The goal of this study was to identify gene-specific determinants of MuV neuroattenuation and neurovirulence by constructing a series of recombinant full-length cDNA clones consisting of various combinations of genes derived from a highly neurovirulent wild-type MuV strain (88-1961 [r88]) and a highly neuroattenuated MuV vaccine strain (Jeryl Lynn [rJL]). Viruses rescued from these clones were then tested for neurovirulence in rats. The strategy for identifying gene-specific determinants of MuV neuroattenuation involved the construction of a cDNA clone of r88 in which r88 genes were replaced individually and in combination with corresponding genes from rJL. Exchange of the r88 F or P genes or the genomic 3′ and 5′ ends with the corresponding rJL-derived sequences had no neuroattenuating effect. In contrast, various degrees of neuroattenuation were achieved with all other individual gene replacement regimens (N, M, SH, HN, and L), ranging from an approximate 25% reduction in neurovirulence associated with N gene replacement to an approximate 50% reduction in neurovirulence associated with L gene replacement. The fact that individual rJL-derived genes were incapable of fully neuroattenuating r88 indicated that neuroattenuation of mumps virus is a multigenic trait. As expected, greater reductions in r88 neurovirulence were achieved with multiple gene replacements, ranging from an approximate 60% reduction in neurovirulence associated with the M/HN or F/HN gene combination to an approximate 90% reduction in neurovirulence associated with the N/M gene combination, a reduction far exceeding more extensive gene replacement combinations, including the full complement of envelope-associated genes (M/F/SH/HN) or all replication complex-associated genes (Le/N/P/L/Tr). Interestingly, the degree of neuroattenuation imparted by the N/M gene combination was substantially greater than an additive effect of the two genes alone (25% and 40%, respectively), suggesting a role of an N/M interaction in virus virulence. Given that the paramyxovirus M protein physically interacts with the N protein to facilitate virus assembly via associating the RNP with areas of the host cell membrane containing the viral glycoproteins (2
), it is noteworthy that in comparison to parental r88, the N/M gene combination chimeric virus grew to 100-fold-lower titers in vivo
. Whether efficient viral assembly or budding is impaired in r88 viruses expressing the rJL N and/or M genes requires further investigation. Notably, we cannot exclude the possibility that other combinations of replication complex- and envelope-associated genes not tested here would lead to nearly complete attenuation of the r88 virus as well, but the creation and in vitro
characterization of chimeric recombinant viruses representative of the absolute number of all possible combinations of rJL-derived and r88-derived genes was not practical.
Interestingly, neither the N/M gene combination nor most other genes that substantially affected r88 neurovirulence were capable of conferring neurovirulence to the rJL strain. This highlights what must be significant differences in the process of neuroattenuation versus neurovirulence induction, which has also been suggested in studies of rabies virus by Yamada et al. (48
). They demonstrated that the glycoprotein (G) gene from a neurovirulent strain was capable of converting an attenuated virus into a neurovirulent virus, yet the converse situation, i.e., replacement of the G gene of the neurovirulent virus with that from the attenuated virus, did not neuroattenuate the neurovirulent virus (48
). That none of the many r88 genes and gene combinations tested in this study converted the rJL strain into a fully neurovirulent virus contrasts with data from a study by Lemon et al., who reported conversion of the attenuated rJL virus to a highly neurovirulent phenotype upon combined replacement of the F and HN genes (a combination tested in our study) with those of the wild-type rodent brain-adapted Kilham MuV strain (21
). The bulk of the effect was imparted by the F gene, a gene that had no effect on the phenotype of either r88 or rJL in our hands. Thus, it appears that genetic determinants of MuV neurovirulence are not universal but rather are strain specific, as has been reported with rabies virus (40
A caveat to interpretation of results from this study is the possible influence on virus phenotype by random mutations that arise in the course of rescuing virus from cDNA and subsequent expansion in vitro
