The Emergence of H7N7 Highly Pathogenic Avian Influenza Virus from Low Pathogenicity Avian Influenza Virus Using an in ovo Embryo Culture Model

Outbreaks of highly pathogenic avian influenza virus (HPAIV) often result in the infection of millions of poultry, causing up to 100% mortality. HPAIV has been shown to emerge from low pathogenicity avian influenza virus (LPAIV) in field outbreaks. Direct evidence for the emergence of H7N7 HPAIV from a LPAIV precursor with a rare di-basic cleavage site (DBCS) was identified in the UK in 2008. The DBCS contained an additional basic amino acid compared to commonly circulating LPAIVs that harbor a single-basic amino acid at the cleavage site (SBCS). Using reverse genetics, outbreak HPAIVs were rescued with a DBCS (H7N7DB), as seen in the LPAIV precursor or an SBCS representative of common H7 LPAIVs (H7N7SB). Passage of H7N7DB in chicken embryo tissues showed spontaneous evolution to a HPAIV. In contrast, deep sequencing of extracts from embryo tissues in which H7N7SB was serially passaged showed retention of the LPAIV genotype. Thus, in chicken embryos, an H7N7 virus containing a DBCS appears naturally unstable, enabling rapid evolution to HPAIV. Evaluation in embryo tissue presents a useful approach to study AIV evolution and allows a laboratory-based dissection of molecular mechanisms behind the emergence of HPAIV.


Introduction
Avian influenza viruses (AIVs) are segmented negative sense RNA viruses belonging to the family Orthomyxoviridae [1]. They are divided into subtypes determined by the external viral glycoproteins, namely haemagglutinin (HA:H1-H16) and neuraminidase (NA: N1-N9), and normally occur as low pathogenicity (LP)AIVs, causing few (if any) clinical signs in wild birds and poultry [2,3]. The H5 and H7 subtypes may mutate to the highly pathogenic (HP)AIV variant typified by severe morbidity and

Viruses
The eight H7N7 viruses used in this study are listed in Table 1. A 12-plasmid RG system was used to rescue recombinant viruses [57], comprising eight rescue plasmids (pPol I) based on A/chicken/England/11406/2008 H7N7 (England-08; GISAID accession # EPI712884-EPI712891 [26]) with four helper plasmids (pPol II) that express the viral polymerase genes and nucleoprotein from A/Victoria/3/75 (H3N2) [58]. Plasmids were synthesised by GeneArt TM Gene Synthesis (ThermoFisher Scientific, Paisley, UK). Modifications to the HA pPol I plasmid were performed using QuickChange II XL Site-Directed Mutagenesis Kit (primers available on request) following the manufacturer's instructions (Agilent Technologies, Stockport, UK). All 12 plasmids were transfected into 293T cells in 12-well plates using FUGENE HD transfection reagent (Promega, Southampton, UK). After overnight incubation, cells were removed and co-cultured with MDCK cells in 25 cm 2 flasks. After 6-8 h incubation the cells were washed and 5 mL fresh serum-free DMEM was added. Where appropriate, TPCK trypsin was added at 0.5 µg/mL (Fisher Scientific, UK). Rescued "synthetic" viruses were passaged once in 9-11-day-old EFEs to increase their titer and also to obtain an avian origin envelope as per the wildtype (wt) viruses. Viruses were full-genome sequenced to confirm the introduced genetic changes.

Wildtype (wt) virus
England-08 CCCGAAATCCCAAAG AGA AAG AAA AGA / GGCCTATTT P E I P K R K K R / G L F Typical European and Asian LP CS motif + The three different CS motifs were rescued on two H7N7 HA backgrounds. H7N7 MB , H7N7 DB and H7N7 SB with the HA gene sequence obtained from clinical material from the outbreak, and H7N7 MB +, H7N7 DB + and H7N7 SB + with the HA gene sequence obtained from the EFE-amplified stock, which included three nucleotide changes. Two of these nucleotide changes were non-synonymous; I77V and A125T (H7 numbering from mature protein), the latter resulting in the acquisition of a potential N-linked glycosylation site at position 123 [26].

Titration According to Plaque-Forming Units (pfu)
Infectious viral titres were determined by plaque assays using 100% confluent MDCK cells. Ten-fold dilution series of viruses were added to wells in duplicate and overlay DMEM media containing 2.5% agar (Oxoid, Basingstoke, UK). Where appropriate, TPCK trypsin was included at 2 µg/mL (Fisher Scientific, UK). After 72 h incubation, cells were stained with 1% crystal violet (Sigma, Poole, UK) and visible plaques counted. Viruses were titrated in triplicate and titres were expressed as pfu/mL.

Multi-Cycle Growth Kinetics
Titrated viruses infected confluent MDCK cells at a multiplicity of infection (MOI) of 0.0001. Where appropriate, TPCK trypsin was added at 0.5 µg/mL. Supernatants were removed from each well at 8, 24, 48, and 72 h post infection (hpi).

M-Gene Reverse Transcription RealTime (RRT)-PCR
M-gene RRT-PCR was carried out using primers and probe [60] as described [59]. Viral RNA was quantified as relative equivalent units (REUs) against a 10-fold dilution series of EID 50 quantified RNA [61].

