1. Computational and Systems Biology
Download icon

Trajectories of childhood immune development and respiratory health relevant to asthma and allergy

  1. Howard HF Tang  Is a corresponding author
  2. Shu Mei Teo
  3. Danielle CM Belgrave
  4. Michael D Evans
  5. Daniel J Jackson
  6. Marta Brozynska
  7. Merci MH Kusel
  8. Sebastian L Johnston
  9. James E Gern
  10. Robert F Lemanske
  11. Angela Simpson
  12. Adnan Custovic
  13. Peter D Sly
  14. Patrick G Holt
  15. Kathryn E Holt
  16. Michael Inouye  Is a corresponding author
  1. Baker Heart and Diabetes Institute, Australia
  2. The University of Melbourne, Australia
  3. University of Cambridge, United Kingdom
  4. Imperial College London, United Kingdom
  5. University of Wisconsin School of Medicine and Public Health, United States
  6. Telethon Kids Institute, University of Western Australia, Australia
  7. National Heart and Lung Institute, Imperial College London, United Kingdom
  8. The University of Manchester, United Kingdom
  9. The University of Queensland, Australia
  10. The London School of Hygiene and Tropical Medicine, United Kingdom
  11. The Alan Turing Institute, United Kingdom
Research Article
Cite this article as: eLife 2018;7:e35856 doi: 10.7554/eLife.35856
7 figures, 6 tables, 2 data sets and 4 additional files

Figures

Figure 1 with 3 supplements
Non-parametric mixture-model-based clustering of CAS dataset, based on 174 features.

SPT = skin prick test. White spaces within the heatmap indicate missing data. Rows represent individuals; columns represent clustering features with general categories as labelled on grey background. Variables with grey background are clustering features ordered by category or type of variable first (e.g. all HDM IgE-related variables grouped together), then by timepoint (earlier to later, from left to right). Variables with lilac background indicate resultant cluster membership and outcome variable (age-5 wheeze). Heatmap values are scaled relative to range and median values for each feature; the median is coloured beige-yellow, the median +range red, and median – range blue. For sex, −1/blue = female, 0/yellow (median) = male.

https://doi.org/10.7554/eLife.35856.003
Figure 1—figure supplement 1
Scatterplot of principal components analysis (PCA) of the complete-case CAS dataset (N = 186), with points coloured by npEM clusters Each point represents an individual.

The first two PCs (shown) account for 16.7% of the total variance.

https://doi.org/10.7554/eLife.35856.004
Figure 1—figure supplement 2
Silhouette widths of clusters generated by npEM.

j = cluster number; nj = cluster size; avei∈Cjsi = average silhouette width among members i of cluster Cj. Overall average silhouette width across all clusters is also given.

https://doi.org/10.7554/eLife.35856.005
Figure 1—figure supplement 3
Overview of study methodology.

Dashed arrows indicate non-critical elements of our method.

https://doi.org/10.7554/eLife.35856.006
Graphical summary of proposed clusters *‘Early’ specifically refers to ‘within the first 6 months of life’.
https://doi.org/10.7554/eLife.35856.007
Figure 3 with 1 supplement
Incidence of multiple phenotypes, including parent-reported wheeze.

(A) Physician-diagnosed asthma (B) defined wheeze phenotypes (C) in relation to food and inhalant sensitisation (D) stratified by cluster and time in the CAS dataset. Points indicate observed proportion; bars indicate 95% CI (binomial distribution). Wheeze phenotypes defined as: no wheeze = no wheeze at ages 1 to 3, or age 5; transient wheeze = any wheeze at ages 1 to 3, but not age 5; late wheeze = wheeze at age 5, but not ages 1 to 3; persistent wheeze = any wheeze at both ages 1 to 3 and age 5. Food sensitisation defined as peanut IgE ≥0.35 kU/L at any age, or cow’s milk, egg white, peanut SPT > 2 or 3 mm for age ≤2 or>2 respectively. Inhalant sensitisation defined as HDM, cat, couchgrass, ryegrass, mould or Phadiatop IgE ≥0.35 kU/L at any age, or mould SPT (Alternaria or Aspergillus spp.)>2 or 3 mm for age ≤2 or>2, respectively.

https://doi.org/10.7554/eLife.35856.008
Figure 3—figure supplement 1
Relationship of clusters to food sensitisation, eczema and wheeze.

