Preclinical Pulmonary Fibrosis Circulating Protein Biomarkers

To the Editor:

Idiopathic pulmonary fibrosis (IPF) is characterized by progressive, irreversible scarring of the lung parenchyma that can require invasive diagnostic testing (1). Interstitial lung abnormalities (ILAs) have been described in the general population (2). Among asymptomatic first-degree relatives of patients with familial interstitial pneumonia (FIP), 14% have radiologic ILAs and 35% have interstitial abnormalities on biopsy (3). In the Framingham population, fibrotic ILAs were present in 1.8% of subjects ≥50 years of age (4) and associated with increased risk of death (5, 6), suggesting ILAs may be a harbinger of IPF.

Because ILAs include ground glass and diffuse centrilobular nodularity can be present without fibrosis of the lung, we created the term “preclinical pulmonary fibrosis” (PrePF) (7) to identify first-degree relatives of patients with FIP (a high-risk cohort) not known to have interstitial lung disease who have features of lung fibrosis on high-resolution computed tomography.

We used proteomic analyses of plasma to identify circulating markers of IPF and then determine if IPF-associated proteins are predictive of PrePF. Some of the results of these studies have been previously reported in the form of an abstract (8).

Methods and Findings

Subjects with IPF (American Thoracic Society and European Respiratory Society criteria) (1) and first-degree relatives of patients with FIP with no known interstitial lung disease were recruited at University of Colorado, National Jewish Health, and Vanderbilt University (COMIRB #15-1147; NJH IRB 1441a; Vanderbilt IRB #020343) (7). PrePF was defined as evidence of fibrosis (reticular abnormality or traction bronchiectasis, or honeycombing) on high-resolution computed tomography (7).

Samples were proteolyzed with the iST Kit in 96-well format (PreOmics) and analyzed by mass spectrometry (Q Exactive HF, Ultimate 3000; ThermoFisher) in a data-independent acquisition mode (9, 10). Protein identification was performed by peptide mapping (Spectronaut Pulsar) to an in-house plasma spectral library at a precursor Q value cutoff of 0.01 and using the match-between run option at a 0.1 percentile threshold. Label-free quantification was performed on the intensities of summed fragment spectra.

Raw intensity data were normalized via a local (retention time-dependent) method and log transformed (9). Intensities were compared in IPF versus unaffected plasma, controlling for age, sex, and family relatedness in a linear mixed-effects model. Analyses were performed in the RStudio (v.3.2.2) and the R (v.3.5.3) environment using the lme4 package. Proteins differentially detected (false discovery rate [FDR] < 0.05) in the IPF versus unaffected analysis were then tested in PrePF versus unaffected plasma using the same model.

Plasma samples were filtered to include the oldest unaffected member per family while maximizing the number of PrePF subjects. Top differentially detected, uncorrelated proteins were used to generate a predictive model. The caret R package was used to train models and receiver operating characteristic curves. Models were developed with only age and sex, and uncorrelated proteins were iteratively added. The model with the highest area under the curve (AUC) was selected.

A total of 328 samples were analyzed. Six were excluded because of hemolysis and six because of internal quality control failures, leaving 316 samples in the analysis. Of these, 34 had IPF, and 282 were first-degree relatives of patients with FIP (240 without radiologic lung fibrosis, 42 had PrePF). Those with PrePF or IPF were older and more likely to be male and have the IPF-associated MUC5B promoter variant rs35705950 (minor allele frequency 0.29 and 0.32, respectively, vs. 0.21 in unaffected subjects). Unaffected subjects were from families with FIP, so they were enriched for the MUC5B promoter variant compared with other studies (11).

Comparison of IPF (n = 34) to first-degree relatives without lung fibrosis (n = 240) revealed 25 plasma proteins differentially detected (FDR < 0.05) (Table 1). These 25 proteins were examined in the first-degree relatives with PrePF (n = 42) versus those without lung fibrosis (n = 240), revealing that 12 of the 25 plasma proteins remained differentially detected (GSN [gelsolin], S100-A9, CRKL [Crk-like protein], LBP [LPS-binding protein], C1QC [C1q subcomponent subunit C], S100A8, BASP1 [brain acid soluble protein 1], SPARC or osteonectin [secreted protein acidic and rich in cysteine], APOA4 [apolipoprotein A-IV], C9, ALB [albumin], and CRISP3 [cysteine-rich secretory protein 3]) (Table 2). The directionality of the plasma protein differences remained constant in terms of affected (IPF or PrePF) versus unaffected subjects.

Table 1.

