key: cord-0979449-rosvxp6c
authors: Miller, I. F.; Metcalf, C. J. E.
title: No current evidence for risk of vaccine-driven virulence evolution in SARS-CoV-2
date: 2020-12-03
journal: nan
DOI: 10.1101/2020.12.01.20241836
sha: 0698b9062bc4a95573d420192f8797117052e04d
doc_id: 979449
cord_uid: rosvxp6c

Vaccines that reduce clinical severity but not infection or transmission could drive the evolution of increased rates of pathogen-inflicted damage, or virulence. Preliminary evidence suggests that COVID-19 vaccines might have such differential effects, conferring greater protection in the lower respiratory tract, where viral growth most impacts severity, than in the upper respiratory tract, where infection is chiefly determined. However, the evolution of increased virulence can only occur under certain conditions, which include the existence of a positive association between transmission and severity linked to viral genetic variation. Here, we review the current evidence for these conditions, which does not point to a risk of vaccine driven virulence evolution. An evo-epidemiological model also indicates that upper respiratory tract protection can minimize or negate selection for increased virulence should these conditions be met. Despite low apparent risks, SARS-CoV-2 virulence should be monitored, and transmission-limiting characteristics should be prioritized for second-wave vaccines.

Since its emergence in late 2019, SARS-CoV-2 has spread globally, resulting in over 53 million cases of its associated disease COVID-19, and over 1.3 million deaths (1). Beyond the morbidity and mortality associated with the pandemic, nations and sub-national states heavily affected by COVID-19 have experienced catastrophic economic collapse, critical pauses in educational services, and pervasive psychological damage. Vaccine-induced herd immunity has been 5 consistently identified as the only acceptable course of action for mitigating the pandemic and allowing society to begin the recovery process. The race for SARS-CoV-2 vaccines has proceeded at an unprecedented pace, and several candidate vaccines have completed their 'phase 1' and 'phase 2' trials to confirm their safety and immunogenicity, and several have completed their final 'phase 3' trials to determine their efficacy to protect individuals from disease (2, 3). 10 While phase 1-3 clinical trials will provide the evidence necessary to confirm the ability of vaccines to safely prevent disease in individuals, potential population level and longer-term consequences of vaccine introduction, particularly viral evolution, should also be considered.

Potential negative outcomes include vaccine escape via antigenic evolution, which would result in a decrease or loss of vaccine efficacy (4) , and the evolution of increased virulence, which 15 could result in more severe health outcomes and a higher infection fatality ratio (IFR). In this paper, we focus on the latter outcome, and characterize the potential for vaccines to drive the evolution of SARS-CoV-2 virulence.

