For individuals living with HIV-1, access and adherence to antiretroviral therapy (ART) has dramatically reduced morbidity,1,2 mother-to-child transmission,3 and sexual transmission4 while increasing life expectancy.5 Because of their significant positive impact, the US Department of Health and Human Services recommends ARTs for all adults and adolescents living with HIV.6 ART is also fundamental to worldwide treatment targets put forth by the United Nations, which aim for 90% of individuals with HIV to receive a diagnosis, 90% of individuals with HIV diagnoses using ARTs continuously, and 90% using ARTs reaching viral load suppression by 2020.7 Current recommendations for first–line combination ART include a recently developed antiretroviral class known as integrase strand transfer inhibitors (INSTIs).6 However, HIV-1 mutants conferring resistance to INSTIs have already been detected,8–10 sparking concern that INSTI drug resistance might become widespread and threaten ART effectiveness. Here, we forecast prevalence of HIV-1 mutants resistant to INSTI class drugs among individuals living with HIV-1 to inform and maintain effectiveness of ART regimens.
INSTIs target the HIV-1 enzyme integrase (IN), preventing retroviral DNA from integrating into host DNA, thereby blocking viral replication.11,12 As of 2018, 3 drugs in the INSTI class have been approved for use by the US Food and Drug Administration and completed Phase 3 trials: raltegravir (RAL) in 2007, dolutegravir (DTG) in 2013, and elvitegravir (EVG) in 2014.13 These drugs have shown high effectiveness in Phase 3 clinical trials, with >80% of patients showing undetectable viral loads (<50 copies/mL) after of 48 weeks of use.14 However, multiple major and accessory INSTI resistance mutations have been identified in vitro14,15 and detected in vivo, appearing in 1%–2% of ART-naïve individuals with HIV-1 and 1%–10% of ART-experienced individuals with HIV-1.8–10 These data indicate that most resistance is acquired or developed in individuals undergoing ART; however, transmitted INSTI resistance has been observed.16,17 Because the length of time before detection of transmitted INSTI resistance was the same duration as the length of time before detection of transmitted mutants resistant to older ART drug classes, resistance to INSTI class drugs might increase to high levels currently detected for nucleoside/nucleotide analogue reserve transcriptase inhibitors (NRTIs) and non-nucleoside reverse transcriptase inhibitors (NNRTIs).18 Ideally, we would use historical data to predict the length of time that we expect INSTIs to remain effective. But, because of the recent introduction of widespread INSTI treatment, the low prevalence of INSTI-resistant mutants in drug-naïve populations, and the combination of acquired and transmitted drug resistance (TDR), it is difficult to predict the duration of INSTI effectiveness.
One approach is to use mathematical models to forecast the prevalence INSTI-resistant HIV-1 infections. Compartmental models based on the classic susceptible, infectious, removed model19 are ideal for representing HIV-1 transmission and can be extended to include diagnosis status,20 ART,21 and drug resistance.22 Here, we use data from published government reports to parameterize our models to mimic HIV-1 dynamics in Washington, DC, United States. In 2016, 1.9% of people residing in Washington, DC, were living with HIV-1, which exceeds the World Health Organization's definition of a generalized epidemic (>1% prevalence).23 Epidemic levels have been reached in black men (4.4% prevalence), Hispanic/Latino men (2.1% prevalence), black women (1.9% prevalence), and white men (1.7% prevalence).23 The most common mode of transmission is sexual contact.23 Nine HIV-1 mutants associated with INSTI class resistance were documented in a cohort of individuals living with HIV-1 in Washington, DC.24 By the end of 2014, the most common acquired INSTI-resistance mutations (F121Y and E92Q) affected 0.9% of the cohort, and the most common transmitted INSTI-resistance mutations (Q148R and E92Q) affected 0.9% of the cohort.24
Using this compartmental model of HIV-1 dynamics, we examine 2 questions about antiretroviral drug resistance. First, we ask what the prevalence of INSTI drug resistance will be over the next 30 years considering variability in diagnosis rates and ART use. Second, we ask how much INSTI drug resistance is caused by TDR and how much is caused by regimen-acquired drug resistance (ADR). Answering these questions will inform initiatives to expand access and effectiveness of antiretroviral drugs while also identifying factors that promote antiretroviral drug resistance.
