Genomic characterization of mumps viruses from a large-scale mumps outbreak in Arkansas, 2016

Abstract

In 2016, a year-long large-scale mumps outbreak occurred in Arkansas among a highly-vaccinated population. A total of 2,954 mumps cases were identified during this outbreak. The majority of cases (1,676 (57%)) were school-aged children (5–17 years), 1536 (92%) of these children had completed the mumps vaccination schedule. To weigh the possibility that the mumps virus evaded vaccine-induced immunity in the affected Arkansas population, we established a pipeline for genomic characterization of the outbreak strains. Our pipeline produces whole-genome sequences along with phylogenetic analysis of the outbreak mumps virus strains. We collected buccal swab samples of patients who tested positive for the mumps virus during the 2016 Arkansas outbreak, and used the portable Oxford Nanopore Technology to sequence the extracted strains. Our pipeline identified the genotype of the Arkansas mumps strains as genotype G and presented a genome-based phylogenetic tree with superior resolution to a standard small hydrophobic (SH) gene-based tree. We phylogenetically compared the Arkansas whole-genome sequences to all publicly available mumps strains. While these analyses show that the Arkansas mumps strains are evolutionarily distinct from the vaccine strains, we observed no correlation between vaccination history and phylogenetic grouping. Furthermore, we predicted potential B-cell epitopes encoded by the Arkansas mumps strains using a random forest prediction model trained on antibody-antigen protein structures. Over half of the predicted epitopes of the Jeryl-Lynn vaccine strains in the Hemagglutinin-Neuraminidase (HN) surface glycoprotein (a major target of neutralizing antibodies) region are missing in the Arkansas mumps strains. In-silico analyses of potential epitopes may indicate that the Arkansas mumps strains display antigens with reduced immunogenicity, which may contribute to reduced vaccine effectiveness. However, our in-silico findings should be assessed by robust experiments such as cross neutralization assays. Metadata analysis showed that vaccination history had no effect on the evolution of the Arkansas mumps strains during this outbreak. We conclude that the driving force behind the spread of the mumps virus in the 2016 Arkansas outbreak remains undetermined.

1. Introduction

In 2016, a year-long mumps outbreak ensued in the state of Arkansas (Health 2018a; Alkam et al. 2017) that resulted in a total of 2,954 confirmed cases within a highly-vaccinated community. Overall, about 98% of cases were confirmed by standard PCR of the N gene, 1% by clinical symptoms and less than 1% by serology. Most affected individuals (1,676 (57%)) were children, 1,536 (92%) of these children had received at least two doses of the mumps measles and rubella (MMR) vaccine (Fields et al. 2019). Mumps is a highly contagious infection that spreads via the respiratory route. Once mumps is diagnosed, it is genotyped via RT-PCR of the small-hydrophobic (SH) gene (Boddicker et al. 2007). This stems from the variability in the genomic region of SH relative to the rest of the mumps virus (MuV) genome (Orvell, Kalantari, and Johansson 1997; Strohle, Bernasconi, and Germann 1996). The MuV expresses two surface glycoproteins, hemagglutinin-neuraminidase (HN) and fusion protein (F), that promote the spread of the virus by facilitating infection of the host cells. These surface glycoproteins are crucial to the activity of the MuV vaccine as neutralizing antibodies target epitopes on the surface of HN and F (Herrera et al. 2010; Elango et al. 1988; Li et al. 2009). The mumps vaccine is administered at childhood in two doses in most developed countries. The only vaccine strain used in the United States is termed Jeryl-Lynn (genotype A) (Amexis et al. 2002; Hviid, Rubin, and Muhlemann 2008). MMR is highly effective as its introduction in the late 1960’s has dramatically reduced incidence - mumps had almost been eliminated by 2001 (Rubin et al. 2015; (CDC) 2018; McNabb et al. 2007). Nevertheless, large sporadic outbreaks, even within vaccinated populations, began to appear in the United States and across the world in the past decade (Boxall et al. 2008; Marin et al. 2008; Rota et al. 2009; Alkam et al. 2017).

