Emergent Properties of the Human Immune Response to HIV Infection: Results from Multi-Agent Computer Simulations

by Ashley Callaghan


Results obtained from multi-agent computer simulations of the immune response to HIV infection suggest that the emergence of certain properties, or lack thereof, may play a critical role in determining the relative success of the response.

The normal immune response to infection by a virus or bacterium (antigen) is characterized by a complex web of interactions involving numerous cells and molecules of the host's defence system. The response results in the emergence of clones of cells and molecules that together manage to remove the antigen from the body. However, the situation with regards to infection by HIV is different. While the immune system generally manages to bring the viral load (ie the level of HIV) to low levels within weeks to months of initial infection, HIV is never completely eliminated but instead remains present in low concentrations. The reason HIV manages to escape the immune response is most likely due to two factors - its incredible rate of replication and high rate of mutation. New strains of HIV constantly emerge during infection as a result of the error-prone nature of replication and the rapid turnover of virus in infected individuals. These strains have the potential for immune escape, as the clones of cells that develop against the wild-type virus will generally lack the ability to respond to new strains of the virus.

The Need for Models of the Immune Response to HIV
The question now arises as to why we would wish to model this system? The main reason we need models of the immune response to HIV is that a lot of experiments that biologists would like to carry out in order to test hypotheses cannot be performed. The reasons for this include ethical considerations, and the fact that biological experiments are very often incredibly expensive and time consuming. If suitable computer or mathematical models existed which could faithfully reproduce what is known from the literature, then it may be possible to carry out certain experiments 'in silico' (ie computer-based simulations), instead of the more usual in vitro and in vivo experiments.

Multi-Agent Computer Simulations
With the computational resources available to us today, we are in a position to construct computational models that incorporate many of the key entities that participate in the immune response to HIV infection. The agent-based models we employ for our simulations model each cell as a unique entity with a set of characteristics that, together with clearly specified rules, define its functional behaviour. This approach allows us to incorporate a number of important characteristics, including cell-surface receptors, physical location in the system, affinity for a particular antigen, and so on. The use of advanced techniques such as parallel or distributed computing means we have now reached a stage where it is possible to design simulations incorporating cell population levels that are fast approaching the level found in the human body.

Emergent Properties during the Immune Response
To investigate possible causes that may explain why different individuals show such a wide range of responses to HIV infection, we performed numerous simulations in which different 'patients' were infected with the same quantity of an identical strain of the virus. By different patients, we mean different runs of the simulation, with the only variable being the random seed used to determine the characteristics of the various cells that together make up the immune system. In order to demonstrate the critical role that the emergence of a particular property plays in the relative success of the response, results for three of these hypothetical patients are discussed.

Figure 1a Figure 1b
Figure 1: Graphs of viral dynamics and immune response for three hypothetical patients during the acute phase of infection: (a (left)) viral load per mL of plasma; (b (right)) number of activated T Killer cells (CD8) per mL of plasma.

Figure 1a illustrates that the simulations quite accurately capture the viral load dynamics associated with the acute stage of infection as described previously. That is, a rapid rise in viral load is followed by a sharp decline to what is often referred to as a 'set point', resulting from the emergence of T Killer cells and HIV neutralizing antibodies.

Figure 1b shows the emergence of activated T Killer cells. These cells attempt to destroy cells infected with HIV before they have a chance to release new virions into the peripheral blood. Comparison of these two figures suggests that the emergence of a sufficiently large T Killer response is critical in bringing viral load to a low level at the set point. In contrast with patients 1 and 2, the T Killer response that emerges in the case of patient 3 is very poor, with the result that the viral load at the set point is significantly higher than in the other two cases. It has previously been shown that the higher the viral load at the set point, the worse the long-term prognosis with regards the progression of the patient to AIDS. This therefore suggests that the emergence of a sufficient T Killer response during the early stages of infection may play a critical role in determining the length of time before the patient develops full-blown AIDS.

Further Work
As noted previously, one of the most striking features of HIV is its very high rate of mutation. Throughout the course of the infection, mutant strains of the virus are constantly emerging. The relative success of these mutant strains across different patients is a subject that warrants future investigation.

Please contact:
Ashley Callaghan, Dublin City University / IUA, Ireland
E-mail: Ashley.Callaghan@computing.dcu.ie