Main

The existence of multiple co-circulating strains or phylogenetic lineages is common for many pathogens, particularly for rapidly evolving RNA viruses. As viruses evolve, immune responses generated against a past variant may become less effective, which creates a complex system, with different antigenic variants interacting through the cross-immunity that is generated within hosts1,2. In the past decade, the increasing ubiquity of viral genetic data has created opportunities to interrogate how ecological processes, such as competition for susceptible hosts3, shape both the epidemiological and evolutionary dynamics of many viruses. In multi-strain dynamics, epidemics occur when a novel viral variant evolves and evades host immunity created by its predecessors4 or when the fitness of an existing variant is modulated by the changing immunity of the population independent of the ability of the virus to mutate5.

Potential ecological and evolutionary interactions among co-circulating viral strains are rarely investigated, particularly in animal populations, even though these interactions probably drive transmission dynamics through both immune-mediated competition and natural selection. These processes may ultimately shape the temporal and spatial distribution of viral genetic diversity across multiple scales, particularly when there are underlying spatiotemporal heterogeneities in the susceptibility of hosts as a result of previous patterns of viral circulation. The challenges in controlling influenza A in swine due to vaccine inefficacies and emergence of distinct divergent viral communities attributed to viral ‘mixing’ in pigs are a good example of the potential benefits of understanding multi-strain dynamics of viruses6,7,8,9.

While definitions of ‘strain’ vary widely and are often pathogen-specific, in this Review, we broadly define strain as when a pathogen occurs in identifiable phylogenetic lineages or clades that also differ phenotypically10. Immunogenic or antigenic phenotype variation may alter the fitness of a genetic variant in terms of its ability to compete with other variants. Phenotypic variation in virulence, transmissibility or other infection attributes may also confer fitness advantages (or disadvantages) and could be considered the basis for strain structure. While phylogenetic structure can be useful for reconstructing transmission history and patterns of dispersal, we would not consider the existence of phylogenetic structure to constitute multi-strain dynamics in the absence of phenotypic variation among lineages. We also do not consider the evolution of multi-strain dynamics unless those clades can potentially co-occur in the same host population.

While recent reviews of multi-strain dynamics of pathogens have focused on mathematical modelling frameworks for investigating strain–host interactions2,11, we synthesize immunological, ecological and evolutionary drivers and implications of multi-strain dynamics in rapidly evolving viruses. We first contrast conceptual differences and similarities between multi-strain dynamics from ecological versus evolutionary perspectives, then outline scales in which multi-strain dynamics occur and summarize immunological, phylogenetic, and mathematical modelling approaches used to quantify interactions among strains. While multi-strain dynamics may occur across a range of pathogens, we focus our discussion on multi-strain viruses. RNA virus–host systems are particularly likely to exhibit multi-strain dynamics because their high mutation rate allows for ecological and evolutionary processes to occur on the same timescale.

Ecological versus evolutionary dynamics

Although differences between ecological and evolutionary perspectives on dynamics of multi-strain pathogens is somewhat arbitrary given that both processes occur simultaneously, this conceptual division is useful in summarizing key theories and methodological approaches surrounding multi-strain dynamics. In both perspectives, past infection by one variant results in only partial cross-immunity to a related strain, and such partial cross-protection is expected to result in a change in the susceptibility, infectivity and/or clinical signs in the partially immune host. Ecological multi-strain dynamics generally encapsulate situations where a discrete number of antigenic alternatives or strains exist in the population and strains are assumed not to evolve phenotypically (only neutral or nearly neutral evolution occurs on the timescale of interest). Cross-immunity among strains is variable and fitness is frequency-dependent based on residual immunity in the population developed against previous strains, as has been suggested for human influenza12. Questions of interest focus on how and why the relative frequency of different strains changes through space and time. By contrast, evolutionary multi-strain dynamics focuses more on how competition and natural selection among genetic variants can drive genetic change, allowing for the emergence of new genetic variants or strains through time (Fig. 1). ‘Immune escape’ occurs when a novel antigenic variant evolves that is no longer controlled by individual/herd-level immunity13,14. In some instances, small mutations (resulting in minimal genetic change) may result in considerable antigenic changes if substitutions occur in immunogenic sites. In such cases, genetic distance may not be a useful measure of the extent of cross-immunity among strains.