Percentages denote proportion of cluster displaying phenotype (numbers in brackets denote actual sample numbers). Food sensitisation defined as peanut IgE ≥0.35 kU/L at any age, or cow’s milk, egg white, peanut SPT > 2 or 3 mm for age ≤2 or>2, respectively. Subphenotypes defined for food sensitisation, eczema and wheeze as: no phenotype = phenotype absent at all ages; transient = any incidence of phenotype at the earlier ages (1 to 3 for wheeze, 6 m to three for eczema, 6 m to two for sensitisation), but not age 5; late = phenotype at age 5, but not the earlier ages; persistent = any incidence of phenotype at both earlier ages and age 5.

https://doi.org/10.7554/eLife.35856.009
Figure 4 with 2 supplements
HDM IgE (A), IgG (B) and IgG4 (C); and peanut IgE (D) and IgG4 (E) stratified by cluster and time, in the CAS dataset Points indicate means; bars indicate 95% CI (t-distribution).
https://doi.org/10.7554/eLife.35856.010
Figure 4—figure supplement 1
Correlation patterns between IgE vs IgG4 (A) and IgE vs IgG (B) at age five *p<0.05 for Spearman correlation with Holm correction for multiple testing.

Note the slightly stronger heat along the main diagonals of both heatmaps, especially for HDM.

https://doi.org/10.7554/eLife.35856.011
Figure 4—figure supplement 2
Distinct biological signals of HDM IgE, IgG4, SPT, and Th2 cytokine (IL-13).

(A) Th2 cytokine (IL-13) to HDM stimulation of PBMCs in vitro, vs.HDM IgE responses in vivo, stratified by age of testing, and SPT result (positive denoted by ≥2 mm at age <2 or 3 mm at age 5). Dotted line represents traditional threshold for HDM IgE-positive result (0.35 kU/L). Note the significant number of individuals on either side of the dotted line for both HDM SPT-positive and negative subgroups. (B) Th2 cytokine (IL-13) to HDM stimulation of PBMCs in vitro, vs. HDM IgG4 responses in vivo, stratified by age of testing and HDM IgE result (positive denoted by ≥0.35 kU/L).

https://doi.org/10.7554/eLife.35856.012
Figure 5 with 4 supplements
LRI frequency (A), wheezy LRI (wLRI) frequency (B), and HDM IgE (C), stratified by age-5 wheeze status, cluster and time, in the CAS dataset.

Points indicate means; bars indicate 95% CI (t-distribution). #p<0.05 for repeated-measures ANOVA across timepoints from the first 3 years of life (see Table 4). *p<0.05 for Mann-Whitney-Wilcoxon comparison within each timepoint.

https://doi.org/10.7554/eLife.35856.015
Figure 5—figure supplement 1
A ‘simple’ decision tree generated by recursive partitioning from CAS data, with breakdown of tree clusters by actual CAS npEM-derived clusters.

(A); scatterplot showing separation of CAS clusters by decision split thresholds (B). Percentages in Panel A may not sum up to 100% because some individuals have missing values for decision node variables, hence making them impossible to classify. In Panel B, note that left-most column of points represent values of HDM IgG4 that were less than the limit-of-detection (LOD) for that assay (0.0003 μg/L) and were subsequently assigned to half the LOD (0.00015 μg/L). Most of these points belonged to individuals from CAS1.

https://doi.org/10.7554/eLife.35856.016
Figure 5—figure supplement 2
Decision tree generated by recursive partitioning from CAS data, excluding Phadiatop assay variables.
https://doi.org/10.7554/eLife.35856.017
Figure 5—figure supplement 3
A ‘comprehensive’ decision tree generated by recursive partitioning from CAS data, given CAS npEM-derived clusters and age-5 wheezing status.

CAS1.x, 2.x, and 3.x, and tree cluster 1.x, 2.x and 3.x, refer to the intersection of npEM-generated clusters and age-5 wheeze status, and their analogous decision tree cluster, respectively. The second digit (x.0 or x.1) refers to age-5 wheeze status, with ‘1’=present wheeze and ‘0’=no wheeze. Boxes, bars and digits with red outline indicate those with predicted (tree cluster) or actual (CAS npEM cluster) age-5 wheeze. Note that the tree did not predict for non-wheezing CAS3, so there is no tree cluster 3.0, and all CAS3 individuals were automatically assigned to a wheezing tree cluster.

https://doi.org/10.7554/eLife.35856.018
Figure 5—figure supplement 4
Comparison of predictors for age-5 wheeze in CAS and COAST clusters.

Colour coding and numbers in cells indicate odds ratio (OR) of predictor for age-5 wheeze in GLM, with sex, maternal and paternal history of asthma, and (for CAS) BMI as covariates. Non-grey cell with number indicates statistically significant association (p<0.05). Grey cell with number indicates non-significant (p>0.05); grey non-numbered cell indicates test not done due to lack of data.

https://doi.org/10.7554/eLife.35856.019
PBMC expression of IL-5.

