key: cord-0287658-m8i0dda6
authors: Liou, T. G.; Argel, N.; Asfour, F.; Brown, P. S.; Chatfield, B. A.; Cox, D. R.; Daines, C. L.; Durham, D.; Francis, J. B.; Glover, B.; Helms, M.; Heynekamp, T.; Hoidal, J. R.; Jensen, J. L.; Kartsonaki, C.; Keogh, R.; Kopecky, C. M.; Lechtzin, N.; Li, Y.; Lysinger, J.; Molina, O.; Nakamura, C.; Packer, K. A.; Paine, R.; Poch, K. R.; Quittner, A. L.; Radford, P.; Redway, A. J.; Sagel, S. D.; Szczesniak, R. D.; Sprandel, S.; Taylor-Cousar, J. L.; Vroom, J. B.; Yoshikawa, R.; Clancy, J. P.; Elborn, J. S.; Olivier, K. N.; Adler, F. R.
title: Associations of Sputum Biomarkers with Clinical Outcomes in People with Cystic Fibrosis
date: 2022-05-26
journal: nan
DOI: 10.1101/2022.05.25.22275540
sha: eeb621a21a3d9b6f087af4522ae0742189810735
doc_id: 287658
cord_uid: m8i0dda6

Background: Airway inflammation promotes bronchiectasis and lung injury in cystic fibrosis (CF). Amplification of inflammation underlies pulmonary exacerbations of disease. We asked whether sputum inflammatory biomarkers provide explanatory information on pulmonary exacerbations. Patients and Methods: We collected sputum from randomly chosen stable adolescents and adults and prospectively observed time to next exacerbation, our primary outcome. We evaluated relationships between potential biomarkers of inflammation, clinical characteristics and outcomes and assessed clinical variables as potential confounders or mediators of explanatory models. We assessed associations between the markers and time to next exacerbation using proportional hazard models adjusting for confounders. Results: We enrolled 114 patients, collected data on clinical variables [December 8, 2014 to January 16, 2016; 46% male, mean age 28 years (SD 12), mean percent predicted forced expiratory volume in 1 s (FEV1%) 70 (SD 22)] and measured 24 inflammatory markers. Half of the inflammatory markers were plausibly associated with time to next exacerbation. Age and sex were confounders while we found that FEV1% was a mediator. Three potential biomarkers of RAGE axis inflammation were associated with time to next exacerbation while six potential neutrophil-associated biomarkers indicate associations between protease activity or reactive oxygen species with time to next exacerbation. Conclusion: Pulmonary exacerbation biomarkers are part of the RAGE proinflammatory axis or reflect neutrophil activity, specifically implicating protease and oxidative stress injury. Further investigations or development of novel anti-inflammatory agents should consider RAGE axis, protease and oxidant stress antagonists.

Airway inflammation erodes lung health in cystic fibrosis (CF). Infecting organisms initiate and amplify inflammation in infants with CF (1) . Infections persist, interact (2) and lead to airway obstruction, mucus impaction, antibiotic resistant microbial biofilms and bronchiectasis. Consequences include strenuous treatment burdens, reduced quality of life, hospitalizations for pulmonary exacerbations and early mortality (3) .

Nine variables summarize clinical CF and durably and successfully predict survival (4, 5) but do not explain inflammatory pathways underlying clinical disease and outcomes. Seeking explanations quantitatively linking inflammation to clinical outcomes, we previously found that high mobility group box 1 (HMGB1) adjusted by number of prior pulmonary exacerbations was associated with time to next exacerbation in a proportional hazards model (6) . Other biomarkers may help identify and explain inflammatory pathway roles in CF. In particular, calprotectin (a heterodimer of S100A8 and S100A9), neutrophil elastase (NE) were associated with subsequent pulmonary exacerbations (7) (8) (9) in single center studies. None of the study cohorts fully represented people with CF, and no study considered confounding (10) or mediation (11) of inflammatory relationships by clinical covariables.

