Characteristics of the cohort and outcome over follow up

The average age for patients who initiated mepolizumab was 57 years, 68% of these individuals were female, median eosinophil count was 269 cells per mcl, and patients had 2.4 exacerbations in the year prior to therapy initiation. The average age of the omalizumab group was 42 years, 90% of the group identified as White, mean BMI was 28.2 kg/m2, and the individuals had 1.5 exacerbations on average in the year prior to initiation of omalizumab (Table 1). Over one year of follow-up, 10 (45.5%) of the 22 mepolizumab patients demonstrated some improvement in exacerbations over one year of follow up with 7 (70%) of these 10 patients having a halving or more of their baseline exacerbation rate in the one year on mepolizumab. The remaining twelve patients included six (27.3%) who had similar exacerbation rates over follow up as at baseline and six patients (27.3%) who had higher exacerbations over the first year of therapy. Over one year of follow-up, 11 (37.9%) of the 29 omalizumab patients had a reduction in their exacerbations over follow up with 8 (73%) of these 11 patients having a halving or more of their baseline exacerbation rate. Ten of the remaining 18 individuals had similar exacerbation rates over follow up as at baseline and 8 fared worse in the year on omalizumab.

Table 1 Baseline characteristics of participantsProteins associated with response: mepolizumab

Lysyl oxidase like protein-2 (LOXL2; unadjusted p: 1.93 × 10E−05, FDR-corrected: 0.028) and myostatin (MSTN; also known as GDF-8- Growth derived factor-8; unadjusted: 3.87 × 10E−05, FDR-corrected: 0.028) were associated with increased risk of post-mepolizumab exacerbation (Fig. 1). Eleven proteins met significance at the FDR threshold of < 0.20 (Additional file 1: Table S1). This includes neutrophil collagenase (MMP-8), the only protein with high levels corresponding with lower risk, mucosa-associated lymphoid tissue lymphoma translocation protein 1 (MALT1), and interferon-related proteins (IFNA7 and IRF4) (Fig. 1 and Additional file 1: Fig. S1). The exacerbation rate was significantly lower in patients with low levels (below median) of LOXL2, myostatin, and ADAMTSL2. For MMP8, the exacerbation rate was higher in patients with low levels (Fig. 2). While patients with eosinophil counts of ≥ 300 cells per mcL was associated with greater reductions in exacerbations on mepolizumab, this did not reach statistical significance.

Fig. 1

Proteins associated with response to mepolizumab. Modeled using negative binomial regression with post-mepolizumab exacerbation rate as a dependent variable and adjusted for age, sex, eosinophil counts, and pre-mepolizumab annualized exacerbation rate. Proteins with false discovery rate (FDR) < 0.20 were considered statistically significant. Diamonds indicate proteins that are significant at FDR < 0.05. Squares indicate proteins with significant association with FDR < 0.20. Circles (‘None’) indicate non-significant proteins. Effect sizes were presented as beta coefficients where positive (negative) values indicate association with increased (decreased) post-mepolizumab exacerbation. The vertical dashed lines represent no difference. The grey horizontal dashed line represents FDR < 0.20 and the black horizontal dashed line represents FDR < 0.05

Fig. 2

Forest plot showing the relationship between the exacerbation rate ratio and unadjusted levels of the top 5 proteins associated with mepolizumab response. ‘Low’ indicates protein level below the median and ‘high’ indicates levels at or above the median. Exacerbation rate ratios < 1.0 indicate better response to mepolizumab, that is greater reductions asthma exacerbation. *comparing groups, e.g., high vs low: p < 0.05. BMI, body mass index; CI, confidence interval; Eos, eosinophil counts

GSEA demonstrated 4 significantly enriched functional groups associated with response to mepolizumab (Additional file 1: Fig. S2). The strongest associations were with extracellular matrix organization and extracellular structure organization (both with NES 1.93, unadjusted p = 7.63 × 10E−05, FDR-corrected: 0.07, Additional file 1: Table S2). The gene-concept network map highlighted functional groups extracellular matrix, extracellular structure organization, and external encapsulating structure organization with leading edge interactions through LOXL2, ADAMTSL2, and CAV2 (Additional file 1: Fig. S3).

