sequence all or parts of the reverse transcriptase and protease genes and provide information about which mutations are present. Relatively inexpensive and rapid, genotyping can provide clear and useful information for some antiretroviral drugs. For example, if the assay shows an M184V mutation, then the virus is resistant to 3TC. Non-nucleoside reverse transcriptase inhibitors (NNRTIs) also show high-level resistance in the presence of a single mutation (K103N). However, genotypic results can become quite complicated. For AZT and protease inhibitors, high-level resistance may require at least three mutations in a single genome. Knowledge of how mutations affect resistance is also incomplete, and the data are only useful if they can be interpreted. For reasons like this, genotypic test results often require expert analysis.
The quality of labs analyzing the data can also vary significantly, especially if a sample is a mix of wild-type and mutant virus. This was evident in an ICAAC presentation (Abstract 1168), which was mentioned by Dr. Douglas Richman at the FDA meeting. Test panels -- one with ten resistant mutations in the protease gene and one with 25 resistant mutations in the reverse transcriptase gene -- were sent out to dozens of labs in the US and Europe for genotypic resistance testing. Each panel further contained five samples containing 100% wild-type virus, 100% mutant virus, and 75/25, 50/50, and 25/75 mixtures. Of the 33 labs that responded, all did very well when analyzing the samples that were 100% wild-type, and most were pretty good when the samples were 100% resistant. However, accuracy declined significantly for the mixed samples. For example, in the 50/50 mixtures, only 49% of the labs detected mutations in the protease gene and 37% detected mutations in the reverse transcriptase gene. While the accuracy of results did not seem to be influenced by the type of genotypic assay used, Dr. Richman pointed out that the experience of the labs and technicians is crucial. "Those of us who have done sequencing, " he said, "know that having a good, experienced person doing it gets the best results."
But how do you know how good a lab is? Unfortunately, Richman declined to "name names." And until the tests are FDA approved, there is no way of knowing, short of finding an expert willing to give advice. Yet this could change in the near future. In September, the FDA reclassified genotypic assays from Class III to Class II devices, making approval both easier and faster. The FDA has also made it clear that it expects manufacturers to seek approval, although none have yet submitted an application and, for the time being, this has not been mandated.
Phenotypic assays, which in many ways complement genotypic tests, are not approved by the FDA either, and the agency is not looking to require approval for these "home brew" tests in the near future. Due to the technological complexity of phenotypic testing, they must be done in-house and are more expensive and take longer than genotypic tests. Instead of looking at the genetic make-up of the virus, phenotypic assays determine how susceptible, in vitro, a particular virus is compared to a reference (wild-type) virus. Reverse transcriptase and protease genes from the patient are inserted into an HIV envelope gene, and the hybrid HIV virus' ability to replicate is then tested in the presence of currently approved drugs. A report provides the fold-resistance of the virus, usually in terms of IC50 (the drug concentration needed to reduce HIV activity by half; the higher the IC50, the more resistance HIV is exhibiting to that drug). General guidelines are that if the virus shows ten-fold or greater resistance, then it is highly resistant, four- to ten-fold is moderately resistant and below four-fold is susceptible, although this really depends on how many fold a drug's plasma concentrations are over its IC90. In fact, these guidelines have no established clinical utility. (Virco uses a four-fold cutoff and ViroLogic uses a 2.5-fold cutoff; but these numbers only indicate statistical confidence that the results are not due to test variability.) Moreover, it is likely that relevant cutoffs will vary between drug classes, between drugs within a class and even from patient to patient.
Finally, both genotypic and phenotypic assays share two important drawbacks. First, the tests vary in their ability to detect subpopulations of resistant virus. A genotypic method known as sequencing, in which most or all of the genome is tested, only detects species that make up 20 to 50% of the total viral population. Genotypic differential hybridization, which probes specific codons, can detect down to two to five percent; however, the test can only detect mutations in specific parts of the genome. Phenotypic tests, on the other hand, can generally detect subpopulations that make up 10 to 20% of the total population. Because of this, results may be virtually useless for some people, either because therapy has been suspended and wild-type virus has become predominant or because the mutant population is still too small to be detected. In such cases, the resistant subpopulation can quickly become dominant if suboptimal therapy is prescribed (the problem, of course, is that no one would know the therapy is suboptimal based on the resistance assay). Second, the tests require viral loads of at least 1000 copies (ViroLogic estimates that it can successfully test greater than 90% of samples with a viral load between 500 and 1000 copies). So, at this point, resistance tests are not helpful for predicting failure for those who are currently succeeding on therapy.
