Rabies remains a globally important zoonosis, despite being one of the oldest recognized infectious diseases (
27,
55). The majority of rabies in terrestrial animals and humans is caused by classical rabies virus (RABV), a lyssavirus in the family
Rhabdoviridae. Since the 1950s, many related lyssaviruses which are capable of causing clinical rabies have been identified. The majority of those viruses have been isolated from bats (Chiroptera), including four divergent viruses, which were isolated in separate geographic locations throughout Eurasia in the past 18 years (
2,
29,
31). The Chiroptera, therefore, represent a global reservoir for lyssaviruses, creating the potential for “spillover” infection to terrestrial mammals, including humans. Occasionally transmission between members of a new host species will occur, with potential for a subsequent adaptation in that species (
35). Phylogenetic evidence suggests that one or more host-switching events from bats into terrestrial mammals were originally responsible for the ongoing global epidemic of terrestrial RABV (
6).
Pre- or postexposure prophylaxis, using vaccination and passive immune globulin administration according to World Health Organization (WHO) guidelines, is currently the only effective way to prevent rabies after infection with a lyssavirus (
1). The efficacy of both active and passive immunization is likely to be affected by antigenic differences between viruses. The lyssavirus trimeric glycoprotein is the primary surface antigen, the major target for neutralizing antibodies (
8), and is involved in cell binding and entry (
34,
36,
53). Antigenic sites on the glycoprotein have been described using monoclonal antibody escape mutants (
8,
16,
47,
51). These studies have elucidated two major sites (sites II and III) and multiple minor sites. Although estimates of antigenic differences can be made using information regarding these known antigenic sites, protein structure, and amino acid properties, predictions of the relative importance of those sites and specific mutations within those sites cannot be quantitatively tested without a method to reliably measure antigenic effect.
The use of serological cross neutralization data to measure antigenic difference is limited by the reliability of the serological test and, more importantly, by paradoxes, or irregularities in the data. These irregularities include higher heterologous than homologous titers and individual variations between sera raised against the same antigen (
22,
52). Hence, serological data are considered to have low resolution, and they are often used only qualitatively. Despite these difficulties, studies have attempted to further quantify antigenic differences among lyssaviruses. Badrane et al. (
5) showed a linear correlation between the glycoprotein amino acid identity and four cross neutralization titers. Other studies have demonstrated variable serological cross-reactivity between European bat lyssaviruses (EBLV) and RABVs (
10,
11) and suggested that antigenic relationships between EBLV-1 and EBLV-2 may not be fully reflected in the genetic relationships (
41). Recent investigations into the efficacy of biologics against the Eurasian lyssaviruses showed an array of relatedness between lyssavirus species, with, for example, a murine anti-Aravan virus (anti-ARAV) serum neutralizing Khujand virus (KHUV) and ARAV equally but an anti-KHUV serum being less effective at neutralizing ARAV than KHUV (
22). Until recently, however, there were no established tools for the quantitative analysis of antigenic data.
Integrating antigenic data with direct sequencing data, here we quantify the antigenic and genetic variation among a global panel of lyssaviruses, including representatives from all lyssavirus species. Furthermore, we address two key issues in the development of antilyssavirus biologics: the appropriateness of animal models and the development of efficacious alternatives to human rabies immune globulins (HRIGs).
DISCUSSION
We have described the antigenic relationships among a panel of 25 lyssaviruses using serological binding assay data and antigenic cartography. This approach has quantified clinically important antigenic differences between lyssaviruses; shown that those differences are equivalent for mouse, rabbit, and pooled human sera; and allowed integration of quantitative antigenic data with genetic distances.
These data give a precise estimate for the correlation between genetic and antigenic distances (95% CI for
r = 0.81 to 0.88), an improvement in accuracy over the current estimate (95% CI for
r = 0.39 to 1.00 and
P = 0.08 calculated from published data [
5]). Increased accuracy allows the evaluation of glycoprotein amino acid sequence homology as a predictor of antigenic difference. Fitting a linear regression model to our data demonstrates that on average, a 4.8% change in the glycoprotein ectodomain amino acid sequence will cause one antigenic unit of difference between viruses (equivalent to a 2-fold change in antibody titer) (95% CI, 0.93 to 1.07 AU;
P < 0.001) (Fig.
