Antigenic diversification is correlated with increased thermostability in a mammalian virus

Abstract

The theory of plastogenetic congruence posits that ultimately, the pressure to maintain function in the face of biomolecular destabilization produces robustness. As temperature goes up so does destabilization. Thus, genetic robustness, defined as phenotypic constancy despite mutation, should correlate with survival during thermal challenge. We tested this hypothesis using vesicular stomatitis virus (VSV). We produced two sets of evolved strains after selection for higher thermostability by either preincubation at 37°C or by incubation at 40°C during infection. These VSV populations became more thermostable and also more fit in the absence of thermal selection, demonstrating an absence of tradeoffs. Eleven out of 12 evolved populations had a fixed, nonsynonymous substitution in the nucleocapsid (N) open reading frame. There was a partial correlation between thermostability and mutational robustness that was observed when the former was measured at 42°C, but not at 37°C. These results are consistent with our earlier work and suggest that the relationship between robustness and thermostability is complex. Surprisingly, many of the thermostable strains also showed increased resistance to monoclonal antibody and polyclonal sera, including sera from natural hosts. These data suggest that evolved thermostability may lead to antigenic diversification and an increased ability to escape immune surveillance in febrile hosts, and potentially to an improved robustness. These relationships have important implications not only in terms of viral pathogenesis, but also for the development of vaccine vectors and oncolytic agents.

Introduction

Viruses must maintain particle stability in order to survive in the environment and to carry out their replication cycles within hosts. Thermal fluctuations are some of the main environmental perturbations faced by viruses, and are particularly relevant for mammalian viruses, which are subject to periodic increases in temperature during febrile episodes. Similarly, thermostability is important for phages that infect thermophilic microbes [1]. Experimental data also suggest that thermal adaptation plays an important role in the evolution of arboviruses, which alternate between vectors and hosts whose temperatures can be quite disparate [2]. More recently, climate change may be increasing the extent of selection for thermostability in viruses and their hosts [3]. The ability of a virus to develop stability under thermal selection depends on its evolutionary history [4]. From an applied point of view, thermostability is highly desirable in vaccines and viral vectors that may be used therapeutically [5–8].

The ability of a population to accumulate mutations without affecting phenotype is known as mutational or genetic robustness [9,10]. The extremely high mutation rates of many RNA viruses ensure that most progeny genomes will contain mutations relative to their parents [11]. While some of these mutations may be beneficial and increase viral fitness, empiric data suggest that the vast majority of newly generated mutations are highly detrimental to subsequent replication [12]. Increased robustness is the result of an increased neutral mutation rate at the expense of the beneficial mutation rate, the deleterious mutation rate or both [13]. The importance of mutational robustness as a buffer against mutational fitness effects is often illustrated using fitness landscapes, which relate genotypes to fitness. The “ground level” is a representation of the range of genotypes in sequence space and the “altitude” at any given location is the fitness associated with that genotype. Selective pressures determine the contours of the landscape. A population with higher robustness would spread out over a flat fitness peak and a fitter, less robust one would occupy a sharp fitness peak. Both classical population genetics and quasispecies theory predict that replication at high mutation rates will select for mutational robustness [9,14], a phenomenon termed “survival of the flattest.” Indeed, experiments with vesicular stomatitis virus (VSV) [15,16], and the phage Φ6 [17] have provided examples of high fitness populations being out-competed by less fit, but more mutationally robust competitors.

The theory of plastogenetic congruence posits that gains in genetic robustness will show a direct correlation with gains in thermostability, because proteins and nucleic acids should have the same response to destabilization regardless of whether it is due to mutation or increased temperature [18]. While there is less theoretical work on plastogenetic congruence for proteins, which have 20 as opposed to 4 monomeric units, data consistent with the overall model have been obtained using both proteins [19,20] and RNA [21–23]. There is less work in cellular systems, which are limited by the sensitivity of most cell lines to increases in temperature [24]. Plastogenetic congruence may be particularly relevant to the evolution of RNA viruses, because they evolve increased genetic robustness under selection. The correlation between thermostability and genetic robustness has been tested, at least partially, in a few cases using RNA viruses. Qβ phages selected for increased physical stability, including thermostability and survival at extreme pH, had increased genetic robustness [25,26]. On the other hand, Φ6 phages selected for increased genetic robustness had increased adaptability during selection under elevated temperature, which would imply a change in phenotype [27,28]. The literature is even scarcer for animal viruses. Work with VSV showed an imperfect correlation between mutational robustness and thermostability [29]. Specifically, virus strains that evolved under selection were found to have increased their genetic robustness without an overall change in thermostability. Even more surprisingly, virus strains that had evolved under random drift lost robustness and increased overall thermostability. However, within this set of strains, there was a strong correlation between thermostability and robustness.

To our knowledge there are no published data that test whether selection for thermostability results in concurrent selection for genetic robustness in animal viruses, a knowledge gap we address here. Our subject is VSV, the prototypic member of the Rhabdovirus family, which has been used extensively as an experimental model in virus evolution [30]. VSV has a nonsegmented, negative-stranded RNA genome [31]. The virions are enveloped in a membrane derived from the host cell and contain the viral genome in a complex with nucleoprotein (N), phosphoprotein (P), and the large segment of RNA-dependent RNA polymerase (L). The interior of the membrane is lined with a matrix protein (M), which aids in viral fusion and budding, and the exterior is lined with glycoprotein (G), which carries the receptor-recognition site and is the main target of the host antibody response. VSV is an arthropod-borne virus and undergoes cycles of insect and mammalian infection, usually in livestock [32]. Interestingly, the influence of the mammalian host on the evolution of VSV appears to be minimal [33,34].