Intravenous Pathogenicity Index (IVPI) Determination
Standard IVPI tests were carried out [2], with each SPF chicken inoculated intravenously with 0.1 mL allantoic fluid diluted to a specified titer or a haemagglutinating activity titer >2 4 . Welfare considerations and IVPI scoring were applied as described [56], with an IVPI of ≥1.2 considered as an HPAIV phenotype [2].

Virus Isolation and Propagation in EFEs
Nine-to eleven-day-old specific pathogen-free (SPF) EFEs were used for virus isolation, growth and titration of median egg infectious dose (EID 50 /mL) by standard means [2,62].
2.9.2. Allantoic Fluid Passage Experiments of H7N7 DB H7N7 DB RG virus was blind passaged ten times in 14-day-old (14do) EFEs. Thirty eggs were inoculated via the allantoic cavity with 0.1 mL per egg containing 10 5 EID 50 H7N7 DB virus. Following embryo death, or after 72 hpi, 1 mL allantoic fluid was harvested from each egg and pooled. Subsequent passages were carried out in two groups each consisting of 15 14do EFEs inoculated with either 0.1 mL neat allantoic fluid from the previous passage (neat group) or this material diluted 1:5000 in PBS (diluted group). At each passage, egg death was recorded and HA activity of the allantoic fluid was determined [2].

Tissue Tropism in EFEs
Nine-day-old (9do) and 14do EFEs were each inoculated with 0.1 mL of either 10 1 EID 50 or 10 4 EID 50 virus. Embryos were harvested after 37 • C incubation at 24, 48, and 72 hpi. The allantoic fluid and membranes (allantoic and amniotic) were harvested and the membranes fixed in 10% buffered formalin. Each embryo was removed from the egg, humanely killed by disruption of membranes and decapitation (cessation of circulation), embryo head and body tissues were fixed in the same pot of 10% buffered formalin. The method was later refined to include only 14do embryos inoculated with 0.1 mL 10 1 EID 50 and harvested at 72 hpi. Embryos were bisected longitudinally; one half of the head and body was fixed in 10% buffered formalin for IHC as described [61] and the other remained unfixed for subsequent homogenization for further characterisation, which included EFE passage, RNA extraction, DNA Sanger or deep sequencing.

Sanger Sequencing
DNA sequencing using the BigDye v3.1 kit (Applied Biosystems, Warrington, UK) used gene-specific primers (available on request), with analysis on an ABI Prism 3130 Genetic Analyser (Applied Biosystems, Warrington, UK). The Lasergene package version 12 (DNAStar, USA) was used for nucleotide sequence analysis.

Deep Sequencing
Deep sequencing was carried out on the MiSeq Illumina platform directly from extracted viral RNA or PCR-amplified across the CS region (see supplementary materials). DNA library preparation was performed using the Illumina Nextera ® XT Library Prep Kit (Illumina, Cambridge, UK) according to the manufacturer's instructions. Indexed libraries were quantified and pooled in equimolar concentrations (2 nM final concentration) and sequenced in multiplex with 2 × 150 base-paired end reads on the MiSeq Illumina platform. Adaptors and primers were removed and reads with a Phred quality score below 30 were excluded using Trimmomatic [63]. Reads shorter than 120bp were also excluded. Raw sequence reads were mapped to the genome of A/chicken/England/11406/2008 H7N7 (accession numbers EPI712884-EPI712891) England-08 using a BWA-MEM algorithm [64]. The consensus sequence was extracted from the resultant bam file using a modified SAMtools script [65] (vcf2consensus.pl available at: https://github.com/ellisrichardj/csu_scripts/blob/master/ vcf2consensus.pl). The frequency of background variants generated by experimental and technical error were established (see supplementary materials, Tables S1 and S2) and the derived error level, 1.92%, was applied to sequenced samples.

Statistics
Statistical analysis for multi-cycle growth kinetics were performed using GraphPad Prism version 6.04. Individual Log values were transformed using the equation Y=Log(Y). The statistical significance of virus growth was assessed by a two-way ANOVA with Bonferroni post-test correction for multiple comparisons. The binomial test of proportions was used to calculate the sample size required to estimate a prevalence of 0% with a 97.5% one-sided confidence interval of 0.05-0% of infected embryos (n = 72).

Rescue of Recombinant RG Viruses
An RG-copy of wt England-08 containing a MBCS was successfully rescued along with RG viruses containing the DBCS from the LPAIV precursor to the HPAIV outbreak (H7N7 DB ) and a typical SBCS of circulating viruses in Eurasia (H7N7 SB ). Previous isolation of wt England-08 in EFEs resulted in sequence changes in the HA gene compared to the sequence obtained from the original clinical material. Consequently, the three different CS motifs were rescued on both H7N7 HA backgrounds (Table 1). Full genome sequencing of each of the rescued recombinant viruses revealed no other genetic changes in all eight genome segments.