To potentially validate calprotectin, NE or HMGB1, or identify other biomarkers as explanatory for clinical CF outcomes, we performed a prospective observational multicenter study incorporating randomized selection to minimize observer bias and maximize generalizability (12). We collected sputum during clinical stability as the least invasive sample closest to sources of persistent airway inflammation and evaluated clinical covariables for confounding and mediation of biomarker effects leading to a pulmonary exacerbation. This strategy does not seek a prediction model (13), but rather searches for insights into pathways underlying pulmonary exacerbations in CF, seeking specific targets for further investigation and potentially novel interventions.

We compared clinical measurements that predict CF survival (4, 5) between participants with at least one pulmonary exacerbation in the year following enrollment and those without exacerbations. We obtained estimates of the distribution of time to next pulmonary exacerbation stratified by clinical variables, followed by evaluations using logistic regression (18) and proportional hazards models (19) . For logistic regression, we used the occurrence of the next exacerbation during study follow up as the outcome with clinical variables as the inputs, and for proportional hazards models, we used time to next exacerbation as the outcome with the clinical variables as explanatory variables.

We evaluated collinearity among biomarker measurements adjusted for detection limits with Pearson Product Moment Correlations (20) . We considered p < 0.01 after Bonferroni correction as statistically significant.

Age, sex and prior pulmonary exacerbations may confound time-to-next pulmonary exacerbation but by definition cannot be mediators thus should be included to adjust models (10) . To help assess whether other clinical variables were confounding, mediating or both (11) for associations of biochemical markers with time to next pulmonary exacerbation, we fitted univariable linear regression models of age, weight-for age z-score, diabetes and FEV 1 % as outcomes with each potential biomarker as explanatory variable. We fitted quasi-Poisson models of prior pulmonary exacerbations with each potential biomarker. We tested each potential biomarker as an adjustment to each of the remaining potentially mediating clinical variables (weight-for-age zscore, diabetes and FEV 1 %) in proportional hazards models for time to next pulmonary exacerbation. We excluded variables that substantially reduced biomarker effect sizes and significance, indicating a mediation effect.

We fitted a regression model for each biomarker with 5-year prognostic risk score variables (4, 5) dependent, using the appropriate regression method as appropriate: linear regressions for age, FEV 1 % and weight-for-age z-score; quasi-Poisson regression for pulmonary exacerbation count in the year prior to enrollment; logistic regressions for pancreatic sufficiency, diabetes and status of infections with MSSA and Burkholderia cepacia complex. Because of their clinical importance (3, 21) , we evaluated relationships between (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. MRSA and Pseudomonas aeruginosa status with potential biomarkers using logistic regression.

To develop potentially explanatory models for our primary outcome, time to next pulmonary exacerbation, we fitted univariable proportional hazards models (19) for each biomarker, adjusted for confounding variables and excluded mediating variables identified by mediation analysis (11) . For each set of models, we accounted for multiple testing using FDR analysis (22) for cutoff values of 0.001, 0.01, 0.05, 0.1 and 0.2 to understand the support for discovered associations for different biomarkers.

We assessed sensitivity of models to anti-inflammatory treatments including inhaled and oral steroids, oral azithromycin and other antibiotics and inhaled antibiotics including aztreonam and tobramycin in any form.

We examined ivacaftor or ivacaftor-lumacaftor combination medication effects because they alter airway physiology (3). In each case, we examined the models for independent effects and interactions with potential biomarkers that could indicate mediation of medication effects. We assessed the impact of using Global Lung Initiative equations (23) for estimating FEV 1 % instead of NHANES III equations (15) .

Some sputum samples were small, preventing some biomarker measurements. We preferentially assayed NE, HMGB1 and Calprotectin (12). To evaluate missingness associated bias, we performed multiple imputation by chained equations (24) for completely missing biomarker values using times to next exacerbation, nonmissing biomarker values, age, sex, FEV 1 %, weight-for-age z-score and number of prior pulmonary exacerbations. We performed three analyses, first, using a dataset with deletion of records with biomarkers that were completely or partially unmeasured, second, using records with deletion of records with completely unmeasured biomarkers but retaining records with partial measurements and third, with all records with imputed values as needed, seeking evidence that missingness was other than completely at random. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 26, 2022. ;

We enrolled 114 from Mountain West CF Consortium (MWCFC) patients and collected sputum samples from December 8, 2014 through January 16, 2016. All patients completed follow up to the next pulmonary exacerbation after enrollment or were censored at study end which occurred up to 2.65 years after enrollment.