Two of the possible mechanisms behind the re-emergence of mumps in vaccinated populations include: waning vaccine-induced immunity, and immune escape. Waning immunity suggests that the immunity engendered by the vaccine is fading and thus the immune response is sub-optimal, which leads to contraction of the virus. The likelihood of waning immunity increases with time since vaccination. In this scenario, adults are more likely to become infected with the virus although mumps is typically a childhood disease (Lewnard and Grad 2018). Immune escape suggests that the MuV had genetically drifted from vaccine strains and its presented epitopes are no longer recognized by the immune system. This implies that the effectiveness of the currently available vaccine is compromised, and a novel vaccine must be designed to target the new MuV epitopes (May, Rieder, and Rowe 2018). In most of the recent outbreaks in the United States, a large number of the affected population were ages 18–29 years (Huong Q. McLean 2013; Savage et al. 2005; Sane et al. 2014), which suggests that waning vaccine-induced immunity is behind these outbreaks (Lewnard and Grad 2018). The 2016 Arkansas outbreak, however, is unique as the majority of those affected were school-aged children (Health 2018b). According to the Arkansas Department of Health, 90–95% of affected children were vaccinated with at least two doses of the MMR vaccine. This reduces the likelihood of suboptimal herd immunity among the affected population despite what was previously predicted (Majumder et al. 2017). These two points led us to hypothesize that the mumps virus that spread in Arkansas may have acquired genomic variations which resulted in its evasion of vaccine-induced immunity, leading to the 2016 outbreak.

We performed in silico analysis to investigate the possibility that immune escape contributed to the spread of mumps during the Arkansas outbreak. To address this, we sequenced samples from patients who contracted mumps during the Arkansas outbreak and analyzed the sequenced genomes using computational methods to deduce the likely mechanism. We chose the MinION sequencing platform (Oxford Nanopore Technologies) because it is portable and produces whole-genome sequences rapidly. However, this is likely more relevant for infections that cause severe acute illness unlike mumps, which causes chronic sequelae. Although we developed our pipeline in a central laboratory, we predict that hospitals in rural settings equipped with basic laboratory instruments may apply our method (Figure 1). Our computational analyses consist of two arms: comparative phylogenomics and immunoinformatics. Comparative phylogenomic methods identify the evolutionary distance between the Arkansas outbreak strains and other mumps strains, using a suite of informative molecular phylogenetic markers. Branching patterns along phylogenetic trees help determine whether the outbreak strains diverged from vaccine strains, which may indicate that the MuV evolved such that its genome is distinct enough from the vaccine strains to allow immune escape. Immunoinformatic analyses via epitope prediction aids in determining which specific epitopes are associated with a given outbreak. For in-silico assessment of the possibility that the Arkansas MuV evolved to escape immunity engendered by the vaccine, we compared potential B-cell and T-cell epitopes along genome sequences especially focusing on the surface glycoproteins (HN and F).

Figure 1.

Whole-genome sequencing analysis pipeline using a portable sequencing device.

2. Materials and Methods

2.1. Viral genome amplification

De-identified buccal swab samples of patients who had tested positive for mumps during the 2016 Arkansas outbreak were provided by the Arkansas Department of Health. The MuV samples fall into one of four groups: (1) children (0–17 years old) with no reported history of vaccination; (2) children (0–17 years old) who received at least two doses of the MMR vaccine; (3) adults (18+ years old) with no reported history of vaccination; (4) adults (18+ years old) who received at least two doses of the MMR vaccine. Viral RNA was extracted using the Quick-RNA Viral kit (Cat. R1034) from ZYMO Research according to the manufacturer’s protocol. Extracted RNA was reverse-transcribed to cDNA and PCR-amplified using the SuperScript™ III One-Step RT-PCR System with Platinum Taq High Fidelity DNA Polymerase (Cat. 12574030). Two sets of five pairs of primers targeting conserved regions of the MuV genome were designed and used in the RT-PCR reaction to amplify the whole genome in five overlapping fragments. These fragments were pooled and sequenced according to the protocol described below. Some samples were amplified via long PCR (Barnes 1994; Cheng et al. 1994) using only three of the abovementioned primer sets. While other samples were amplified using a set of five overlapping primers kindly provided by Dr. Biao He (Zhang et al. 2013). The primer sequences are summarized in Supplementary Table S1 along with a figure of their positions in Figure S1. Note that the final nucleotides of the genome are not covered by our primers and are therefore recovered from the reference-based genome assembly in our genomes.

2.2. Whole-genome sequencing

The pooled fragments of the amplified viral genomes were subjected to sequencing using the third-generation sequencing platform MinION (Oxford Nanopore Technologies). Four to twelve genomes representing a respective number of mumps samples were barcoded and sequenced on one flow cell using the 1D Native Barcoding Genomic DNA (with EXP-NBD103 and SQK-LSK108) kit. The manufacturer’s protocol was followed in the preparation of sequencing libraries. Samples were sequenced for 24hrs and raw signal data was subsequently collected for computational downstream analyses.

