Cell lines

Adherent MDCK (No. 84121903) or Vero (No. 88020401) cells were purchased from ECACC (Salisbury, UK). The cells were either grown in T75- and T175- flasks at 5% CO₂ and 37°C or in roller bottles with closed caps at 37°C. For cell growth GMEM (Invitrogen/Gibco, #22100-093, Darmstadt, Germany) was supplemented with 5.5 g/L glucose (Roth, #X997.3, Karlsruhe, Germany), 2 g/L peptone (IDG, #MC33, Lancashire, UK), 10% FCS Invitrogen/Gibco, #10270-106, Darmstadt, Germany) and 4 mg/mL NaHCO₃ (Roth, #6885.3, Karlsruhe, Germany).

Virus

Infections were performed with influenza A/Puerto Rico/8/34 (H1N1) from either the National Institute for Biological Standards and Control (NIBSC, #06/114, South Mimms, UK) or the Robert Koch Institute (RKI, Amp. 3138, Berlin, Germany) according to Genzel et al. [27]. In the following these virus strains are referred to as PR/8/34 (NIBSC) or PR/8/34 (RKI), respectively. Briefly, for infection and virus production the serum-free growth medium was supplemented with a final concentration of 5 U/mL trypsin (Invitrogen/Gibco, #27250-018, Darmstadt, Germany), which was prepared in phosphate buffered saline (PBS) according to the activity given by the supplier.

Virus growth for stability testing of HA N-glycosylation pattern and cell line adaptation studies

All infections were performed when cells were confluent using 0.2 mL virus seed for T75 flasks, 0.4 mL for T175 flasks or 1 mL for roller bottles (a volume-based passaging was chosen since the time required for determination of the number of infectious virus particles by TCID50 or plaque assays was too long for the experimental setup). Furthermore, Genzel et al. showed that the multiplicity of infection (moi) has no effect on final HA virus titers, only on the time needed to reach this maximum HA value [28]. To generate the initial virus inoculum (passage 1), MDCK cell-adapted virus seed was thawed and propagated for 24 h in MDCK cells using a moi of 2 or 0.03 for PR/8/34 (RKI) and (NIBSC), respectively. All successive passages (2 to 11) were infected with virus culture supernatant of the previous cultivation provided that HA-titers were above 1.8 log HA units/100 µL (1.8 HAU) - the minimal HA-titer required for N-glycan analysis. Overall, the initial MDCK cell-adapted virus was passaged five times in Vero cells and subsequently for back-adaptation up to five times in MDCK cells. As a control the virus seed was passaged in MDCK cells only (initial moi 0.3). In order to monitor the impact of harvest time points, MDCK (initial moi 0.3) and Vero (initial moi 0.1) cell-adapted virus seed was propagated in MDCK and Vero cells, respectively and isolated at different time points within the time frame of interest for MDCK as well as Vero cell-derived virus. All controls were performed in roller bottles using PR/8/34 (RKI).

Hemagglutination assay

Influenza virus from cell culture supernatant was titrated by hemagglutination based on the method by Kalbfuss et al. [29]. Samples were serially diluted 1:2^0.5 in PBS resulting in a final volume of 100 µL using round-bottomed 96-well plates. 1.9–2.1×10⁶ red blood chicken cells in a total volume of 100 µL in PBS were added and incubated for 3 h to 4 h at room temperature. Light extinction was measured at 700 nm and plotted over the logarithm of the inversed dilutions. The end point was defined as the point of inflection - the last dilution (d) showing complete hemagglutination - corresponding to 1 HAU (HAU  =  HA Units  =  -log₁₀ d/100 µL). A sample's HA-titer is calculated by considering the dilution factor of the end point. Consequently, if e.g. the end point is reached at a dilution of 1/64 (1/10^1.8), this corresponds to an HA-titer of 1.8 HAU for the undiluted sample.

