Systematic Identification of H274Y Compensatory Mutations in Influenza A Virus Neuraminidase by High-Throughput Screening

Abstract

Compensatory mutations contribute to the appearance of the oseltamivir resistance substitution H274Y in the neuraminidase (NA) gene of H1N1 influenza viruses. Here, we describe a high-throughput screening method utilizing error-prone PCR and next-generation sequencing to comprehensively screen NA genes for H274Y compensatory mutations. We found four mutations that can either fully (R194G, E214D) or partially (L250P, F239Y) compensate for the fitness deficiency of the H274Y mutant. The compensatory effect of E214D is applicable in both seasonal influenza virus strain A/New Caledonia/20/1999 and 2009 pandemic swine influenza virus strain A/California/04/2009. The technique described here has the potential to profile a gene at the single-nucleotide level to comprehend the dynamics of mutation space and fitness and thus offers prediction power for emerging mutant species.

INTRODUCTION

Influenza A viruses, including the well-known 1918 Spanish influenza virus, cause mortality and significant societal cost globally every year (1). Recently, the appearance of swine-origin H1N1 influenza A virus also drew the world's attention (2). Drugs have been developed to offer protection against influenza virus infection (3, 4). However, rapid viral mutations pose a challenge for drug efficacy (5–9). There is a need to develop methods for prediction of the emerging influenza A virus mutants.

Oseltamivir (Tamiflu; Hoffmann-La Roche) was discovered in the late 1990s and has been purchased by more than 70 governments at a cost of multiple billions of dollars (10, 11). Oseltamivir biochemically targets neuraminidase (NA), the viral integral membrane protein, which functions in the releasing of viral particles from the host cells (12, 13). In 2007 to 2008, influenza A virus carrying residue 274 substituted from histidine to tyrosine (H274Y [N2 numbering]), which confers oseltamivir resistance and had long been thought to compromise viral fitness, spread throughout the world (14, 15). The presence of H274Y mutations not only raises a public health concern but also highlights the role of permissive mutations in viral evolution (16).

To understand the dynamic nature of the influenza A mutations, phylogenetic analysis or in vitro selection has been commonly employed (16–22). However, the phylogenetic approach does not allow the evaluation of all possible mutations and has a prediction power restricted by dependency on preexisting clinical isolates. In vitro selection, on the other hand, has limited sensitivity due to its dependence on stochastic spontaneous mutation. In this report, we describe a new experimental approach which couples saturation mutagenesis with next-generation sequencing to study the evolution of the viral NA protein in the oseltamivir-resistant influenza A virus. Our results reveal that the replication fitness of influenza A/WSN/33 (H1N1) virus can be fully restored by the substitution R194G (numbering based on WSN sequence) or E214D (N1 numbering) or partially restored by L250P (N1 numbering) or F239Y (N1 numbering). This high-throughput screening approach is also sensitive to mutations with a minor beneficial effect and allows single-nucleotide profiling of the entire gene or genome. One representative compensatory mutation identified by this method, E214D, is further characterized and suggested to have an important role in the emergence of swine-origin 2009 H1N1 influenza virus.

MATERIALS AND METHODS

Cell lines, plasmids, and subclonings.

The C227 cell line, stably expressing dominant-negative IRF-3, is derived from human embryonic kidney (293T) cells (23). The H1N1 strain A/WSN/33 (WSN) eight-plasmid reverse-genetics system was kindly provided by Hoffmann et al. (24). The A/New Caledonia/20/1999 (NC99) NA segment and hemagglutinin (HA) segment, all eight segments of A/Texas/36/1991 (TX91), and all eight segments of A/California/04/2009 (swine-origin influenza virus) were kind gifts from Jesse Bloom at the University of Washington. A 4-μg volume of WSN NA plasmid containing H274Y was used as a template for 28 PCR cycles with error-prone polymerase Mutazyme II (Stratagene) to create the insertion library. Bsa1 restriction enzyme sites were added to the PCR primers (forward, 5′-GTGTGTGGTCTCAGGGAGCGA-3′; reverse, 5′-GTGTGTGGTCTCGTATTAGTA-3′) and cloned into BsmB1-digested pHW2000 vector (24). Transformation was conducted with electrocompetent MegaX DH10B T1R cells (Invitrogen). Around 800,000 colonies were collected to generate the NA plasmid library. A total of 10 clones were submitted for Sanger sequencing (Laragen) to approximate the mutation frequency. N-terminal Flag-tagged NA was created by cloning the WSN NA gene into pcDNA 5/TO (Invitrogen). Individual mutant constructs were created by site-directed mutagenesis (Agilent).