2.3. Bioinformatics

2.3.1. Reference-based genome assembly.

Raw signal files were collected from the Oxford Nanopore output and subjected to local basecalling using the Albacore v2.1.0 (https://github.com/rrwick/Basecalling-comparison) software. To trim adapters that were attached to sequenced DNA during library preparation, the Porechop software was applied (v0.2.1, https://github.com/rrwick/Porechop). Both Albacore and Porechop software demultiplex barcoded reads each using different criteria. To ensure the accuracy of the classification of reads into their proper barcodes, only reads that Albacore and Porechop agreed on their classification into a certain barcode bin were collected. These reads were then filtered by applying a phred quality score threshold. The Pauvre tool was used to visualize the phred quality scores relative to read lengths. Reads with quality scores equal to or below the threshold, set at 9 through inspection of the histogram produced by Pauvre (https://github.com/conchoecia/pauvre), were discarded. In addition, reads with length less than 2500bps were discarded as the PCR-amplified genome fragments were at least 2500bps-long. Duplicate reads were then removed using the seqkit tool (Shen et al. 2016). Thus, only high-quality reads were used for further downstream analyses. GraphMap, which is a mapping algorithm specifically designed for nanopore sequencing data, was used for alignment to a reference genome (Sovic et al. 2016). We first produced the reference genome (published in (Alkam et al. 2017)) by sequencing an Arkansas mumps sample using both MinION and MiSeq, reads from the latter were used to polish genomes obtained from MinION. The published script align-trim.py (from https://github.com/zibraproject/zika-pipeline/tree/master/scripts) was used to remove primer sequences. Next, the Nanopolish tool was used to generate consensus genomes using raw signal data obtained from the MinION (Loman, Quick, and Simpson 2015). Nanopolish iterates through signal data to detect modifications such as single-nucleotide polymorphisms (SNPs) and insertion-deletions (indels) in the sequenced genomes. These modifications are incorporated into the sequenced reads to generate consensus Arkansas sample genomes.

2.3.2. Phylogenetic analyses.

To examine the evolutionary relationships of the newly-sequenced Arkansas mumps genomes to all other mumps strains, we generated maximum-likelihood phylogenetic trees. All publically-available, complete MuV genomes were downloaded from the NCBI database. A nucleotide multiple sequence alignment of the concatenated coding sequences (CDS) and intergenic regions of the downloaded and sequenced Arkansas genomes was aligned using MUSCLE (Edgar 2004). We applied the partition scheme that specifies partitions corresponding to 1^st, 2^nd and 3^rd codon positions from the protein coding genes and intergenic regions to the multiple sequence alignment. The best partitioning schemes were identified using ModelFinder via IQ-TREE (Lanfear et al, 2012), which starts with the full partition model and subsequently merges until the model fit does not increase any further prior to constructing a phylogenetic tree. The output maximum-likelihood phylogenetic trees with 1000 bootstrap replicates were visualized using Dendroscope (Stamatakis 2014; Huson et al. 2007). For building the SH gene tree in Figure 2, ModelFinder via IQ-TREE first identified the right partition scheme of the SH gene from partitions corresponding to 1^st, 2^nd, and 3^rd codon positions, and identified K3Pu+F+G4 as the best-fit substitution model for the collapsed one partition. A maximum-likelihood tree of non-redundant SH genes was built based on the identified best partition model using IQ-TREE, and then was rooted by midpoint rule in Figure 2. For the tree in Figure 3, we chose the right partition scheme based on the partition model that specifies partitions at each intergenic region and at each codon position of individual genes. This resulted in a total of 27 partitions (7 genes with 3 codon positions and 6 intergenic regions were considered). Note that only one of the isoforms of the P gene was included. ModelFinder via IQ-TREE identified six partitions with the corresponding models as follows: TIM+F+G4: NP_1_P_2_M_1_F_1_HN_1_L_1, K3Pu+F+I: NP_2_F_2_HN_2, TVM+F+G4: NP_3_M_3_F_3_SH_2_HN_3_Int_6_L_3, GTR+F+G4: Int_1_Int_2_Int_3_Int_4_SH_1_SH_3_Int_5, K3P+G4: P_1_P_3, TN+F+I: M_2_L_2 where NP, P, M, SH, F, HN and L represent the MuV genes; Int represents each of the intergenic regions of the MuV genomes; _1, _2, _3 represent the first, second and third codon positions within genes respectively. With the right partition scheme identified, a maximum-likelihood phylogenetic tree with 1000 bootstrap replicates was constructed for a non-redundant dataset of the complete coding sequences of 7 MuV genes and intergenic regions, and then was rooted by midpoint rule in (Figure 3).

Figure 2.

