Adaptive reassortment in TULV

The persistence of reassortants in natural populations requires continued genomic compatibility based on highly specific RNA-RNA and RNA-protein interactions [1]. Most reassortants between deeply diverged viruses cannot maintain this compatibility, thus suffer fitness costs and are quickly purged from populations [1,9]. Similarly, hybridization in hosts can impose additional fitness constraints on associated parasites due to novel genetic and immunological environments [11,12]. Despite these hurdles, we observed a reassorted hantavirus (TULV-CEC) that not only persisted but has become dominant within a 12.5 km broad region of the host hybrid zone, displacing both parental clades. This pattern suggests that the TULV-CEC M-segment provides a fitness advantage in hybrid hosts. Concrete verification of this hypothesis would, however, require further testing in vitro. Viral replication and shedding of different TULV clades and the reassortant would need to be quantified in cell cultures or living hosts of different genetic backgrounds to confirm a fitness advantage of the reassortant in hybrid hosts (see [30]).

The deep phylogenetic divergence between the M-segment of the CEC reassortant and other TULV-EST.N M-segments from the region (Fig 3) suggests evolution within hybrid hosts for an extended period of time. Viral reassortments with adaptive functions have been primarily observed for the influenza A virus [4,31,32], but are also sporadically documented for a variety of other viral families, including reoviridae [33] and peribunyaviridae [34]. Among the hantaviruses, reassortments documented in nature so far have been attributed to selectively neutral processes [15]. However, the few studies that examined hantavirus reassortment in nature lack the in-depth characterization necessary to establish their adaptive potentials [15,16,35,36]. Our findings suggest that reassortment-driven adaptation in hantaviruses may be more common than previously recognized.

Additional reassortment types

The presence of multiple reassortment types within the TULV-CEN.N/TULV-EST.N contact zone suggests a higher degree of segment compatibility than for previously reported TULV clade interactions [18–20]. While M-segment reassortment is the most frequently observed form in hantaviruses [15], we also identified reassortants involving the S- and L-segments (TULV-CEE-1 & 2, Fig 3). TULV-CEE-2 consists of segments, which are nearly identical to local variants in the same population, suggesting that the causative reassortment event occurred not long ago (S4 Table). Notably, three out of four samples contained S-segments from both parental virus clades (Fig 3). These three might represent hosts with double infections for TULV-EST.N and TULV-CEE-2, in which case the M- and L-segment between both could be too similar to distinguish. Alternatively, these three may represent transiently diploid genomes for the S-segment. Transient segment diploidy has been observed in vitro as an intermediate state during reassortment [17,37,38]. These viruses are expected to eventually segregate into stable reassortants or parental genotypes [38], which may have already occurred in the haploid TULV genomes from the TULV-CEE-2 sampling sites.

Structural variation and functional implications

Our results suggest that structural differences in a short region at the N-terminus of the TULV glycoprotein may be responsible for an adaptive advantage of TULV-CEC, resulting in its spread in the hybrid zone. This genome region exhibits substantial structural and genetic divergence (Figs 4, S4 & S5). We observed strong signals of positive selection at key residues (S8 & S9 Tables) that are in line with previous reports [18,20], all of which point to this region being a hotbed of evolutionary activity. In TULV, this region is part of the N-terminal ectodomain of the Gn-subunit [39,40]. Protein modeling revealed that the N-terminal ectodomain forms a distinct and consistent shape for each TULV clade, positioned within an exposed surface pocket between two glycoprotein spikes (Fig 4B). The functional role of this region remains unclear, but its spatial localization suggests a role in host receptor binding or immune evasion. The N-terminal domains of the Gn-Gc subunits have been implicated in glycoprotein maturation, viral entry, and immune interactions in other hantaviruses [41–44]. Experimental verification of this regions function could entail initial deletion followed by an analysis of post-deletion fitness to verify a concrete role in hantaviral infection. This could be followed up by an assay on potential interaction partners to confirms its role in the metabolic pathways of viral infection, replication and transmission.

Our models predicted the start of the protein at residue 13, preceded by a signal peptide of 12 residues. However, predictions of the protein start are inconsistent across hantaviruses and individual studies. The TULV M-segment coding region starts 15 bp upstream compared to other hantaviruses, aligning our predictions with the peptide start at residue 18 first described via protein crystallography of the Haantan virus [45]. Crystallography studies initially placed the start at residue 25 in Puumula virus (PUUV) [46], but later also revised it to 18 [47]. Other orthohantaviruses show further variation: Maporal virus (residue 22) and Andes virus (residue 23) [48]. These inconsistencies have led to a general neglect of the N-terminus in hantavirus glycoprotein studies. However, the absence of structural variation in other parts of the protein suggests that this region could play a primary role in shaping the adaptive landscape of TULV reassortants and may carry a similar level of importance across the entire hantavirus family.

