Massively parallel sequencing for monitoring genetic consistency and quality control of live viral vaccines

Abstract

Intrinsic genetic instability of RNA viruses may lead to the accumulation of revertants during manufacture of live viral vaccines, requiring rigorous quality control to ensure vaccine safety. Each lot of oral poliovirus vaccine (OPV) is tested for neurovirulence in animals and also for the presence of neurovirulent revertants. Mutant analysis by PCR and restriction enzyme cleavage (MAPREC) is used to measure the frequency of neurovirulent mutations at the 5′ untranslated region (UTR) of the viral genome that correlate with the level of neurovirulence determined by the monkey neurovirulence test. However, MAPREC can only monitor mutations at a few genomic loci and miss mutations at other sites that could adversely affect vaccine quality. Here we propose to use massively parallel sequencing (MPS) for sensitive detection and quantification of all mutations in the entire genome of attenuated viruses. Analysis of vaccine samples and reference preparations demonstrated a perfect agreement with MAPREC results. Quantitative MPS analysis of validated reference preparations tested by MAPREC produced identical results, suggesting that the method could take advantage of the existing reference materials and be used as a replacement for the MAPREC procedure in lot release of OPV. Patterns of mutations present at a low level in vaccine preparations were characteristic of seed viruses used for their manufacture and could be used for identification of individual batches. This approach may represent the ultimate tool for monitoring genetic consistency of live viral vaccines.

High mutation rates inherent to replication of RNA viruses create a wide variety of mutants that are present in virus populations, which are often referred to as quasispecies (1). The diffuse, “cloud-like” nature of viral populations allows them to rapidly adapt to changing replicative environments by selecting preexisting variants with better fitness (2–4). Many important virus properties cannot be explained by a mere consensus sequence, but require knowledge about the microvariants present in viral stocks. One such property is virulence, which is critically relevant to vaccine development and manufacture.

Attenuated Sabin strains used in manufacture of oral poliovirus vaccine (OPV) (5–7) can regain neurovirulence during growth in vaccine recipients and propagation in cell cultures (8). It occurs because of the accumulation of mutants with higher neurovirulence (9, 10), which include changes in the 5′ untranslated region (UTR) (10). Because of this genetic instability, every batch of OPV must be tested for neurovirulence in monkeys (11) or transgenic mice susceptible to poliomyelitis (12, 13). Previously, we developed a highly sensitive molecular method, mutant analysis by PCR and restriction enzyme cleavage (MAPREC), which enabled us to quantify the 5′ UTR revertants in monovalent batches of OPV and demonstrate that their content in vaccine lots directly correlates with the results of the monkey neurovirulence test (MNVT) (14). The content of 472-C revertants in type 3 OPV is usually below 0.8% in batches of vaccine that passed the MNVT, whereas batches that failed the test usually contain more than 1% of 472-C revertants (15). Similar methods were developed to quantify mutants with 480-A in Sabin 1 and 481-G in Sabin 2 strains (16, 17) that were also shown to be determinants of attenuation. MAPREC is currently recommended by the World Health Organization (WHO) for screening of batches of OPV before they can be released for use in humans (11, 18).

Despite high sensitivity and accuracy of MAPREC, it can be used to monitor only a limited number of genomic loci known to contain the markers of attenuation and misses mutations at other sites, which can also contribute to loss of the attenuated phenotype. For most viruses, determinants of attenuation are unknown, which limits the utility of this approach for other live viral vaccines. Nevertheless, ensuring molecular consistency of vaccine batches by requiring that no mutation accumulates beyond the level present in past batches with good clinical record could help maintain vaccine safety (19). Therefore methods enabling detection of unstable genomic loci in attenuated strains could be used for the monitoring of genetic stability. Such studies are usually conducted by serial passaging of attenuated strains in vivo and in vitro, followed by screening for the presence of small numbers of mutants that may have accumulated in viral populations (16, 17, 20, 21). Traditional sequencing approaches can detect mutants only if their content exceeds 20–30% and are therefore of limited utility. Other approaches based on analysis of electrophoretic mobility in gels (22, 23) do not always allow precise location of mutations. Matrix-assisted laser desorption/ioniation-time of flight (MALDI-TOF) mass spectrometry (24) and hybridization with microarrays of short oligonucleotides (25, 26) are more sensitive, but are relatively laborious and may require follow-up by direct sequencing.

