A total of 176 (91 females, 51.7% and 85 males, 48.3%) HIV-1 infected patients receiving care at Instituto de Medicina Integral Professor Fernando Figueira (IMIP) were recruited in Recife and nearby cities, Pernambuco state, Northeast Brazil, for a case–control, observational study between 2011 and 2015.

Inclusion criteria were: age over 18 years old, not reporting illicit drug use, and not being pregnant. Informed consent to participate in the study and to have the medical charts reviewed was obtained from all included participants. A peripheral blood sample was collected from each patient. IMIP Research Ethics Committee approved the study protocol (number 3629-13).

Each patient answered a standard questionnaire about sex, age and age at ART start, socioeconomic status, smoking, and drinking habits. The data extracted from medical charts covered the whole period between the first and last plasma viral load (pVL) and CD4+ T cell absolute counts and percentage (relative to all white blood cells) measurements (retrospective follow-up).

Other abstracted data included ART regimens received and their refill prescriptions and serological status for the following etiologic agents: HBV, HCV, CMV (IgG and IgM), HTLV-1/2, Toxoplasma gondii (IgG and IgM) and Treponema pallidum (VDRL) and occasional AIDS-defining conditions following the Centers for Disease Control (CDC) 1993 Revised Classification System.15

All patients achieved persistent viral suppression, which was defined as maintaining undetectable (also known as “target not detected”) plasma viral load measurements (pVL<50copies/mL as measured by RT-PCR). Current pVL quantification methods adopted in Brazil, such as Abbott RealTime HIV-1 Viral Load Assay16 have detection limits of 40copies/mL, but since our sample is retrospective, we adopted the 50copies threshold to account for measurements using earlier methods. No patient presented viral load rebound, which was defined as two consecutive pVL>1000copies/mL at any time after initial suppression, according to Brazil's Ministry of Health's guidelines at the time of patients’ recruitment.17 In the study setting, pVL measurements, CD4+ T cell counts and other laboratory tests were generally performed every three or four months, at the physician's discretion.

Immunologic response was defined as CD4+ T cell percentage achieving 30% or higher for two consecutive measurements during follow-up and immunological non-response if otherwise, following Brazilian Ministry of Health's guidelines.17 If a patient already had pre-ART CD4+ T cell percentages >30% (early start patients), immunological response was considered as an absolute gain of 200cells/μL, according to a previous study.5 CD4+ T cell percentages were preferred instead of absolute counts because percentages are less variable over time.18

Thus, 67 patients were categorized into immunological non-response and 109 into immunological response groups and their follow-up duration was recorded for further survival analysis.

Genes and SNPs were selected through literature search and functional criteria. We selected 46 SNPs distributed in 19 genes of antiretroviral pharmacodynamic pathways: ABCB1 (rs1128503, rs2214102, rs2235048 and rs3842), ABCC1 (rs129081, rs113264879, rs4148380, rs8056298, rs212091 and rs16967632), ABCG2 (rs115770495, rs1448784 and rs2231142), CYP1A2 (rs762551), CYP2A6 (rs8192726), CYP2B6 (rs8192709, rs28399499, rs34097093, rs28399502, rs707265 and rs1042389), CYP3A4 (rs4646437), SLC22A6 (rs11568629, rs11568628 and rs4149170) and NR1I3 (rs3003596)19–27 and immunological activation and homeostasis: CCL5 (rs2107538), FAS (rs2234767 and rs1800682), IFNG (rs2069709), IL10 (rs2222202, rs1800871 and rs1800890), IL10RA (rs3135932 and rs9610), IL15 (rs10519613 and rs10833), IL1B (rs16944), IL2 (rs2069762), IL4 (rs2243250), IL7R (rs1494555, rs11567762, rs6897932, rs3822731, rs987106 and rs3194051).28–30

Additionally, 48 SNPs that served as ancestry informative markers (AIMs) were also genotyped (all variants are listed on Supplementary Table 1). Briefly, these SNPs help estimating ancestry proportions in admixed populations such as the one enrolled in our study, controlling for population structure and reducing bias during genetic association analysis.31 Genomic DNA was extracted through Promega® Wizard Genomic DNA Purification Kit (Fitchburg, Wisconsin, USA), following manufacturer's instructions. Genotyping was performed through VeraCode® platform of GoldenGate® Illumina Inc (San Diego, California, USA) technology, following manufacturer's instructions.