Characteristics differed between patients who had at least one pulmonary exacerbation or had none within one year post-enrollment. The differences suggested that the 5-year prognostic risk score (4, 5) , FEV 1 %, number of exacerbations in the year prior to enrollment and weight-for-age z-score are potential clinical confounders, mediators or both for time to next exacerbation ( Table 1) . No other variable incorporated in the 5year prognostic risk score or methicillin resistant Staphylococcus aureus (MRSA) or Pseudomonas aeruginosa significantly differed between patients with and without an exacerbation during one year follow up.

Kaplan-Meier analysis, logistic regression and proportional hazards modeling found that FEV 1 %, number of prior pulmonary exacerbations, weight-for-age z-score and diabetes were associated with time to next exacerbation within a year of enrollment (Supplemental Figure 1 , Supplemental Table 3A -B). Proportional hazards modeling confirmed these associations (Supplemental Tables 3B-C) and confirmed these specific clinical markers as potential confounders, mediators or both (10, 11) . The remaining prognostic risk score variables, MRSA or Pseudomonas aeruginosa infection status were not associated with time to next exacerbation.

Biomarker measurements outside the limits of detection were adjusted after log transformation (Supplemental Table 4 ), to preserve partial information and retain as many patients as possible for analysis. Table   5 ) suggesting that confounding and mediation may exist among the biomarkers themselves.

We found multiple univariable associations among biomarkers by proportional hazards modeling for time to next pulmonary exacerbation. The results suggest that some potential biomarkers may be causal variables for time to next pulmonary exacerbation (11) ( Table 2 and Supplemental Table 6 ).

We evaluated confounding and mediation by clinical variables (Supplemental Figure 2) and found associations between different subsets of our markers with each clinical variable except sex (Table 3 and   Supplemental Table 7A -E). With a false discovery rate (FDR) <0.001, three biomarkers were significantly associated with FEV 1 %, secretory leukoprotease inhibitor (SLPI), NE and myeloperoxidase (MPO), and for an FDR <0.01, an additional three were significant, soluble receptor for advanced glycation end-products (sRAGE), matrix metallopeptidase 9 (MMP9) and extracellular newly identified receptor for advanced glycation end-products (ENRAGE) (Supplemental Figure 3 ). S100A8 was likely to be associated with weightfor-age z-score (Table 3) . Altogether, these findings suggest a potential for confounding or mediation by these clinical markers (11) .

By definition, age, sex, and prior pulmonary exacerbations, cannot mediate associations between biomarkers and time to next exacerbation. Thus, we treated them as confounders in further modeling with biochemical markers of inflammation.

We evaluated weight-for-age z-score, diabetes and FEV 1 % as potential confounders, mediators or both.

In proportional hazards models of weight-for-age z-score and diabetes for time to next exacerbation, the hazard ratios and p-values for the biomarkers used as adjustment variables (Table 4 and Supplemental Tables 8A-C) were similar to those found in earlier models of the biomarkers alone for time to next exacerbation ( Table 2 and   Supplemental Table 6 ). Weight-for-age z-score and diabetes are unlikely to be confounders or mediators of inflammation.

Estimated biomarker hazard ratios were weaker and uniformly non-significant when used as adjustments in proportional hazards models of FEV 1 % for time to next exacerbation (Table 4 and Supplemental Tables 8A-C) . FEV 1 % partially mediates the association between inflammation and time to next exacerbation and should be excluded from further models (11) . (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. 

Multivariable proportional hazards models of each potential biomarker including age, sex and prior pulmonary exacerbations as confounders for time to next pulmonary exacerbation (10) found 14 inflammatory markers with hazard ratios for time to next exacerbation with p < 0.2 (Table 5 and Supplemental Table 9 ).

Among these, setting false discovery rate (FDR) < 0.2 sets an 80% likelihood suggesting eight of the first ten have true associations with time to next exacerbation (Supplemental Figure 4) . Setting the FDR < 0.01 shows ENRAGE and MPO each have > 99% chance of being true findings (Table 5) .