to generate a working virus stock. Notably, this is a concern for all studies involving generation of RNA virus stocks and is a consequence of the inherent low fidelity of RNA-dependent RNA polymerases (RdRp) (6
). The average mutation rate per genome per round of replication for RNA viruses has been estimated to be approximately 0.76 (13
). A careful examination of the chromatograms generated in our study revealed an average of two mutations (mostly in the form of nucleotide heterogeneity) in each virus stock, indicating the presence of variant virus subpopulations. The use of more sensitive technologies, such as massively parallel sequencing, would likely reveal many more sites of heterogeneity, reflective of the evolution of the cloned virus after numerous rounds of replication. It is well documented that different virus populations within a quasispecies can impact virus phenotype (8
). Although many of the mutations identified in our study were coding changes, in nearly all cases, the wild-type nucleotide was predominant, perhaps limiting any effect of these particular mutations on the measured neurovirulence of the virus. Dissection of the effect on virus phenotype by any specific mutation is complicated by the intrinsic high error rate of the viral RdRp. If one were to construct a new plasmid encoding any of the mutations of interest for further study, upon virus rescue and expansion, de novo
mutations would likely arise elsewhere in the genome, making it impossible to discern effects imparted by the engineered mutation versus those imparted by new mutations arising at different sites. In this study, we attempted to address the possibility of randomly arising mutations influencing virus phenotype by performing at least two independent virus rescues from each cDNA plasmid and to demonstrate phenotypic similarity between the two. This assumes that identical mutations do not occur upon rerescue. Indeed, of all the rescued virus pairs sequenced, we found only one instance of an identical mutation [G12590
A in r88+JL(L)-1 and r88+JL(L)-2] (), which most likely did not affect neurovirulence, given that the variant with the mutation was selected against within the first 3 days of replication in rat brain.
With few exceptions, repeat rescues of viruses behaved similarly in vivo. Significant intrapair differences in neurovirulence scores were seen only with viruses rescued from plasmids p88+JL(P), pJL+88(HN), and p88+JL(SH). Viruses rescued from p88+JL(P) and pJL+88(HN) ultimately did not appear to be of consequence, given that both viruses rescued from the former were highly neurovirulent, removing any argument supporting the independent involvement of the rJL P, V, and I proteins in significant attenuation, and for the latter, both viruses were highly attenuated, removing any argument supporting the r88 HN protein as a significant independent virulence factor.
The case is not as clear for the multiple rescues made from p88+JL(SH), which demonstrated much greater differences. Sequencing of these viruses revealed a total of two heterogeneous sites in r88+JL(SH)-1, both leading to coding changes in the L protein (S792P and M1035I), and a single heterogeneous site in r88+JL(SH)-3, also leading to a coding change in the L protein (R1078I). In all cases, the mutant nucleotide dominated over the wild-type nucleotide, and all were located within conserved domains, based on analogy to other paramyxoviruses and rhabdoviruses (41
). Thus, rather than the rJL SH gene being responsible for the observed attenuation, it is quite possible that the mutations arising in the L gene during virus rescue were responsible. While two heterogeneous sites were also identified in the nonattenuated r88+JL(SH)-2 virus (one leading to a coding change in the L protein [T1370K] and the other to a coding change in the P protein [Q244R]), in both cases, the wild-type nucleotide dominated, thus the consensus sequences of the L and P genes in r88+JL(SH)-2 were identical to that of the parental r88 virus (r88modΔNhe). Notably, several nucleotide heterogeneities also were identified in viruses rescued from the r88 cDNA clone containing the rJL N/M gene combination. It is important to note that these as well as the mutations identified in the other chimeric viruses most likely are of a random nature, rather than of a compensatory nature, given that we also identified similar frequencies of mutations in several parental nonchimeric viruses.
