Validation of Systemic Complement Signatures in the Progression of Chronic Obstructive Pulmonary Disease

To the Editor:

Individuals with chronic obstructive pulmonary disease (COPD) vary substantially in rates of spirometric lung function decline (1). Rapidly progressing individuals endure worse outcomes, including higher hospitalization and mortality rates (2). A more comprehensive understanding of the mechanisms driving progression is essential to inform novel treatment approaches aimed at halting accelerated lung function decline.

Using samples that were collected at enrollment in the SPIROMICS study, we previously identified a multivariate 52-protein signature that predicted the risk for accelerated FEV₁ decline (≥70 ml/yr) with high cross-validation accuracy (3). An enrichment analysis of this progression signature indicated that differences in spirometric progression rates were uniquely associated with baseline alterations in the complement pathway. This finding was supported by a subsequent principal-component analysis (PCA) involving 22 baseline complement proteins, which confirmed significant changes in the expression of blood-derived complement proteins (“complement profiles”) associated with progression phenotypes. Complementary efforts to refine our progression signature yielded smaller five- and 10-protein signatures with prognostic potential, although most complement proteins were removed during this stage. Nonetheless, the lack of external validation limited insights into the clinical utility of these composite biomarkers. In this study, we utilized plasma samples from two prospective cohorts to examine the reproducibility of our previously identified baseline biomarker signatures and complement profiles in predicting longitudinal FEV₁ decline.

Some of the results of these studies have been previously reported in the Ph.D. dissertation of the first author (4).

We obtained plasma samples from participants in the SPIROMICS and COPDGene cohorts who were independent of those analyzed in our previous work (3). Institutional review boards at participating institutions approved the study, and participants provided written informed consent. Individuals with a smoking history, available baseline SOMAscan (SomaLogic) plasma measurements (by means of the SOMAscan 1.3k platform), and 5-year follow-up spirometry were classified into three groups on the basis of their annualized rate of post-bronchodilator FEV₁ decline (ΔFEV₁) and presence of longitudinal airflow obstruction (Figure 1A): the COPD greater decliner (COPD-GD) group, the COPD lesser decliner (COPD-LD) group, and the group of tobacco-exposed persons with preserved spirometry (TEPS). Further details regarding model inclusion criteria, cohort demographics, and SOMAscan datasets have been previously published (3, 5–7).

Figure 1.

Systemic complement profiles are replicated in independent samples from two unique analysis groups. (A) Illustration of individuals included in discovery (3) and validation analyses from SPIROMICS and COPDGene cohorts. Patients were excluded on the basis of missing plasma data, baseline or longitudinal spirometry measurements, or PRISm criteria (for COPDGene only). Participants were then classified into three groups on the basis of their annualized rate of post-bronchodilator FEV₁ decline (ΔFEV₁) and presence of airflow obstruction at both baseline and 5-year follow-up (V5): chronic obstructive pulmonary disease–greater decliners (COPD-GD; ΔFEV₁ ≥ 70 ml/yr and FEV₁/FVC < 0.7), COPD lesser decliners (COPD-LD; ΔFEV₁ < 70 ml/yr and FEV₁/FVC < 0.7), and a control group of tobacco-exposed persons with preserved spirometry (TEPS; FEV₁/FVC > 0.7 at baseline and V5). Eligible COPDGene participants were matched to SPIROMICS participants included in the discovery and validation groups on the basis of sex and specified ranges of age (±10 yr), FEV₁/FVC (±0.07), and FEV₁ percent predicted (±18% for COPD and ± 25% for TEPS). (B) Receiver operating characteristic curves for the performance of prognostic signatures of FEV₁ decline with five-protein (dashed lines) and 10-protein (solid lines) signatures in the SPIROMICS (n = 16; gray) and COPDGene (n = 61; black) validation groups. All validation models had areas under the curve ≈ 0.5. For the five-protein signature, SPIROMICS had 20% sensitivity and 82% specificity, whereas COPDGene had 10% sensitivity and 95% specificity. For the 10-protein signature, SPIROMICS had 0% sensitivity and 73% specificity, whereas COPDGene had 21% sensitivity and 90% specificity. Performance measurements were derived from receiver operating characteristic curves on the basis of the partial-least squares discriminant analysis models’ performance with the validation datasets. (C and D) Measurements from 22 complement proteins (C1q, C1qBP, C1r, C2, C3d, C3b, C3, C3a, iC3b, C3a desArg, C4, C4b, C5, C5a, C5-6, C6, C7, C8, C9, Factor B, Factor D, and Properdin), which were taken in plasma samples from (C) COPD-GD (n = 5; red circles) and COPD-LD (n = 11; blue squares) from the SPIROMICS validation group and (D) COPD-GD (n = 19; red circles) and COPD-LD (n = 42; blue squares) from the COPDGene validation group, fit to the principal-component analysis model previously generated on the SPIROMICS discovery group (3). P values are reported from a permutation test (n = 2,000 permutations) between groups’ mean scores across the first and second principal component (PC1 and PC2, respectively). The labels on the plots denote group means. (E and F) Kernel density plots of the scores across the PC1 from COPD-GD and COPD-LD in the (E) SPIROMICS and (F) COPDGene validation groups, compared with scores from the original principal-component analysis model generated on the discovery analysis (dashed lines). *A detailed analysis of the discovery population has been previously published (3).

