Recombinant vector vaccines and within-host evolution

Many recombinant vector vaccines are capable of replication within the host. They consist of a fully competent vector backbone engineered to express an antigen from a foreign transgene. From the perspective of viral replication, the transgene is not only dispensable but may even be intrinsically detrimental. Thus vaccine revertants that delete the transgene may evolve to dominate the within-host population and in doing so reduce the antigenicity of the vaccine. We apply mathematical and computational models to study this process, including the dynamics of vaccine and revertant growth plus the dynamics of innate and adaptive immunity. Although the selective basis of vaccine evolution is easy to comprehend, the immunological consequences are not. One complication is that, despite possible fitness differences between vaccine and revertant, the opportunity for vaccine evolution is limited by the short period of growth before the viral population is cleared. Even less obvious, revertant per se does not interfere with immunity to vaccine except as the revertant suppresses vaccine abundance; the magnitude of this interference depends on mechanisms and timing of viral suppression. Adaptive immunity targeting the foreign antigen is also a possible basis of vaccine inferiority, but it is not worsened by vaccine evolution. Overall, we find that within-host vaccine evolution can sometimes matter to the adaptive immune response targeting the foreign antigen, but even when it does matter, simple principles of vaccine design and the control of inoculum composition can largely mitigate the effects. Author Summary Recombinant vector vaccines are live replicating viruses that are engineered to carry extra genes derived from a pathogen – and these produce proteins against which we want to generate immunity. These genes may evolve to be lost during the course of replication within an individual, and there is a concern that this can severely limit the vaccine’s efficacy. The dynamics of this process are studied here with mathematical models. The potential for vaccine evolution is somewhat reduced by the short-term growth of the vaccine population before it is suppressed by the immune response. Even when within-host evolution can be a problem, the models show that increasing the vaccine inoculum size or ensuring that the inoculum is mostly pure vaccine can largely avoid the loss of immunity arising from evolution.


Viral density
Revertant Vaccine (Antigen threshold) Figure 1: Independent growth of vaccine (blue) and revertant (green). The revertant virus has the superior growth rate, but in the absence of interference between the two, vaccine growth is unimpeded and immunity is triggered. growth and thereby suppresses antigen production. 103 The challenges are thus to understand (i) when and how much vaccine evolution occurs; (ii) whether and to 104 what extent that evolution affects the abundance of vaccine virus; and (iii) the extent to which change in the 105 vaccine abundance affects the generation of adaptive immunity against the antigen. The arguments presented 106 above are qualitative and only superficially identify the scope of the problem. Quantitative understanding 107 ultimately rests on analysis of mathematical models. However, as the models have many interacting processes 108 -minimally innate immunity, adaptive immunity and intrinsic growth differences between vaccine versus Perhaps surprisingly, therefore, an observation of intrinsic vaccine inferiority is not necessarily the norm. 115 Populations of recombinant viruses are commonly stable in culture, at least for a few transfers [9-20], but 116 potentially indefinitely [21,22]. Of course, short term population retention of antigen expression may mask an 117 underlying long term instability, so most of these observations merely set limits on the possible magnitude of 118 inferiority. Yet even if vaccine selective 'neutrality' turns out to be fleeting, merely a mistaken impression 119 from short-term observations, we will find that the phenomenon of short-term stability mirrors a solution to 120 minimize vaccine evolution within the host. levels. Again, the problem is complicated by the limited duration of the infection: reduced antigen production 127 due to vaccine evolution depends not only on interference between the two genomes but also on overall growth 128 and the extent to which it affects the level of immunity to vaccine and vector. A mechanism that forces 129 interference between vaccine and revertant can also limit the total amount of viral growth, thereby limiting 130 evolution.

131
Evolution of vaccine versus revertant thus depends on details, in particular, the specific mechanism by which 132 revertant interferes with vaccine growth. We describe three different mechanisms that have been proposed: 133 innate immunity, resource limitation, and adaptive immunity to vector components. For many vaccines, each 134 mechanism will impede revertant and vaccine equally as a collective population, thus ensuring interference. 135 136 It was initially believed, implicitly if not explicitly, that the adaptive immune response played the dominant 137 role in the control of viruses and other infections. In the 1990's, Janeway and Medzhitov identified shared 138 pathways for the control of pathogens between vertebrates and Drosophila, even though Drosophila lacks 139 an adaptive response [23]. This led to a resurgence of interest in the role of innate immunity in the initial 140 control of infections. Later modeling studies of influenza infections suggested yet another mechanism, that 141 the dynamics of these infections could be largely described by simple resource limitation models, of the type 142 used in ecology for population growth [24,25]. The realization that all three different processes might suppress 143 viral infection led to more careful examination of the roles of different factors in the early control of acute 144 infections [26][27][28][29]. The relative role of each mechanism in clearing infections is the basis of ongoing discussion, 145 but it is widely accepted that the roles differ among infections by different viruses and that each mechanism 146 is potentially important for some viruses. limitation is expected to affect vaccine and revertant similarly.