Model of HIV-1 Transmission, ART Adherence, and Drug Resistance
To determine dynamics of HIV-1 incidence and drug resistance, we designed a model that included transmission of HIV-1 responsive to INSTI treatment and transmission of HIV-1 strains that carry a mutation that confers drug resistance to raltegravir (RAL), elvitegravir (EVG), and/or dolutegravir (DTG). Henceforth, we refer to the strains of HIV-1 sensitive to INSTI treatment as “responsive” strains, but we acknowledge that these strains might carry mutations and might be resistant to other ART classes. Our model represents HIV-1 transmission with individuals categorized into 8 compartments depending on their HIV-1 status. Susceptible (S) individuals have never contracted responsive HIV-1 nor HIV-1 with INSTI drug resistance characteristics and are ART-naïve. Acutely infected individuals harbor either responsive HIV-1 (A1) or HIV-1 with INSTI drug resistance characteristics (A2) and are ART-naïve. Individuals with prolonged HIV-1 infection that adhere with ART also harbor responsive HIV-1 (C1) or HIV-1 with INSTI drug resistance characteristics (C2). Individuals with prolonged HIV-1 that do not adhere to ART also harbor responsive HIV-1 (Y1) or HIV-1 with INSTI drug resistance characteristics (Y2). Finally, removed (R) individuals die from complications of HIV-1 infection (Fig. 1).
We assumed that transmission is frequency dependent and symbolized by a coefficient, β that represents the rate of contacts multiplied by the probability that a susceptible individual will acquire infection given contact with an infectious individual.25 Let β represents the transmission coefficient for both responsive HIV-1 and HIV-1 with INSTI resistance. Let i represent a specific HIV-1 strain, such that i = 1 represents responsive HIV-1 and represents HIV-1 with INSTI resistance. We assume that acutely infected individuals (i = 2) and chronically infected nonadhering individuals (Ai) drive transmission, and that acutely infected individuals (Ai) contribute more to transmission. We also assume that the drug-resistant mutant (i = 2) is less fit than the responsive strain (i = 1), implying that only a fraction δ of individuals carrying HIV-1 with INSTI resistance will transmit this resistant strain while the other 1 − δ individuals carrying HIV-1 with INSTI resistance will transmit a responsive strain26 (δ < 1). Then, for a strain of HIV-1 with INSTI resistance,where α < 1. Please note that λ2 represents TDR.
For responsive HIV-1, the force of infection (λ1) is a function of acutely and chronically infected nonadhering individuals living with responsive HIV-1 (A1, Y1) and a fraction 1-δ of the acutely and chronically infected nonadhering individuals living with INSTI-resistant HIV-1 (A2, Y2), such that
A full description of the model can be found in the supplementary material (see Text S1 and Equations S1–S7, Supplemental Digital Content, http://links.lww.com/QAI/B388).
Model Parameterization and Initial Conditions
We obtained values for all infection parameters—except β—from the peer-reviewed literature as reported in the supplementary material (see Table S1, Supplemental Digital Content, http://links.lww.com/QAI/B388). We obtained data on HIV-1 infections from in Washington, DC, from publicly available government reports and used them to estimate values for β (see Texts S2 and S3, Supplemental Digital Content, http://links.lww.com/QAI/B388). We obtained values for demographic parameters from publicly available government reports as reported in the supplementary material (see Table S1, Supplemental Digital Content, http://links.lww.com/QAI/B388). We set initial conditions to represent HIV-1 infection and drug resistance dynamics in Washington, DC, using values from census data and publicly available reports (see Table S3, Supplemental Digital Content, http://links.lww.com/QAI/B388).