Adding inflammation modifying treatments as confounders to the proportional hazards models adjusted by age, sex and number of prior exacerbations did not substantially modify coefficients for any biomarker. Use of corticosteroids, inhaled, oral or both, chronic azithromycin, inhaled aztreonam, tobramycin, each alone or alternating every other month had no significant independent associations with time to next exacerbation nor interactions with the biomarker variables in each model. Coefficients (Table 5 ) remained stable throughout sensitivity testing suggesting that results are not due to anti-inflammatory treatments.

Seven patients used ivacaftor or ivacaftor-lumacaftor, but their use had no independent effect or interaction, and individual biomarker variable effects were stable throughout testing. Use of Global Lung Initiative equations (23) to calculate FEV 1 % instead of NHANES III led to similar results and identical interpretations and conclusions.

We repeated models to understand the impact of partially missing information due to biomarker measurements outside detection limits and completely missing data due to insufficient samples. Analyses excluding patients with partially missing data produced similar results for models derived from using complete data sets. Analyses with missing data imputed using multiple imputation with chained equations produced (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Tables 6-9 ). We found no evidence of bias for any relationship examined.

We measured 24 sputum biomarkers from 114 randomly chosen, clinically stable adolescents and adults with CF to validate prior observations for calprotectin, NE and HMGB1 (6) (7) (8) and to explore causal inference, confounding (10) and mediation (11) between an expanded list of inflammatory biomarkers and clinical outcomes. After evaluation for confounders and mediators of biomarker effects, we found potential associations with ten different molecules: ENRAGE, MPO, sRAGE, ICAM1, NE, chitinase-3-like 1 protein (YKL40), thymus and activation regulated chemokine (TARC), MMP9, IL1β and IL5 (Table 5 ). At least eight are associated with time to next pulmonary exacerbation (Supplemental Figure 4) . FEV 1 % mediates biomarker effects and was excluded from every model, while age, sex, number of prior pulmonary exacerbations were confounders and were included (10, 11) . Results were insensitive to anti-inflammatory treatments or ivacaftor, although the latter was infrequently used.

Severe CF lung disease continues to steadily decrease as treatments with CFTR modulators increase (25) . FEV 1 % remains the single most common measure of disease severity, but it strongly mediates and obscures biomarker effects (Table 4 ). FEV 1 % poorly distinguishes between rapid or slow worsening of lung disease (4). As CF disease severity lessens (26), FEV 1 % will become less useful, and biomarkers measuring inflammation with appropriate adjustments for confounding will be increasingly useful for understanding clinical status ( Figure 1 ) and providing attractive intervention targets.

Biomarkers associated with time to next exacerbation fall into several categories (Figure 2 ). ENRAGE and sRAGE are constituents of RAGE-axis related inflammation. ENRAGE is a high avidity RAGE ligand that indicates neutrophil activity (27) . It is elevated in CF airways in infancy (28) and is associated with low FEV 1 and CF-related diabetes (29) . ENRAGE increases during pulmonary exacerbation and falls with antibiotic treatment (30) . Beyond CF, higher ENRAGE is associated with cigarette smoking-induced injury (31), higher coronary artery disease risk (32) and higher chronic kidney disease mortality (33) . ENRAGE increases in (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 26, 2022. ; https://doi.org/10.1101/2022.05.25.22275540 doi: medRxiv preprint bronchoalveolar lavage in the acute respiratory distress syndrome (ARDS) and decreases with recovery (34) . In SARS-CoV-2-associated ARDS, it rises with increasing severity and falls with recovery (35) . Table 5 Figure 3) were associated with FEV 1 % with a FDR set at 0.001. For FDR set at 0.01, we found another three biomarkers, sRAGE, MMP9 and ENRAGE. Anti-protease SLPI opposes NE, MMP9 and proteinase 3 (PR3) activities (7, 37, 38) and was associated with higher FEV 1 % (Table   3 ). These findings suggest that RAGE-axis, protease-anti-protease imbalance and reactive oxygen species sources of inflammation might cause lower FEV 1 % (Supplemental Figure 5 ).