454 sequencing and data analysis.

Neuraminidase cDNA was PCR amplified into two nonoverlapping fragments with a multiplex identifier (MID) and the constant region for sequencing flanking the primer according to the primer design instructions for the 454 sequencing system (Integrated DNA Technologies). Briefly, each PCR fragment was approximately 700 bp long, and a bidirectional read was used during the sequencing process. 454 FLX sequencing was performed at UCLA GenoSeq Core. Sequence alignment was done through BLAST 2.2.24. A custom perl script was employed for downstream analysis. After setting the Phred quality score cutoff at 20, the mean Phred quality score was 32.46, which corresponds to a 0.06% base-calling error rate. Assuming the sequencing error rates for each of the 3 changes (e.g., A to T, C, or G) are evenly distributed, the base-calling error rate for each exact change is 0.06%/3 = 0.02%. Using the Poisson distribution model (λ = 0.0002 × total number of base calls at a particular position), a minimum read cutoff with a 5% false-discovery rate can be set for each position.

Viral selection and growth experiment.

To produce WSN virus, C227 cells were transfected with Lipofectamine 2000 (Invitrogen) using the neuraminidase mutant plasmid library plus the remaining 7 plasmids of the influenza virus reverse-genetics system. Media were replaced at 24 and 48 h posttransfection. Supernatant containing infectious viral particles was harvested at 72 h posttransfection and filtered with a 0.45-μm-pore-size mixed cellulose ester (MCE) filter. 293T-MDCK coculture cells were transfected to produce NC99 (HA segment and NA segment from NC99 and other 6 segments from TX91) or pdmNA-WSN virus (NA segment from A/California/04/2009 and all of the other 7 segments from WSN). At 24 h posttransfection, cell growth media were replaced with viral growth media (Dulbecco's modified Eagle's medium [DMEM] supplemented with 0.3% bovine serum albumin [BSA], 0.01% serum, 2 μg/ml tosylsulfonyl phenylalanyl chloromethyl ketone [TPCK]-trypsin, and penicillin-streptomycin). Supernatant containing infectious viral particles was harvested at 60 h posttransfection and filtered with a 0.45-μm-pore-size MCE filter.

For infection or selection experiments, A549 cells were used for WSN and MDCK cells were used for NC99 and pdmNA-WSN virus for infection and titering. Briefly, three phosphate-buffered saline (PBS) washes were conducted at 2 h postinfection, and virus in the supernatant was harvested at 24 h postinfection. Titers were determined as described in reference 25. Oseltamivir was a kind gift from Hoffmann-La Roche Pharmaceuticals.

Neuraminidase activity assay.

Equal amounts of NA plasmids were transfected into 293T cells using Lipofectamine 2000 (Invitrogen). At 24 h posttransfection, the media were removed and the cells were collected by resuspension in isotonic buffer (15 mM morpholinepropanesulfonic acid [MOPS], 145 mM sodium chloride, 2.7 mM potassium chloride, 4.0 mM calcium chloride, and 2% heat-inactivated fetal bovine serum). NA activity was assayed using the fluorogenic substrate methylumbelliferyl-α-d-N-acetylneuraminic acid (MUNANA) (purchased from Sigma) following a previously described protocol (http://www.nisn.org/documents/A.Hurt_Protocol_for_NA_fluorescence.pdf). The fluorescence readout was subtracted from background and normalized to the cell concentration via cell counting. For the total cell lysate NA activity assay, the same protocol was used except that the cells were lysed in TEB (50 mM Tris [pH 7.5], 150 mM NaCl, 1.0% Nonidet P-40, 10% glycerol, 10 μg/ml aprotinin, 10 μg/ml pepstatin, 0.5 mM phenylmethylsulfonyl fluoride).