The evolutionary limits to rates of disease associated damage or mortality, termed 'quantitative virulence', have intrigued scientists for decades. In the absence of any associated 20 costs, pathogens are expected to evolve towards greater transmissibility but also reduced host damage (since mortality, and in some cases severe symptoms, curtail opportunities for transmission) to maximize both the rate and duration of their spread. However, for many pathogens, host damage either can enhance transmission (e.g. by inducing coughing) or is an unavoidable consequence of transmission (e.g. cellular damage resulting from viral replication). 25 Thus, incremental increases in transmission bear increasing costs of damage. This association results in a positive, saturating relationship between transmission rate and mortality (or severity) rate--the canonical 'virulence transmission tradeoff'--and has the effect of limiting the evolution of virulence (Fig. 1) . The saturating nature of this relationship makes an intermediate degree of virulence evolutionarily optimal, as the benefits of increased transmission are balanced against 30 the costs of host death (or other severe outcomes truncating transmission); a non-saturating . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted December 3, 2020. ; https://doi.org/10.1101/2020.12.01.20241836 doi: medRxiv preprint relationship would not limit virulence evolution (5, 6) . Direct evidence for the existence of virulence-transmission trade-offs is limited, but suggestive evidence has been found in many systems (7) . Considering this phenomenon in the context of vaccine deployment is important, as 15 theory predicts that immunity which reduces disease but not transmission (8), or reduces disease to a greater extent than transmission (9), can drive virulence evolution. Vaccinal immunity to SARS-CoV-2 could have this combination of effects, if its impacts on infection in the upper respiratory tract (URT) and lower respiratory tract (LRT) are not equal. The partitioning of virulence and transmission effects between respiratory tract compartments is known to exist for 20 other respiratory diseases such as influenza (10, 11) . In the case of SARS-CoV-2, protection in the LRT is thought to reduce disease severity, potentially reducing the costs of increased transmission. However, protection in the URT, the primary location of infection colonization, . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted December 3, 2020. ; https://doi.org/10.1101/2020.12.01.20241836 doi: medRxiv preprint might lead to a reduction in infection risk, or even complete sterilizing immunity, perhaps negating any gains in transmission rate the virus might be able to acquire via increased virulence (2) . This pattern of higher viral infectivity in the URT compared to the LRT reflects a decreasing gradient of angiotensin-converting enzyme 2 (ACE2, the receptor protein utilized by SARS-CoV-2 for cellular entry) expression from the URT to the LRT (12). Non-human primate 5 challenge studies investigating the efficacy of early COVID-19 vaccine candidates found that vaccinal immunity reduced viral replication in the LRT to a greater extent than in the URT (2, 13) . The patterns of lower URT than LRT protection for these candidate vaccines may be rooted in the type of immune responses they induce. All are delivered intramuscularly, which generally and predominantly stimulates the production of IgG antibodies. The LRT system is primarily 10 protected by these IgG antibodies, while the URT is primarily protected by IgA antibodies involved in mucosal immunity (2) . Other tissues where SARS-CoV-2 infection can cause acute injury (14) are protected by IgG antibodies, indicating immunological protection against damage in the lower respiratory and the rest of the body might be correlated.

While vaccines will be a critical factor shaping the landscape of immunity that drives 15 SARS-CoV-2 evolution, naturally acquired immunity could also contribute to selective pressures. Evidence suggests that similarly to early SARS-CoV-2 candidate vaccines, natural immunity provides greater LRT protection than URT protection, although the differential might not be as large. In macaques re-challenged with SARS-CoV-2, immunity eliminated or significantly reduced viral replication in the LRT, and reduced replication in the URT, but to a 20 lesser extent (15) . Consistent with this pattern, a longitudinal study of healthcare workers in the UK found that 0/1246 individuals with anti-spike protein IgG antibodies and 89/ 11052 seronegative individuals became symptomatically infected, indicating that natural immunity provides robust protection against disease (16). The same study also found that natural immunity provides substantial but incomplete immunity against infection, as the incidence of 25 asymptomatic infection was observed to be roughly four times higher in the seronegative group compared to the seropositive group (16). Natural infection stimulates the production of both IgG and IgA antibodies (2) , which likely explains how natural infection generates more balanced URT and LRT protection.

Differential effects of candidate COVID-19 vaccines in the URT and LRT are only one 30 of several features that would have to combine for the virus to evolve increased virulence in . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted December 3, 2020. ; https://doi.org/10.1101/2020.12.01.20241836 doi: medRxiv preprint response to vaccine deployment. First, increased transmission must be biologically feasible and a product of genetic variance (e.g. heritable). Second, evolutionary increases in transmission must lead to increases in virulence (e.g. virulence must be an unavoidable consequence of transmission). Third, if a trade-off exists between virulence and transmission, the current balance of the two traits must be such that increasing rather than decreasing transmission rate leads to 5 greater total pathogen spread and thus fitness. Fourth, selection for an increase in transmission must be sufficient to overwhelm the effects of chance (drift). Integrating these last two features requires constructing a model that includes both the epidemiological and evolutionary impacts of vaccination on the virus.

As the world prepares to rapidly roll out COVID-19 vaccines, characterizing the 10 likelihood of vaccine driven virulence evolution becomes critical for public health preparedness.