A MBCS Reflects a Trypsin-Independent Phenotype In Vitro
H7N7 SB and H7N7 DB required trypsin to produce plaques ( Figure 1a) and grew to lower titres in MDCK cells than H7N7 MB in the absence of trypsin at 24 hpi (p < 0.001), 48 hpi and 72 hpi (p < 0.0001) (Figure 1b). In contrast, monolayers infected with H7N7 MB produced plaques without trypsin ( Figure 1a) with growth independent of trypsin ( Figure 1b). H7N7 MB achieved a slightly higher, although not statistically significant, titre in the absence of trypsin.

A MBCS in the HA Protein Is Required for a HP Phenotype In Vivo
Five groups of chickens were inoculated as per the IVPI test. IVPI chickens were previously infected with England-08 [26] and are included to compare pathogenicity of the various H7N7 viruses ( Figure 2). Following infection with England-08 and H7N7 MB+ , 100% mortality occurred, although the former all died very rapidly by 1 dpi. While the H7N7 MB attained 50% mortality, all three registered IVPI scores of >1.2 as evidence of being HPAIVs (Figure 2). Following the observation that H7N7 MB + gave a higher IVPI than H7N7 MB , the H7N7 DB + and H7N7 SB + rescued viruses were then selected for IVPI testing. Neither virus showed clinical signs throughout the duration, with both registering IVPI values of 0.00, indicating an LPAIV phenotype (Figure 2).

A MBCS in the HA Protein is Required for a HP Phenotype in Vivo
Five groups of chickens were inoculated as per the IVPI test. IVPI chickens were previously infected with England-08 [26] and are included to compare pathogenicity of the various H7N7 viruses ( Figure 2). Following infection with England-08 and H7N7MB+, 100% mortality occurred, although the former all died very rapidly by 1 dpi. While the H7N7MB attained 50% mortality, all three registered IVPI scores of >1.2 as evidence of being HPAIVs (Figure 2). Following the observation that H7N7MB+ gave a higher IVPI than H7N7MB, the H7N7DB+ and H7N7SB+ rescued viruses were then selected for IVPI testing. Neither virus showed clinical signs throughout the duration, with both registering IVPI values of 0.00, indicating an LPAIV phenotype ( Figure 2).

H7N7DB Passage in Allantoic Fluid
The effect of in ovo passage of H7N7DB was initially assessed by sequential passage of allantoic fluid in 14do EFEs. No substantial change in embryo death time or allantoic fluid HA titres (all >1:64) across 10 passages were seen. Sanger sequencing of passage 10 material did not reveal any genetic changes at the consensus level, with the HA sequence being identical to the initial inoculum. Four non-synonymous changes were observed outside of the CS region (S31I, F92V, N123D, G196E; H7 numbering of the mature protein) but with no evidence of any MBCS emergence via in ovo passage.

Comparative Pathotyping of H7N7 Wt and Recombinant RG Viruses in Chicken Embryo Tissues
Nine-day-old and 14do EFEs were inoculated with two titres, 10 1 EID50 and 10 4 EID50, of the HPAIVs wt England-08 and RG H7N7MB as well as the LPAIV wt Sweden-08 virus. Three embryos per titre were harvested after 37 °C incubation at 24, 48 and 72 hpi. Virus-specific IHC revealed differences between the HPAIVs and LPAIV which reflected pathotype distinction. In addition to viral tropism in the allantoic and amniotic membranes, both HPAIVs displayed systemic multi-organ distribution, including parenchymal and endothelial cells which defined an HP phenotype ( Figure  3A,B). Widespread endothelial staining was prominent in lung, heart, spleen, liver and kidney tissues. The viral distribution in embryos for both the HPAIVs were comparable and no marked differences in antigen distribution were seen between the 9do and 14do EFEs infected with either England-08 (hexagons), H7N7 MB (squares), H7N7 MB + (diamonds), H7N7 DB + (circles) and H7N7 SB + (triangles). Chickens infected with viruses containing an MBCS (England-08 [26], H7N7 MB and H7N7 MB +) resulted in varying levels of mortality. Chickens infected with di-basic-(H7N7 DB +) or single-basic-(H7N7 SB +) containing viruses did not display any clinical signs and survived until 10 dpi when the experiment was terminated.

H7N7 DB Passage in Allantoic Fluid
The effect of in ovo passage of H7N7 DB was initially assessed by sequential passage of allantoic fluid in 14do EFEs. No substantial change in embryo death time or allantoic fluid HA titres (all >1:64) across 10 passages were seen. Sanger sequencing of passage 10 material did not reveal any genetic changes at the consensus level, with the HA sequence being identical to the initial inoculum. Four non-synonymous changes were observed outside of the CS region (S31I, F92V, N123D, G196E; H7 numbering of the mature protein) but with no evidence of any MBCS emergence via in ovo passage.

