Study design, setting, participants and data sources
The Manchester Asthma and Allergy Study (MAAS)40 is a population-based birth cohort from the greater Manchester area, consisting of a mixed urban-rural population within 50 square miles of South Manchester and Cheshire, United Kingdom located within the maternity catchment area of Wythenshawe and Stepping Hill Hospitals. Validated questionnaires were interviewer-administered to collect information on parentally-reported symptoms and physician-diagnosed illnesses. We assessed allergic sensitisation by skin prick tests (SPT). The study was approved by the South Manchester Local Research Ethics Committee and parents gave written informed consent.
Screening and recruitment
All pregnant women were screened for eligibility at antenatal visits (8-10th week of pregnancy) between October 1st 1995 and July 1st 1997. Of the 1499 couples who met the inclusion criteria (≤ 10 weeks of pregnancy, maternal age ≥ 18 years, and questionnaire and skin prick data test available for both parents), 288 declined to take part in the study and 27 were lost to follow-up between recruitment and the birth of a child. A total of 1184 children born into the study had at least some evaluable data.
Sensitisation was ascertained by SPT at ages 1, 3, 5, 8, 11 and 16 years for 7 allergens (Dermatophagoides pteronyssinus, cat, dog, grass pollen, mixed moulds, milk, and egg [Bayer, Elkahrt, Ind, US]). From age 8 years, SPTs were additionally performed for tree pollen (birch) and peanut (total of 9 allergens tests). We defined sensitisation as a mean wheal diameter 3 mm larger than that elicited by the negative control to at least 1 of the allergens tested.
Children were followed prospectively, and attended review clinics at ages 1, 3, 5, 8, 11, and 16 years of age. At age 1 year, only children with either both atopic parents, or no atopic parents who lived in homes without a pet were invited to attend clinical follow up. At all other time points for all other measures all children were invited to participate.
Sensitisation was defined as having a positive skin prick test to any of the above-mentioned allergens at ages 1, 3, 5, 8, 11 and 16 years of age.
Asthma was defined by fulfilling at least two out of three following criteria: current wheeze, current use of asthma medication, or physician-diagnosed asthma ever at ages 1, 3, 5, 8, 11 and 16 years of age.
Outcomes derived using machine learning and data-driven methodologies
In this study population, we have previously described clusters of allergic sensitisation (using SPTs and IgE to whole allergen extracts41 and component-resolved diagnostics (CRD)42), allergic diseases (eczema, wheeze and rhinitis)43 and asthma exacerbations.44 In the current analysis, we used the following multivariable outcomes
Atopy clusters 41 : (1) non-dust mite atopic vulnerability, (2) dust mite atopic vulnerability, (3) multiple late atopic vulnerability, (4) multiple early atopic vulnerability and (5) no latent atopic vulnerability.
Atopic march clusters 43 : (1) no disease, (2) atopic march, (3) persistent eczema and wheeze, (4) persistent eczema with later-onset rhinitis, (5) persistent wheeze with later-onset rhinitis, (6) transient wheeze, (7) eczema only and (8) rhinitis only.
IgE clusters 42 : (1) Multiple sensitisation, (2) predominantly house dust mite sensitisation, (3) predominantly grass and tree sensitisation and (4) lower-grade sensitisation.
Exacerbations clusters 44 : represented by the following clusters at age 8; (1) no wheeze, (2) wheeze no exacerbations, (3) infrequent exacerbations and (4) early onset frequent exacerbations.
Quantification of chemokines in plasma samples
A total of 905 plasma samples originating from the time of birth (cord blood, n = 376), 1 year (n = 195) and 8 years (n = 334), were analysed for their chemokine content (Figure S1). Circulating plasma levels of CCL18 were measured using an in-house DuoSet ELISA kit (R&D Systems) and an in-house multiplex bead assay was setup for analysis of circulating CCL17, CCL22, CXCL10 and CXCL11, as described in detail below.
