Peripheral blood mononuclear cells (PBMCs) were isolated from PLWH receiving ART (HIVART) and seronegative individuals [healthy donor (HD)]. All HIVART participants were receiving stable standard-of-care ART and had maintained plasma HIV-1 RNA < 50 copies/ml and a CD4 T cell count of > 300/μl for at least 6 months before enrollment. All experimental protocols were approved by local Institutional Biomedical Review Boards (ethics numbers: 14–0741, 11–0228, 13–3613, 12-1660, and 10-01330) and performed in accordance with the relevant guidelines. HD used for assay standardization were recruited by the UNC CFAR HIV/STD Laboratory Core (IRB 96-0859, http://unccfar.org/portfolio/hiv-std-laboratory-core/) and New York Blood Center (https://nybloodcenter.org).

The HIV-1 JRCSF infectious clone (pYK-JRCSF, #2708) was used in all presented data All viruses were propagated in HEK293T cells following plasmid transfection using Lipofectamine Plus reagent (Invitrogen Life Sciences, US) according to the manufacturer’s recommendations. Virus was collected from cell culture supernatant 48 hr post-transfection. JRCSF virus titers were determined by the TCID50. A luciferase assay was performed using TZM-bl cells with the Bright-Glo luciferase assay kit (Promega, USA), and the TCID50 value was determined by using the Reed-Muench method (13, 14). JRCSF titers ranged from 6.2 to 6.5 log10 TCID50/ml.

Cryopreserved PBMCs were thawed (Day 0) using Benzonase (25IU/ml of culture media) and rested overnight (18-20 hr) at 37°C. The next day (Day 1) PBMCs were counted using a Muse^® Cell Counter (Millipore Sigma, US) in accordance with manufacturer’s instructions. Samples were counted independently, 2×, then results were averaged. CD4 T cells were isolated from the PBMC by negative selection (MACS, Milteny-Biotec 130-096-533). Cell viability was typically > 95% and purity was > 96% after the isolation. CD4 T cells were resuspended at 1 to 2 ×10⁶/ml in R-10⁺ media (RPMI-1640 + 10%FBS + L-Glu + Pen/Strep + HEPES + Sodium pyruvate) in a 6-well plate (5× 10⁶/well maximum) for 72 hr +/- 3 hr, stimulated with 3 to 5 μg/ml of phytohaemagglutinin (PHA, L8902-5MG, Sigma-Aldrich, USA), IL-2 (20 IU/ml) and IL-7 (5ng/ml) (further details in Supplementary Material 1 ). Note, 3μg/ml PHA was used to stimulate CD4 T cells from HD and 5 μg/ml for stimulation of cells from HIVART participants. Generally, the cell recovery after 72 hr (Day 4) stimulation ranged between 70-120% of the input cells, and viability ranged between 70-90%.

After 72 hr, cultured CD4 T cells were washed in 10ml R10⁺ media 3× and counted. A total of 0.1 ×10⁶ cells were set aside to be used as uninfected control cells. Cells were infected with HIV-1 JRCSF at 5x 10⁶/ml in a 15ml Falcon tube (maximum number of cells and volume used for infection/15ml Falcon tube was 2× 10⁶ and 0.4ml, respectively) by spinoculation for 2hr at 27°C in 1200g with IL-2 (20 units/ml) and Polybrene (4μg/ml) (NC9840454, Santa Cruz Biotechnology), at MOI 0.03.

PBMCs were thawed as described above on Day 3. The next day (Day 4), PBMCs were counted independently, 2×, then results averaged. CD8 T cells were isolated from the PBMCs by positive selection in accordance with manufacturer’s instructions (MACS, Milteny-Biotec, 130-045-201). Cell viability was typically > 90% with purity > 98%.