Fig. 1: Influence of partial cross-immunity on evolutionary dynamics of multi-strain pathogens.
figure 1

a, Antigenic drift/shift: after infection by a specific variant at t0, the virus begins to accrue genetic mutations as it replicates, creating a viral cloud (t1). Partial cross-immunity can exert evolutionary selection pressures by which more-divergent variants are likely to propagate through time (t1 → t2, either within or between hosts) due to their ability to evade host immunity. b, This process can result in a shift in the antigenic phenotype of viral populations through time. c, Pathogens characterized by antigenic drift/shift often exhibit ladder-like phylogenetic trees wherein older strains go extinct and are replaced by newer strains, as suggested for influenza viruses122. d, Ecological antigenic shift: immunity in the population creates ecological pressure for antigenically divergent strains to increase in frequency through time (t1 → t2). e, This results in shifts in the antigenic phenotype as a new dominant strain in the population takes over. f, Pathogens characterized by ecological antigenic shifts probably exhibit more symmetrical/balanced phylogenetic trees with longer branches, as hypothesized for sublineages within porcine reproductive and respiratory syndrome virus type 248,61,122.

Full size image

Common to both perspectives are (1) the existence of variable levels of cross-immunity between strains or variants and (2) the fact that viral variants that are more effective at evading host immunity (induced by previous exposure to a related strain) have higher fitness and thus can outcompete other variants either within an individual or at the population level. Depending on the nature of cross-immunity, this can lead to fitness advantages for strains and variants occurring at low frequencies as compared with more common strains or variants towards which host immunity is already strong. Theory predicts that due to imperfect cross-immunity and frequency-dependent fitness among co-circulating strains, rare strains bearing novel antigenic mutations are expected to be able to spread more widely in the host population but then subsequently decline as herd immunity rises15. Cyclic or chaotic changes in the frequency of different strains occur in host–pathogen systems with intermediate levels of immune selection. These changes can complicate and restrict our ability to interpret and predict the outcome of interventions, including vaccination1,12,16 or selection for disease resistance traits in hosts, which is increasingly implemented in animal-based agriculture17.

Scales of action and impact of multi-strain dynamics

Multi-strain dynamics can be quantified across multiple scales, from within-host processes to host-to-host transmission within a single population or between populations. Both ecological and evolutionary processes can occur at each of these scales. Different scales are visualized in Fig. 2, where greater similarity in colour indicates hosts with higher levels of cross-immunity to each other’s viruses (based on past by exposure to more-similar antigenic viral variants) as compared with two hosts with more-divergent colours. Viral populations replicating within an individual host (Fig. 2a) form a viral cloud of highly related genetic variants, sometimes referred to as quasi-species18. Some genetic mutations may alter a variant’s antigenic phenotype, allowing immune escape to occur. Due to their ability to evade host immunity, the relative frequency of escape mutants may increase within the host and thus increase the likelihood that they are transmitted (however, see ref. 19 for a discussion of how within-host adaptation of viral populations may be detrimental to between-host transmission). Despite diminishing viral populations as infection progresses, surviving variants are likely to have mutations favoured in the immune-mediated selection process, as observed for porcine reproductive and respiratory syndrome virus20,21. An escape mutant that emerges from within-host evolutionary processes has the potential to propagate within the population due to its antigenic novelty against which the population has limited cross-immunity (light blue individual in Fig. 2b).

Fig. 2: Scales at which multi-strain dynamics occur.
figure 2

a, Strains within hosts. b, Strains among hosts. c, Strains among populations. Greater similarity of colours represents higher levels of cross-immunity conferred by exposure to more-similar antigenic viral variants. Thickness of arrows represents the relative likelihood of spread between hosts or populations with different immunological histories. Successful transmission of the light blue variant between hosts is more likely if the recipient host has previous exposure to a more dissimilar virus, such as the darker blue variant (b). Similarly, the dark blue variant is more likely to be transmitted to a population largely exposed to the dissimilar green variant than to a population exposed to the more-similar variant (c).