PPI network of mepolizumab response module

We evaluated the interactome of the eleven proteins associated with mepolizumab’s response (mepolizumab response module). Five (55%) of these proteins (IFNA7, IRF4, CD79B, and MALT1, and BIRC3) were directly connected without including other interactors (P for enrichment: < 0.001). Three of these five proteins (IRF4, CD79B, and MALT1) were associated with functional enrichment of the B-cell receptor signaling pathway (FDR p-value: 0.016). Including interactors in the network resulted in ten nodes and 33 edges, the MCL algorithm generated 5 unsupervised clusters (Fig. 3). Three of these clusters had 3 or more proteins in its network (Additional file 1: Table S3). The first cluster (PPI enrichment: < 1.0 × 10E−17) included myostatin and five other proteins involved in the activin and TGF-beta signaling pathways (enrichment strength: 2.03, FDR P: < 0.001) and in cytokine–cytokine receptor interaction (enrichment strength: 1.67, FDR P < 0.001). The second cluster was associated with enrichment in the TNF signaling pathway and included caspase-related proteins (RIPK1 and RIPK3). This cluster was related to cluster-5 which included CARD9 and MALT1. Cluster 4 included two proteins: CD79A and CD79B and was enriched for the B cell receptor signaling pathway (Additional file 1: Table S4).

Fig. 3

Protein–Protein Interaction (PPI) network of mepolizumab response. Network including interactions with at least a score of high confidence (≥ 0.70), 10 interactors in first shell, and 5 in the second shell. Clusters (color-coded) were generated using the Markov Cluster Algorithm (MCL) embedded in the STRING database

Proteins associated with response: omalizumab

For omalizumab, 172 proteins met significance at the FDR-corrected threshold of P < 0.05 and 5 of these proteins at the P < 0.001 threshold. Higher levels of CD9 antigen (unadjusted: 5.30 × 10E−07, FDR-corrected: 0.0006) and mucin-1: region 3 (MUC1; unadjusted: 1.15 × 10E−06, FDR-corrected: 0.0006) were associated with better response to omalizumab, while higher levels of leukotriene B4 receptor 1 (LTB4R; unadjusted: 1.12 × 10E−06, FDR-corrected: 0.0006), inhibitor of growth protein 1 (ING1; unadjusted: 2.43 × 10E−06, FDR-corrected: 0.0008), and the sulfotransferase family 1A, SULT1A1 (unadjusted: 3.24 × 10E−06, FDR-corrected: 0.0009) were associated with worse response to omalizumab (Fig. 4 and Additional file 1: Table S5). Clustering of the top 80 proteins revealed some clustering by change in exacerbations but with no distinct cut-offs (Additional file 1: Fig. S4).

Fig. 4

Proteins associated with response to omalizumab. Modeled using negative binomial regression with post-omalizumab exacerbation rate as a dependent variable and adjusted for age, sex, body mass index, and pre-omalizumab annualized exacerbation rate. Proteins with false discovery rate (FDR) < 0.20 were considered statistically significant. Diamonds indicate proteins that are significant at FDR < 0.05. Squares indicate proteins with significant association with FDR < 0.20. Circles (‘None’) indicate non-significant proteins. Effect sizes were presented as beta coefficients where positive (negative) values indicate association with increased (decreased) post-omalizumab exacerbation. The vertical dashed lines represent no difference. The grey horizontal dashed line represents FDR < 0.20 and the black horizontal dashed line represents FDR < 0.05

The exacerbation rate was lower in patients with high levels of CD9 and MUC1. For SULT1A1, the exacerbation rate was higher in patients with high levels (Fig. 5). The association between IgE ≥ 100 or < 100 and the change in exacerbations was not significantly different.

