Metabolomics to identify omalizumab responders among children with severe asthma: A prospective study

To the Editor,

In this study, we aimed to assess the potential value of baseline urinary metabolomic profiles in predicting the response to omalizumab in children with severe asthma. Although most children show a good response to this drug, ¹ 10–15% are partial or non‐responders. ² Identifying the patients most likely to benefit from omalizumab is crucial to personalise the treatment and optimise cost‐effectiveness. So far, clinical or biochemical predictors of response to omalizumab have not been definitively recognised. Metabolomic analysis, using an untargeted approach, has the potential to enable the identification of metabolic features associated with relevant clinical outcomes. ³

In this multicentre prospective study (Ethics Committee approval 4329/AO/17), after obtaining informed consent, we consecutively recruited patients 6–17 years old with severe asthma participating in a national database of the Italian Society for Paediatric Respiratory Diseases, who were candidates for omalizumab. Children had been treated for at least 4 months with a daily ICS dose ≥500 mcg of fluticasone propionate or equivalent. A urine sample was collected before starting omalizumab. After 52 weeks of follow‐up, or sooner if omalizumab was stopped before, children were classified as responders or not, using a multidimensional approach. ⁴ We considered the variation from baseline of GINA score (which evaluates asthma control in the last 4 weeks) and CASI ⁵ score (which evaluates control in the last 2 weeks together with lung function, exacerbations and treatment). We defined as responders children with at least a 1‐point reduction in CASI score ⁶ and controlled or partially controlled according to GINA. Since QoL is significantly affected in severe asthma, ⁷ in case of inconsistency, we considered as responders those with an increase of at least 0.5 points ⁸ in Pediatric Asthma Quality of Life Questionnaire (PAQLQ) score. ⁹

Urine analysis was performed through high‐resolution mass‐spectrometry (Q‐ToF Synapt G2, Waters, Milford, USA) interfaced with the chromatography Acquity UPLC using the reverse phase column HSS T3 (Waters, Milford, USA)

Children's baseline characteristics were compared by t‐test, Mann–Whitney test and Fisher's exact test; in case of differences in their distribution, sub‐sampling was performed to match responders and non‐responders (a standard procedure in metabolomic studies to avoid false discoveries due to experimental design bias).

For metabolomic analysis, a two‐group comparison and a one‐class modelling approach were applied. The former assumes that responders and non‐responders belong to two different groups, each metabolically well‐defined, and it was based on univariate methods (Mood's median test controlling the false discovery rate by Storey method, adjusted p < .15). The one‐class approach assumes that responders belong to a metabolically homogeneous group while non‐responders are scattered around it, and it was based on multivariate (Principal Component Analysis) and univariate methods for outlier detection. With the univariate approach, the distribution of each metabolic variable in responders was modelled by kernel density estimation; the probability of each non‐responder being an outlier for that variable was estimated and false discovery rate was controlled by Storey method (adjusted p < .15).

The relevant variables were annotated searching our database of commercial standards, METLIN database and Human Metabolome Database.

52 children were included in the analysis (10 non‐responders) (Table 1).

TABLE 1.

Children's characteristics, clinical variables and identified metabolites

Baseline characteristics	Responders (n = 42)	Non‐responders (n = 10)
Males (n, %)	23 (55%)	4 (40%)
Age (mean, SD)	12.2 (2.7)	10.9 (2.3)
BMI (mean, SD)	20.7 (4.2)	19.8 (2.9)
Total IgE (IU/ml) (mean, SD)	701.5 (516.3)	494.1 (472.8)
Allergic comorbidities a (n, %)	26 (62%)	5 (50%)
Sensitised to perennial allergens (n, %)	41 (98%)	8 (80%)
Parental smoking (n, %)	29 (69%)	7 (70%)
Parental asthma (n, %)	19 (45%)	4 (40%)

Clinical variables	Baseline	End of follow‐up b	Baseline	End of follow‐up b
Number of steroid courses in the previous 12 months (mean, SD)	4.2 (4.2)	0.7 (1.2)	4.5 (3.3)	3.4 (4.0)
LABA (n, %)	40 (95%)	32 (76%)	9 (90%)	9 (90%)
Montelukast (n, %)	29 (69%)	9 (21%)	4 (40%)	4 (40%)
FEV1 (% pred) (mean, SD)	91 (19)	96 (16)	91 (14)	85 (11)
GINA score U/P/C (n)	37/5/0	5/9/28	8/2/0	8/2/0
CASI score (mean, SD)	9.8 (3.8)	4.7 (1.7) c	9.6 (4.1)	8.8 (4.6) c
PAQLQ (mean, SD)	4.7 (1.3)	6.5 (0.5) c	4.7 (1.0)	4.1 (0.7) c

