Mathematical and computational models were used to derive numerical values for the replication kinetics of different influenza virus strains in human bronchial epithelial cells. The relative values for each free parameter, particularly p and R0, among the three influenza virus strains studied are highly informative for comparison of strain-specific pathogenicities. The absolute values reported here for these parameters and cell types are considered to be robust since they were derived by an unbiased process, but since the values may be model specific, they await confirmation in other models.
human bronchial epithelial cells in air-liquid interface cultures allowed us to sample virus more frequently and completely than in the in vivo
), and it avoided possible effects of antibody or cell-mediated immune responses, which may provide more variability in an in vivo
). NHBE cell cultures are likely more relevant than cell lines for studying kinetics of influenza virus replication, although no attempt was made here to compare primary to immortalized cells. NHBE cells support the replication of 2009 H1N1 and seasonal strains (21
), but there is disagreement on whether replication of avian strains is supported in well-differentiated NHBE cells (8
). We have found replication of two highly pathogenic H5N1 avian influenza virus strains, HK/483 (Fig. ) and A/Vietnam/1203/2004 (data not shown), in 7 independent experiments performed using both sources of NHBE cells. Documenting of replication may depend on the timing of sampling (Fig. ), enumeration of all secreted viral particles (Fig. ), and temperature of the cell culture (36
). Replication of human seasonal strains has been shown to be consistently greater than that of avian strains in NHBE cells (8
Cell tropism of different influenza virus strains is determined largely by preferential binding of hemagglutinin proteins to specific sialic acid-decorated glycan receptors. NHBE cultures reproduce the receptor-type distributions found in human respiratory mucosal tissues (39
). In addition, except for in vivo
immigrant humoral and cellular innate immune elements, the in vitro
NHBE cells present in-life barriers to viral entry and spread in mucosal tissue. Avian strains preferentially bind the α-2,3-linked sialic acid receptors in the first cycle of infection (26
). In spite of the loss of cilia during influenza virus infection, we identified by immunohistochemistry early nuclear staining of the avian virus N protein in ciliated cells (Fig. ). Human-adapted strains preferentially bind α-2,6-linked sialic acid on nonciliated secretory cells, although receptor preferences may shift during adaptation to different hosts (26
). Human mucin proteins predominantly display the α-2,3-linked sialic acid receptors (11
). This would appear to favor the binding of avian strains to mucus-associated fractions, but no evidence for this preference was found when comparing the avian, seasonal, and 2009 H1N1 strains (Fig. ). Thus, rather than serving exclusively as a binding decoy for H5N1 strains, mucus and mucin proteins may serve other trapping functions beyond specific receptor binding and deserve further investigation.
We used the ODE model to describe aggregate infection kinetics. ODEs are valuable for studying viral dynamics in a well-mixed system, particularly for blood-borne pathogens, such as HIV and hepatitis viruses (4
). However, spatial effects are not represented, yet they are likely important in the experimental system where influenza virus infection spreads though a monolayer of epithelial cells. Thus, a model that accounts for the spread of infection to nearby susceptible cells may better represent influenza dynamics (4
). In addition, resistance to infection has been shown to pass directly from infected cells to adjacent neighbors (32
), further justifying a spatially explicit model. CA modeling corroborated the results of the ODE model, but in the 2009 H1N1 infection, R0
values in the CA model were lower than those in the ODE model, possibly reflecting the constraints on lateral spread of virus in the in vitro
system more effectively. In contrast, the wide orders-of-magnitude variation in the absolute values of e
between the two models is explained by the different way each model treated e
, making comparison between the models not relevant. In the ODE model, e
refers to the inhibitory effect among all cells at once, whereas the CA model considers each cell at a given instant in time.
Oscillations in the experimentally derived virus output from avian virus-infected cells were consistently seen but were statistically significant only for donor 2 (Fig. ). Mild oscillations were also noted in both models of all three viruses studied (Fig. ). This observation suggests overlapping cycles of replication that occur due to the delay in production between the cycles. The CA model predicted an infected cell life span (eclipse plus secreting phase) of 20 to 24 h in almost every simulation, and the eclipse phase, τ1, estimated from the experimental data was approximately 10 h. Absorption of secreted virus into cells infected during the next cycle would produce a temporary decrease in viral levels and apparent oscillatory behavior.
We observed experimentally that 2009 H1N1 produced more virus after 24 h in culture than the other strains, and the models calculated a higher R0
for the 2009 H1N1 strain than for the avian and seasonal strains. Productivity varies among strains of the same hemagglutinin subtype, and some seasonal strains may even approximate the replication efficiency of 2009 H1N1 strains (21
). The amount of secreted virus at any given time is determined by the balance of virus production and virus decay (Fig. ), and variable decay among different strains has not been rigorously tested. Greater levels of secreted virus on the respiratory mucosa may be a key determinant of the contagiousness of the infected host. Productivity of virus secretion on the respiratory mucosa, however, may not be the only determinant of transmission by exhaled aerosol. In controlled exposure chamber experiments in which susceptible ferrets were exposed to exhaled aerosols from infected ferrets, 2009 H1N1 transmission was more efficient than seasonal virus transmission (F. Koster et al., unpublished data). The exhaled aerosols contained the same levels of viral RNA for the two strains, suggesting that transmission was primarily dependent on more-efficient viral replication in the susceptible host.