Relative Advantages and Disadvantages of Genotypic and Phenotypic Assays
VIRADAPT, GART and Other Studies
Nevertheless, data on the utility of geno- and phenotyping have been mounting. At the FDA advisory meeting, summaries of two prospective genotypic studies -- VIRADAPT and GART -- were provided. In addition, ten retrospective and two prospective studies were reanalyzed by the Resistance Collaborative Group in order to make sense of them collectively.
Dr. Phillipe Clevenbergh presented 12-month data from the VIRADAPT study (see also Treatment Issues, July/August 1999). In this trial, heavily pretreated patients who were failing therapy were randomized into a control arm (n=43) or a genotypic arm (n=65). For the genotypic arm, data from a complete sequencing of the major part of the reverse transcriptase and all of the protease gene were made available, and an algorithm was used to interpret the results. Patients in the genotypic arm received therapy based on the results while the control group received the standard of care. After six months, it was clear that those in the genotypic arm were doing significantly better, so genotypic data was made available to all patients (31% of the control patients, however, had finished their 12-month follow-up by this time). The genotypic arm showed a consistent reduction in viral load throughout the 12 month period, with a mean reduction of 1.15 log, whereas the control arm had a reduction of 0.67 log at 6 months and an additional drop of 2.98 log in the last six months. Similarly, 30% of patients in the genotypic arm had a viral load below 200 copies throughout the 12 months. Fourteen percent of the control arm were below 200 copies by six months, compared to 30% at 12 months. Both the absence of a primary protease mutation and genotypic-guided treatment were independent predictors of virologic success.
Clevenbergh noted, however, that 70% of patients failed to go undetectable even with genotypic testing, and he suggested that reasons other than drug resistance -- e.g., suboptimal drug levels -- should be considered, especially since a correlation has been found between low plasma levels of protease inhibitors and a rebound in viral load. And Clevenbergh had data to back up his suggestion. A pharmacological substudy was conducted on 81 patients at one site who had at least three protease inhibitor plasma trough levels taken over the 12 month period. In both the control and genotypic arms, about 32% had suboptimal protease inhibitor concentrations. Viral load drops were significantly better for those in both arms who had optimal drug concentrations (1.2 log reduction at 12 months versus 0.36). Those who were both in the genotypic group and had optimal protease inhibitor concentrations fared much better than those with suboptimal concentrations in the control group (none of the latter attained undetectable viral loads). The other two groups -- control arm with optimal concentrations and genotypic arm with suboptimal concentrations -- fell somewhere in the middle and had similar outcomes. Clevenbergh concluded that optimal protease inhibitor concentrations, genotypic-guided treatment and the absence of a primary protease inhibitor mutation all independently predicted successful virologic outcome.
The next presentation, a summary of the CPCRA's GART study, was given by Dr. John Baxter. Patients had to have a three-fold increase in viral load while taking a protease inhibitor and two nucleoside analogs for 16 weeks. They had a median CD4 count of 230, a median viral load of approximately 28,000 copies/ml and a total history of at least 12 months of antiretroviral therapy. Unlike the VIRADAPT study, however, about half were on their first drug regimen.
Initially, each patients' doctor specified the salvage regimen he or she would give without genotypic information, and blood was drawn for viral load and genotypic data. Virologists then reviewed each patient's genotypic results and treatment history, and they prepared a report listing mutations, an interpretation of the data and up to four treatment suggestions. Next, patients were randomized into a GART (n=78) or no-GART (n=75) arm. Those in the no-GART arm received the treatment that had already been defined by their doctor. For the GART arm, doctors received the GART report, but they did not have to follow it. And while 83% said they were influenced by the GART report, only 54% actually prescribed one of the suggested salvage regimens. (Reasons for not following the suggestions included concerns about toxicity, patient preference and saving drug classes -- especially NNRTIs -- for later.)
At 12 weeks (the duration of the GART study), those in the GART arm had a mean viral load reduction of 1.19 log, compared to 0.61 in the no-GART arm. The genotypic arm also did better for a variety of predefined subcategories -- a CD4 count above or below 200, a viral load above or below 25,000, having failed one or more than one protease inhibitor, having or not having a major mutation on the reverse transcriptase or protease gene, and taking or not taking a non-nucleoside. In addition, 34% of those in the GART arm had a viral load below 500 copies, compared to 22% in the no-GART arm (the p-value was 0.1, which does not indicate statistical significance). Thus, the success of genotyping was not substantially better in GART than in VIRADAPT even though all patients in the latter group were heavily pretreated and half of those in the former were on their first regimen. (Baxter mentioned that patients who had few options didn't even do that well in the genotypic arm even though doctors were more likely to follow expert advice for advanced patients.)
