The majority of mutations do not cause changes in the amino acid sequence and there is evidence of negative selection in the evolution of most genes (Donker and Kirkwood, 2012; Song and Hao, 2009). Genomic regions that tend to accumulate missense mutations include those encoding key epitopes on the surface of the neutralization antigens. Such mutations have been implicated in changes to affinity to antibodies and likely contribute to the antigenic drift of field strains due to immune selection.

The reason behind this phenomenon is unclear. One explanation operates with genetic or phenotypic incompatibility between some variants of the neutralizing antigen specificities (Iturriza-Gómara et al., 2001). An alternative scenario is that there is in vivo selection of combinations that result in masking of VP4 or VP7 neutralization epitopes. Strains hiding their neutralization epitopes after reassortment may evoke a weaker neutralizing antibody response, thus enhancing their fitness over less advantageous antigen combinations (Bányai et al., 2009).

There are two major routes by which heterologous rotaviruses may be acquired by a new host: (1) direct transmission of the rotavirus strain to a new host species and (2) reassortment between the rotavirus strains of any heterologous host species. Most interspecies direct transmission events lead to dead-end infection. Fully zoonotic strains are rarely detected in surveillance studies of human rotavirus infections (Bányai et al., 2012; Dóró et al., 2014). The low transmissibility in alternative hosts may be partially governed by host genetic factors or viral genes, for example, the one encoding NSP1 (see Chapter 2.8). Zoonotic transmission coupled by reassortment is a more efficient means for the introduction of new antigen specificities to which humans are immunologically naïve. An example is the introduction of G9 VP7 specificity during the 1990s from an animal, most likely the porcine, host (Iturriza-Gómara et al., 2000; Mijatovic-Rustempasic et al., 2011). Reassortants carrying a single or few animal-origin genes within a genetic background typical of human rotavirus strains may be more transmissible in the new host.

The zoonotic potential of RVAs has important implications for human rotavirus epidemiology. The diversity of rotaviruses in any host species, including humans, is readily refreshed by independent interspecies transmission events at any geographic location, coupled with global transmission of adapted strains. Thus, eradication or even elimination of rotavirus infections through vaccination and other human-targeted interventions is unlikely due to the large number of host species whose homologous rotaviruses and gene pools are viable in the human host.

The discovery of serotype-specific epitopes on the other surface antigen, VP4, was an important finding showing that the two surface antigens elicit neutralizing antibodies independently. However, this initially had little effect on routine epidemiological studies given that reliable MAb panels to distinguish VP4 (P) serotypes had not been developed. A limited number of studies using cross neutralization or cross-reactive MAbs in EIAs revealed that many human rotavirus strains carried the P1 (P1A and P1B) serotype specificity. Furthermore, some preferred combinations between G and P serotypes were observed: the G1, G3, and G4 VP7 specificities were more likely associated with P1A, whereas the G2 VP7 specificity showed a tendency to combine with the P1B VP4 gene (Estes and Kapikian, 2007). In animal rotavirus strains, a different distribution of VP4 serotypes (eg, P9 in pigs, and P6 to P8 in cattle) and surface antigen combinations were identified (Hoshino and Kapikian, 1994; Papp et al., 2013b,c).

Another set of MAbs developed against the VP6 protein of RVAs led to the recognition of antigenic variants of VP6 (Greenberg et al., 1983). Subgroup I strains were most commonly seen among animal RVAs and some human RVAs, whereas subgroup II strains were detected in most human and some porcine and lapine RVAs. A few other strains, mainly of animal origin, were characterized as [subgroup I + II] or [subgroup nonI, nonII] (Hoshino and Kapikian, 1994).

Altogether, studies using cross-neutralization assays and MAb-EIAs revealed differences in serotype and subgroup distribution among various host species.

A significant reduction in costs of traditional (ie, Sanger) nucleotide sequencing has opened new avenues in rotavirus strain typing. Over the past 30 years, >40,000 nucleotide sequences of rotaviruses have been deposited in GenBank, primarily for RVAs. Nearly 30,000 sequences were deposited within the past 5 years, showing a very strong motivation to understand the evolution of RV strains in the postvaccine marketing era. Sequencing and phylogenetic analysis have uncovered new type specificities and significant diversity within rotavirus genotypes. In large-scale sequencing and phylogenetic analyses, several genetic lineages and sublineages could be distinguished within most, if not all, G and P type specificities. Furthermore, classification of rotaviruses into lineages and sublineages typically permits the distinction of strains collected from various host species.

