The squirming facts exceed the squamous mind,
If one may say so...
-- Connoisseur of Chaos, Wallace Stevens
Sorting winners from losers has never been easy in Québec, starting with the storied battle on the Plains of Abraham. Canadian schoolchildren can recite the details of that 1759 tussle:
British general James Wolfe sweet-talked 385 soldiers into scaling the 53-meter precipice that soars from the St. Lawrence River to the Plains of Abraham. Heading the French forces, General Louis-Joseph de Montcalm left only 100 defenders at the cliff's verge, certain that Wolfe would never essay such a daring dodge.1 But the Brits got a foothold on the Plains, sent another 5,000 troops shinnying up behind, and routed the French.
The fracas marked the end of French dominion in North America. Yet who really won remains hard to say. Montcalm succumbed to grievous wounds, as did Wolfe. In the Treaty of Paris, France traded the sere continental wastes of "New France" for the lush but tiny Guadaloupe. The British queen's head now graces Canada's coins. But for some reason they still speak French on the streets of Québec.
English was the official language of the XIV International HIV Drug Resistance Workshop, but the site was the Château Frontenac on rue des Carrieres in Québec Ville. And the workshop pivot proved familiar -- sorting winners from losers in puzzling out resistance to antivirals.
One would think -- after scaling the slick cliffs of resistance for 15 years -- High Science would have a purchase on the grassy Plains of Sapience. We have had genotyping and phenotyping (vanilla and virtual), fuzzy logic and flexible semiparametrics, inhibitory quotients (plain, genotypic, virtual, and normalized), neural networks, machine learning, causal inference, and case-based learning. We have had algorithms ad hoc, ad lib, ad rem, and ad unguem. We have had controlled trials.
But what have they taught about reckoning resistance? What do they tell about planning the next regimen?
If a writer groping for some sizzle to start yet another resistance article posed these questions, one might read on insouciantly. But these questions came from the very clerisy of resistance research -- deep thinkers like John Mellors (University of Pittsburgh), keen clinicians like Andrew Zolopa (Stanford University).
After sitting through 10 minutes of Zolopa's studied review of resistance scores and cutoff decoctions, the rarely reticent Mellors rose. At best, he proposed, these teeming tries to interpret resistance have yielded only incremental progress -- for those brave enough to breast the consequent "quagmire of data."
Zolopa, whose pithy gloss on expert resistance analysis bared both its muscle and its dystrophy,2 went further. Accreting interpretive systems have not edged knowledge forward, he feared. They have made him more confused.
This is a scary confession from one of the field's brightest guys -- someone you might call to shepherd you through a brambly genotype. Zolopa's lament suggests less that he is losing ground than that less savvy medics underrate the odds they face.
One can cipher those odds' length from many a recent proposal. Take, for example, Boehringer Ingelheim's advice that clinicians thumb a list of 21 mutations at 16 protease positions when sizing up potential resistance to tipranavir (TPV) (see abstract 27).3 Some scoff at the suggestion. But with eight other protease inhibitors (PIs) already on the market -- and a world of cross-resistant genotypes at large -- a 21-item list may end up seeming mercifully brief.
Attempts to hew clean paths through genotypic thickets leave many unimpressed. Brendan Larder (HIV Resistance Response Database Initiative, Cambridge, UK), who with others discovered thymidine analog mutations (TAMs), scoffed at simplifying maps of TAM-1 and TAM-2 paths. Deenan Pillay (Health Protection Agency, London), owner of a most penetrating mind, judged such paths not simplifying but simplistic.
Workshop attempts to divine the clinical import of resistance to boosted PIs -- or to familiar nucleoside/nucleotide reverse transcriptase inhibitors (NRTIs) -- impressed with their cunning but left attendees wondering about their worth.
Then there's the K65R saga -- now approaching Homeric proportions even though most research suggests it doesn't really arise much.
Of course this 14th edition of the fabled Resistance Workshop did more than sow doubt, brew discord, or divine dead ends. Two hundred attendees learned that:
Everyone in this analysis stuck with their assigned regimen for 24 weeks. Coakley defined the clinical cutoff as the lowest fold-change in viral susceptibility to ATV above and below which he discerned statistically distinguishable differences in viral control rates.
The 043 trial cross-exam involved 131 people who started unboosted ATV at a mean ATV fold change of 2.9. These people had tried PIs for a median 2.6 years, but only 36% were taking a PI when they began unboosted ATV. About one third had four or more major protease mutations. At the 24-week mark, 83 (66%) had a viral load under 400 copies/mL.
Using Fisher's exact test to fish for clinically relevant ATV cutoffs in the 1.6- to 5.1-fold range, Coakley settled on 2.2 as the sharpest cut point. Among people with a baseline fold change below 2.2, 76% had a sub-400 load at week 24. Among those whose baseline fold change measured 2.2 or more, 45% met the week-24 response criterion.
The 045 trial inquest probed for cutoffs in 111 people who started ATV/RTV with an average ATV fold change of 4.1. This group had tried a median of two regimens and logged 2.6 years of PI experience. Yet they had only a median of two main PI mutations, and only 44% had three or more mutations from the Stanford database of 16 ATV- or LPV-linked mutations.
Taking the same statistical approach as in the 043 analysis, Coakley homed in on a cutoff of 5.2. Among people whose susceptibility to ATV lay below 5.2 on enrollment, 77% had a viral load below 400 copies/mL at week 24, compared with 12% with a baseline cutoff of 5.2 or more. Coakley prudently cited three caveats to this analysis:
Combined, these factors may have lowered the clinical cutoff for ATV/RTV, and somewhat less for ATV alone. So the 2.2- and 5.2-fold thresholds can be considered conservative estimates.
With Swiss and French coworkers, Anne-Genevieve Marcelin (Geneva University Hospital) took a different path to truffle for ATV response predictors in PI-experienced people -- a genotypic resistance score [abstract 7]. Her heuristic began with 62 people who switched to ATV/RTV with a viral load above 1,000 copies/mL because of virologic failure (84%) or intolerance (16%).
The group had tried a median of two PIs and had a median baseline load of 4.3 log copies/mL (about 20,000 copies/mL). A hefty proportion -- 69% -- took TDF with ATV/RTV. After three months on their new regimen, 82% had at least a 1-log (10-fold) drop in viral load.
