Poster M6 - Exploring mechanistic parameters and clinical observations as potential biomarkers of response to CCR5Δ32 HSCT in HIV infection

Introduction: In July 2022, a fifth patient who received a potential HIV cure treatment was revealed to have no detectable HIV over a year after the cessation of anti-retroviral drugs. What all five apparently-cured patients have in common is that they received stem cell transplants – of umbilical cord blood or bone marrow or both – from donors homozygous for a deletion in CCR5. Because CCR5 is used for viral entry for some HIV subtypes, these CCR5Δ32/Δ32 stem cells generate CCR5Δ32/Δ32 immune cells that are resistant to viral entry and therefore do not get infected, do not produce infectious progeny virus, and reduce overall viral burden in the body. Though it has only been tested in a small number of patients, this evidence suggests the potential of hematopoietic stem cell transplantation (HSCT) using HIV-resistant cells in attaining long-term HIV remission. However, HSCT does not guarantee remission; six other patients who received allogeneic transplantation of CCR5-deficient stem cells presented poor outcomes and unfortunately all died within a year. Conversely, anti-HIV mutation may not be required to achieve HIV remission; a recent patient remained HIV-free after receiving a transplantation without the CCR5Δ32 mutation. Given the invasiveness of the procedures required to carry out HSCT, our mechanistic computational model allows us to simulate complex virtual clinical trials in silico, with validated virtual patients capturing variability across HIV patient population. Here, we explored potential biomarkers that drive the observed heterogeneity of therapeutic success under HSCT among the patients.

Materials and

Methods: Using clinical data from the Multicenter AIDS Cohort Study (MACS) cohort, our lab previously developed and validated a mechanistic computational model that captures the dynamics of cell-cell and cell-virus interactions during HIV infection. As part of this modeling process, the MACS patients were divided into four subgroups based on the length of time from HIV seroconversion to AIDS diagnosis. Within each of these subgroups we generated a population of 1,000 virtual patients, each with different parameter values defining their HIV model dynamics, but each fitting within the distribution of the clinical training dataset. These virtual populations were subsequently validated against another clinical dataset, and used to simulate potential therapies. We investigated potential biomarkers of two types: mechanistic parameters and clinical observations. We hypothesized that mechanistic parameters may be predictive of the therapeutic outcomes as they are fundamentally built into the model and define the interindividual variability (IIV) in infection dynamics between HIV patients. However, mechanistic parameters, such as the rate of infectivity of CD4+ T cells and the rate of homeostatic CD4+ T cell death, are difficult to measure in clinical settings. This led us to also explore the predictive power of clinical observations, which are not necessarily mechanistic, but are more readily accessible from the patients. Examples of clinical observations include: total CD4+ T cell counts prior to cART, after cART, and at the end of therapy.

Results, Conclusions, and Discussions: Using these virtual-patient mechanistic computational models of HIV infection, here we study how the outcome of anti-HIV stem cell therapy – its likelihood of success or failure – is not the same for each virtual patient and is driven by different model parameters. As previously mentioned, HSCT followed by cART cessation does not work for every HIV patient, so we asked: for whom is it most likely to work, and which parameters govern the response to therapy? In our simulations, we can vary the value of each parameter and observe how the change is reflected in the key metrics. Of the total 32 parameters in the model, we identified several that are potential drivers of HSCT response and therefore potential biomarkers of therapeutic success (measured by low viral load and high T-cell recovery). In other words, multiple parameters that when changed lead to decreased pre-ATI and post-ATI viral loads and increased time to viral rebound (or prevention of viral rebound altogether). In general, we found that most of these model parameters have low population variability and low sensitivity of key metrics. Global sensitivity and Partial Least-Squares Discriminant Analysis (PLS-DA) confirm that no single model parameter particularly determines the response to HSCT therapy; rather, the therapeutic outcomes are determined largely by the interactions between model parameters. In contrast, PLS-DA suggests that clinical observations are better predictors (i.e. better than individual mechanistic parameters) of which patients are most likely to benefit from CCR5Δ32-donor HSCT (Figure 1).

In conclusion, we quantified how model parameters and clinical observations in our mechanistic model drive the trajectory of HIV infection and therapeutic outcomes. We suspect that clinical observations are better predictors of therapeutic success because they serve as an integrator that combines information from many model parameters, and the interactions between them. Through understanding the relationships between therapeutic outcomes and model parameters in each virtual patient, the model and these biomarkers allow us to identify the best candidates for treatment, to better optimize a general approach to therapy, and potentially to personalize the optimal course of anti-HIV treatments for each patient.

Acknowledgements (Optional):