Early childhood lung function is a stronger predictor of adolescent lung function in cystic fibrosis than early Pseudomonas aeruginosa infection

Objective Pseudomonas aeruginosa has been suggested as a major determinant of poor pulmonary outcomes in cystic fibrosis (CF), although other factors play a role. Our objective was to investigate the association of early childhood Pseudomonas infection on differences in lung function in adolescence with CF. Methods Two populations of subjects with CF were studied: from the Gene Modifier Study (GMS), 346 F508del homozygotes with severe vs. mild adolescent lung disease, and from the Colorado Newborn Screen Study (NBS) 172 subjects diagnosed with CF by newborn screening. Associations of Pseudomonas infection and lung function in early childhood with lung function in adolescence were investigated using multivariate linear regression analyses. Results Among GMS subjects, those with severe adolescent lung disease had worse lung function in childhood (FEV1 25 percentage points lower) compared to subjects with mild adolescent lung disease, regardless of early childhood Pseudomonas status. Among NBS subjects, those with lowest adolescent lung function had significantly lower early childhood lung function and faster rate of decline in FEV1 than subjects with highest adolescent lung function; early Pseudomonas infection was not associated with rate of FEV1 decline. The strongest predictor of adolescent lung function was early childhood lung function. Subjects with a higher percentage of cultures positive for Pseudomonas before age 6 or a lower BMI at 2–4 years old also had lower adolescent lung function, though these associations were not as strong as with early childhood lung function. Conclusions In separate analyses of two distinct populations of subjects with CF, we found a strong correlation between lower lung function in early childhood and adolescence, regardless of early childhood Pseudomonas status. Factors in addition to early Pseudomonas infection have a strong impact on lung function in early childhood in CF. Further exploration may identify novel underlying genetic or environmental factors that predispose children with CF to early loss of lung function.


Results
Among GMS subjects, those with severe adolescent lung disease had worse lung function in childhood (FEV 1 25 percentage points lower) compared to subjects with mild adolescent lung disease, regardless of early childhood Pseudomonas status. Among NBS subjects, those with lowest adolescent lung function had significantly lower early childhood lung function and faster rate of decline in FEV 1 than subjects with highest adolescent lung function; PLOS
We have previously shown in a large, retrospective case-control study that earlier age of Pa infection (before 5 years of age) was strongly associated with severe (vs. mild) CF lung disease in adolescence and adulthood [29]. However, we hypothesized that multiple other factors are involved in determining lung function in early childhood in CF, and that early Pa infection alone would not account for differences in lung function in early childhood that persist into adolescence, even in a cohort of children diagnosed by newborn screening with subsequent aggressive early management of CF lung disease. Our primary objective was to investigate the association between early childhood lung function, early Pa infection, and adolescent lung function in subjects with CF. We also aimed to investigate other factors associated with early lung function in children with CF. To accomplish this, we tested our hypothesis in two separate datasets: the Gene Modifier Study (GMS), a retrospective, case-control study with annualized data, used in our original publication [29], and the Colorado Newborn Screen (NBS) Study, a prospectively collected, encounter-based study.
Percentage of cultures positive for S. aureus prior to 6 years of age, Number of respiratory cultures before 6 years of age.
• Early childhood nutritional status: Mean BMI Z-score from 2 to 4 years of age, Mean height Z-score from 2 to 4 years of age, Mean weight Z-score from 2 to 4 years of age [42].
• Absolute change in FEV 1 percent predicted from early childhood to adolescence: defined as mean early childhood FEV 1 percent predicted minus the mean adolescent FEV 1 percent predicted; • Rate of change in FEV 1 (percent predicted/year): defined as the absolute change in FEV 1 percent predicted / (mean age at adolescent PFTs-mean age at early childhood PFTs) Other variables included for both analyses were gender, race (Caucasian vs. non-Caucasian), ethnicity (Hispanic vs. non-Hispanic), year of birth, age at diagnosis, means of diagnosis (newborn screen, meconium ileus at presentation, and symptoms [respiratory, gastrointestinal, or failure to thrive]), CFTR genotype, and age of first recorded respiratory culture.