Testing mutations with a prevalence above 8%, Marcelin gleaned 25 substitutions at 13 positions that dented response to ATV/RTV: 10F/I/V, 16E, 33I/F/V, 46I/L, 54L/V/M/T, 60E, 62V, 71I/T/V/L, 82A/T, 84V, 85V, 90M, and 93L. The strongest statistical tie to poor response involved mutations at seven sites -- 10F/I/V, 16E, 33I/F/V, 46I/L, 60E, 84V, and 85V (P = 8.04 x 10-9, which means 804 lies eight zeros away from the decimal point). Notably, no genotypic roadmap to ATV resistance lists mutation milestones at positions 16, 60, and 85. The 14 mutations in the resistance score authored by Bristol-Myers Squibb did not predict this response better than Marcelin's septet.
One-log response rates tailed off sharply in people who piled up two, three, or more of these seven treacherous mutants:
The number of other active drugs in the regimen also predicted a 1-log response (P = 0.0012), as did the number of previous PIs (P = 0.04). In people with three or more mutations from Marcelin's menacing seven, none with no other active drugs responded, 29% with one active drug responded, and 60% with two or three active drugs responded (P = 0.024).
Inhibitory quotients -- which factor measures of resistance and drug levels -- remain a much-mulled metric of virologic response. Isabelle Pellegrin (Bordeaux University Hospital, France) brandished a genotypic inhibitory quotient (GIQ) -- which uses a mutation set for the resistance measure -- to plot response to ATV/RTV in 90 people who had tried a PI (91%) and/or a nonnucleoside reverse transcriptase inhibitor (NNRTI) (62%) before beginning ATV/RTV with a median RNA load of 3.8 log copies/mL (about 6,300 copies/mL) [abstract 8]. They had taken a median of five earlier regimens.
After six months of boosted RTV, the median load dipped 1.2 log copies/mL, and 76% of the study group had fewer than 400 copies/mL. Pellegrin counted 66% as virologic responders with a six-month viral load below 50 copies/mL.
Using the 14-position ATV/RTV resistance score of the French national AIDS trial group (Agence Nationale de Recherches sur le SIDA [ANRS]), Pellegrin found that people with fewer than six of these mutations trimmed their viral load by a median 1.7 log copies/mL, while the median load crept up 0.01 log copies/mL in people with six or more mutations (P = 0.01). While 75% with fewer than six mutations qualified as responders, only 20% with six or more mutations met the response criterion (P = 0.002). The six-mutation cutoff predicted a six-month response with a specificity of 95% but a sensitivity of only 35%.
Univariate analysis sorted out two mutations not on the ANRS list that foretold a six-month load still above 400 copies/mL -- I62V (P<0.05) and V77I (P<0.2). At the same time none of Pellegrin's nonresponders had the ANRS mutations at protease site 36, 63, or 71. And the potential resistance hot spots Marcelin proposed -- at positions 16, 60, and 85 -- did not rank in Pellegrin's analysis.
Pellegrin figured GIQ as ATV trough divided by the number of mutations in the ANRS inventory. By this calculus 88% of people with a GIQ above 0.33 had a sub-50-copy load at six months versus 50% of those with a GIQ under 0.13 (P = 0.03). These GIQ limits predicted response with 91% sensitivity but only 36% specificity.
Atazanavir trough, peak, or area under the curve did not correlate with six-month response. But 80% of people with a mutation score below six and a trough above 0.7 µg/mL had a six-month load under 50 copies/mL compared with 47% whose trough lay below 0.2 µg/mL.
What can clinicians make of these diverse response divinings? Or need they fuss about them at all? Atazanavir/RTV has emerged as a favored first-line option, noted Douglas Richman (University of California, San Diego). And research retailed at the workshop (see abstract 29 below) and earlier suggests it lacks the punch of LPV/RTV in people with PI experience. So these doughty efforts to define response cutoffs may seem wasted effort.
Even among clinicians considering an ATV/RTV rescue regimen, suggested Jonathan Schapiro (National Hemophilia Center, Tel Aviv), cutoffs conjured so far may find few takers. Because these three studies -- and in fact all efforts of this ilk -- rest on arbitrary response definers, the resulting cutoffs retain a random hue. Different definitions of success, Schapiro proposed, would yield different cutoffs.
ViroLogic's Coakley did not dispute this assessment. Whenever research tries to nail down response thresholds, some caprice creeps in. Nothing makes this point more clearly than the largish proportion of people above or below discovered cutoffs who wrong-headedly respond or fail to. In Coakley's study 23% with baseline susceptibility to ATV under the desired 5.2-fold cutoff did not notch a 24-week load below 400 copies/mL. The dismal specificity in Pellegrin's GIQ reckonings, and the flaccid sensitivity in her six-mutation score, do not inspire confidence.
Such uncertainties moved the University of Pittsburgh's Mellors to urge defining a continuum of probability for salvage responses, rather than defining cutoffs. Cut points, he maintained, "are grossly misunderstood by clinicians." But if clinicians find cut points tricky, others observed, they will find probability curves opaque.
The population studied represents another fraught variable in sleuthing down response predictors. Daniel Kuritzkes (Brigham and Women's Hospital, Boston) argued, for example, that the moderately resistant groups Coakley analyzed may not be the best grist for multivariate mills. Differences between Marcelin's and Pellegrin's cohorts could explain why certain novel mutation candidates popped up in one study but not the other.
Of course the vagaries these studies spotlight in guesstimating success apply not only to ATV/RTV but to any antiretroviral prescribed for people with resistant virus.
After 48 weeks no one with five or more protease mutations responded to ATV, the FDA's Lisa Naeger reported, in an analysis using the agency's time to loss of virologic response (TLOVR) criterion at the 400-copy threshold. But 28% with five or more mutations responded to LPV/RTV. Response rates proved similar in the two groups among people who began the boosted PIs with fewer than five mutations.
Certain mutation clusters undermined ATV/RTV more than others:
Baseline fold change in susceptibility also discriminated between responses to these two boosted PIs. People with less than a five-fold change in susceptibility to either ATV or RTV had equivalent responses to the two drugs. But people with more than five-fold resistance at baseline did better with LPV/RTV.
To sum up the ATV reports (on a sour note), three studies fashioned three discouragingly distinct sets of mutations that undercut this PI's antiviral vigor (Table 1). Yet at least two of three research teams saw trouble with mutations at five positions, marked in bold type. Substitutions at site 46 signal sure peril for ATV/RTV. When joined by other mutations, these studies and others suggest, the L90M flip may be the king of PI cross-resistance.
