Respiratory syncytial virus (RSV), a negative strand ribonucleic acid (RNA) virus in the pneumoviridae family, is the most important cause of respiratory tract infection during infancy, causing annual winter outbreaks lasting 18 to 24 weeks in the United States [1-5]. In the United States, 50 to 70% of the 4 million newborns each year are infected during their first winter, and 1 to 3% are hospitalized, 4 to 7% are seen in emergency departments, and 10 to 16% require physician office visits because of RSV [6]. Although mortality is low in the United States (approximately 50 deaths annually), in developing countries RSV is estimated to cause as many as 118,000 deaths, 6 million cases of severe acute lower respiratory illness, and 3 million hospitalizations annually in children younger than 5 years [7,8]. In addition, severe RSV infection has been linked to development of asthma and implicated recently in development of chronic obstructive lung disease [9-12].

Major risk factors for severe illness include prematurity, cyanotic congenital heart disease, severe neuromuscular disease, immune compromise, and bronchopulmonary dysplasia [13,14]. However, approximately 70% of hospitalized infants have no overt risk factors for severe illness, although young age at infection, environmental factors (secondhand tobacco smoke, lack of breast feeding, household crowding, and low socioeconomic status), and viral, host genetic, and immune factors have been associated with severe disease [14,15]. In addition, high RSV viral load has also been associated with more severe disease in several but not all studies [16], and although group A RSV strains are more common than group B strains among hospitalized infants, the relationship of viral genetic differences to disease severity has not been demonstrated conclusively [17-22].

RSV is not considered highly cytopathic in airway epithelium, whereas host immune responses are thought to be a major contributor to disease pathogenesis [23,24]. Single nucleotide polymorphisms in a number of cytokine and chemokine genes (IL-1, IL-4, IL-8, IL-13, IL-18, RANTES, CCR5), Toll-like receptor 4, and vitamin D receptor have been associated with increased risk of severe disease [25-28]. In contrast, high levels of cord blood neutralizing antibody are associated with delayed onset of hospitalization with RSV and diminished illness severity [29-31]. High levels of maternally derived RSV-specific antibody at infection have also been associated with diminished illness severity in 2 recent reports [31,32]. The most compelling evidence of the beneficial effect of antibody comes from studies of prophylactic administration of immunoglobulin with high titers of RSV neutralizing antibodies or monoclonal antibody to high-risk infants that demonstrate approximately 50% reductions in hospitalization from RSV [33,34]. Some studies suggest that a type 2 helper (Th2)–biased response during primary infection may also be a contributing factor to disease severity [35-37]. Finally, more severe RSV disease has been associated with greater abundance of haemophilus influenzae and streptococcus pneumoniae in the nasal microbiota during RSV infection [38-40].

The Assessing Predictors of Infant RSV Effects and Severity study was designed to simultaneously measure a number of host demographic, environmental, and innate and adaptive immune parameters in conjunction with viral factors and the respiratory microbiota in relation to disease severity in full-term healthy infants younger than 10 months undergoing primary RSV infection. We hypothesize that assessment of the interplay of these factors will provide insight into the pathophysiology of RSV disease in this population.

The study was performed in Rochester, New York, encompassing 3 winter RSV seasons spanning from 2012 to 2016. To capture the full spectrum of RSV severity from very mild outpatients to hospitalized infants, 3 cohorts of infants were recruited. Inclusion and exclusion criteria are shown in Textboxes 1 and 2. The investigational review boards of the University of Rochester, Highland Hospital, and Rochester General Hospital (RGH) approved the study.

Birth cohort: Infants were enrolled at birth in the late summer through midwinter (approximately August 15 to February 1). Infants were enrolled at 3 hospitals: the University of Rochester Medical Center’s (URMC’s) Strong Memorial Hospital and Highland Hospital, and RGH.

Supplemental cohort: A second cohort, designated the supplemental cohort, was recruited from infants seen with acute respiratory illness not requiring hospitalization at URMC’s Golisano Children’s Hospital and RGH emergency departments or at pediatric clinics at URMC’s Golisano Children’s Hospital and RGH, and the Elmwood Pediatric Practice, a private pediatric office affiliated with URMC.

Hospital cohort: Infants were enrolled on admission to URMC’s Golisano Children’s Hospital and RGH with documented RSV infection.