The SPIROMICS and COPDGene validation groups comprised 29 individuals (COPD-GDs, n = 5; COPD-LDs, n = 11; and TEPSs, n = 13) and 114 individuals (COPD-GDs, n = 19; COPD-LDs, n = 42; and TEPSs, n = 53), respectively. To ensure concordance of population demographics across studies, the COPDGene validation group reflects a subset of participants matched to the SPIROMICS participants (discovery and validation) on the basis of sex and specified ranges of age, FEV₁/FVC, and FEV₁ percent predicted (Figure 1A and Table 1). The final validation groups were well matched for all demographics except race (anticipated from the design of COPDGene) and inhaled corticosteroid (ICS) use; however, previously, we found no impact of ICS use on our analyses (3). We assessed the performance of our previous models by obtaining validation metrics on each dataset independently (3).

Table 1.

Baseline Participant Characteristics

Characteristic	SPIROMICS* (N = 114)	COPDGene (N = 114)	P Value†
Age, mean ± SEM	61 ± 8.6	62 ± 8.7	0.79
Currently smoking, n (%)	74 (65)	78 (68)	0.67
BMI, kg/m2, mean ± SEM	28 ± 5.2	29 ± 5.2	0.61
Sex, male, n (%)	51 (45)	51 (45)	1
Race, White/other, n (%)	85/29 (75)	102/12 (89)	0.006
ICS use, yes, n (%)	27 (24)	4 (4)	<0.001
FEV1 percent predicted, mean ± SEM	86 ± 21	83 ± 22	0.30
FEV1/FVC, mean ± SEM	0.67 ± 0.13	0.67 ± 0.13	0.86
FEV1, L, mean ± SEM	2.5 ± 0.78	2.5 ± 0.89	0.85
Visit 5 FEV1, L, mean ± SEM	2.3 ± 0.86	2.3 ± 0.91	0.89
Time from baseline to follow-up, yr, mean ± SEM	6.2 ± 0.94	5.4 ± 0.71	<0.001
ΔFEV1, ml/yr‡	−50.9 ± 46	−50.6 ± 43	0.97

Definition of abbreviations: BMI = body mass index; COPD = chronic obstructive pulmonary disease; GD = greater decliner; ΔFEV₁ = annualized decline in FEV₁; ICS = inhaled corticosteroids; LD = lesser decliner; TEPS = tobacco-exposed persons with preserved spirometry.

^*SPIROMICS analysis group includes participants from the discovery (n = 85) and validation (n = 29) groups.

^† We used a two-sample, two-tailed t test or chi-square test to determine significant differences. Bold type denotes a statistically significant difference between SPIROMICS and COPDGene analysis groups (P < 0.05).

^‡ Reported values only include those for the COPD-GD and COPD-LD groups, as ΔFEV₁ was not used to classify participants in the TEPS group. ΔFEV₁ values for the SPIROMICS and COPDGene analysis groups, including participants in the TEPS group, were −40 (±47) and −39 (±42), respectively.