Identified metabolites at baseline	Responders (n = 42) Median [IQR]	Non‐responders (n = 10) Median [IQR]
L‐Histidine (178.0590 m/z; RT 0.523) [ID: POS413]	0.23 [0.20–0.34]	0.13 [0.10–0.19] d
Uric acid (169.0361 m/z; RT 0.827) [ID: POS1370]	0.22 [0.10–0.30]	0.42 [0.35–0.58] d
L‐Kynurenine (209.0930 m/z; RT 1.885) [ID: POS 2148]	0.05 [0.03–0.13]	0.018 [0.007–0.037] d
3‐Dimethylallyl‐4‐hydroxyphenylpyruvate (249.1139 m/z; RT 5.582) [ID: POS7858]	0.12 [0.10–0.15]	0.18 [0.16–0.22] e
Aspartylglycosamine (316.1141 m/z; RT 0.677) [ID: NEG3634]	0.28 [0.18–0.35]	0.13 [0.07–0.22] e
Aspartyl‐Threonine (215.0661 m/z; RT 0.789) [ID: NEG3420]	0.21 [0.15–0.27]	0.12 [0.02–0.15] e

Abbreviations: BMI, body mass index; GINA, Global Initiative for Asthma; U, uncontrolled asthma; P, partially controlled asthma; C, Controlled Asthma; CASI, Composite Asthma Severity Index; RT, retention time; ID, variable identifier coded as [ionization mode][identifier].

^a Allergic rhinitis and/or atopic dermatitis and/or food allergy.

^b End of follow‐up: evaluation at 52 weeks or when omalizumab was stopped because of poor asthma control (5/10 non responders stopped omalizumab after 4 to 9 months).

^c Variation from baseline in responders versus non‐responders: p < .01 for CASI; p < .001 for PAQLQ.

^d Responders versus non‐responders p = .01.

^e Responders versus non‐responders p < .001.

Metabolic features were differently expressed in responders and non‐responders.

The two‐class comparison approach highlighted 17 discriminating features (Figure 1); among these, six metabolites were identified (putative biomarkers), four higher in responders and two in non‐responders (Table 1).

FIGURE 1.

Boxplots showing the distributions of the features selected by Mood's test controlling the false discovery rate by Storey method (adjusted‐p < .15); green bars: non‐responders (NR); yellow bars: responders (R); variables are indicated as [ionization mode][ID][group]

Biomarkers increased in responders were dipeptides and amino acids, in keeping with previous studies that suggest an altered amino acid metabolism in asthma. ¹⁰ The better characterised were the histamine precursor L‐Histidine, a possible marker of an altered histamine pathway and L‐Kynurenine.

On the other hand, uric acid, a metabolite with a possible role in innate and adaptive T2 response amplification, ¹¹ was a putative biomarker in non‐responders.

Moreover, the one‐class modelling approach highlighted 100 features differently expressed in at least three non‐responders with respect to responders (annotated metabolites in Table S1) and the Q‐chart built by PCA was promising for non‐responder detection (Figure S1).

Details of methods and results are reported in Appendix S1.

In conclusion, we found that children with severe asthma responding to omalizumab showed a different metabolomic urinary profile at baseline compared to non‐responders. A metabotype enriched in amino acids was associated with a good response to omalizumab, while the uric acid metabolic pathway was involved in non‐responders.

For the first time, this study paves the way to a possible a‐priori identification of children who are most likely to benefit from omalizumab based on their metabolic arrangement. Further studies, also based on targeted approaches, may expand these results.

AKNOWLEDGEMENTS

We thank the IPSAN members: Antonio Capizzi and Maria Angela Tosca, Istituto Giannina Gaslini, Genova; Salvatore Cazzato, Salesi Children's Hospital, Ancona; Renato Cutrera, Bambino Gesù Children's Hospital, Roma; Lucia Leonardi, Sapienza University, Roma; Salvatore Leonardi, University of Catania; Gian Luigi Marseglia, University of Pavia; Francesca Mori, Meyer Children's University Hospital, Firenze; Maria Francesca Patria, University of Milano; Giorgio Piacentini and Laura Tenero, University of Verona; Giovanna Pisi, Cystic Fibrosis Center, University Hospital, Parma; Giampaolo Ricci, University of Bologna. Open Access Funding provided by Universita degli Studi di Padova within the CRUI‐CARE Agreement.

CONFLICT OF INTEREST

GF reports financial and non‐financial support from AstraZeneca, Chiesi, Boehringer Ingelheim, GSK, Novartis, Sanofi, outside this work: AV reports financial or non‐financial support from Chiesi, Fidia Sooft, Momento Medico, Lusofarmaco, outside this work; EL reports financial and non‐financial support from AbbVie, Angelini, Boehringer, Chiesi, GlaxoSmithKline, Lusofarmaco, Novartis, Restech, Sanofi, Vertex, outside this work; SZ reports financial support from Sanofi, outside this work; AdM reports financial support from GlaxoSmithKline and Sanofi, outside this work. SC, EdP, FR, SB, PP, GDC, VC, AL, GG, MS, report no conflict of interest for this manuscript.

Supporting information

Appendix S1