Here, we examined changes in genetic robustness in VSV after selection for thermostability over 25 passages. Populations that were repeatedly passaged at 40°C not only became more thermostable, but were also more fit than their ancestors under normal replication conditions. This thermostability was greater than that seen in control passaged populations, although the fitness gain was less than that seen in control passages, demonstrating a cost to the development of thermostability. Nearly all of the thermostable VSV populations shared a fixed, nonsynonymous mutation in the nucleocapsid open reading frame. Consistent with the theory of plastogenetic congruence, these heat-adapted VSV were more mutationally robust than the wild type ancestor, although this increase in robustness was not significantly greater than that of control populations. Interestingly, the thermostable populations were also more resistant to neutralization by monoclonal antibodies and polyclonal sera relative to control adapted populations. These data suggest a previously unrecognized relationship between robustness, thermostability, and antigenic diversification in RNA viruses.

Results

Selection of vesicular stomatitis virus with increased thermostability

We used two different selective regimens to evolve VSV populations with increased thermostability (Figure 1). In the first, we selected for virion stability by incubating VSV at 37°C for 48 hours prior to infection of BHK cells. We refer to these populations as “37-adapted.” In the second regimen, all passages were carried out in cells incubated at 40°C. These “40-adapted” viruses completed the entire replication cycle at an elevated temperature. We performed 25 passages with six independent lineages for each regimen. These populations were then compared to either the WT ancestor or VSV populations passaged on BHK cells for 25 passages without thermal selection – “Adapted Controls.”

Fig. 1. Schematic for experimental evolution.

Regimens used to generated the 37-adapted and 40- adapted populations as well as the adapted controls.

We quantified the thermostability of each evolved population by measuring virus survival after incubation at either 37°C or 42°C and used these data to derive a thermostability coefficient (% virus survival/day). We chose to test thermostability at 42°C as it represents the limit of what a virus would encounter in mammalian infection. Wild type VSV was quite sensitive to elevated temperature, maintaining only 18% virus survival/day at 37°C and less than 0.5% virus survival/day at 42°C (Figure 2). The 37-adapted viruses failed to demonstrate statistically significant thermostability at 37°C with an average of 24% virus survival/day (p = 0.4509, Figure 2a), but they had a statistically significant increase in stability at 42°C to an average value of 1.1% virus survival/day (p = 0.0278, Figure 2b). All adapted strains were more thermostable than WT to at least one temperature: strain 37C at 37°C, strains 37D and 37E at 42°C, and strains 37A, 37B, and 37F at both temperatures. All of the 40-adapted viruses were more stable to the 37°C (Figure 2a), and three of six were more stable to the 42°C (Figure 2b) treatment than the average of the WT ancestor, with averages of 35% virus survival/day (p = 0.0026) and 1.2% survival/day (p = 0.0334), respectively. We had shown previously that adapted controls had no overall increase in thermostability at 37°C or 39°C [29]. The 37-adapted strains were not significantly more thermostable than these adapted controls at 37°C (p = 0.9900) while the 40-adapted strains were (p = 0.0103).

Fig. 2. Thermostability of evolved VSV strains at selected temperatures.

WT and adapted strains were tested for thermostability (a) at 37°C and (b) 42°C as described in the text. As confirmed by a two-sample t-test, the 37-adapted strains showed an increase in thermostability at 42°C (p = 0.0278), while the 40-adapted strains showed increases in thermostability at both 37°C (p = 0.0026) and 42°C (p = 0.0334). At 37°C the 37-adapted viruses were not more thermostable than control adapted populations (p = 0.9900) while the 40-adapted viruses were more thermostable (p = 0.0103). (*): p < 0.05, (**): p < 0.01.

The cost of increased thermostability

Evolutionary theory generally predicts the existence of trade-offs during adaptation to distinct environments [35]. There is at least one example showing such a tradeoff during the development of thermostability in phage W6 [36]. To determine whether increased thermostability imposed a fitness cost to VSV replication in the absence of thermal selection, we measured the fitness of each evolved population at 37°C (Figure 3; fitness of the WT ancestor is indicated with a broken line). Fitness for two of the strains, 37B and 40F, could not be determined using our standard direct competition assay (see last section of Results). We found that all of the 37-adapted (average = 2.02, range = 1.4 – 2.9) and 40-adapted lineages (average = 2.34, range = 1.3 – 3.3) gained fitness relative to the WT ancestral strain (two-sample t test, p = 0.0242 and p = 0.0137 for the 37-adapted and 40-adapted strains, respectively). The extent of fitness increase in the 37-adapted strains was similar to that of the 40-adapted strains.

Fig. 3. Fitness of thermostable VSV strains.

Fitness was determined for each adapted strain. As controls, we show eight independent WT replicas passaged 25 times under identical conditions, except for the thermal selective pressure (e.g. without preincubation of virions at 37°C before infection and keeping the temperature at 37°C during infection). Population means were compared using a two-sample t-test (37-adapted p = 0.0242, 40-adapted p = 0.0137). However, both strains also exhibited lower fitness gains than seen in the control populations (37-adapted p = 0.0014, 40-adapted p = 0.0047), suggesting a fitness cost to thermal adaption. (*): p < 0.05, (**): p < 0.01

Because these populations were able to increase thermostability and replicative fitness simultaneously, these data argue against a classic trade-off between the two traits under the conditions employed. However, the fitness gains observed in the 37-adapted viruses and 40-adapted viruses were not as high as fitness gains observed in control populations that had been selected for the same number of passages under identical conditions, except for thermal selective pressure (average = 4.2, range = 2.6 – 5.6, p = 0.0014 and p = 0.0047 for the 37-adapted and 40-adapted viruses, respectively, Figure 3). The larger fitness gains in the controls reflect adaptation of these populations to BHK cells under standard replication conditions. These data demonstrate that while adaptation to elevated temperature does not result in a classic fitness trade-off, it does impose a cost by limiting fitness gain.