IPF versus No Fibrosis, Significant Proteins in Plasma

Protein	Coefficient	P Value	FDR
GSN	−0.28	2.82 × 10−12	1.04 × 10−9
C1QC	−0.33	1.52 × 10−6	0.0003
KNG1	−0.18	3.33 × 10−6	0.0004
CLEC3B	−0.31	2.35 × 10−5	0.0022
A2M	0.36	5.44 × 10−5	0.0025
APOA4	−0.32	4.03 × 10–5	0.0025
FBLN1	0.25	5.94 × 10−5	0.0025
YTHDC2	−0.25	5.00 × 10−5	0.0025
CRKL	−0.30	5.99 × 10−5	0.0025
SPARC	0.59	7.34 × 10−5	0.0027
PRSS3	0.51	0.0001	0.0041
ALB	−0.14	0.0002	0.0051
LBP	0.27	0.0003	0.0082
APOA2	−0.22	0.0006	0.015
BASP1	−0.42	0.0007	0.011
APOA1	−0.21	0.0010	0.021
S100A8	−0.83	0.0010	0.021
CRISP3	−0.50	0.0010	0.021
CTBS	0.34	0.0012	0.024
C9	0.24	0.0014	0.024
PGLYRP2	−0.20	0.0014	0.024
S100A9	−0.65	0.0014	0.024
FGG	0.20	0.0015	0.025
HP	0.33	0.0023	0.035
IGKV1D_13	0.76	0.0028	0.042

Definition of abbreviations: FDR = false discovery rate; IPF = idiopathic pulmonary fibrosis.

Differentially detected proteins discovered in IPF versus no lung fibrosis plasma protein analysis are shown. Analysis was controlled for age, sex, and family relatedness in a linear mixed-effects model; raw P values are listed as well as adjustment for multiple testing.

Table 2.

PrePF versus No Fibrosis, Plasma Protein Analysis

Protein	Protein Name	Coefficient	95% CI	P Value	FDR
GSN	Gelsolin	−0.14	−0.22 to −0.07	0.0002	0.003
S100A9	Protein S100-A9	−0.73	−1.11 to −0.35	0.0002	0.003
CRKL	Crk-like protein	−0.23	−0.37 to −0.10	0.0006	0.005
LBP	LPS-binding protein	0.21	0.08 to 0.35	0.0013	0.006
C1QC	Complement C1q subcomponent subunit C	−0.22	−0.35 to −0.09	0.0011	0.006
S100A8	Protein S100-A8	−0.67	−1.13 to −0.25	0.0021	0.009
BASP1	Brain acid soluble protein 1	−0.32	−0.55 to −0.10	0.0042	0.015
SPARC	SPARC	0.35	0.09 to 0.61	0.0075	0.024
APOA4	Apolipoprotein A-IV	−0.18	−0.32 to −0.05	0.0093	0.026
C9	Complement component C9	0.18	0.04 to 0.31	0.011	0.027
ALB	Serum albumin	−0.08	−0.15 to −0.02	0.014	0.031
CRISP3	Cysteine-rich secretory protein 3	−0.32	−0.61 to −0.04	0.023	0.049
APOA1	Apolipoprotein A-I	−0.12	−0.24 to −0.01	0.026	0.050
PRSS3	Trypsin-3	0.27	0.03 to 0.51	0.029	0.051
YTHDC2	Probable ATP-dependent RNA helicase YTHDC2	−0.12	−0.24 to −0.01	0.034	0.058
PGLYRP2	N-acetylmuramoyl-l-alanine amidase	−0.13	−0.25 to −0.01	0.038	0.057
CLEC3B	Tetranectin	−0.14	−0.27 to −0.01	0.044	0.062
APOA2	Apolipoprotein A-II	−0.12	−0.23 to −0.002	0.047	0.062
A2M	Alpha-2-macroglobulin	0.16	0.0 to 0.32	0.047	0.062
CTBS	Di-N-acetylchitobiase	0.13	−0.05 to 0.31	0.147	0.184
HP	Haptoglobin	0.14	−0.06 to 0.34	0.180	0.214
FGG	Fibrinogen gamma chain	0.06	−0.06 to 0.18	0.327	0.371
FBLN1	Fibulin-1	0.05	−0.06 to 0.17	0.351	0.381
IGKV1D-13	Ig kappa variable 1D-13	0.11	−0.30 to 0.52	0.603	0.628
KNG1	Kininogen-1	−0.006	−0.08 to 0.07	0.874	0.873

Definition of abbreviations: CI = confidence interval; FDR = false discovery rate; PrePF = preclinical pulmonary fibrosis.

Proteins found to be significant in the analysis of subjects with idiopathic pulmonary fibrosis versus those without pulmonary fibrosis were examined in the plasma of subjects with PrePF versus those without pulmonary fibrosis. Analysis was controlled for age, sex, and family relatedness in a linear mixed-effects model; raw P values are listed as well as adjustment for multiple testing. Proteins with FDR < 0.05 are italicized.