Comparative Pathotyping of H7N7 Wt and Recombinant RG Viruses in Chicken Embryo Tissues
Nine-day-old and 14do EFEs were inoculated with two titres, 10 1 EID 50 and 10 4 EID 50 , of the HPAIVs wt England-08 and RG H7N7 MB as well as the LPAIV wt Sweden-08 virus. Three embryos per titre were harvested after 37 • C incubation at 24, 48 and 72 hpi. Virus-specific IHC revealed differences between the HPAIVs and LPAIV which reflected pathotype distinction. In addition to viral tropism in the allantoic and amniotic membranes, both HPAIVs displayed systemic multi-organ distribution, including parenchymal and endothelial cells which defined an HP phenotype ( Figure 3A,B). Widespread endothelial staining was prominent in lung, heart, spleen, liver and kidney tissues. The viral distribution in embryos for both the HPAIVs were comparable and no marked differences in antigen Viruses 2020, 12, 920 9 of 24 distribution were seen between the 9do and 14do EFEs infected with either dose. In contrast to the HPAIV infections, embryos infected with Sweden-08 showed dominant replication in the allantoic and amniotic epithelia from 24 hpi, independent of dose or age (shown as duplicate embryo findings; Figure 3C,D). These observations are consistent with typical LPAIV in ovo infection and are defined as an LP phenotype. Occasional dissemination of Sweden-08 was observed in some embryonic organs comparable to HPAIV-infected embryos e.g., in lung tissue ( Figure 3C). Absence of endothelial tropism was another overall outcome of LPAIV infection, although staining of discrete foci of hepatocytes and endothelial cells in the liver of one embryo was detected ( Figure 3D), indicative of a mixed LP/HP phenotype. The above IHC results obtained from in ovo infection with the two wt viruses (one LPAIV and one HPAIV) and the one recombinant HPAIV (Figure 3), in young and older aged embryos at two different doses, enabled clear distinction of three phenotypes: LP, HP and a mixed LP/HP phenotype.

Emergence of HPAIV after H7N7 DB Passage in Chicken Embryo Tissues
Following the identification of the mixed LP/HP phenotype in embryo tissues infected with Sweden-08, the role of the DBCS in the emergence of HPAIV was investigated. Two different titers of H7N7 DB (10 1 EID 50 or 10 4 EID 50 ) were inoculated in 9do and 14do EFEs and harvested at 37 • C for 24, 48 and 72 hpi (n = 36 total). The majority of infected embryos showed characteristic LP phenotypes by IHC ( Figure 4A). At 72 hpi, two embryos displayed a mixed LP/HP phenotype: one inoculated at 9do with 10 4 EID 50 (H7N7 DB 9do) and another inoculated at 14do with 10 1 EID 50 (H7N7 DB 14do). Positive cells were stained in the membranes of the mesenchyme and blood vessels with widespread labelling in blood vessel endothelium. The H7N7 DB 9do embryo had widespread multifocal labelling of endothelial cells and labelling in tubular epithelial cells. Moderately positive amnions in addition to positive sloughed cells in the parabronchi, hepatocytes and numerous splenic cells were observed. Positive labelling was also found in the brain, heart, kidney, intestine and bursa tissues ( Figure 4B). The H7N7 DB 14do embryo also showed positive labelling in skeletal muscle cells, brain tissue, bursal follicles, spleen and hepatocytes in addition to amniotic cells ( Figure 4C). To attempt reproducibility of the above mixed LP/HP phenotype findings, a repeat inoculation of nineteen 14do EFEs was done with 10 1 EID 50 of H7N7 DB and harvested at 72 hpi. Six of nineteen (32%) embryos displayed a mixed LP/HP phenotype by IHC.
RNA extracted from FFPE tissue or tissue homogenates from both experiments was Sanger-sequenced across the CS. Sequence was obtained from one 9do embryo and five 14do embryos that showed a mixed LP/HP phenotype at 72hpi and demonstrated that the H7N7 DB virus had acquired additional basic amino acids to create MBCS sequence motifs. Two different MBCS were sequenced: PEIPKKKKR/GLF and PEIPKRKKR/GLF, both of which occurred naturally during the 2008 H7N7 UK outbreak [26]. RNA was extracted from total tissue homogenates for deep sequencing from amplicons. Concordant with Sanger sequencing, tissue homogenates from two embryos with a mixed LP/HP phenotype (H7N7 DB E3 and E10) contained a majority MBCS population at 53.10% and 69.24% respectively by deep sequencing ( Figure 5 and Table S3). However, deep sequencing revealed that a high proportion of DBCS also remained in the tissues at 43.50% and 29.59%, respectively ( Figure 5 and Table S3).
Interestingly, for another embryo (H7N7 DB E14), only the FFPE block from the body section displayed a mixed LP/HP phenotype ( Figure 4D) and MBCS by Sanger sequencing. The head and membrane tissues had very few positively stained endothelial cells ( Figure 4D) compared to the same sections in other embryos ( Figure 4B,C). DNA Sanger sequencing revealed retention of the DBCS. Deep sequencing showed that the head and membrane FFPE tissues contained a majority DBCS population as expected (72.77%) but also enabled detection of MBCS variants, albeit at a lower proportion of the total variant (26.52% MBCS population; Figure 6 and Table S3). The body section (visceral organs) contained a majority MBCS population at 98.94%, which was in accord with Sanger sequencing, with only 0.02% DBCS detected (i.e., below the error threshold of 1.92%; Figure 6 and Table S3).    UK outbreak [26]. RNA was extracted from total tissue homogenates for deep sequencing from amplicons. Concordant with Sanger sequencing, tissue homogenates from two embryos with a mixed LP/HP phenotype (H7N7DB E3 and E10) contained a majority MBCS population at 53.10% and 69.24% respectively by deep sequencing ( Figure 5 and Table S3). However, deep sequencing revealed that a high proportion of DBCS also remained in the tissues at 43.50% and 29.59%, respectively ( Figure 5 and Table S3). Interestingly, for another embryo (H7N7DB E14), only the FFPE block from the body section displayed a mixed LP/HP phenotype ( Figure 4D) and MBCS by Sanger sequencing. The head and membrane tissues had very few positively stained endothelial cells ( Figure 4D) compared to the same Deep sequencing showed that the head and membrane FFPE tissues contained a majority DBCS population as expected (72.77%) but also enabled detection of MBCS variants, albeit at a lower proportion of the total variant (26.52% MBCS population; Figure 6 and Table S3). The body section (visceral organs) contained a majority MBCS population at 98.94%, which was in accord with Sanger sequencing, with only 0.02% DBCS detected (i.e., below the error threshold of 1.92%; Figure 6 and Table S3). Despite the mixed LP/HP phenotype and emergence of MBCS in the embryo tissues, the corresponding allantoic fluids collected from six of the nineteen (32%) embryos inoculated with H7N7DB in the repeat experiment still retained the DBCS and showed no evidence of mutation to MBCS by Sanger sequencing. Deep sequencing of the allantoic fluid for H7N7DB E3 and H7N7DB E10 also showed retention of the DBCS (Table S3).  Despite the mixed LP/HP phenotype and emergence of MBCS in the embryo tissues, the corresponding allantoic fluids collected from six of the nineteen (32%) embryos inoculated with H7N7 DB in the repeat experiment still retained the DBCS and showed no evidence of mutation to MBCS by Sanger sequencing. Deep sequencing of the allantoic fluid for H7N7 DB E3 and H7N7 DB E10 also showed retention of the DBCS (Table S3).