Genetic determinants of thermostability

We determined the complete genome sequences of the 37-adapted and 40-adapted populations to identify the genetic basis for increased thermostability. We found a total of 97 mutations (Table 1), including 96 nucleotide replacements or polymorphic sites and one deletion in 87 positions distributed throughout the genomes (Table 1). Among the single nucleotide changes, 71 were changes in the consensus sequence (i.e. the majority base), while the rest were polymorphic sites in which the mutant bases were less frequent than the WT at each position. Four substitutions were in noncoding regions (Table 1). The first- second- and third-positions of the codons exhibited similar numbers of replacements, with 33, 26, and 32 respectively. We found 23 synonymous mutations, including 18 substitutions in the consensus sequence, but all were unique and the majority was found in strain 40F (Table 1). Transversions (53) outnumbered transitions (43), in part due to the parallel A→U substitution found at position 604 (Table S1). These values are in contrast with the results of clonal analysis, where transitions exceed transversions by a factor of about 8.5:1 (Table S2, see below). Despite the excess of transversions in the thermostable strains there were no C→G or G→C substitutions, and only one U→A replacement. It is important to note that many additional single nucleotide variants in each lineage exist at frequencies below the limit of detection of Sanger sequencing, which may also contribute to the thermostability phenotype.

Table 1. Mutations and amino acid changes found in thermostable VSV strains.

Mutations found in VSV populations after 25 passages of thermal selection. Based on the relative frequencies of WT and mutant peaks in the chromatograms, we scored the presence of mutations as follows: 1 indicates a mutant peak that is consistently present in all readings but at a very low frequency, 2 indicates a mutant peak that is at a substantial frequency but lower than that of the WT, 3 indicates a mutant peak that is at a frequency equal to that of the WT, 4 indicates that the WT is still present but that the mutant peak is at a higher frequency, and 5 indicates that the mutant peak has replaced the WT. AA = Amino Acid, Syn. = Synonymous, NC = Non-Coding

Mutation	AA Change	Strains	Mutation	AA Change	Strains
G 286 A	G → R	40B (1)	U 5104 C	F Syn.	40F (5)
U 391 G	L → V	40F (5)	G 5283 A	R → K	40F (5)
G 543 A	L Syn.	40D (1)	G 5340 U	R → M	40A (1)
A 604 U	I → F	37A (5), 37B (5),	G 5450 U	G → W	40A (1)
		37C (4), 37D (5),	U 6096 G	I → R	40A (1)
		37E (5), 37F (4),	C 6190 U	N Syn.	40E (4)
		40A (4), 40B (5),	U 6198 G	I → S	40A (1)
		40C (5), 40D (5),	A 6224 C	M → L	37C (2)
		40E (5)	A 6315 G	K → R	37D (5)
G 699 A	S Syn.	40F (5)	A 6681 C	N → T	40B (1), 40D (2)
C 702 A	F → L	40F (5)	A 6945 C	N → T	40F (5)
U 924 C	S Syn.	40F (5)	U 7060 C	Y Syn.	40F (5)
A 988 C	R Syn.	40F (5)	C 7141 U	D Syn.	40F (5)
A 1016 U	Q → L	37D (5)	A 7312 C	E → D	37F (1), 40F (5)
A 1173 G	P Syn.	40A (4)	G 7510 A	M → I	40C (4)
A 1250 U	Y → F	40F (5)	A 7717 G	E Syn.	40F (5)
C 1345 U	NC Region	37D (5)	G 7720 U	E → D	37C (2)
A 1370 C	NC Region	40F (5)	A 8272 G	T Syn.	40C (1)
G 1627 A	D → N	40F (5)	U 8275 C	R Syn.	40F (5)
U 1870 G	F → V	40F (5)	U 8325 A	M → K	37C (2)
U 1927 G	Y → D	40F (5)	G 8624 A	A → T	40A (3)
A 1977 G	V Syn.	37C (2)	C 8715 U	T → I	37F (3)
G 2081 A	G → E	40B (3)	U 8800 C	A Syn.	40B (1)
C 2151 U	Y Syn.	40F (5)	A 9057 G	K → R	40E (4)
C 2381 A	S Syn.	40F (5)	A 9132 C	H → P	40C (1)
G 2433 A	E → K	37A (2)	G 9154 U	P Syn.	40F (5)
U 2594 G	N → K	37C (2)	G 9223 U	K → N	37D (2)
C 2607 U	P → S	40F (5)	A 9647 C	I → L	40D (1)
U 2937 G	Stop → G	37F (2)	G 9833 A	V → I	37E (4)
G 3036 U	NC Region	40F (5)	A 9855 U	E → V	40F (3)
C 3149 A	H → Q	40F (5)	G 9856 U	E → D	40F (3)
G 3268 U	S → I	37A (5)	G 9972 A	R → Q	40F (3)
A 3283 U	Q → L	40F (5)	C 10035 U	S → F	40F (3)
A 3751 U	Y → F	40A (3)	C 10059 U	T → M	40F (3)
U 3780 C	C → R	40E (2)	U 10066 C	V Syn.	40F (3)
G 3846 A	D → N	40F (5)	U 10123 C	Y Syn.	40F (3)
G 4122 U	V → F	37B (5)	U 10135 C	I Syn.	40E (4)
C 4180 A	T → K	40E (5), 40F (5)	A 10277 U	N → Y	40D (4)
C 4390 U	T → I	37F (2)	C 10439 U	L Syn.	40C (4)
U 4397 C	F Syn.	40F (5)	U 10611 G	I → S	40F (5)
U 4512 G	F → V	40F (5)	A 10672 C	K → N	40A (3)
A 4692 G	NC Region	40F (5)	10775–11013	Deletion	40C (5)
U 5074 G	H → Q	40F (5)	C 10812 U	S → L	40D (1)
A 5092 G	A Syn.	40A (1)	G 10892 A	E → K	40D (1)