Due to its high concentration in the circulation, the CCL18 chemokine was measured with a separate DuoSet ELISA kit (DY394, R&D systems). On the first day, Corning® 96 Well Half Area Clear Flat Bottom Polystyrene High Bind Microplates (Corning Life Sciences, Kennebunk, ME, USA) were coated with 4 µg/ml monoclonal anti-human CCL18/PARC antibody (MAB 394, clone: 65407 R&D systems, Minneapolis, MN, USA) diluted in carbonate-bicarbonate buffer (28 mM Na2CO3 72 mM NaHCO3, pH 9.6, Sigma Aldrich). The plate was shaken for 1 h at RT, and thereafter left in RT without shaking overnight. The following day, the plate was washed 4 times in PBS-T (Medicago, Uppsala, Sweden), and subsequently blocked by adding skimmed cow’s milk diluted in PBS (Medicago, Uppsala, Sweden). One hour of incubation on a plate shaker at RT followed. Next, a seven-point standard curve, diluted in steps of 1:2, was prepared (125 − 1.9 pg/ml) in PBS + 1% BSA (Probumin, Merck Millipore, Darmstadt, Germany). After washing the plate, standards, samples and blanks were added in duplicates to the plate, which was followed by a one-hour incubation at RT while shaking. Washing was thereafter followed by addition of 200 ng/ml biotinylated anti-human CCL18/PARC antibody (polyclonal goat IgG, BAF394, R&D systems Minneapolis, MN, USA) diluted in HPE buffer (High Performance ELISA buffer, Sanquin Plesmalaan, The Netherlands), and one hour incubation at RT while shaking. Upon washing the plate, streptavidin-poly-horse radish peroxidase-conjugate (SA-poly-HRP, Sanquin Plesmalaan, The Netherlands) was added and let incubate for 30 min at RT while shaking in the dark. Another wash step later, the substrate for SA-poly HRP, TMB (3,3’,5,5’-Tetramethylbenzidine, Sigma Aldrich, St. Louis, MO, USA) was added to the plate and 30 min of shaking incubation at RT in the dark followed. The reaction was stopped using 1.8 M H2SO4, and the plate was read at 450 nm in the microplate absorbance reader (Tecan Sunrise, Austria) and subsequently analysed. The standard curve was calculated by means of five parametric statistics. In general, data from duplicate samples were considered reliable when the CV was below 15%. However, in total 4 samples (2 samples at birth and 2 samples at 8 years of age) were included in further analyses despite their higher CV (all below 17%). The detection limit was 3.9 pg/ml, and the inter-assay variation was below 30%.
Multiplex Luminex bead assay (CCL17, CCL22, CXCL10 and CXCL11)
An in-house multiplex bead assay was setup for CCL17, CCL22, CXCL10 and CXCL11 for analysis in plasma samples. In preparation for the in house multiplex bead assay, four bead sets were coupled with 5 µg capture antibody (Clones; CCL17: 54026, CCL22: 57226 and CXCL11: 87328 (R&D Systems, Abingdon, UK) and CXCL11: 4D5/A7/C5, BD Pharmingen, NJ, USA) per million beads for the respective chemokine of interest. The coupling procedure was validated by running standard curves with some control samples. For the main analyses, 1.2 µm pore-size filter plates Millipore multiscreen, Millipore corporation, Bedford, MA, USA) were moistened by addition of PBS + 1% BSA (Probumin, Merck Millipore, Darmstadt, Germany), while shaking for a while at RT. Dilution of the seven point standard curves (in steps of 1:3) for the respective chemokines; 1400 − 1.9 pg/ml for CXCL10 (Cat.no: 266-IP), CCL17 (Cat.no: 364-DN) and CCL22 (Cat.no: 336-MD), and 4000 − 5.5 pg/ml for CXCL11 (Cat.no: 672-IT) followed (all standards: R&D Systems, Abingdon, UK). Upon removal of the buffer from the plate using vacuum manifold, standards, samples and blanks were added to the plates in duplicates. Thereafter, a mixture of beads diluted to a concentration of 2000 beads per bead set and well, was added to the plate. After one hour of incubation on a plate shaker in RT, the plate was shaken over night at 4 °C. The following day, a mixture of the detection antibodies for CCL17 (500 ng/ml, Cat.no: BAF364, R&D systems), CCL22 (200 ng/ml, Cat.no: BAF336 R&D systems), CXCL11 (500 ng/ml, Cat.no: BAF672, R&D systems) and CXCL10 (1000 ng/ml, clone: 6D4/D6/G2 BD Pharmingen, NJ, USA), was added to the plate upon washing and removal of fluid by vacuum manifold. The plate was incubated for 1 h in the dark, while shaking at RT. Upon washing and removal of the fluid, SA-PE (Life Technologies, Oregon, US) was added to the plate to a concentration of 1 µg/ml, and a 30 min incubation at RT while shaking followed. Prior to analysis of the plate in a Luminex 200 instrument (Luminex Corporation, Austin, Texas, US), the plate was washed again, and the beads were resuspended in 75 µl PBS + 1% BSA. The acquired data were evaluated utilizing the Masterplex software (Version 126.96.36.199, Hitachi, South San Francisco, CA, US). Standard curves were calculated by means of five parametric statistics. The standard curve from the Th1-associated chemokine CXCL11 was weighted (1/y) for all plates, as to increase the resolution at the lower end of the curve when retrieving the sample data from the Luminex analyses. Data from duplicate samples were considered reliable when the CV was below 15%. For CXCL11, three samples were included in further analyses despite their higher CV (a CV of less than 16%, two samples at 1 year and one sample at 8 years of age). The detection limits of the chemokines were as follows; CCL17: 1.0 pg/ml, CCL22: 1.9 pg/ml, CXCL10: 5.8 pg/ml and CXCL11: 16.5 pg/ml. Undetectable samples were given half cut-off levels, represented by half of the previously mentioned detection limits. Inter-assay variation was below 30%.
