HIV drug resistance

HIV drug resistance is the central problem of treating a virus that mutates as it replicates. Every patient on antiretroviral therapy (ART) carries the seeds of resistance from the first day of treatment; whether resistant virus ever takes over depends on the balance between drug pressure and adherence.

Why resistance happens

HIV drug resistance is best understood not as the virus “becoming resistant” to a drug but as the selection of pre-existing resistant variants under drug pressure. Three biological facts drive everything that follows.

HIV replicates fast and inaccurately. A person with untreated HIV produces on the order of ten billion virions per day, and the viral reverse transcriptase copies the RNA genome into DNA at an error rate of roughly one mutation per genome per replication cycle, with no proofreading. Within each infected person the virus exists not as a single sequence but as a quasispecies, a swarm of related variants. By chance that swarm already contains every possible single-nucleotide substitution, including the changes that confer drug resistance.

Resistant variants are usually less fit. Most resistance mutations carry a measurable fitness cost: they slow replication, weaken an enzyme, or otherwise impair function. In the absence of drug, fit wild-type virus outcompetes them and they remain at very low frequencies, often well under 1%.

Drug pressure inverts the balance. When a patient starts effective antiretrovirals, wild-type virus is suppressed below the limit of detection, and the previously rare resistant variants gain a selective advantage: they replicate a little while the wild type cannot. If drug levels are reliably high enough to suppress even the resistant variants, no resistance emerges. If drug levels are inadequate, whether through incomplete adherence, malabsorption, drug interactions or sub-therapeutic dosing, the resistant variants are selected, expand, and eventually dominate. Resistance is selected, not created.

Three clinical consequences follow.

• Adherence is decisive. Resistance is, more often than not, a marker of inadequate drug pressure. The reflex dolutegravir drug-level testing workflow used in South African practice turns this into an algorithm: when drug levels are undetectable in a patient supposedly failing on TLD (tenofovir, lamivudine and dolutegravir), the result reflects non-adherence, and a genotype is not informative.
• Resistance is archived. Once a resistant variant has been selected, the integrated provirus carries that mutation for life. Even if drug pressure is removed and wild-type virus dominates again, the resistance is archived in the reservoir and re-emerges when the drug is restarted.
• Some mutations cripple the virus. A small number of resistance mutations, the most familiar being M184V, confer high-level resistance to one drug but reduce viral fitness so much that the virus is impaired even in the presence of drug. This is the basis of “recycling” lamivudine (3TC) in second-line regimens.

The language of mutations

Every resistance report uses the same shorthand, M184V, K103N, L90M, Q148R, which follows one pattern: [wild-type amino acid][codon number][mutant amino acid].

M184V illustrates the convention.

• 184 is the codon number in the protein. For reverse transcriptase, codons are numbered from the start of the protein; codon 184 sits in the catalytic site, in a loop called the YMDD motif that coordinates the incoming nucleotide during DNA synthesis.
• M is the wild-type amino acid, the standard one-letter code for methionine.
• V is the mutant amino acid that has replaced it, the one-letter code for valine.

So M184V means methionine at codon 184 has been replaced by valine.

Each amino acid has a standard one-letter and three-letter code; the one-letter code is universal in molecular biology and is the form used in a resistance report. The twenty are:

Letter	Amino acid		Letter	Amino acid
A	Alanine		M	Methionine
R	Arginine		F	Phenylalanine
N	Asparagine		P	Proline
D	Aspartate		S	Serine
C	Cysteine		T	Threonine
E	Glutamate		W	Tryptophan
Q	Glutamine		Y	Tyrosine
G	Glycine		V	Valine
H	Histidine		I	Isoleucine
K	Lysine		L	Leucine

The amino acid sequence is set by the underlying DNA: every three nucleotides (a codon) encode one amino acid. To change methionine to valine at RT codon 184, the triplet changes from ATG (methionine) to GTG (valine), a single A-to-G substitution in the first base. When reverse transcriptase makes its inevitable copying error and substitutes G for the original A, that one change breaks the YMDD loop’s ability to discriminate between 3TC and natural dCTP and confers over a hundred-fold loss of 3TC activity.

On a report, then, M184V records the protein-level consequence of a single nucleotide change in one variant of a patient’s quasispecies, selected to dominance by drug pressure under inadequate adherence.

Mechanisms of resistance by class

Each drug class is escaped in a characteristic way. Four mechanisms recur.

Discrimination. The virus distinguishes the drug from its natural substrate and rejects it. This is used by the NRTI (nucleoside reverse transcriptase inhibitor) mutations M184V, K65R and L74V: the active site of RT is tightened so that the natural nucleotide, slightly differently shaped from the analogue, is preferentially incorporated and the drug excluded.