Mutation 604A→U was identified in 11 out of the 12 populations, and the corresponding AUU that codes for an isoleucine became a UUU codon that codes for a phenylalanine in amino acid position 181 of the N ORF (Table 1). Here, the fixation of a single mutation across multiple independent lineages is a strong indicator of positive selection and suggests that the nucleocapsid protein plays an important role in viral thermostability. However, given the range in thermostability observed among these 11 populations (see Figure 2), its absence in 40F, and the results from mutants with single substitutions other mutations must contribute to the thermostability phenotype. The other parallel mutations were: 4180C→A, 6681A→C and 7312A→U. All parallel mutations, except 6681A→C have been found previously during experimental evolution of VSV under different conditions (Table S3).

Interestingly the only strain lacking mutation A604U mutation was replica 40F. This virus has a large number of mutations throughout the genome, including 32 fixed sites and a cluster of seven polymorphic sites (with similar frequencies for WT and mutant nucleotides (nts) in each) in the L ORF (Table 1). Six of the fixed substitutions, including three nonsynonymous mutations, are within the N ORF, further supporting the importance of the nucleocapsid to viral thermostability.

The relationship between thermostability and robustness

According to the theory of plastogenetic congruence, selection for thermostability should lead to increased mutational robustness [18]. We used a mutagen sensitivity assay [29] to measure the robustness coefficients of each of the 37-adapted and 40-adapted populations.

While there was variation among individual lineages, the 40-adapted populations were more robust as a group than the wild type ancestor (two-sample t-test, p = 0.0057) (Figure 4a). The differences between WT and the 37-adapted viruses were not statistically significant (p = 0.1668). We combined the data for the two sets of strains to further explore the relationship between robustness and thermostabilty at each temperature. In the first analysis, we compared robustness of the complete set of viruses (37A-F and 40A-F) with their thermostability at 37°C. Here, there was no significant correlation (Figure 4b), with a correlation coefficient of 0.008 (Pearson’s test, p = 0.4895). In contrast, there was a direct correlation between robustness and thermostability at 42°C (Figure 4c), with a correlation coefficient of 0.537 (Pearson’s test, p = 0.0292). Importantly, adapted control virus populations also showed an increase in robustness over ancestral WT (p = 0.0119), and neither the 37-adapted viruses (p = 0.7088) nor the 40-adapted viruses (p = 0.1245) were significantly more robust than the passaged controls. This supports our previous work suggesting a link between adaptation and robustness [29]. These results also suggest that thermostability is not the most important factor for determining robustness, and that any relationship that exists between the two is complex and can vary with assay or selection conditions.

Fig. 4. Mutational robustness of thermostable VSV strains.

(a) The robustness of WT and thermostable strains were tested as survival in the presence of 5FU. There was no significant increase in robustness for 37-adapted viruses (p = 0.1668) but there was an increase in robustness for 40-adapted viruses (p = 0.0057). Adapted controls were also more robust than WT (p = 0.0119), and neither 37-adapted (p = 0.7088) or 40-adapted (p = 0.1245) virus populations were significantly more robust than the adapted controls. (b) The relationship between thermostability at 37°C and robustness for 37- and 40-adapted strains. A Pearson test revealed no correlation between robustness and thermostability at 37°C for this complete set of thermostable strains (r = 0.0081, p = 0.4895). (c) The relationship between thermostability at 42°C and robustness for 37- and 40-adapted strains. A Pearson test revealed a significant correlation between robustness and thermostability at 42°C for the complete set of thermostable strains, represented by the dashed line (r = 0.5370, p = 0.0292). (*): p < 0.05, (**): p < 0.01

Differences in mutagen sensitivity can be due to differences in robustness, but can also be the result of differences in polymerase fidelity. Viral strains with high-fidelity polymerases are typically more resistant to mutagen than their more error-prone WT counterparts simply because fewer mutations are incorporated. We tested whether changes in fidelity could be contributing to the observed differences in mutagen sensitivity by sequencing 60–90 molecular clones from each population. Because all genomes are sampled regardless of fitness, the frequency of unique mutations approximates the viral mutation rate. Using this assay, we found that the mutation rate of each evolved population was similar to that of the WT ancestor (Table 2). Recognizing that we only had 34% power to detect a two-fold difference in mutation frequency, we also compared the evolved strains as groups, where we had 80% power. Here, the 37-adapted group and 40-adapted group had nearly identical mutation rates to WT, with values for WT of 1.75 × 10⁻⁴ mutations/nucleotides sequenced and 1.70 × 10⁻⁴ mutations/nucleotides sequenced for both the 37-adapted and 40-adapted groups (Fisher exact test, p = 0.911). Furthermore, for the complete set of viral strains there was no correlation between mutagen sensitivity and mutation rate, with a Pearson coefficient of 0.2663 (p = 0.1896, Figure 5). Thus, we ruled out altered fidelity as a cause of differences in mutagen sensitivity, and we confirmed that this assay is a valid method to quantify mutational robustness.