Assessment of rotavirus evolution and molecular epidemiology has been made possible by some large-scale whole genome based studies. These studies clearly demonstrated the relative conservation in the genotype constellations of backbone genes (ie, other than VP7 and VP4) of medically important RVA strains. Reassortment among cocirculating strains belonging to identical genotype constellations are fairly common (Iturriza-Gómara et al., 2001) whereas reassortment events among strains belonging to different genotype constellations are typically rare (Esona et al., 2013; Mijatovic-Rustempasic et al., 2011).

Many whole genome based studies have focused on the diversity and evolutionary origin of unusual human rotavirus strains. These studies, together with the increasing amount of genome sequence information collected from animal host species, have permitted new insights into the evolution of some unusual genotype specificities, for example, the canine or feline origin of many human G3P[3] and G3P[9] strains. In addition, bats in Africa and China have been recognized to carry rotaviruses with both usual and unusual human genes. More recently, pigs have been identified as a major source and reservoir of strains infecting humans, showing a complex multidirectional transmission pattern between these two host species (Matthijnssens et al., 2008a; Papp et al., 2013a; Theuns et al., 2015).

Further improvements and cost reduction of high-throughput sequencing technology will facilitate whole genome based comparisons, determination of preferred genotype constellations, identification of vaccine-derived genes in field strains, and tracking of evolution of vaccine strains and other rotavirus strains in the postvaccination surveillance era (Bányai and Gentsch, 2014).

G1P[8] is the most common genotype, accounting for 38% of all genotyped strains (Bányai et al., 2012). However, the relative importance of G1P[8] strains has been declining in recent years. This genotype accounted for 65% of strains in studies published between 1989 and 2004 (Santos and Hoshino, 2005). The reasons for this decline are not well understood.

We can only speculate about the driving forces underlying these regional differences in rotavirus genotype distribution. Improved genotyping methods and increased surveillance efforts have led to the contribution of data from more and more countries in recent years, including greater representation from low income countries, which tend to exhibit a greater diversity of genotypes. Zoonotic transmission is likely responsible for the majority of rare and NT strains. Likewise, G8 strains appear to be closely related to bovine RVA strains, and therefore should be more common in countries where humans and animals live in close proximity, or where transmission can occur from unimproved water sources. Furthermore, mixed infections and infection with reassortant strains are expected to be more common in countries with a higher rate of rotavirus transmission.

In settings with 10 or more years of continuous surveillance of RV genotypes, cycling of the predominant genotypes is observed. The period of oscillations in the prevalence of different genotypes has been found to vary between 3 and 11 (or more) years (Pitzer et al., 2011a). Therefore, at least 4 years of surveillance data are needed to reliably determine the underlying distribution of RV genotypes in any given location.

Nevertheless, when RV strains collected from a single locale are compared to contemporary and archival sequences available for other regions of the world, the observed genetic diversity is similar (McDonald et al., 2012). The global spread of RV strains was particularly apparent following the recent emergence of G9 and G12 strains. In both cases, one particular lineage was responsible for the worldwide spread of these novel G-types (Matthijnssens et al., 2010). Within just over a decade, both of these strains had spread throughout the globe (Matthijnssens et al., 2009,  2010). However, larger than normal outbreaks of RVGE were not associated with the emergence of these new genotypes, which suggests that heterotypic immunity provides at least some protection against novel strains.

Although there is a broad scientific consensus that both vaccines are effective in reducing the burden of severe RVGE, there is still controversy regarding the long-term effects of vaccination on the circulating RV genotype distribution (Matthijnssens et al., 2012a). Vaccine effectiveness has been found to be equally high against both homotypic and heterotypic strains (Leshem et al., 2014). Nevertheless, transient changes in the genotype distribution have been observed following large-scale vaccination with Rotarix and/or RotaTeq.

The introduction of RotaTeq was followed by an increase in the prevalence of the G3P[8] genotype in some places, including the United States and some Australian states (Hull et al., 2011; Kirkwood et al., 2011), raising concern that this could be linked to lower rates of seroconversion to the G3 component of the pentavalent vaccine (Matthijnssens et al., 2012a). However, similar observations have not persisted. Other genotypes, including G9P[8], G12P[8], and G2P[4], have predominated for a year or two in certain locations, but overall G1P[8] has remained the predominant genotype in countries using RotaTeq.