The results were further analyzed in terms of change in viral load by the number of active drugs (defined as drugs to which study virologists determined HIV was sensitive or possibly resistant) prescribed. In both the control and genotypic arms, there was a 0.37 log reduction in viral load for each active drug added to the regimen, although 86% of those in the genotypic arm received three or more active drugs, compared to 45% in the control arm. Baxter noted, however, that there was also a 0.17 log reduction for each additional "inactive" drug, which suggests that even they might provide some benefit to the patient.
The design of the study, as well as its short duration, leave some questions. Will the viral load reductions be sustained for six months or a year, or longer? An answer might be forthcoming since patients are being rolled over to a longer term monitoring study. However, one reason for the short time frame was that researchers didn't think people would be willing to participate in a longer study without getting access to their genotypic results. For the same reason, it is doubtful that a longer term study will have much of a comparison arm. And how significant was the fact that almost half of the health care providers didn't follow expert advice? Baxter mentioned that there was a trend, although it was not statistically significant, suggesting that patients did better when health care providers followed expert advice. Conversely, what would have happened if recommendations were followed, without regard to either future therapeutic options or patient preference? Clearly, these factors need to be considered. And what if the control arm had the benefit of expert opinion on everything except genotypic results? Without a control, the significance of expert advice can't be determined.
In the next presentation, led by John Mellors, ten retrospective and two prospective studies were reanalyzed according to a data analysis plan in an attempt to make sense of their collective results. Researchers examined all the drugs in the studies in order to determine a genotypic or phenotypic sensitivity score. For each drug that a patient's virus was sensitive to, one was added to the score; otherwise, zero was added. In other words, the higher the patients' scores, the more drugs they were likely to respond to. Actually, things were slightly more complicated. In determining the genotypic sensitivity score, an algorithm was created based on a -- sometimes hard-won -- consensus; and in some cases, the 0/1 dichotomy was modified (for example, the 184V mutation from a patient taking adefovir received a 1.5 since there is evidence that this mutation makes the virus hypersensitive to adefovir -- although the FDA may beg to differ ; see sidebar). And there were two phenotypic sensitivity scores. The first was based on the minimum cutoff of the assay used (2.5-fold for Virco's test and four-fold for ViroLogic's) and the second used a ten-fold cutoff.
The research team presented results after results that consistently showed that both higher genotypic and higher phenotypic sensitivity scores correlated with a reduced likelihood of virologic failure. Moreover, these scores predicted failure better than baseline viral load or log changes in baseline over a 24-week period. In the phenotypic studies, the four-fold cutoff correlated with failure better than the 10-fold cutoff. Nevertheless, no solid conclusions can be drawn from this, especially since clinically important cutoffs are likely to vary with different drugs, and for some newer drugs a higher cutoff might be appropriate. Cutoffs may also vary depending on whether a patient is treatment experienced.
These data suggest, at the very least, that resistance testing can be better than nothing. That is, in some situations they might provide information about which drugs are ineffective. Nevertheless, Dr. Richard D'Aquila provided a sobering presentation on the complexity involved. Some of the problems have been mentioned already. For example, while the implication of some mutations is clear-cut, this is not always the case. Furthermore, mutations can interact in ways that are not understood: a mutation that does not confer resistance by itself, for instance, could contribute to resistance in the presence of other mutations. To make matters worse, currently available assays fail to detect minor species of drug resistant virus, so even if we had perfect knowledge, the tests don't always provide the necessary data. Drug levels can also play an important role. For protease inhibitors and NNRTIs trough levels are probably most significant, whereas for nucleoside analogs cellular triphosphate levels need to be examined. And it is possible that newer drugs that attain higher blood concentrations could overcome resistance, although this does not mean that resistance to such drugs won't develop.
D'Aquila added to the list of confounding factors. Resistance testing should be done while a patient is on failing therapy because, once therapy is stopped, wild-type virus can quickly outgrow resistant virus, pushing it below detectable levels. The timing of the blood sample with respect to viral breakthrough can also effect how a resistance test should be interpreted. For example, resistance to 3TC is often seen early after viral rebound, but there could also be small levels of protease inhibitor resistant virus that will show up if therapy continues.
The use of combination therapy also complicates matters since the drugs can have interactive effects and a virus that is resistant to one drug might be especially sensitive to another drug. Moreover, while no one at the meeting suggested doing monotherapy studies, the use of combination regimens makes it all the more difficult to understand the relationship between genotype and phenotype. While phenotypic assays could theoretically test the effects of drug combinations on resistant virus, this would be extremely complex, expensive and time-consuming.