5). A linear regression model applied to the log
2 of previously published data gives a similar mean of a 2-fold change in antibody titer per 5.5% change in glycoprotein amino acid sequence homology, but with a much larger confidence interval (−0.8 to 4.4 2-fold dilutions;
P = 0.296). Despite a good correlation between genetic and antigenic distances among the lyssaviruses, over 30% of the variance in antigenic distance cannot be predicted by the number of amino acid substitutions between viral glycoprotein ectodomains, illustrating the difficulty in interpreting antigenic differences using the gene sequence alone. Although these and previous studies have used the entire glycoprotein sequence, the techniques applied here could be applied to specific regions of the glycoprotein, for example, previously reported antigenic sites.
Integration of genetic and antigenic data allows identification of viruses where there are differences between genetic and antigenic relationships. For example, the antigenic positions of IRKV and ARAV, which are closer to EBLV-2, and KHUV, which is closer to RABV would not be expected from genetic relationships. Phylogenetic analysis using the glycoprotein ectodomain here, similar to that using the entire glycoprotein and entire nucleoprotein gene previously (
29), suggests that KHUV is more closely related to EBLV-2 than ARAV or IRKV is to EBLV-2. However, both genetic and antigenic studies are limited by the existence of only one representative of some viruses, a problem that also applies to divergent classified lyssavirus species such as LBV (
38).
Comparison with other RNA viruses illustrates the close antigenic relationships among the lyssaviruses. Studies of human influenza A(H3N2) viruses showed a similar correlation between the amino acid sequence of the main antigenic component of the virus (the hemagglutinin HA1 domain) and the antigenic distance between viruses (
r = 0.81), but a different slope. On average, only 2.9 amino acid substitutions (<1% of the hemagglutinin HA1 domain) caused a 1-unit change in antigenic difference (
52). Antigenic distances measured by different binding assays and for different pathogens may not be directly comparable, meaning that one antigenic unit (2-fold dilution) derived for influenza virus will not equal one antigenic unit (2-fold dilution) from lyssavirus studies. Influenza viruses and lyssaviruses both have trimeric surface glycoproteins. However, an average of 13 amino acid substitutions in influenza virus hemagglutinin will cause lack of cross-reactivity, compared with lyssaviruses, where viruses with as many as 100 amino acid substitutions still show significant cross-reactivity. Such conservation in the phenotype of the key antigenic determinant of lyssaviruses is consistent with a low immune selective pressure upon lyssaviruses (
9,
14,
24) in comparison with pathogens such as influenza virus that are under large selection pressure (
42,
52). Although low immune selective pressure could be expected from the natural history of RABV in terrestrial animals, where infection classically leads to death, the dynamics of RABV in terrestrial animals may not be applicable to bats, particularly in light of evidence of high seroprevalence against RABV and other lyssaviruses in apparently healthy conspecifics (
23,
30,
54).
A limitation of all
in vitro antigenic studies is the potential effect of adaptation to cell culture on the viral genomic sequence. Sequences generated for this study were taken from cell culture supernatant used in neutralization assays or after single passage of supernatant in mouse brain, to ensure valid comparison of genetic and antigenic data. Although we cannot rule out potential differences between those viruses in cell culture and the original isolate, previous studies using rabies virus have shown no change in the glycoprotein consensus sequence despite 20 passages in cell culture (
26).