Excision. The virus removes the incorporated drug. This is used by the thymidine analogue mutations (TAMs), M41L, D67N, K70R, L210W, T215Y/F and K219Q/E, which arose under zidovudine (AZT) and stavudine pressure. Once an NRTI has been incorporated into the nascent DNA and stopped chain elongation, the mutated RT uses a phosphorolytic reaction to remove it and resume synthesis. TAMs accumulate stepwise, with full resistance needing three or four, and their pattern of accumulation explains why AZT-experienced patients often respond less well to subsequent tenofovir regimens.

Allosteric pocket disruption. This is used by the NNRTIs (non-nucleoside reverse transcriptase inhibitors). K103N, Y181C, G190A and others alter the shape of a hydrophobic pocket on RT that the NNRTI binds. Because the pocket sits next to the active site rather than within it, a single mutation can sharply reduce drug binding without much affecting catalysis, which is why the NNRTIs have the lowest genetic barrier of any class: one mutation loses one drug.

Active-site reshaping. This is used by the PIs (protease inhibitors) (substrate-cleft and flap mutations such as D30N, V32I, M46I/L, V82A, I84V, L90M) and the INSTIs (integrase strand-transfer inhibitors) (Q148H/R/K, N155H, Y143C/R, plus secondary mutations). Each major mutation removes a few ångströms of substrate grip; because the enzyme must still process its natural substrate, multiple mutations are needed to escape entirely, giving these classes intermediate to high genetic barriers.

Cross-resistance and the principle of switching class

A central principle governs regimen change: cross-resistance is the rule within a class and the exception between classes.

Within a class, the same active-site or pocket residues serve every drug that binds there. K103N abolishes both efavirenz and nevirapine, since any drug binding the same NNRTI pocket is lost. The Q148 pathway degrades raltegravir, elvitegravir and, with enough secondary mutations, dolutegravir. PI resistance mutations accumulate stepwise, so that viruses with several mutations are progressively less susceptible to atazanavir, lopinavir and darunavir together.

Between classes, the drugs target entirely different proteins (RT, protease, integrase, capsid, the envelope spike, host receptors) and resistance does not cross: a mutation that abolishes NNRTI activity has no effect on PIs, INSTIs, capsid inhibitors or the entry and attachment classes.

This underlies regimen design at virological failure: when a class is lost, switch class, and aim for at least two, preferably three, fully active drugs.

Testing for resistance: phenotypic and genotypic

Resistance is measured in two complementary ways.

Phenotypic testing measures susceptibility directly. The patient’s pol gene is amplified from plasma by PCR (polymerase chain reaction), cloned into a laboratory backbone, and the resulting recombinant virus is grown in cell culture against a range of concentrations of each drug. The concentration that inhibits 50% of viral replication (IC50) is compared with the wild-type IC50 to give a fold-change, which cut-offs (lower, upper and biological) translate into a clinical interpretation: fully susceptible, partially susceptible or fully resistant. The two best-known commercial assays are PhenoSense and Antivirogram.

Phenotypic testing is the gold standard but slow and expensive, with a turnaround of weeks and a cost of several hundred US dollars per sample, so it is reserved for research, drug development and complex clinical cases. Its fold-change range is drug-dependent: tenofovir reaches only about a five-fold loss, while 3TC and AZT span more than two hundred-fold.

Genotypic testing measures the sequence and predicts resistance from it. Plasma RNA is reverse-transcribed and the pol gene amplified by PCR; the amplicon is sequenced, almost always by Sanger, with next-generation sequencing (NGS) increasingly used in research and reference laboratories. The nucleotide sequence is translated to amino acids, mutations are identified against a reference, and an algorithm predicts susceptibility for each drug.

Genotypic testing is comparatively fast and cheap, with a turnaround of days to weeks and a cost of a few hundred US dollars. It identifies established mutations from the consensus sequence but misses minority variants below about 20%. It is the workhorse of clinical practice worldwide and in the South African public sector.

The distinction is that phenotype is direct measurement and genotype is sequence plus prediction, both resting on the same body of evidence linking mutations to phenotypes.

Sanger sequencing

The Sanger method has been the workhorse of clinical sequencing since the late 1970s, and its principle is straightforward.

DNA polymerases extend a growing strand by adding nucleotides (A, T, C, G) that match the template, using the 3′ hydroxyl group of the previous nucleotide as the attachment point. Sanger sequencing exploits a chemical trick: dideoxynucleotides (ddNTPs) are modified versions of A, T, C and G that lack the 3′ hydroxyl, so when a polymerase incorporates a ddNTP the chain stops dead, with no 3′ OH to attach the next base.