Model Simulations and Output Analyses
We performed stochastic simulations of HIV-1 dynamics as detailed in the supplementary material (see Text S4, Supplemental Digital Content, http://links.lww.com/QAI/B388). The first set of simulations represented HIV-1 dynamics and drug resistance (ADR + TDR) for INSTI class drugs for 30 years and quantified the proportion of HIV-1 cases with INSTI drug resistance prevalence at the end of the simulation. We altered initial conditions and quantified the effect on the proportion of HIV-1 cases with INSTI drug resistance. We also conducted sensitivity analyses to quantify how parameter uncertainty affected model output. We varied parameters by 10% and 20% and ran 100 simulations at each new parameter value. We report the mean, minimum, and maximum effect of changing the parameter value. The second set of simulations represented HIV-1 dynamics and drug resistance (ADR only) for INSTI class drugs. To simulate only ADR, we removed the TDR transmission route (λ2 = 0), the compartment representing the individuals acutely infected with resistant HIV-1 (A2 = 0), and any terms including A2 (Equation 2). We quantified the amount of INSTI resistance caused TDR by comparing results of the first simulation (ADR + TDR) to the results of the second simulation (ADR only). All parameter estimations, model simulations, and sensitivity analyses were conducted in R version 22.214.171.124
We forecast the prevalence of INSTI drug resistance over the next 30 years. Simulations indicated that the proportion of HIV-1 cases resistant to raltegravir (RAL), elvitegravir (EVG), and dolutegravir (DTG) were similar (Fig. 2). After we simulated 30 years of HIV-1 dynamics, the mean proportion resistant to RAL was 0.41 (minimum: 0.21; maximum: 0.57), the mean proportion resistant to EVG was 0.44 (minimum: 0.26; maximum: 0.60), and the mean proportion resistant to DTG was 0.44 (minimum: 0.25; maximum: 0.65). Variability among the proportion resistant after 30 years for RAL, EVG, and DTG was similar. For RAL, there was 0.36 between the minimum value simulated and maximum value simulated. For EVG, there was 0.34 between the minimum value simulated and maximum value simulated. For DTG, there was 0.40 between the minimum value simulated and maximum value simulated. Most simulations forecast considerable INSTI resistance after 30 years of use (Fig. 2).
We varied initial conditions and quantified the proportion of HIV-1 cases with INSTI resistance (Fig. 3). We only plotted results for HIV-1 cases that are responsive to INSTIs (A1, C1, Y1) in Figure 3 because results for HIV-1 cases with INSTI resistance (A2, C2, Y2) were identical. When we compared results from simulations with varying initial conditions (Fig. 2) to results from baseline simulations (Fig. 2), we found similar mean proportions of HIV-1 cases with INSTI resistance after 30 years. The greatest difference between mean values was observed in simulations representing EVG resistance (see Figure S1, Supplemental Digital Content, http://links.lww.com/QAI/B388). We observed similar minimum proportions resistant with the greatest difference observed in simulations representing RAL resistance (see Figure S2, Supplemental Digital Content, http://links.lww.com/QAI/B388). We found a discrepancy in the maximum proportions resistant—up to 14.6%—with the greatest differences observed in simulations representing RAL and EVG resistance, respectively (see Figure S3, Supplemental Digital Content, http://links.lww.com/QAI/B388). Although the mean and minimum values of the proportion of HIV-1 cases with INSTI resistance was not affected by initial conditions, the maximum was affected and substantially increased the range of values that the model produced (Fig. 3).
We performed sensitivity analyses on all parameters. The parameters that most often affected model output were p, the proportion of those living with HIV-1 in Washington, DC, on ART, and τ1, the rate of converting from an INSTI sensitive strain to an INSTI-resistant strain for chronically infected ART-experienced cases (Table 1). Sensitivity for all parameters is reported in the supplementary material (see Tables S4 and S5, Supplemental Digital Content, http://links.lww.com/QAI/B388).
When we simulated INSTI resistance caused only by ADR and used initial conditions that represented HIV-1 cases in Washington, DC, in 2016, simulations indicated that the proportion of HIV-1 cases with INSTI resistance (Fig. 4) was almost identical to simulations that included ADR and TDR (Fig. 1). After we simulated 30 years of HIV-1 dynamics with only ADR, the mean proportion resistant to RAL was 0.41 (minimum: 0.22; maximum: 0.57), the mean proportion resistant to EVG was 0.44 (minimum: 0.27; maximum: 0.60), and the mean proportion resistant to DTG was 0.44 (minimum: 0.25; maximum: 0.65). Most simulations forecast considerable INSTI resistance after 30 years of use and indicated that TDR contributes very little, if at all, to the overall proportion of HIV-1 cases with INSTI resistance. (Fig. 4).