Confirming the Phenotype of the HP Progeny Viruses by IVPI
Virus propagation of H7N7 DB E3 and H7N7 DB E10 tissue homogenates that contained a majority population of MBCS was attempted in EFEs, however the harvested allantoic fluids contained DBCSs by Sanger sequencing. A further attempt to isolate a MBCS containing virus from the tissue homogenate was performed by enriching for MBCS populations in CEFs without trypsin prior to inoculation of EFEs. Sanger sequencing revealed the presence of an MBCS from the CEF supernatants inoculated with both tissue homogenates. Upon one further passage in EFEs, the allantoic fluid from H7N7 DB E10 successfully produced the MBCS: PEIPKKKKR/GLF, but allantoic fluid from H7N7 DB E3 retained the DBCS. By deep sequencing, the population of MBCS in the cell supernatant of H7N7 DB E3 slightly increased from 53.10% to 59.51%, but when the cell supernatant was inoculated into EFEs and the allantoic fluid harvested, the population of MBCS decreased to 32.28% ( Figure 5 and Table S3). The DBCS population appeared to have been preferentially selected and amplified to constitute the majority population at 63.99%. In contrast, the population of MBCS in the H7N7 DB E10 sample greatly increased from 69.24% in the tissue homogenate to 97.64% in the CEF supernatant. When inoculated into EFEs this only slightly decreased to 94.59% ( Figure 5 and Table S3).
The mixed genotype progeny virus from H7N7 DB E10, named H7N7 DB-MB hereafter, was tested by IVPI. The final IVPI result was 1.31 and confirms that H7N7 DB-MB is a HPAIV in chickens with a greater IVPI than the H7N7 MB (IVPI 1.26) (Figure 2). Mortality of chickens infected with H7N7 DB-MB was slower than H7N7 MB , however, 60% (n = 6/10) of chickens infected with H7N7 DB-MB died compared to 50% (n = 5/10) infected with H7N7 MB .

Generation of HPAIV from H7N7 SB Was Unsuccessful after Multiple Passages in Chicken Embryos
Seventy-two 14do EFEs were inoculated with 10 1 EID 50 H7N7 SB , harvested at 72 hpi, and were classified as an LP phenotype ( Figure 7A,B). However, 18 (25%) showed enhanced labelling in the amniotic epithelium, skin, lung, liver, intestine and bursa epithelium ( Figure 7B). Four of the 18 embryos (22%) that showed the most prominent staining in these organs showed retention of the SBCS by both Sanger and deep sequencing ( Figure 8 and Table S4A-D). The enhanced level of replication in systemic tissues warranted further investigation by an additional embryo passage. Reserved frozen embryo tissue from one of these embryos (P1: E14) was re-passaged into fourteen 14do EFEs. Two of the 14 embryos (14%) showed a mixed LP/HP phenotype by IHC ( Figure 7C), essentially as defined previously. RNA was extracted from both the fresh tissue homogenate embryo half and the FFPE embryo tissue half. Sanger and deep sequencing from both embryo halves showed retention of the SBCS (P2: E14, E1 and P2: E14, E11) ( Figure 8 and Table S4E-I).
Virus-specific staining in the brain without evidence of a CS mutation prompted one further passage (P3) to determine whether this genotype could be selected. Tissue homogenates from both embryos that showed a mixed LP/HP phenotype at passage two were passaged in a further fifteen EFEs. At P3, only one of 15 embryos (~7%) (P3: E14, E11, E9) showed a mixed LP/HP phenotype ( Figure 7D). Epithelial and endothelial staining was observed in the amniotic membrane and the skin. A low number of positively stained neurons in the brain, spleen, liver and kidney were observed, but an SBCS genotype was detected by Sanger and deep sequencing with no evidence of mutation to an MBCS (Figure 8 and Table S4J-M). Full genome consensus sequencing was performed for three SBCS passaged tissue homogenates that demonstrated a mixed LP/HP phenotype by IHC (P2: E14, E1; P2: E14, E11; P3: E14, E11, E9). Non-synonymous mutations resulted in five and eight amino acid changes at P2 and P3 respectively, but the SBCS remained unchanged (Table 2).