The cohort started 700/100 mg of FPV/RTV twice daily with a median of three major PI mutations (interquartile range [IQR] 0 to 4) and a median FPV/RTV score of 4 (IQR 1 to 5). The median viral load stood at 4.8 log copies/mL (about 63,000 copies/mL), and the median CD4 nadir lay at 100 cells/mm3. After 12 weeks of FPV/RTV 25% had a viral load below 50 copies/mL, 37% had fewer than 400 copies/mL, and 33% had at least a 1-log viral load dip.
Mutations that correlated with 12-week response did not mirror the ANRS list (Table 2). Yet these discrepancies at least partly reflect the study design, which banned people with six or more of the ANRS mutations.
In a univariate analysis Pellegrin and colleagues found that nearly any venomous variable one can imagine predicted a poor 12-week response to FPV/RTV -- number of previous regimens; viral load zenith; earlier treatment with PIs and NNRTIs; number of NRTI mutations; number of minor, major, or total PI mutations; four or more mutations from the newly defined octet; FPV trough, peak, or 12-hour exposure; and low GIQ (trough/number of FPV/RTV mutations).
Pellegrin did not offer a multivariate analysis, but she proposed three cutoffs for a 12-week response:
The sensitivity and specificity quotients suggest a bare mutation count foretells success better than FPV level alone or FPV level divided by mutation score (the GIQ). Results like these underlie some researchers' sense that GIQ has little use in the clinic.
Valdez began with a bulky data set rolling together people who took TPV/RTV as their only PI in four trials -- the two phase 3 RESIST studies of enrollees with triple-class resistance in whom two PIs had flopped but with two or fewer mutations at positions 33, 82, 84, and 90; a companion trial for people with more than two mutations at those sites; and a phase 2 dose-finding study with no entry limits involving positions 33, 82, 84, or 90.
This analysis aimed to test Boehringer Ingelheim's busy mutation list of 21 substitutions at 16 positions -- 10V, 13V, 20M/R/V, 33F, 35G, 36I, 43T, 46L, 47V, 54A/M/V, 58E, 69K, 74P, 82L/T, 83D, and 84V. But along the way Valdez turned up evidence that baseline fold change in susceptibility to TPV predicted at least a half-log nip off viral loads during the RESIST trials' first eight weeks.
Among 252 people with less than threefold resistance at baseline, 233 (92%) met the modest half-log response benchmark. Yet a hefty proportion with higher baseline resistance -- 74 of 109 (68%) -- also enjoyed a half-log viral load ebb. The significantly better response in the group with lower baseline resistance (P<0.001) did not depend on cotreatment with the fusion inhibitor enfuvirtide (ENF).
Using the same eight-week half-log response marker, Valdez confirmed that more mutations from Boehringer Ingelheim's laundry list chip away at chances of viral control:
People with up to three of the 21 menacing mutants managed a reasonable response after 24 weeks of TPV/RTV (Table 3). The response rate slipped sharply when four or more listed mutations accrued. People with eight or more of the 21 mutations had essentially no response.
Boehringer Ingelheim's 21-mutation-long agglomeration steamed even resistance mavens with steel-trap minds and typify the "data quagmire" Mellors complained about. Many note that the list omits L90M, a founding member of Boehringer Ingelheim's "universal" quartet of minatory mutants. Douglas Mayers, the company's viral chief, reasonably concurs that L90M thwarts TPV as surely as it does other PIs; it merely failed to emerge in the statistical analysis that crowned the 21. Everyone yearns for further work that may narrow the lineup of usual suspects in resistance to this PI. But the dreaded 21 may yet prove a bellwether of 21st-century resistance reckonings.
The analyses involved treatment-experienced people from nine databases in North America and Europe -- 583 people with a pre-ABC viral load of 4.4 log copies/mL (about 25,000 copies/mL) and an average 1.6-log change after eight weeks of an ABC regimen; and 400 people with a pre-ddI viral load of 4.2 log copies/mL (about 16,000 copies/mL) and a mean 1.8-log viral load change after eight weeks of a ddI combo. No one in the ABC analysis used that NRTI before, and no one in the ddI study had tried ddI.
The Forum for Collaborative HIV Research experts devised regression models fixed on eight-week change in viral load, comparing resistant virus with intermediate-resistant or sensitive virus. They adjusted the models for baseline viral load and number of active drugs in the regimen. This eight-week scrutiny will be followed by a 24-week analysis.
Catechizing eight algorithms, Costagliola discovered that predicted mean changes in viral load for ABC-resistant versus -sensitive virus ranged from -0.02 to +0.66. The gap between predicted mean changes in viral load for intermediate-resistant versus sensitive virus yawned almost as widely, from +0.10 to +0.64.
Univariate analysis of the eight algorithms confirmed displeasingly disparate correlations between intermediate or sensitive virus and mean change in viral load. For five of these eight algorithms, Costagliola found no significant correlation between change in viral load and viral susceptibility. This statistical cross-exam did suggest that mutations at positions 70 and 215 would best be dropped from rules for interpreting resistance to ABC.
Algorithm reliability grew gloomier with the six systems used to predict response to ddI. The range for mean change in viral load for sensitive versus resistant virus proved tighter than with ABC (-0.12 versus +0.28). But univariate analysis found no correlations between change in viral load and viral susceptibility for any of the algorithms tested.
Dismayed, Costagliola and colleagues tossed out the whole data set and plugged in numbers from 300 people in the NARVAL resistance test trial.5 They picked the NARVAL data because the trial enrolled people with and without ddI or ABC experience. That meant the study population was more heterogeneous than the original cohorts in baseline resistance to either drug. The researchers reasoned the more diverse NARVAL cohort might prove more fertile ground for validating the algorithms.
With the NARVAL data set the mean change in viral load for resistant versus sensitive virus narrowed considerably (+0.36 to +0.52). And univariate analysis figured significant correlations between viral load change and viral susceptibility for three of the six algorithms.
This brave inquest into algorithmic riddles left some attendees slack-jawed. The University of California, San Diego's Richman, a founding father of HIV resistance research, went so far as to wonder whether clinicians can rely on algorithms at all -- given the gaping predictive variability Costagliola confirmed. She countered that further analysis -- with further expansion of the data set to include more people already treated with ddI or ABC -- may resolve some discrepancies.