Western blot analysis.

293T cells on a 12-well plate were transfected with equal amounts of Flag-tag NA expression plasmid using Lipofectamine 2000 (Invitrogen). At 24 h posttransfection, cells were lysed by TEB and equal volumes of total cell lysate were heated with SDS loading buffer for 5 min and loaded onto a 10% polyacrylamide gel. Milk (5%) was used for blocking. Mouse anti-Flag antibody (Sigma) and sheep horseradish peroxidase-conjugated anti-mouse immunoglobulin G (GE Healthcare) were used as primary and secondary antibodies. For the loading control, mouse anti-beta actin antibody (Sigma) was used as the primary antibody.

Pulse-chase experiment.

293T cells on a 6-well plate were transfected with equal amounts of Flag-tag NA expression plasmid using Lipofectamine 2000 (Invitrogen). At 24 h posttransfection, cells were washed with PBS and incubated at 37°C and 5% CO₂ for 1 h with 800 μl DMEM (minus cysteine [Cys]/methionine [Met], 5% fetal bovine serum [FBS]). [³⁵S]Met/Cys Express label (PerkinElmer) was then added (0.1 mCi/well), and the mixture was incubated for 1 h (pulse). Cells were washed with PBS, and the radioactive media were replaced with cell growth media (DMEM with 10% FBS) after the 1-h pulsing. At the indicated time point, cells were washed with PBS and resuspended in PBS. Cell pellets were obtained by centrifugation for 5 min at 3,000 rpm and were stored at −80°C when necessary. Cell pellets from different time points were lysed in parallel on ice for 20 min using TEB (50 mM Tris [pH 7.5], 150 mM NaCl, 1.0% Nonidet P-40, 10% glycerol, 10 μg/ml aprotinin, 10 μg/ml pepstatin, 0.5 mM phenylmethylsulfonyl fluoride). Equal amounts of total protein from different samples were then incubated with TEB and loaded onto an EZview Red Anti-Flag M2 Affinity gel (Sigma) for 2 h. Beads were washed 3 times with TEB. Elution was done in 200 μl SDS loading buffer (50 mM Tris-HCl [pH 6.8], 2% SDS, 10% glycerol, 1% β-mercaptoethanol, 12.5 mM EDTA, 0.02% bromophenol blue), and the mixture was heated at 90°C for 5 min. Samples of equal volumes were loaded onto a 10% polyacrylamide gel. The resultant gel was dried at 60°C using an FB-GD-45 gel dryer (Fisher Scientific) and exposed for 3 days. Quantification was done by ImageJ (26).

RESULTS

Screening for compensatory mutations.

H274Y mutation was introduced to the NA gene of H1N1 A/WSN/33 (WSN) using site-directed mutagenesis. Random mutagenesis utilizing error-prone polymerase was done on the H274Y NA gene to generate the mutant library with an average mutation frequency of approximately 5.6 per clone. The mutant library was transfected into C227 cells (23), which are subclones of 293T cells stably expressing dominant-negative IRF-3 and are able to increase the viral titer by ∼10%, to generate the viral mutant library using the eight-plasmid reverse-genetics system (24). The viral mutant library was then processed through successive rounds of 24-h infections in human lung carcinoma A549 cells at a low multiplicity of infection (MOI) (0.003 to 0.01) to reduce transcomplementation, with increasing oseltamivir concentrations (from 250 nM to 2 μM) throughout selection (Fig. 1A).

Fig 1.