164
Adaptive immunity. 165 Adaptive immunity can be induced by the wild-type vector and the vaccine virus. Adaptive immune responses 166 specific to antigens expressed by the wild-type vector will presumably affect the vaccine and revertant equally 167 -because the vaccine encodes a complete vector genome, and the revertant is also a complete vector. As with 168 the preceding pair of mechanisms, adaptive immunity common to both revertant and vaccine will operate so 169 that revertant abundance will depress vaccine. Adaptive immunity to the vaccine antigen will be considered 170 shortly. 171 one mechanism may take precedence over the others, simply because it is activated earlier or enforces a lower 173 limit on viral density than the others. However, there are different stages or degrees of vaccine suppression, 174 so an early mechanism may act to control the infection without clearing it, and another mechanism may 175 act later to clear. Because of the delay in developing an adaptive response, viral suppression by adaptive 176 immunity typically occurs later than effects of innate immunity or resource limitation and so might seem  The preceding paragraphs omitted adaptive immunity to the antigen. By its very nature, adaptive immunity 182 suppresses vaccine growth. But adaptive immunity to the antigen is specific to the vaccine and is thus 183 another reason -besides intrinsic fitness effects -that the vaccine may have lower fitness than revertant.

184
The evolutionary consequences should be the same for both types of inferiority, reducing the long term 185 generation of antigen levels. But the interesting twist is that adaptive immunity to the antigen might feed 186 back negatively on itself to limit its own growth -immunity against a virus is intrinsically inhibitory, so 187 adaptive immunity against the vaccine will limit vaccine growth and thus limit antigen build-up that would 188 fuel further immunity. One question is whether this self-inhibition is worsened with vaccine evolution.

189
The effect is biologically complicated because adaptive immunity to the antigen does not necessarily translate 190 into selection against the vaccine. Selection against the vaccine per se operates only when adaptive immunity 191 specifically targets the vaccine genome over the revertant genome, and this selection need not occur -either 192 because adaptive immunity is so delayed that it is never manifest during vaccine growth, or because the 193 antigen is physically decoupled from its genome when attacked by the adaptive response. Without imposing 194 selection on the vaccine, antigen-directed immunity will not affect vaccine evolution.  196 We now employ quantitative models to evaluate the intuitive ideas presented above. Given the high 197 dimensionality of the problem, we are especially interested in how well intuition works and whether generalities  The models assist us by forcing us to specify assumptions for how the viruses and immunity interact, and 208 by allowing us to rigorously explore outcomes in different scenarios. However, there is uncertainty in the 209 model structure, many parameter values are unknown, and different viruses will behave somewhat differently.

210
Consequently, we focus on broad generalities that arise from many simulations and illustrate these for a 211 few specific cases, reserving the supplement for further details. The presentation below briefly discusses 212 the individual dynamics of individual trials for illustration but then moves to plots that reveal differences

Evolution from intrinsic fitness effects can matter 217
In the trials used for illustration, we allow innate immunity to control the infection and adaptive immunity  The dynamics of virus and immunity are shown in the absence of revertant (i.e. no evolution). (Right) The revertant is included, but at two different levels. The solid lines correspond to little evolution: the vaccine has a small cost (intrinsic cost =1%, initial level of W is 0.1 that of initial vaccine, and the mutation rate is 10 − 6 per day). The dotted lines correspond to major evolution: the vaccine has a 20% intrinsic cost, the mutation rate is 10 −3 , and the initial level of the revertant is 10 fold that of the vaccine.   245 We focus on infections of short duration. Factors that limit the duration of infection include resource 246 limitation, and innate and adaptive immunity. For the most part these factors act equally against vaccine 247 and revertant virus. Only one factor, adaptive immunity to the vaccine antigen (X), acts specifically on 248 the vaccine virus and not the revertant. Intuition suggests that this adaptive immunity to the antigen 249 can potentially suppress the vaccine's growth and give an advantage to the revertant. As with intrinsic 250 fitness costs, this selection might feed back to limit vaccine growth and thus limit the development of further immunity by allowing revertant to grow and interfere with vaccine. This section considers whether these 252 arguments are supported by the model.