Virus purification, N-glycan analysis and data evaluation

Virus purification by gradient step centrifugation, sample preparation, N-glycan analysis and data evaluation were performed according to Schwarzer et al. [30], [31], but applying an optimized protocol. This included the substitution of 20 mM sodium hydrogen carbonate _(aq) with 50 mM ammonium hydrogen carbonate _(aq) during enzymatic in-gel-deglycosylation and N-glycan extraction as well as the reduction of the final sample volume for PNGaseF-digestion to 60 µL, which resulted in an increased enzyme concentration. Labeled samples were desalted by size exclusion chromatography as described by Schwarzer et al. [30]. Briefly, capillary gel electrophoresis with laser-induced fluorescence detection (CGE-LIF) was performed in an ABI PRISM 3100-Avant genetic analyzer (Applied Biosystems, Foster City, California, USA). The purified samples were diluted in HiDi formamide (#4311320C, Applied Biosystems, Foster City, California, USA). The N-glycan pool of an HA sample was separated by capillary gel electrophoresis, resulting in capillary electropherograms, which – after migration time normalization - corresponds to the N-glycosylation pattern, in which relative fluorescence units (RFU) are plotted over the normalized migration time (t_mig) in base pairs (bp). In N-glycosylation patterns, one peak corresponds to at least one distinct N-glycan structure. The total peak height (TPH) is calculated as the sum of all peaks and is set to 100%. Differences in relative abundances are expressed in percentage of relative peak height (RPH). Accordingly, overlays of multiple N-glycosylation patterns allow a direct comparison of different HA N-glycan pools. In the following peaks are annotated as shown in figure 1. Data were processed as published previously [30]–[32].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Peak annotation of MDCK and Vero cell-specific HA N-glycosylation patterns of influenza A/PR/8/34 (H1N1).

Relative fluorescence units (RFU) plotted over normalized migration time in basepairs (bp). The shifted overlay of the normalized capillary electropherograms, represents MDCK cell-specific (black) and Vero cell-specific (red) glycosylation patterns.

https://doi.org/10.1371/journal.pone.0027989.g001

Pyrosequencing and sequence assembly

Segment 4 (HA) of influenza A was sequenced using the Genome Sequencer FLX instrument (Roche, Mannheim, Germany). For DNA preparation the protocol of Höper et al. [33] was applied with the modifications described by Leifer and colleagues [34]. RT-PCR was conducted as described by Höper et al. [35]. For reverse transcription of the RNA genome segment, the Transcriptor High Fidelity cDNA Synthesis Kit (Roche, Mannheim, Germany) and for amplification of the cDNA the iProof High-Fidelity Master Mix (Bio-Rad Laboratories GmbH, München, Germany) was used. The sequencing libraries were generated according to Wiley et al. [36] followed by DNA binding to library capture beads and the recovery of single-stranded template DNA (sstDNA) library. For bead-bound clonal amplification, the DNA libraries were subjected to duplicate emulsion PCRs (emPCR) with the GS emPCR kit I (Roche, Mannheim, Germany), according to the manufacturer's instructions (2 copies per bead). After bead recovery and enrichment, the beads were sequenced using a GS LR70 sequencing kit (Roche, Mannheim, Germany) and the appropriate instrument run protocol provided by the vendor. The resulting sequencing reads were sorted according to the genome segments and assembled into one contig (i.e., a set of overlapping sequencing reads) per segment using the GS FLX sequence assembly software newbler (version 2.3; Roche, Mannheim, Germany). During the assembly, primer sequences were trimmed-off the raw data. All sequences were deposited in the GISAID EpiFlu database (www.gisaid.org), the accession numbers are compiled in the supplementary (table S1). For quasispecies analysis, a mapping of the raw sequencing reads along the reference sequence was performed using the GS FLX reference mapper software (version 2.3; Roche, Mannheim, Germany) with the default parameters (also during the mapping primer sequences were trimmed off the raw data). Briefly, ”high confidence” differences were defined by a combination of flow signal, quality score and difference type information: generally, at least three non-duplicate reads with the difference are required. Besides, there must be both forward and reverse reads showing the difference, unless there are at least seven reads with quality scores over 20. If the difference is a single-base over- or undercall (e.g. CC instead of C, or A instead of AA), the reads with the difference must form the consensus of the sequenced reads (i.e. at that location the overall consensus must differ from the reference) and the signal distribution of the differing reads must vary from the matching reads and the number of bases in that homopolymer of the reference (GS FLX Software Manual 2.17.1.14).

In the following, amino acid substitutions during the adaptation processes will be annotated by the one letter amino acid code for the original residue (in passage 1), followed by the position number of the residue, followed by the one letter code of the substituting amino acid in the later passage. Insertions and deletions are indicated by a minus sign (-) at the first or last position, respectively.

