Respiratory Syncytial Virus Infections

We developed an age-structured SIRS model to describe the transmission dynamics of RSV (Fig. 2A). The model assumes individuals are born with protective maternal immunity (M), which wanes exponentially (with a mean duration of 3–4 months) [65] (link), leaving the infant susceptible to infection (S_n, where n is the number of previous infections). Following infection with RSV, individuals develop partial immunity, which reduces the rate of subsequent infection and the duration and relative infectiousness of such infections, consistent with epidemiological studies and previous models of RSV transmission [26] (link), [27] (link), [29] (link). We assume a progressive build-up of immunity following one, two, and three or more previous infections (I_n) [37] (link), [66] (link)–[68] (link). Both age and number of previous infections were assumed to influence the risk of developing severe lower respiratory disease (D) possibly requiring hospitalization [37] (link), [66] (link), [69] (link). We parameterized the model based on data from cohort studies conducted in the US and Kenya (Table 2) [37] (link), . Transmission-relevant contact patterns were assumed to be frequency-dependent and were parameterized based on self-reported data on the number and age of conversational partners from one European study [54] (link), [55] (link); no such study has been conducted among a widely representative cohort in the US.
We initially fit our model to the age-stratified hospitalization data from all nine states with complete data from 1989–2009 in order to estimate the mean transmission rate, relative infectiousness of first and second versus subsequent infections, seasonality parameters, and reporting fraction (i.e. proportion of individuals with severe lower respiratory tract disease who are hospitalized, coded as RSV, and reported in our dataset), which are key unknown parameters. We then fixed the relative infectiousness and fit the model to the hospitalization data from each of the nine states plus Florida individually, using the other estimated parameters from the cumulative data fit as our starting conditions to estimate state-specific transmission rates, seasonality parameters, and reporting fractions. For each fit, we seeded the model with one infectious individual in each age group and used a burn-in period of 40 or 41 years, examining the fit using both even- and odd-year burn-in periods to allow for the biennial pattern of epidemics present in some states, and selected the best-fitting model for each state. We also explored longer burn-in periods and examined the model output to ensure that the equilibrium quasi-steady state had been reached.
Seasonality in the instantaneous rate of transmission of RSV was modeled using sinusoidal seasonal forcing with a period of 1 year (52.18 weeks) as follows: , where β₀ is the baseline transmission rate, b is the amplitude of seasonality, and φ is a seasonal offset parameter (a measure of timing of peak transmissibility), and t is the time (in years) [31] (link), [32] (link). We constrained φ to be between −0.5 and 0.5, where φ = 0 represents January 1 and φ = −0.5 and φ = 0.5 both represent July 1. These parameters were estimated by fitting our model to the state-specific data.
We used maximum likelihood to determine the best-fitting models. For each set of parameters, the likelihood of the data given the model was calculated by assuming the number of hospitalizations in each age class (a) during each week (w), x_a,w, was Poisson-distributed with a mean equal to the model-predicted number of severe lower respiratory tract infections due to RSV (D_a,w) times the reporting (or hospitalized) fraction (h), , as has been described previously [27] (link), [36] (link). The log-likelihood (log(L)) of the model was given by the equation:
While this observation model may fail to capture the true variability in the distribution of cases, other observation models (e.g. negative binomial) would require estimating an additional parameter, which we do not feel is justified. We used the “fminsearch” command in MATLAB v7.14 (MathWorks, Natick, MA) to minimize the –log(L), which employs a direct simplex search method.
Next, we fit the model to the laboratory data on RSV-positive tests from 38 states. The laboratory data did not contain detail on the age of cases; therefore, we could not derive reliable estimates of the baseline transmission rate by fitting our model to these data, since estimates of the baseline transmission rate and reporting fraction are inherently confounded. (The age distribution of cases, pattern of epidemics, and mean incidence rate inform estimates of the baseline transmission rate, while the mean incidence also informs estimates of the reporting fraction.) Instead, we estimated the baseline transmission rate for each state from the relationship we observed between population density and R₀ among the ten states with hospitalization data (S3 Fig.). We fixed the R₀ for the District of Columbia at the maximum observed R₀ (for New Jersey). We then estimated the amplitude of seasonality, seasonal offset parameter, and reporting fraction by fitting our model to the rescaled laboratory data. We examined the sensitivity of our results to the method we used to correct for changes in testing and reporting effort for RSV over time by also fitting our model to the raw number of RSV-positive tests reported, instead applying the estimated scaling factor to the model output (i.e. dividing the model output by the scaling factor for each week). The log-likelihoods of the fitted models were similar (S5 Table), and the results were qualitatively the same (S8 Table).
We examined the correlation between the estimated model parameters for each state and the significant climatic variables from the univariate statistical analyses. We calculated the Pearson's correlation coefficient and associated p-value for each state-specific parameter estimate and climatic variable of interest. We also examined the ability of the model to capture the biennial pattern of RSV epidemics present in some states by comparing the strength of the biennial cycle in the observed and predicted RSV time series. The strength of the biennial cycle was calculated as the ratio of the biennial to annual Fourier amplitude [73] , [74] . Finally, we examined whether monthly deviations from average climatic conditions could help explain the difference between observed and predicted monthly RSV activity across states.