Conclusion

Our findings provide strong evidence for adaptive reassortment in a non-pathogenic hantavirus, driven by hybridization of its rodent host, and highlights the importance of fine-scale ecological and genetic context in understanding viral adaptation. The emergence and persistence of the reassorted TULV-CEC strain suggests that under the right conditions, reassortment can act as a mechanism for viral adaptation to novel host environments in hantaviruses. Hybrid zones, as natural settings of elevated and recombined genetic diversity of hosts, may play an underappreciated role in facilitating viral evolution and should be considered as key sites for studying the emergence of new viral phenotypes. Furthermore, our structural analyses revealed that a short region at the start of the N-terminal ectodomain of the TULV glycoprotein is a hotspot of adaptive change, suggesting a key role in host-virus interactions and emphasizing the need for further functional characterization of this region.

Sample acquisition

A total of 1284 common voles (Microtus arvalis) were collected from 58 trapping locations at the border of Germany and the Czech Republic (Fig 1). This region, termed the “Saxony transect”, was expected to encompass the hybrid zone between the Central and Eastern evolutionary lineages of the vole host, as well as the contact zone of the TULV clades TULV-CEN.N, TULV-EST.N, and TULV-EST.S, as suggested by prior studies [18,19,25]. Snap traps were used for vole collection, and specimens were stored at -20°C immediately after retrieval. All samples are archived at the Institute of Ecology and Evolution at the University of Bern.

TULV screening and phylogenetic analysis

528 adult common voles from 56 sites were screened for TULV infections. Only adult voles weighing at least 20g (summer) or 18g (autumn) were included, as juveniles show low TULV prevalence (<1%), likely due to maternal antibodies and limited exposure [49]. RNA was extracted from lung tissue using a modified QIAzol protocol [25]. TULV infection status was determined by RT-PCR amplification and Sanger sequencing of at least 427 base pairs (bp) of the S-segment nucleocapsid gene, following the assay described in [50]. A single sample (MarDSl04) could be amplified via RT-PCR, but not at sufficient quantity or quality for successful Sanger sequencing (S1 Table).

For reassortment detection, all TULV-positive samples underwent additional PCR amplification of M- and L-segments using primers C1/C2 [51] and HanLF1/HanLR2 [52], respectively, yielding fragments of at least 356 bp (M-segment) and 305 bp (L-segment). In the case of the M-segment, three individuals yielded no results using Sanger sequencing, although subsequent whole genome sequencing of one of the samples showed a complete TULV genome including the M-segment. To prevent loss of information for phylogenetic clade assignment due to the complete deletion of sites containing one or more missing or ambiguous nucleotides, a total 0.64% of sequence content was imputed based on the closest genetic relative [20]. Phylogenetic assignment of these segments was performed with MrBayes v3.2.7a [53] on the CIPRES platform [54]. We performed MCMC sampling for up to 10⁸ generations in four independent runs comprising four chains, implementing reversible-jump sampling over the entire general time-reversible substitution model space [55]. After discarding a burn-in fraction of 25%, samples were recorded every 10³ generations. Chains converged after 2 460 000, 1 965 000 and 5 410 000 generations for the S-, M- and L-segment respectively. Phylogenetic trees were drawn and edited using the online platform iTOL v5 [56]. We created maps for visualization of the viral clade distribution using the geosphere package [57] in R and topographic backgrounds were modified after the osmdata package [24].

Sequencing of host mitochondrial DNA

We extracted host DNA according to a standard phenol-chlorophorm protocol (modified after [58]). We used mtDNA in order to assess the evolutionary lineages of voles across the Saxony transect. Host evolutionary lineage was determined by sequencing [59] at least 458 bp of the mitochondrial cytochrome b gene in 183 individuals. All TULV infected individuals, as well as a minimum of two individuals per population when available were sequenced to show the distribution of host lineages in the Saxony transect. Phylogenetic analysis followed the same pipeline as viral segments, with lineage assignment based on reference sequences [26]. Chains converged after 12 850 000 generations. The information on mtDNA lineages was used in context with previous publications on the Central-Eastern M. arvalis hybrid zone as a proxy for establishing genetic admixture and hybridization of the host in the Saxony transect [18–20,22]. Mitochondrial, sex-chromosomal and autosomal markers are geographically consistent in assigning evolutionary lineages of M. arvalis across Europe [26,59–63] except in the immediate zone of hybridization. The extensive mixture between mtDNA lineages within populations across the sampling region cannot be explained without a history of hybridization between evolutionary lineages, given the low mobility and dispersal distance of common voles [64,65].