(A) and IL-4 mRNA (B), as well as IL-13 protein (C), in response to stimulation HDM, stratified by cluster and time (CAS) Cord = cord blood sample collected at birth.Points indicate means; bars indicate 95% CI (t-distribution).

https://doi.org/10.7554/eLife.35856.022
Description of npEM-derived clusters in external cohorts: in MAAS, incidence of wheeze.

(A), asthma diagnosis (B), and HDM IgE levels (C); in COAST, incidence of asthma diagnosis (D), proportion of individuals with detectable aeroallergen-specific IgE levels (E), and PBMC protein expression of IL-13 following HDM stimulation above unstimulated control (F) MAAS cohort (N = 934) was classified using npEM model from CAS, into MAAS1 (N = 199, 21%), MAAS2 (N = 692, 74%) and MAAS3 (N = 43, 5%); these correspond to CAS clusters CAS1, 2 and 3, respectively. COAST cohort (N = 285) was similarly classified into COAST1 (N = 105, 37%), COAST2 (N = 151, 53%) and COAST3 (N = 29, 10%).

https://doi.org/10.7554/eLife.35856.023

Tables

Table 1
Comparison of selected demographic and clinical variables in CAS clusters
https://doi.org/10.7554/eLife.35856.013
VariableAge (y)Cas1 (N = 88)Cas2 (N = 107)Cas3 (N = 22)P-value (unadjusted)Feature?
Prop. (95% CI)Prop. (95% CI)Prop. (95% CI)OverallCas1 vs. 2Cas1 vs. 3Cas2 vs. 3
Sex = male55% (44–65%)51% (42–61%)86% (71–100%)7.3E-030.676.8E-033.7E-03Yes
Maternal asthma51% (40–62%)41% (32–51%)59% (37–81%)0.190.190.630.16Yes
Paternal asthma22% (13–30%)44% (35–54%)23% (3.7–42%)2.2E-031.3E-0310.093Yes
Wheeze133% (23–43%)30% (21–39%)55% (32–77%)0.0920.760.0840.046No
525% (15–35%)21% (13–30%)76% (56–96%)7.1E-060.592.6E-053.4E-06No
1012% (3.4–21%)18% (8.4–27%)50% (24–76%)3.1E-030.461.5E-030.011No
Asthma515% (7–23%)13% (5.9–20%)52% (29–76%)4.1E-040.837.7E-042.1E-04No
1010% (2.3–18%)15% (6.1–23%)56% (30–81%)2.6E-040.591.8E-047.9E-04No
Eczema6m39% (28–49%)45% (35–54%)91% (78–100%)2.4E-050.477.9E-069.0E-05Yes
134% (24–44%)30% (21–39%)82% (64–99%)2.5E-050.547.2E-051.4E-05Yes
528% (18–37%)24% (16–33%)71% (50–92%)2.1E-040.733.3E-047.9E-05No
Atopic rhinoconjunctivitis530% (20–40%)39% (29–49%)76% (56–96%)6.4E-040.212.7E-043.2E-03No
Mean (95% CI)Mean (95% CI)Mean (95% CI)OverallCas1 vs. 2Cas1 vs. 3Cas2 vs. 3
BMI (kg/m2)316 (16–17)16 (16–17)16 (16–17)0.860.650.680.8No*
416 (16–17)16 (16–16)17 (16–17)0.590.760.320.39No
516 (16–16)16 (16–16)16 (15–17)0.710.560.480.67No
1018 (17–19)18 (17–18)18 (17–19)0.890.7510.62No
Number of older siblings00.93 (0.72–1.1)0.53 (0.38–0.69)0.77 (0.32–1.2)4.5E-031.0E-030.370.25Yes
20.85 (0.66–1)0.5 (0.34–0.65)0.77 (0.32–1.2)2.8E-036.5E-040.480.16Yes
50.68 (0.5–0.85)0.39 (0.25–0.54)0.67 (0.23–1.1)0.0165.1E-030.750.12No
Geom. mean (95% CI)Geom. mean (95% CI)Geom. mean (95% CI)OverallCas1 vs. 2Cas1 vs. 3Cas2 vs. 3
Vitamin D (nmol/L)160 (55–64)59 (55–63)59 (52–67)0.930.980.760.7No
257 (54–61)58 (55–61)47 (40–55)0.0120.825.4E-034.4E-03No
589 (83–95)84 (79–89)77 (69–84)0.0570.460.0160.056No
  1. BMI = body mass index; feature?=whether variable was used as a clustering feature or not; geom. mean = geometric mean; prop. = proportion. For categorical variables, associations were tested using Fisher exact test; for continuous variables, Kruskal-Wallis and Mann-Whitney-Wilcoxon. Bold text indicates statistical significance (p<0.05); italics indicate near-significance (p<0.10). *Not used as clustering feature, as BMI is a derived variable. Height and weight at age three were used instead.

