Simultaneously, a new class of potent antiviral drugs, HIV protease inhibitors, was introduced. Protease inhibitors have proven, for the first time, that potent, sustained suppression of HIV replication can be achieved in a clinical setting for at least 2 years [1, 2] and that progression to AIDS and death can be dramatically reduced [3-5]. For this and other reasons, the number of AIDS-related deaths declined in the United States in 1996, the first such decline in the history of AIDS [6]. Thus, there is reason to be cautiously optimistic that with appropriate therapies, HIV and AIDS may ultimately be controlled.

For anti-HIV drugs to have long-term benefits, however, they must be used in a way that will completely suppress viral replication. This is because HIV, like any other microorganism, develops resistance to even the most potent suppressive agents if it is allowed to replicate, even slowly, in the presence of those agents. Drug resistance evolves through the accumulation of replication errors (mutations) in the viral population and the preferential survival of the mutants best able to replicate in the presence of the drug. This replication sets the stage for new rounds of mutation and selection, increasing the level of drug resistance over time. Thus, the evolution of drug resistance is a strict Darwinian process, driven by viral replication, mutation, and selection [7, 8].

Because mutations arise as replication errors, they are a direct result of the viral replication process itself. The probability that any new mutation will occur is thus a direct function of the extent of ongoing viral replication. Potent inhibition of viral replication with therapy can greatly suppress the emergence of resistance in a nonspecific manner [9]. Indeed, if a regimen is potent enough to block all viral replication, it should be possible, in theory, to prevent the emergence of resistance altogether.

But simply having drugs with great antiviral potency is not enough. This is because, for many drugs, mutations engendering moderate- to high-level resistance are likely to exist, even before therapy. It has been estimated that because of the high viral mutation rate, rapid viral turnover, and high viral load in HIV-infected persons, every possible single nucleotide substitution mutation occurs 10⁴ to 10⁵ times per day in a given patient [7]. Thus, most, if not all, single amino acid substitution mutations are likely to preexist. If any of these mutations confers a significant degree of resistance to the drug or drugs used, the virus will be allowed to replicate and to evolve greater resistance. Under these circumstances, even a very potent regimen may be inadequate to contain the viral population for an extended period. This is true of all known non-nucleoside reverse transcriptase inhibitors and also of many protease inhibitors and nucleoside reverse transcriptase inhibitors.

Long-term viral suppression, therefore, requires that a high genetic barrier to resistance be established. Simply put, the regimen should require the viral population to acquire as many new mutations as possible before the level of resistance becomes high enough to overcome the block imposed by therapy [8]. This can be accomplished in several ways.

First, using drugs that require the virus to undergo multiple mutations to achieve high-level resistance maximizes the efficacy of the drug for the existing viral population and minimizes the probability of breakthrough. In this respect, a drug to which resistance develops after only a single amino acid substitution is expected to be more vulnerable to resistance than an equipotent drug that requires the virus to undergo multiple mutations to achieve the same degree of resistance.

Second, the need for multiple mutations can be increased further by combining different drugs that inhibit independent targets. Three distinct therapeutic classes of drugs with nonoverlapping sets of resistance determinants exist: protease inhibitors, nucleoside inhibitors, and non-nucleoside reverse transcriptase inhibitors. There is no evidence that mutations compromising the effects of members of one class will reduce the utility of members of any other class. This is one of the most important benefits of divergent combination therapy: When many simultaneous mutations are required, the probability of preexisting resistance in a therapy-naive patient becomes negligible and the effect of the drug combination is maximized.

However, many patients are not naive to therapy. Their extensive drug experience and the consequent resistance of their viruses to multiple drugs may be the greatest obstacle to long-term viral suppression. The presence of preexisting resistance mutations in the viral population effectively lowers the genetic barrier to resistance with respect to a wide variety of possible treatment regimens and sometimes even to the most efficacious combination therapies.

Drug failure necessitates attempts to substitute other drugs, often of the same therapeutic class, for one failed regimen. Although there is extensive cross-resistance among most of the nucleoside analogue reverse transcriptase inhibitors, there might be some combinations of these inhibitors that could be used successfully in sequence. However, all of the non-nucleoside reverse transcriptase and protease inhibitors have been shown to select for extensive in-class cross-resistance. Because they select amino acid substitutions that are shared by other drugs of the same class [10-16], acquisition of any of these shared substitutions lowers the genetic barriers to resistance to the other members of that class and hastens the development of resistance. In many cases, it also leads to directly measurable cross-resistance. Although each drug has been shown to select for some amino acid substitutions that are not generally seen with other drugs, their patterns show more similarities than differences.

It is not surprising, then, that in the "real" world, therapies containing protease inhibitors have not been as successful as clinical trials have shown they can be. In one recent study, the percentage of therapeutic successes (defined as sustained viral RNA reductions) was only about half that seen in clinical trials of the same regimens [17]. This was attributable to the improper use of these potent new drugs, which resulted primarily from the addition of single new agents to preexisting regimens; from the previous use of similar drugs, including protease inhibitors; and from poor adherence.