Table 2.

Mutation frequencies of adapted strains

Strain	Clones Sequenced	Nucleotides Sequenced	Unique Mutations	Mutation Frequency	p value vs. WT1	p value vs. WT1-3*
37A	81	60993	12	1.97 E-4	0.8479	0.7271
37B	92	70380	9	1.28 E-4	0.2210	0.4832
37C	84	63168	13	2.06 E-4	0.8522	0.6088
37D	81	62046	11	1.77 E-4	0.5642	1
37E	85	65875	12	1.82 E-4	0.7013	0.8648
37F	65	48555	6	1.24 E-4	0.2706	0.5485
40A	78	60216	8	1.33 E-4	0.1994	0.4529
40B	81	59130	13	2.20 E-4	1	0.4872
40C	79	56327	11	1.95 E-4	0.8447	0.7200
40D	78	59280	10	1.69 E-4	0.5510	1
40E	71	54173	6	1.11 E-4	0.1873	0.4341
40F	84	64176	13	2.03 E-4	0.8514	0.7319
WT1	86	66822	15	2.24 E-4	-	-
WT2	81	59778	10	1.67 E-4	-	-
WT3	61	44648	5	1.12 E-4	-	-
WT1-3*	228	171248	30	1.75 E-4	-	-

^*Three replicates of WT were performed (WT1, WT2, WT3). WT1-3 is the combination of these three replicates.

Fig. 5. Correlation between robustness and mutation frequency.

A Pearson test showed no correlation between mutation frequency, shown as number of mutations per 1000 bases and robustness, expressed as sensitivity to mutagen (r = 0.2663, p = 0.1896).

Antigenic diversification of thermostable virus populations

Our fitness assay relies on a difference in monoclonal antibody sensitivity between test and reference virus [37]. The WT ancestor is sensitive to the I1 Mab, which recognizes a linear epitope in the A1 antigenic site of the G glycoprotein, and before performing fitness assays on the evolved lineages, we tested whether they maintained their sensitivity to this monoclonal antibody. We measured the frequency of resistant variants in each population using the I1 monoclonal antibody. Surprisingly, 14% of the viruses in the 37B population and 60% of those in the 40F population were resistant (Figure 6), which precluded the possibility of measuring fitness for these two strains. The other ten thermostable populations also tended to exhibit somewhat reduced sensitivity compared to the WT ancestor. For these ten populations, the frequency of resistant mutants was < 0.4%, which is sufficiently low to be excluded as a significant source of error in fitness assays. The differences in I1 sensitivity between WT, with a Monoclonal Antibody Resistant Mutant (MARM) frequency of 0.06% and the 37-adapted viruses, with a MARM frequency of 2.5% or 40-adapted virus, with a MARM frequency of 10.1%, were not seen in adapted controls (0.06%), and are unlikely to result from passaging on BHK-21 cells. These differences were not supported statistically, but the tendency among thermostable strains to have increased resistance in this single epitope led us to ponder whether resistance towards other antibodies could have evolved as well.

Fig. 6. I1 sensitivity of thermostable VSV strains.

WT and adapted strains were titrated in the presence and absence of I1 Mab. Mutant frequency was determined as the titer in the presence of I1 divided by the full titer in the absence of I1. These ratios were then log-transformed due to a very large range of values. Using a two-sample t-test, no significant difference was seen in the average of any adapted group (37-adapted p = 0.4180, 40-adapted p = 0.3291; adapted controls p = 0.7808). The outliers are 37B and 40F.

To study antigenic diversification within a larger number of epitopes, we measured resistance to polyclonal sera from two pigs [38] and one horse [39], which are natural hosts for VSV. We also tested a serum raised against our WT strain in a rabbit [40]. Unfortunately, we did not have sufficient quantities of the horse serum to interrogate the adapted control populations. As a group, the 37-adapted strains exhibited a less remarkable pattern of antibody sensitivity (Figure 7). They were more resistant to the rabbit serum, but showed no difference in resistance to the either of the pig sera or the horse serum. Looking at individual lineages, the 37D population had a very high level of resistance (19.7%) to the second pig polyclonal serum (Pig 2). The more thermostable 40-adapted populations were also more resistant to both pig and horse sera than WT, but were not significantly more resistant to the rabbit serum, indicating a significant level of antigenic diversification. While we did observe a small overall increase in rabbit serum resistance in control adapted populations in this study, previous work with this same serum showed no changes compared to WT [40]. None of the control populations showed high resistance - the highest values were 2 to 3-fold compared to several orders of magnitude in thermostable strains. There were no changes in sensitivity to sera from either pig (there was not enough horse serum to run the tests). Overall, it seems that the general increases in antibody resistance seen in this study cannot be attributed solely to adaptation on BHK-21 cells. Also the 40-adapted populations, which are more thermostable, tended to have increased resistance to a variety of sera, suggesting a direct correlation between thermostability and antibody escape.

Fig. 7. Sensitivity of thermostable VSV strains to polyclonal sera from natural hosts.