253
Any real vaccine that elicits immunity against the antigen may also experience an intrinsic fitness cost.

254
The effect of immunity on evolution would then be confounded with the effect of intrinsic fitness effects on 255 evolution, making it difficult to isolate one from the other. The models do not face this problem, however.

256
They can be parameterized so that the only possible selection against the vaccine comes from immunity (by 257 setting c = 0). Vaccine populations can also be freed of revertant by omitting revertant from the inoculum 258 and setting the mutation rate to 0. Thus, we can measure the effect of adaptive immunity on vaccine growth 259 from trials that lack revertant and then compare those results with trials that include revertant.

260
There are several background points to note about the model structure. First, adaptive immunity specific 261 to vaccine (X) develops at a rate proportional to the vaccine abundance (V ) and parameters s and φ X .

262
In contrast the impairment of vaccine growth depends on the level of immunity (X) and the parameter 263 (k X ). Thus, immunity can develop even when there is little or no impairment, i.e., when k X → 0. Second, than vector -revertant is interfering less.

289
In sum, therefore, immunity to the vaccine (X) is reduced by itself and by evolution (presence of revertant).

290
The self-limiting effect of anti-vaccine immunity depends heavily on the impairment parameter. The two 291 effects do not interact to make the problem worse than from their separate effects. is high). Second, the vaccine should elicit a large response to this antigen. This requires that the antigen 300 rapidly elicits immunity (i.e. has low φ X , and in terms of immunology it should be an immunogenic antigen), 301 and also requires a high vaccine viral load to generate a large response. Engineering this requires tackling 302 a trade-off between avoiding vaccine clearance (i.e. having a low k X ) but allowing for rapid clearance of 303 the pathogen (having a high k P ). Vaccines designed to express the antigen in a form that is different from 304 that in the pathogen might help solve this problem. Thus, to elicit immunity to influenza, one might design   The effect of revertant frequency in the inoculum is evident in Figure 6: the magnitude of immunity to the 322 vaccine increases by orders of magnitude as the initial frequency of the revertant is decreased.

323
Evolution can also be reduced by increasing the inoculum size. To achieve a threshold antigen level, a large  Composition of the vaccine has the larger effect for these parameters, as indicated by the contours being more horizontal than vertical. An intrinsic fitness cost of c = 0.1 was set for these trials. Smaller c values would lead to higher vaccine and immunity levels across the graphs.
Whether and how well controlling the inoculum will work in practice will depend on details. Solutions may 333 be quantitative rather than absolute. Intuition is useful for guidance but needs to be confirmed by formal 334 analyses, guided by data for the specific implementation.  Our results revealed that that for a broad parameter regime, within-host evolution is unlikely to cause 360 a significant loss of vaccine efficacy (i.e. reduction in the level of immunity to the inserted transgene).

361
Furthermore, undesirable consequences of vaccine evolution may often be easily remedied by ensuring the 362 frequency of the revertant virus in the inoculum is low and by increasing the size of the inoculum. We 363 also suggest that further gains in vaccine efficacy can be achieved by appropriate engineering of the vaccine 364 antigen, allowing it to elicit immunity that clears the pathogen but not the virus vaccine, although such 365 engineering may not be easy.

366
One major outcome of our analysis was that intuition about vaccine evolution was not easily translated into 367 intuition about immunity. Indeed, even intuition about evolution often failed because that intuition was based 368 on vaccine versus revertant fitness, but the vaccine growth phase was short enough that differential fitness 369 had little effect on evolution. Even more fundamentally, intuition sometimes failed because the development 370 of immunity to vaccine could be unaffected by the revertant. Thus, our intuition suggested that vaccine 371 inferiority could stem from both an intrinsic fitness disadvantage and a disadvantage due to adaptive immunity 372 to the transgene/antigen. Both effects were found to impair the development of immunity to vaccine, but not 373 necessarily for the reasons suggested by our intuition.  Resources start with a fixed amount and are depleted by vaccine and revertant growth, without replenishment: The vaccine virus grows on resource R at rate r, depleted by mutation, death, and all 3 types of immunity: Adaptive immunity specific to vaccine grows according to its present value and a discounted value of the 438 current vaccine density: Adaptive immunity common to vaccine and revertant grows according to its present value and a discounted 440 value of the current vaccine plus revertant densities: Innate immunity, also common to vaccine and revertant, grows according to current levels of vaccine and 442 revertant, with diminishing growth as a limit is approached. Innate immunity also decays: These models follow the usual assumptions of SIR models, except that susceptible hosts (host cells in our case) 444 are modeled as Resource. As is typical in these models, variables for 'free' virus are omitted, an assumption 445 based on the quasi-steady state approximation (Perelson 2002