Full size image

Successful transmission of different variants between hosts can be influenced by bottlenecks in host susceptibility and infectivity. Reduced infectivity could be expected if the host’s immunity (for example, based on the history of exposure or host genetics) towards the viral variant is sufficient to reduce viral replication and shedding, making transmission less likely. Susceptibility bottlenecks can occur in between-host infection chains, for example, if past exposure to a similar variant influences a host’s susceptibility to a new variant. Because of variability in cross-immunity, for example, a host infected by the light blue variant in Fig. 2b is more likely to transmit to hosts that have immunity to more dissimilar variants (for example, darker blue) than to hosts with more-similar immunologic histories (Fig. 2b). Thus, the ultimate success of the light blue variant in spreading within the population is expected to be higher when the frequency of light blue is low because fewer individuals would have developed immunity against it. If we amplify this concept to consider between-population dynamics, we can expect heterogeneities in population immunity to shape the invasion success of different variants in new populations. For example, a dark blue variant may be much more likely to invade a population that has strong herd immunity towards green variants than a population that has an immunological history of blue variants (Fig. 2c). Across all these scales, antigenic novelty is expected to confer some degree of fitness advantage, which will allow more-divergent variants to propagate within and between hosts and consequently shape the invasion success of different strains across populations.

Quantifying immunogenic interactions between strains

Viral entry into cells can occur by various mechanisms, depending on the virus species, including fusion of the virion membrane with the cell membrane and receptor-mediated endocytosis22. Cell entry of RNA viruses typically involves binding of viral surface proteins to a cellular receptor22,23,24,25,26,27, which also triggers host immune responses25,28,29,30. These responses often include antibodies that bind to the surface antigens (proteins) on a virion to hinder binding of the viral proteins to the cellular receptors, preventing infection of the cell31,32. Consequently, whether infection with one strain of a virus can influence the host’s susceptibility to another viral strain may depend, among other things, on the extent to which the host’s immune system recognizes and inhibits infection with the newly infecting strain. Thus, partial immunity among genetically/immunogenically similar strains33,34 can shape the fitness of different strains and influence the likelihood that multiple strains co-circulate in a population.

To quantify antigenic distance between strains, binding and cross-neutralization assays are often used to measure the cross-reactivity of immune reactions elicited by different strains. Briefly, hyperimmune serum is generated against specific viruses (hypothetical strains A, B, C) by exposing naïve animals. Different viruses are then cross-reacted with serially diluted sera, and the highest neutralization titre is identified. A comparison between neutralization titres achieved by serum A on virus A (homologous titre) and the titres achieved for serum A against virus B or C (heterologous titre) can be interpreted as an indicator of antigenic difference/similarity between the strains35. These assays have been extensively employed in the study of influenzas, and cross-immunity profiles across a panel of different strains are often mapped through the application of antigenic cartography36,37,38. Antigenic cartography, a computational technique used for graphical visualization of antigenic distances obtained from inhibition assays39, can be used to visualize the genetic and antigenic differences among co-circulating variants and identify clusters of variants with similar immune profiles40. Data from panels of cross-reactivity assays can be combined with genetic mapping and epidemiological data and analysed using machine learning and other statistical approaches to identify specific amino acid changes that underlie antigenic phenotypes and potentially result in the emergence of different viral variants41,42,43. These tools can be used to refine the relationship between genetic and antigenic variation among co-circulating strains of a virus in a population.

Evolutionary processes of multi-strain pathogens

Evaluating the existence of multiple strains co-circulating in a population is a complex process because ecological and host factors may influence how evolution manifests in different strains, and interactions between strains against the backdrop of host population structure needs to be disentangled to understand the evolutionary trends of the virus44. However, quantifying the nature of multi-strain dynamics and drivers of co-evolution of multiple strains is of epidemiological relevance when dealing with outbreaks of infectious diseases in populations, as observed for SARS-CoV-245, dengue fever3,46 and influenza47. Multiple co-circulating strains can also influence disease severity through antibody-dependent enhancement or shaping the evolution of virulence (Box 1). In the following, we summarize different tools and approaches that can be applied to investigate evolutionary dynamics of multi-strain viruses.

Bayesian phylodynamic models provide a versatile framework for the study of pathogens over time through the inclusion of ecological or host-specific factors that may influence viral evolution in a landscape14,48,49,50,51,52, including how host traits, population structure and environmental characteristics impact the emergence, spread and turnover of viral populations53,54,55,56. The flexibility of including discrete or continuous traits50,53,57, estimating viral population change under structured coalescent models58 and including structured birth–death models59 in these analyses allows for more nuanced estimation of the prevalence of various mutations in viral populations. These methods also can be used to reconstruct viral population dynamics; identify emergence, population expansions and extinction events of different strains; and quantify the sustained co-circulation of distinct viral populations while accounting for variation in host population structure60. While the overall amount of genetic diversity through time may be somewhat constant for endemic multi-strain pathogens, analysing each lineage separately can help visualize these emergence–extinction cycles, as seen with porcine reproductive and respiratory syndrome virus61.