Maximum likelihood tree of SH gene. A non-redundant set included SH genes of all complete genomes publicly available, Arkansas MuV SH genes, and SH genes of the draft MuV genomes from Washington outbreak, and were generated with 100% cutoff. Non-NCBI accession IDs in red correspond to Arkansas outbreak strains sequenced using our pipeline. Arkansas strains are labeled with the following suffixes: _G1 corresponds to strains isolated from unvaccinated children, _G2 corresponds to strains isolated from vaccinated children, _G3 corresponds to strains isolated from unvaccinated adults, _G4 corresponds to strains isolated from vaccinated adults. Labels in blue represent ones annotated as Jeryl-Lynn vaccine strains, and ones in magenta annotated as genotype G by GenBank. Labels in green represent strains from March 2016–2017 Washington outbreak that are annotated as genotype G except for KY880986 annotated as genotype K, and Accession IDs in orange from October, 2016 Massachusetts outbreak that are annotated as genotype G. The tips are labeled following the style of WHO names (Jin et al. 2015; WHO 2012) which includes information of accession number, collection date (represented as weeks.year), and genotype with asterisk for reference strains for all genotypes if available.

Figure 3.

Maximum likelihood tree of complete coding and intergenic sequences. The phylogenetic tree was built using a non-redundant dataset of all complete MuV genomes publicly available, and the sequenced Arkansas strains. The non-redundant dataset included the complete coding sequences and intergenic regions of the MuV and was generated with a 99.95% cutoff. Non-NCBI accession IDs, KY996510 and KY996512 in red correspond to Arkansas outbreak strains sequenced through using our pipeline. Arkansas strains are labeled with the following suffixes: _G1 corresponds to strains isolated from unvaccinated children, _G2 corresponds to strains isolated from vaccinated children, _G3 corresponds to strains isolated from unvaccinated adults, _G4 corresponds to strains isolated from vaccinated adults. Seven among 8 strains from Group 3 were clustered into a group whose representative member is 20171128_BC05_G2, and one member from Group 3 was clustered into a group whose representative is 20180104_BC08_G1 Labels in blue represents ones annotated as Jeryl-Lynn vaccine strains, and ones in magenta annotated as genotype G by GenBank. Accession IDs in orange are strains from October, 2016 Massachusetts outbreak that are annotated as genotype G by GenBank. WHO names were used (Jin et al. 2015; WHO 2012). (*) represents reference strains.

2.3.3. Immunoinformatics.

Predictions of T-cell and B-cell epitopes of the MuV genomes was conducted using the NetMHC-4.0 (Andreatta and Nielsen 2016) and BepiPred-2.0 (Jespersen et al. 2017) respectively. NetMHC-4.0 is an artificial neural network that predicts binding affinity of the input peptide sequences to MHC class I and class II molecules. We limited the NetMHC predictions to 9-mers (peptides with a length of 9 amino acids) because it is rare for longer or shorter peptides to bind to MHC (Buus et al. 2003; Lundegaard et al. 2008). Epitopes predicted to have strong binding to MHC were selected for further analyses. Strong binders were defined as epitopes with a rank<0.5 (Andreatta and Nielsen 2016). For B-cell epitope prediction, we explored the software BepiPred-2.0, which was a random forest prediction model trained with epitopes that were annotated from solved crystal structure data of antigen-antibody binding. Predicted epitopes were filtered out of the output dataset and subjected to downstream analyses. We applied the cutoff of 0.5, such that peptides were considered epitopes if their probability was above 0.5.

3. Results

3.1. The Arkansas outbreak mumps strains are phylogenetically distinct from vaccine strains.

We devised experimental and computational pipelines for rapid whole-genome sequencing of the Arkansas MuV using the third-generation sequencing platform MinION (Oxford Nanopore Technologies) (Figure 1). We sequenced the MuV genomes obtained from 51 samples of patients who had tested positive for mumps in the 2016 Arkansas outbreak, deposited in GenBank under the BioProject: PRJNA517140. To verify the accuracy of the sequences obtained using the MinION platform (Oxford Nanopore Technologies), three samples were sequenced using both MinION and the next-generation sequencing platform, MiSeq (Illumina). We compared their reference-based assemblies where KF481689 was used as a reference genome. The whole-genome sequences obtained from both technologies were compared using MUMmer (Germann et al. 1996). The average sequencing depth of Nanopore samples ranged from 381 to 1378 with a mean of 997 and standard deviation of 220. The average depth of Illumina samples ranged from 4903 to 7314 with a mean of 6266 and standard deviation of 1268, which is much larger than Nanopore sequencing depth because we sequenced one sample on one Illumina run to test primer sequences with the Arkansas outbreak samples. We observed two single-nucleotide polymorphisms (SNPs) between Nanopore and Illumina sequencing data of two samples at positions 8420 and 1183, and at positions 8388 and 11801 for one sample. These are negligible considering the MuV genome is 15,383 bps in length. Importantly, none of the aforementioned SNPs were on the coding sequences of the HN (6614–8362) or F (4546–6162) proteins. We therefore concluded that the whole-genome sequences obtained using the Oxford Nanopore Technologies are reliable, and we implemented this platform only in further experiments.