Nancy Shulman (Stanford University, Stanford, California) and US AIDS Clinical Trials Group (ACTG) colleagues tackled the troubling question of resistance to ddI by retrospectively pooling data from 444 NRTI-experienced people enrolled in five ACTG trials [abstract 49].
They used data from three trials to indict likely resistance mutations by stepwise regression analysis. Then they probed the merits of their picks by comparing baseline mutations with RNA results in two trials of ddI monotherapy -- ACTG 173 and ACTG 307. They also aimed to figure how much individual mutations contribute to viral susceptibility to ddI.
The resistance score they settled on summed mutations at nine positions: Substitutions at reverse transcriptase sites 43, 67, 74, 75, 184, 210, 215, and 223 contributed to resistance, while a mutation at position 83 lessened resistance. This score correlated strongly with fold-change in susceptibility to ddI (r = 0.67, P<0.0001).
Next Shulman compared the ACTG score with the formula based on four-week viral load change in the Jaguar trial of add-on ddI6 (41L + 67N + 69D + 74V + 215Y/F + 219Q/E - 70R - 184V). The ACTG score with the position 184 mutation (r = 0.43, P = 0.002) or without 184 (r = 0.41, P = 0.005) tracked with virologic response more closely than the Jaguar score (r = 0.25, P = 0.09).
In experiments designed to find out, Parikh reckoned kinetics for single nucleotide incorporation of AZT triphosphate (TP), deoxyadenosine triphosphate (dATP), and tenofovir diphosphate (TFV-DP) in reverse transcriptase of:
Parikh found that K65R heightened selectivity against TFV-DP by decreasing incorporation without affecting binding. TAM 67 reverse transcriptase antagonized K65R by restoring incorporation of TFV-DP. TAM 41 also antagonized K65R -- but less than TAM 67 -- by bridling selectivity for dATP versus TFV-DP. K65R also more than doubled selectivity against AZT-TP [abstract 85].
Can clinicians exploit the mutual antagonism of K65R and TAMs? It seems so, two workshop studies suggested, but how mutant virus manifests its wiles in humans always trumps the tangles that lab work may tease apart. For example, Parikh's paradigm suggests that K65R would have a tough time emerging during failure of a regimen combining AZT and TDF. And that would be a good thing since K65R renders virus less susceptible all NRTIs except AZT.8 Bernard Masquelier (Bordeaux University Hospital, France) put that theory to the test in a nonrandomized trial enrolling 24 treatment-naive people starting AZT/3TC (as Combivir) plus TDF [abstract 20].
This group began their triple nukes with a median viral load of 4.3 log copies/mL (about 20,000 copies/mL) and median CD4 sum of 443 cells/mm3. Masquelier searched for resistance mutations in RNA and proviral DNA before treatment and six months into therapy. Despite this group's naive status, five people had TAM-tainted virus -- four with T215D/C and three with M41L.
Three people stopped treatment or switched to another regimen before month six. An intent-to-treat analysis set the six-month sub-50-copy rate at 63%, and an on-treatment analysis figured an 83% response rate. Masquelier recorded four rebounds from below 50 copies/mL. No new resistance mutations arose in three of these people, but K65R cropped up in one. That meant K65R appeared in 4% of people overall but in 25% of rebounders. So AZT does not hoist a shield impervious to K65R. But the antiviral vigor of this three-NRTI mix against TAM-laden virus bears notice.
In a study outlined in the workshop abstract book but not posted at the meeting, Schlomo Staszewski (JW Goethe University Hospital, Frankfurt) found that AZT can handcuff virus carrying K65R but no TAMs [abstract 17]. He outlined three case reports of people who added only AZT to a failing regimen and kept taking their other drugs. All three smartly pushed their viral load beneath 50 copies/mL, even though their TAM-less virus bore mutations conferring resistance to all other drugs they continued.
K65R and multiple TAMs almost never share the same viral genome. Earlier work by Parikh found K65R and T215Y/F plus two other TAMs rubbing genomic shoulders in only 24 of more than 56,000 viral isolates tested.9 But K65R and TAMs -- with the possible exception of codon 215 changes -- cohabitate quite cheerfully in people with ample NRTI experience. Andrew Nevins (Stanford University, Stanford, California) confirmed that matrimony in a retrospective study of 39 much-treated people whose virus carried K65R [abstract 16].
Sixteen of the 39 (41%) had at least one TAM, and five had three or more. But no one had T215Y/F. K65R appeared linked to familiar TAMs at positions 41, 67, 70, 210, and 219, although only cloning can verify those links. People who also had the M184V mutation and those whose new regimen had a higher genotypic susceptibility score had the best chance of reaching a sub-50 viral load with salvage therapy. But no other factor tested predicted a better salvage response -- including AZT, TDF, or any other specific drug in the salvage regimen, or a new class of drugs in the new regimen. Nevins cautioned the study may be too small to confirm or rule out the impact of such variables.
Does the path taken make a clinical difference? It depends on whom you ask. Separate studies at the 2005 Conference on Retroviruses and Opportunistic Infections hinted that 3TC ties up TAM-2 mutants better than emtricitabine (FTC),10 an edge perhaps also held by d4T over AZT.11 But a TAM-path sortie by the UK Collaborative Group on HIV Drug Resistance found the TAM topography so choked with overlapping paths that mapping only two seemed "simplistic" [abstract 130].
Threshing through 2,379 genotypes of treatment-experienced people, David Dunn (MRC Clinical Trials Unit, London) discovered 116 distinct TAM patterns, none of which looked much like a well-toed path. He traced only 12 patterns that made up more than 2% of the trails plotted.
TAM-1 mutations turned up in 728 tests (31%) and TAM-2 mutations in 685 (29%). Mutations from both TAM bins littered far more genotypes -- 956 (40%). Indeed, the most common single mutation covey -- at a prevalence of 14% -- consisted of three TAM-1 mutants (M41L, L210W, and T215Y) plus one TAM-2 interloper (D67N).
What do you call that mutant array, wondered TAM trailblazer Brendan Larder. More than 10 years of clinical work show it's a nasty virus -- and the Collaborative Group survey confirmed it's a favored route. But you can't call it TAM-1 or TAM-2. Kuritzkes of Brigham and Women's Hospital, on the other hand, still sees some merit in the TAM-path distinction, though he thinks it remains unvalidated.