Raw genotyping data were extracted with Illumina® Genome Studio 2.0 software and exported to an Excel® worksheet. After processing, the dataset was exported into PLINK software, version 1.9032 to perform quality control (QC) filtering. Samples and variants with less than 90% global call rates were removed from further analysis. Variants with significant departure from Hardy–Weinberg equilibrium were also removed, using an exact test with p-value<0.001 as threshold.

The individuals in our sample presented a major European ancestry contribution (mean proportion 55.2±18.5%), followed by African (32.3±16.4%) and Amerindian (12.4±10.5%), as expected, due to our previous works with other samples of the same general population.31

The median age at treatment start was 33 years old (interquartile range, IQR=27–39), and the patients tended to start ART with relatively low CD4+ T cell counts/μL of peripheral whole blood (median=272, IQR=127.5–432.5), and 29.5% of the patients started with less than 200cells/μL. In contrast, only 10.8% of the patients started with more than 500cells/μL. Similarly, CD4 percentages relative to total lymphocyte count were also low, (median=19.4%, IQR=12.4–27.5%), and only 13.1% of the patients started therapy with more than 30% CD4+ T cells. Most AIDS-defining conditions diagnoses were not available (67.6%) in the medical charts, but at least 22.1% of the patients started ART with C stage conditions.

Regarding ART regimens, the majority of the patients received zidovudine+lamivudine (AZT+3TC) as first line NRTI backbone following the guidelines at that moment (68.2%), followed by tenofovir+lamivudine (TDF+3TC, 19.9%) and abacavir+lamivudine (ABC+3TC, 0.6%). AZT monotherapy was the starting treatment for 2.8% of the patients, and information on the first regimen received was not available for 8.5% of the patients. The third drug option was mostly a NNRTI (53.5%), such as efavirenz (EFZ) or nevirapine (NVP), and ritonavir boosted PI such as lopinavir or atazanavir (LPV/r or ATV/r) prescribed to 35.2% of the patients. Table 1 also displays univariate comparison of patients’ characteristics.

Males had more than two-fold increased risk of presenting immunological non-response (61.2% non-responders and 40.4% responders; odds ratio (OR)=2.32; 95% CI=1.20–4.56; p=0.01). Both groups had similar ages at treatment start (median 34 years and 33 years, respectively; p=0.45). As expected, non-responders started treatment with lower absolute CD4+ T cell count than responders (median 179cells/μL vs. 342.5cells/μL, respectively, p<0.001), and most non-responders started therapy with less than 200cells/μL (49.2% vs. 17.4%; p<0.001). Patients in both groups had similar ancestry backgrounds (mean African ancestry proportion 33.5% in non-responders vs. 31.7% in responders, p=0.68).

Non-responders started treatment with more advanced disease than responders; 31.3% of non-responders presented C (AIDS-indicator conditions) CDC system stage vs. 16.5% of responders, although the difference did not reach statistical significance (p=0.28).

Patients started with similar first-line ART regimens (p=0.20). There was neither difference in the frequency of NNRTI and PI (p=0.40) as third drug nor AZT+3TC and TDF+3TC (p=0.56) as backbones.

The median time on treatment and of laboratory tests was 33 months (IQR=19.8–60.5). As expected, ART greatly benefited patients. Aggregating laboratory measurements from the last observation in each patient, the median CD4+ T cell count rose to 578.5cells//μL of blood (IQR=393.2–807.5), an increment by nearly 113% from pre-ART values mentioned above (p<0.001). Similarly, CD4+ T cell percentages rose to 36.7% (IQR=23.0–35.8), an increment by nearly 62% (p<0.001). These post-ART changes are summarized in Table 2.