Sera from two separate pigs, Pig 1 (a), and Pig 2 (b), a horse (c), and a rabbit (d) were used to test antibody sensitivity in WT and the adapted virus strains as described previously. (a) The 37-adapted viruses mean sensitivity to Pig 1 serum was not significantly different than WT (two-sample t-test, p = 0.5451), but the 40-adapted virus mean sensitivity was significantly less, indicated by the greater number of antibody resistant mutants present (p = 0.0418). The outlier is 37D; the most resistant strain among the 40-adapted is 40F. The control-adapted population did not show any significant differences in sensitivity (p = 0.7644). The 40-adapted population was also significantly more resistant to the serum than the control adapted viruses (p = 0.0008).(b) The 37-adapted population was not significantly more resistant to the second pig serum (p = 0.0817) but the 40-adapted population exhibited reduced sensitivity to this serum (p = 0.0176). The most resistant among the 40-adapted strains is 40F. The control-adapted population was not more resistant than ancestral WT (p = 0.2062). Both the 37-adapted (p = 0.0007) and the 40-adapted (p < 0.0001) populations are significantly more resistant to this serum than the control adapted populations. (c) 37-adapted viruses exhibited no difference in sensitivity to the antibodies in the horse serum compared to ancestral WT (p = 0.5856), while the 40-adapted viruses again showed greater resistance (p = 0.0215). The most resistant strain among the 37-adapted strains is 37D, and the least resistant among the 40-adapted strains is 40E. Control adapted populations were not able to be tested against this serum. (d) The 37-adapted viruses mean sensitivity to rabbit serum was significantly less than ancestral WT (p = 0.0222), but the 40-adapted virus mean sensitivity was not significantly different than WT (p = 0.2210). The control adapted populations also showed a significant decrease in sensitivity when compared to WT (p = 0.0390). Neither population was more resistant than the control adapted population (37-adapted p = 0.1529, 40-adapted p = 0.3531).(*): p < 0.05; (***): p < 0.001

Discussion

We have employed two different selective regimens to generate thermostable VSV strains. Both regimens were designed to optimize selection by the use of large populations (200,000 PFU/passage) [41] replicating at low MOI (0.1 PFU/cell) [42–44]. The first regimen targeted the stability of virions in the extracellular environment. The second selected for thermostability over the course of a complete replication cycle, a situation that would represent selection during febrile periods for mammalian viruses. In addition to the differences in temperature, other factors may separate the two types of selection. For instance, selection on RNA folding and structure – particularly for mRNAs – or on RNA-protein interactions is likely to be stronger during replication. Conversely, the availability of cellular heat shock proteins and other chaperones during replication may weaken selection and buffer the deleterious effects mutation by correctly folding mutated viral proteins [45]. Heat-shock proteins may also have indirect selective effects, for instance, through the induction of innate immune response [46] or when required for optimal replication. For example, fever during measles virus infection induces the expression of the cellular heat shock protein, hsp72, which augments viral replication and virulence through its interaction with the viral N protein [47]. While chaperones appear to be the universal response of cells to thermal pressure [48,49], their role in the biology of RNA viruses is complicated, with evidence for beneficial and detrimental effects on viral replication [46,47,50].

Sequencing of the thermostable genomes demonstrated that almost all the replicas had a nonsynonymous transversion mutation in the N ORF, which strongly suggests that this mutation is contributing to thermostability in some way. This mutation could play a role in protein-protein or protein-RNA interactions, but in silico modeling suggests that the corresponding mutant N protein may be more thermodynamically stable (Amy Gilson & Eugene Shakhnovich, personal communication). This mutation has been previously identified in VSV strains adapting to BHK-21 cells (Table S3), although in only 4 of more than 200 evolved genomes for which complete sequences are available (see references in Table S3). These data suggest that it has pleiotropic effects on thermostability and BHK-21 adaptation. While increased capsid stability is a common strategy to improve overall thermostability in naked, icosahedral viruses [5,7,8,51–55], we found that an enveloped virus with a helical capsid can follow the same path.

It is highly unlikely that the 604A-U substitution contributes to antibody escape because N is an internal virion protein, and while the results point to it as a main determinant of thermostability, its absence from 40F and the wide range of thermostability values require contributions from other mutations. Indeed, two G glycoprotein mutations tested mediate increased thermostability and also contribute to antibody escape when tested individually, providing the link between thermostability and antigenic diversity. First, mutation 3853G→A is the MARM U marker and maps inside the I1 epitope [56,57]. It has no effect on fitness or robustness, but it acts pleiotropically to confer thermostability and Mab I1 escape [29]. Second, Hwang and Shaffer isolated mutation 4180C→A (Table S3) under immune selection and demonstrated that it confers both increased thermostability and increased resistance to antibodies [6]. Other G mutations may also contribute to one or more of the observed phenotypes.

The large number of mutations identified in 40F remains unexplained. Initially we hypothesized that this would be a hypermutator, but, with the caveat imposed by relatively low sensitivity of the method, the results of clonal analysis showed similar error rates in this mutant compared to WT and other thermostable populations (Table 2). Nevertheless, even if we had found an increased mutation rate, we would still need to explain the fixation of 32 mutations and the co-dominance of seven additional mutations in only 25 passages. This level of fixation is remarkable considering that the most mutations we had previously found in an adapted strain were 16, and only after 80 passages [58].

Antigenic diversity can arise in the absence of immune pressure during adaptation in cell culture [59,60]. Our results are in agreement with these prior observations and are explained once again by pleiotropy. Here, we have found that viruses subjected to thermal stress undergo a significant degree of antigenic diversification, which may lead to escape from adaptive immune responses. In this manner, we propose that some viruses may use a single selective factor to coevolve at least two desirable traits: thermostability and antibody resistance. This idea warrants further investigation as it would represent a new paradigm in the study of the antiviral response.