A few studies at the Resistance Workshop offered contradictory answers.
Julie Strizki (Schering-Plough) had good news and bad news about two CCR5 compounds labeled SCH 351125 and SCH 417690 (SCH-D or vicriviroc) [abstract 59]. The good news is that resistance to these agents evolves at a plodding pace -- in the lab. The bad news came in two flavors: Virus resistant to Schering's antagonists proved cross-resistant to other offerings in this class, and one resistant species appeared to favor CXCR4 as readily as CCR5.
Serial passage studies with the Schering-Plough antagonists showed that resistance emerged slowly as mutations in gp120 accrued. Resistant virus did not surface at all for up to 16 weeks in some passage studies. Clonal analysis traced most mutations to gp120's V3 loop, but Strizki could find no V3 mutations in some resistant virus. That suggested undiscovered mutations outside gp120 may also hamstring the Schering-Plough drugs.
HIV resistant to the SCH hopefuls remained susceptible to AZT, FTC, and indinavir (IDV). But Strizki confirmed their cross-resistance to other Schering CCR5 antagonist compounds and to a Merck CCR5 drug. She also found that the CXCR4 inhibitor AMD3100 did not stifle replication of virus resistant to Schering-Plough's CCR5 antagonists -- good evidence that these mutants did not swap coreceptor preference and begin homing to CXCR4. But certain escape mutants grown in X4 cells retained their R5-linking talents, a dismaying development suggesting dual (X4 + R5) tropism.
Unlike Strizki, Pfizer's Mike Westby saw no signs of cross-resistance in studies of maraviroc, the CCR5 plugger formerly tagged UK-427,857 [abstract 65]. He cooked up four clones from maraviroc-resistant virus and confirmed that they retained resistance to the Pfizer drug. Then he tested the clones' susceptibility to three CCR5 antagonists -- Schering's SCH-C and SCH-D (vicriviroc) and GlaxoSmithKline's GW873140 -- and to the fusion inhibitor ENF and the PI saquinavir (SQV). All four clones remained sensitive to these drugs with less than a fivefold gain in 50% inhibitory concentration. This retained susceptibility, Westby observed, indicates these mutants also retain their CCR5 tropism.
Molecular modeling by the Pfizer team showed that maraviroc, SCH-C, SCH-D, and GW873140 nestle into the same CCR5 pocket. But they sit there in slightly different postures. Westby proposed these subtle yaws in binding conformation explain the lack of cross-resistance. But in a workshop colloquy on CCR5 research, he cautioned that his findings do not rule out cross-resistant blossomings in clinical trials.
Painstaking experiments by Donald Mosier (Scripps Research Institute, La Jolla, California) may soothe some jumpy nerves among CCR5 drug developers [abstract 60]. His results suggested that fears of receptor switching as a resistance gambit against CCR5 antagonists "may be overstated."
Mosier used site-directed mutagenesis to mint all 32 possible "mutation intermediates" in the sesquipedalian sequence leading from R5 to X4 virus. Then he sized up these mutants to see how well -- or poorly -- they invaded CCR5- and CXCR4-expressing target cells and how sensitive they were to the CCR5 inhibitor PSC-RANTES.
Figuring that the 32 mutants offered 120 viable routes from R5 to X4 tropism, Mosier found only four (3%) of those routes "operational." Yet given HIV's polymorphous perversity, four tracks from R5 to X4 may be four too many.
GlaxoSmithKline, a third aspirant in the dash to market a CCR5 meddler, sent Kathryn Kitrinos to the workshop with another mote of evidence arguing against resistance by coreceptor switching [abstract 61]. Her work focused on a person who took the lowest dose of GW873140 -- 200 mg once daily -- in a dose-ranging study of the drug.
Samples gleaned at screening and on treatment day one contained virus with an exclusive preference for the CCR5 coreceptor. But virus collected on treatment day 10 readily snagged either CCR5 or CXCR4. On the 24th and last treatment day, this person's virus seemed to shed its X4 tropism and again home solely to CCR5.
But as all resistance savants now know, standard assays see only those species that make up about 20% or more of a person's viral pool. Deploying a more discriminating assay to size up virus cloned from these samples, Kitrinos confirmed dual tropism for X4 and R5 in one of 23 clones (4%) as therapy began, six of 22 clones (27%) on treatment day 10 (when the standard assay spotted dual tropism), and two of 24 clones (8%) on day 24.
In other words this person's viral horde had a thin sprinkling of CXCR4-tropic species before treatment with GW873140 ever began. That tiny population burgeoned to detectable levels during the first week of therapy, then waned to its original pennyweight proportions. Comparing gp160 sequences of sampled virus by phylogenetic analysis, Kitrinos bolstered her contention that dual-tropic variants arose from virus lurking in scant pretreatment caches.
This careful and convincing analysis must be read in the right context. Kitrinos proved GW873140 monotherapy did not drive CCR5-tropic virus into dual tropism -- a trait that may spur faster progression by doubling HIV's binding options. The X4/R5 virus remained fully susceptible to the CCR5 antagonist in cells expressing only CCR5 coreceptors.
But these results apply to one person taking a low dose of GW873140 unlikely to see clinical use. And even though monotherapy did not engender resistance to the antagonist by selecting X4-favoring mutations, it did (briefly) give wing to a recondite X4 flock. Whether that happens with some frequency when CCR5 antagonists reach the clinical arena -- and whether such outgrowth puts progression on a faster flyway -- are questions that can't be answered now.
That estimate came from Davey Smith (University of California, San Diego), an earnest young man who did not blush at his bold report (detailed below). Like most researchers in this field, Smith must be inured to mankind's Dionysian mold, the progenerative drive that ensures survival of the species -- human and viral. And now that imperfect antiretrovirals taken by impassioned people guarantee resistance, the risk of transmitted mutants imperils sex partners and needle sharers alike.
Rigging an epidemiologic model to factor treatment uptake and response, Andrew Leigh Brown (University of Edinburgh) analyzed prevalence of primary infection with resistant virus and transmission of drug-resistant HIV in a San Diego cohort [abstract 115]. The model predicted declining rates of primary resistance only if fitness of transmitted virus drops over time. If fitness of resistant virus stays steady, rates of infection with resistant virus will billow "continuously." Without sharp-eyed screening for transmitted resistance, Leigh Brown warned, "uncontrolled primary resistance and treatment failure" will ensue.