Phylogenetic branching patterns can be analysed to provide insights on multi-strain dynamics and immune-mediated selection. This analytical approach has been used extensively to describe the ladder-like phylogeny of seasonal influenza and some coronaviruses associated with immune-mediated selection62,63. In these examples, specific lineages of the viruses circulate over relatively short periods before being replaced by a new strain, creating a ladder-like tree (Fig. 1c). As a result, the most recent common ancestor for contemporary variants is relatively recent. However, phylogenies may not always exhibit step-like temporal topologies since ancestral clades may continue to persist even as descendant clades expand, as observed for foot-and-mouth-disease virus and different lineages within the same serotype of dengue virus64,65, with immune-mediated competition dictating the fitness of different clades through time (Fig. 1f). Two strains are expected to be antigenically differentiated to co-circulate within a host population without one going extinct. In addition, host genetic diversity, environmental heterogeneities, and spatial structure of the host population may contribute to diversifying evolution (increasing genetic and antigenic distance), with the impact of the latter two dependent on the pathogen’s transmission mode and dispersal capabilities66,67,68.

Selection pressures and resulting mutations responsible for adaptation or immune evasion are not always easily identifiable from phylogenetic trees alone69. Therefore, we describe four approaches that can complement phylodynamic models to evaluate rates of viral evolution depicted on phylogenetic trees: (1) Tajima’s D is a statistical test used to calculate the genetic deviation of a population from a neutrally evolving population70 and can be used to identify non-random mutations, bottlenecks and selective pressure driving the evolution process71. Tajima’s D relies on two measures of genetic difference between organisms: the mean pairwise differences in genetic sequences and the number of differentiating sites. (2) On the basis of the tree topology, fitness models can be used to estimate the rate of population expansion and fitness of a viral variant in a population72,73. The local branching index (LBI), for example, is a statistical calculation to estimate the fitness of a node (an ancestor) in a phylogenetic tree by calculating the size of a node’s neighbourhood (number of descendants/progeny of a node) over a given period in time73,74. Mutations that increase viral fitness are associated with higher LBI, and LBI has been shown to correlate with other metrics of fitness74. Since nodes with higher LBI are likely to be ancestors for future clades74, LBI can be used to predict expansions of different clades in a phylogenetic tree. Co-circulation of strains may be expected in cases where several contemporary nodes have near equal LBI. (3) The fixation index (FST) is a measure of changes in a population associated with the population’s genetic structure. Locus-by-locus FST using analysis of molecular variance can be used to identify potential genomic regions that determine the difference in accumulation of group-specific genes by a pathogen75,76,77. By comparing locus-by-locus differences, one can distinguish between groups of genomes isolated from a host population and determine the presence of one or multiple strains. (4) The rate of synonymous (dS) versus non-synonymous (dN) mutations can also elucidate dynamics of viral evolution78. Synonymous mutations are nucleotide substitutions that do not change the amino acid coded for by the respective codon while non-synonymous mutations result in changes in the amino acids. Synonymous mutations are generally considered neutral as they do not affect protein phenotype (although this is not always the case79,80,81), and the rate at which such ‘neutral’ mutations occur is typically interpreted as the expectation for background rates of change. Non-synonymous mutations may impact viral fitness if they are deleterious or beneficial, and thus may experience negative or positive selection pressures82. Calculating the codon-level dN/dS ratios can help identify whether selective pressure in a population is driving viral evolution. Higher-than-expected rates of non-synonymous change, usually inferred when dN/dS > 1, can be interpreted as evidence of positive or diversifying selection on that codon, suggesting that mutations resulting in amino acid changes are favoured83. Positive selective pressure at antigenic sites is indicative of immune-mediated selection. Combining dN/dS analysis with host/environmental factor analysis can further identify drivers of strain/variant co-circulation84.