To determine the Arkansas MuV genotype, we first explored the traditional genotyping approach based on the small hydrophobic protein (SH) (Figure 2). The dataset for building this tree consists of the coding sequences of the SH gene of all publicly-available complete MuV, along with the newly-sequenced Arkansas strains. We also included 66 SH coding sequences of strains from a 2017 Washington outbreak whose CDS were available in GenBank. The tree includes 105 strains available from the 2016 Massachusetts outbreak which were reduced to only 4 SH genes in (Figure 2) by 100 % sequence identity. However, the same strains were reduced to 63 by 100 % sequence identity of CDS sequences and intergenic regions (data not shown), and to 12 strains in (Figure 3) by 99.95% sequence identity of CDS sequences and intergenic regions. The 99.95% cutoff in (Figure 3) resulted in 9 Arkansas outbreak representative sequences where 20171128_BC05_G2 is a representative of 29 Arkansas outbreak strains, 20180104_BC08_G1 of 9 strains, 20171006_BC02_G2 of 6 strains, 20171006_BC01_G2 of 2 strains. The rest of the representative sequences are singletons. In addition, we observed that the Arkansas strains grouped with genotype G strains in the SH gene tree (Figure 2). This is expected as most mumps outbreaks in the United States that occurred in recent years belonged to genotype G (Dayan et al. 2008).

Next, we built a phylogenetic tree of all complete coding and intergenic sequences in Figure 3. This tree includes all complete MuV genomes publicly available in GenBank at the time of analysis along with the newly-sequenced 51 Arkansas mumps strains generated through our pipeline were included (Figure 1). Note that the tree in Figure 3 does not contain strains from 2017 Washington outbreak since complete genomes of those strains were not available at the time of analysis. We confirmed the genotype of the Arkansas mumps strains as genotype G based on whole genome sequences, which is in agreement with the traditional SH-based genotyping method (Figure 3). The set of 51 SH genes from the outbreak is redundant and was reduced to a set of 5 SH genes with a 100% sequence similarity cutoff where 20180104_BC12_G4 is a representative of 44 Arkansas SH genes, 20180104_BC01_G1 of 3 Arkansas SH genes, and 20171128_BC02_G3 of 2 Arkansas SH genes (Figure 2). The rest of the representative sequences are singletons. Thus, the SH gene tree in (Figure 2) contains only 5 branches from the Arkansas outbreak. In contrast, a set of 51 whole genomes was reduced to a set of 44 genomes with a 100% sequence similarity cutoff (data not shown). The whole-genome tree showed that the Arkansas MuV strains are phylogenetically distinct from the vaccine strains (Figure 3). Thus, comparative genomic data indicates that the Arkansas strains evolutionarily diverged from the vaccine strains. We observed no evolutionary distinctions between the aforementioned four groups of Arkansas samples (vaccinated children: group G1, unvaccinated children: group G2, vaccinated adults: group G3, unvaccinated adults: group G4). Therefore, vaccination history had no effect on the evolution of the Arkansas MuV during this outbreak, which indicates that immune evasion is probably not the driving force behind the Arkansas 2016 outbreak.