Table 2
Comparison of HDM-associated immunological variables in CAS clusters
https://doi.org/10.7554/eLife.35856.014
VariableAgeCas1
(N = 88)
Cas2
(N = 107)
Cas3
(N = 22)
P-value (unadjusted)Feature?
Geom.
mean (95% CI)
Geom.
mean (95% CI)
Geom.
mean (95% CI)
OverallCas1 vs. 2Cas1 vs. 3Cas2 vs. 3
Total antibody
IgE (kU/L)6m1.2
(0.69–2)
2.2
(1.4–3.6)
21
(12–35)
1.2E-070.0446.7E-082.2E-06Yes
10.6
(0.29–1.3)
2
(1.1–3.7)
43
(17–109)
2.0E-090.0194.3E-095.3E-08Yes
26.6
(3.5–12)
17
(12–25)
187
(131–267)
1.2E-110.0444.2E-111.4E-10Yes
535
(23–55)
60
(46–80)
451
(278–731)
2.2E-080.0961.9E-081.5E-07No
1085
(46–154)
150
(103–217)
800
(405–1.6E + 03)
1.4E-040.111.3E-042.8E-04No
HDM antibody
IgE (kU/L)6m0.018
(0.016–0.02)
0.019
(0.016–0.022)
0.033
(0.019–0.059)
1.9E-030.477.9E-044.2E-03Yes
10.019
(0.017–0.023)
0.019
(0.016–0.022)
0.26
(0.075–0.93)
1.3E-090.472.5E-074.5E-09Yes
20.024
(0.019–0.031)
0.042
(0.029–0.06)
7.1
(2.7–19)
2.6E-160.0782.5E-153.5E-13Yes
50.072
(0.041–0.13)
0.23
(0.12–0.45)
31
(7.8–127)
4.2E-090.0153.8E-095.1E-07No
100.37
(0.17–0.8)
1.3
(0.51–3.4)
52
(19–144)
2.9E-060.0685.7E-079.7E-05No
IgG (mg/L)10.21
(0.2–0.23)
0.23
(0.21–0.25)
0.29
(0.21–0.39)
0.0420.340.0120.07Yes
20.32
(0.27–0.37)
0.49
(0.41–0.59)
0.89
(0.57–1.4)
1.9E-062.1E-043.8E-067.0E-03Yes
50.55
(0.42–0.7)
0.59
(0.46–0.74)
1.7
(0.88–3.3)
1.5E-030.676.4E-049.0E-04No
101.6
(1.3–1.9)
2.1
(1.8–2.5)
2.8
(1.9–4.2)
1.0E-020.0230.0110.18No
IgG4 (μg/L)6m1.5E-04
(1.5E-04–1.5E-04)
1.7E-04
(1.3E-04–2.1E-04)
4.6E-04
(9.0E-05–2.4E-03)
4.9E-030.375.2E-030.024Yes
11.5E-04
(1.5E-04–1.5E-04)
6.9E-04
(3.2E-04–1.5E-03)
0.081
(4.6E-03–1.4)
1.8E-105.2E-046.6E-122.2E-05Yes
23.4E-04
(1.8E-04–6.6E-04)
4.8
(1.7–13)
61 (8.9–419)1.8E-251.5E-228.6E-189.8E-05Yes
52
(0.48–8.1)
168
(111–256)
539
(317–917)
1.1E-151.3E-121.0E-081.9E-04No
HDM cytokine
response^
IL-13 protein (pg/ml)^00.22
(0.066–0.73)
0.22
(0.076–0.63)
0.085
(0.011–0.66)
0.680.760.410.45No
6m0.064
(0.022–0.18)
0.06
(0.025–0.14)
19 (1.4–244)4.6E-060.981.7E-054.1E-06No
50.13
(0.046–0.37)
0.32
(0.11–0.87)
12 (1.2–117)2.1E-040.297.7E-055.1E-04No
IL-5 protein (pg/ml)^00.043
(0.018–0.11)
0.026
(0.013–0.052)
0.018
(5.0E-03–0.068)
0.440.360.290.57No
6m0.018
(9.2E-03–0.034)
0.013
(8.9E-03–0.02)
0.21 (0.012–3.7)7.9E-040.48.1E-033.5E-04No
50.028
(0.014–0.057)
0.042
(0.02–0.087)
2.3
(0.25–22)
3.2E-060.455.7E-062.0E-05No
IL-13 mRNA^01.7E-03
(1.1E-04–0.026)
6.0E-03
(4.8E-04–0.075)
6.7E-03
(3.3E-05–1.4)
0.850.60.680.94No
6m1.0E-04
(8.8E-06–1.1E-03)
3.2E-04
(3.8E-05–2.6E-03)
2
(0.015–266)
3.2E-040.51.7E-043.8E-04No
50.036
(1.6E-03–0.8)
0.11
(8.8E-03–1.4)
2.9E + 03
(742–1.1E + 04)
6.8E-050.599.9E-052.5E-05No
IL-4 mRNA^01.4E-06
(6.9E-07–3.0E-06)
1.9E-06
(7.8E-07–4.4E-06)
1.0E-06 (1.0E-06–1.0E-06)0.710.650.60.47No
6m4.6E-06
(1.0E-06–2.1E-05)
5.1E-06
(1.4E-06–1.8E-05)
0.54 (6.5E-03–44)6.2E-090.944.7E-071.0E-07No
52.3E-04
(1.7E-05–3.0E-03)
4.7E-04
(5.3E-05–4.3E-03)
5.3 (0.082–345)4.9E-040.724.5E-043.2E-04No
IL-5 mRNA^02.5E-04
(2.1E-05–2.9E-03)
2.6E-04
(2.8E-05–2.5E-03)
1.2E-05 (3.1E-07–4.6E-04)0.470.960.240.25No
6m5.2E-05
(5.6E-06–4.8E-04)
3.1E-05
(5.2E-06–1.8E-04)
0.33 (1.3E-03–83)1.5E-040.852.3E-041.1E-04No
50.021 (9.9E-04–0.43)0.07
(5.7E-03–0.85)
246 (7–8.7E + 03)1.3E-040.497.1E-051.1E-04No
Prop. (95% CI)Prop. (95% CI)Prop. (95% CI)OverallCas1 vs. 2Cas1 vs. 3Cas2 vs. 3
HDM SPT past atopy threshold
Wheal ≥ 2 mm6m2.3%
(0–5.4%)
1.9%
(0–4.5%)
14%
(0–29%)
0.04310.0540.035No*
210%
(3.8–17%)
15%
(8.1–22%)
86%
(71–100%)
2.9E-120.398.2E-121.5E-10No*
Wheal ≥ 3 mm513%
(5.2–20%)
28%
(18–37%)
81%
(63–99%)
1.5E-080.0224.6E-091.0E-05No
1036%
(23–49%)
51%
(38–63%)
78%
(57–99%)
7.4E-030.112.7E-030.06No
  1. Feature?=whether variable was used as a clustering feature or not; geom. mean = geometric mean; PBMC = peripheral blood mononuclear cells; prop. = proportion; SPT = skin prick or sensitisation test. For categorical variables, associations were tested using Fisher exact test; for continuous variables, Kruskal-Wallis and Mann-Whitney-Wilcoxon. Bold text indicates statistical significance (p<0.05); italics indicate near-significance (p<0.10). ^PBMC cytokine responses to HDM above unstimulated control; birth samples (age 0) taken from cord blood (CBMC). *Not used as clustering features, as these were derived variables; the variables from which they were derived (HDM IgE and IgG4) were used instead.