And freshly transmitted resistant virus ends up not only in semen, but also in long-lived cells, according to research by Jade Ghosn (Pitié-Salpêtrière Hospital, Paris) [abstract 161]. With PRIMO cohort colleagues, she figured that resistant strains picked up in primary infection "massively fuel" peripheral blood mononuclear cells (PBMCs).
The study involved 518 people with primary HIV infection, 44 of whom (8.5%) got handed resistant virus. A median 46.5 days elapsed between the estimated date of infection and inclusion in the cohort. In plasma samples Ghosn unmasked NRTI resistance mutations in 27, NNRTI mutations in 13, and PI mutations in 23. Of the 27 people with NRTI mutations in plasma, 21 had exactly the same mutations already hunkered down as DNA in PBMCs. Of the 13 with NNRTI mutants in plasma, 12 had identical sets in PBMCs. And of the 23 with PI mutations in plasma, 18 had the selfsame mutations in DNA.
During at least one study visit, 29% reported unprotected anal or vaginal sex with a person without HIV or with an unknown HIV status; 6% owned up to unsafe sex with an uninfected or serostatus-unknown partner during more than 75% of visits. Multivariate analysis pinpointed five predictors of unsafe sex with such a partner in the preceding four months at the following odds ratios (OR) and 95% CIs:
Taking sildenafil and younger age -- but not popping crystal meth -- raised risks of unprotected sex only with HIV-infected partners:
A substantial minority of people, 38%, had mutations conferring resistance to the first three antiretroviral classes, while another 26% had NRTI- and PI-related mutations, 15% NRTI- and NNRTI-educed mutations, and 1% NNRTI- and PI-inspired mutations. But how prevalent is transmission of multidrug-resistant (MDR) virus outside gay hot spots like San Francisco? And how frequent is transmission of MDR virus that can use both coreceptors -- like the virus transmitted in the storied New York case?12
Ron Kagan (Quest Diagnostics, San Juan Capistrano, California) addressed both questions by analyzing more than 153,000 viral sequences from HIV-infected people [abstract 128]. He charted a drop in rates of MDR HIV, defined as virus bearing mutations engendered by four or more NRTIs, four or more PIs, and one or more NNRTIs. Prevalence of such maculate breeds stood at 12% in 1998, then faded to 10% in 2001 and 2002, and slid all the way to 4.6% so far in 2005. Sequence scrutiny tied the New York dual-tropic virus to 12 other sequences traced to four people in this vast database.
Michael Kozal (Yale University, New Haven, Connecticut) and coworkers at the University of Western Ontario used a looser definition of multidrug resistance -- virus resistant to at least one drug in each of the first three classes -- in their study of 393 people (56% male, 79% heterosexual) [abstract 127].
Of the 250 people who had sex during the study, 112 (45%) had unprotected anal, vaginal, or oral sex, exposing 354 of 1,225 sex partners (29%) to HIV. Thirty-nine people practicing risky sex (35%) did so with drug-resistant virus, including 13 (11.6%) with double-class resistance and two (1.8%) with triple-class resistance. People with resistant virus exposed 71 of 354 partners (20%) to their mutations, 29 of them to double-barreled mutants and two of them to triple threats.
Using the same definition of multidrug resistance, Gayatri Jayaraman (Public Health Service of Canada, Ottawa) tracked a low but abiding rate of transmission to newly infected people [abstract 123]. Of 1,738 just-diagnosed, untreated people studied from 1998 through March 2004, more than 8% had at least one resistance-conferring mutation. Jayaraman first spotted multidrug resistance transmission in 1999 at a prevalence of 0.9%. That rate climbed to 1.9% by 2002 and has stayed there since.
Assays that flushed out as little as 0.05% of viral quasispecies boosted the estimate of double-or-greater-class resistance from 12% to 15% and of triple-class resistance from 2% to 4%. And those jumps reflect analysis of only four mutations, including none of the oft-transmitted NNRTI mutations.
Karin Metzner (University of Erlangen-Nuremberg, Germany) used similarly acute assays to smoke out wisps of resistant virus bearing L90M, M184V, or K103N (the NNRTI cutthroat) in 49 untreated acute seroconverters [abstract 110]. She found one or more of these mutations in 10 people (20.4%), all at levels that standard sequencing would miss. One person (2%) had L90M, five (10.2%) had K103N, and six (12.2%) had M184V.
Robert Grant (Gladstone Institute, San Francisco) cautioned that researchers must still validate the accuracy of these super-mutant snoopers.
Jan Gerstoft (Copenhagen University) figured transmission risk by reckoning the proportion of treated people with a viral load above 1,000 copies/mL and thus likely to pass along resistant species when infecting another person [abstract 121]. This comprehensive analysis rests on a nationwide cohort of all 3,722 Danes seeking medical care for HIV from 1995 through 2003. That group included 2,729 taking a potent antiretroviral regimen.
The transmission risk group -- those with a viral load topping 1,000 copies/mL -- peaked at 425 in 1998 (24% of people at that point) and tumbled to 233 (9%) in 2003. The relative change over those years measured 0.81 per year (P = 0.000). A quick chute in PI failures accounted for most of this improvement. The number of people who endured a PI failure crescendoed at 277 (14%) in 1999 and slipped to 152 (5.9%) in 2003 for a relative change of 0.80 per year (P = 0.000). The number of NNRTI failures crested at 119 (4.6%) in 2002, then dipped to 108 (4.2%) by 2003, but this drop was not significant.
Sage antiretroviral policy planning may contribute to this stellar record. Danish clinicians do not advocate treatment interruptions. In 2003 only 6.4% of antiretroviral-treated people were on a drug holiday.
Better responses to antiretrovirals after 2000 may also account for lower recent transmission of resistant species in Québec [abstract 120]. Jean-Pierre Routy (McGill University, Montreal) sorted resistance records of 230 untreated people diagnosed with HIV infection at eight centers from January 1997 through March 2005. Nineteen (8%) had primary drug resistance.
Univariate analysis comparing people infected with resistant and nonresistant virus found no effect of age, gender, ethnicity, education, income, or HIV transmission route. People infected with resistant virus had a higher CD4 count (610 versus 440 cells/mm3, P = 0.009) and a lower viral load (4.12 versus 4.75 log copies/mL, P = 0.003) than those infected with wild-type virus. An earlier year of infection also favored primary resistance in the univariate analysis.