A Sanger reaction combines:

• the template DNA (in HIV testing, the PCR-amplified pol gene);
• a short primer;
• a DNA polymerase;
• the four normal dNTPs (deoxynucleoside triphosphates) in plentiful supply; and
• a small amount of each fluorescently labelled ddNTP, each base coloured differently (commonly A green, C blue, G black, T red).

The polymerase extends the primer along the template, sometimes adding a normal dNTP and continuing, and sometimes, rarely and at random, adding a fluorescent ddNTP and stopping. The result is a population of fragments of every possible length, each terminated by a fluorescent base.

Run through a capillary electrophoresis instrument, the fragments separate by length, shortest first; each passes a laser that excites the fluorescent terminator, and a detector records its colour. Signal plotted against length gives a chromatogram, a series of coloured peaks, one per base, read off as the sequence.

Read length on modern instruments reaches about 700 bp (base pairs), more than enough to cover the resistance-bearing parts of pol. The detection threshold is about 20%: a variant present in under a fifth of the population will not show clearly above the wild-type signal. Mixed populations appear as overlapping peaks at the same position, flagged in the report with mixture codes (for example R for A or G, Y for C or T).

Next-generation sequencing

Next-generation sequencing, also called deep sequencing, reads millions of DNA fragments in parallel. The technical detail varies by platform (Illumina, Ion Torrent and others), but the clinically relevant feature is depth of coverage. Sanger reads each position once, from the consensus of all the template molecules in the reaction; NGS reads each position thousands of times, each read from an individual template molecule, so minority variants present in as little as 1%, sometimes lower, appear as discrete reads.

NGS is changing what a resistance test can be asked. Minority drug-resistance mutations, at frequencies Sanger would miss, are increasingly recognised as clinically relevant for the low-barrier classes, particularly the NNRTIs. Costs have fallen sharply and reference centres are moving to NGS as routine; Sanger remains the standard in most clinical laboratories, including the South African public sector, though the direction of travel is clear.

The Stanford HIV Drug Resistance Database (HIVdb)

A clinical resistance test produces a sequence, and that sequence must be turned into an interpretation. The dominant tool for this worldwide is the Stanford HIV Drug Resistance Database, maintained by Robert Shafer’s group at Stanford.

What it does. A sequence is submitted in FASTA format, plain text beginning with a > header followed by the nucleotide sequence of the patient’s pol gene. The Stanford algorithm aligns the sequence to a reference (HXB2), identifies mutations, and for each drug returns one of five susceptibility levels:

• Susceptible: no resistance.
• Potential low-level resistance: mutations of uncertain or mild effect.
• Low-level resistance: a modest reduction in susceptibility; the drug may still contribute.
• Intermediate resistance: a clear reduction; the drug is likely to fail if used alone.
• High-level resistance: full loss of activity; the drug should not be used.

How it works. The algorithm uses an additive Mutation Penalty Score. Every recognised drug-resistance mutation carries a penalty for each drug it affects, derived from clinical and laboratory evidence; the penalties for all mutations in the sequence are summed per drug, and the total maps to one of the five levels. The weights are grounded in the IAS-USA Drug Resistance Mutation List, an expert-consensus reference updated annually.

Why it matters. Stanford HIVdb is free, online, fast and used worldwide, and the resistance interpretation on a laboratory report almost certainly relies on it, directly or indirectly. Beyond interpreting a sequence, the site carries mutation lists by class, a curated literature database, an annotated comment archive, and tools for primer-ID NGS analysis, at hivdb.stanford.edu.

Other interpretation and surveillance resources:

• IAS-USA Drug Resistance Mutations List (iasusa.org): the canonical expert-consensus reference.
• ANRS Geno2pheno (geno2pheno.org): an alternative European interpretation tool.
• WHO HIVResNet: national surveillance reports on transmitted and acquired drug resistance.

Reading a resistance report

A typical South African resistance report contains three elements:

1. The mutations identified: major and minor (or accessory), listed by class and viral protein (RT, protease, integrase).
2. The drug-by-drug susceptibility predictions: one of the five Stanford levels for each drug.
3. A clinical comment: a brief interpretation and regimen suggestion from the reporting virologist.

Interpreting one turns on four questions.

• What in the next regimen is fully active? Ideally at least two, preferably three, active agents.
• What is the genetic barrier of the new regimen? A regimen anchored by a high-barrier agent (dolutegravir, darunavir/r, bictegravir) can tolerate one suboptimal partner; a regimen of low-barrier agents needs three active drugs.
• What can be recycled? M184V is the classic example: 3TC is kept because its fitness cost helps suppress the virus and re-sensitises it to zidovudine, with a modest effect on tenofovir. PIs without major mutations are usually still active.
• What does the history suggest is archived? A patient with previous NNRTI exposure but no current NNRTI mutations may still harbour archived K103N; a sequence-negative result is not archive-negative.