Table 3
Comparison of selected respiratory-disease-related variables in CAS clusters
https://doi.org/10.7554/eLife.35856.020
VariableAge (y)Cas1 (N = 88)Cas2 (N = 107)Cas3 (N = 22)P-value (unadjusted)Feature?
Mean (95% CI)Mean (95% CI)Mean (95% CI)OverallCas1 vs. 2Cas1 vs. 3Cas2 vs. 3
URI (events per y)12.9 (2.4–3.3)2.6 (2.2–3)2.5 (1.7–3.3)0.590.340.50.96Yes
23.2 (2.6–3.7)2.6 (2.2–3)2.5 (1.2–3.8)0.190.190.120.34Yes
32.7 (2.2–3.2)2.8 (2.4–3.3)2.2 (1.3–3.2)0.450.410.590.24Yes
42.1 (1.7–2.6)2.2 (1.8–2.7)1.7 (0.77–2.7)0.50.940.260.27No
51.6 (1.1–2)1.5 (1.2–1.9)0.67 (0.2–1.1)0.0810.760.0470.026No
LRI (events per y)11.6 (1.2–1.9)0.98 (0.76–1.2)2 (1.3–2.6)4.0E-030.0210.172.6E-03Yes
21.4 (0.98–1.7)1 (0.81–1.2)2.2 (1.6–2.9)2.5E-030.836.1E-032.0E-04Yes
31 (0.76–1.3)0.6 (0.4–0.8)1.8 (1.1–2.6)6.1E-040.020.0392.7E-04Yes
40.87 (0.52–1.2)0.46 (0.3–0.63)2 (1.1–2.8)1.7E-050.33.5E-041.6E-06No
50.42 (0.24–0.6)0.36 (0.24–0.48)0.86 (0.44–1.3)0.01910.0117.5E-03No
Wheezy LRI (wLRI, events per y)10.47 (0.3–0.63)0.24 (0.15–0.34)0.64 (0.19–1.1)0.0540.0360.610.065Yes
20.68 (0.45–0.91)0.41 (0.26–0.56)1 (0.56–1.5)5.2E-030.0630.0661.7E-03Yes
30.59 (0.37–0.81)0.3 (0.17–0.44)1.4 (0.78–2.1)4.6E-050.0652.5E-036.6E-06Yes
40.52 (0.25–0.79)0.32 (0.18–0.46)1.9 (0.95–2.8)4.5E-080.869.3E-073.3E-08No
50.28 (0.13–0.42)0.23 (0.13–0.33)0.76 (0.36–1.2)2.3E-030.992.0E-031.2E-03No
Febrile LRI (fLRI, events per y)10.36 (0.22–0.51)0.28 (0.16–0.4)0.55 (0.28–0.81)0.0250.240.0716.4E-03Yes
20.36 (0.23–0.5)0.33 (0.22–0.43)0.95 (0.46–1.4)0.0116.1E-033.8E-03Yes
30.38 (0.21–0.55)0.16 (0.09–0.23)0.52 (0.13–0.92)0.060.0630.440.04Yes
40.3 (0.13–0.47)0.15 (0.064–0.24)0.43 (0.16–0.7)0.0210.180.0914.9E-03No
50.19 (0.082–0.3)0.14 (0.06–0.21)0.19 (0–0.42)0.830.550.910.8No
Prop. (95% CI)Prop. (95% CI)Prop. (95% CI)OverallCas1 vs. 2Cas1 vs. 3Cas2 vs. 3
>20% Streptococcus in first infection-naive NPA sample7w11% (0.34–23%)15% (3.3–26%)44% (3.9–85%)0.0810.750.0420.065No
6m7.6% (1.6–14%)18% (10–26%)14% (0–31%)0.120.0450.391No
% Healthy NPAs with infection-associated MPGs0–249% (38–59%)32% (24–39%)62% (47–76%)1.2E-030.0130.25.5E-04No
2–446% (37–55%)44% (37–51%)45% (29–61%)0.90.670.920.8No
  1. Feature?=whether variable was used as a clustering feature or not; geom. mean = geometric mean; ARI = acute respiratory infection (lower or upper); LRI = lower respiratory infection; MPG = microbiome profile group; NPA = nasopharyngeal aspirate; prop. = proportion; URI = upper respiratory infection; 7w = 7 weeks. For categorical variables, associations were tested using Fisher exact test; for continuous variables, Kruskal-Wallis and Mann-Whitney-Wilcoxon. Bold text indicates statistical significance (p<0.05); italics indicate near-significance (p<0.10). *Not used as clustering features, as these were derived variables; the variables from which they were derived (URI, LRI, wLRI, fLRI) were used instead.