Multivariate analysis singled out two independent predictors of infection with resistant virus. Being infected after 2000 cut the risk of resistance 70%, and a lower viral load trimmed the risk 46%.
The 12% rate of infection with resistant virus in France13 means more than one in 10 people may start a regimen including one or more already enfeebled pills. That's what happened to 35 of 45 people (78%) handed resistant virus by a sex or needle partner, reported Marie-Laure Chaix (René Descartes University, Paris) [abstract 114]. She compared them with 259 people in the same two cohorts -- all of them infected since 1996.
Of the 35 people with resistant virus, 33 had NRTI mutations, 10 had PI mutations, and three had NNRTI mutations. Chaix did not analyze the 10 people infected with resistant virus who started a regimen containing no mutation-threatened drugs. As in Routy's study of Québecois with primary resistance (see abstract 120 above), higher CD4 count and lower viral load favored infection with resistant virus in Chaix's cohort.
After six months of treatment 81% in the resistant group versus 95% in the nonresistant group had a viral load below 400 copies/mL (P = 0.02). Respective proportions with a sub-50 load were 57% and 79% (P = 0.02). At treatment month three people with wild-type virus had an estimated 0.71-log lower viral load than people with resistant virus (P<0.01). When Chaix adjusted the analysis for gender, age, and baseline CD4 count and viral load, people starting treatment without resistant virus had a 0.6-log lower viral load at three months (P<0.01).
A multivariate analysis adjusted for the same variables plus time since HIV infection determined that infection with resistant virus and a higher baseline viral load independently narrowed the odds of reaching a sub-400 load after three months of therapy (Table 4). Chaix saw a trend toward a better three-month response among older people.
Why did so many people infected with resistant virus start treatment with mutant-menaced drugs? Because no one had their virus genotyped before therapy began. A second study by Chaix rated empiric first-line therapy especially risky in people infected with MDR virus [abstract 122]. But three of five people who started treatment without the benefit of genotyping did just fine despite being saddled with multi-resistant virus. And two other people infected with pan-resistant virus maintained a stable disease course for two years despite foregoing therapy.
This analysis started with 797 people infected with HIV since 1996. Ten of them (1.25%) cached virus resistant to at least one drug in the first three classes of antiretrovirals. Among five people who began treatment without resistance testing, Chaix retrospectively found that three had virus resistant to two drugs and two had triple-class resistance.
These five people had a sluggish median one-year CD4 gain of 33 cells/mm3. But three of them pushed their viral load below 50 copies/mL within six months and maintained that response through a median 36 months of follow-up. The other two, however, still had a viral load above 1,000 copies/mL after one and three years of follow-up. Chaix did not distinguish responders from nonresponders on the basis of active drugs in their starting regimen.
Among the other MDR people not lost to follow-up, two did not start therapy. Their CD4 counts stayed above 500 cells/mm3 for two years. Then one count sank to 268 cells/mm3 after 30 months of infection.
That conclusion rests on a study of five gay men by Davey Smith (University of California, San Diego), who edified workshop attendees with the estimate that 16,000 gallons of HIV-laden semen (60,000 liters) get spilled daily [abstract 115]. Smith also figured that 8,000 people will get inseminated with HIV today. To reckon consequences of this Rabelaisian incidence, he tracked mutant virus in semen of five men who got NNRTI-linked mutations when infected by other men.
Using population sequencing -- not a supersensitive resistance assay -- Smith recorded mutations in semen of one man up to 1,179 days after the estimated date of HIV infection -- even though virus in blood samples proved sensitive to NNRTIs at that point. Drug-susceptible virus did not turn up in semen of another man until an estimated 1,193 days after infection. The other three men still have NNRTI resistance mutations in semen. Resistant virus, Smith concluded, can last over three years in semen and apparently sticks around in semen even longer than in blood.
Pillay's analysis rests on the UK HIV Drug Resistance Database, set up in 2001 to bank resistance test results from clinical practice. This stockpile now brims with over 15,000 test results covering about 85% of antiretroviral-treated people in Britain. Pillay defined MDR virus as HIV spotted with mutations from the first three antiretroviral classes. He toted a genotypic sensitivity score (GSS) by adding
So a higher GSS is better.
When the 628 people studied learned they had MDR HIV, they had a median CD4 count of 238 cells/mm3 (IQR 110 to 376 cells/mm3) and a median viral load of 4.2 log copies/mL (about 15,850 copies/mL, IQR 3.5 to 4.8 log). They had tried a median of eight antiretrovirals (IQR two to 14) and had spent a median of 4.5 years on therapy (IQR 3.1 to 6.8).
Pillay recorded 54 deaths after MDR diagnosis in a median follow-up of 24 months. Estimated probability of death measured 3% one year after MDR diagnosis, 8% two years after, and 13% three years after.
Regression analysis to thresh out survival predictors involved 321 people with at least six months of follow-up after they learned their MDR status. To no one's surprise, a higher CD4 count at that point (P = 0.013), a lower viral load (P = 0.006), more recent year (P = 0.024), and lower number of drugs tried (P = 0.002) improved the odds of survival. Years of treatment, GSS of the original regimen, number of active drugs in the original regimen, change in number of inactive drugs, and boosted PI therapy did not independently influence survival.
This analysis also uncovered a trend (P = 0.143) toward prolonged survival in people who switched regimens after their MDR verdict came in. Compared with people who continued the same regimen with MDR virus, two groups had a better shot at longer survival:
In contrast, people who switched antiretrovirals but ended up with a lower GSS had about a 50% higher risk of death than those who stuck with the same regimen (RR 1.49, 95% CI 0.17 to 13.0). Everyone knows that changing drugs won't slow HIV if the new regimen doesn't work. This analysis suggests that changing drugs and picking up new mutations curtails survival. On the other hand shifting to different drugs and improving the GSS made longer life more likely -- and hiking the GSS by a single point correlated with a higher CD4 count 24 weeks after the change.
Pillay believes his results "support an active management strategy of patients diagnosed with MDR HIV-1, including resistance test-guided optimization of therapy." But the results also suggest that misguided tries at improving a failing regimen may imperil a person's lease on life. With the resistance data quagmire growing more glutinous daily, clinicians need expert help more than ever.