Descriptive analyses of demographic variables and allergy outcomes (at ages 1, 3, 5, 8, 11 and 16) in the children of the sub-study, compared to the excluded children or the entire MAAS cohort, were performed using a Chi-square test or Student’s t-test, for discrete and continuous variables, respectively (Tables S1-S3).
To assess the normality of the chemokine data, Shapiro-Wilk’s test was applied to the measurements. As most of the chemokine concentrations were non-normally distributed, univariate comparisons of chemokine levels at the different ages were performed using the non-parametric Kruskal-Wallis test with Dunn’s post hoc test for multiple comparisons, whereas Mann-Whitney U tests were used for cross-sectional comparisons. Correlations between the chemokines were assessed using the Spearman’s rank correlation.
The chemokine levels were thereafter assessed in relation to asthma and sensitisation outcomes at ages 1, 3, 5, 8, 11 and 16 years of age. Furthermore, CCL18 levels at all ages were related to previously machine learning derived clusters, more specifically sensitisation clusters41, atopic diseases clusters43, CRD IgE clusters42 and exacerbations clusters.44
Binomial logistic regression analyses were performed on the natural log transformed chemokine data in relation to binary sensitisation and asthma variables, to study whether chemokine levels at birth, 1 year or 8 years of age predicted outcomes in terms of sensitisation or asthma at ages 8, 11 and 16 years of age, respectively. All chemokines were included at each age, as significant outcomes were consistent regardless if the models included one chemokine or all chemokines at once (results not shown). The models were adjusted for parental atopy at recruitment, parental smoking at recruitment and sex. These variables were chosen on the basis of differences seen in the demographic data and previous knowledge from the cohort.45,46 As not to over-fit the models, we chose to merge the maternal and paternal variables on atopy and asthma into one parental variable each. We did not correct for pet ownership separately or as a merged variable, as only cat-ownership was significantly different, and as not to over-fit the models. Coefficients represent the increased/decreased odds of the respective outcome per log-unit increase in chemokine levels.
The longitudinal analyses were performed by using generalised estimating equations (GEE), which takes into account the correlations reported for the chemokines. Population-averaged GEE models were developed to investigate whether the effect of natural log-transformed chemokine levels on the development of asthma (ages 5–16 and 8–16) or sensitisation (ages 3–16 and 8–16) changed over time. This was investigated in different time intervals to study the role of the chemokines both contemporaneously as well as later in life. For asthma, the interval 5–16 years was chosen, as we excluded the development of pre-school wheeze as a surrogate for asthma development. For sensitisation, we chose to use the interval 3–16 years of age, as the prevalence of sensitisation at age 1 was generally lower than later in life and the sample size was smaller than at the other ages. For both outcomes, the 8–16 years interval was chosen, as chemokine measurements at age 8 were studied. The models were adjusted for parental atopy at recruitment, parental smoking at recruitment and sex. Coefficients represent the increased/decreased odds of sensitisation per log-unit increase in chemokine levels.
Analyses were conducted in GraphPad Prism 847, IBM SPSS Statistics version 2548, and Stata 15 software.49