Discussion
An H7N7 HPAIV RG-system was developed to successfully introduce mutations at the CS and in the HA gene to attempt to study HPAIV emergence. H7N7 viruses containing an MBCS replicated in vitro in the absence of trypsin, but DBCS-and SBCS-containing viruses required trypsin. There is very limited prior literature describing H7 viruses possessing a DBCS, and whether an extra basic amino acid at the CS contributes to enhanced HA cleavage resulting in increased pathogenicity for poultry. In vitro growth curve assays, measured using molecular tools, demonstrated low-level growth of H7N7 DB and H7N7 SB without trypsin (Figure 1b). Accumulation of progeny viral RNA over time suggested higher viral loads for H7N7 DB than for H7N7 SB in vitro in the absence of trypsin, although this was not statistically significant. Replication of LPAIVs in the absence of trypsin has been observed previously [66,67], with speculation that egg-cultured LPAIV can undergo subsequent replication in vitro in trypsin-free medium due to carry-over of proteases from the egg fluid [68]. Both H7N7 SB and H7N7 DB displayed an LPAIV phenotype and remained trypsin-dependent for plaque formation and optimal growth. Importantly, both viruses were not pathogenic for chickens (IVPI 0.00).
The successful spontaneous mutation of the DBCS to a MBCS via in ovo passage in this study suggested that this DBCS may represent an intermediate that is already "primed" to mutate further, allowing for basic amino acid residue accumulation. The rare detection of wt viruses containing the DBCS previously described only in waterfowl on two occasions [54,55] suggests that it is unstable or unfavorable for maintenance in galliforme poultry. Mutation to an MBCS from a DBCS occurs rapidly following transmission to chickens and is readily selected as the dominant viral population. Further work to evaluate mutation from LPAIV to HPAIV in different avian species is merited to explore the hypothesis that DBCS emergence may be restricted to waterfowl as opposed to gallinaceous poultry. However, whilst the DBCS may represent an intermediate, passaging of the DBCS virus resulted in an increase in the minority population of SBCS in one embryo (1.25% to 1.41%; Figure 5A) and whether this is a true result or an increase in sequencing error remains to be determined.
Examining the nucleotide composition encoding the two MBCS motifs, it is proposed that PEIPKKR/GLF mutated to PEIPKKKKR/GLF first by a duplication insertion event and then mutated to PEIPKRKKR/GLF by base substitution, although mutation to PEIPKRKKR/GLF directly from PEIPKKR/GLF cannot be ruled out. Further investigation of the molecular mechanisms involved in the stepwise mutation of the CS is required. The minimum number of basic amino acids required to result in an HPAIV phenotype have been investigated [42,69,70]. However, identifying the optimal nucleotide composition that allows for emergence of an HPAIV phenotype could improve our understanding on how basic amino acids accumulate at the CS in nature. Evidence from the H7N7 UK outbreak in 2008 was also unclear on which motif occurred first, although PEIPKKKKR/GLF was detected more often (n = 35) than PEIPKRKKR/GLF (n = 10) by consensus sequencing [26]. This observation was comparable to the in ovo passage where mutation to PEIPKKKKR/GLF was more frequent (n = 5) than PEIPKRKKR/GLF (n = 1). However, the selection and mutational pressures that occurred at the outbreak (29-week-old layers) are different to those in an experimental in ovo passage with RG viruses. Whilst it is accepted that live birds, rather than embryos, are key in the selection pathway from LPAIV to HPAIV, the in ovo embryo culture approach served as an alternative and reduced in vivo experimentation.
Interestingly, embryo age did not appear to influence the ability of H7N7 DB to mutate as this was seen in both 9do and 14do EFEs. This contrasted with earlier studies where an H5N2 virus acquired HPAIV status after passage in 14do but not in 8-10-day-old EFEs [45,71]. An increase in in ovo pathogenicity was observed when H7N7 DB was passaged in the EFE model, however it is imperative to determine whether this can also occur in hatched birds. Passage of H7N7 DB in chickens would be an essential experiment to validate this in ovo model as a substitute for in vivo experiments such as IVPI. This could contribute to the principles of replacement, reduction, and refinement of animals in scientific research [72] and provide supporting evidence to refine international approaches for defining highly pathogenic status [2].