Using the cor function in R and using a cutoff of 0.5, we found two correlated proteins (GSN and S100A8) and removed them from predictive modeling. Plasma samples were reviewed to create a data set with only one member per family while maximizing cases of PrePF, leaving 31 first-degree relatives with PrePF and 99 without evidence of lung fibrosis. The 12 proteins significant among subjects with PrePF were included in predictive modeling. When compared with a model using age and sex alone, including the top four proteins (S100A9, LBP, CRISP3, and CRKL) improved the model performance based on AUC. The AUC for the model including age, sex, and the four proteins was 0.86 (95% confidence interval [CI], 0.82–0.89) versus 0.77 (95% CI, 0.72–0.82) for the model using only age and sex; the lack of overlap in 95% CIs for the AUCs indicates improved predictive utility for the model including the four proteins (Figure 1). Adding MUC5B genotype to the models did not improve predictive ability (AUC, 0.79; 95% CI, 0.74–0.83). Adding MUC5B genotype to the aforementioned four proteins with age and sex did not improve the AUC (0.82; 95% CI, 0.78–0.86).

Figure 1.

Predictive model for preclinical pulmonary fibrosis using top plasma proteins and patient characteristics. When compared with a model utilizing age and sex alone, including the top four proteins (S100A9, LBP, CRISP3, and CRKL) in addition to age and sex in a predictive model for preclinical pulmonary fibrosis improved the receiver operating characteristic curve performance based on comparing areas under the curve (AUCs). The AUC for the model including age, sex, and the four proteins was 0.86 (95% confidence interval [CI], 0.82–0.89; sensitivity, 0.77; specificity, 0.90; blue line) versus the AUC of 0.77 (95% CI, 0.72–0.82; sensitivity, 0.68; specificity, 0.89; black line) for the model using only age and sex. The negative predictive value and the positive predictive values of the age and sex model were 0.90 and 0.66 versus 0.93 and 0.70 for the model including age, sex, and protein levels. Adding MUC5B genotype to age and sex did not significantly improve the predictive ability of the model (red line) compared with including only age and sex (black line) or including the top four proteins (blue line).

To interrogate the consistency of the findings in a different blood sample, serum samples from first-degree relatives with PrePF (n = 26) and subjects without fibrosis (n = 129) were analyzed in a similar fashion to plasma proteins. Ten of the previously discovered 12 proteins were able to be detected in serum samples; S100A9 and S100A8 could not be measured in serum and so could not be compared. Nine of these 10 serum proteins showed consistent changes in directionality. Seven of those nine approached statistical significance but did not meet an FDR < 0.05 (ALB, GSN, C9, LBP, CRISP3, CRKL, SPARC); one did reach significance (C1QC, FDR = 0.02) (Table 3).

Table 3.

Serum Protein Analyses, PrePF versus No Fibrosis

Protein	Coefficient	P Value	FDR	Same Direction as Plasma?
ALB	−0.07	0.04	0.07	Yes
APOA4*	0.06	0.34	0.40	No
GSN	−0.09	0.04	0.08	Yes
C9	0.18	0.06	0.09	Yes
LBP	0.20	0.03	0.07	Yes
C1QC	−0.14	0.002	0.02	Yes
CRISP3	−0.32	0.04	0.07	Yes
BASP1	−0.04	0.56	0.58	Yes
CRKL	−0.13	0.08	0.12	Yes
SPARC	0.27	0.01	0.05	Yes

Definition of abbreviations: FDR = false discovery rate; PrePF = preclinical pulmonary fibrosis.

Of the 12 significant proteins identified in plasma analysis, 10 were able to be detected in serum samples to allow for comparison between groups. Analysis was controlled for family relatedness.

^*Indicates different directionality than in the plasma samples.

Discussion

Circulating proteins have been associated with IPF but are not in clinical use (12). We focused on a high-risk cohort, first-degree relatives of patients with FIP, and found that in addition to age and sex, circulating proteins (S100A9, LBP, CRISP3, and CRKL) may be useful in identifying subjects with PrePF.

The identification of PrePF may play an important role in the development of clinical care of pulmonary fibrosis because other investigators have illustrated that ILAs in first-degree relatives of both patients with FIP and subjects with IPF progress (6, 13, 14). Clinically, how to address PrePF and/or ILAs is an important question because approved medical therapies (nintedanib and pirfenidone) slow down disease progression but do not reverse existing fibrosis. Therefore, there is rationale to study the role of early treatment in this disease before patients develop irreversible lung fibrosis.

One limitation of this study is that the subjects included in these analyses were not true “control subjects”—those without disease were first-degree relatives from families with FIP. As numerous studies have now illustrated (3, 7), first-degree relatives from families with FIP are at high risk for developing abnormal lung parenchyma. However, the “No Fibrosis” family members included in this investigation were those that did not have radiologic evidence of lung fibrosis. Though this may be considered a limitation of study design, we believe that this would bias our study toward the null hypothesis and would not lead to false-positive findings.

This study is also limited by the lack of a validation cohort, and validation in independent cohorts are required before these findings can be generalized. Further validation is particularly important because serum data showed consistent trends for most but not all of the plasma protein findings.

In conclusion, circulating plasma proteins are differentially detected in IPF, and some are common to subjects with IPF and PrePF. Further study and validation of these findings in independent cohorts is necessary.