In this paper, we review the evidence surrounding the evolution of SARS-CoV-2 transmission and the existence of a trade-off between this trait and disease severity. Next, we present an evoepidemiological model that incorporates the differential effects of vaccinal and natural immunity in the URT and LRT and identify desirable features of vaccines that would minimize selection 15 for increased virulence while simultaneously providing personal and population level protection.

Virulence effects unassociated with transmission will only reduce pathogen fitness, because they diminish opportunities for spread. Therefore, we only expect selection for increased virulence if 20 mutations that lead to this viral trait are also associated with increases in transmission (a phenomenon called pleiotropy). The first question then becomes whether mutations in SARS-CoV-2 can increase viral transmission. While most SARS-CoV-2 genomic variants tracked to date are thought to be neutral (17), a few mutations have been posited as candidates for increasing viral spread. In particular, D614G, a mutation in the viral spike protein has rapidly 25 increased in frequency and become the dominant variant worldwide (18) . The spread of virus carrying this mutation has been attributed to increased transmission (19, 20) , and the D614G variant has indeed been found to enhance replication in vitro (19, 21) . Clinical data indicates that patients infected with the this variant have increased viral titers in the URT compared to those infected with the ancestral variant (20) , and data from a hamster model system suggests that the 30 increase in viral titers might be limited to the URT (21) . While this evidence indicates that it is . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted December 3, 2020. ; https://doi.org/10.1101/2020.12.01.20241836 doi: medRxiv preprint plausible that D614G increases SARS-CoV-2 transmissibility, difficulties in disentangling increased transmission from the impact of founder effects suggests that further evidence is necessary to establish a conclusive link to population-scale relevant increases in transmission (22) . However, these challenges also mean that it is possible that other mutations associated with increased transmission might be missed. 5 A mutation in the viral spike protein receptor binding domain, N439K, has arisen independently at least twice and is currently circulating in Europe and the U.S. When expressed in the background of D614G, this mutation is associated with higher viral loads in vivo, and faster initial replication rates in vitro compared to the ancestral variant expressed in the same background. N439K has also been found to reduce the neutralizing efficacy of monoclonal 10 antibody and convalescent sera therapies, although the consequences of this for transmission, especially from infected to immunologically naive individuals are unexplored (23) .

The next question is whether SARS-CoV-2 transmission and virulence are associated, and if so, 15 whether their relationship constitutes a trade-off. Direct evidence for such a linkage or trade-off between transmission and virulence in SARS-CoV-2 (e.g. paired evolutionary increases in both traits) is lacking. At the population scale, there is weak correlative evidence associating frequency of the D614G mutation and case fatality rates (24, 25), but further analysis is needed to confirm this association while accounting for changes in medical practices. Another mutation, 20 ORF1ab 4715L, shows similar patterns, but whether this mutation affects transmission has not been investigated. Clinical data suggests that neither the D614G mutation nor concurrent N439K and D614G mutations are associated with a change in the severity of disease (23, 26) . However, even if a trade-off (or positive association) exists, evolutionary patterns may not clearly indicate such a link if the virulence and transmission are sub-optimal (e.g. they fall on the interior of the 25 set delineated by the trade-off curve). Only after transmission is maximized for a given virulence strategy would evolutionary changes be expected to trace the curve defining the trade-off (Fig.1) .

Beyond such population scale data, growing empirical data on the within-host dynamics of SARS-CoV-2 infection provides an alternative window onto the relationship between transmission and virulence. Mechanistically, increases in transmission could result from an 30 increase in receptor binding affinity, faster replication rates, slower clearance rates, or . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted December 3, 2020. 