3.2. Possible immune evasion of the Arkansas mumps strains.

To further investigate the possibility of immune escape, we predicted potential locations of epitopes presented to both T-cells and B-cells focusing on the MuV surface glycoproteins: hemagglutinin-neuraminidase (HN) and the fusion protein (F). All complete MuV genomes available in GenBank at the time of analysis, as well as the sequenced Arkansas genomes were subjected to these immunoinformatic analyses. We observed a discrepancy in the number of predicted B-cell epitopes between the Arkansas strains and the vaccine strains, such that certain epitopes in the vaccine HN proteins are absent in all Arkansas strains (Figure 4). There is evidence that the other MuV surface glycoprotein, F protein, is an immunodominant antigen (May, Rieder, and Rowe 2018; Cusi et al. 2001; Kulkarni-Kale et al. 2007; Santak, Orvell, and Gulija 2015). Therefore, we conducted the same B-cell epitope prediction analysis described above on the F protein. Here, we observe similar results to those found in the HN protein such that there are distinct regions where there is a discrepancy between predicted epitopes for the Arkansas strains compared to the vaccine strains (Figure 5). Note that an amino acid position is predicted to be an epitope based on a probability which the Bepi-Pred software assigns. This probability is assigned while considering the surrounding region. Therefore, an amino acid which is conserved across the Arkansas and vaccine strains may still be a potential epitope as there may be differences in amino acids in that region (Supplementary Figure 1). These analyses were applied to all other MuV proteins. We observed no significant differences in other proteins. We therefore conclude that the antigen prediction data of Arkansas MuV strains does not definitively indicate the occurrence of immune evasion by the Arkansas MuV strains, but does not exclude the possibility either. Next, we conducted similar analyses on the same genomes to predict epitopes that bind to T-cells with high affinity. We observed discrepancies in predicted T-cell epitopes on the HN protein in 4 amino acid positions, and on the F protein in 4 amino acid positions (data not shown). The significance of these observations is unclear. Taken together, we observed differences in predicted B-cell and T-cell epitope occurrence between the Arkansas outbreak MuV strains and the Jeryl-Lynn vaccine strains. While these observations are interesting, they are insufficient to determine whether the Arkansas mumps outbreak was a result of vaccine-induced immunity or not. Further investigation of the significance of our findings is warranted.

Figure 4.

Potential B-cell epitopes on the HN surface glycoprotein. A maximum-likelihood phylogenetic tree of the coding sequences and intergenic regions of 7 MuV genes was constructed. The tree includes all sequenced Arkansas strains and all publicly-available strains of the Jeryl-Lynn vaccine. The JL-5 major component and JL-2 minor component strains are marked in red. Arkansas strains are labeled with the following suffixes: _G1 corresponds to strains isolated from unvaccinated children, _G2 corresponds to strains isolated from vaccinated children, _G3 corresponds to strains isolated from unvaccinated adults, _G4 corresponds to strains isolated from vaccinated adults. The corresponding heatmap indicates the presence (blue) or absence (light gray) of predicted antigens recognized by B-cells on the HN surface glycoprotein. The BepiPred-2.0 software was used to predict potential B-cell epitopes. The horizontal axis of the heatmap represents the amino acid position of predicted antigens on the HN protein.

Figure 5.

Potential B-cell epitopes on the F surface glycoprotein. A maximum-likelihood phylogenetic tree of the coding sequences and intergenic regions of 7 MuV genes was constructed. The tree includes all sequenced Arkansas strains and all publicaly-available strains of the Jeryl-Lynn vaccine. The JL-5 major component and JL-2 minor component strains are marked in red. Arkansas strains are labeled with the following suffixes: _G1 corresponds to strains isolated from unvaccinated children, _G2 corresponds to strains isolated from vaccinated children, _G3 corresponds to strains isolated from unvaccinated adults, _G4 corresponds to strains isolated from vaccinated adults. The corresponding heatmap indicates the presence (blue) or absence (light gray) of predicted antigens recognized by B-cells on the F surface glycoprotein. The BepiPred-2.0 software was used to predict potential B-cell epitopes. A horizontal axis of the heatmap represents the amino acid position of predicted antigens on the F protein.