In the past few years, a set of new technologies for massively parallel sequencing (MPS), which are also referred to as high-throughput, next generation, or ultradeep sequencing, have emerged. They allow rapid generation of extremely large amounts of sequence information (27) and are primarily used to sequence genomes of higher organisms and for metagenomic studies of populations of various bacteria and viruses (28, 29). These new technologies are also very promising for analysis of viral quasispecies and have been used to study HIV (30, 31), severe acute respiratory syndrome (SARS) (32), hepatitis B (33), and other RNA viruses (29). MPS was also used for detection of adventitious agents in vaccines (34). Here we demonstrate that MPS can be used to identify very small amounts of mutant viruses in preparations of live viral vaccines as well as for their accurate quantification. The ability to quantify potentially harmful mutations in vaccine batches makes this method suitable for quality control and ensuring manufacturing consistency of live viral vaccines.

Results

In the first experiment, pyrosequencing (technology developed by 454 Life Sciences, a Roche company, further referred to as 454 Life Sciences) was used to analyze full-length PCR amplicons prepared from two samples of type 3 OPV: one that passed the monkey neurovirulence test and another that failed. Both were made from passage level 3 of the Sabin original seed stock (SO+3). The lot that passed the MNVT was manufactured in primary monkey kidney cells, whereas the failed lot was grown in WI-38 human diploid cells. Fig. 1A shows that the number of times each nucleotide was determined in positive and cDNA strands was about the same, but varied between 2,000 and 10,000 depending on the genomic location with the total for both strands being between 6,000 and 15,000. Both samples contained a number of mutants, the most prominent of which was the one with C₂₄₉₃U mutation, which changes threonine to isoleucine at amino acid 6 of the capsid protein VP1 (35, 36). The mutation does not affect neurovirulence, but the virus with 2493-U rapidly accumulates upon growth in cultured cells (15). Importantly, in agreement with our previous MAPREC results, there was a difference in the content of 472-C mutants, which correlated with the MNVT results. The vaccine lot that passed the test contained 0.35% 472-C, whereas the lot that failed it contained 2.4%.

Fig. 1.

MPS analysis of two batches of type 3 OPV performed by pyrosequencing. (A) The number of times each nucleotide was read in forward (green) and reverse (red) orientations. (B and C) Mutational profiles for vaccine batches that failed and passed the MNVT, respectively. Here and in all other figures the contents of mutants is shown by colored bars: mutations to A shown in orange, mutations to C in red, mutations to G in blue, and mutations to U in green.

There were a number of other mutations reaching several percent of the population in addition to a low level of less than 0.5% present at all nucleotide positions. This background level may represent true heterogeneity due to the quasispecies nature of the viral population. However, it could also include mutations that arose during reverse transcription and PCR amplification or represent the intrinsic inaccuracy of the pyrosequencing method. To determine the origin of the background, Sabin 3 cDNA plasmid, PCR amplicons prepared from the plasmid, and the virus stock recovered by transfecting HeLa cells with T7 transcripts of the plasmid, both before and after passaging in Vero cells, were analyzed by MPS. Table 1 shows that there was a difference in the level of background sequence heterogeneity detected in the plasmid DNA, PCR amplicons, and virus stocks. The average frequency of mutations in DNA plasmids was 0.05%, whereas there were 0.104% mutations in PCR product and 0.121% in the rederived virus. Although there was no apparent bias for the type of mutations detected in plasmid DNA (Fig. 2), in all other samples prepared by PCR amplification there were significantly more transitions (purine–purine and pyrimidine–pyrimidine changes) than transversions (purine–pyrimidine and pyrimidine–purine changes). After passaging the virus in HeLa cells (seven passages), the average content of mutants increased to 0.155% and after passaging in Vero cells (five passages), was 0.197%. The excess of mutants in the passaged virus could be accounted for by their increase at a select number of genomic loci, including 472-C, 2493-U, and several others, while the overall background level remained relatively unchanged. Therefore, mutants at roughly 0.1% represent intrinsic errors of sequencing procedures and mutations introduced by PCR amplification. Mutants in excess of this level may reflect the true sequence heterogeneity of viral quasispecies.