Sample storage

Virus containing cell culture supernatants as well as virus isolates were stored at −80°C; HA N-glycans for glycosylation pattern analysis were stored at −20°C. Sealed Coomassie-stained polyacrylamide gels with HA bands (#161-1155 or #161-1158, BioRad, Hercules, Canada) were stored at 4°C.

Stability of HA N-glycosylation pattern over multiple virus passages and over different harvest time points

One essential condition for the characterization of adaptation processes of influenza viruses A/PR/8/34 (H1N1) to different host cell lines is the stability of the HA N-glycosylation pattern over multiple virus passages in one cell line. For MDCK cell-derived PR/8/34 (RKI), HA-titers of 10 successive virus passages ranged between 2 and 2.4 HAU at 24 hpi (data not shown). Furthermore, all HA N-glycosylation patterns feature the same 15 main peaks (no. 1-4, 6-16) throughout all passages (figure S1A). The standard deviations (SD) of RPHs range between 0.25% and 1.14% (figure S1B). Overall, these results demonstrate a high stability of the HA N-glycosylation pattern over 10 successive virus passages in the same host cell system.

Furthermore, previous experiments showed that for guaranteed N-glycan analysis samples with a minimum HAU-value of 1.8 are required (data not shown). With differences in replication dynamics during adaptation to a new cell line, i.e. slow replication dynamics during initial adaptation steps and fast progress of infection towards the end, different time points for sampling were required to achieve a minimum of 1.8 HAU. Hence, the impact of harvest time point on HA N-glycosylation patterns was investigated for Vero cell-adapted PR/8/34 (RKI) virus in Vero cells, and for MDCK cell-adapted virus in MDCK cells. For all sampling points (24, 96 hpi) during MDCK cell-based virus replication, the HA N-glycosylation pattern was dominated by the same 15 main peaks (no. 1 - 4, 6 - 16; figure S2A), exhibiting SD for RPH between 0% and 2.3% (figure S2B).

In contrast, the pattern from Vero cell-derived HA was defined by 16 main peaks (no. 5, 7, 8, 11, 15 - 25, figure S3A). Regarding the complete time span (sampling points 48, 96, 200, 360 hpi) the SD for RPH ranged from 0.1% to 6% (figure S3B). Considering only the time span until 96 hpi, SD comparable to the MDCK time series were observed (SD<2.7%). Taking into account the comparatively long sampling period of 360 h, the higher SD were most likely caused by N-glycan degradation or synthesis. In particular, the RPH of peaks 5 and 7 steadily increased (it almost doubled for peak 7) and the RPH for peaks 16, 20, 22 to 25 steadily decreased over time. However, even despite those changes of RPH, the HA N-glycosylation pattern itself was stable over 360 h.

Overall, results demonstrated a highly reproducible HA N-glycosylation pattern for both cell lines concerning the number of glycan structures. For Vero cells, but to a certain degree also for MDCK cells (due to smaller time frame), the relative amounts (RPH values) of single structures present in the glycan pool, were dependant on the harvest time point of the virus-containing supernatant.

Adaptation of PR/8/34 from MDCK to Vero cells and back

In 2010 Genzel et al. showed that viruses often require adaptation to new host cells to optimize yields and time of harvest. In order to further characterize this adaptation process and to investigate possible biological mechanisms MDCK cell-adapted virus was propagated for 24 h in MDCK cells (passage 1). As a result HA-titers of 1.9 HAU and 2.2 HAU were obtained for the strains from RKI and NIBSC, respectively (data not shown). Supernatant of this passage served as the virus seed for a first infection of Vero cells (passage 2). For the RKI-strain virus-release was not detected before 288 hpi (0.8 HAU, figure 2A) and the maximal HA-titer of 1.4 HAU was finally reached 360 hpi. For the NIBSC-strain virus release was first detected 216 hpi and virus replication was continued until 360 hpi, exhibiting a final HA-titer of 1.95 HAU (figure 2B). An aliquot of these supernatants served as virus seeds for the next infections of Vero cells (passage 3) and so on (passage 4 to 6). During this virus adaptation to Vero cells, viral fitness of the RKI-strain improved: The time required to achieve specific HAU-values (≥ 1.4 HAU) decreased, whereas maximum HA-titers increased from passage 2 to 4. No significant differences were detected for passage 4 to 6, indicating the completion of the adaptation process (figure 2A). For the NIBSC-strain, the second passage in Vero cells (passage 3) reached 2.1 HAU at 120 hpi, and passage 4 reached 2.74 HAU at 96 hpi. For all subsequent passages the time required to achieve a HA-titer of at least 1.8 HAU as well as the maximal titers were more or less the same. This indicated the completion of the adaptation process (figure 2B). The following five virus passages (7 to 11) were again performed in MDCK cells to monitor virus back-adaptation. Here, for the RKI- as well as the NIBSC-strain all titers ranged between 2 HAU and 2.5 HAU and were reached between 48 hpi and 96 hpi (data not shown). In contrast to the adaptation to Vero cells, no impact on HA-titer level and virus release dynamics was observed during back-adaptation.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. HA-titers and shifted overlays of HA N-glycosylation patterns during influenza virus A/PR/8/34 (H1N1) adaptation from MDCK to Vero cells.