4. Discussion

Here we present a pipeline for whole-genome sequencing, genotyping and phylogenetic analysis of the mumps virus. We also computationally investigated whether immune escape was the reason behind the spread of MuV in the 2016 Arkansas outbreak. We devised a protocol for rapid-whole genome sequencing of the MuV that allows us to analyze the phylogenetic and immunologic characteristics of the Arkansas outbreak strains (Figure 1). We sequenced 51 MuV whole-genomes extracted from buccal swab samples of individuals who had contracted mumps during the 2016 Arkansas outbreak using third-generation sequencing. This enabled us to quickly and accurately genotype the MuV. A recent report demonstrates the importance of whole-genome sequencing of the MuV during outbreaks (Dilcher et al. 2018). We find that whole-genome sequencing provides a higher resolution of branching patterns compared to SH-based genotyping. This may be advantageous in monitoring outbreak evolution. The SH gene tree (Figure 2) displays poor phylogenetic resolution since it shows incomplete separation. Outbreaks in three different states that occurred during the same time period (Massachusetts, Washington, and Arkansas during 2016–2017) were nearly collapsed with other strains of genotype G. On the other hand, the whole-genome tree in (Figure 3) clearly shows higher resolution in distinguishing strains within and between outbreaks, which is crucial for purposes of epidemiological surveillance and tracking outbreak origins. Arkansas strains first formed a monophyletic clade and then grouped with the Massachusetts strains forming a monophyletic clade, barring one Massachusetts strain in (Figure 3). This resolution was not achieved in the SH gene tree. However, subclades formed within the Arkansas clade did not correspond to metadata (age, vaccination status, collection date). To quantify phylogenetic diversity, we calculated Faith’s index (Swenson 2009) for both trees in (Figure 2 and 3) although they are not directly comparable since the two trees were built using different datasets. As expected, the SH tree has a larger Faith’s index than the tree in (Figure 3), especially for clades within genotype G (data not shown), due to polytomies of the tree in (Figure 2) (data not shown). Upon examining the whole-genome based tree (Figure 3), we speculate that the mumps outbreaks in Massachusetts and Arkansas may be linked. However, this is not necessarily true as the proximity of the Arkansas and Massachusetts strains in (Figure 3) may also be explained by the wide circulation of the same mumps virus across the United States, as the virus may have not been confined to areas of sporadic outbreak such as Arkansas or Massachusetts. The Arkansas genomes grouped with strains from genotype G, this is expected as the majority of mumps outbreaks in the United States in recent years were caused by strains belonging to genotype G. Therefore, our rapid whole-genome sequencing workflow is applicable and superior to the current method of SH genotyping. Our protocols are optimized for the mumps virus but this pipeline may be applied to other infectious agents that cause more severe acute complications during active outbreaks.

The reasons behind the re-emergence of mumps are not clear. The two most plausible theories to explain mumps outbreaks in vaccinated populations are waning vaccine-induced immunity and immune escape. Conflicting evidence in the literature provides support for both theories. An example of evidence towards waning immunity is the American Committee on Immunization Protection (ACIP) recent recommendation of a third booster MuV vaccine in individuals in areas of active outbreaks (Marin M 2018). Injection of a third dose in individuals at risk of contracting MuV during outbreaks has limited the spread of infection, however, there is no evidence of the effectiveness of long-term boosting in the general population (Cardemil et al. 2017; Barskey, Glasser, and LeBaron 2009). On the other hand, some reports show antigenic diversion of the MuV, which supports the immune escape theory (Santak et al. 2013; May, Rieder, and Rowe 2018). The lack of a definitive theory to explain the spread of mumps among vaccinated populations necessitates investigation. Therefore, we set out to determine the contribution of immune escape to the 2016 Arkansas outbreak. We applied comparative genomic and immunoinformatic tools on the 51 sequenced Arkansas MuV genomes to either confirm or rule out the possibility of immune escape. Phylogenetic analyses showed that the Arkansas strains were evolutionarily distinct from the vaccine strains, which is expected as the MMR vaccine was developed decades ago (Buynak and Hilleman 1966). However, this may allude to immune escape (Figure 4, 5). On the other hand, the distribution of the Arkansas strains across the tree did not correlate with vaccination history in neither children nor adults. This negates the possibility that MuV divergence was driven by evasion of vaccine-induced immunity.