The cardinal rule of regimen change is to never add a single active drug to a failing regimen: that is functional monotherapy and selects for resistance to the added drug.

Key resistance mutations

The following mutations are the most relevant to current South African clinical practice. Agents used internationally but not yet in South African practice (bictegravir, doravirine, fostemsavir, ibalizumab) and the full every-position lists are held in Stanford HIVdb. Each row gives the class, mutation, drugs affected, mechanism and clinical significance.

Class	Mutation	Drugs affected	Mechanism	Clinical significance
NRTI	M184V/I	3TC, FTC	Discrimination at the YMDD motif, giving very high-level 3TC/FTC resistance	Re-sensitises to zidovudine, modestly to tenofovir; reduces viral fitness; keep 3TC in the regimen
NRTI	K65R	TDF (mainly), ABC	Discrimination; only about two-fold TDF loss but the commonest TDF-failure mutation	More common in subtype C than B (favoured by an upstream poly-A run); increases zidovudine susceptibility
NRTI	TAMs: M41L, D67N, K70R, L210W, T215Y/F, K219Q/E	All NRTIs (excision)	Phosphorolytic removal of the incorporated NRTI from the nascent DNA chain	Accumulate stepwise; Type-1 TAMs (41/210/215) hit TDF and ABC harder than Type-2 (67/70/215F/219); a legacy of the zidovudine/stavudine era, still relevant to second-line interpretation
NNRTI	K103N	NVP, EFV	Allosteric; disrupts the NNRTI binding pocket	The commonest transmitted NNRTI mutation in South African surveillance
NNRTI	Y181C/I/V	NVP (high), EFV (modest)	Allosteric	Often emerges with K103N
NNRTI	G190A/S	NVP, EFV	Allosteric	A common EFV-failure mutation
NNRTI	V106A/M	NVP, EFV	Allosteric	V106M is favoured in subtype C (one base change in C versus two in B)
PI	L90M, M46I/L, I84V, V82A	ATV/r, LPV/r, DRV/r (stepwise)	Substrate-cleft and flap mutations	Accumulate stepwise; multiple mutations needed for full PI failure
PI	I50L vs I50V	I50L to ATV; I50V to DRV, LPV	Active-site reshaping (in opposite directions)	I50L is ATV-specific and leaves DRV/LPV active; I50V hits DRV/LPV but spares ATV
INSTI	Q148H/R/K + secondaries (G140S/A, E138K, L74I/M, E92Q, T97A, N155H)	RAL, EVG (fully); DTG with two or more secondaries	Active-site reshaping	The high-risk pathway for DTG-based failure; Q148K plus two or more secondaries compromises DTG even at double dose
INSTI	N155H (± E92Q)	RAL, EVG; DTG less affected	Active-site reshaping	The second main RAL/EVG-failure pathway
INSTI	R263K	DTG	Active-site reshaping	Rare in clinical TLD failure; carries a fitness cost; the main DTG-monotherapy escape mutation
Capsid	M66I (dominant), with L56I, Q67H, K70N/H, N74D/S, T107N	Lenacapavir	Capsid hexamer interface disruption	M66I dominates emergent failure (CAPELLA, CALIBRATE); N74D was the Purpose 2 PrEP-breakthrough signature; capsid genotyping is not part of standard SA pol testing and needs a dedicated capsid assay

Local resistance patterns

South African resistance epidemiology has several features worth knowing.

Subtype C predominates. Most published resistance data, and the algorithms built on them, are anchored on subtype B virus. Subtype-C virus shows a few characteristic differences in which mutations arise, because codon usage influences how readily a given substitution is made:

• K65R emerges more readily in subtype C than B at tenofovir failure (an upstream poly-A run favours it).
• V106M is the favoured NNRTI mutation in subtype C at efavirenz failure (one base change in C versus two in B).
• G190S is the favoured G190 substitution in subtype A.

These differences are subtle: drug efficacy is equivalent across subtypes, and Stanford HIVdb handles all subtypes well.

Pre-treatment NNRTI resistance has grown globally over the past decade, with prevalence above 10% now documented in several South African surveys. This is one reason the shift from efavirenz-based to dolutegravir-based first-line therapy matters: dolutegravir’s much higher genetic barrier makes pre-existing NNRTI resistance irrelevant to first-line success.

