Serum peptides as candidate biomarkers for relapsing polychondritis

Highlights

• •Identified serum peptides useful for diagnosis of relapsing polychondritis (RP)
• •The peptides are fragments of fibrinogen α.
• •A candidate diagnostic peptide biomarker was 83.3% sensitive and 71.7% specific.

Abstract

Introduction

Relapsing polychondritis (RP) is an intractable disease characterized by recurrent inflammation of cartilaginous tissue throughout the body. It is difficult to accurately diagnose RP, and no useful biomarkers have yet been identified.

Objectives

We analyzed serum peptide profiles to identify novel candidate biomarkers for RP.

Methods

Thirty-seven patients with RP, 42 patients with rheumatoid arthritis (RA), and 35 healthy control (HC) subjects were divided into training and testing sets. Seven patients demonstrating granulomatosis with polyangiitis (GPA) were used for validation. The ion intensity of serum peptides was comprehensively measured by matrix-assisted laser desorption/ionization time-of-flight/time-of-flight mass spectrometry and applied to a supervised multivariate analysis. Peptides of interest were analyzed by liquid chromatography-tandem mass spectrometry.

Results

In the training set, models developed based on 11 (RP/HC-11P model), 9 (RP/RA-9P model), and 14 (RP/nonRP-14P model) peptides, out of 160 peptides detected were able to completely discriminate the RP group from the HC, RA, and nonRP (HC + RA) groups. Almost all of the 15 identified discriminatory peptides comprising these models were fragments of proteins associated with coagulation. Four models, each consisting of 4 out of 10 identified peptides of the RP/nonRP-14P model (models RP/nonRP-4P-2, -10, -11, and -38), provided ≥ 70.0 % sensitivity and specificity when applied to the validation set (the testing set and the GPA group) (AUROC, 0.779–0.815). Notably, the RP/nonRP-4P-2 model provided 83.3 % sensitivity and 71.7 % specificity in the validation set (AUROC, 0.802).

Conclusions

Serum peptides are useful as candidate biomarkers for discriminating RP and may be involved in the pathophysiology of RP.

Introduction

Relapsing polychondritis (RP) is an autoimmune disease characterized by recurrent and progressive inflammation of cartilaginous tissue throughout the body [1,2,3]. Diagnostic criteria for RP have been proposed [4,5] in which the diagnosis is based on the inflammatory symptoms affecting six tissue or organs: the pinna, joints, nose, eyes, respiratory tract, and inner ear, as well as histological findings. However, it is impractical to accurately diagnose RP in the early phase of the disease, before the typical symptoms and findings have developed. The differential diagnosis of RP from rheumatoid arthritis (RA) and granulomatosis with polyangiitis (GPA) is also difficult especially in cases with polyarthritis or saddle nose. A considerable percentage of RP patients are cytoplasmic-anti-neutrophil cytoplasmic antibody (C-ANCA) positive or PR3-ANCA positive [6,7,8], which further complicates differential diagnosis.

At present, diagnostic biomarkers for RP have not been established. Thus far, autoantibodies to type II collagen [9] and matrilin 1 [10] have been proposed as candidate biomarkers. Serum proteins such as cartilage oligomeric matrix protein [11], macrophage migration inhibitory factor [12], and soluble triggering receptor expressed on myeloid cells-1 [13], as well as urinary type II collagen neoepitope [14] have been reported as putative markers of RP activity. However, the clinical use of these proteins as biomarkers for RP has not been realized because they are frequently detected in other diseases (e.g., RA and GPA) [9,12,13] or because further validation studies are needed [11,14].

Here, we comprehensively analyzed serum peptides in effort to detect a novel and disease-specific biomarker for RP. Although 99.0 % of the serum proteome consists of 20 proteins, including albumin and immunoglobulin, the remaining 1 % comprises of a diverse spectrum of known and unknown proteins and peptides at low concentrations, which could serve as an excellent source of biomarkers [15]. Previously, we detected serum peptide(s) that may function as disease-specific biomarkers for microscopic polyangiitis, dementia with Lewy bodies, and other conditions [16,17,18,19]. Herein, we report on the identification of novel candidate biomarker panels for RP, each consisting of four serum peptides. We are hopeful that our analysis will provide useful candidates for the future development of a blood biomarker for the diagnosis of RP.

Materials and methods

Patients

Eighty-seven patients with RP, RA, or GPA and 37 healthy control (HC) subjects were enrolled in this clinical study (Supplementary Table 1). The diagnoses of RP [4,5], RA [20,21], and GPA [22,23] were made according to their respective criteria. Subjects who were not suffering from rheumatic diseases, or any moderate to severe diseases, were selected as HC subjects. This research was approved by the Ethics Committee of St. Marianna University School of Medicine (approval number 3864). Peripheral blood was obtained from the above participants with their informed consent. The study protocol conformed to the ethical guidelines of the Declaration of Helsinki revised in 2013.

The analysis of serum peptide profiles

Serum peptides were purified by weak cation exchange (MB-WCX, Bruker Daltonics, Ettlingen, Germany). The ion intensity of peptides was measured using a matrix-assisted laser desorption/ionization time-of-flight/time-of-flight mass spectrometer (Ultraflex I, UltrafleXtreme; Bruker Daltonics; Billerica, MA, USA). The ion intensity was compared among the RP, HC, and RA groups by ClinProTools (Bruker Daltonics). Peptides were designated as p + peptide mass/charge ratio (i.e., p1206 indicates a peptide with 1206 m/z). Amino acid sequences of peptides were analyzed using a nano-HPLC system (Eksigent nano-LC 400 system, AB Sciex, Framingham, MA, USA) connected to a Triple TOF5600 system (AB Sciex).