All current rabies vaccine virus strains are based on classical rabies viruses (
18). Evidence suggests these vaccines are fully effective against virtually all RABVs tested to date (
4,
11,
33,
37,
46,
50) but not against distantly related viruses in phylogroup II or WCBV (
5,
43). In addition, the reported variable efficacy of vaccines against the EBLVs and DUVV (
11,
17,
32) and recent evidence of variable efficacy against ARAV, KHUV, and IRKV (
22) suggest that there may be a gradual loss of protection as viruses become antigenically distant from vaccine strains. This gradation in protection is the case for other viruses (
42,
52). Although protection can ultimately be tested only in challenge models, reliably quantifying the antigenic differences among divergent lyssaviruses is an important step toward predicting differences in vaccine protection (
5).
In addition to active vaccination, passive immunization against rabies remains a critical part of postexposure prophylaxis (
1). The gold standard, human rabies immune globulin (HRIG), is expensive and in short supply, prompting attempts to find less expensive alternatives, including cocktails of monoclonal antibodies (
21,
40). In some studies dose-dependent survival has been demonstrated following immune globulin administration in animal models (
21,
48), suggesting that the potency of passive immune globulin may be related to the neutralizing titer. If this is true, antigenic differences as measured by neutralizing antibodies are of direct relevance to protection provided by passive immune globulin treatments. HRIG titers correspond to rabbit titers and are therefore predictable using antigenic maps. Further investigation is necessary to determine whether candidate monoclonal antibody cocktails can be predicted with similar accuracy.
Sera from animal models are widely used to investigate antigenic differences between lyssaviruses (
5,
22). However, all require extrapolation to species of clinical interest, which has thus far been largely unsubstantiated. Here we have demonstrated that antigenic differences between viruses determined by a variety of species are equivalent. These findings validate the use of nonhuman experimental animals as a model for determining antigenic variation that is relevant to humans.
The Chiroptera are increasingly implicated as reservoirs for many zoonotic viral diseases (
12). The lyssaviruses provide one globally widespread example, with at least one antigenically distinct lyssavirus having been isolated from bats on all continents except Antarctica. The possibility of future elimination of lyssaviruses from bats by human intervention is at best very optimistic. The threat posed by lyssaviruses in bats is therefore global and likely to be continuous. Recent human deaths due to DUVV (
45) and EBLV-2 (
19) highlight the significance of spillover into humans. More alarming is the possibility of a bat lyssavirus adapting to a terrestrial host, as is hypothesized to have created the current global epidemic of terrestrial RABV (
6), with widespread and prolonged consequences. A recent example of a bat RABV switching its host to skunks (
35) illustrates the ease with which this adaptation can occur. With knowledge of the alarming clinical manifestations and extremely high mortality rates caused by lyssaviruses, along with increasing globalization and the altered interface between humans and wildlife, a better understanding of the antigenic as well as genetic relatedness among lyssaviruses is vital. The methods presented here provide a quantitative method to test predictions regarding the antigenic effects of amino acid substitutions, and those antigenic effects in turn can be used to make predictions regarding the efficacies of biologicals.
Acknowledgments
D.L.H., J.L.N.W., and N.L. were supported by the Cambridge Infectious Diseases Consortium, a DEFRA Veterinary Training and Research Initiative. D.A.M., L.M.M., and A.R.F. were supported by DEFRA grants SEO420, SEO423, and SEO424. D.J.S. and C.A.R. were supported by an NIH Director's Pioneer Award, part of the NIH road map for medical research, through grant DP1-OD000490-01; 223498 EMPERIE, an FP7 grant from the European Union; and program grant P0050/2008 from the Human Frontier Science Program. J.L.N.W. was also supported by the Alborada Trust and the RAPIDD program of the Science and Technology Directorate, Department of Homeland Security. C.A.R. was also supported by a University Research Fellowship from the Royal Society, London, and a research fellowship from Clare College, Cambridge. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
For technical support, we are indebted to the VLA Reagent Production Unit, the VLA Sequencing Department, Colin Black, Hooman Goharriz, Erasmus Medical Centre, and Eugene Skepner. We also thank Charles Rupprecht, Nicholas Johnson, Thomas Mueller, David Hayman, and Sharon Brookes for helpful discussions and contribution of reagents.