Table 1.

Frequencies of mutations identified by pyrosequencing in plasmid DNA, PCR product, and viruses

		Mutation frequencies
Sample	Total nucleotides read, 106	Substitutions, %	Insertions and deletions, %
Plasmid	28.1	0.050	0.424
PCR product	40.1	0.104	0.860
Plaque	18.2	0.121	0.719
HeLa p1	20.7	0.143	0.694
HeLa p7	36.6	0.155	0.858
Vero p5	19.0	0.197	1.232

Fig. 2.

Frequency of different types of nucleotide substitutions detected by MPS (Roche 454) analysis in plasmid DNA (A), PCR-amplified DNA (B), and virus rederived from T7 transcript of the plasmid (C). Bars represent the frequency of all possible transitions and transversions.

It is obvious from the results in Table 1 that there was a significant number of deletions, which exceeded the number of nucleotide substitutions. Because poliovirus contains one long ORF, the presence of insertions or deletions suggested a methodological artifact. To study this phenomenon, we analyzed a set of eight poliovirus samples by both pyrosequencing and Illumina methods. Various characteristics of both datasets shown in Table 2 demonstrate that the average sequence length was significantly higher for pyrosequencing (195 vs. 38), but the Illumina method produced roughly 60 times more sequence information. Therefore, the average number of times each nucleotide was sequenced was 885 for pyrosequencing and 56,650 for Illumina. The number of substitutions detected by each method was roughly the same, but the number of insertions and deletions was significantly lower in the samples analyzed by Illumina: 0.01% vs. 1.49% in pyrosequencing. Therefore, although both methods appear suitable for analysis of viral quasispecies, Illumina produced a better quality and greater number of sequences and thus may be the preferred method.

Table 2.

Comparison of pyrosequencing and Illumina methods for MPS

	Average length		Number of sequences, 103		Total nucleotides read, 106		Average coverage per nucleotide		% mutations (substitutions)		% insertions and deletions
Sample	Roche 454	Illumina	Roche 454	Illumina	Roche 454	Illumina	Roche 454	Illumina	Roche 454	Illumina	Roche 454	Illumina
1	218	38	69	25,177	11.5	850.6	774	57,154	0.20	0.30	1.17	0.01
2	199	38	101	26,361	15.2	881.8	1,025	59,254	0.23	0.20	1.61	0.01
3	223	38	108	26,398	18.6	912.5	1,254	61,390	0.17	0.19	1.75	0.01
4	179	38	77	26,272	10.5	868.5	709	58,429	0.21	0.18	1.67	0.00
5	196	38	95	25,667	14.2	850.9	956	57,246	0.18	0.16	1.26	0.00
6	207	38	79	26,065	12.3	719.8	830	48,423	0.22	0.19	1.40	0.01
7	159	38	79	24,359	9.2	855.8	616	57,577	0.23	0.17	1.55	0.00
8	180	38	102	24,669	13.6	799.6	917	53,796	0.26	0.30	1.47	0.01
Average	195	38	89	25,621	13.2	842.4	885	56,659	0.21	0.21	1.49	0.01

Results shown in Fig. 1 suggest that MPS could be used for the quantification of the proportion of the critically important 472-C mutants, which determine the level of neurovirulence of the virus stock. To assess the accuracy of this quantification, US neurovirulence reference NC2 (that by definition represents marginally acceptable vaccine) was tested along with two WHO International reference preparations for MAPREC: one that represents acceptable (96/572) and another unacceptable vaccine (96/578). By design, these WHO reference preparations for MAPREC bracket the proportion of mutants in the NC2 reference. Fig. 3 shows profiles of the sequence heterogeneity of these samples in the vicinity of nucleotide 472. The contents of 472-C in NC2 as determined from MPS data were 0.687% and was very close to our historical MAPREC data (0.729 ± 0.156%). The contents of 472-C for the 96/572 reference was 0.556% (0.642 ± 0.104% according to MAPREC) and for the 96/578 reference was 1.072% (1.010 ± 0.161% according to MAPREC). Therefore, there was a very close match between our previous MAPREC data and the results obtained by MPS.