Table 4
Analysis of selected predictors for age-5 wheeze within each CAS cluster, with demographic covariates (sex, BMI, parental history of asthma)
https://doi.org/10.7554/eLife.35856.021
Selected predictors for age-5 wheeze
Cas1 (N = 88)
Or (95% CI)
P-valueCas2 (N = 107)
Or (95% CI)
P-valueCas3 (N = 22)
Or (95% CI)
P-valueOr (95% CI)P-value
LRI (events per y)10.97 (0.71–1.3)0.841 (0.61–1.5)0.990.48 (0.13–1.1)0.161 (0.81–1.2)0.92
21.2 (0.88–1.6)0.261.5 (0.97–2.5)0.0690.99 (0.34–2.6)0.981.4 (1.1–1.7)5.3E-03
32 (1.3–3.2)2.3E-032.6 (1.5–5.3)2.7E-030.98 (0.4–2.6)0.962 (1.5–2.7)3.8E-06
42 (1.4–3.4)2.0E-033.6 (1.8–8.3)6.5E-041.9 (0.57–8.4)0.322.5 (1.8–3.6)1.5E-07
Wheezy LRI (events per y)11.3 (0.68–2.4)0.431.1 (0.35–3)0.832.6 (0.62–58)0.341.5 (0.98–2.3)0.06
21.2 (0.8–2)0.331.6 (0.89–2.9)0.122.4 (0.67–16)0.241.6 (1.2–2.2)5.6E-03
32.8 (1.6–5.6)1.3E-033 (1.4–8)0.0161.2 (0.43–4.6)0.762.7 (1.8–4.2)4.1E-06
42.5 (1.5–5)4.0E-036.3 (2.5–21)6.8E-047.1 (1.2–169)0.13.9 (2.5–6.7)5.4E-08
Febrile LRI (events per y)11.6 (0.77–3.6)0.210.84 (0.28–1.9)0.717.3 (0.78–178)0.121.5 (0.93–2.4)0.098
21 (0.44–2.2)14.8 (1.8–15)3.9E-031.6 (0.48–10)0.52.3 (1.4–3.9)1.2E-03
32 (1–4.8)0.084.3 (1.2–15)0.024.2 (0.55–519)0.372.4 (1.4–4.3)2.3E-03
41.8 (0.97–4.1)0.0922.6 (0.88–8.3)0.0821.1 (0.11–18)0.932.2 (1.3–4)5.9E-03
Quartile of % healthy NPAs with infection-associated MPGs0–21 (0.54–1.8)0.981.3 (0.72–2.4)0.36NANA1.3 (0.89–1.8)0.19
2–40.45 (0.19–0.88)0.0351 (0.51–2.1)0.9NANA0.8 (0.53–1.2)0.24
HDM IgE (kU/L)*6m8 (0.85–94)0.0740.93 (0.14–3.6)0.923.4 (0.26–180)0.42.3 (0.99–5.8)0.054
11.5 (0.22–7.8)0.650.54 (0.039–2.3)0.5139 (2.5–22000)0.0822.7 (1.5–5)0.00089
20.93 (0.28–2.5)0.892 (1.2–3.7)0.0161.4 (0.38–4.8)0.622 (1.5–2.8)2.80E-05
31.4 (0.68–2.9)0.321.5 (0.9–2.4)0.121.5 (0.4–5.2)0.551.7 (1.3–2.2)1.00E-04
41.9 (0.94–4.1)0.0861.9 (1.2–3.1)0.0111.4 (0.31–5.5)0.641.9 (1.5–2.5)3.70E-06
HDM IgG4 (μg/L)*6mNA (NA-NA)0.550.053 (NA-6.5e + 24)0.9928 (1.7e-34-NA)0.991.4 (0.88–2.6)0.17
1NA (NA-NA)0.611.1 (0.8–1.5)0.50.9 (0.58–1.3)0.61.2 (1–1.4)0.053
21.1 (0.71–1.6)0.671.1 (0.85–1.4)0.610.4 (0.038–1.2)0.261.1 (1–1.3)0.056
31.1 (0.85–1.5)0.351.1 (0.77–2)0.640.94 (0.19–2.3)0.91.1 (0.98–1.2)0.1
41.2 (0.98–1.5)0.0820.89 (0.7–1.1)0.330.46 (0.031–5.4)0.531.1 (1–1.3)0.034
HDM IgG (mg/L)*125 (0.32–1.6E + 04)0.193.3 (0.16–46)0.385.6E-03 (8.4E-06–0.57)0.0582 (0.31–11)0.44
20.8 (0.15–3.5)0.780.97 (0.24–3.7)0.960.79 (0.031–18)0.881.3 (0.6–2.9)0.48
32.3 (0.14–35)0.540.48 (0.057–2.5)0.433.9 (0.26–96)0.342.1 (0.89–5)0.089
  1. BMI = body mass index; HDM = house dust mite; LRI = lower respiratory infection. Association analyses performed via generalised linear models (GLM) with demographic covariates: age-5 wheeze ~predictor + sex (male) +BMI at age 3 + paternal history of asthma +maternal history of asthma. Bold text indicates statistical significance (p<0.05); italics indicate near-significance (p<0.10). *Odds ratio (OR) is for every 10-fold increase in IgE, IgG4 or IgG.