Our study forecasted the prevalence of INSTI resistance among those living with HIV-1 using initial conditions and parameter values that represented disease dynamics in Washington, DC. Our model forecast a substantial proportion of HIV-1 cases with INSTI resistance after 30 years that was similar across all 3 INSTIs that we investigated. The proportion of cases resistant to RAL (mean: 0.41) was almost identical to the proportion resistance to EVG (mean: 0.44) and DTG (mean: 0.44). When we varied the initial conditions for our simulations, the mean and minimum proportion of cases resistant to each INSTI did not change. However, the maximum of proportion of cases resistant in the simulations did change in a manner that increased variability in simulation output. When we performed parameter sensitivity analyses, the most sensitive parameter was p, the proportion of those living with HIV-1 in Washington, DC, on ART, regardless of INSTI prescribed. Finally, we compared results from simulations with both ADR and TDR to simulations with only ADR. We found almost no difference in the proportion of cases resistant, indicating that ADR caused most—if not all—of INSTI resistance in our simulations.
The proportion of HIV-1 cases with INSTI resistance that our model forecasted was similar to the proportion of HIV-1 cases resistant to other ART classes observed in Washington DC in 2014. Resistance to nucleoside reverse transcriptase inhibitors (NRTIs) was 29.7% after 29 years of use; resistance to non-nucleoside reverse transcriptase inhibitors (NNRTIs) was 29.3% after 20 years of use.24 Resistance to protease inhibitors (PIs) and entry inhibitors (EIs) was lower at 14.6% after 21 years and 1.5% after 13 years of use, respectively.24 Our simulations predict that INSTI drug resistance over the next 30 years will be comparable with drugs in other ART classes after use for a similar duration of time.
The proportion of HIV-1 cases with RAL and EVG resistance that our model predicted is consistent with current understanding of the genetic barrier to resistance. The genetic barrier for RAL and EVG is similar to the genetic barrier for efavirenz (EFV), an NNRTI.15 After 20 years of use, data indicate ∼30% NNRTI resistance in Washington, DC, which is similar to the mean proportion of HIV-1 case-resistant RAL and EVG predicted by our model after 20 years (Fig. 2). However, the proportion of HIV-1 cases with DTG resistance that our model predicted is less consistent with the genetic barrier to resistance. Although the genetic barrier for DTG is similar to the genetic barrier for PIs,15 after 20 years of use, PIs resistance data from Washington, DC, show lower proportions than predicted by our model. Importantly, however, data from Washington, DC, are still within the lower bounds of our model predictions (Fig. 2). These results indicate that population factors like the proportion of those living with HIV-1 in Washington, DC, on ART might have a greater effect on DTG introduction and use than on RAL and/or EVG introduction and use. Alternatively, a combination of unsystematic factors might have led to DTG resistance to occur nearer our lower predictive bounds than RAL and EVG. Nevertheless, our model simulation results generally align with comparative genetic barriers.
The proportion of HIV-1 cases resistant to INSTIs due to TDR forecasted by our model differed from the proportion of cases resistant to other ART classes through TDR. In Washington, DC, in 2014, 20.5% of individuals exhibited TDR to any ART.24 NNRTIs showed the highest proportion with TDR at 11.7%, followed by NRTIs (7.9%) and PIs (5.7%). Cases of INSTI TDR have also been reported16,28 but have not yet reached the high proportions exhibited by other ART classes.24,29 This discrepancy could arise for 3 reasons. First, it might be an artifact of our modeling procedure. Perhaps the relatively new introduction of INSTIs gives values for initial conditions for our model that obscures TDR. Or, perhaps the population-level resolution of our model obscures TDR because TDR is often observed in transmission clusters29 that might be better simulated using individual-level or network models. Second, the discrepancy might be caused by mutations that confer INSTI resistance but less likely to be transmitted. HIV-1 acquires mutations that confer resistance at a fitness cost and are less likely to be transmitted than responsive strains.30 If the fitness cost for INSTI resistance is higher than fitness costs for resistance to other ARTs, then INSTI TDR would be impeded compared with TDR for other ARTs. Detailed molecular studies that show mutation/fitness tradeoffs would be particularly informative. Third, the discrepancy might be caused by recent public health efforts and ART use that decrease transmission of HIV-1 in general and decrease transmission of mutations conferring resistance to the most recent class of ARTs prescribed—INSTIs—in particular.