Overview of the saturated mutagenesis screening. (A) A DNA plasmid library of the H274Y background neuraminidase gene was generated through error-prone PCR followed by cloning. More than 800,000 clones were collected. The eight-plasmid reverse-genetics system of strain A/WSN/33 (H1N1) was used to generate the viral mutant library by transfection in C227 cells. Multiple rounds of infection were performed in the presence of increasing concentrations of oseltamivir for selection. 454 sequencing was performed to survey mutants within selection pools. (B) Candidate mutations (red) that had a frequency of over 1% and 50 reads containing the mutation were identified. A total of 10 candidates were present in the post-round 2 pool, while 6 candidate mutations remained in the post-round 4 pool. Only missense mutations are shown. (C) Single-nucleotide fitness profiling of neuraminidase by categorizing mutations according to the enrichment ratio of mutation occurrence frequency in posttransfection round to post-round 2 selections. Ratio > 5, score = 3; 5 ≥ ratio > 1, score = 2; 1 ≥ ratio > 0.1, score = 1; 0.1 ≥ ratio > 0, score = −1; ratio = 0, score = −2. Confidence is color coded by the number of reads containing the mutation in the posttransfection pool: >30, purple; 16 to 30, brown; 5 to 15, green (0 to 4, not displayed). (D) Enrichment ratio profile. Ratio > 5, red; 5 ≥ ratio > 1, blue; 1 ≥ ratio > 0.1, green; 0.1 ≥ ratio, black. Only missense mutations are shown.

We used 454 pyrosequencing to survey the mutant pools at different stages of selection: the DNA plasmid library, the viral library after transfection, selection round 2, and selection round 4. Nucleotides with low-quality base calling (Phred quality score < 20) were filtered out such that a reliability score for each base call of least 99% was achieved. To examine the selection progress at different stages of selection, a preliminary analysis was done to digitally identify beneficial mutations with an occurrence frequency of over 1% and covered with ≥50 reads. Among the mutations identified, 10 satisfied the cutoff in the post-round 2 selection pool and 6 remained in the post-round 4 selection pool (Fig. 1B and Table 1). Therefore, we utilized the data from the post-round 2 selection pool for the rest of the study to achieve higher sensitivity. R194G, which was previously shown to be able to compensate for the H274Y fitness deficiency in WSN (16), appeared to be a positive candidate in both the post-round 2 and post-round 4 selection pools, thus supporting the validity of our approach.

Table 1.

Summary of round 2 candidates' occurrence frequencies

Point mutation	Amino acid change	Occurrence frequency ratio
		DNA plasmid	Transfection	Round 2	Round 4
A602T	E214D	0.00159	0.00442	0.629	0.504
C588G	R194G	0.000789	0.000942	0.165	0.445
A1305G	N450D	0.00228	0.00298	0.081	0.066
A602C	E214D	0.000159	0.000327	0.058	0.03
A1077G	M373V	0.00097	0.00173	0.066	0.0162
T117C	S37P	0.00266	0.00347	0.029	0.015
G564T	A202S	0.00138	0.00141	0.028	0.013
T709C	L250P	0.00116	0.00236	0.038	0.0017
A156G	N50D	0.0023	0.00296	0.01	0.00014
T676A	F239Y	0.00207	0.00235	0.011	0

Single-nucleotide profiling and systematic identification of beneficial mutations.

To detect weakly beneficial mutations and also permit gene profiling at a single-nucleotide resolution level, the enrichment ratio of post-round 2 occurrence frequency to posttransfection occurrence frequency was computed for each mutation. To limit statistical errors, only mutations that had a <5% false-discovery rate in the posttransfection pool were considered. We reasoned that a mutant had to acquire a relative growth advantage to maintain its occurrence frequency and avoid being outcompeted. As a result, in addition to a strict cutoff (ratio > 5), we also set a less stringent cutoff (ratio > 1). A total of 9 mutations satisfied the strict cutoff (ratio > 5), and an additional 13 mutations satisfied the less stringent cutoff (ratio > 1) (Fig. 1D and Table 2). A lot of mutations disappeared (ratio = 0) or had a decrease of at least 10-fold in occurrence frequency (ratio < 0.1) after 2 rounds of selection (Fig. 1C and D). These mutations most likely occurred at codons encoding essential residues for viral growth. Utilization of the recently developed sequencing platform with higher throughput would improve the sensitivity.

Table 2.