In this issue, Shafer and colleagues [18] provide textbook examples of the dangers of suboptimal sequential therapy. Over the course of several years, patients received multiple drugs. These were usually in combinations, but often, only one new drug was introduced at a time. The result was essentially "sequential monotherapy." And because the genetic barrier imposed at each stage was low, the viral populations were permitted to acquire resistance to each new drug in turn. The resulting viral populations were thus resistant to almost every available antiretroviral drug, even some drugs to which patients had not been exposed. Although it may have seemed that this therapeutic strategy could preserve future therapeutic options, it actually limited those options by allowing viral replication to occur. In retrospect, it is easy to see the fallacy behind this approach, but at the time they were (sequentially) introduced, each new drug was often the only available option for many patients who had already exhausted existing drugs. Understandably, therefore, each new drug was used, and used up, as it was introduced.

When a therapy fails because of resistance, it can be extremely difficult to choose an effective salvage regimen, especially in heavily therapy-experienced patients. Consequently, there has been much recent interest in the use of in vitro testing methods to guide these decisions. The two types of resistance testing available, phenotypic and genotypic, yield different kinds of information. Phenotypic tests measure the ability of drugs to block viral replication in cell culture, whereas genotypic tests determine the actual mutations present in the viral nucleic acid.

Virologic breakthrough is often accompanied by a loss of viral drug susceptibility in in vitro tests, but this is not always the case. We have found instances in which viral RNA rebound due to resistance was not reflected by measurable phenotypic resistance in the virus in vitro [19]. This is not surprising, given the inherent high variability and insensitivity of these tests. Moreover, immeasurably small changes in growth rates can have an enormous impact on the genetic composition of viral populations [7]. Thus, phenotypic drug "susceptibility" does not ensure good clinical virologic response.

In contrast, genetic tests can identify the mutations themselves, including those whose effects are phenotypically immeasurable. But for the data to be interpreted, the genetic basis for drug resistance must be well understood. Unfortunately, because of the complexities of mutational interactions, genotypic resistance patterns are often difficult or impossible to interpret, even for experts.

This lack of correlation between mutation data and clinical significance is seen especially with the protease inhibitors. Many mutations have been identified as correlates of resistance to some protease inhibitors and not others. But all too often, these apparent differences reflect nothing more than our incomplete state of knowledge about the genetic basis of resistance to some drugs, especially those that are the least well characterized. The less that is known about a drug, the simpler its resistance pattern seems to be and the lower its perceived potential for engendering cross-resistance. A rapidly growing body of evidence has shown that this perception of low potential for cross-resistance is in error, however; in fact, widespread cross-resistance exists among all four approved protease inhibitors in a clinical setting [20]. Unfortunately, that same misperception has been the primary justification for using genetic testing methods to guide decisions to switch among the protease inhibitors.

Apart from our incomplete knowledge, an even greater problem limits our ability to identify drugs to which patients will respond, and this problem applies equally to both phenotypic and genotypic tests. Because of the random appearance of different mutations in different viruses, HIV-1 populations consist of multiple, genetically distinct subpopulations evolving in parallel [7, 8]. Any sample taken from such a population is therefore heterogenous, and none of the available testing methods is likely to detect variants representing less than 10% to 25% of the population. However, even variants representing fewer than 1 in 1000 viruses have been shown to be clinically significant [19]. Thus, failure to detect resistance is not evidence of its absence.

Because of the many uncertainties associated with the clinical interpretation of virologic tests for resistance, the results of these tests can be misleading. Until the usefulness of these tests is shown in controlled clinical trials, it is premature to rely on these tests to guide clinical decision making.

Given the poor prospects for long-term salvage of failed therapy, the best strategy is to prevent resistance from occurring in the first place. Fortunately, an understanding of basic virology and genetics provides a sound rationale for accomplishing this.

First and foremost, it is critical to maximize the potency of therapy from the beginning. This can be done by choosing the most potent drugs possible, using them at their maximum tolerated doses, and using them in combination. The objective should be nothing short of maximal suppression of viral replication because the first attempt is likely to be the most successful.

Second, to maximize genetic barriers to resistance, drugs that require the virus to undergo multiple mutations to achieve resistance should be combined with the most efficacious agents in other therapeutic classes. Whenever possible, drugs should be from classes to which the patient is therapy-naive or should be as different from previously used drugs as possible.

Finally, these aggressive therapeutic regimens should be made as tolerable and user-friendly as possible (without compromising efficacy) to encourage long-term adherence and continued viral suppression.

The advent of potent new antiretroviral therapies, especially the protease inhibitors, combined with an improved understanding of the biology and genetics of HIV, have revolutionized the treatment of HIV infection and AIDS. These developments have given rise to a renewed sense of optimism but also to great trepidation that these gains could be rapidly lost if not used wisely. To apply these powerful tools and principles to clinical practice in a way that is both consistent with the biology of the virus and tolerable to patients is a challenge, but it is also an unprecedented opportunity to make a substantial and lasting impact on HIV disease.