The accuracy of our model forecasts is contingent on several key assumptions. We assumed consistent adherence or nonadherence to INSTIs and stable in- and out-migration of individuals living with HIV-1. We assumed independent drivers of resistance for each INSTI without considering cross-resistance. We assumed specific values for parameters that affected simulated HIV-1 dynamics and drug resistance trends, such as the proportion of individuals with and without INSTI-resistant strains of HIV-1 and the rate at which ART-experienced individuals with responsive HIV-1 acquire INSTI resistance mutations. Only time will tell whether our forecasts are accurate. But, even if our forecasts lack complete accuracy, our model remains useful in understanding HIV-1 dynamics and ART resistance. Our work exposed the lower threshold and variability that we can expect for INSTI resistance. Our work also highlights HIV-1 modeling studies' dependence between molecular and epidemiological studies of HIV-1. For example, τ1 was found to be an influential parameter for model outcomes; there is considerable uncertainty in the value of the parameter due to population-level variation in the proportion of individuals living with HIV-1 adhering to ART and the prevalence of drug resistance. Molecular mechanisms causing resistance other than the one based on our parameter values might affect our model results. As such, further molecular quantification of this value will be important for projections of INSTI drug resistance.
Our model shows general qualitative patterns of increasing INSTI resistance that are applicable to any location. The quantitative pattern of increasing INSTI resistance will be based on a location-specific value for p, the proportion of those living with HIV-1 in the local community on ART. The expected increase in HIV-1 cases with resistance to prescribed INSTIs testifies to the importance of monitoring and minimizing the proliferation of HIV-1 mutants that threaten the efficacy of treatment regimens. Continued monitoring will allow for maintenance of effective ART regimens, real-time comparisons of the amount of INSTI resistance caused ADR versus TDR, and real-time comparisons of INSTI resistance versus resistance to other antiretroviral drugs, and inform the patterns and processes underlying HIV-1 ADR resistance.
The authors thank Kristine Yoder for providing insight on HIV-1 biology.
1. Deeks SG, Phillips AN. HIV infection, antiretroviral treatment, ageing, and non-AIDS related morbidity. BMJ. 2009;338:a3172.
2. Wester CW, Koethe JR, Shepherd BE, et al. Non-AIDS-defining events among HIV-1-infected adults receiving combination antiretroviral therapy in resource-replete versus resource-limited urban setting. AIDS. 2011;25:1471–1479.
3. Townsend CL, Byrne L, Cortina-Borja M, et al. Earlier initiation of ART and further decline in mother-to-child HIV transmission rates, 2000–2011. AIDS. 2014;28:1049–1057.
4. Cohen MS, Chen YQ, McCauley M, et al. Prevention of HIV-1 infection with early antiretroviral therapy. N Engl J Med. 2011;365:493–505.
5. Nakagawa F, May M, Phillips A. Life expectancy living with HIV: recent estimates and future implications. Curr Opin Infect Dis. 2013;26:17–25.
6. Panel on Antiretroviral Guidelines for Adults and Adolescents. Guidelines for the Use of Antiretroviral Agents in Adults and Adolescents Living With HIV. Services DoHaH. Available at: http://www.aidsinfo.nih.gov/ContentFiles/AdultandAdolescentGL.pdf
. Accessed March 8, 2019.
7. Joint United Nations Programme on HIV/AIDS (UNAIDS). 90-90-90: An Ambitious Treatment Target to Help End the AIDS Epidemic Geneva, Switzerland: UNAIDS; 2016.
8. Menza TW, Billock R, Samoff E, et al. Pretreatment integrase strand transfer inhibitor resistance in North Carolina from 2010–2016. AIDS. 2017;31:2235–2244.
9. Lepik KJ, Harrigan PR, Yip B, et al. Emergent drug resistance with integrase strand transfer inhibitor-based regimens. AIDS. 2017;31:1425–1434.