Illness surveillance and surveillance visits: Infants enrolled in the birth cohort were followed by passive and active surveillance for development of respiratory symptoms throughout their first winter (mid-November until April 15). Parents were asked to call when their infant developed any of the following symptoms: nasal congestion, nasal discharge, cough, wheezing, sustained rapid breathing, or fever. A study nurse evaluated the infant in the research clinic or at a home visit, and a nasal swab was collected for RSV diagnosis. Supplemental cohort infants had nasal swabs collected at the time of evaluation in the emergency department or physician’s office. Nasal swabs were tested for the presence of RSV RNA using an RSV-specific reverse-transcriptase polymerase chain reaction (RT-PCR) assay or by rapid antigen testing (Quidel) and confirmed by RT-PCR, as described [41]. The hospital cohort infants were identified as RSV-infected by the clinical virology laboratory using either of 2 commercial RT-PCR assays (Focus Simplexa RSV/influenza duplex PCR or the Pasteur Merrieux Biofire multiplex PCR assay).

The chronology of study visits for the birth cohort (cord blood and 1-month visit) and illness visits for all RSV infected infants in each of the cohorts are exemplified in Figure 1.Enrollment visit: At enrollment a study nurse and 1 of the physician investigators (MTC, EEW, ARF) explained study objectives and procedures to the parents and obtained written informed consent. Demographic data were collected including gestational age, birth weight, household size including number and age of siblings or other children, tobacco use by mother/household members, and breastfeeding frequency. For infants in the birth cohort, a cord blood sample and a buccal swab were obtained (Figure 1).

One-month visit: The birth cohort was seen at the age of 1 month at a time when asymptomatic and a nasal swab, nasal wash, nasal brush, and buccal swab was collected.

Acute respiratory Illness visits: When infants in the birth cohort and supplemental cohort were seen for respiratory illness, a nasal swab was obtained for same day RSV diagnosis. If RSV was identified, the infant was scheduled for an RSV positive illness visit within 24 hours at the research clinic or during a second home visit.

RSV positive illness visits: RSV-infected Infants were evaluated at 3 time points: acute visit when RSV was first identified, at 12 to 16 days, and approximately 28 days after illness onset. At the acute illness visit, clinical information was collected (date of illness onset, presence of nasal congestion/discharge, cough, wheezing, rapid breathing, apnea, cyanosis, fever, lethargy, poor feeding). A physical examination was performed noting weight, pulse, respiratory rate, temperature, room air oxygen saturation (SaO2), presence of cyanosis, nasal discharge, rales or rhonchi, wheezes, and chest retractions. Biological samples were collected in the following sequence: buccal swab, nasal swab from 1 nostril, nasal wash followed by nasal brush from the contralateral nostril, and 2 to 3 ml venous blood sample. Parents were given an illness diary card to record signs of illness for a 10-day period beginning retrospectively from illness onset. Medical records of hospitalized infants and those seen in the emergency department or physicians’ offices were reviewed and clinical information recorded. For hospitalized infants the following information was also collected: amount and duration of supplemental oxygen, respiratory support and intensive care requirements, fluid and antibiotic administration, and worst values for each of the respiratory (respiratory rate, pulse, wheezing, rales/rhonchi, retractions, and SaO2) and systemic signs (fever, lethargy, difficulty feeding). Length of hospitalization and results of complete blood count, blood cultures, and chest radiographic findings, if available, were recorded. Interim history and findings on physical examination and room air SaO2 were recorded; a 2 to 3 ml venous blood sample obtained at the second visit; and a nasal swab, nasal wash, nasal brush, and 2 to 3 cc venous blood sample was obtained at the third visit.

Multiple study procedures were performed on enrolled subjects (Table 1).

RSV RT-PCR: Diagnosis of RSV infection and designation as group A or B RSV was made by RT-PCR using nasal swab samples as previously published [41].