Fig. 3.

Patterns of mutations in the vicinity of nucleotide 472 in Sabin 3 genome revealed by pyrosequencing. “Passed” WHO reference for MAPREC 96/572 (A); US National neurovirulence reference for type 3 OPV NC2 (B), and “failed” WHO reference for MAPREC 96/578 (C).

There was also good agreement between the contents of mutants at another unstable genomic position, 2,493, which was mentioned above. Although this mutation does not contribute to neurovirulence, its content in vaccines represents a marker of seed virus. Vaccines produced from the so-called Sabin original (SO) seed typically contain from 30 to 90% of these mutants, whereas vaccines prepared from plaque-purified rederived Sabin original (RSO, or “Pfizer”) seed stock contain mostly 2,493-C and only 1–4% of 2,493-U (15). Fig. 4 shows the results of MPS analysis of NC2 prepared from SO seed virus and a batch of RSO-based vaccine. There was 45.3% of 2,493-U in NC2 and 2.7% in the RSO vaccine lot according to the MPS analysis, very close to our previous MAPREC data. The profile of other mutations present in SO- and RSO-based vaccines was different, suggesting that the method could be used as a seed virus identity test. The pattern of mutations present in vaccines made from the same seed virus was highly reproducible. This is illustrated by a cluster of mutations in the viral 3D polymerase gene (Fig. 5). The mutations A₅₄₆₇G, U₅₄₇₃C, and U₅₄₇₆C (marked 1, 2, and 3 in the figure) are located in the third positions of codons and are silent. Mutations U₅₄₇₃C and U₅₄₇₆C (2 and 3) were present in the SO+2 stock (Fig. 5A) from which all SO-based vaccines were made. These mutations were never present in the same RNA molecule; therefore, the proportion of the virus population that contains these mutations is greater than 10%. A vaccine made in primary monkey kidney cells at SO+3 level by manufacturer A contains about the same amount of these two mutants (Fig. 5B), whereas vaccine made by manufacturer B contains one additional A₅₄₆₇G mutant (Fig. 5C). Passaging of this vaccine 10 times in Vero cells did not result in further accumulation of these variants, and their content even decreased slightly (Fig. 5D). A vaccine lot made in human diploid cells by manufacturer C contained a significantly increased amount of A₅₄₆₇G (Fig. 5E). Not surprisingly, vaccine prepared from plaque-purified RSO seed virus (Fig. 5F) did not contain mutants 2 or 3. However, it contained a small amount of mutant 1, which probably accumulated in the course of limited passaging used to prepare master and working seed viruses and to produce a lot of vaccine. This vaccine batch was made by manufacturer B, similar to the batch shown in Fig. 5C, suggesting that production conditions used by this manufacturer favor selection of this mutant. Fig. 5 G and H shows results with vaccines made by manufacturers D and E, which use plaque-purified Sabin 3 seed viruses different from RSO. One of these vaccines did not harbor any of the three mutants, whereas another contained a small amount of mutant 1 and 100% of mutant 2 plus one other mutant A₅₄₈₂U, which were presumably selected at random during plaque purification and amplification. Thus, analysis of nucleotide heterogeneities could be used to identify the origin of vaccine stocks and to monitor consistency of vaccine production conditions to avoid selection of mutants.

Fig. 4.

Mutational profiles of two batches of type 3 OPV prepared from SO seed virus (A) and plaque-purified RSO seed virus (B).

Fig. 5.

Patterns of silent mutations in a region of 3D/polymerase gene. Vaccines are: SO+2 (A); SO+3, manufacturer A (B); SO+3, manufacturer B (C); same as C but passaged 10 times in Vero cells (D); SO+3, manufacturer C (E); RSO, manufacturer B (F), plaque-purified seed stock, manufacturer D (G); plaque-purified seed stock, manufacturer E (H).