Following spin-oculation, CD4 T cells were washed 3× in R10+. JRCSF-superinfected CD4 T cells and isolated CD8 T cells were counted independently, 2×, then results averaged. CD4 T cells were resuspended at 10⁶ cells/ml and CD8 T cells at 2×10⁶ cells/ml. To minimize cell loss from transferring cells, cell cultures were set-up in 5ml round bottom FACS tubes (12x75 mm style, Falcon, 352058) and subsequent intracellular staining at Day 7 performed in the same FACS tubes. For each participant examined, the following cell cultures, typically 13 in total, were established. All cells were cultured in R-10+ supplemented with 20 units/ml IL-2.

We adapted the cost-effective p24 intracellular staining protocol of Yang and colleagues (5). Highly statistically significant correlations have been reported between p24 intracellular staining and p24 culture supernatant levels as measured by ELISA (2, 5) and gag RNA measured by quantitative PCR (15); measures used in other viral inhibition assays. In this assay, cells were harvested at Day 7 and stained first with Zombie NIR™ Fixable Viability Kit (Biolegend, 423106), fixed with 4% paraformaldehyde (PFA)/lysolecithin (20 μg/ml) at RT, then resuspended in cold 50% methanol for 15min. Further permeabilization was achieved with 0.1% Nonidet P-40, and cells were then stained with antibodies to p24 antigen (KC-57- RD1, Beckman Coulter, 6604667) followed by antibodies to CD3, CD4 and CD8 receptors (conjugated to BV421, BD Biosciences; AF421, Biolegend and BV510, Biolegend respectively). Samples were acquired on a Fortessa and analyzed using FlowJo (version X10.0.7r2). Compensation controls were prepared for each fluorochrome using anti‐mouse Ig, κ compensation particles (552843, BD™ CompBead). (Gating strategies illustrated in Supplementary 1 ). The frequency of infected CD4 T cells was defined as the percentage of HIV-1 p24+ cells among live, single CD3+ CD8^negative lymphocytes using the uninfected CD4 cell culture that had a p24+ frequency of < 0.1 across all assays, to set the p24 gate.

Two sets of HIV peptides were generated (Sigma-Genosys, USA): 18-mer peptides overlapping by 10 amino acids were synthesized (Sigma-Genosys, USA) to match the HIV Clade B consensus sequence (386 peptides) and previously defined HIV CD8+ optimal peptides (9- to 11-mer peptides) (16). Optimal CD8 T cell peptides describe previously defined HIV epitopes were grouped by protein, 109 Gag/Nef (CTL-A) peptides or 103 non-Gag/Nef (CTL-B) peptides. A reactive/positive T cell response to a peptide pool was defined as the average of replicate wells > 30 SFU/10⁶ PBMC and 4× the average of mock wells (16). Zero values were not accepted in any replicate of antigen-stimulated wells. No data were excluded due to high background.

Intracellular cytokine staining (ICS) was performed as previously described (17). Similar to the VIA assay, cells were cultured and stained in 5ml round bottom FACS tubes. Briefly, a pool of peptides was added to 1 × 10⁶ PBMCs in 0.4ml at a final peptide concentration of 1.0 μg/ml. Cells were immediately stained with CD107a-APC, monensin/Brefeldin A and cultured for 6 hr at 37°C, 5% CO₂. Next, cells were stained with Zombie NIR viability dye at room temperature for 20 min, then labeled with CD3, CD4, CD8 (conjugated to PerCP-Cy5.5, Biolegend; PE-Cy5, Biolegend and BV510, Biolegend respectively); and BV650-conjugated CD14, 16, 19 and 56 (dump channel, Biolegend) at room temperature for 15 min. Cells were fixed, permeabilized and stained intracellularly with IFN-γ, TNFα, MIP-1β, perforin (conjugated to PE, Biolegend; PE-Dazzle 594, Biolegend; PE-Cy7, BD Biosciences; BV421, Biolegend respectively). In all assays, fluorescence minus one controls (FMO) were used for gating of CD107a and cytokines. Samples were acquired on a Fortessa and analyzed using FlowJo (version X10.0.7r2). Compensation controls were prepared for each fluorochrome using anti‐mouse Ig, κ compensation particles (552843, BDTM CompBead). For all cultures, >30,000 CD8 positive events were acquired and in the final gate, a minimum of 15 events.