Analyses (NBS, GMS)
Bivariate analyses used Student's t-test and χ 2 testing to compare means or proportions, respectively. Multiple comparisons were accounted for in the NBS cohort using ANOVA testing with Bonferroni correction. Comparisons consisted of:

GMS population
Of the 629 F508del homozygous subjects initially enrolled and described elsewhere [29], 346 met inclusion criteria (Fig 1). Over 60% of subjects had at least one culture positive for Pa prior to 6 years of age. GMS subjects were stratified into four subgroups by Pa infection status at age 6 (ever vs. never Pa positive prior to 6 years of age) and adolescent lung disease group (severe vs. mild). Subjects in the four sub-groups did not differ significantly by gender distribution, age at diagnosis, age of definition of adolescent lung function, or presence of other respiratory pathogens in early childhood (data not shown). Age of first culture was younger in subjects who were Pa positive in early childhood whether they had mild or severe lung function in adolescence (Mean age (years) at first culture in the mild adolescent lung disease group was 2.22 vs. 3.51 for Pa positive and negative subjects, respectively, p<0.05. Mean age at first culture in the severe adolescent lung disease groups were-2.18 vs. 3.51 for Pa positive and negative subjects, respectively, p<0.05). Similarly, Pa positive subjects had a higher mean number of respiratory cultures before age 6 regardless of lung function in adolescence (number of cultures among subjects with mild adolescent lung disease-4.56 vs. 3.11 for Pa positive and negative subjects, respectively, p<0.05. Number of cultures among subjects with severe adolescent lung disease -4.86 vs. 2.57 for Pa positive and negative subjects, p<0.05).
Fig 2 illustrates differences in early lung function among the four GMS subgroups. Among subjects who were Pa positive in early childhood, mean FEV 1 at 6-8 years of age was lower in those with severe vs. mild adolescent lung disease (72.4% ± 15.8 and 99.4% ±13.1, respectively) (p<0.0001). Among Pa-negative subjects, there was a similar difference in early childhood FEV 1 between those with severe vs. mild adolescent lung disease (79.1% ± 16.7 and 103.2% ± 14, respectively) (p<0.0001). In comparing subjects with severe lung disease later in life, mean early childhood FEV 1 was~8 percentage points lower in those subjects who were Pa-positive before age 6 (p = 0.01); however, there was no significant difference in early FEV 1 between Pa positive and negative subjects with mild adolescent lung disease.

NBS population
Of 434 subjects included in the Colorado NBS database, 172 were included in final analysis; the majority excluded were not old enough to have "adolescent" lung function data available at 10 years of age or older (Fig 3).
Comparison by adolescent lung function quartile. NBS subjects were first divided into even quartiles by adolescent lung function (n = 43 in each quartile), then further divided by Pa status in early childhood. When comparing NBS subjects by lung function quartile in adolescence, subjects with lower adolescent lung function were born earlier and had a higher percentage of cultures positive for Pa prior to age 6 than adolescents with higher lung function ( Table 2). There were no significant differences in nutritional status, age of first culture, number of cultures before age 6, percentage of subjects Pa-positive before age 6, or age at first Papositive culture between the adolescent lung function quartile groups. Subjects with lower adolescent lung function had significantly lower FEV 1 at 6 to 8 years of age than those in the higher adolescent lung function quartiles (Table 2).