Fig. 5

Forest plot showing the relationship between the exacerbation rate ratio and unadjusted levels of the top 5 proteins associated with omalizumab response. ‘Low’ indicates protein level below the median and ‘high’ indicates levels at or above the median. Exacerbation rate ratios < 1.0 indicate better response to mepolizumab, that is greater reductions asthma exacerbation.. *comparing groups, e.g., high vs low: p < 0.05. BMI, body mass index; CI, confidence interval; IgE, immunoglobulin E

GSEA demonstrated 60 significantly enriched functional groups at FDR-corrected P < 0.05 and 145 at P < 0.20 associated with response to omalizumab (top 5 gene sets represented in Additional file 1: Fig. S5; top 20 gene sets shown in Additional file 1: Table S6). The strongest associations were with intracellular transport (NES 2.02, unadjusted p = 4.63 × 10E−08, FDR-corrected: 0.0001), peptidyl-threonine phosphorylation (NES 2.08, unadjusted p = 6.49 × 10E−06, FDR-corrected: 0.0036), and phosphorus metabolic process (NES 1.52, unadjusted p = 8.43 × 10E−06, FDR-corrected: 0.0036). The network map highlighted functional groups intracellular transport, phosphorylation, phosphate-containing compound metabolic process, and cellular macromolecule metabolic processes with leading edge interactions through ING1, MAPK3, SYK, HEXIM1, DERL1, and SULT1A1. (Additional file 1: Fig. S6 and Table S6).

PPI network of omalizumab response module

The omalizumab response module included 428 nodes and 1864 edges with a P-value for enrichment of < 1.0 × 10E-17. Unsupervised MCL clustering generated 87 clusters with 40 of these clusters (46%) having a node degree of 3 or more, that is had at least two interactions within the network. When limited to proteins that met the FDR corrected significance threshold of < 0.05, there were 187 nodes, 204 edges, and 20 clusters with ≥ 3 proteins in the network (Fig. 6 and Additional file 1: Table S7). The top 4 clusters had 8 or more proteins. The core of Cluster 1 (20 proteins) and Cluster 4 (8 proteins) demonstrated multiple functionally enriched pathways involving protein kinases which included the AKT serine-threonine protein kinase family (AKT1 and AKT2), the Mitogen‑activated protein kinase (MAPK) signaling pathway (MAPK1, MAPK3, MAPK8, MAPK14), and the Protein Kinase C family (PRKCA and PRKCB, and PRKCQ). Cluster 2 (9 proteins) included MUC1, and two members of the casein kinase-II (CK2) class of the serine-threonine protein kinase family (CSNK2A1 and CSNK2B) (Additional file 1: Table S7). The enriched pathway with the highest strength was the Fc epsilon R1 signaling pathway (FDR P 2.88 × 10E−21) (Additional file 1: Table S8). Proteins involved in this pathway included multiple MAPK and AKT genes and the phospholipases: PLA2G4A and PLCG2.

Fig. 6

Protein–Protein Interaction (PPI) network of omalizumab response. Network including interactions with at least a score of high confidence (≥ 0.70), 10 interactors in first shell, and 5 in the second shell. Clusters (color-coded) were generated using the Markov Cluster Algorithm (MCL) embedded in the STRING database

Overlap between mepolizumab and omalizumab

At the FDR-corrected p-value threshold < 0.05, no protein was associated with both omalizumab and mepolizumab response. At an FDR-corrected P < 0.20 threshold, the mucosa-associated lymphoid tissue lymphoma translocation protein was associated with both omalizumab and mepolizumab response (Fig. 7). Higher levels of MALT1 were associated with poorer response to both mepolizumab (unadjusted P 0.00047, FDR-corrected: 0.15) and omalizumab (p 0.048, FDR-corrected: 0.17). MALT1 in addition to CARD9 made up Cluster 5 in the PPI network of mepolizumab response module and was closely related to Cluster 2 which included BIRC3, TNFRSF1, and the serein/threonine-protein kinases, RIPK1 and RIPK3 (Additional file 1: Table S3). For omalizumab, MALT1 was part of the core enrichment of 11 of 52 gene sets significant at the FDR-corrected p < 0.05 level but was not part of the top 20 clusters associated with the omalizumab response module.

Fig. 7

Overlapping proteins associated with the change in exacerbations in the mepolizumab group and in the omalizumab group. a Venn diagram indicates the numbers of significant proteins and overlapping proteins. b Forest plot of the singular overlapping protein (mucosa associated lymphoid tissue lymphoma translocation protein 1- MALT1) associated with change in exacerbations in both the mepolizumab and the omalizumab groups. Squares indicate proteins with significant association with FDR < 0.20. Effect sizes are presented as beta coefficients with 95% confidence intervals where positive (negative) values indicate association with increased (decreased) exacerbations following biologic initiation. The vertical dashed lines represented no difference