Geographic cline analysis

To assess the spatial transition of vole evolutionary lineages and TULV clades, geographic cline analyses were performed using the HZAR package in R [66]. Sampling locations were projected onto a one-dimensional transect axis, optimized to minimize geographic distance between the TULV-CEN.N and TULV-EST.N clades (Fig 1) [18,20,22]. Genomic segments from TULV-EST.S were not included in the analysis. Distances are given between the projection points and the western transect start in km. The proportion of each TULV clade (based on S, M, and L-segments) and host mtDNA lineage at each site was modeled using four cline models with increasing complexity: null (no cline), model 1 (free cline center and width), model 2 (free minimum and maximum frequency), and model 3 (additional exponential tail parameters). Likelihood scores of all cline models were compared for each analysis and cline parameters were estimated for the model with the highest likelihood, performing 10⁵ generations of MCMC sampling in three independent chains and with a burn-in period of 10⁴ iterations. We tested pairwise concordance of cline centers and widths across all analyses using a likelihood-ratio test (LRT) in R. We compared a null model of a concordant cline through two combined datasets to an alternative model of individual cline widths and centers for both datasets. We used two times the difference between the log-likelihoods of the alternative and the null models as the test statistic. Significance was determined based on a χ2 distribution with two degrees of freedom.

TULV whole genome sequencing

To confirm reassortment patterns, whole-genome sequencing was performed on all individuals with partial segment evidence of reassortment, as well as on additional individuals from populations within 10 km of the clade contact and four reference sequences. Preparation of libraries, whole genome sequencing and genome assembly followed the hybrid sequence capture protocol in [67], using custom baits to capture and enrich viral sequences in libraries. Libraries were sequenced on an Illumina MiSeq (Illumina, San Diego, CA, USA) with 2 x 300 cycles by the Next Generation Sequencing Platform of the University of Bern. Hybrid sequence capture generated a total of 2,695,193 TULV sequence reads (596–192,233 per sample, median = 35,919), with an average read depth of 795× (range: 7×–2494×) across genomes (S5 Table). De novo genome assembly was conducted using Iterative Virus Assembler [68] and subsequently mapped back against the initial viral consensus genomes in order to infer quality and mapping statistics. Missing nucleotides in three genomes (MarDLw01, MarDLw02, MarDLw03) were imputed based on their closest genetic relatives. We calculated the percentage of sites with a read depth of at least 3x and 20x, average genomic coverage and total read count for each genome using R.

TULV phylogenetic analysis

We carried out a phylogenetic analysis of TULV genomes both for the complete cds and the derived AA sequences of each viral segment from all genomes. Reference sequences consisted of published TULV genomes with a complete cds from Central Europe [18,28,67]. Four TULV genomes from M. obscurus in China and two PUUV were additionally included as outgroup genomes [69–72]. Bayesian phylogenetic inference for nucleotide sequences was performed as described for segment analysis. Chains converged after 670 000, 965 000 and 360 000 generations for the S-, M- and L-segment respectively.

For AA sequences we conducted the phylogenetic analysis in MEGA X [73]. We derived tree topologies using the Maximum Likelihood method based on the JTT matrix-based model [74] with 1000 bootstraps. Initial trees for all segments were obtained by applying Neighbor-Joining and BioNJ algorithms to a matrix of pairwise distances estimated using the JTT model and keeping the topology with the superior log likelihood value. Tree drawing and editing was performed identically to the TULV segment fragments (see above).

TULV sequence diversity and signatures of selection

Genome wide nucleotide diversity and divergence between TULV clades was calculated across the cds of all TULV segments in DnaSP version 5 [75]. Sliding-window analyses (30-bp window, 10-bp step) were used to evaluate d_N/d_S ratios (ratio of non-synonymous to synonymous substitutions) and D_XY (average number of nucleotide substitutions per site). AA divergence in the form of mean p-distance within and between TULV clades was calculated in MEGA X. Signatures of selection were inferred using the branch-site model [76] of CodeML, which is part of the PAML package version 4.9 [77]. The RAxML software version 8.2.12 on the CIPRES platform was used for the construction of phylogenetic trees for selection inference. Model likelihoods were compared to a null hypothesis using a likelihood ratio test and χ2 distribution. We used Bayes Empirical Bayes (BEB) [78] inference to identify sites under positive selection. We performed two additional scans for selection to test for rate variation at synonymous sites using FUBAR [79] and MEME [80] in HYPHY [81] on the Datamonkey webserver [82]. Posterior probabilities > 0.85 or p < 0.1 were considered as evidence of positive selection for sites.