Percent CD8 T cell-mediated virus inhibition is estimated by

where the ratio of f_{HIV – CD 4+ CD 8} to f_{HIV – CD 4} is interpreted as an estimated relative risk (RR). This relative risk is a ratio of two composite frequencies; the numerator (f_{HIV – CD 4+ CD 8}) is the composite frequency of p24+ cells in superinfected CD4 T cells co-cultured with CD8 T cells, and the denominator (f_{HIV – CD 4}) is the composite frequency of p24+ cells in CD4 T cells cultured alone.

These composite frequencies were computed by taking a weighted average of p24+ frequencies in CD4 T cells across all replicates using

where the weights, w_i , are the CD4 T cells in replicates 1 through n, and the frequencies (i.e., proportions or percentages divided by 100), p_i , are the frequency of p24+ cells in superinfected CD4 T cells in each of these replicates. Equation (2) is equivalent to taking the sum of all p24+ cells across all replicates and then dividing by the sum of all CD4 T cells across replicates. To produce two separate composite frequencies, this would be done separately for superinfected CD4 T cells cultured alone and for superinfected CD4 T cells co-cultured with CD8 T cells. See Supplementary Material 2 for this alternate formula.

For example, assume data for 3 replicates has been collected for a single participant. In superinfected CD4 T cells cultured alone, the participant has 23,000, 25,000, and 20,000 CD4 cells. The frequency of p24+ cells for these 3 replicates is 0.040 (4.0%) or 920 of 23,000 cells, 0.044 (4.4%) or 1,100 of 25,000 cells, and 0.047 (4.7%) or 940 of 20,000 cells, respectively. In co-cultures of infected CD4 T cells with CD8 T cells, the participant has 17,000, 18,000, and 15,000 CD4 T cells. The frequency of p24+ cells for these wells is 0.030 (3.0%) or 510 of 17,000 cells, 0.025 (2.5%) or 450 of 18,000 cells, and 0.023 (2.3%) or 345 of 15,000 cells, respectively.

Then, the frequencies needed for Equation (2) are calculated as

and as

Percent inhibition can then be calculated by substituting the two frequencies calculated in Equations (3) and (4) into Equation (1) to obtain

A simulation study was used to investigate several other weighting approaches including weights of w_i = 1 (simple arithmetic mean) and weights of w i = 1 / σ ^ i 2 (inverse variance weighting), where σ ^ i 2 represents the variance estimate of the estimated p24+ frequency for the i^th replicate. The p24+ cells were assumed to follow a binomial distribution, Binomial(N_i , p_i ), where N_i represents the number of CD4 T cells and p_i represents the frequency of p24+ cells for the i^th replicate. Thus, p_i was estimated by maximum likelihood as the ratio of p24+ cells to CD4 T cells with variance estimated by σ ^ i 2 = p ^ i ( 1 − p ^ i ) N i , where p ^ i represents the maximum likelihood estimate of p_i . Although inverse variance weighting is optimal in a statistical sense (18), simulations involving the experimental data resulted in approximately equal performance (measured by mean squared error) when weighting with CD4 T cells [Equation (2)] versus weighting with the inverse variance of the p24+ frequency.

Uncertainty of the estimates of percent inhibition can be quantified using the following 95% confidence interval (CI).

with lower and upper limits calculated using

where exp(·) is the exponential function and ln(·) is the natural logarithm. The standard error of ln ( f H I V − C D 4 + C D 8 f H I V − C D 4 ) , the natural logarithm of a relative risk, is estimated (19) as

where the totals (c_i ) in Equation (9) are calculated across all replicates as

c 1 =   t o t a l   p 24 +   c e l l s   i n   s u p e r i n f e c t e d   C D 4   T   c e l l s   c u l t u r e d   a l o n e ,

c 2 =   t o t a l   s u p e r i n f e c t e d   C D 4   T   c e l l s   c u l t u r e d   a l o n e ,

c 3 =   t o t a l   p 24 +   c e l l s   i n   s u p e r i n f e c t e d   C D 4   T   c e l l s   c o ‐ c u l t u r e d   w i t h   C D 8   c e l l s ,   a n d

c 4 =   t o t a l   s u p e r i n f e c t e d   C D 4   T   c e l l s   c o ‐ c u l t u r e d   w i t h   C D 8   c e l l s .