MDCK cell-adapted virus seed from RKI (A) as well as from NIBSC (B) replicate poorly in the first Vero cell passage (♦, passage 2). Passage (3) to (6) exhibit increased virus fitness (◊, ▴, Δ, ▪). For the HA-assay the 95% confidence interval is maximum 15%. (C, D) The normalized capillary electropherograms altogether represent 11 virus passages of the virus from RKI (C) and NIBSC (D): (1) virus seed, (2) to (6) adaptation to Vero cells, (7) to (11) back-adaptation to MDCK cells. MDCK and Vero cell-derived HA N-glycosylation patterns dominated by 15 peaks and 16 peaks, respectively.

https://doi.org/10.1371/journal.pone.0027989.g002

Host cell-specificity of HA N-glycosylation patterns

All HA N-glycosylation patterns of MDCK cell-derived virus samples were similar. The same applied to the Vero cell-derived virus samples (figure 2C). In agreement with earlier studies the HA N-glycosylation pattern was strictly host cell-specific and changed significantly with the switch to the new host cells [28], [31]. However, of all Vero cell-derived HA N-glycosylation patterns for the RKI- as well as the NIBSC-strain passage 2 revealed the biggest differences in RPH (figure 3A). Here, the RPH of peak 5 and 7 was almost twice as high as for all subsequent passages in Vero cells. This is in agreement with the time series in Vero cells (figure S3) where these RPH almost doubled until 360 hpi. Furthermore, the low abundant glycan structure represented by peak 11 was missing in passage 2 of both strains (figure 2B, D; figure 3). This is most likely due to a drop below the detection limit. During the adaptation of the RKI-strain the height of peak 16 decreased by a factor of two in passage 2 (figure 3A, figure 2C), which was in agreement with the steady decrease of peak 16 during time course experiments (figure S3). Besides, peaks 17 and 18 were missing in passage 2 of the RKI-strain. In the time series in Vero cells these peaks represented low abundant structures with only 0.2 – 2.7% of TPH (figure S3). This probably indicates again a drop below the detection limit.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Relative peak heights of influenza A/PR/8/34 (H1N1) HA N-glycosylation patterns during virus adaptation.

(A) for the isolate from RKI: In passages (1) and (7) to (11) virus was propagated in MDCK cell culture (▪). In passages (2) to (6) virus was propagated in Vero cell culture (□). Most of the 25 different major peaks are host cell-specific. Only the HA N-glycan structures represented by peak no. 7, 8, 11, 14 to 16 and 19 to 21 are present in virus samples from both host cells. (B) for the isolate from NIBSC: In passage (1) and (7) to (11) virus was propagated in MDCK cell culture (▪). In passage (2) to (6) virus was propagated in Vero cell culture (□). Most of the 25 different major peaks are host cell-specific. The HA N-glycan structures represented by peak no. 7, 8, 11, 14 to 16 and 19 to 21 are present in virus samples from both host cells.