Phylogenetic dating of TULV

Estimates for the time of most recent common ancestor (tmrca) of clade and cluster splits within TULV were calculated using the BEAST v1.10.4 software [83]. We followed the parameters established in [27], using a substitution rate of 1.51 × 10⁻³ substitutions per site per year for TULV, as well as a slower substitution rate estimate for hantaviruses of 2.7 × 10⁻⁴ [70]. The latter substitution rate naturally led to estimations of a notably higher TMRCAs, but proportions of TMRCAs between clades and clusters were consistent between higher and lower substitution rates. Only estimates for 1.51 × 10⁻³ are shown in the results, as they are directly comparable to [27]. We implemented a Bayesian skyline coalescent tree prior alongside a relaxed lognormal molecular clock for MCMC sampling of 10⁸ generations. Samples were recorded every 5000 generations.

Homology modeling, structural alignments and molecular dynamics simulations

To assess structural differences in the M-segment glycoprotein among TULV clades and reassortants, we conducted homology modelling and 3D visualization of the TULV M-segment and its N-terminal ectodomain. We selected six M segment sequences which reflect the full diversity spectrum for AA in the N-terminal ectodomain and over 90% of all AA diversity across the entire M-Segment from the Saxony transect, as well as the original TULV genome Moravia [28] as a TULV-EST.S reference. Homology modelling and molecular dynamics (MD) simulations were performed in YASARA Structure version 20 using adapted macros and structures implemented from [84,85]. Homology models were built using the standard “hm_build.mcr” macro based on the hantavirus glycoprotein structure “6ZJM” template [48]. Resulting homology models displayed three different subunit configurations (four in Configuration 1, two in Configuration 2, one in Configuration 3). These different configurations likely reflect stochastic variations in fitting or different states of the protein, as the hantavirus glycoprotein can transition through different structural variations [47]. It is unlikely that these variations reflect actual changes in protein folding, as we even observed different configurations between the near identical TULV-EST.S proteins from the Saxony transect and our reference Moravia (4 AA difference).

Pairwise structural alignments, including RMSDs and sequence identities between the homology models were calculated using MUSTANG [86]. RMSD of atomic locations across the entire protein tetramer (residue 13–1104) were compared to the RMSD of first nine AA of the Gn-subunits ectodomain (N-terminus). Only the results of the alignment for the four proteins in Configuration 1 are displayed in S10 Table, because structural alignments require identical configuration of all subunits for meaningful evaluation. Individual subunits of the three remaining proteins, when aligned, showed similar levels of low RSMDs, but randomness in the predicted configurations artificially inflates the pairwise RMSDs of the complete proteins. This problem can normally be circumvented by using the configuration of a predicted protein for all future predictions. This was not applicable for our set of proteins, as it would also artificially force the flexible N-terminal ectodomain to change shape to match the reference thus eliminating the key variation between proteins.

MD simulations of all N-terminal regions were performed using the “md.run.mcr” macro and AMBER14 force field, with a simulation duration of 500 ns [87]. Time simulation steps were set to 1.35 fsec. The simulation box was ‘Cube”-shaped and extended at least 10 Å to each side of the model (extension = 10), was filled with 0.9% NaCl and the TIP3P water model was used at physiological pH 7.4. Further settings were: temperature at 298K, pressure at 1 bar, density = 0.997, cutoff 8Å- periodic cell boundary and longrange coulomb forces (particle-mesh Ewald). The solute was kept from diffusing and crossing periodic boundaries using the CorrectDrift function [84,85]. The four flanking amino acids were fixed as anchor point during the MD simulations. To determine structural differences between the regions of interest, root mean square fluctuations (RMSF) were calculated by the “md_analyze” macro. The RMSF is the fluctuation of every heavy atom compared to the mean structure within a simulation cell. The average RMSF of the constituent atoms is used to determine the RMSF per solute residue. False positives with artificially high scores were further minimized by considering the number of structural neighbors in the model ensemble. The finished protein models can be found in the dryad repository [88].