Summary of candidates' read count ratios identified by fold enrichment between posttransfection pool and post-round 2 pool

Point mutation	Amino acid change	Read count in posttransfection	Read count ratioa
C588G	R194G	7	175.3
A602T	E214D	27	142.5
A1077G	M373V	5	38.4
A1305G	N450D	14	27.3
G564T	A202S	11	19.7
T709C	L250P	20	16.0
T117C	S37P	38	8.5
A66T	I20F	8	7.7
T78G	L24V	7	6.1
A125T	Q39H	12	4.1
T676A	F239Y	25	4.0
A156G	N50D	35	3.4
A591G	I211V	14	3.3
A636G	T226A	10	2.2
A813G	T285A	16	2.1
T709A	L250Q	17	1.3
C114A	H36N	39	1.3
A601G	E214G	10	1.2
T99C	S31P	34	1.1
T153A	C49S	15	1.1
T153C	C49R	49	1.1
A813T	T285S	13	1.0

^aRead count ratio values represent the ratio of the mutation frequency in the posttransfection pool to that in the post-round 2 pool.

It was surprising to see E214G (ratio = 1.2) in the candidate list (Table 2). Mutation E214G, which spread globally in circulating influenza virus strains (27), was reported to have a neutral effect on virus harboring the H274Y NA mutation (16). However, our data in combination with the clinical prevalence of E214G suggest that the mutation may provide a weakly beneficial effect for the H274Y influenza virus. The beneficial effect of L250Q (ratio = 1.3) described in our data is also supported by its fixed presence in all swine-origin 2009 H1N1 clinical isolates (data not shown).

Validation of individual mutant identified from the screen.

Compensatory mutant candidates with a mutation frequency > 1% and at least 50 reads of coverage in the post-round 2 selection pool were reconstructed using site-directed mutagenesis. R194G and E214D fully restored and F239Y and L250P partially restored the growth deficiency of the H274Y mutant while not affecting the replication kinetics in the wild-type (WT) background (Fig. 2A to C). (Note that the E214G and L250Q mutations mentioned above have amino acid substitutions different from those of E214D and L250P at the same positions). Since our random mutagenesis method does not restrict the number of mutations per viral mutant, it is possible that false-positive candidates could be considered bystander mutations that happen to appear in the fittest genome. It is also possible that the level of the beneficial effect of some candidates may be below the sensitivity of our validation experiment. Previously identified compensatory mutations, V234M and R222Q (16), did not appear in our screen, likely due to the strain differences. While seasonal influenza virus strains (A/New Caledonia/20/1999 and A/Brisbane/59/2007) were used in the previous studies to study the compensatory effect of V234M and R222Q (16, 28), we employed a mouse-adapted WSN strain in this study. Indeed, V234M and N222Q (WSN carries an asparagine [N] instead of an arginine [R] in this position) do not restore the growth deficiency of H274Y in strain A/WSN/33. A similar observation was reported in a swine-origin 2009 H1N1 strain, A/California/4/2009, using a neuraminidase functional assay (29). Among the validated compensatory mutations, all but L250P are distant from H274Y (Fig. 2D). A close spatial arrangement (side chains separated by 4.4 Å) suggested that V234M, a proposed permissive mutation (16), and F239Y might exert their compensatory effect through similar molecular mechanisms (Fig. 2D).

Fig 2.

Validation and location of the compensatory mutation. (A) A549 cells were infected with the wild-type WSN strain (WT) or the indicated mutants at an MOI of 0.01 with 1 μM oseltamivir where indicated. Virus was harvested at 24 h postinfection. Titer was measured by a 50% tissue culture infectious dose (TCID₅₀) assay using A549 cells. Error bars indicate the standard errors of the means (SEM) of the results of 2 independent experiments. (B and C) A549 cells were infected with the wild-type WSN strain (WT) or the indicated mutants at an MOI of 0.005. Virus was harvested at the indicated postinfection time point. Titer was measured by a 50% tissue culture infectious dose (TCID₅₀) assay using A549 cells. (D) The spatial locations of different mutations were projected on the neuraminidase crystal structure (Protein Data Bank [PDB] no. 3CL0) (32). Red, 274Y; orange, 250L; blue, 214E; green, 239F; purple, 194R; yellow, 234V. Oseltamivir is shown in sphere form. The image is created using PyMOL version 1.3.

NA activity of individual mutants identified from the screen.