Classic trade-offs, in which adaptation to a new environment results in loss of fitness in the original one, are often found in bacteria and DNA phages [51,52,54,55,61–63]. These trade-offs are less frequently observed in RNA viruses, and when they are noted, alternation between environments often results in coadaptation to each environment (reviewed in [64]). It is becoming increasingly clear that many viruses can adapt simultaneously to multiple conditions, and our results support this paradigm. Thermostable viruses increased in fitness (Figure 3), and the results of genomic sequencing are consistent with the absence of a tradeoff. These results provide support to the concept that the contribution of mutations to fitness in viruses is the result of their effect on protein stability [65]. However, there is still a cost to generalism [54,66] because the fitness gains are much less than those found in the absence of thermal pressure. Adaptation of E. coli to different thermal environments produced similar results in that tradeoffs were sometimes absent [67,68]. The cost of thermal adaptation may explain why, in the absence of tradeoffs, there may be selection for specialists that do well in some environments and not others [35].

We have also explored the relationship between thermostability and genetic robustness. In a test of the plastogenetic congruence model [18], we found that selection for thermostability led sometimes, but not always, to increased genetic robustness. Thus, there is a correlation, but an imperfect one. These results are consistent with our previous observation that strains selected for increased robustness during replication under selection did not change overall thermostability, and strains that lost robustness during drift increased their thermostability. Within these drifted strains there was an excellent correlation between thermostability and robustness [29]. The decoupling of robustness and thermostability may be due to the strength of selection on different regions of the genome. For instance, different HIV-1 proteins have different levels of robustness, with the capsid protein being the most fragile [69]. The sequences that code for antigenic sites on influenza virus are selected for very high fragility, presumably because mutation of these sequences is more likely to cause amino acid changes and enable antibody escape [70]. Intriguingly, our results show a similar pattern in VSV, where the decoupling of genetic robustness and thermostability may be related to the evolution of antigenic variation. Others have proposed decoupling mechanisms that may be relevant for other systems, such as oligomerization [71] or epigenetic modification [72].

The broader relevance of the observations we have presented is twofold. VSV has been used as both a vaccine vector and oncolytic virus [73–75], and increased thermostability and antibody resistance may be desirable phenotypes in these applications [6]. We have found that it is possible to simultaneously select for both. Robustness is also a desirable property in viruses that will be used clinically, because robustness implies stability. We have shown that while robustness does not always coevolve with thermostability, it may, and that the two properties are not in conflict. The second aspect of relevance relates to the ability of viruses to evade host immunity during febrile periods, either by subverting cellular stress responses or by evolving to escape antibody.

Materials and Methods

Cells and viruses

Baby hamster kidney cells (BHK-21) were a gift from John Holland (University of California, San Diego) and were maintained in minimal essential media (MEM) supplemented with Hanks salts, 7% heat-inactivated bovine calf serum (BCS), and 0.05% proteose peptone #3 (PP3, Difco). Plaque assays were performed with the same media, but without supplemental PP3. The l1 monoclonal antibody was harvested as conditioned media from hybridoma cells provided by Douglas Lyles (Wake Forest University) [56]. All VSV populations in this study were derived from the Mudd-Summers strain of the Indiana serotype. The wild-type virus (WT), which was used to derive the other populations, has a fitness value set at 1 [37]. Passaged WT control populations, which replicated in BHK-21 cells for 25 passages but were not subjected to specific thermal selection regimes, have been described previously [76]. Briefly, these control populations were passaged on BHK-21 cells at low MOIs in order to increase in fitness in that environment.

Fitness determinations

Viral fitness is measured as a competition as described previously [37]. Briefly, viruses were mixed with a monoclonal antibody resistant mutant, MARM U, and used to infect BHK-21 cells at a population size of 2 × 10⁵ PFU. After 24 hours, the progeny virus was then diluted to the same population size to infect a fresh monolayer of BHK-21 cells. The ratio of resistant and susceptible viruses could be measured at each passage by performing titrations in the presence and absence of neutralizing antibody. Plotting the log-transformed ratios against time yields a linear relationship, and the antilog of the slope of this plot yields a fitness value.

Virus passages

We followed two experimental regimens to select for thermostable populations. In the first, we took a sample of the WT virus stock at 10¹⁰ plaque forming units (PFU)/mL and divided it into six microcentrifuge tubes, each being the progenitor of an independent replica. The tubes were incubated at 37°C for 48 hours, which typically caused a 99% reduction in viral titer. After treatment, the viable subpopulation was diluted, and ~2 × 10⁵ PFU were used to infect a T-25 monolayer of BHK cells at a multiplicity of infection (MOI) of 0.1 PFU/cell. Progeny populations were recovered at maximum titer, which in this case correlated with complete cytopathic effect (CPE), typically at 20–24 hours. A sample of this passaged population was again placed in a microcentrifuge tube and incubated at 37°C for 48 hours. Titrations were done after each infection to verify the virus concentrations at each passage and adjust dilutions as needed to maintain all passages at a constant population size and an MOI of 0.1 PFU/cell. These lineages are termed 37A through 37F. The only exception to the above protocol was that replica 37D required 48 hours for complete CPE over the final five passages.

In the second regimen, a sample of WT virus was diluted to a concentration of 10⁶ PFU/mL and six samples of 200 μL were used to infect six independent flasks at an MOI 0.1 PFU/cell. In this case, the infected cells were incubated at 40°C. Each flask contained the progenitor of one evolved replica. We allowed replication until maximum titers, and we continued infections with a constant population size and at an MOI of 0.1 PFU/cell for 25 passages. These lineages are referred to as 40A through 40F.