Although genotypic and phenotypic tests are complementary in many ways, they can also give conflicting results. For example, a phenotypic test could show that a virus is susceptible to a drug while a genotypic test shows it is resistant. The reason could be that there is a mixture of wild-type and mutant at the relevant codon (in which case the genotypic test might -- or might not -- provide an early warning of future resistance). Another possibility is that mutations could interact, resulting in increased resistance or susceptibility. Conversely, the genotypic test could indicate susceptibility while the phenotypic shows resistance because the mutations that lead to resistance are not yet understood, a common problem with newer drugs. Or, again, the discordance could be due to a mixture of wild-type and mutant virus if the genotypic test is better at amplifying the minority strain.
Anatomical compartments raise another potential problem. The inability to detect resistant virus in the blood doesn't mean that it is not elsewhere in the body (e.g., semen or cerebrospinal fluid) from where it could emerge. Currently, it looks as if protease inhibitor resistant virus is unlikely to be found in the semen if it is not in the blood, but there are fewer studies of other compartments.
Despite the complexity involved with resistance testing, their imperfections and the importance of taking other factors into account, D'Aquila concluded that "perfection shouldn't be the enemy of the good." Improved tests and methods of interpretation are likely to lie in the near future, and for some populations, currently available resistance assays can be a valuable tool.
Who Could Benefit...
According to guidelines by the International AIDS Society-USA, the two populations most likely to benefit from resistance testing are those who are newly infected (including newborns and people infected in an occupational setting) and those who are changing drug regimens due to viral rebound. Because resistance testing requires a detectable viral load, it offers little for people who are on highly successful therapy. Similarly, due to an inability to detect minor populations of resistant virus, the tests could miss important mutations in people who are treatment naive but have an established infection or who have been off therapy for more than a couple weeks. In both cases, wild-type virus has probably outgrown the drug resistant virus, making the latter difficult to detect. However, the resistant virus is still present, and could quickly become dominant again if ineffective drugs are prescribed.
Due to the cost of the tests, the most controversial population is the recently infected. How likely is it that a person has been infected with a drug resistance virus? This was the topic of Dr. Susan Little's FDA presentation. Clearly, infection with drug resistant HIV is possible through sexual contact, IV drug use, from mother to infant or in a health care setting. The earliest reported case was of AZT resistance in 1992 and since then cases of primary infection with resistance to all classes of drugs have been documented. But is it common enough to justify expensive, sometimes unreimbursable, tests? Reanalyzing studies from North America, Europe and Australia, Little estimated that about five percent of recently infected people have clinically relevant drug resistant HIV. Furthermore, several large studies in the US and Europe have shown that one to four percent of patients are infected with a virus that is resistant to more than one class of antiretroviral drugs. Little noted, however, that studies tend to over-represent gay, white men in major metropolitan areas, typically in the Western US, so things could be much different in other geographic areas or with people who are infected through heterosexual sex, intravenous drug use, perinatally or in an occupational setting.
Little did not suggest widespread resistance testing for newly infected patients. However, in determining the utility of resistance testing in a recently infected person, the prevalence of resistant virus in the relevant population is worthy of consideration. If the local prevalence of drug resistant virus is five to ten percent, testing could be warranted. In addition, a slower than normal response to an initial drug regimen could be an indication of drug resistance and, therefore, merit resistance testing. Finally, she recommended more studies to monitor the prevalence of resistant virus, especially with respect to geographic location and risk exposure, as well as to determine the clinical significance of primary infection with resistant, especially moderately resistant, virus.
And Who Will Pay?
With genotypic assays priced at around $400 and a phenotypic assay around $900, cost is certainly an issue. Moreover, because the tests are not FDA-approved, getting reimbursement can be an uphill battle. With growing evidence of the tests' value, however, more insurance companies are willing to pay for them. Aetna, for instance, covers both genotypic and phenotypic assays in cases of treatment failure (after other possible explanations have been ruled out). Of note, the company based this decision on the International AIDS Society-USA's 1998 recommendations, which only hesitantly endorse resistance testing. Presumably, the more aggressive recommendations that are expected will convince more insurance companies to cover the cost of the tests.
In the meantime, both ViroLogic (which does phenotyping) and Virco (which offers genotypic as well as phenotypic testing) offer very limited "compassionate access" programs for people with low incomes and/or no insurance. Physicians or community organizations typically contact Virco on behalf of patients who are unable to pay for the tests. ViroLogic has set up a toll-free number (1-877-436-6243) both for low-income people with no insurance and to help those who are insured get prior approval or reimbursement for their phenotypic test.
This article was provided by Gay Men's Health Crisis. It is a part of the publication GMHC Treatment Issues. Visit GMHC's website to find out more about their activities, publications and services.