To measure the NA activity of different mutants, a fluorometric enzyme activity assay based on the substrate 2′-(4-methylumbelliferyl)-α-d-N-acetylneuraminic acid (MUNANA) was used (30). Unlike R194G, E214D does not fully restore the NA activity to the wild-type (WT) level but is still capable of restoring the viral fitness to that of the WT. In addition, the compensatory effect of F239Y and L250P in viral replication was not directly correlated with the cell surface NA activity (Fig. 2A and B and Fig. 3A). Therefore, we suspect that either the WT has an excess of cell surface NA activity for viral replication or that the cell surface NA enzyme activity is not the sole determinant of viral neuraminidase activity for effective replication. Control experiments utilizing the N-terminal Flag-tagged WSN NA showed that the differences of the functional neuraminidase activity in cell surface and total protein lysates are not due to the differences in protein expression level or stability (Fig. 3B to E).

Fig 3.

Functional enzymatic activity of neuraminidase mutants. (A to C) A MUNANA assay was employed to measure (A) the cell surface activity of WT NA and different NA mutants; (B) the cell surface activity of of N-terminal Flag-tag WSN NA; and (C) the activity of N-terminal Flag-tag WSN NA in total cell lysate. (D) Western blot analysis was performed to measure the protein expression level of different NA mutants in total lysates of cells 24 h posttransfection. Mouse anti-Flag antibody was used for detecting NA. Mouse anti-beta-actin antibody was used as a loading control for detecting actin. Data from an experiment representative of two independent experiments are shown. (E) An S³⁵-labeled protein pulse-chase experiment was performed to measure the protein degradation rate of NA mutants. The S³⁵-labeled protein level was normalized to the 0-h time point for each mutant. Data from an experiment representative of two independent experiments are shown. All experiments were done in 293T cells. All error bars in this figure indicate the standard errors of the means (SEM) of the results of 3 independent experiments. The P values for the data in panel A were calculated by a one-sided Student's t test. *, P < 0.05.

Clinical prevalence of validated compensatory mutations.

F239Y and L250P offer a compensatory effect for H274Y viral replication in vitro but have a low prevalence in patient samples (Table 3). This finding raised the possibility of their emergence in circulating strains and the need of monitoring these two amino acid residues in clinical isolates. Whereas R194G has been described in the past (16, 31), we turned our attention to E214D, which is present in only 1.3% of seasonal H1N1 strains but is fixed in the swine-origin 2009 H1N1 influenza virus (Table 3). Interestingly, it was reported that H274Y did not significantly decrease the fitness of swine-origin influenza virus (32–34). We hypothesized that the difference between seasonal influenza virus and swine-origin influenza virus at position 214 allows the swine-origin influenza virus to carry H274Y without significantly compromising viral growth.

Table 3.

Presence of identified compensatory substitutions in clinically relevant isolates

Mutation	% seasonal influenza virusa	% seasonal influenza virus (H274Y)	% 2009 swine influenza virusa	% 2009 swine influenza virus (H274Y)
R194G	99.6	100	99.9	100
E214D	1.3	0	100	100
F239Y	0.8	0.3	0	0
L250P	5.9	0	0	0
V234 M	61.2	97.5	0	0
R222Q	63.4	97.2	0	0
H274Y	27.8	100	3.5	100

^aData were retrieved from the NCBI Influenza Virus Resource on 1 July 2011 (17). A total of 1,140 seasonal influenza virus and 1,006 swine-origin influenza virus full-length neuraminidase sequences were included.

Position 214 in swine-origin influenza virus and seasonal influenza virus.

Consistent with our hypothesis, in pdmNA-WSN (NA segment from A/California/04/2009 and all of the other 7 segments from WSN), the D214E-H274Y double mutant has a replication deficiency whereas H274Y or D214E alone does not (Fig. 4A). In addition, the D214E-H274Y double mutant has a cell surface NA activity less than 5% of that of WT, which correlates with its fitness deficiency (Fig. 4A and C). We also found that E214D is able to compensate for the deficiency of H274Y in NC99 (a seasonal influenza virus model in which the HA and NA segments are from A/New Caledonia/20/1999 and the other 6 segments are from A/Texas/36/1991) (Fig. 4B) and also to rescue the cell surface NA activity to around 80% of the wild-type level (Fig. 4D).