Demographic and clinical data will be assessed by descriptive analysis using means and SE, medians and inter-quartile ranges for continuous variables, and proportions for categorical variables. We will use graphical methods such as histograms, Q-Q plots, and box-plots to visualize the data and identify potential data problems such as outliers, missingness, and skewness. For continuous variables, we will test their normality by Shapiro-Wilk test and Kolmogorov-Smirnov test. If problems are detected, appropriate data preparation steps such as outlier removing, data imputation, and log-transformations will be considered. For those variables that pass the normality test, we will perform 1-way analysis of variance (ANOVA) F test followed by Bonferroni post hoc testing for pairwise group comparisons. When groups exhibit unequal variances, Welch ANOVA method will be used instead. For non-normal variables, nonparametric Kruskal-Wallis test with Dunn post hoc test will be used instead. Extended Fisher exact test will be used to compare the proportions of categorical variables such as gender and race between cohort groups. Pearson and Spearman correlation analysis will be used to assess associations between 2 continuous variables such as severity and gene expression levels. Multivariate linear regression will be used to model the associations between covariates and continuous outcome variables, controlling for possible confounding effects such as age and sex. ANOVA F test and regression t test will be used to assess the significance of the overall and specific linear association between the covariates and the outcome variables. Multivariate logistic regression will be used to model associations between covariates and categorical outcome variables such as hospitalization, controlling for possible confounding effects. The Wald test will be used to assess significance of the association between covariates of interest and the outcome variables. P values <.05 will be considered statistically significant for standard statistical analyses. Some research aims involve analyzing high-throughput data such as RNAseq-based transcriptomic data and 16S-based microbiome data, and we will apply suitable multiple testing procedures, such as the Benjamini-Hochberg procedure, to control false discovery rate at a prespecified level (0.05) [58]. These high-throughput data may have high-level of between-sample variations, in part because of technical issues such as technical noise and batch effects. These undesirable variations can be reduced by a combination of stringent quality assurance analysis and specialized data transformation techniques. Specifically, samples with insufficient quality metrics (low read number, mapping rate or poor sample-wide correlation) will be excluded from analysis. Before statistical analysis, we will explore the dataset by principle component analysis (PCA), principal coordinates analysis, and hierarchical clustering to identify any unwarranted structure or association. If necessary, batch correction methods such as ComBat [59] and surrogate variable analysis [60] may be applied. Due to the non-normal nature of these data, specialized normalization methods and analytical pipelines will be used whenever appropriate [53,60-69]. All statistical analyses will be performed in R 3.3.0 (R Foundation for Statistical Computing, Vienna, Austria) and SAS 9.3 (SAS Institute).

Primary hypothesis: Our primary hypothesis is that we will be able to identify a number of factors associated with disease severity during primary RSV infection in full-term healthy infants. The first step toward this goal was to develop a Global Respiratory Severity Score (GRSS) reflecting the severity of the entire illness as the main outcome variable in most of our analyses [70]. PCA and multivariate logistic regression was used to determine the 9 optimal clinical parameters and their relative weights that comprise the GRSS (available as a Web-based algorithm [71]). We will assess each of the following datasets for their association with disease severity: (1) CD4⁺, CD8⁺ T cell, B cell ,and NK cell gene expression during RSV infection; (2) nasal epithelial cell gene expression before, during, and after RSV infection; (3) CD4⁺ and CD8⁺ cytokine synthesis as measured by flow cytometry during and after RSV infection; (4) serum antibody to RSV F and G proteins and the conserved central CX3C G protein region; (5) serum neutralizing titers to RSV A2 and B1 viruses during and after infection; (6) nasal microbiota composition before, during, and after RSV infection; (7) viral load; and (8) complete RSV genomic sequences. Each analysis will be adjusted for confounding demographic variables such as sex, gestational age, breast feeding history, age at infection, history of exposure to tobacco smoke, coinfection with other respiratory viruses, and the presence of pathogenic bacteria such as streptococcus pneumoniae, haemophilus influenzae, and moraxella catarrhalis as determined by direct PCR. Most importantly, we will seek to integrate several datasets in complex analyses to assess various interactions between virus, nasal epithelial cell and T cell gene expression, nasal microbiota, and adaptive immune responses as they relate to disease severity. Due to the inclusion of multiple high-dimensional datasets, we plan to assemble novel data integration pipelines that use state-of-the-art dimension reduction methods such as multi-omics factor analysis [72] and multi-set canonical correlation analysis [73] to extract the most informative features from individual high-throughput datasets, and afterwards, use statistical learning techniques such as penalized regression [74-76] and support vector machine classification [77] to predict disease severity. Strict cross-validation criterion will be used to evaluate the performance of the proposed methods. Recently, we developed FUNNEL, which is a time-course gene set enrichment method that has the capability to incorporate both between-gene correlation and weights [78]. We propose to design a new weighting method based on supervised principal component analysis [79] and extend FUNNEL for cross-sectional data and use this new method to study the biological functions of key genes with the largest absolute loadings in the extracted principle components/canonical vectors.