The SE approximation assumes that the totals (c_i ) for Equation (9) are sufficiently large. As a worked example, to compute the 95% confidence interval for % inhibition using the data in the previous section, the cell counts are first totaled to give:

c 1 =   920   +   1 , 100   +   940   =   2 , 960 ,

c 2 =   23 , 000   +   25 , 000   +   20 , 000   =   68 , 000 ,

c 3 =   510   +   450   +   345   =   1 , 305 ,   a n d

c 4 =   17 , 000   +   18 , 000   +   15 , 000   =   50 , 000.

These totals are then substituted into Equation (7) (lower limit) and Equation (8) (upper limit) to give

Finally, substituting these lower and upper limits into Equation (6) and solving gives

For this particular participant, we would report a percent inhibition of 40.0% (95% CI: 36.0%, 43.7%). It is important to note that these confidence intervals are not expected to be symmetric around the estimate due to the exponential function (i.e., non-linear transformation).

Per usual, the (frequentist) confidence interval for a particular sample may not contain the true value for percent inhibition. If data were collected on a large number of participants (e.g., 1,000 participants), then approximately 95% (or 950) of the confidence intervals computed using Equation (6) should contain the true value. Simulations with 1,000 iterations were used to calculate an empirical estimate of the coverage probability of the confidence interval ( Figure 2C ). These simulations demonstrate that the mean coverage probability (dashed red line) approximates the theoretical value of 95%.

An Excel template, titled “VIA Replicate Calculator” with formulas for computing percent CD8 T cell-mediated virus inhibition [Equation (1)] and associated 95% CI [Equation (6)] can be downloaded from our GitHub repository at (https://github.com/glab-hiv/via). The first worksheet in this template provides detailed instructions on how to dynamically apply these formulas to raw data. The formulas are written to appropriately update dependent on the participant ID and culture conditions, so it is not necessary that the user modify cell references separately for each participant. To use this calculator, users will paste their data into the “Main data” worksheet and then follow the instructions on the “Instructions” worksheet to obtain estimates for % inhibition and a corresponding 95% CI.

An Excel file titled Supplementary Material 3 “Master Supplementary”, also found at (https://github.com/glab-hiv/via), includes a sample dataset that contains only the columns necessary to compute virus inhibition and associated 95% CI. Users should employ a similar format for their raw data when using the template Supplementary Material 4 “VIA Replicate Calculator”. Detailed instructions are provided in Excel template.

To evaluate assay performance (for estimation of % inhibition) in future datasets, participant data was used to stochastically generate CD4 T cell and p24+ cells. To account for both clustering and overdispersion, a negative binomial mixed-effects model (with random effect for participant ID) was fit to the CD4 count data. Parameters from this model were used to stochastically generate CD4 counts from a negative binomial distribution. Lastly, the p24+ counts were stochastically generated from a binomial distribution, Binomial(N_i , p_i ), where N_i are the simulated CD4 T cells and p_i are the prespecified p24+ frequencies for the i^th replicate.

A schematic of the VIA protocol used in this study is provided in Figure 1 with a detailed protocol provided in Supplementary Material 1 . This protocol is a modified version of previously published VIA protocols in which HIV infection is measured by intracellular staining of CD4 T cells for HIV Gag p24 (2, 5). We used CCR5-tropic HIV JRCSF in these studies to mimic both the level of cell infection and more common co-receptor usage observed following in vitro infection with autologous or transmitted/founder viruses (20–22).