Fig 4.

Effect of mutation at position 214 of swine-origin influenza virus and seasonal influenza virus. (A and B) MDCK cells were infected with (A) pdmNA-WSN viruses at an MOI of 0.0005 or (B) A/New Caledonia/20/1999 virus (WT) at an MOI of 0.0001. Virus was harvested at 24 h postinfection. Titer was measured by a 50% tissue culture infectious dose (TCID50) assay using MDCK cells. (C and D) A MUNANA assay was performed to measure the cell surface neuraminidase activity of (C) A/California/04/2009 NA or (D) A/New Caledonia/20/1999 NA. All error bars indicate the standard errors of the means (SEM) of the results of 3 independent experiments.

DISCUSSION

NA H274Y mutation has been identified as the major mutation that confers H1N1 influenza A virus oseltamivir resistance (14). While H274Y has been thought to damage viral fitness, clinical isolations in recent years suggest that permissive mutations exist such that the drop in the fitness effect from H274Y can be minimized (16, 28). Phylogenetic analysis which relies on existing clinical isolations is often used to discover beneficial mutations (16–20, 22). However, the searching space is often limited by the defined number of clinical isolates being sequenced. In vitro selection or serial passaging could be employed for discovering beneficial mutations in virus (21), but it is limited by the stochasticity and frequency of spontaneous mutation. In addition, the selection process restricts the sensitivity to mutants that offer a minor beneficial effect, due to the highly competitive environment of selection. The high-throughput screening method presented in this study provides an alternative approach to screen for beneficial mutations in an unbiased fashion.

A few recent studies have demonstrated the concept of high-throughput surveillance of a mutant library such as functional mapping for the human WW domain using a phage display system (35) and genome functional element screening using random transposon insertion (36, 37). This is the first time, to the best of our knowledge, that this approach has been applied with single-nucleotide resolution to a whole gene. By surveying the mutant pool in early rounds of selection with 454 sequencing, we were able to avoid an out-competing effect exerted by the optimum mutation and discover suboptimum mutations, such as F239Y and L250P. By increasing the scale of the sequencing sampling with newer sequencing technologies, which would provide greater sequencing depth, it is possible to build a single-nucleotide profile of the whole genome and predict the dynamics of viral mutation space under different conditions. The technique employed in this study is widely applicable to different fields where reverse-genetics systems exist such as human immunodeficiency virus (38), hepatitis C virus (39), etc. It also has the potential to be adapted to screen for mutations in cellular genes.

The screen performed here identified a novel mutation, E214D, which can fully restore the H274Y viral replication efficiency. In contrast to the seasonal H1N1 strain, aspartic acid at position 214 is present in every clinical isolate of the swine-origin 2009 H1N1 strain. It could provide an explanation for the phenomenon that H274Y imposed a less deleterious effect on swine-origin influenza virus than on seasonal influenza virus (32–34). The novel compensatory mutations presented in this study enhance the forecasting power for the emergence of oseltamivir-resistant influenza virus. Influenza virus harboring a compensatory mutation is likely to acquire the oseltamivir-resistant mutation NA H274Y under conditions of oseltamivir treatment and to be less responsive to the treatment without any compromise in fitness. Therefore, to prevent future spread and minimize the social cost, precautions should be taken before treating an influenza virus-infected patient with oseltamivir when most isolates in an area are acquiring the compensatory mutations.

In summary, we present a new approach allowing the identification of beneficial mutations in a comprehensive fashion with a high sensitivity for detection of mutations with suboptimal potency. Besides identifying a novel compensatory mutation, E214D, which can fully restore the viral replication fitness of the H274Y mutant, our data also revealed mutations that could offer a smaller degree of beneficial effect, such as F239Y and L250P. The results of this study not only demonstrate the usage of high-throughput functional profiling to study a gene at single-nucleotide resolution but also broaden the understanding of NA protein evolution of oseltamivir-resistant influenza virus by offering a tool for quasispecies dynamic prediction.