For Rotarix, a consistent and more prolonged increase in the relative prevalence of the fully heterotypic G2P[4] genotype has been observed following large-scale vaccination (Dóró et al., 2014), particularly in Brazil, Belgium, and some Australian states. However, the magnitude and duration of the increase varied between locations. In Australia, the increase G2P[4] was seen for two seasons following vaccine introduction (2007–09) and again in 2012, but G1P[8] and G3P[8] were the most common genotypes in 2009–11 and 2013, respectively, in states using Rotarix. A clear increase in G2P[4] after vaccine introduction has also been noted in Belgium, and has been sustained for at least seven seasons, 2006–13 (Zeller et al., 2010; Pitzer et al., 2015). In Brazil, G2P[4] has been the predominant genotype for at least six years since vaccine introduction (2006–11), with prevalence ≥50% in all years except 2009 (Gómez et al., 2014). Furthermore, G1P[8] has been relatively rare, having been detected in ≤20% of samples in all seasons since vaccine introduction in Brazil (Carvalho-Costa et al., 2011), while another fully heterotypic genotype (G8P[4]) emerged in Northeast Brazil in 2012 (Gurgel et al., 2014). The high prevalence of G2P[4] RVA strains in Brazil after introduction of the universal RV vaccination with Rotarix was not unique but also recorded in other South American countries where no RV vaccination program had been established (Matthijnssens et al., 2009).

Finally, a number of vaccine-derived reassortant strains have also been isolated from individuals both with and without a history of vaccination. In particular, vaccine-derived reassortant G1P[8] strains have been described in patients from the United States, Australia, and Finland following vaccination with RotaTeq. Reassortment between wild-type G1P[8] and the Rotarix vaccine strain has also occurred (Dóró et al., 2014). This has important implications for surveillance, as more advanced laboratory methods may be needed to differentiate between wild-type and vaccine-derived strains.

A variety of mathematical models for rotavirus infections have been developed. These models vary in their structure and parameters based on different data sources and interpretations of the literature. Most models were developed with the aim of predicting the direct and indirect effects of vaccination. Different models predict similar reductions in RVGE in the first 5 years following vaccine introduction, but predictions for the long-term impact of vaccination have varied depending on assumptions about the relationship between age at infection and the severity and reporting of RVGE, as well as the duration of vaccine-induced immunity (Pitzer et al., 2012).

Models are also useful for explaining observed patterns of data. For instance, one model was able to demonstrate that the southwest-to-northeast pattern of rotavirus epidemics in the United States prior to vaccine introduction could be linked to underlying variation in the birth rate (Pitzer et al., 2009). Higher birth rates typical of states in the southwest lead to a quicker build-up of numbers of fully susceptible infants, who serve as important drivers of transmission. Models also predicted that rotavirus epidemics could become biennial at high levels of vaccine coverage, which has aided in the interpretation of patterns that have emerged since vaccine introduction.

Models have also attempted to explain the interaction among different RV genotypes, for example, modifying this in light of a couple of recent publications (e.g. Pitzer et al., 2011a). This model was able to show that the observed cycling of genotypes could be explained by differences in homotypic versus heterotypic host immunity. If immunity is stronger against a second infection with the same genotype as that causing the first infection (ie, if homotypic immunity is stronger than heterotypic immunity), then over time there will be a build-up of population-level immunity to the prevailing strain. This provides a selection pressure in favor of the rarer genotypes, allowing them to infect more people. Thus, over time one of the rarer genotypes is expected to increase in prevalence, while the prevailing genotype decreases in prevalence, and then the cycle repeats (Fig. 2.10.5).

Models suggest that vaccination with a monovalent vaccine, such as Rotarix is expected to exert different selection pressures on the rotavirus population than vaccination with a multivalent vaccine, such as RotaTeq. This selection pressure can occur even in the absence of strain-specific variation in vaccine effectiveness against RVGE (Pitzer et al., 2015). If the relative risk of infection with a fully heterotypic strain (ie, G2P[4]) is greater than the risk of second infection with G1P[8] following vaccination with Rotarix, then this can explain the predominance of G2P[4] in the years following vaccine introduction in countries, such as Belgium. Nevertheless, both vaccines are estimated to provide a broadly heterotypic immunity in most fully vaccinated infants and to remain effective despite any possible changes to the genotype distributions (Pitzer et al., 2015). In order to understand and predict the impact of vaccination on the distribution of rotavirus genotypes, it is important to take into account differences in the risk of infection (and thus transmission), and not just differences in vaccine effectiveness against RVGE.

Mathematical models are only as good as the data that are used to inform them. Often it is necessary to make simplifying assumptions in order for the models to be tractable. Individual rotavirus strains are continually evolving over time through evolutionary processes, such as point mutations and reassortment, which can be difficult to represent in a mathematical model. Nevertheless, models provide a “test-bed” for examining hypotheses regarding how individual-level characteristics of infection can affect the population-level spread of disease, and can help to highlight important unknown factors of outbreak developments.