Mathematical models of multi-strain pathogens

Mathematical models have been instrumental in understanding the dynamics of disease outbreaks and spread. They facilitate the estimation and prediction of changes in pathogen population size, the speed and duration of epidemics and the impact of control measures. Despite the ubiquity of strain structure85, models that incorporate such diversity have remained focused on a few prevalent human diseases, such as influenza4,5, human papillomavirus86,87, Dengue fever3,88 and human immunodeficiency virus (especially in the context of the emergence of treatment-resistant strains)89,90. Due to their inherent complexity and differences in assumptions about model structure, models of multi-strain disease can exhibit a wide variety of dynamics, from globally stable equilibria to cyclic or chaotic fluctuations in the frequency of different strains. Thus, multi-strain dynamics are difficult to predict.

Multi-strain disease models can track either individuals (agent-based models91) or changing proportions of different infection states (compartmental models), but the underlying dynamics are similar: individuals/groups of the population (hereafter just ‘individuals’ for simplicity) are divided into a finite set of possible classes on the basis of their exposure history. In the simplest case, this mirrors the commonly used single-strain SIR framework where each individual is either susceptible to a pathogen, currently infectious, or recovered and no longer capable of being infected or infecting others. Considering a pathogen with two strains, one might use an SI1I2R model in which individuals are delineated into one of four classes: susceptible to both diseases, infectious with one of the two potential strains or immune to further infection from either.

The preceding example highlights two key considerations that arise when modelling multi-strain diseases. First, what is the optimal model structure in terms of the number and resolution of the classes? This in turn depends on how one classifies previous infections (does it matter which strains an individual has been exposed to, the order of infection or simply how many?) and has dramatic consequences for the computational complexity of a model92. In addition, as with single-strain models, one must consider whether and how to implement population structure and heterogeneity among individuals (for example, differences in susceptibility)44,93,94. Second, how should cross-immunity be modelled? Cross-immunity can vary in degree (how much less likely is infection with strain B following infection with strain A?), duration (is immunity waning or lifelong?) and implementation (does immunity affect susceptibility or infectivity?).

In the face of this complexity, multi-strain disease modellers have frequently focused on systems with only two competing strains (for example, refs. 90,91,95,96,97,98,99,100) and employed simplifying assumptions, such as the discretization of a finite strain space. Put another way, strains are typically modelled as a set of strains that are all categorically different from one another (but see ref. 101). This is typically accomplished by assuming that infectious agents are clustered into functionally equivalent antigenic phenotypes85. Models of multi-strain disease are more disposed to non-stationary dynamics (for example, cycles/chaos) than are their single-strain counterparts102,103, driven largely by the degree of cross-immunity. When infection by one genetic variant provides near-complete immunity to another, stable and discrete strains emerge, whereas intermediate levels of cross-immunity lead to cyclic or chaotic fluctuations in strain prevalence1. Importantly, however, this effect can be overridden if strains differ too much in their epidemiological parameters103.

The incorporation of evolution into models of multi-strain disease introduces a wide range of additional complexities, but, in general, the framework for modelling evolving pathogens consists of two linked modules: one for the epidemiology, as discussed above, and one for the evolution. The proximity of this linkage depends on the nature of the evolutionary module, which can range from explicitly modelling nucleotide substitutions104 to allowing epidemiological parameter values to evolve (for example, transmissibility)99 or to adding a new parameter corresponding to an abstract phenotype101,105 or genotype space106,107. One of the more studied areas of multi-strain dynamics is the evolution and emergence of novel variants within a treatment and resistance paradigm.

To improve fit to empirical systems, some models incorporate spatial structure, which can promote strain coexistence108. Cyclical patterns of strain dominance, for example, can be produced in the absence of immune interactions if host population structure is introduced. In a model of Dengue virus, for example, spatial substructuring of the population explained stochastic differences between neighbouring areas in the prevalence of different serotypes, even in the absence of immune-mediated competition3. Finally, host contact networks can introduce another layer of complexity through the influence of local network structure on disease spread109.

Population structure and stochasticity

Host population structure can have major impacts on how multi-strain dynamics manifest by impacting the frequency with which strains serially infect or co-infect hosts. For example, host contact networks can impact the strength of immune-mediated selection pressure by influencing how rapidly the network becomes locally saturated with immune hosts109 and thus increase the likelihood of escape mutants to evolve. In virulence evolution (Box 2), the severity of the trade-off between competition among strains within co-infected subpopulations and transmissibility between subpopulations is reduced if between-population spread occurs frequently110. In other words, increased opportunities for viral dispersal between subpopulations may favour increased virulence110.

