The systematic use of metabolomic epidemiology, biobanks, and electronic medical records for precision medicine initiatives in asthma: findings suggest new guidelines to optimize treatment


 The application of large-scale metabolomic profiling provides new opportunities to realize the potential of omics-based precision medicine with regard to asthma. We leveraged over 14,000 individuals from four distinct epidemiological studies. We identified and independently replicated seventeen steroid metabolites that were significantly reduced in individuals with prevalent asthma. Importantly steroid levels were reduced among all individuals with asthma regardless of medication use; however, the largest reduction was associated with inhaled corticosteroids use (ICS) that was further confirmed in a four-year ICS clinical trial. Cortisol levels extracted from electronic medical records confirmed that cortisol is reduced among asthmatics taking ICS over the entire 24-hour period, compared with all other groups. Clinical-grade adrenal suppression in asthmatics on ICS, resulting from substantial reductions in steroid metabolites, represents a larger public health problem than previously recognized. Regular cortisol testing may identify at-risk individuals, enabling personalized treatment modifications and improving overall patient care.


Introduction
Asthma imparts a tremendous global health and economic burden, affecting over 350 million people worldwide [1][2][3][4] , with annual asthma-related costs surpassing $80 billion/year in the USA. Any improvement in treatment will therefore have signi cant public health rami cations. While several genetic variants have been determined to in uence an individual's asthma liability [5][6][7][8] , asthma also has substantial environmental triggers 9 and the majority of cases arise from complex interactions between both factors.
Metabolomics, the systematic analysis of small molecules in a biological sample, provides an integrated pro le of biological status, re ecting the 'net results' of genetic, transcriptomic, proteomic, and environmental interactions, making it ideally suited to the study of asthma. The clinical potential of asthma metabolomics is well-demonstrated through the improvement in predictive accuracy of asthma phenotypes and the improved understanding of underlying biology it has provided relative to genetics or environmental factors alone 8 . While a number of metabolomic signatures of asthma-relevant phenotypes have been reported to date 8 ; there are several drawbacks to these studies that have limited their overall clinical impact, including a lack of independent replication 10-13 and small sample sizes 11 . The identi cation and validation of metabolic signatures for asthma phenotypes using multiple large, wellcharacterized epidemiological cohorts and clinical trials with comprehensive coverage of the global metabolome is imperative for the translational potential of metabolomics to be fully realized 11,[13][14][15][16][17][18][19][20][21] .
As a composite measure that captures environmental exposures, the metabolome can inform our understanding of the biological impact that speci c medications, such as inhaled corticosteroids (ICS), may have on individuals with asthma. Long-term use of ICS among moderate to severe asthmatics is common. While several studies have examined the biological impact that prolonged ICS use may have, signi cant limitations in study design have limited their overall utility, resulting in con icting ndings 22-Page 4/17 26 . Moreover, the current guidelines on ICS use are inconsistent on dosing and effective symptom control 27 , resulting in confusion.
Metabolomic ndings from epidemiological studies may be further enhanced for clinical translation by integrating them with Electronic Medical Records (EMR) 28 , as many validated and clinically approved 'non-omic' biomarkers and clinical tests have quanti able relationships with metabolites and metabolomic pathways 29,30 . In this study, we utilized metabolomics to investigate prevalent asthma and the impact of ICS use, leveraging multiple cohorts in combination with clinical tests available via EMR to explore a metabolomic-driven precision medicine approach for optimized asthma management.

Metabolite associations in the EPIC-Norfolk cohort
We identi ed 973 metabolites that remained for analysis after quality control. In EPIC-Norfolk, 35 (3.6%) were signi cantly associated with prevalent asthma after multiple testing corrections using Bonferroni threshold (Table S3). Thirty-four of these metabolites were signi cantly reduced in asthmatics compared with controls (range of ORs=0.65-0.81; range of P-values=1.4x10 -27 -1.5x10 -7 ) and were annotated to canonical curated pathways for corticosteroids, pregnenolone, and androgenic steroids ( Figure S1A).

Replication in MGBB-Asthma cohort
Seventeen of the 35 signi cant metabolites were replicated in MGBB-Asthma at an FDR threshold of 5%; 15 of which also met the more stringent Bonferroni threshold ( Table 2). All 17 metabolites were reduced in asthmatics compared to controls ( Figure 2A) and were annotated to major steroid hormone biosynthesis sub-pathways, speci cally corticosteroid, pregnenolone, and androgenic steroid pathways (Table 2, Figure S2).