Supplementary Material 1 details modifications made to the protocol to minimize cell loss and improve assay reproducibility. Here we focus on a relatively simple yet impactful change to the protocol, specifically the introduction of independent culture replicates for both the HIV-CD4 and HIV-CD4:CD8 T cell co-cultures. Overall, for 6 cultured replicates, the coefficient of variation (%COV) of the frequency of p24 positive cells was less than 20% for the majority of participants (n=70, mean 9.20%, range 2.04 to 27.9%) in either HIV-CD4 T cell or HIV-CD4:CD8 T cell co-cultures in both healthy donors (HD) and PLWH on ART ( Figure 2A ). However, this level of variation across cultures can considerably impact the % inhibition that results if replicate cultures are not performed. This is exampled in Figure 2B which shows data for Participant ID (PID) 231. The %COV across replicates for PID231 was 2.5% for HIV-CD4 and 12.5% for HIV-CD4:CD8 culture replicates. However, calculating % inhibition using the highest p24 frequency in the HIV-CD4 T cell culture and lowest p24 frequency in HIV-CD4:CD8 co-culture (red text) or vice versa (blue text), resulted in a % inhibition of 34.9% and 56.6%, respectively. Differences in % inhibition following independent repeats of single-culture assays were even greater, producing up to 40% difference in % inhibition (data not shown).

The inclusion of replicate co-cultures to minimize intra-assay variability impacts the calculation of % inhibition in two ways. First, for each replicate, different numbers of cells were acquired (all cells were acquired to maximize cell yield). To account for the different contributions of each replicate culture to our analysis, we calculated a weighted average of p24+ cell frequencies across the HIV-CD4 T cell culture replicates. This weighted average was again calculated for the HIV-CD4:CD8 co-culture replicates. The averages were then used to calculate % inhibition as detailed in Methods.

Next, we constructed a confidence interval (CI) associated with % inhibition estimates generated from replicate cultures of both HIV-CD4 T cell cultures and HIV-CD4:CD8 T cell co-cultures. Simulations using the experimental data confirmed that 95% of the CIs contained the true value for % inhibition ( Figure 2C ). The derived 95% CI appropriately captures the variation in p24+ frequencies across both HIV-CD4 and HIV-CD4:CD8 T cell replicate cultures in our assay. In Figure 2D , we apply calculation of % inhibition [Equation (1)] and apply confidence intervals [from Equation (6)] to the data shown for PID 231 in Figure 2B . The calculated virus inhibition was 47.6% (95% CI: 45.2%, 49.9%). To examine whether these calculations improved reproducibility, we re-tested virus inhibition in PID231 three months later using a different lot of JRCSF. The calculated virus inhibition was similar at 45.5% (95% CI: 43.6%, 47.2%) indicating good reproducibility.

Using this assay, CD8 T cell inhibition of JRCSF infected CD4 T cells was assessed in HD (n=21) and HIVART donors (n=14). Across the cohort, HIV infection of CD4 T cells measured at day 3 post-infection ranged from 1.08 to 15.16% (median: 5.1%). On average, 990 (range: 240 to 2,742) p24+ cells and 22,500 (range: 4,271 to 52,394) CD4 T cells were acquired per participant examined (n=35). While we did not routinely examine the purity of cell isolations, the flow-based analysis enabled analysis of the CD4:CD8 T cell ratio following co-culture. In co-culture assays, set up in a ratio of 1 CD4 target cell to 2 CD8 T cells, the Day 7 ratio remained very close to 1:2 ( Figure 3A , %CD4 T cells in total T cells: HD T cell=34.7%, HIVART=32.1%, n=35), suggesting a high and consistent purity of CD4 and CD8 T cell isolation. These ratios also indicate that neither CD4 nor CD8 T cell populations were undergoing dynamic changes during co-culture (Days 4-7).

When % inhibition was then calculated using Equation (1), we observed that all HIVART participants produced a positive (> 0%) virus inhibition (n=14, range: 8.2% to 73.3% inhibition), consistent with our previous reports that HIVART participants maintain functional HIV-specific CD8 T cells over time (16, 23) ( Figure 3B ). The level of inhibition in HD was mostly negative (n=21, mean: -13.7%, range: -49.9 to 20.9%), though some participants produced % inhibition > 0%, suggesting false positives or non-specific CD8 T cell mediated virus inhibition ( Figure 3B ). The negative % inhibition values observed in HD reflected that higher HIV replication (i.e., higher p24+ frequencies) was observed in HIV-CD4:CD8 T cell co-cultures than HIV CD4 T cell cultures, suggesting the addition of CD8 T cells (that are not specific for HIV) enhanced HIV replication. Note, we also observed higher HIV replication in CD8-depleted PBMC compared with CD4 T cell-only cultures ( Supplementary Figure 1A ). Overall, HIVART participants produced consistently positive and higher levels of % inhibition than HD ( Figure 3B ) (n=35, p<0.0001, two-tailed, exact Mann-Whitney test).