https://doi.org/10.1371/journal.pone.0027989.g003

For each peak in both host-specific glycosylation patterns a comparison of the maximal SD in controls (biological replicates and the time courses) with the SD observed during adaptation experiments is shown in table S2. A more than 3-fold higher SD compared to the controls was considered significant. For the RKI-strain, during adaptation to Vero cells three peaks (11, 14 and 18) showed a more than 3-fold higher SD of RPH (table S2; Adaptation series H1N1, RKI) compared to the same peaks of the time series experiment (figure S3). Closer examination of the 15 main peaks of all MDCK cell-derived HA N-glycosylation patterns during back-adaptation revealed a more than 3-fold higher SD in RPH for peaks 6 and 14 (table S2; Adaptation series H1N1, RKI) compared to the respective time series and biological reproducibility experiments (figure S3). Of these peaks, only number 6, with an average RPH of 12.13%, represents a high abundant glycan structure above 5% of the TPH. For the NIBSC-strain, during forward adaptation, a 3-fold higher SD was found for the high abundant glycan structure represented by peak 14 (RPH >5%), as well as for another low abundant glycan structure represented by peak 19 (RPH <5%). Regarding the back-adaptation to MDCK cells the low abundant glycan structures of peaks 7 and 14, as well as the high abundant glycan structures of peak 6, 10 and 13 (RPH >5%) exhibited a more than 3-fold higher SD compared to controls.

Overall, however, a clear trend during the forward and backward adaptation process was not evident (figure 2C, figure 3). Furthermore, good reproducibility of the host cell-specific HA N-glycosylation pattern during virus adaptation to different host cell lines was demonstrated for both PR/8/34 influenza virus strains. Obviously, it is mainly the host cell line which determines the HA N-glycosylation patterns of a specific virus strain while the harvest time point or the viral adaptation status have only a minor impact on the respective RPH of glycan structures. Finally, the discrepancy of the HA N-glycosylation pattern for NIBSC- as well as the RKI-virus between the first passage in Vero cells (passage 2) and the patterns of all subsequent Vero cell adaptation passages (passages 3 – 6) can not completely be explained by differing harvest time points and remains to be further investigated.

Changes in quasispecies composition

To investigate the impact of forward and backward adaptations in both cell lines segment 4 coding for the HA was sequenced for passage 1, 6 and 11. Focus was in particular on changes in potential HA N-glycosylation sites. The consensus sequences of segment 4 of the initial, MDCK cell-derived virus seeds from passage 1 are shown in the supplementary (figure S5). Regarding the consensus amino acid sequences of HA, the initial virus seeds from RKI and NIBSC differed in 7 amino acid positions. The differences comprise K147-, A156E, E158K, I208L, R269M, F309Y and S398T (RKI vs NIBSC, figure S5). For the RKI-strain, sequencing analysis clearly indicated that the virus population from the first initial MDCK passage 1 was uniform concerning RNA segment 4, i.e. only one virus variant was detected above the detection limit. In contrast segment 4 of the initial virus seed from NIBSC (passage 1) already comprised several virus variants. The following variants were detected with the specified frequencies (table 1, B): Y24H (22%), T397A (1.3%), T397S (0.6%), D455Y (21.4%) and N460D (12.2%).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Changes of quasispecies composition on HA level of Influenza A/PR/8/34 (H1N1) (RKI, A) and (NIBSC, B) during virus adaptation.

https://doi.org/10.1371/journal.pone.0027989.t001

After complete adaptation to Vero cells (passage 6), 80% of the RKI-strain virus population carried amino acid substitution K460E, where a positively charged lysine was replaced by the negatively charged glutamic acid. Another population of 19% carried the S457L substitution, where polar serine was replaced by non-polar leucine (table 1A). These two substitutions were uncoupled. No single read was detected, which carried both substitutions. Hence, 99% of sequenced viruses carried either one of these two substitutions, indicating a crucial region of the HA for adaptation to efficient virus growth in Vero cells. Only single reads were detected that carried the original sequence of the first passage. This suggests that on the one hand, the initial virus did replicate in Vero cells, but only poorly. Considering the dilutions of the initial virus seed during the 5 Vero passages without any replication, its concentration would have been below 0.1 virion/mL in passage 6 and would most likely not have been detected at all. On the other hand, only a change in very few single amino acids in this HA-region is necessary to increase virus fitness for sufficient growth in Vero cells. In passage 6 of the adaptation of the NIBSC-strain a new variant carrying a V395M substitution was detected with a frequency of 41.5%, which introduced an additional sulfur containing residue into the HA2 peptide chain. Furthermore, the D455Y virus seed subpopulation dropped to 6.1%, but a new variant D455H, in which aspartic acid was substituted by the basic amino acid histidine, was detected that dominated the virus population with 52%. In addition, the N460D subpopulation increased to 41.1% (table 1B).