Uncertain evidence for genetic variation in SARS-CoV-2 transmission, combined with mixed 25 evidence for a trade-off between virulence and transmission all indicate no immediate cause for concern that vaccine-driven selection might drive the evolution of greater virulence in SARS-CoV-2. However, growing clarity around differences associated with transmission and severity in the URT and LRT, and differential effects of vaccines in these two compartments, combined with the unprecedented scale and speed at which the SARS-CoV-2 vaccine is likely to be 30 deployed point to considerable value in bounding expectations for the trajectory of selection on . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted December 3, 2020. ; https://doi.org/10.1101/2020.12.01.20241836 doi: medRxiv preprint the pathogen under different vaccine characteristics. Here, we develop a general theoretical framework to explore the degree to which different combinations of vaccinal protection in the URT and LRT (represented by parameters (,* and +,* ) might shape selection on pathogen virulence, if transmission can respond to selection and is involved in a trade-off with virulence ( Fig. 2) . We replicate our analyses across a range of assumptions about the shape of the 5 virulence-transmission trade-off by varying how optimal virulence, !"#$% , compares to observed virulence, !&' = 0.005 (Fig. 1) . Our analysis is motivated by SARS-CoV-2, but separation of contributions to virulence and transmission from different compartments (e.g., in URT and LRT) is likely relevant for a largely array of viruses. We expect that candidate COVID-19 vaccines will not be approved unless they confer robust protection against disease, indicating a high degree of LRT protection. Preliminary reports indicate that the first available vaccines may reduce the rate of disease by 90% or more (32) .

When vaccinal immunity confers this degree of protection in the LRT ( +,* > 0.9), we observe 5 that as the degree of vaccinal protection in the URT increases, either the strength of selection against greater virulence grows or the strength of selection for greater virulence shrinks. This pattern holds for all values of !"#$% , and its magnitude increases with vaccine coverage (Fig. 3) .

This suggests that robust URT protection, resulting in similarly large reductions in infection and transmission ( (,* > 0.75), is a highly desirable vaccine effect. Unsurprisingly, we find that (Fig. S1 ) and the strength of natural immunity (Figs S2-3) .

As the global spread of SARS-CoV-2 continues and the widespread rollout of early vaccines is imminent, evaluating the risk of vaccine-driven virulence evolution increases in priority. We synthesized information surrounding the factors necessary for virulence evolution to occur, finding evidence for the evolution of transmission but no definitive support for links 5 between transmission and virulence or a trade-off between the two traits. Evidence to date does not warrant concern about first wave vaccines driving the evolution of increased virulence. Our evo-epidemiological model identified URT protection as a desirable characteristic of 'second wave' vaccines expected to be administered to large portions of the population, as it would mitigate selection for increased virulence should transmission be subject to an evolutionary 10 tradeoff with disease severity. Overall, we find that vaccines that provide robust protection in both the LRT and URT could achieve the goal of reducing disease severity while minimizing prospects for the evolution of increased virulence.

URT protection would also carry significant public benefits, as it would mean that vaccines reduce transmission and lead towards herd immunity. This impact on transmission, While we focus on the evolution of quantitative virulence in this paper, other aspects of viral evolution should also be considered when evaluating candidate SARS-CoV-2 vaccines. 25 Antigenic evolution leading to 'vaccine escape' has occurred in response to other human vaccines (37, 38) and the prospects for similar outcomes in SARS-CoV-2 should be evaluated.

Antigenic evolution may also have implications for virulence evolution if it leads to antibody dependent enhancement (39) , consequent increases in viral load, and ultimately greater disease severity and potentially greater transmission. The evolution of resistance is theorized to be more 30 likely to occur in response to drug treatment than vaccination because drugs often target a single 13 pathogen epitope while vaccines induce an immunological response with a border target set (40).