Table 5
Key findings from cluster analysis
https://doi.org/10.7554/eLife.35856.024
Certain childhood populations may be broadly split into three clusters, each representing a unique trajectory of immune function and susceptibility to respiratory infections: low-risk non- atopic Cluster 1 with transient wheeze; low-risk but allergy-susceptible Cluster 2 with mixed wheeze; and strongly-atopic high-risk Cluster 3 with persistent wheeze.
Cluster 3 is consistent with an early-sensitised and multi-sensitised phenotype.
HDM hypersensitivity is an important predictor of wheeze in allergic or allergy -susceptible individuals.
Food and peanut hypersensitivities are important contributors to membership in high-risk Cluster 3. This may be pathophysiologically related to eczema, multi-sensitisation and the atopic march.
In CAS, IgG4 flags for clusters with susceptibility to atopic disease (CAS2 and CAS3), while early and multiple-allergen elevation in IgE predicts frank atopic disease. The pathophysiological role of IgG4 remains unclear.
Allergic and infective processes act additively to intensify airway inflammation during respiratory pathogen clearance. Some (Cluster 3) may be more susceptible to this effect than others that lack strong allergic sensitisation (Cluster 1).
Tests for atopy (IgE, SPT, cytokines) do not overlap perfectly. Therefore, atopy may be better defined by the composite result from a battery of tests encapsulated in a predictive model, rather than just a single test or threshold.
The microbiome acts differently on asthma risk depending on cluster membership. In CAS, early-life asymptomatic colonisation with infection-associated MPGs is associated with risk of persistent wheeze in allergy-susceptible clusters (CAS2, CAS3), while it is potentially protective in non-atopic children (CAS1)
Different childhood populations may share similar trajectories of asthma susceptibility, but there may be subtle differences in terms of the types of tests, allergens, or biological signals that are most informative (SPT, IgE, cytokines, etc.).
Key resources table
Reagent type
(species) or resource
DesignationSource or referenceIdentifiers
Biological sample (Homo sapiens)Childhood Asthma Study (CAS)DOI:10.1016/j.jaci.2005.06.038Microbiome sequencing data: NCBI
GenBank SRP056779
Biological sample
(Homo sapiens)
Childhood Origins of Asthma Study
(COAST)
PMID:12688623NA
Biological sample (Homo sapiens)Manchester Asthma
and Allergy Study (MAAS)
PMID:12688622NA
Software,
algorithm
The R project for Statistical ComputingISBN:3-900051-07-0RRID:SCR_001905
Software,
algorithm
ggplot2ISBN:978-3-319-24277-4RRID:SCR_014601
Software,
algorithm
mixtoolsDOI:10.18637/jss.v032.i06NA
Software,
algorithm
rpartTherneau and Atkinson, 2015.
Package 'rpart'. URL: https://cran.r-project.org/web/packages/rpart/rpart.pdf
NA
Software,
algorithm
epiDisplayChongsuvivatwong, 2015.
Package 'epiDisplay' URL:
https://cran.r-project.org/web/packages/epiDisplay/epiDisplay.pdf
NA