Comparison by adolescent lung function quartile and early Pa status. Fig 4 illustrates differences in early childhood lung function by adolescent lung function quartile and early childhood Pa status. Early childhood FEV 1 was significantly lower in subjects with poorer adolescent lung function who were Pa positive before age 6 (mean early childhood FEV 1 78% (±13), 89% (±7), 94% (±11), and 107% (±12) for lowest to highest adolescent lung function groups, p<0.0001). Though a similar pattern was seen in the Pa negative group, differences did not reach statistical significance (mean early childhood FEV 1 93% (±7), 95% (±16), 97% (±11), and 103% (±13) for lowest to highest adolescent lung function group, p = 0.065). When comparing Pa positive and negative subjects by adolescent lung function quartile, Pa positive subjects in the lowest adolescent lung function quartile had significantly lower early childhood lung function than Pa negative (78% (±13) vs. 93% (±6), p< 0.001); differences in early childhood FEV 1 between Pa positive and negative subjects in the other quartiles were not statistically significant (p = 0.09, 0.18, 0.86 for 2 nd lowest, 3 rd , and highest quartiles, respectively).
There was a significant decline in FEV 1 percent predicted from early childhood to adolescence in the two lowest adolescent lung function quartiles, with a decrease of 25 percentage Subjects from the parent GMS study included in current analysis were limited to those who a) had lung function data between age 6 and 8, b) were enrolled in the GMS study at 12 years of age or greater, and c) had respiratory culture data available prior to age 6. Definition of adolescent lung function occurred at time of GMS enrollment; age cutoff was created to separate time of definition of adolescent lung function from early childhood lung function.
https://doi.org/10.1371/journal.pone.0177215.g001 Mean FEV 1 percent predicted at 6 to 8 years of age is plotted by classification of lung disease in adolescence (severe vs. mild) and Pa infectious status up to age 6 (ever vs. never culture positive for Pa).
https://doi.org/10.1371/journal.pone.0177215.g002 points and 10 percentage points in lowest and 2 nd lowest quartiles, respectively (p<0.0001 for both). There was no significant change in mean FEV 1 from early childhood to adolescence in the higher two adolescent lung function quartiles (Table 3). Similarly, rate of decline in FEV 1 from childhood to adolescence (percent predicted per year) was significantly higher in subjects in the two lowest adolescent lung function quartiles (Table 3). Table 4 shows annual rate of decline (percent predicted per year) in FEV 1 from early childhood to adolescence by adolescent lung function quartile and early childhood Pa status. Subjects with the lowest adolescent lung function had significantly faster annual rates decline in FEV 1 than all other subjects; mean FEV 1 percent predicted in the two highest quartiles appeared nearly stable from childhood to adolescence (by both percent predicted and rate of Subjects in Colorado NBS analysis were limited to those 10 years of age or older at the time of analysis in order to have "adolescent" lung function data available. Subjects included in analysis were required to have data on respiratory cultures before age 6, BMI at age 2 to 4, lung function at age 6 to 8, and lung function over age 10. https://doi.org/10.1371/journal.pone.0177215.g003 decline). Importantly, rate of decline did not differ significantly within adolescent lung function quartiles when comparing subjects by early childhood Pa status (Table 4).
For all comparisons discussed above, similar patterns were seen when comparing best early childhood FEV 1 as opposed to mean (data not shown).
Linear regression. Variables included in the final linear regression model (after backwards elimination of covariates) were: adolescent FEV 1 , early childhood FEV 1 , BMI Z-score at 2-4 years of age, percentage of cultures positive for Pa before age 6, number of cultures obtained before age 6, number of PFTs included in the early childhood lung function variable, number of PFTs included in the adolescent lung function variable, gender, year of birth, and age at diagnosis. We found a strong association between lower early childhood FEV 1 and lower adolescent FEV 1 (p<0.001, Table 5). Higher percentage of cultures positive for Pa prior to age 6 was also significantly associated with lower adolescent FEV 1 (p = 0.02), though the effect appeared markedly less than that of early childhood FEV 1 . Contribution to the explained  5). Lower BMI in early childhood also showed a significant association with lower adolescent lung function (p = 0.005). Neither the simple presence of Pa prior to age 6 (yes/no dichotomous variable for at least one positive culture) nor the presence or percentage of cultures positive for S. aureus prior to age 6 were significantly associated with adolescent FEV 1 .