Contrary to pre-study expectations (6) (7) (8) , Calprotectin, NE and HMGB1 were not associated with time to next pulmonary exacerbation. However, the study redemonstrated associations between these molecules and prior pulmonary exacerbations (Table 3 and Supplemental Table 7D ), and they are markers of neutrophil activity and part of the RAGE-axis of inflammation. These partial validation results reinforce that this study differs from prior studies in at least two substantial ways. First, we recruited randomly selected clinically stable (39) rather than mixed stable and unstable serial patients. Second, our analyses evaluated confounding and mediation by clinical covariables (10, 11) .

Our study has limitations. Because we are pursuing explanatory airway inflammation models (13), no methods of model selection are sufficient (10) to confidently produce a multivariable panel of inflammatory biomarkers. However, our current observational work will be useful prior to launching new observational and other types of investigations to discover a multivariable biomarker model. Although many study patients took support section that follows participated in any way with this trial. Table 6 for results for all biomarkers.

All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 26, 2022. ;  (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 26, 2022. ; 

We categorized patients by hazard ratios higher or lower than the median estimated using a full model including log transformed values of ENRAGE adjusted by age, sex and number of prior pulmonary exacerbations as confounders. We drew Kaplan-Meier plots and compared with plots based on hazard ratios estimated using model A. excluding ENRAGE data, B. excluding prior pulmonary exacerbations, C. including only ENRAGE data, and D. including only prior pulmonary exacerbations. The numbers at risk for A and D differ from other plots because median hazard ratios including exacerbation count data did not divide patients into equal groups. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. 

Airway microbiota across age and disease spectrum in cystic fibrosis

Microbial Interactions in the Cystic Fibrosis Airway

Cystic fibrosis

Predictive 5-year survivorship model of cystic fibrosis

Evaluation of a five-year predicted survival model for cystic fibrosis in later time periods

Sputum biomarkers and the prediction of clinical outcomes in patients with cystic fibrosis

Sputum biomarkers of inflammation and lung function decline in children with cystic fibrosis

Sputum and serum calprotectin are useful biomarkers during CF exacerbation

Measurement of Serum Calprotectin in Stable Patients Predicts Exacerbation and Lung Function Decline in Cystic Fibrosis

Control of Confounding and Reporting of Results in Causal Inference Studies. Guidance for Authors from Editors of Respiratory, Sleep, and Critical Care Journals

The moderator-mediator variable distinction in social psychological research: conceptual, strategic, and statistical considerations

Spirometric reference values from a sample of the general U.S. population

Exploratory data analysis

Nonparametric Estimation from Incomplete Observations

The Regression Analysis of Binary Sequences

Regression models and life-tables

Modern Applied Statistics with S

Association between respiratory tract methicillin-resistant Staphylococcus aureus and survival in cystic fibrosis

Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing

Multi-ethnic reference values for spirometry for the 3-95-yr age range: the global lung function 2012 equations

Multiple imputation of missing blood pressure covariates in survival analysis

Modeling long-term health outcomes of patients with cystic fibrosis homozygous for F508del-CFTR treated with lumacaftor/ivacaftor

Time to get serious about the detection and monitoring of early lung disease in cystic fibrosis

Activation of critical, host-induced, metabolic and stress pathways marks neutrophil entry into cystic fibrosis lungs

Novel neutrophil-derived proteins in bronchoalveolar lavage fluid indicate an exaggerated inflammatory response in pediatric cystic fibrosis patients

Advanced glycation end products are elevated in cystic fibrosis-related diabetes and correlate with worse lung function

Expression of S100A12 (EN-RAGE) in cystic fibrosis

RAGE: developmental expression and positive feedback regulation by Egr-1 during cigarette smoke exposure in pulmonary epithelial cells

EN-RAGE: a novel inflammatory marker for incident coronary heart disease

Plasma S100A12 and soluble receptor of advanced glycation end product levels and mortality in chronic kidney disease Stage 5 patients

Neutrophil-derived S100A12 in acute lung injury and respiratory distress syndrome

Systems biological assessment of immunity to mild versus severe COVID-19 infection in humans

Induced sputum matrix metalloproteinase-9 correlates with lung function and airway inflammation in children with cystic fibrosis

Myeloperoxidase oxidation of methionine associates with early cystic fibrosis lung disease

Interaction of secretory leukocyte protease inhibitor with proteinase-3

Planning of Experiments

Research electronic data capture (REDCap)--a metadata-driven methodology and workflow process for providing translational research informatics support