A receiver operating characteristic (ROC) curve ( Figure 3C ) of study data demonstrates that the optimized assay can reliably predict values of specific CD8 T cell mediated virus inhibition >8% (rounded down from 8.24%). This threshold was determined by maximizing the true positive rate and minimizing the false positive rate. The estimates of within-sample sensitivity and specificity were 100.0% (14/14) and 90.5% (19/21), respectively. In other words, when this assay is repeated in the future with 6 replicates per participant, one can confidently detect HIV-specific CD8 T cell mediated inhibition for percentages greater than 8%.

CD8 T cell inhibition over time was measured in PLWH on ART ( Figure 3D ). Participants all initiated ART in chronic infection and at the time of VIA testing had been durably suppressed for 3.3-10.5 years. In each participant, CD4 T cell target cells from a single timepoint were cultured with CD8 T cells isolated at 4-5 timepoints over a 6-12months window. CD8 T cell inhibition was relatively stable over the time window tested, with %COV in six participants averaging 9.4%, range 0.9 to 17.3%. We examined whether % inhibition correlated with ex vivo IFN-γ production measured within the same timeframe ( Figure 3E ). Consistent with previous reports (10), the CD8 T cell inhibition of JRCSF did not correlate with the summed magnitude of T cell response to overlapping peptides spanning the HIV clade B proteome measured by IFN-γ ELISpot (n=12, r=-0.03, p=0.9; two-tailed Spearman’s rank correlation). Similarly, no evidence of correlation was observed between virus inhibition and either cytolytic function CD107a+perforin^low, (n=9, r=-0.08, p=0.8; two-tailed Spearman’s rank correlation) or MIP-1β+TNF-α (n=9, r= -0.09, p=0.8; two-tailed Spearman’s rank correlation) release of HIV-specific CD8 T cells measured in 6-hour ICS ( Figures 3F, G ). Note, IFN-γ measurements were correlated between the ELISpot and ICS assay ( Supplementary Figure 1B ) (24).

A final question, which is a common challenge in flow cytometry when examining low frequency cells is, ‘What is the minimum infection rate (%p24+) and minimum cells necessary to see a “real” change in percent virus inhibition?’. Given our use of replicate cultures, we also examined how this minimum changed depending on the number of replicates, i.e., were six replicates needed if either a higher number of p24+ cells were acquired or a higher % inhibition was observed?

Assuming 20,000 CD4 T cells were acquired, simulations were constructed so that the minimum acceptable HIV infection rate was 0.5% (p24+ cells in HIV-CD4 T cells). At 90% power, for 6 replicates, a minimum infection rate of 2.43% is needed to observe a true virus inhibition of 8% ( Table 1 ). Again, assuming a denominator of 20,000 CD4 T cells were acquired, this translates to a minimum of 486 p24+ cells ( Table 2 ). If the number of replicates is decreased to a standard of 3 replicates per participant, a higher infection rate of 4.63% (p24+ in CD4 T cells cultured alone) is necessary ( Table 1 ). This translates to a minimum of 926 p24+ cells to have 90% power to observe 8% virus inhibition ( Table 2 ). Given that 20% virus inhibition is more likely to achieve a true positive threshold than 10% inhibition (reminder, % inhibition is a relative risk calculation), the absolute number of p24 cells required decreases as this threshold increases. Percent virus inhibition ≥ 30% requires that only 100 p24+ cells (our minimum assigned threshold) are acquired, whether 2 through 6 culture replicates are used.