When does humoral memory enhance infection?

Abstract

Antibodies and humoral memory are key components of the adaptive immune system. We consider and computationally model mechanisms by which humoral memory present at baseline might increase rather than decrease infection load; we refer to this effect as EI-HM (enhancement of infection by humoral memory). We first consider antibody dependent enhancement (ADE) in which antibody enhances the growth of the pathogen, typically a virus, and typically at intermediate 'Goldilocks' levels of antibody. Our ADE model reproduces ADE in vitro and enhancement of infection in vivo from passive antibody transfer. But notably the simplest implementation of our ADE model never results in EI-HM. Adding complexity, by making the cross-reactive antibody much less neutralizing than the de novo generated antibody or by including a sufficiently strong non-antibody immune response, allows for ADE-mediated EI-HM. We next consider the possibility that cross-reactive memory causes EI-HM by crowding out a possibly superior de novo immune response. We show that, even without ADE, EI-HM can occur when the cross-reactive response is both less potent and 'directly' (i.e. independently of infection load) suppressive with regard to the de novo response. In this case adding a non-antibody immune response to our computational model greatly reduces or completely eliminates EI-HM, which suggests that 'crowding out' is unlikely to cause substantial EI-HM. Hence, our results provide examples in which simple models give qualitatively opposite results compared to models with plausible complexity. Our results may be helpful in interpreting and reconciling disparate experimental findings, especially from dengue, and for vaccination.

Fig 1. Conceptual diagram.

Classically, infected cells (I) trigger the growth of antibodies (A_c and A_N), and antibodies inhibit infection. In the case of ADE, antibodies can instead enhance infection (shown by green plus signs). In the case of suppressive memory (effect shown by the red minus sign), the cross-reactive response (A_c) directly inhibits the response of de novo antibodies (A_N); this is in addition to any indirect effect mediated by infection load.

Fig 2. Experimental and simulated in vitro ADE.

In Panel A, a single cycle West Nile virus, a flavivirus related to dengue, was grown in K562 DC-SIGN R cells with different molar concentrations of antibody which yielded peak infection load at intermediate concentrations of antibody. Dengue, grown for 24 hours in PBMC with different concentrations of dengue pooled convalescent serum, likewise shows a similar pattern, as seen in Panel B. In Panel C, our ADE model was used to simulate in vitro experiments in which antibody level is kept constant and infection load is measured after 2 days. Qualitatively all three panels show the same pattern: lower levels of antibody increased viral growth but higher levels were neutralizing. (West Nile data digitized from Fig 3A in [18]. Dengue data digitized from Fig 4A in [19].)

Fig 3. Passive antibody can enhance infection.

The figure shows experimental data, Panel A, and simulation results from the ADE model in Panel B. Adding lower levels of passive antibody (e.g. maternal antibody) enhances infection, but higher levels of passive antibody are protective. (Experimental data from Table 1 in [8].)

Fig 4. ADE but no EI-HM.

The figure shows simulation results from the ADE model. In these simulations, cross-reactive antibodies behave the same as de novo antibodies (s_c = s_n and k_ci = k_ni). Boosting the baseline level of cross-reactive humoral immunity does not enhance infection in these simulations despite the presence of ADE.

Fig 5. Less neutralizing, cross-reactive antibodies, modeled with k_c2 = 0.05, produce some EI-HM.

The figure shows simulation results from the ADE model when including cross-reactive antibodies. When the cross-reactive antibody is half as neutralizing as the de novo antibody (k_c2 = 0.125), there is no EI-HM. But, when the cross-reactive antibody is 5 times less neutralizing than the de novo antibody (k_c2 = 0.05), there is some enhancement of infection at low levels of baseline cross-reactive antibody.

Fig 6. High growth rates of the non-antibody immune response, such as s_R = 0.64, can produce dramatic EI-HM.

Panel A shows simulation results from the ADE model. When the non-antibody immune response is strong (s_R = 0.64), controlling infection several days before antibody levels become neutralizing, there is the possibility of dramatic EI-HM of up to 11 fold. In contrast when s_R ≤ 0.26, there is no EI-HM. Peak viremia has a greatly varying range from experimental challenges with dengue in rhesus macaques, as shown in Panel B. In these experiments peak viremia was greatly increased in secondary DENV2 infections but below detection in secondary DENV3. Both the simulation and the experimental data suggest the possibility of dramatic EI-HM but also its inconsistency. Experimental data from [20].

Fig 7. Suppressive memory can produce EI-HM even in the absence of ADE.

The figure shows simulation results from the model described in § ‘Suppressive memory’. In this case the combination of suppressive memory and lower potency of the cross-reactive antibody leads to EI-HM at lower levels of baseline cross-reactive antibody even though there is no ADE. The presence of non-antibody immune responses (s_R > 0) can lessen or even prevent this effect.