Treatment-emergent INSTI resistance on TLD is uncommon but rising in southern African post-marketing surveillance. Emergent dolutegravir resistance is now detected in programmatic cohorts, particularly in patients switched to TLD while viraemic and in paediatric populations, and most often after long periods of very poor adherence. The reflex dolutegravir drug-level testing workflow is partly an attempt to separate these patients from the much larger group of non-adherent patients who would otherwise be flagged for a genotype that returns wild-type.

Long-acting PrEP and the capsid question is emerging. HPTN 083 and 084 documented integrase resistance in cabotegravir-PrEP breakthroughs; in the lenacapavir trials (CAPELLA, CALIBRATE), M66I is the dominant emergent capsid mutation, with N74D, the Purpose 2 PrEP-breakthrough signature, also reported. From the ATLAS-2M analysis, the risk of cabotegravir-LA virological failure concentrates in patients combining archived rilpivirine resistance, subtype A6 virus, and a body mass index of 30 kg/m² or higher, with any two of the three conferring high failure risk. As long-acting agents scale up, the South African resistance laboratory will need to add the capsid region to its standard pol genotype, a capability now being developed.

Managing a detectable viral load

A detectable viral load on ART is not always virological failure; three patterns must be distinguished.

A viral blip is a single transient detectable viral load, usually between 50 and 199 copies/mL, that returns to undetectable on the next measurement. Blips occur in 10 to 30% of suppressed patients on long-term ART and are not, on their own, associated with virological failure or resistance. They probably reflect a combination of random assay noise, transient release from the reservoir, and intercurrent illness or vaccination. The response is to repeat the viral load, almost always undetectable, and continue the regimen.

Persistent low-level viraemia is a viral load between 50 and 999 copies/mL on serial measurement. It carries an increased risk of subsequent virological failure and is an indication for adherence assessment and intervention, with a lower threshold for drug-resistance testing than in suppressed patients.

Virological failure is two consecutive viral loads at or above 1,000 copies/mL after at least nine months on a stable regimen, with documented adherence support. It triggers the reflex dolutegravir drug-level testing workflow in current South African practice:

1. Two viral loads at or above 1,000 copies/mL, and the next sample is drawn in EDTA (ethylenediaminetetraacetic acid) tubes for both viral load and drug-resistance testing.
2. The laboratory performs a reflex dolutegravir drug level (DLT) on the drug-resistance-test specimen.
3. If dolutegravir is undetectable, the result is reported as non-adherence, no genotype is performed, and the patient returns for enhanced adherence counselling.
4. If dolutegravir is detectable, the laboratory proceeds to a full HIV drug-resistance test (reverse transcriptase, protease and integrase regions).
5. The clinician interprets the result against the patient’s history and refers to the ARV Drug Resistance Committee (ADReC) for third-line authorisation if indicated.

The governing principle is to investigate adherence first. Dolutegravir virological failure with detectable drug levels and proven adherence is unusual; when it does occur, the genotype is informative and the regimen change is meaningful.

Where to go for the detail

For the full mutation lists, complete interpretation rules and latest data, the principal external resources are:

• Stanford HIV Drug Resistance Database (hivdb.stanford.edu): submit sequences, browse mutation tables by class, and read the rationale for each penalty.
• IAS-USA Drug Resistance Mutations List (iasusa.org): the annually updated expert-consensus reference.
• ANRS Geno2pheno (geno2pheno.org): the European alternative interpretation tool.
• WHO HIV Resistance Network (HIVResNet): surveillance reports on transmitted and acquired drug resistance.
• Southern African HIV Clinicians Society: webinars and review articles for the South African clinical perspective.

References and recommended reading

• Clutter DS, Jordan MR, Bertagnolio S, Shafer RW. HIV-1 drug resistance and resistance testing. Infection, Genetics and Evolution 2016;46:292 to 307. A comprehensive review of mechanisms, testing methods, mutations and surveillance. It predates first-line dolutegravir at scale and the long-acting era (cabotegravir-LA, lenacapavir, bictegravir, doravirine); those agents are supplemented from current knowledge.
• Wensing AM, Calvez V, Ceccherini-Silberstein F, et al. 2025 Update of the Drug Resistance Mutations in HIV-1. Topics in Antiviral Medicine 2025;33(2). The IAS-USA mutation panel: the canonical, annually updated reference, now formally covering capsid inhibitors (lenacapavir), attachment inhibitors (fostemsavir) and post-attachment inhibitors (ibalizumab) alongside the established classes.
• Stanford HIV Drug Resistance Database (Shafer group). hivdb.stanford.edu. The interpretation tool of choice worldwide and the source of the five-level susceptibility framework.