These conflicting findings prompted us to turn to publicly-available immunoinformatic software to predict epitopes recognized by either B-cells or T-cells. We focused our search to the two immunodominant antigens, the HN and F surface glycoproteins. We expect to observe a discrepancy in the representation of epitopes on HN between the vaccine strains and strains that potentially evaded vaccine-induced immunity. Indeed, this is evident in (Figure 4) as predicted B-cell epitopes in certain regions of the vaccine HN proteins are either present or absent in the Arkansas strains. We observed differences in predicted B-cell epitopes in both HN and F when comparing the JL-5 major and JL-2 minor component strains of the Jeryl-Lynn vaccine (Figure 4, 5). We report sharp differences in predicted B-cell epitopes between the two Jeryl-Lynn vaccine component strains. Both JL-5 and JL-2 are present in MumpsVax (Level B) which is included in the vaccine administered across the United States, MMR. The variation in the nucleotide sequences of the JL-2 and JL-5 components is established, along with biological and molecular differences (Afzal et al. 1993; Amexis et al. 2002; Chambers, Rima, and Duprex 2009). Interestingly, we observed differences in predicted epitopes in a region that had been previously shown to elicit an immune response in vivo when expressed as a construct independent of the whole protein (Herrera et al. 2010). This region is 176 amino acids in length and spans across positions 255 to 431. Our data showed that predicted epitopes in this region are present in Arkansas strains whereas they are absent in the major component of the vaccine strains (JL-5) (Figure 4, positions 261–425). While the significance of this observation is unclear, discrepancies in antigen presentation in this experimentally-verified immunogenic region may result in attenuated activation of the humoral immune response upon exposure to the Arkansas MuV strains. Indeed, Dilcher at al. conclude that immune escape is a possibility based on structural differences of predicted epitopes between the JL-5 major component of the vaccine strain and the genotype G wild-type strain (Dilcher et al. 2018). We also observed a variation in predicted epitopes on the 513 position, which is adjacent to the crucial 512 position that is part of the active site of HN binding to the sialic acid of the host cell (Figure 4) (Kubota et al. 2016). Since the amino acid at position 513 is conserved among Arkansas and vaccine strains, we speculate that alterations in the residues surrounding the 513 site led to discrepancies in predicted epitope probability. We observed no differences in predicted antigens in previously described neutralization sites on the F protein, namely residues 221, 323 and 373 (Figure 5) (Santak, Orvell, and Gulija 2015). Homan et al. report differences in predicted B cell linear epitopes and potential T-cell epitopes between certain MuV strains and the JL-5 vaccine (Homan and Bremel 2014). Particularly, they found differences on the 275 amino acid residue that falls within the important B-cell neutralizing epitope region: 265–288 (Kulkarni-Kale et al. 2007). Similarly, we report differences in predicted B-cell epitopes in the same region on position 274 between the Arkansas strains and the JL-5 major component strains. Collectively, these observations hint that the Arkansas strains may have escaped immunity engendered by the vaccine. Our findings support a previous publication (Gouma et al. 2018) that compares genotype G strains with genotype A strains, which contain the Jeryl-Lynn vaccine. They report differences in amino acid sequences of functional regions in the HN and F proteins, this may reduce the immunogenicity of the vaccine strains. In addition, poor in vitro cross-neutralization of the wild-type and vaccine strain, JL-5, was reported (Vermeire et al. 2018). It should be stressed that our conclusions are based on computational predictions and must be experimentally verified before making conclusions about whether the Arkansas outbreak was driven by immune escape. In order to assess if the genomic signatures of epitopes have any bearing on immune evasion, robust laboratory experiments must be carried out. For example, cross-neutralization assays with sera from vaccines and from patients who were infected with the relevant wild-type strain. Finally, we focused on the MMR vaccine which is used in the United States. However, vaccines that contain only the JL-5 major component, such as RIT4385, are used around the world (Homan and Bremel 2014). This should be considered when weighing waning immunity or immune escape by comparing outbreak strains to vaccine strains.

These observations are insufficient to draw a conclusion on whether or not the 2016 Arkansas outbreak was caused by MuV evasion of vaccine-induced immunity. Nonetheless, our findings indicate that we cannot rule out the possibility that the Arkansas MuV strains may have escaped immunity engendered by the vaccine. A combination of decreased antibody titers at adolescence with changes in the epitope presentation may precipitate the spread of infection. This remains unclear and further studies that combine immunoinformatics and experimental evidence are necessary to understand this unexpectedly reoccurring public health issue.

4. Conclusions

Here we present evidence that the rapid, whole-genome sequencing method we devised is superior to the current genotyping method. We find that our pipeline provides higher resolution in branching patterns than the currently-used SH-based surveillance method. This, along with genomic characterization of the mumps virus, would provide the most complete picture of transmission. This is crucial and may aid in limiting the progression of active outbreaks. We applied this method on mumps virus strains collected from patients infected during the 2016 Arkansas outbreak. Comparative genomic and immunoinformatic analyses of the sequenced genomes show that the 2016 Arkansas mumps virus strains are phylogenetically distinct and differ in positions of predicted B-cell epitopes in the surface glycoprotein Hemagglutinin-Neuraminidase from the Jeryl-Lynn vaccine strains. We show that vaccination history had no effect on the evolution of the outbreak strains. The cause of the 2016 Arkansas mumps outbreak remains undetermined. Therefore, further research into the driving forces behind the re-occurring mumps outbreaks across the United States is warranted.

Highlights.

• The 2016 Arkansas mumps virus strains are phylogenetically distinct from the Jeryl-Lynn vaccine strains.

• The 2016 Arkansas mumps virus strains differ from the Jeryl-Lynn vaccine strains in positions of predicted B-cell epitopes in the surface glycoprotein Hemagglutinin-Neuraminidase.

• Vaccination history had no effect on the evolution of the Arkansas mumps strains during this outbreak.

• We developed a whole genome sequencing and characterization pipeline using a portable sequencing device, possibly applicable to infectious agents during active outbreaks.

• The cause of the spread of the mumps virus in the 2016 Arkansas outbreak remains undetermined.