After back-adaptation of the RKI-strain to MDCK cells the dominating virus population (81%) still carried the K460E substitution in the HA. The minor subpopulation (19% after Vero-adaptation) carrying the S457L substitution decreased to 9% after back-adaptation. The remaining population of 10% represented the initial virus seed sequence from passage 1 (table 1A). This demonstrates that the K460E variant allows good virus replication in both cell lines. In contrast, the S457L variant seems to be less efficiently replicating in MDCK cells. These results, particularly the fitness of the K460E variant in MDCK as well as in Vero cells, strongly suggests that a mutation in this region was acquired, rather than an already existing virus subpopulation of the initial virus seed was selected. This mutation finally allowed sufficient virus replication in passage 2 of the adaptation series to reach an HA-titer of 1.4 HAU at 288 hpi. After back-adaptation of the NIBSC-strain to MDCK cells the V395M, the D455Y, the D455H, and the N460D variants decreased to 11.3%, 3.2%, 44.1%, and 10.5%, respectively. In contrast, the T397S variant that was not detected in passage 6, came-up again and made up for 5.4% in passage 11 (table 1B). This T397S substitution abolished the only difference of the consensus sequences in the HA2 chain from the two influenza viruses A/PR/8/34 (H1N1) strains from NIBSC and the RKI (figure S5). Furthermore a new variant carrying the K459E substitution was detected with a frequency of 44.2%.

All substitutions detected during adaptation of both A/PR/8/34 (H1N1) strains are located in the HA2 chain, neither inside nor in close proximity of any N-glycosylation site. They are, however, located in the inside of the HA trimer within or in close proximity to the fusion peptide pocket: within the subunits' contact site for the RKI-strain (figure S4A-D) and within the subunits' and monomer contact sites for the NIBSC-strain (figure S4E-H). This position belongs to the fusion subdomain [37]. Substitutions in this region of HA subunit contacts have been described to be crucial for the stability of the structure of the native protein. Among other factors, the optimal stability depends on pH or temperature [2], [38]-[40]. In this regard lower pH environments for instance require higher native structure stability, whereas elevated pH-values require less stable structure conformations to mediate membrane fusion. In 2010 Reed et al. demonstrated using recombinant H5N1 influenza viruses that substitutions within the fusion peptide pocket and the α-helix of HA2 alter the pH of activation of HA, which in term effects influenza virus pathogenicity as well as transmissibility in mallards [41]. Furthermore, Thoennes et al. showed that different substitutions at HA2 position 111 of a H3N2 influenza virus strain significantly effected fusion pH, suggesting, a key role of this residue for neutral pH structure stability [42]. Therefore, it remains to be investigated, whether the substitutions occurring during the adaptation processes from MDCK to Vero cell-based replication and back alter acid stability of HA [43], pH of activation, or membrane fusion [44]. Or, whether they simply counteract steric hindrance [45] caused by Vero-specific changes in HA N-glycosylation, e.g. on residues 28 and/or 40 in the stem region of HA to achieve low-pH conformation required for membrane fusion.

An alignment of the segment 4 sequence of the initial, MDCK cell-derived virus PR/8/34 (RKI) from passage 1 with the viral sequence published in 1981 [46] demonstrated homology with three exceptions. These included a substitution at position 146, where threonine was substituted by asparagine (T146N). In 2001, Schickli et al. [47] detected an asparagine instead of the threonine and lysine at this position (T146N, K147-). Furthermore, they described a second substitution D204E present in this virus strain (passage 1). The third substitution was located at position 158, where lysine was substituted by glutamic acid (K158E).

In summary, the genome sequences of RNA segment 4, coding for HA, are homogenous in the initial virus seed of the RKI-strain (passage 1). This suggests clearly that the virus seed consisted of only one population. The adaptation processes resulted in additional virus variants, due to mutations in the HA stem region. However, no potential HA N-glycosylation sites or amino acids in their close neighborhood were affected. In contrast, the initial virus seed population (passage 1) of the NIBSC-strain of the adaptation series comprises more subpopulations concerning the HA-encoding RNA segment 4 than the RKI virus seed. Previously, these two (theoretical identical) influenza virus seeds have been described to differ significantly in infection characteristics such as interferon (IFN) and apoptosis induction, expression of interferon stimulated genes, final virus yields [48], [49], and the activation of general host cell response [50], [51]: PR/8/34 from NIBSC was characterized to induce higher levels of IFN and Mx expression, to induce apoptosis earlier, and to reach lower final titers than the PR/8/34 from RKI. Seitz et al. hypothesized that two amino acid substitutions in the non-structural protein 1 (NS1) might be related to these differences. Our present findings of a set of slightly different virus variants in the NIBSC seed suggest that the lower average virus yields are due to a broadened quasispecies of the PR/8/34 (NIBSC) virus seed which comprises low-yield-virus variants. The heterogeneity of PR/8/34 (NIBSC) persisted during the whole adaptation processes: from passage 1 to 11, samples comprised multiple virus subpopulations. Thereby, sizes of subpopulations as well as selection and extinction of specific virus variants varied in dependence of adaptation pressure, i.e. the host. Similar findings were reported by Bull et al. for adaptation studies with bacteriophages. They observed the highest level of convergence for bacteriophages grown on the same host and furthermore finally deduced that most substitutions during forward and backward adaptation to different hosts were most probably driven by the adaptation pressure of host-switching [52].