Metabolite-ICS associations in MGBB-Asthma and CAMP
In MGBB all corticosteroid, androgenic, and pregnenolone steroids were lower in both asthmatics on and off ICS when compared with controls (Table S4, Figure 2B). These reductions were larger and more signi cant among asthmatics on ICS than among asthmatics off ICS; however, steroid reductions were still evident in asthmatics off ICS when compared with controls. While progestin steroids also demonstrated this general trend in all comparisons between asthmatics and controls, the reduction was not as robust (Table S5- Table 1. The asthmatics and controls differed by sex (P=3.9x10 -14 ), race (P=4.3x10 -5 ), ICS intake (P<2.2x10 -16 ) and adrenal insu ciency diagnosis (P=0.03). We determined that asthmatics with ICS intake had the lowest minimum cortisol levels throughout a 24-hour period. Pairwise comparisons between the subgroups demonstrated that asthmatics on ICS had signi cantly lower minimum cortisol levels compared to asthmatics without ICS intake (b=-1.74 mcg/dL; 95% CI=-2.90, -0.59; P=3.3x10 -3 ) and controls without ICS intake (b=-2.86 mcg/dL; 95% CI=-3.96, -1.76; P=3.7x10 -7 ) throughout the 24-hour period with the largest difference occurring in early morning, the time when individuals are most vulnerable to asthma attacks 31,32 ( Figure   3). A comparison with controls on ICS was marginally signi cant (P=0.055); however, this group may have other comorbid and confounding diagnoses that make this relationship di cult to assess.
The Research Patient Data Registry (RPDR) identi ed 16,929 individuals with asthma and 23.4% (3,969) met the criteria for regular ICS use, as de ned for MGBB-Asthma. Of the 1,021 (25.7%) asthmatics with ICS use that were tested for adrenal insu ciency, 31.0% (n=316) met the clinical diagnostic criteria for adrenal suppression (de ned as a peak level<500 nmol/L) 33 .

Discussion
In this study, we used four independent cohorts to demonstrate the translational utility of metabolomics. We identi ed signi cant and robust reductions in steroid metabolites among asthma cases using ICS that place them at serious risk for adrenal suppression with as many as thirty percent reaching clinically diagnosable levels. There were three key ndings. First, seventeen steroid metabolites had substantially reduced levels in prevalent asthma cases compared with controls, including marked reductions in the two primary hypothalamic-pituitary-adrenal axis (HPA) steroid hormones that are biomarkers for adrenal suppression 34 , DHEA-S and cortisol. Second, we observed that this reduction in steroids was primarily, but not exclusively driven by ICS use in asthmatics, suggesting that the reduced steroid levels represent both a fundamental characteristic of the asthma phenotype and a result of ICS use. Third, not only were cortisol levels consistently reduced among asthmatics on ICS compared with all other groups throughout an entire 24-hour diurnal period, but this group had the largest variation in levels throughout the day, with a steep reduction in cortisol levels during the early morning hours, a peak time for asthma exacerbations to occur 31,32 Prior studies have correlated low cortisol with decreased asthma control 35 , suggesting that this reduction may pose a further threat during this vulnerable time 31,32 . The global reduction in cortisol levels was so pronounced that on average, peak cortisol levels among asthmatics on ICS during an entire 24-hour period, were lower than the average minimum cortisol levels among any of the other groups at any time.
To date, multiple studies have investigated the potential adverse side effects of ICS use for asthma; however, the overall utility of these studies to assess adrenal suppression has been hampered by small or modest samples sizes, short trial periods 22,25 , and a limited range of ICS dose 22,36 , with few resounding conclusions; a more comprehensive interrogation is therefore recommended 27 . While acknowledged as a potential harmful side effect, clinical suppression of the HPA-axis from ICS therapy alone has been considered unusual with minimal long-term systemic rami cations on adrenal function 22,23 that may only be apparent at high doses 25,26 . In contrast to the existing scienti c literature, in our population there was a consistent, pronounced, and clinically relevant reduction (as de ned by a pronounced in increase in adrenal suppression diagnoses) in adrenal function over a range of ICS use and dose. The utilization of 25 years of EMRs from individuals diagnosed with asthma and ICS use, cortisol measurements from a 4year ICS trial, and extensive clinical-grade cortisol testing from over 2,000 individuals enables a robust evaluation of the long-term impact of ICS use on adrenal function and suggests increased rates of adrenal and sub-adrenal suppression among asthmatics as a result of ICS use.
The majority of moderate to severe asthmatics use ICS as the rst line of treatment to improve control of persistent asthma. It remains an integral part of their long-term treatment protocol 37,38 and one of the most effective and e cacious treatments to date. In the short term, these bene ts likely outweigh the long-term side effects. However, patient treatment is optimized with proactive monitoring of circulating steroid levels to prevent permanent adrenal suppression. We observed that 31 percent of asthma cases with ICS use tested met the clinical criteria for an adrenal suppression diagnosis. While this is likely an overestimate due to potential selection bias in ordering tests for adrenal suppression when clinically suspected, if these were the only diagnosable cases out of all the entire RPDR, including the other 74.3 percent of asthmatics using ICS who were not tested, this would still suggest that eight percent of all asthmatics using ICS have cortisol levels low enough to classify as adrenal insu ciency diagnosis.