At the host level, superspreader hosts or events may play a large role in the spread of a specific strain of a virus111. In such cases, the spreading success of a strain may be more related to host behavioural or physiological attributes than to the fitness of that particular viral strain93. In highly structured livestock populations, for example, farms that ship high volumes of animals and occupy central positions in animal transport networks can disproportionately contribute to spread of a particular strain regardless of the fitness displayed by that particular strain112.

More generally, stochastic events may also be responsible for the apparent success of a given viral strain in a population113. Viral founder effects, population bottlenecks and superspreading events, for example, may influence viral populations in manners not clearly related to viral fitness114,115,116. Depending on how many viral particles are transmitted between two individuals, the transmission event itself may introduce stochasticity (random founder effects) in determining which strains transmit and persist. For example, multiple introductions of SARS-CoV-2 in specific populations leads to, at least in the beginning, outbreaks of strains that just happened to be earlier introduced rather than outbreaks of particularly fit strains117,118,119. Alternatively, transmission between hosts or populations may represent a selective bottleneck wherein a variant’s ability to be transmitted is mediated by characteristics of both the transmitter and recipient. Furthermore, the fitness of a particular variant is contextual and may not be the same within different hosts or populations, especially given hosts/populations vary immunologically, physiologically, behaviourally and genetically (Box 2).

Outstanding questions

Numerous unresolved questions need to be addressed to understand multi-strain dynamics in different host–virus systems. (1) With complex host immune responses and interaction with co-circulating strains, how do co-infection and co-evolution influence the effectiveness of disease management such as vaccination or other control strategies? (2) Although we have described different phylodynamic tools useful for understanding genetic evolution of co-circulating strains, what are the best approaches to investigate and contextualize antigenic evolution in those strains? In addition, are there distinct and measurable phylogenetic tree topologies characteristic of ecological multi-strain dynamics, and how do perturbations in host populations affect tree structure? (3) Host genotypes may non-uniformly influence susceptibility to certain pathogens. How do these host differences affect multi-strain pathogen dynamics at the population level? (4) Host populations may be stratified or substructured for many reasons (natural or artificial). Since strains theoretically evolve to balance transmissibility–virulence trade-offs specific to a given subpopulation, how do changes in host population structure affect the co-evolution/co-circulation of different strains in a population? (5) How quickly and to what extent does the fitness of a particular strain vary between individual hosts and across space and time? What are the most suitable approaches to quantify and predict the role of viral fitness in the establishment of multiple strains in a population or subpopulation? Can these tools be used to predict future success or invasion potential of different strains?

Concluding remarks

Although multi-strain dynamics are likely to occur in many rapidly evolving pathogens, the implications of immune-mediated competition among co-circulating strains for shaping spatiotemporal dynamics, maintenance of genetic diversity and emergence of novel variants are often overlooked. However, such multi-strain dynamics are critical for predicting the invasion success of novel genetic variants and anticipating outcomes of vaccination programmes. In this Review, we synthesized the interacting ecological and evolutionary processes that constitute multi-strain dynamics. To predict sequential or cyclic dominance of different strains, it is essential to understand the interplay between population immunity and the emergence of novel strains, as well as to understand the ecological dynamics among co-circulating strains that interact via frequency-dependent fitness advantages related to partial cross-immunity. Although the availability of sequence data has increasingly enabled studies of pathogen evolution and molecular ecology, examining the complex interactions occurring in multi-strain systems is challenging both theoretically and empirically. By highlighting the different components and scales of understanding multi-strain dynamic in viruses, we call attention to the need for more holistic studies in the future. Methodological approaches are rapidly developing113,120, with the evolution of SARS-CoV-2 variants now providing the quintessential exemplar of multi-strain dynamics (Box 3). However, there are many fundamental questions still to be answered to more fully understand the interplay between the immunology, evolution and epidemiology of multi-strain pathogens. Whereas previous research has focused largely on human host–pathogen systems, such as influenza1,5,104, dengue3 and rotavirus13, research on multi-strain dynamics in animal populations provides a rich area to further explore fundamental questions and generalizable insights for multi-strain pathogens121. Investigating these questions will improve our ability to anticipate the behaviour of multi-strain pathogens.