10. Zoufaly A, Kraft C, Schmidbauer C, et al. Prevalence of integrase inhibitor resistance mutations in Austrian patients recently diagnosed with HIV from 2008 to 2013. Infection. 2017;45:165–170.
11. Pommier Y, Johnson AA, Marchand C. Integrase inhibitors to treat HIV/Aids. Nat Rev Drug Discov. 2005;4:236.
12. Hare S, Gupta SS, Valkov E, et al. Retroviral intasome assembly and inhibition of DNA strand transfer. Nature. 2010;464:232.
13. Antiretroviral drugs used in the treatment of HIV infection. Secondary Antiretroviral drugs used in the treatment of HIV infection 2018. https://www.fda.gov/ForPatients/Illness/HIVAIDS/Treatment/ucm118915.htm
. Accessed March 8, 2019.
14. White KL, Raffi F, Miller MD. Resistance analyses of integrase strand transfer inhibitors within phase 3 clinical trials of treatment-naive patients. Viruses. 2014;6:2858–2879.
15. Clutter DS, Jordan MR, Bertagnolio S, et al. HIV-1 drug resistance and resistance testing. Infect Genet Evol. 2016;46:292–307.
16. Young B, Fransen S, Greenberg KS, et al. Transmission of integrase strand-transfer inhibitor multidrug-resistant HIV-1: case report and response to raltegravir-containing antiretroviral therapy. Antivir Ther. 2011;16:253–256.
17. Boyd SD, Maldarelli F, Sereti I, et al. Transmitted raltegravir resistance in an HIV-1 CRF_AG-infected patient. Antivir Ther. 2011;16:257–261.
18. Hurt CB. Transmitted resistance to HIV integrase strand-transfer inhibitors: right on schedule. Antivir Ther. 2011;16:137–140.
19. Keeling MJ, Rohani P. Modeling Infectious Diseases in Humans and Animals. Princeton, NJ: Princeton University Press; 2011.
20. Gonsalves GS, Crawford FW. Dynamics of the HIV outbreak and response in Scott County, IN, USA, 2011–15: a modelling study. The Lancet HIV. 2018;5:e569–e77.
21. Medlock J, Pandey A, Parpia AS, et al. Effectiveness of UNAIDS targets and HIV vaccination across 127 countries. Proc Natl Acad Sci. 2017;114:4017–4022.
22. Smith RJ, Okano JT, Kahn JS, et al. Evolutionary dynamics of complex networks of HIV drug-resistant strains: the case of san francisco. Science. 2010;327:697.
23. District of Columbia Department of Health HIV/AIDS H, STD, and TB Administration (HAHSTA). Annual Epidemiology & Surveillance Report: Data Through December 2016. Washington, DC: Strategic Information Division. 2017.
24. Aldous AM, Castel AD, Parenti DM, et al. Prevalence and trends in transmitted and acquired antiretroviral drug resistance, Washington, DC, 1999–2014. BMC Res Notes. 2017;10:474.
25. Begon M, Bennett M, Bowers RG, et al. A clarification of transmission terms in host-microparasite models: numbers, densities and areas. Epidemiol Infect. 2002;129:147–153.
26. Leigh Brown AJ, Frost SDW, Mathews WC, et al. Transmission fitness of drug-resistant human immunodeficiency virus and the prevalence of resistance in the antiretroviral-treated population. J Infect Dis. 2003;187:683–686.
27. R: A Language and Environment for Statistical Computing [program]. Vienna, Austria: R Foundation for Statistical Computing; 2017.
28. McGee KS, Okeke NL, Hurt CB, et al. Canary in the coal mine? Transmitted mutations conferring resistance to all integrase strand transfer inhibitors in a treatment-naive patient. Open Forum Infect Dis. 2018;5:ofy294.
29. Levintow SN, Okeke NL, Hué S, et al. Prevalence and transmission dynamics of HIV-1 transmitted drug resistance in a southeastern cohort. Open Forum Infect Dis. 2018;5:ofy178.
30. Yang W-L, Kouyos RD, Böni J, et al. Persistence of transmitted HIV-1 drug resistance mutations associated with fitness costs and viral genetic backgrounds. PLoS Pathog. 2015;11:e1004722.