Thermostability

Thermostability was measured by incubating aliquots of virus at the specified temperature, either 37°C or 42°C, in glass tubes, and taking aliquots every 24 hours. These samples were titrated on BHK cells and the log titers plotted against incubation time. We then calculated the thermostability coefficient as the antilogarithm of the slope of the linear regression. This coefficient is expressed as % surviving virus/day.

Mutagen sensitivity assay

BHK cells were grown to semiconfluency as above. Growth medium was replaced with virus growth medium supplemented with varying concentrations (0, 10, 35, 100 μg/mL) of 5-fluorouracil (5FU). The cells were incubated in this medium at 37°C for 6.5 hours prior to infection. The medium was removed and the treated cells were inoculated with 2 × 10⁵ PFU of virus. The inoculated cells were left at room temperature for 10 minutes to allow for virus attachment, followed by 45 minutes at 37°C to allow for virus entry. Fresh virus growth medium with the appropriate concentration of 5FU was then added, and the cells were incubated at 37°C. Samples were taken at maximum titer, which usually corresponded to complete CPE. In conditions where no CPE was observed, samples were taken after 3 days.

Genomic sequencing

The full-length sequences of the 12 heat-adapted genomes were determined as described previously [58]. Briefly, viral RNA was isolated using the Qiamp viral RNA mini kit (Qiagen) and used as a template for reverse transcription using Superscript III reverse transcriptase (Invitrogen). The resulting cDNA was used to amplify 10 overlapping fragments covering the entire length of the virus genome. The fragments were sequenced by MCLAB using a set of primers as described previously [77] except for minor adjustments to match the consensus sequence of our laboratory strain. Multiple reads were obtained for each position and analyzed using MacVector (v9.0) with visual inspection of chromatograms for each variant reading to identify mutations that had not reached high enough frequency to produce a change in the consensus and were observed as polymorphic sites.

Measurement of viral mutation frequency

Viral RNA was harvested from clarified culture supernatants using Qiamp Viral RNA kits, and cDNA were synthesized using random hexamer primers and Superscript III reverse transcriptase. Fragments corresponding to the P-N region of the VSV genome (nt 1314 – 2178) were amplified using Phusion DNA polymerase (NEB) and either primer pairs N1314Fm (5′ GTCAGAGTTTGACAAATGACCC 3′) and G119R (5′ GCAATAATGGTAATTGGAAGGA 3″) or N1314Fm and P793R (5′ GACTCTCGCCTGATTGTA 3′). Polymerase chain reaction products were purified using the GeneJET PCR Purification Kit (Thermo) and terminally adenylated by incubation with 500 μM dATP and Taq DNA polymerase for 10 minutes at 72°C. These products were cloned into the pCR4-TOPO-TA vector using the TOPO-TA Cloning Kit for Sequencing (Invitrogen). Direct sequencing of colonies with primers N1314Fm and P793R was performed by Eton Biosciences (Newark, NJ). Sequences were aligned over a 750–800 bp region that had adequate quality reads for all clones and mutations were identified using SeqMan Pro version 10.1.1 (DNASTAR). Only mutations present in both the forward and reverse reads of a clone were counted, and mutations found in multiple clones were excluded, as they could represent polymorphisms as opposed to newly generated mutations.

Antibody resistance

Antibody resistance was measured by titrating virus strains in both the presence and absence of the I1 monoclonal antibody [56] used in fitness assays [37], as well as polyclonal sera from a VSV-infected horse [39] and two pigs [38] (kindly provided by Dr. Luis Rodriguez). The convalescent horse serum was recovered from animals inoculated with a natural field strain, USA-1997. Both convalescent pig sera were recovered from animals inoculated with derivatives of a hybrid San Juan/Orsay strain of the Indiana serotype. We also tested a rabbit polyclonal serum obtained after three immunizations with 100 μg of wt VSV in complete Freund’s adjuvant [40]. To avoid the confounding effects of phenotypic mixing and hiding [78,79], antibody was not added until after the initial round of virus entry. Briefly, the virus was applied to semiconfluent BHK-21 cells in the absence of antibodies, incubated at room temperature for 10 minutes and 37°C for 45 minutes as described above. The infected monolayers were then covered with MEM+FBS supplemented with monoclonal antibody to cause maximum neutralization (e.g. when additional antibody did not have any further effect on plaque numbers). For polyclonal sera, available in limited amounts, special care was taken to ensure the same numbers of pfu during infection and each serum was diluted 1/20 in the overlay. The degree of antibody resistance was quantified as the titer of virus in the presence of antibody divided by the titer of the virus in the absence of antibody.

Data analysis

Statistical analyses were performed using GraphPad Prism 5.0 and R using the statistical methods indicated in the Figure legends and text.

Author Summary.

Thermostability is likely to be an important trait for viruses that must survive in extreme environments or replicate in febrile hosts. We used experimental evolution of vesicular stomatitis virus to identify evolutionary correlates of thermal adaptation. Selection for thermostability resulted in coevolved fitness and, surprisingly, antigenic variation. There was a partial correlation between thermostability and mutational robustness (the ability to mutate without changing the phenotype) and no correlation between thermostability and mutation rate. Nearly all of the thermostable viruses had a mutation in the nucleocapsid protein as well as other mutations elsewhere in the genome that may explain the congruent selection of thermostability, fitness and antibody escape. Our results open the intriguing possibility that viruses are taking advantage of selection during fever to escape the immune system.

Highlights.

• Vesicular stomatitis virus was evolved under thermal selection

• There was a partial correlation between mutational robustness and thermostability in evolved populations

• Selection for thermostability resulted in coevolved fitness and, surprisingly, antigenic variation.