Discussion

In this communication we report the use of MPS for analysis of mutants present in OPV and demonstrate that the method was highly sensitive and allowed us to simultaneously detect and accurately quantify mutants at all positions of the entire viral genome. Close match between the MPS results obtained for the calibrated reference materials previously tested in MAPREC suggested that MPS could be used instead of MAPREC for lot release of OPV and take advantage of the existing validated reference reagents. In a recent report, OPV and other viral vaccines were analyzed by MPS for the presence of sequence heterogeneities, but the authors failed to detect any connection between the presence of minority sequence variants and vaccine quality (34). This may have been due to either insufficient sensitivity of the MPS method used, or the absence of reference materials previously tested by other methods. In either case the present communication uniquely reports the utility of MPS for quality control of viral vaccines.

One critical advantage of MPS over MAPREC is that it allows all nucleotide positions in complete viral genomes to be screened in one assay and therefore addresses concerns that mutations at unknown genomic loci could emerge but remain undetected and thus compromise vaccine quality. We were able to distinguish patterns of mutations that were characteristic for specific seed virus; therefore, groups of vaccine lots produced from the same seed could be identified by the specific patterns of mutations that they contain. For instance, lots produced from the Sabin original seed stock of type 3 poliovirus contained three closely located mutations in the 3D polymerase gene: A₅₄₆₇G, U₅₄₇₃C, and U₅₄₇₆C. These three silent mutations were never present in the same RNA molecule, and together they constitute about 10–20% of the virus population. They do not appear to provide a clear or universal selective advantage in vitro because after 10 passages in Vero cells, their contents remained mostly unchanged. However, at least one of the mutants (A₅₄₆₇G) was consistently and independently selected under manufacturing conditions used by two vaccine manufacturers. This example shows that MPS could be used to detect the presence of a low frequency of signature mutations that reflect subtle differences in manufacturing conditions and therefore be used for monitoring molecular consistency of viral vaccines. Unlike OPV, other live viral vaccines are not routinely tested for residual neurovirulence at the level of individual vaccine batches, but phenotypic tests to ensure genetic consistency of vaccine manufacture are an important part of qualification of seed viruses and are performed when significant changes to production conditions are introduced. Profiles of mutations present in vaccine batches could be analyzed by MPS and used to ensure consistency in the manufacturing process, which is a cornerstone of vaccine safety. The ability of MPS to measure small quantities of mutants could also be used in basic virus research to study effects of different mutations on virus fitness by monitoring their selection at different growth conditions.

Analysis of a plasmid DNA, a PCR product amplified from it, as well as virus stocks recovered from the plasmid after T7 transcription revealed that there was on average 0.05% of mutations in the homogeneous plasmid DNA preparation, presumably due to the inherent rate of base-calling errors. This number is consistent with the average Phred sequence quality value for the data analyzed in this work. The rate of mutations in the PCR amplicon produced from the plasmid was higher (about 0.1%). Prevalence of transitions over transversions in all PCR-amplified samples probably reflects the intrinsic bias of nucleotide misincorporation typical for most polymerases. The background level of mutations produced by PCR amplification and base-calling errors inherent to any sequencing technology puts a lower limit on the number of mutations that could be detected by this method.

Two versions of MPS were used in this study: pyrosequencing (454 Life Sciences, a Roche company) and Illumina methods. Whereas pyrosequencing produced longer sequences, the Illumina approach generated roughly 60 times more sequence information, allowing more sensitive detection of minority sequence variants. Although the accuracy of base calls was roughly similar, pyrosequencing erroneously identified a significant number of insertions and deletions, especially in regions where the same base is repeated several times. This shortcoming is inherent to the base-calling method used in pyrosequencing and was largely compensated by software used for data processing. Therefore, the Illumina approach may be preferable, but pyrosequencing can also be successfully used for screening and quantification of mutants in viral vaccines. In this work, we have used both full-length PCR amplification, as well as amplification of two overlapping segments that covered the entire genome; there was no difference in the results between these two approaches.

Practical aspects of implementing this approach must be considered. The MPS equipment is highly sophisticated, expensive, and requires maintenance and operation by trained personnel. Although it is still relatively expensive, its cost, which varies depending on the number of samples being analyzed in one run, is significantly lower that that of neurovirulence tests. It is widely expected that the cost of MPS will undergo a dramatic reduction with the introduction of next generation methods based on ion torrent technology or single-molecule sequencing.

In conclusion, making MPS analysis a part of vaccine lot release, characterization of new vaccine strains, and qualification of seed stocks promises to significantly increase molecular consistency and therefore ensure the safety of live viral vaccines, and also move them one step closer to being well-characterized biological products.