Development of the RSV-iiiQ was in accordance with current regulatory guidance and best practices and included a review of the literature and input from experts and patients. A literature review was conducted to determine relevant symptoms and impacts of RSV infection and to identify any existing measures designed to assess the RSV patient experience (see Supplemental Table S1 for the search strategy for the literature review). The first stage included a PubMed database search conducted in December 2017 and restricted to English-language articles that used human participants and were not comments, letters, or editorials and that were published in a 10-year window. Next, desktop research was conducted to identify relevant grey literature, such as conference presentations or government reports focusing on the impact of RSV, including the patient’s ability to perform activities of daily living (ADLs) and psychosocial well-being.

Previously published studies were used to verify the abundance of proteins that make up the N-degron pathway in the non-infected and SARS-CoV-2 infected groups: (i) Saccon et al. [51 (link)] (Calu-3, Caco-2, Huh7, and 293FT cell lines, proteomics); (ii) Nie et al. [52 (link)] (autopsy 7 organs, 19 patients, proteomics); (iii) Leng et al. [53 (link)] (lung tissue, 2 patients, proteomics); (iv) Qiu et al. [54 (link)] (lung tissue, 3 patients, proteomics); (v) Bojkova et al. [55 (link)] (Caco-2 cells, proteomics); (vi) Wu et al. [56 (link)] (lung tissue, colonic transcriptomics); and (vii) Desai et al. [57 (link)] (lung tissue, transcriptomics). To verify modulation of the N-degron pathway in other viral infections, including MERS-CoV/SARS-CoV/H1N1 influenza virus/Respiratory syncytial virus (RSV), data from the following studies were evaluated: (viii) Zhuravlev et al. [58 (link)] (MRC-5, A549, HEK293FT, and WI-38 VA-13 cell lines, H1N1 influenza virus, transcriptomics); (ix) Li et al. [59 (link)] (A549 and 293T cell lines, H1N1 influenza virus, transcriptomics); (x) Krishnamoorthy et al. [60 (link)] (comparative among coronaviruses, transcriptomics); (xi) Ampuero et al. [61 (link)] (time course of RSV infection in the lung, transcriptomics); (xii) Besteman et al. [62 (link)] (RSV infected neutrophils, transcriptomics); and (xiii) Dave et al. [63 (link)] (RSV infected alveolar cell, proteomics). Deep proteome data from non-infected cell lineages were recently made available publicly by Zecha et al. [64 (link)] to model SARS-CoV-2 infection in Vero E6 (kidney epithelial cell, African green monkey), Calu-3 (lung adenocarcinoma), Caco-2 (colorectal adenocarcinoma), and ACE2-A549 (lung carcinoma expressing ACE2 to gain cellular entry). The iBAQ intensities of proteins that make up the N-degron pathway were evaluated without infection to access the basal levels of arginylation-related proteins. Experimentally arginylated proteins were retrieved from the datasets of Seo et al. [65 (link)] and Wong et al. [66 (link)] to access the regulation levels of these proteins during SARS-CoV-2 infection. Single-cell RNA-seq data from nasopharyngeal samples provided by Chua et al. [67 (link)] were reanalyzed to identify cell clusters expressing the ATE1 enzyme.