All of the candidate SARS-CoV-2 vaccines that are in or have completed phase 3 trials induce an immune response targeted only at the viral spike protein (2) , and as such could be viewed as having effects intermediate between drugs and other vaccines. However, the necessity of the spike protein for cellular entry (41) might make SARS-CoV-2 vaccines unlikely to drive the 5 evolution of resistance relative to other vaccines that include many proteins. While evidence thus far suggests that low viral genetic diversity might limit the evolution of vaccine escape (17) 

Figures S1-S3

To evaluate the potential evolutionary consequences associated with SARS-CoV-2 vaccine candidates, we develop a compartmental ordinary differential equation model that broadly reflects SARS-CoV-2 epidemiology, and the protective effects of naturally acquired and vaccine-5 induced immunity (Fig. 2) . We extend previous work by explicitly separating the impacts of vaccinal and natural immunity in the upper and lower respiratory tracts, and the impact of these protective effects on infection, transmission, and mortality. We assume that infection in the LRT and URT additively contribute to transmission rate, (Eqn. S1). The parameter defines the fractional contribution of the LRT to transmission. The rate of removal due to disease associated 10 mortality (which might also reflect reductions associated with severity, changes in behavior, etc.)

is assumed to be equal to virulence, . Transmission rate in both the URT and LRT follow the classic increasing and saturating function of generally termed the virulence-transmission tradeoff ( Fig. 1) , which can limit the evolution of pathogen virulence as described above. As increases, transmission rate initially rises rapidly but then approaches an asymptote, while transmission Finally, as the URT seems to be the key driver of transmission, to model effects of immunity on transmission, we assume that force of infection ( ) experienced by vaccinated individuals is reduced by a scalar (1 − (,* ) for vaccinated individuals and by (1 − (,-) for individuals with naturally acquired immunity.

After defining the relationships between immunity, transmission, infection, and mortality, we incorporate them into a compartmental epidemiological model (Fig. 2B , Eqn. S1-S3).We set model parameters to broadly reflect SARS-CoV-2 epidemiology across a one week time-step, 5 which roughly corresponds to the viral serial interval, implying that the recovery rate = 1 (43).

Using mortality as a first benchmark for how transmission might be curtailed by pathogen week time-horizon negligible). As this model is aimed at making short term predictions about the effects of vaccination on selection for virulence, we ignore births, non-COVID-19 associated deaths, and the waning of immunity. Note that convalescent individuals can become reinfected in 15 the absence of immunological waning if immunity provides incomplete protection against reinfection ( (,-< 1).

The dynamics of this model hinge on the relationship between transmission and virulence. While the true nature of this relationship is unknown, we can construct a reasonable set of boundaries to the assumed form for trade-off (Eqn. S4, reflected in Eqn. S1) and 5 parameterizing it according to the estimated value of R0 and various assumptions about the optimality of the observed early-pandemic SARS-CoV-2 virulence strategy, !&' .

To accomplish this, we first construct a set of optimal virulence ( !"#$% ) values to explore that reflect assumptions that the observed degree of SARS-CoV-2 virulence ( !&' ) is either below the optimum, at the optimum, or above the optimum:

As is equivalent to IFR in our model parameterization, we set !&' = 0.005, which is the approximate overall IFR observed in the United States (45) . For a given assumption about the value of 15 !"#$% , we use the next-generation matrix approach (46) to set the value of b2 to that which maximizes R0 at !"#$% and b1=1. By definition, R0 is calculated assuming a completely susceptible population. Next, we used these same methods to set the value of b1 to that which corresponded R0 =2.5 at !&' . This assumed value of R0 is broadly consistent with a range of estimates (47-49). We checked to ensure that the value of b2 was insensitive to the value of b1, 20 that !"#$% maximized R0, and that !&' correspond to R0 = 2.5 at the identified values of b1 and b2. Figure 1 shows the virulence transmission trade-off curves constructed from the three assumptions about the value of !"#$% ; these assumptions map to a wide range of possible shapes of the trade-off.

To characterize the landscape of short-term selection on pathogen virulence in response to vaccination across a range of characteristics of vaccinal protection in the URT and LRT, we identify the strength of selection (fitness differential) associated with a doubling of virulence or the IFR. We initialize the model with 10% of individuals convalescent (roughly consistent with 25 observations from the U.S. in July 2020 (50)), and all non-vaccinated individuals susceptible.