Discussion
We have shown in two separate, distinct cohorts of CF subjects that lower FEV 1 in early childhood (6 to 8 years of age) is strongly associated with lower FEV 1 in adolescence, and that this association is only partially explained by infection with Pa before age 6; thus, other genetic and/or environmental factors, likely in early childhood, must be playing a major role. Importantly, we have also shown a striking divergence in loss of lung function among CF patients, as Early lung function predicts adolescent disease in CF subjects with worse lung function in early childhood had a significantly faster rate of FEV 1 decline from childhood to adolescence compared to those with better early childhood lung function. This finding holds even for those diagnosed in infancy, was the majority of NBS subjects presented for management and treatment prior to the development of respiratory symptoms, and those with worse lung function early still had a faster rate of decline. Our findings suggest that predisposing genetic and/or environmental factors and events in infancy and preschool in children with CF may have a strong impact on lung function from early childhood through adolescence and adulthood, even in children diagnosed by newborn screening.
Our results are consistent with Burns' finding that 97% of infants with CF in their clinic population had serologic or microbiologic evidence of Pa exposure before age three [43], which would suggest that growth of Pa on respiratory culture cannot be the driving force in Table 3. Colorado NBS Population: Absolute change and annual rate of change (percent predicted per year) in FEV 1 percent predicted from early childhood to adolescence by adolescent lung function quartile. Absolute change in FEV 1 was defined as mean FEV 1 percent predicted age 6-8 minus mean FEV 1 percent predicted adolescent . Annual rate of change was defined as absolute change in FEV/ (mean age adolescent PFTs-mean age early childhood PFTs).  Early lung function predicts adolescent disease in CF differences in early lung function. In the GMS dataset, subjects with severe adolescent lung disease had mean early childhood FEV 1 approximately 25 percentage points lower than adolescent subjects with mild disease, regardless of early childhood Pa status. In the NBS dataset, early lung function was lower in subjects with lower adolescent lung function. We acknowledge that there is an association between early Pa infection, early childhood lung function, and adolescent lung function, as subjects with lower adolescent lung function and early Pa infection appeared to have worse early childhood lung function in both study populations. This confirms that infection with Pa is one of many factors associated with lower lung function in adolescence, but our findings emphasize that, even when adjusting for chronicity or prevalence of positive cultures, Pa is not the major cause of lower lung function in children that persists into adolescence.

Mean
Previous studies have shown that lung function in healthy subjects remains relatively stable when comparing percent-predicted values from childhood to adolescence [44,45]. Studies in children with CF have noted an association with higher baseline FEV 1 and more rapid rate of decline over the next 3-6 years [27,46]. However, we found that subjects with CF with lower adolescent lung function showed significantly greater rate of decline in FEV 1 percent predicted and annual rate of decline in FEV 1 percent predicted from early childhood to adolescence, while those subjects with higher adolescent lung function showed no significant decline from early childhood to adolescence. This suggests that children with CF and lower FEV 1 are not simply tracking along a given lung function trajectory, but have active progression and worsening of their lung disease in comparison to their healthier counterparts with CF. Our findings may differ from previous studies in follow-up time from early childhood to adolescence was relatively longer-perhaps uniformly bridging adolescence in our population resulted in different patterns of lung function decline. The NBS population was also limited to a single state population, and thus our study population was smaller and possibly more homogenous. Finally, our analysis approaches may have differed.
The strongest predictor of low lung function in adolescence was low lung function in early childhood. Importantly, there were no significant differences in rate of decline when comparing Pa positive and negative subjects within adolescent lung function quartiles. Instead, subjects with lower adolescent lung function (in both Pa positive and negative groups) both started with lower early childhood lung function and showed a faster rate of decline in lung function than their counterparts with higher adolescent lung function. Those subjects with a higher percentage of respiratory cultures positive for Pa prior to age 6 also had lower lung function in adolescence, though that association was not as strong as that with early childhood lung function, and the dichotomous presence/absence of Pa prior to age 6 was not associated with adolescent lung function. This suggests that infectious burden, and possibly inflammatory response, plays a role in outcomes, though compliance with prescribed preventative treatments such as chest physiotherapy could also explain persistent decline in some children. Our findings are consistent with previous studies implicating chronic Pa infection in poorer pulmonary outcomes in patients with CF [15,27,29,47], and implicating inflammation as a strong predictor of lung function decline [48].