During the 3 seasons, 226 infants were enrolled in the birth cohort. Of these, 126 (55.8%) attended the scheduled 1-month visit, indicating a relatively high early attrition rate. Following the 1-month visit, 150 respiratory illnesses were reported in 81 infants (64% of active subjects) during their first winter season, each of which was evaluated during a home visit or clinic visit. A total of 36 infants had a single illness, 29 had 2 illnesses, 11 had 3, and 5 had 4 or more illnesses. A total of 19 of the 81 (23.5%) evaluated infants were shown to be RSV positive, and the overall incidence of RSV infection in the active birth cohort was 15% (19/126). Among the 19, 4 were hospitalized during the RSV infection, 1 of whom was admitted to the pediatric intensive care unit (PICU) for noninvasive respiratory support.

The supplemental cohort was recruited during the second and third winter seasons when it was apparent that the early withdrawal rate in the birth cohort was higher than anticipated. A total of 149 infants with respiratory illness met inclusion/exclusion criteria for the supplemental cohort. Of these, 60 (41%) were documented to have RSV infection, and 42 (70%) were enrolled in the full study. A total of 24 (57%) infants were enrolled from physician offices, whereas 18 (43%) were enrolled from the emergency departments. A total of 2 infants were subsequently hospitalized, 1 of who was admitted to the PICU and intubated.

During 3 seasons, 78 RSV-infected infants meeting inclusion/exclusion criteria were enrolled at the time of hospitalization. A total of 9 were cared for in the PICU and 4 intubated. The average length of stay for all hospitalized infants, including those from the birth and supplemental cohorts, was 4.2 (SD 0.6) days, with a median of 2.3 days. None of the infants died. The demographic characteristics of the RSV-positive subjects in the 3 cohorts are shown in Table 2.

Although for most factors the differences between cohorts were relatively minor (ie, gestational age, mother’s age, household size, number of siblings in the home), there were several significant differences. Birth by cesarean section was significantly lower in the supplemental cohort (6/42) compared with the birth cohort (9/19) and the hospital cohort (25/78; P=.001 and P=.05, respectively). Maternal smoking was more common in both the supplemental (8/42) and hospital cohorts (15/78) compared with the RSV-infected birth cohort (0/19), both significant (P=.05 and P=.04, respectively).

A large number of biological samples were collected from the enrolled subjects, including 217 cord blood samples (birth cohort only), 533 buccal swabs, 322 illness blood samples, 661 nasal swabs, 356 nasal wash samples, and 366 nasal brush samples.

As noted, we first completed development of the GRSS [70]. The GRSS distribution of the enrolled infants varied by cohort as expected, with lower severity scores in the birth and supplemental cohorts and higher severity score in the hospital cohort (Figure 2).

The mean (SE) GRSS for the birth and supplemental cohorts (both including hospitalized infants from these cohorts) were 2.6 (0.5) and 2.1 (0.3), respectively, and was 6.2 (0.2) for the hospital cohort.

A comprehensive study of this type, one that requires a large number of simultaneously collected samples at several time points before, during, and after RSV infection, offered several obstacles in recruitment and retention of subjects. We believe the relatively intensive commitment required on the part of young parents, often with the stress of caring for their first infant, resulted in a higher than anticipated attrition rate among the birth cohort. To adjust for loss of subjects in the birth cohort, we modified the enrollment strategy to include infants seen in the outpatient setting for acute respiratory illness or during routine visits for immunization when they were noted to have very mild respiratory signs. This modification of the enrollment strategy resulted in enrollment of mildly ill RSV-infected infants, thus, allowing development of a GRSS with a scale of 0 to 10 that spans the entire severity spectrum. This tool will allow us to utilize the severity endpoint as either a dichotomous or continuous variable in our analyses of the various parameters measured. As these datasets are completed, they will be made available in public repositories for use by other investigators.

We chose to assay host gene expression responses as a primary indicator of health status of subjects. The methods we have developed to interrogate nasal gene expression, which appears to be a reasonable surrogate of the airway and is capable of being recovered from infants of varying age and health status, is highly novel. This approach is clearly capable of providing significant new opportunities to define the molecular status of the airway in ill infants, a major impediment to prior research of infant lung diseases [80]. We also chose to separate peripheral blood cells into major subtypes before interrogation of their gene expression. Although this raised some significant technical challenges, they have been overcome [37,50]. Furthermore, this approach allows us to better achieve our primary objectives of defining the biology of the system and how it is affected by disease state.

Ultimately, we plan to integrate each of the datasets using disease severity (GRSS) as the clinical outcome, allowing us to identify and account for interactions among the various data types during the infant’s response to RSV infection. These analyses should provide important insights into to the complexity of RSV disease pathogenesis and potentially to novel interventions to alter RSV severity.