Although the DBCS successfully mutated to a MBCS in the embryo tissue, it remains unclear why no virus containing an MBCS was detected in the allantoic fluid. Feldmann et al. (2000) described receptor expression and polar budding as factors that can restrict the spread of HPAIV from endothelial cells in embryo tissues. There is also the possibility that epithelial cells in the allantoic and amniotic cavities have already been infected with LPAIV prior to the emergence of HPAIV such that the number of susceptible cells is reduced (due to cell death), thereby restricting HPAIV spread. This may explain why no mutation to MBCS was observed when H7N7 DB was passaged ten times via the allantoic cavity. This observation may reflect a unique tropism of the virus or it may indicate that ten sequential passages in EFEs are insufficient for mutated virus release into the allantoic fluid. Interestingly, previous studies have shown that HPAIV emergence can occur in allantoic fluid after few EFE passages [45]. Since only the allantoic fluid-derived virus was sequenced, it is plausible that MBCS mutants remained undetected within the embryo tissues. Possible MBCS emergence during passage in allantoic fluid cannot be excluded as the pooling of allantoic fluid at each passage in the initial study may have inadvertently diluted any MBCS mutants, preserving DBCS dominance within the viral population. Consensus sequence data from allantoic fluid at the 10th passage demonstrated 100% conservation with input virus across the HA CS. This was further supported through sequencing of the allantoic fluid of infected embryos that had a mixed LP/HP phenotype. Indeed, it was not until the MBCS virus population was enriched by selection in vitro without trypsin that the MBCS virus was propagated to recover an H7N7 DB-MB virus. It has been speculated that cellular protease availability may change during embryo development [73], and that LPAIVs may be degraded in the allantoic fluid of older embryos (14do) [45], thereby allowing HPAIV minority variants to dominate the population.
Influenza A viruses exist as a collection of closely related viral quasispecies, with reports that minority HP viral variants can be maintained in a predominantly LPAIV population [71,74]. A significant impeding impact on H7 HPAIV replication when co-infected with H7 LPAIV has been described in ovo and in vivo [75], suggesting that there may be a threshold at which a particular population is preferentially selected for (or outgrows) the other. However, results from the multi-cycle growth curve demonstrated that neither LPAIVs (H7N7 SB ; H7N7 DB ) nor HPAIV (H7N7 MB ) demonstrated an inherent replicative advantage in vitro when measuring nucleic acid over time (Figure 2). The emergence of an HPAIV from an LPAIV can be attributed to spontaneous CS mutation or the selection of minority MBCS variants in the viral population [44][45][46][47][48]75]. Indeed, both events are likely to occur in nature where the emergence of HPAIV from LPAIV precursors have been reported. The successful spontaneous DBCS-to-MBCS mutation demonstrated that the in ovo passage model can support the emergence of an MBCS and mimics the selection of an HPAIV as seen in the field.
Mutation to an MBCS was not achieved when H7N7 SB was passaged in EFEs. One of the proposed molecular mechanisms for the insertion of multiple basic amino acids is polymerase slippage at purine bases during viral replication to introduce additional purines that translate into basic amino acids [37]. This mechanism may have been involved in the mutation of DBCS to MBCS and is theoretically plausible with the SBCS, although there was no evidence of naturally occurring SBCS-containing viruses during the H7N7 outbreak in 2008 [26]. Furthermore, with the RG viruses, H7N7 SB did not change to DBCS or MBCS where >98% of the population retained the same SBCS ( Figure 8 and Table S4). Therefore, the recombinant H7N7 SB may require additional consecutive purine bases to enable polymerase slippage or further adaptation/compensatory mutations to mutate further. This observation is comparable to investigations of evolution of H5 HPAIV [42], where it was speculated that viruses containing short CS motifs (≤2 basic residues) are more stable and require intense selection pressure for a mutational change at the CS compared to viruses that contain mid-length CS motifs (3-4 basic residues). Indeed, residues other than the MBCS may also be attributed to an increase in pathogenicity. Recombinant H2, H4, H8 and H14 viruses with engineered MBCS on a H5N1 HPAIV background could support an HPAIV phenotype [76] in addition to MBCS engineered into H6 and H9 subtypes [67,77].
Full-genome sequencing of the tissue homogenates from embryos infected with H7N7 SB that showed a mixed LP/HP phenotype also had multiple substitutions that could be contributing to this phenotype in the PA, HA, NA and M proteins ( Table 2). The function of these identified substitutions remains unknown for an avian host. The accumulation of mutations seen from P2 to P3 may suggest that H7N7 SB is unstable and is selecting mutations for increased fitness in ovo. One confounder identified with this embryo model was the unexplained dissemination of LPAIV in the brain cells, particularly in endothelial cells and neurons. Literature suggests that an MBCS is required for intracellular cleavage and replication where trypsin-like serine proteases are absent [10,78]. However, replication in endothelial cells was observed in brain tissue harvested from chickens from an H4N6 LPAIV outbreak [79]. A plausible hypothesis could be that initial infection with LPAIV altered the permeability of the allantoic and amniotic membrane allowing, the virus to replicate in the epithelial layer and disseminated to embryo structures bathed by the amniotic fluid such as the skin, the respiratory and digestive epithelia. The allantoic membrane epithelium extends into the embryonic coelom connected to the allantoic fluid [80], which may explain how LPAIVs which are unable to replicate in endothelial cells are transported through by circulating cells to other sites. The mixed LP/HP phenotypes observed during H7N7 DB or H7N7 SB infection without an MBCS genotype may be explained as spillage from replication sites via macrophages that have disseminated into other tissues, giving the impression of an artefactual extended antigen dissemination. It is unclear whether these LPAIVs are replicating or if they are circulating in an un-cleaved, non-infectious and immature form after one round of replication. Determining the exact mechanism by which LPAIVs can enter systemic organs remains to be investigated. Further work, using relevant ex vivo assays such as chicken tracheal epithelial cells [81] or chicken tracheal organ cultures [82], is warranted.