A comparison of results of both adaptation experiments with influenza virus strains from NIBSC and from RKI shows one interesting similarity, the substitution of lysine by glutamic acid. In particular, the K459E variant from NIBSC that carries a deletion in the HA1 chain (K147-) corresponds to the K460E variant from the RKI. Moreover, also the substitutions D455H and N460D of the NIBSC strain are located in the same region of the HA2 subunit as the substitutions S457L and K460E of the RKI-strain.

For both influenza virus strains the amino acid substitutions in HA during virus adaptation are located exclusively in the HA2 chain (figure S4). Rott et al. [53] described the occurrence of virus variants after adaptation to MDCK cells, which exhibited elevated fusion pH, though these carried exclusively substitutions within the HA1 chain. In contrast, our experiments indicate a crucial role of the HA stem region for virus adaptation from MDCK to Vero cells. Here, inter-monomer or inter-subunit contact is mediated, suggesting HA-initiated fusion as a driving factor of adaptation pressure. Both adaptation series show a significant change of the HA N-glycosylation pattern with the change of host cell system. Additionally, both experiments demonstrate that only a small sequence adaptation is required for successful infection and fast growth to high titers in new host cell lines. These findings are in agreement with results of Wagner et al. and Klenk et al. [45], [54] who reported that N-glycans attached to the stem domain of HA efficiently regulate influenza A virus replication. The authors showed that a loss of N-glycans in the stem region results in increased pH-sensitivity of the virus and that these viruses are also temperature sensitive.

Taken together, our data show that a MDCK cell-adapted virus seed that had not been exposed to Vero cells before, hardly produces any virus progeny in Vero cells. Nevertheless, amino acid substitutions within the stem region of HA can rescue the virus population and ensure efficient virus replication in the new host. Interestingly, for both RKI- and NIBSC-derived virus strains the glycan pattern stabilized with the first passage in the new cell line, with all passage to passage SDs values below 8.5%. The faster release of virus particles and to begin with the further increase in HA-titers during passages two to six indicates that other factors are as well involved in the adaptation process, i.e. changes in the sequence of the viral genome resulting in improved fitness of virus subpopulations. In agreement with Schwarzer et al. the HA glycan profile of Vero cell-adapted virus shows a tendency towards smaller glycan structures compared to MDCK cell-derived viruses [31]. The fitness of these adapted virus variants during backward adaptation to MDCK cells varies – some variants as the K460E (RKI) replicate well and infect either cell line while others grow only poorly in one cell line (e.g. S457L (RKI) in MDCK cells). Whether the substitutions that occurred during the adaptation processes alter acid stability [43], pH of activation, or membrane fusion mechanism [44] of HA or simply counteract steric hindrance caused by Vero-specific changes in N-glycosylation on residues 28 and/or 40 in the stem region of HA for achieving a low-pH conformation, remains to be further elucidated. Finally it should be pointed out, that other factors besides the adaptation status to a host cell line contribute to HA N-glycosylation. Besides the host cell itself - as demonstrated by Schwarzer et al. [31] - one may have to include differences in cell culture media, cultivation systems and cultivation conditions, virus seed preparation, cell density and multiplicity of infection used in the virus production phase. Furthermore, little is known concerning the impact of changes in glycan profiles on properties of live and dead vaccines, i.e. safety and immunogenicity. Nevertheless, monitoring HA N-glycosylation patterns during vaccine production processes allows not only to investigate the impact of process modifications on antigen quality, but also offers a sensitive tool to evaluate consequences of unwanted process variations or process failures.