For a given set of protective effects ( (,* , +,* ) , we calculated RE for !&' using the nextgeneration matrix approach. Next, we used the same methods to calculate RE for = 2 * !&' = 0.01. We then calculated the selection coefficient for increased virulence as the difference between RE at 2* !&' and RE at !&' . 

SARS-CoV-2 vaccines in development

Draft landscape of COVID-19 candidate vaccines

Monitor for COVID-19 vaccine resistance evolution during clinical trials

Selection of Intermediate Rates of Increase in Parasite-Host Systems

Coevolution of hosts and parasites

Virulence evolution and the trade-off hypothesis: history, current state of affairs and the future

Imperfect vaccines and the evolution of pathogen virulence

Vaccine-driven virulence evolution: consequences of unbalanced reductions in mortality and transmission and implications for pertussis vaccines

Pathogenesis of influenza virus infections: the good, the bad and the ugly

Linking Influenza Virus Tissue Tropism to Population-Level Reproductive Fitness

SARS-CoV-2 Reverse Genetics Reveals a Variable Infection Gradient in the Respiratory Tract

Animal models for COVID-19

COVID-19 cardiac injury : Implications for long-term surveillance and outcomes in survivors

SARS-CoV-2 infection protects against rechallenge in rhesus macaques. Science

A SARS-CoV-2 vaccine candidate would likely match all currently circulating variants

Structural and Functional Analysis of the D614G SARS-CoV-2 Spike Protein Variant

Tracking Changes in SARS-CoV-2 Spike: Evidence that D614G

Increases Infectivity of the COVID-19 Virus

A clade of SARS-CoV-2 viruses associated with lower viral loads in patient upper airways

Spike mutation D614G alters SARS-CoV-2 fitness

Making Sense of Mutation: What D614G Means for the COVID-19 Pandemic Remains Unclear

Evolution of virulence when transmission occurs before disease

Vaccine wagers on coronavirus surface protein pay off. Science (80-. )

Recurrent mutations in SARS-CoV-2 genomes isolated from mink point to rapid host-adaptation

Bouwmeester-10 vincken, F. Harders, R. Hakze-van der Honing

Jumping back and forth : anthropozoonotic and zoonotic transmission of SARS-CoV-2 on mink farms

Immune life history, vaccination, and the dynamics of SARS-CoV-2 over the next 5 years

Projecting the transmission dynamics of SARS-CoV-2 through the postpandemic period. Science (80-. )

Vaccine Escape Recombinants Emerge after Pneumococcal Vaccination in the United States

A systematic review of hepatitis B virus (HBV) drug and vaccine escape mutations in Africa: A call for urgent action

A perspective on potential antibody-dependent enhancement of SARS-CoV-2

Structure, Function, and Antigenicity of the SARS-CoV-2 Spike Glycoprotein

Code for: No current evidence for risk of vaccine-driven virulence evolution 5 in SARS-CoV

SARS-CoV-2 (COVID-19) by the numbers

Clinical Characteristics of Coronavirus Disease 2019 in China

Age-specific mortality and immunity patterns of SARS-CoV

On the definition and the computation of the basic reproduction ratio R0 in models for infectious diseases in heterogeneous populations

Reconciling early-outbreak estimates of the basic reproductive number and its uncertainty: Framework and applications to the novel coronavirus (SARS-CoV-2) outbreak: Reconciling early-outbreak estimates of the basic reproductive number and its uncertainty: Framework and applications to the novel coronavirus (SARS-CoV-2) 25 outbreak

The reproductive number of COVID-19 is higher compared to SARS coronavirus

No current evidence for risk of vaccine-driven virulence evolution in SARS

We thank Bryan Grenfell, Jeremy Farrar, Gordon Douglas, Daniel Douek, and AdrianMcDermott for helpful discussions. Figure 2A was created in part using BioRender.com.