In contrast to the H7N7 DB and H7N7 SB viruses, the H7N7 MB virus was virulent in vivo (IVPI: >1.2), proving that this was an HPAIV according to standard international definitions [2]. Interestingly, different IVPI scores were obtained for H7N7 MB and H7N7 MB +. It appears that the substitutions in the HA protein outside of the CS, I77V and A123T, the latter introducing a potential glycosylation, was sufficient to substantially increase the IVPI from 1.26 to 2.70. The A123T substitution has been previously characterized as a poultry-adaptation marker [83] and may be principally responsible for this virulence increase as described in earlier observations [83,84], although the underlying pathogenicity mechanism remains unclear. However, the H7N7 viruses assessed in this study are unique and do not contain a glycosylation site at N149, nor do they have a stalk deletion in the NA protein, which have been associated with increased poultry replication and pathogenicity [85]. Despite H7N7 DB + and H7N7 SB + containing the additional A123T adaptive marker for chickens, it was insufficient to cause clinical disease in IVPI-inoculated chickens. It appears that while this marker enhances pathogenicity when coupled with an MBCS, it is not the sole virulence determinant. Banks and Plowright (2003) analysed 85 H7 and 62 H5 protein sequences for the presence of potential glycosylation sites, but found no direct correlation between glycosylation near the receptor binding site (RBS) and increased pathogenicity. However, after intracerebral passage in one-day-old chickens, additional glycosylation sites emerged and correlated with increased pathogenicity [86]. This observation is analogous to the H7N7 2008 UK outbreak viruses where glycosylation at N123 was not naturally selected in hens, but was selected from a minor variant in a HPAIV sample after EFE passage [26]. This finding indicated that an additional glycosylation site near the RBS is not a prerequisite for HPAIV evolution from LPAIV precursors, but suggested that they may predispose AIVs to increased virulence.
Chickens which were IVPI-inoculated with the wt England-08 isolate derived from the outbreak all died by 1 dpi and registered the maximum IVPI of 3.00. By comparison, the genetically identical RG virus H7N7 MB + caused a more gradual time to death and gave an IVPI of 2.7. These results suggest that the RG virus was not as fit or virulent as wt England-08, although this difference is unlikely to be statistically significant. This has been noted previously with other RG viruses [87]. There is an indication that there is a requirement for quasi-species, a cloud of genetic variants, in the viral population to enhance virulence as described for other viruses [88][89][90].
While genetic pressures are believed to play a key role in the mutation from LPAIV to HPAIV, the immunological status of the host and external pressures such as the environment must also be considered [91,92]. Whether an intermediate between an LPAIV and an HPAIV is an obligatory requirement for stepwise mutation remains unknown. Whether this may begin to occur in wild birds or only in poultry is yet to be fully investigated. Surveillance of AIVs in wild birds and poultry combined with genetic and epidemiological data may enhance our ability to identify sources of outbreaks and LPAIV progenitors [3]. Understanding the genetic markers that facilitate evolution of LPAIV into HPAIV can increase our knowledge of AIVs with pathogenic potential and preparedness for HPAIV. Influenza A viruses carrying an intermediate DBCS could signal increased risk for mutation to HPAIV for H5 and H7 subtypes, certainly if maintained in a poultry population.
Identifying the tissue and cell types that are targeted during infection with a particular AIV strain could provide additional information, such as subtle signs of increased pathogenicity of an LPAIV leading to the emergence of HPAIV populations. The in ovo embryo tissue culture approach evaluated in this study may enable a rapid assessment of novel isolates and can also provide predictive value to inform disease control strategies for different strains of LPAIVs, since current approaches often apply similar statutory measures for the prevention and control of LPAIV and HPAIV [93]. The rigour of such controls reflects the perceived risk these viruses pose for mutation to HPAIV. There are economic and welfare impacts from the imposition of such measures for LPAIV infections in poultry that could be better informed if it were possible to understand and predict the mutation events with more certainty, since relatively few viruses appear to carry a high risk if they are not aggressively controlled.