Page 7/17
The degree of pronounced steroid suppression upon extended ICS use among asthma cases suggests that proactive monitoring of cortisol levels has the potential to identify and prevent adrenal suppression through altered treatment approaches. This becomes even more imperative, as cases of adrenal suppression are often problematic to diagnose, symptoms can be missed due to the range of presentation of the disease, and the broad range of negative side effects can even result in lifethreatening complications that are critical to avoid 27,[38][39][40][41][42] . The institution of regular cortisol testing enables the identi cation of marked decreases in steroid levels prior to signi cant and potentially permanent long-term complications from clinical-grade adrenal suppression. Moreover, yearly monitoring is inexpensive (adrenocorticotropic hormone test (ACTH) simulation test) 43 and manageable in most primary care settings and may decrease the overall public health burden of this ICS adverse effect. Individual optimization of treatment protocols does not necessarily imply omitting ICS use in all cases, but potentially changing the dosage, the frequency, or supplementation with additional medications may result in more e cacious outcomes in some individuals, while preventing adverse adrenal suppression rami cations in others.
While other omic data types are touted for their potential use in precision medicine, this study demonstrates the expediency to clinical translation that metabolomic pro ling offers. Because a large number of clinical tests currently in use are measured metabolites, there are often cases where the metabolites of interest from untargeted analyses are already measured in clinical practice. This was the case in the present study, in which the observed reduction of multiple steroid metabolites led to the exploration of the use of cortisol testing to assess adrenal suppression in asthmatics on ICS. This approach enables investigators to more e ciently translate metabolomic ndings into clinical practice than for other omic data types that would require substantial follow-up time and resources.
Despite the strengths of the reported ndings, several limitations should be noted. First, our discovery cohort (EPIC-Norfolk) did not have information on ICS use. In MGBB, we created a robust algorithm to identify ICS use. We acknowledge that there is likely misclassi cation; however, this misclassi cation would result in a bias towards the null hypothesis and the effects we identi ed would remain highly signi cant. Therefore, we do not believe that this impacted the overall conclusions of the study. Moreover, we veri ed our ndings in an RCT, which represents a more robust study design. Second, the CAMP RCT utilized children and while at the end of the trial, they were mostly adolescents; this differs from the other cohorts that utilized primarily adults. It is important to note that there may be important differences between ICS response in adults and adolescents that should be studied in more detail. Third, metabolomic pro ling was performed in a different laboratory for CAMP and did not have the broader range of steroid metabolites. Despite these differences, we were able to validate and further re ne our ndings over the four populations we utilized. Finally, while it is important to realize the potential of large EMR databases, it is equally important to recognize that this information is derived from an overrepresentation of individuals with illness and may bias the data or result in confounding by indication. Acknowledging this limitation, we excluded individuals with common relevant comorbid conditions, such as COPD.
In conclusion, our results suggest regular monitoring of steroid levels among asthma cases with longterm ICS use, which is not currently commonplace, is merited to identify the optimal clinical regimen for individuals with asthma at risk of serious adrenal suppression and determine the clinical impact of low cortisol levels in this population. This could potentially improve overall health and reduce health care spending. Integrating metabolomics data from epidemiological studies with existing clinical biomarkers obtained via EMRs may enhance the interpretation of metabolomic data as it relates to health and current medical practices, in addition to enhancing comprehension both on the side of researchers and of clinicians.