The cohort had a median baseline CD4 count of 300 cells/mm3 (IQR 146 to 389 cells/mm3) and a median RNA of 3.8 log copies/mL (about 6,300 copies/mL, IQR 3.78 to 4.70 log). Most of the study group, 86%, had at least one NRTI mutation, 52% had one or more PI mutations, and 43% had at least one NNRTI mutation. While 23% had single-class resistance, 42% had virus resistant to two classes and 23% to three classes. Among people with protease mutations, 77% had at least one "universal" protease mutation at position 33, 82, 84, or 90; 23% had two such mutations and 18% three. Two thirds of the cohort carried TAMs.
Eighty people had an AIDS diagnosis or died during 715 person-years of follow-up. Compared with people who had no mutations or mutations conferring resistance to one or two classes, those with triple-class mutations had a 62% higher risk of progression in an analysis adjusted for CD4 count and viral load at genotyping and for prior AIDS diagnosis (95% CI 0.98 to 2.70, P = 0.013). In the same type of analysis, people marked with three or more "universal" protease mutations had a 3.43 higher risk of progression (95% CI 1.61 to 7.29, P = 0.0007) than people with two or fewer of these mutations.
By itself the I84V protease mutation heightened the risk of progression 1.71 times, though that correlation fell shy of statistical significance in a univariate analysis (Table 5). Every additional TAM raised the risk of progression 1.13 times. And resistance to NNRTIs fostered progression almost as much as triple-class resistance.
Winters and ACTG colleagues monitored mutations in 46 people enrolled in study A5102, which randomized participants to add interleukin 2 to their antiretrovirals 16 weeks before the STI or to continue only antiretroviral therapy before the STI. Everyone had a viral load below 200 copies/mL and a CD4 count above 500 cells/mm3 when they enrolled, and the median time on potent therapy measured 48 months.
All study participants rebounded above 500 copies/mL within eight weeks of stopping therapy. Eleven of 41 people with pre-STI PBMC results (27%) had a resistance mutation in their rebound plasma sample. Eight of 17 (47%) suspending a PI regimen and three of 20 (15%) mothballing an NNRTI had mutants at rebound. People with faster CD4 slippage when they stopped treatment proved significantly more likely to have resistance mutations upon rebound (P = 0.0004).
Ten of 11 people with rebound resistance had detectable viremia only two weeks after stopping treatment. The ACTG crew advised monitoring viral load and genotype two weeks into an STI to judge whether it's safe to continue.
All five people with mutations in PBMCs before the STI had mutations in their breakthrough sample, compared with six of 36 people without PBMC mutations before the STI (P = 0.0006). Still, the high resistance rate -- 17% -- in people with no evidence of archived mutations before their STI underlines the risks of this strategy.
At the Resistance Workshop, Harrigan extended this analysis to chart the impact of adherence on specific mutations. The study involved people starting their first antiretrovirals from August 1996 through September 1999. The British Columbia team genotyped 2,805 samples from 1,191 people whose viral load topped 1,000 copies/mL in the first 30 months of treatment. During that span one mutation arose in 298 people, two in 158, three in 83, four in 42, and three in 28. Whereas 885 people had begun a PI regimen, 306 started with an NNRTI.
Harrigan and colleagues rated adherence by two measures -- prescription refills in British Columbia's universal antiretroviral coverage program, and PI or NNRTI levels in the first two blood samples during treatment. A multivariate model adjusting for resistance to each class as a time-dependent variable traced a strong correlation between adherence and resistance to NNRTIs or 3TC, but not to other NRTIs or to PIs.
Compared with 95% or better adherence, 80% to 95% refill-based adherence raised the risk of NNRTI mutations six times (95% CI 3.3 to 10.9, P<0.0001) and of 3TC mutations three times (95% CI 1.9 to 4.7, P<0.0001). When Harrigan figured adherence by refill and drug levels, less than 95% adherence raised the risk of NNRTI mutations seven times (95% CI 3.4 to 14.5, P = 0.0001) and of 3TC mutations 4.5 times (95% CI 2.6 to 7.9, P = 0.0001). If people refilled prescriptions more than 95% of the time but had low PI or NNRTI concentrations, they had a high risk of mutations involving any drug class.
Margolis thinks valproic acid rousts HIV from sleepy CD4s by inhibiting histone deacetylase (HDAC). Stifling HDAC perks up HIV promoter and viral expression without waking up every T cell in town. He tested that theory in a pilot trial involving four people who kept their viral load below 50 copies/mL for an average 46 months with antiretroviral therapy.
As a protective measure Margolis first added the fusion inhibitor ENF to each person's regimen for four to six weeks to bar HIV from uninfected cells during valproic acid therapy. Then study participants took valproic acid for another three months, maintaining drug levels from 50 to 100 µg/mL with weight-based doses around 500 to 750 mg twice daily.
Three of four volunteers had significant drops in resting T cells housing replication-competent HIV. The downturns averaged 64% and ranged from 52% to more than 84%. Falls in infectious units per million cells ranged from 29% (P = 0.035) to more than 84% (P<0.009). Antiretroviral therapy alone doesn't do that. Everyone tolerated valproic acid well. Margolis plans to continue studying the drug, but next without the ENF shield.
Study results appeared in Lancet shortly after the workshop.18 There, Margolis allowed himself a conclusion bolder than any heard since 1996: "This finding, though not definitive, suggests that new approaches will allow the cure of HIV in the future."
Draining the last drop of infectious virus from T-cell reservoirs will not eradicate HIV. The virus has too many hiding places. But siphoning some HIV from resting cell reservoirs could pay other dividends. At least that's what one might conclude from recent work at the University of Washington.19
Studying 37 children with well-controlled viral replication, the University of Washington's Lisa Frenkel and coworkers took a stab at that enduring question -- where does blip virus come from: archived HIV or creeping replication? Frenkel's evidence suggested that both sources may fuel viral spikes during apparently suppressive therapy. And those spikes, she confirmed, are not innocuous. Resistance mutations evolved in virus of one child with frequent blips.
Will tightening the reservoir tap -- with valproic acid or some other agent -- stop inevitable resistance drips? No one knows. That's why the import of studies like these should not be minimized.
The Wallace Stevens line quoted at the top of this article ends, by the way, thus:
... And yet relation appears,
A small relation expanding like the shade
Of a cloud on sand, a shape on the side of a hill.
Mark Mascolini writes about HIV infection (firstname.lastname@example.org).