The importance of S. aureus in CF lung disease has been a source of recent debate [28,30,31,49,50]; we found no significant association between S. aureus infection in early childhood and adolescent lung function, though we were unable to investigate the association with methicillin-resistant S. aureus due to low prevalence in our NBS population and lack of consistent reporting in our GMS population.
Poor nutritional status in early childhood was also associated with lower lung function in adolescence, which is consistent with previous studies [21,22]. Earlier year of birth was also associated with poorer adolescent lung function, which may reflect our study design, as these subjects would have been older when adolescent lung function was recorded, but could also reflect advances in management and treatment. Our findings suggest early childhood lung function, early nutritional status, and density or persistence of early Pa infection were most strongly associated with adolescent lung function in subjects with CF.
Our two distinct datasets show consistent results, even though they represent different birth cohorts (straddling initiation of widespread use of inhaled tobramycin for treatment of Pa infection), had different genetic makeups and geographic origins, and likely reflected different treatment regimens (based on clinical advances from GMS to NBS birth cohorts). Both datasets had distinct advantages and complement the other well. The GMS dataset was large, representing CF centers across the U.S., and genetically homogenous (all F508del homozygotes), but was limited by use of annualized data, relatively scant data in infancy and preschool, and the extremes of phenotype design. The NBS dataset was smaller regional, but contained the full spectrum of CF lung disease (not extremes of phenotype), a representative distribution of CF genotypes, and used encounter-based, non-annualized data. Initiation of CF-directed care such as chest physiotherapy at diagnosis, and prior to development of respiratory symptoms, is another advantage in the NBS dataset and eliminates a potential confounder.
Our study had several limitations. Both GMS and NBS datasets are comprised of clinically collected data; there was not uniformity in data collection frequency or variables collected. PFT data at 6-8 years of age was the earliest age at which we could determine "early" childhood lung function. Because of the annualized data in the GMS study and the extremes of phenotype design, we were not able to explore associations between early childhood events, early childhood lung function, and adolescent lung function in great detail in this population. The NBS study population had more extensive early childhood data; however, there was still substantial variability as to how much data was available on individual subjects. We were unable to include mucoid Pa or methicillin-resistant S. aureus in either analysis due to low prevalence in the NBS population, and limited data available from the CFFPR for the majority of GMS subjects. Similarly, we had limited data on other organisms such as Haemophilus influenza, Aspergillus fumigatus, or Stenotrophomonas maltophilia, and were not able to include these in our analyses, nor were we able to include differences in therapies and treatments. Finally, socioeconomic status was not reliably collected and was therefore not included in our regression model, yet socioeconomic status has been shown to be associated with worse outcomes in CF [18]. Despite these limitations, we believe the consistency of our results in two distinct populations of subjects lend support to our findings.
In summary, we have shown in two distinct populations of subjects with CF that lung function in early childhood is more strongly associated with lung function in adolescence in CF than infection with Pseudomonas aeruginosa before age 6. Most strikingly, we found that subjects with lower lung function in adolescence had lower lung function in early childhood and a more rapid rate of decline of FEV 1 from childhood to adolescence than those subjects with higher lung function in adolescence. Annual rate of decline was not different between Pa positive and Pa negative subjects in early childhood (within adolescent lung function quartile), suggesting that long-term pulmonary outcomes and disease trajectory may be largely determined prior to lung function measures at 6 years of age. Perhaps equally important, these differences were present in a cohort of children diagnosed with CF in infancy and prior to development of respiratory symptoms. Our findings suggest genetic influences and multiple factors in infancy and preschool have a substantial impact on lung function from early childhood onwards in subjects with CF. It will be critical to further study events in early childhood in patients with CF to define these key factors for future intervention.
Supporting information S1