Statistical Analysis
Overview of Analytic Approach A detailed summary of the four cohorts utilized in this study, including cohort generation, details on asthma diagnosis and ICS use, metabolomic pro ling and quality control procedures is provided in the supplement (Methods & Table S1-S2). An illustration of our analytic strategy is presented in Figure 1. Brie y, we utilized a discovery and replication approach to identify metabolites associated with prevalent asthma, using the EPIC-Norfolk (discovery) and biobank-derived Mass General Brigham Biobank-Asthma (MGBB-Asthma) (replication) cohorts. As these ndings implicated potential involvement of steroidassociated metabolites, we then assessed the impact of ICS use on the replicated asthma-associated metabolites within MGBB-Asthma cohort and further evaluated the impact of ICS use on steroid metabolites using cortisol and cortisone measures from the four-year double-blind longitudinal ICS randomized controlled trial (RCT), Childhood Asthma Management Program (CAMP) 44

Discovery and Replication Analyses for Metabolite-Asthma Associations
Multivariable logistic regression models were employed in EPIC-Norfolk to assess the association between log-transformed plasma concentrations of each metabolite with asthma affection status. Models were adjusted for age, sex, body mass index (BMI) and smoking status (current, former and never). We did not adjust for ethnicity, as the EPIC-Norfolk population is mostly White (99.7%). To correct for multiple testing, we applied a stringent Bonferroni signi cance threshold (0.05/number of statistical tests in EPIC-Norfolk). In MGBB-Asthma, multivariable logistic regression models adjusted for age, sex, race, BMI and smoking status were used to replicate the associations between the Bonferroni signi cant metabolites and asthma case status. An association was considered "replicated" if: 1) The effect estimate (Odds Ratio) is in the same direction as the initial association nding and 2) The False Discovery Rate (FDR) 47 is <5%. To quantify the relative reduction in steroid metabolite levels based on ICS use, asthmatics were strati ed into four sub-groups: 1) no asthma/no ICS use; 2) no asthma/ICS use; 3) asthma/no ICS use; 4) asthma/ICS use. Multivariable logistic regression models using pairwise comparisons adjusted for age, gender, race, and BMI were utilized to compare metabolite levels between groups. All models were adjusted for age, gender, race, and BMI.
Establishment of a temporal relationship between cortisone, cortisol, ICS using the CAMP RCT We utilized multivariable linear regression models and individuals from CAMP to further assess the relationship between cortisol and cortisone levels in children randomized to ICS (budesonide) verses those randomized to nedocromil or placebo. Considering both baseline and the end of the four-year clinical trial, the models were adjusted for age, gender, race, BMI, and an interaction variable between age and randomized ICS-use for the end of the trial model, as the children were in various stages of puberty and puberty directly in uences steroid levels.

Clinical Quanti cation of Cortisol: MGBB-Cortisol
We investigated minimum cortisol levels (mcg/dL) recorded in MGBB-Cortisol subjects throughout a 24hour period stratifying on asthma status and ICS use. To account for differences in sample availability across the 24-hour period, subjects were binned into three time categories based on sample collection time: 4:00am-12:00pm, 12:00pm-6:00pm and 6:00pm-4:00am. The mean cortisol levels were subjected to smoothing interpolation using loess curve regression tting. Tukey's HSD 48 test was used to identify signi cant differences between the asthma/ICS subgroupsPairwise comparisons between the subgroups were also performed using generalized linear models, adjusted for collection time, age, gender and race.
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Requests for raw data, analyzed data and materials will be reviewed by the cohort and contact PIs for the studies to determine if the request is subject to intellectual property or con dentiality obligations. Data and materials that can be shared will be released using a Material Transfer Agreement. Appropriate IRB approvals may be required to access de-identi ed data in particular data from electronic medical health records. The EPIC-Norfolk data can be requested by bona de researchers for speci ed scienti c purposes via the study website (https://www.mrc-epid.cam.ac.uk/research/studies/epic-norfolk/). Data will either be shared through an institutional data sharing agreement or arrangements will be made for analyses to be conducted remotely without the need for data transfer.

Code availability
The R Code for data processing and analyses can be made available for research purposes upon request from the corresponding author.

Declarations
Author Contributions: PK and IDS had full access to the data and take responsibility for the data integrity and accuracy of the analysis. JALS and CL contributed to conceptualization of the study; PK performed the quality control and statistical downstream data analyses for Con ict of Interest disclosures: There are no con icts of interests to declare.
Funding/Support and acknowledgements: Effort from PK, JALS and STW is supported by Role of the Funder/Sponsor: The external funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.  Metabolite names with an asterisk * indicate that the metabolite has not been confirmed based on an analytical standard, but we are confident in its identity as confirmed by Metabolon