The median time to achieve immunologic response was 47 months (95% CI=36–64). The time for men and women to achieve immunologic response was almost the same with a median time of 48 months (95% CI=36–110) for men and 46 months (95% CI=30–66) for women (p=0.42). Age groups at ART start was neither associated with time until immunologic response (p=0.67). Similarly, use of AZT instead of TDF (p=0.11) or PIs instead of NNRTIs as third drug (p=0.14) on ART regimens also did not influence time until response.

Serological tests results were not recorded in medical charts of the majority of patients. Although about 65% of the patients did not results for CMV and toxoplasmosis infections, the prevalence of positive IgG was 33.5% for CMV and 27.8% for toxoplasmosis. Syphilis VDRL positivity rate was 13.6% of the total (34.7% untested); 25.0% were immune to hepatitis B virus (HBV) due to vaccination, 5.7% due to natural infection, and the remaining 26.1% were susceptible, but none presented chronic infection (43.2% untested). No patient was positive for anti-HTLV-1/2 (79.5% untested); a single case of anti-HCV positive was detected (47.7% untested). Table 3 details the serological tests results.

Two candidate SNPs, rs11568629 (SLC22A6) and rs16967632 (ABCC1) and four AIMs did not pass genotyping QC and were removed from further analysis. Other variant, rs34097093 (CYP2B6), was also removed because all individuals in the sample had the same genotype (one of the alleles was fixed). The remaining allele and genotype frequencies were all in conformity to Hardy–Weinberg equilibrium according to PLINK software exact test. All frequencies and global call rates are displayed in Supplementary Table 2. Therefore, further analyses were performed using the remaining 43 candidate SNPs and 44 AIMs.

Five SNPs presented statistic association with immunological outcome. Three minor alleles were more frequent in non-responders than responders: rs2243250 (IL4) A allele (48.0% vs. 37.0%, p=0.04), rs1128503 (ABCB1) A allele (40.0% vs. 28.0%, p=0.03) and rs707265 (CYP2B6) A allele (40.0% vs. 23.0%, p=0.02), whereas the other two were more frequent in responders: rs2069762 (IL2) C allele (15.0% vs. 30.0%; p=0.004) and rs4646437 (CYP3A4) A allele (26.0% vs. 34.0%; p=0.04). Allele, genotype frequencies and statistical analyses for these five SNPs are displayed in Table 4 and Supplementary Table 3 displays all genetic association results.

Since non-responders and responders had similar ancestry proportions (for example, mean African contribution 33.5±18.5% vs. 31.7±15.0%, respectively; p=0.68), we believe that there is no hidden genetic structure biasing the genetic association analysis. As mentioned in the next session, individual African genetic ancestry contributions were included in the multivariate Cox proportional hazards model for an additional “genomic control”.35

The multivariate Cox proportional hazards analysis suggested that starting ART with less than 200 CD4+ T cells/μL, IL2 rs2069762 genotypes, ABCB1 rs1128503 genotypes and PI/r use were significantly associated with time to immune response.

Individuals starting ART with less than 200 CD4+ T cells/μL had 61% lower chance of achieving immunologic response when compared with those who started ART with more than 200 CD4+ T cells/μL (HR=0.39, 95% CI=0.21–0.74, adjusted p=0.004). IL2 rs2069762 C/C genotype had an eight-fold higher probability of achieving immunologic response when compared to A/A genotype (HR=8.60, 95% CI=2.06–35.89, adjusted p=0.003) during the follow-up period. ABCB1 rs1128503 A/A genotype was associated with 71% lower chance of achieving immunologic response when compared with G/G genotype (HR=0.29, 95% CI=0.15–0.58, adjusted p=0.0004). Using an ART regimen containing PI/r had almost two-fold higher probability of achieving immunologic response (HR=1.89, 95% CI=1.03–3.50, adjusted p=0.04).

On the other hand, sex (adjusted p=0.51), genetic background (adjusted p=0.85), age at treatment start (adjusted p=0.82), IL4 rs2243250, CYP3A4 and CYP2B6 SNPs, use of AZT as NRTI backbone, and events of detectable pVL following viral suppression were not significantly associated with probability to immune response on multivariate analysis. The multivariate analysis results are summarized in Table 5.