In the human study (2
), nasal washes were collected only once daily following intranasal inoculation with an unknown MOI, measuring the productivity in nasal turbinate and respiratory airway cells. Interestingly, in spite of differences in the frequency of collection, MOI, and cells infected, our seasonal strain R0
calculations in the ODE model (range, 11 to 22) and CA model (range, 12 to 15) are comparable to the seasonal strain R0
calculation (~22) for human nasal cells (2
). Bronchial epithelial cells from the airway portion of the lower respiratory tract and nasal turbinate cells may support the replication of influenza virus strains equally; 2009 H1N1 virus infection of epithelial cells from ferret bronchi and turbinates displayed identical kinetics (J. Tipper et al., unpublished data).
The infected cell life span was treated differently in the ODE and CA modeling approaches, with different results. In the ODE model, virus-producing cells (I2
) were assumed to die at a constant rate, δ, implying an exponential distribution of cell lifetimes. The rate δ was constrained to correspond to a biologically plausible range for infected cell half-life (ln 2/δ) between 6 h and 24 h. The value for δ in each independent fit resulted in the lower boundary of that range. Because the infected cell death rate (δ) and viral production rate (p
) covary in the model, differences between the abilities of strains to produce virus are best seen by examining the values of p
(Table ). For the CA model, best-fit values of the infected cell life span ranged from a mean of 21.7 h for the avian strain to a mean of 27.6 h for the seasonal virus. Recent modeling of influenza virus kinetics has used a fixed life span at 24 h based on experimental data (35
), and experimental data in the murine model of a mouse-adapted seasonal strain were used to calculate an epithelial cell life span of 29 h (28
The infected cell life span has not been measured experimentally in influenza virus-infected epithelial cells in vitro
to our knowledge. Influenza virus infection induces cell apoptosis (19
), and viral strains vary in their ability to retard apoptosis (44
). Attempts to measure relative levels of apoptosis here were inaccurate due to the loss of infected cells from the surface. We noted thinning of the monolayer during the culture, which was more pronounced with the avian strain than with the H1N1 strains (Fig. ). A large number of variables affect the measurement of infected cell life span, including the monolayer integrity, the interval between apoptosis induction and cell dissolution, and the surface cell replacement rate from basal cell differentiation. Epithelial barrier function is preserved after surface cell loss by a process of apoptotic cell extrusion in which dying cells are squeezed out of the epithelium (37
Both models predicted e
to be smaller in the avian strain than in either the 2009 H1N1 strain or the seasonal strain (see Tables and ). This is consistent with published experimental observations finding greater suppression of interferon-mediated defense by virulent avian strains (3
). The ODE model predicted a higher e
value in the 2009 H1N1 strain than in the seasonal strain, while the CA model predicted that these two strains would have similar e
values. The differences in how the two models handle diffusion and spatial distribution of infected cells may account for these differences, but experimental validation is needed to resolve this discrepancy.
Uninfected cell susceptibility to infection, and thus the number of target cells, is determined by the cell phenotype, differentiation state, and activation of antiviral defense mechanisms. Avian (H5N1) and seasonal (H1N1 and H3N2) influenza virus strains have distinct preferences for cell type infection (27
). We observed that the monolayers used in this study contained 25 to 40% ciliated cells and that of the remaining nonciliated cells, 10 to 20% were goblet cells, distributions observed by other investigators (27
). During injury, lung progenitor cells located below the surface are activated, undergo robust clonal expansion, and replace lost infected surface epithelial cells (15
). However, the rate of differentiation at the surface and subsequent susceptibility to influenza virus infection are unknown for our in vitro
experimental system and have been considered to be a relatively minimal contribution to the target cell population in vivo
). Finally, the susceptibility of cells adjacent to an infected cell is altered in a paracrine fashion by local signaling through gap junctions for interferon-mediated defenses (32
). Our observations are consistent with paracrine defense activation, since infected cells identified by immunohistochemistry early in infection occurred in isolation (Fig. ), and we rarely found adjacent infected cells. Modeling of in vivo
data has also found a major role for a paracrine-distributed innate defense, leaving most respiratory epithelial cells uninfected (35
). In view of the multiple unknowns in our experimental system, our models make the assumption that each monolayer had the same number of susceptible cells and that the phenotype of susceptible cells did not change during the course of infection.
Taken together, our results show that in vitro
infection of human well-differentiated bronchial epithelial cell monolayers can be used to derive relative numerical estimates for replication of three influenza virus strains. Mathematical and computational models can provide reproducible results for R0
. Whether these derived parameters are useful as phenotypic characteristics of each strain will require further replications that account for differences in host cell donor characteristics and study of multiple strains to incorporate strain-specific modulations of the host response (43
). These phenotypic indices can then be compared to the transmissibility and virulence of each strain, since transmissibility to humans cannot be quantified experimentally (1
). The higher R0
value of the 2009 H1N1 strain derived here may be one of several parameters predicting the higher epidemiological R0
value derived from field studies in human populations (10
). Likewise, the low R0
value calculated here for an avian strain may predict the rare human-to-human transmission of H5N1 strains (29
). The caveats applicable to this in vitro
experimental model and the computational models include restriction to one tissue source of epithelial cells, the absence of immigrant innate and acquired immune response operating in the intact host, a simplistic representation of the innate host cellular response, and insufficient information about the actual life span of infected cells. Nonetheless, the ability to study and compute kinetics of virus replication in human epithelial cells provides a step in the direction of deriving accurate phenotypes of the pandemic potential of any influenza virus strain.