Materials and Methods

Virus Preparations and Growth.

Reference preparations of OPV were obtained from WHO and Center for Biologics Evaulation and Research (CBER) repositories, and monovalent vaccine lots were obtained from their manufacturers. Serial virus passaging was performed in HeLa and Vero cells by inoculating cell cultures at the multiplicity of infection of about 1 TCID₅₀/cell.

Preparation of DNA for Analysis.

Viral RNA was isolated from cell culture supernatants or from monovalent vaccine lots using PureLink viral RNA/DNA mini kit (Invitrogen). Isolated RNA was reverse transcribed by ThermoScript RNase H− reverse transcriptase (Invitrogen) using A3-As primer (T₃₀CCTCCGAATTAAAGAAAAATTTACCCCTAC). The reaction was carried out for 5 h at 55 °C in a total volume of 20 μL using conditions recommended by the manufacturer. After completion of the RT step, residual template DNA was removed by digestion with 2 U of RNase H (Invitrogen). cDNA was PCR amplified either as a full-length amplicon or in two partially overlapping amplicons. Full-length amplification was performed as previously described (37). For two-amplicon reactions, PCR primers S1-T7: GCGGCCGCTAATACGACTCACTATAGGTTAAAACAGCTCTGGGGTTG and 3797R: CCCTGCTCCATGGCCTCTTCCTCGTAAGC were used for amplification of the left half of the genome and primers 3734F: ATCGGAATCGTGACAGCTGGTGGAGAGGG and S3-R: CCTCCGAATTAAAGAAAAATTTACCCCTACAACAGTATGACCCAATCC were used for the right half. In both cases, reactions were performed with Platinum PCR SuperMix high fidelity kit (Invitrogen). PCR products were purified using QIAquick PCR purification kit (Qiagen), and their quality was checked by agarose gel electrophoresis and by measuring absorbance at 260 and 280 nm using a NanoDrop 1000 spectrophotometer (Thermo Scientific). Two half-length PCR products were mixed in equimolar proportion and subjected to MPS.

Massively parallel sequencing was performed by either pyrosequencing (454 Life Sciences technology) (38) or Illumina/Solexa method (39). Pyrosequencing was done at the Laboratory of Molecular Technology, Advanced Technology Program, SAIC (Frederick, MD). An experimental format identical to the manufacturer's protocol for de novo genome sequencing was used throughout the study. DNA was sheared using an S2 sonicator (Covaris) to produce DNA molecules with an average length of 500 bp. To avoid contamination, each sample was sonicated in a single-use tube. Libraries for sequencing were prepared using GS FLX Titanium Rapid Library preparation kit (454 Life Sciences). Libraries were clonally amplified by emulsion PCR (emPCR) with GS FLX Titanium MV emPCR kit (Lib-L) (454 Life Sciences). Beads with amplified libraries were loaded onto GS FLX Titanium PicoTiterPlate with dividers with separate reaction chambers to accommodate eight individual samples. Sequencing reactions were carried out using FLX Genome Sequencer (454 Life Sciences) with GS FLX Titanium reagents (454 Life Sciences). Initial data treatment (image acquisition, base calling, quality estimation) and data extraction and filtering were performed using software supplied by 454 Life Sciences with default settings.

Analysis using the Illumina/Solexa platform was performed at Macrogen. DNA shearing, library preparation, and cluster generation were conducted according to the manufacturer's recommendations. Sequencing reactions were performed on a Genetic Analyzer IIx (Illumina) using a 38-cycle setting with a single-read option. Each sample was loaded onto a single line of the flow cell for a total of eight samples per run. Initial data analysis was performed by the instrument's software, using default options.

Bioinformatic Treatment.

MPS data were processed using custom software developed in this laboratory. Sequence data with Phred quality scores below 20 were discarded. Newly generated sequences were aligned with amplicon sequences, and mutations at each nucleotide were recorded. Results with significant strand polarity bias, as well as insertions and deletions adjacent to homopolymeric stretches in pyrosequencing data were disregarded to reduce the effect of sequencing artifacts. The results were used to construct plots showing the representation of each mutation in the genome.