We first assessed whether our previously identified prognostic biomarker signatures (3) accurately predicted accelerated progression (≥70 ml/yr) in these independent samples. In the SPIROMICS validation group, the five-protein signature, involving inactivated complement C3b, cadherin-2, leptin, heparin-binding EGF-like growth factor, and teratocarcinoma-derived growth factor 1, derived from a partial-least squares discriminant analysis model exhibited suboptimal performance, producing an area under the curve (AUC) of 0.6, notably lower than the discovery group’s cross-validated AUC of 0.8 (Figure 1B). The 10-protein signature—which included the same five proteins plus carbonic anhydrase–related protein 10, IFN-γ, low affinity immunoglobulin γ Fc region receptor II-B, apolipoprotein L1, and lymphatic vessel endothelial hyaluronic acid receptor 1—also performed poorly, with an AUC below 0.5. Outcomes were similar in the COPDGene validation group, with both signatures performing only slightly better than randomly (five-protein signature, AUC = 0.56; 10-protein signature, AUC = 0.58). Nevertheless, all validation analyses consistently achieved specificities above 70%, indicating the signatures’ capability to characterize the COPD-LD group and suggesting that these signatures may provide insights into proteomic profiles that are modestly protective against progression.

We next examined whether alterations in complement profiles, as determined by the participant scores derived from a PCA model involving 22 baseline complement proteins (3), were linked to COPD progression. In the SPIROMICS validation group, a permutation test comparing the mean scores across the first two principal components of the model revealed significantly different complement profiles between the COPD-GD and COPD-LD groups (P = 0.045; Figures 1C and 1E). Consistent with our discovery analysis, after introducing a TEPS reference group to the PCA model, complement profiles from the COPD-GD group remained significantly distinct from those of the COPD-LD and TEPS groups (P = 0.039; data not pictured). Conversely, the TEPS and COPD-LD groups exhibited similar profiles. In evaluations with the COPDGene cohort, when we performed replication analyses on all individuals who met the inclusion criteria used in the original analysis (n = 532), we were unable to reproduce our results in either the two-group (COPD-GD vs. COPD-LD) or three-group (COPD-GD vs. COPD-LD vs. TEPS) comparisons. However, when the COPDGene validation cohort was matched for key demographic characteristics, especially baseline lung function parameters, between analysis groups (n = 114), an analysis of the matched COPDGene validation group confirmed our previous observations, with complement profiles of the COPD-GD group again differing significantly from those of the COPD-LD group (P = 0.005; Figures 1D and 1F), even after the addition of the TEPS group to the model (P = 0.029; data not pictured). A linear regression analysis also revealed that participant scores across the first principal component were highly correlated with annualized declines in FEV₁ in both the SPIROMICS (P = 0.004) and COPDGene analysis groups (P = 0.02), even after adjusting for age, sex, current smoking status, and ICS use.

To our knowledge, this study is the first to provide reproducible, longitudinal evidence that early alterations in the complement pathway measured at baseline are associated with a risk for accelerated FEV₁ decline approximately 5–6 years later. By demonstrating that complement alterations precede accelerated FEV₁ decline, we extend reported associations between the levels of blood-derived complement proteins and cross-sectional COPD outcomes, such as FEV₁ percent predicted (8, 9) and emphysema severity (10). Although our findings reflect systemic profiles, alterations in the levels of various complement components have been reported in COPD lungs (including C3a, C4, C5a, and MASP-2), suggesting that dysregulation might extend into the pulmonary compartment (11–14). How such aberrations in the blood or tissue might contribute to airflow limitation is unknown. However, murine studies have implicated a potential role for C1q and C3 in cigarette smoke–associated emphysema development through modifying antigen-presenting cell–directed Th17 inflammation (15, 16).

Although the pathway-related alterations were successfully validated in two independent datasets, replicating small prognostic signatures proved more challenging. This difficulty aligns with previous proteomics investigations in large COPD cohorts (17), which suggest that no single protein or combination can reliably predict progression in such a heterogeneous disease. Instead, it is plausible that spirometric decline results from alterations in disparate pathways or their constituents. Together, our results emphasize the value of data-driven models in understanding COPD outcomes and underscore the importance of shifting focus from biomarker discovery to pathway dysregulation in future analyses of spirometric progression.

Our study has several limitations. Chief among them is our inability to assess the function of complement components, as SomaLogic aptamers cannot reliably distinguish complement cleavage products from their parent proteins. The comprehensive demographic matching that was used in this analysis also limited the size of our COPDGene validation group. However, it is important to note that validated findings here are dependent on a validation cohort that is matched for key demographic characteristics, especially baseline lung function parameters, between analysis groups. This finding aligns with a previous COPD study that reported poor replication between systemic plasma proteins in unmatched cohorts (17), underscoring the importance of population matching to validate prognostic models.

In summary, our results suggest that the alterations in blood complement levels precede and are reliably associated with accelerated spirometric decline. Targeted analyses of the complement pathway are indicated in COPD to deconvolve its link to spirometric progression.