Data availability

This study utilises extensive data from human subjects, specifically paediatric cohorts, for which eLife's policies recognise that there can be strong reasons to restrict access. For each of the cohorts involved in our study (CAS, COAST, MAAS), parents were consented on the use of biomedical data for allergy and asthma research, but not for the open sharing of their or their children's data. Studies were run in the late 1990s and early 2000s and we do not have ethics permission to attempt to recontact families to seek consent. Importantly, we note that key data features could risk re-identification of subjects (e.g. demographic data from small communities). However, we have provided public data at the summary level which can be used for subsequent studies, such as replication and meta-analysis. This is standard practice in sensitive data settings, such as genome-wide association studies. These data have been uploaded as Excel spreadsheets to FigShare for ease of data extraction: Supplementary Table 3 https://figshare.com/articles/Supplementary_File_1_1/6934052; Supplementary Table 7 https://figshare.com/articles/Supplementary_File_1_2/6934055

The following data sets were generated
  1. 1
    Figshare
    1. HF Tang Howard
    2. Inouye Michael
    3. E Holt Kathryn
    (2018)
    Data from supplementary table 3.
  2. 2
    Figshare
    1. HF Tang Howard
    2. Inouye Michael
    3. E Holt Kathryn
    (2018)
    Data from supplementary table 7.

Additional files

Supplementary file 1

All table supplements

https://doi.org/10.7554/eLife.35856.025
Supplementary file 2

Comparison of variables (respiratory, immunological, clinical) across CAS clusters.

Analogous to Table Supplement 3.

https://doi.org/10.7554/eLife.35856.026
Supplementary file 3

Predictors for age-five wheeze within each CAS cluster, with demographic covariates (sex, BMI, parental history of asthma).

Analogous to Table Supplement 7.

https://doi.org/10.7554/eLife.35856.027
Transparent reporting form
https://doi.org/10.7554/eLife.35856.028

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)