Intriguingly, in two viruses, a single nucleotide insertion was identified in poly(A) intragenic stop signals, one at the end of the M gene [r88+JL(N/M)-1] and one at the end of the F gene (r88modΔNhe-1). This phenomenon is most likely due to the stuttering of the RdRp at homopolymeric tracts and has been reported previously (6
). Addition of a nucleotide to the mumps virus genome would violate the “rule of six,” thought to be a requirement for efficient replication of those paramyxoviruses that edit their P gene (17
). However, the stringency of the rule of six has not been specifically examined for mumps virus, and in studies of related rubulaviruses, the stringency of this rule does not appear to be absolute (27
). Whether or not the two viruses in question here have compensatory deletions elsewhere, such as at the genomic ends, was not investigated due to expected complications in interpretation of results obtained from 3′ and 5′ rapid amplification of cDNA ends (RACE) results (35
). In any case, since both r88+JL(N/M)-1 and r88+JL(N/M)-2 replicated with the same efficiency and displayed similar neurovirulence scores, and since the neurovirulence score or r88modΔNhe-1 was almost identical to that of r88modΔNhe-2, the nucleotide insertions likely did not have any impact on neurovirulence.
A second limitation to interpreting the data from this study involves the cloning strategy employed, which involved the use of restriction enzyme sites located within the UTRs between ORFs, resulting in the generation of chimeric viruses composed not only of the foreign inserted ORF but also of portions of the flanking UTRs. This is an approach used by many, but the implications are often not considered. Therefore, we cannot rule out the possibility that the apparent gene-specific effects on virus phenotype were in part due to the effects of the UTRs that display sequence differences between rJL and r88. Western blot analyses of select constructs failed to identify an effect of changes in the UTRs on protein expression levels, but this needs to be explored further, including an assessment of mRNA levels by Northern blotting, which will be the focus of future studies. Notably, for technical reasons, this limitation is difficult to circumvent, requiring extensive correction in each clone by site-directed mutagenesis. Nonetheless, this deserves exploration.
Although the pathological mechanism underlying MuV neurovirulence is not well understood, in the rat model, we have previously demonstrated that the r88 virus grows to significantly higher titers in brain tissue than the rJL virus and that the severity of hydrocephalus appeared to be associated with the extent of virus infection of the ventricular ependymal cells and the ability to spread deeper into the brain parenchyma (31
). Nevertheless, it was unclear whether a strict correlation exists between maximum brain virus titers and neurovirulence scores. The recombinant viruses generated for this study provided a unique opportunity to investigate the relationship between virus replication potential and neurovirulence. In vitro
, we observed a poor correlation between virus growth kinetics and neurovirulence scores. Of note, all viruses that expressed the rJL F gene in general exhibited a higher fusogenicity in Vero cells than those expressing the r88 F gene. This caused lysis of fused cells and likely prevented the virus from further replication to reach higher peak titers. However, omission of all rJL F-expressing viruses in the regression analysis () led to an increase in the regression coefficient from 0.5 to 0.7 only (data not shown), indicating that fusogenicity is not a major determinant of virus titer and neurovirulence. Rather, replication complex-associated proteins, as expected, appear to be major determinants of peak virus titers in Vero cells. Nevertheless, the fact that we found a very good correlation between virus growth in the brain and neurovirulence might be due in part to the possibility that viruses expressing the rJL F gene might be unable to cause fusion of infected brain cells. These in vivo
replication data are consistent with previous findings suggesting that the relative neurovirulence of MuV is associated with the extent of infection of ventricular ependymal cells (the primary targets of the virus in rats inoculated intracerebrally) and virus spread from these sites into the subventricular parenchyma (31
). Notably, we did not examine virus growth in other rat tissues; thus, it is quite possible that a similar pattern of virus replication observed in the brain might also be apparent in extraneural tissues.
Results from this study demonstrating multiple virus proteins acting independently and in combination to affect the ability of virus to replicate in rat brain highlight the vast array of possible mechanisms of neuroattenuation. These results highlight the complexity of MuV neurovirulence and neuroattenuation and likely apply to other related viruses. From an applied perspective, further research in this area may provide a basis for the rationale design of new live attenuated vaccines, particularly for MuV, given the apparent difficulty in adequately neuroattenuating this virus, as indicated by the occurrence of aseptic meningitis in recipients of nearly all mumps vaccine strains, with the exception of vaccines derived from the rJL strain (29