Statistical analysis

The significance of differences in peptide ion intensity between the RP group and the HC, RA, or HC + RA (nonRP) group was calculated using Student’s t-test. Multivariate analyses of a principal component analysis (PCA) and an orthogonal partial least square-discriminant analysis (OPLS-DA) were performed using the SIMCA-P⁺ software program (ver. 14; Umetrics, Umea, Sweden). A receiver operating characteristic (ROC) analysis was conducted using the JMP software program (ver. 9.0.2, SAS Institute, Cary, NC, USA). The significance of the prediction values of the biomarker models for RP was calculated using a Wilcoxon signed-rank test.

Results

Cohort setting for generation and validation of candidate biomarkers

In this study, we aimed to discriminate RP from healthy condition and other inflammatory diseases such as RA and GPA using serum peptide profiles. The 117 cases and controls (RP, n = 37; HC, n = 37; RA, n = 43) were divided into the 1st cohort (RP, n = 19; HC, n = 19; RA, n = 22; total, n = 60) and the 2nd cohort (RP, n = 18; HC, n = 18; RA, n = 21; total, n = 57) (Supplementary Table 1). To generate useful models, we used the 1st cohort as the parent cohort for the training set, excluding atypical cases as described below. All cases of the 2nd cohort were used as a testing set to validate the generated models. Furthermore, the testing set together with the seven GPA patients (the validation set) was also used for the validation to examine the usefulness of the models for a group with GPA.

A total of 160 peptides were detected from the 60 serum samples of the 1st cohort (Supplementary Fig. 1). To exclude atypical RP cases from the training set, serum peptide profiles of individual RP patients, HC subjects, and RA patients in the 1st cohort were analyzed by a PCA and compared in each group. All RP patients showed relatively similar serum peptide profiles (Fig. 1A). In contrast, two HC subjects (HC12, HC13) and one RA patient (RA22) were identified as outliers from the results of PCA (Fig. 1B, C). Excluding these three cases from the 1st cohort, 57 patients and subjects were used as the training set (Table 1, Fig. 2).

Fig. 1.

Score scatter plots from PCA using serum peptide ion intensity. Results of the RP group (A), the HC group (B), the RA group (C), and all three groups (D) in the training set. The two individuals in the HC group and one patient in the RA group who were identified as outliers are specifically indicated by their ID numbers.

Table 1.

Clinical information of the subjects.

	RP (n = 37)			RA (n = 42)			GPA (n = 7)			HC (n = 35)
	Training	Testing		Training	Testing					Training	Testing
	(n = 19)	(n = 18)		(n = 21)	(n = 21)					(n = 17)	(n = 18)
Male: Female	7:12	7:11		6:15	6:15		2:5			7:10	6:12
Age# (y)	52.6 ± 10.3	49.9 ± 13.3		66.4 ± 12.4	65.9 ± 10.1		58.0 ± 18.9			52.0 ± 11.6	52.4 ± 14.3
CRP (mg/dL)	0.26 ± 0.51	1.67 ± 3.52		0.36 ± 0.77	0.41 ± 0.58		0.60 ± 1.3			NA	NA
ESR (mm/h)	13.6 ± 9.7	25.5 ± 31.8		20.8 ± 18.7	20.5 ± 14.8		16.3 ± 11.3			NA	NA
Disease duration# (y)	4.5 ± 4.4	6.5 ± 5.4		13.8 ± 9.2	13.1 ± 6.3		13.3 ± 12.3			NA	NA
Medication	33 PSL, 12 CYA, 6 MTX,			24 PSL, 10 BUC, 9 LOX,			4 PSL, 2 MMF, 2 RTX,			NA	NA
	5 COL, 2 TAC, 2 IFX,			7 SASP, 6 TAC, 5 ETN,			1 MTX, 1 TAC, 1 LOX,
	1 AZP, 1 MMF			2 CEL, 1 NPI, 1 ETO,			1 AZP
				1 DIC, 1 COL

Information of the subjects in the training and testing sets and that of 7 GPA patients is shown.

RP, relapsing polychondritis; RA, rheumatoid arthritis; HC, healthy control; GPA, granulomatosis with polyangiitis; y, years; ESR, erythrocyte sedimentation rate; CRP, C-reactive protein; PSL, prednisolone; CYA, cyclosporin; MTX, methotrexate; COL, colchicine; TAC, tacrolimus; IFX, infliximab; AZP, azathioprine; MMF, mycophenolate mofetil; BUC, bucillamine; LOX, loxoprofen; SASP, salazosulfapyridine; ETN, etanercept; CEL, celecoxib; MEL, meloxicam; NPI, rebamipide; ETO, etodolac; DIC, diclofenac; RTX, Rituximab; NA, not applicable.

^#Means ± standard deviations are shown.

Fig. 2.

Serum peptide profiles of the RP, HC, and RA groups in the training set.

One hundred sixty serum peptide ion peaks were detected from 19 RP patients, 17 HC subjects, and 21 RA patients in the training set. The average ion intensity of the peptides in the RP, HC, and RA group is shown.

Discrimination of the RP group using the ion intensity of single peptides and an unsupervised multivariate analysis of peptide ion intensity

For the generation of a biomarker model for RP, we attempted to discriminate the RP group from the other groups based on the ion intensity of serum peptides. To evaluate this possibility, we examined the number of peptides that showed a difference in ion intensity between the RP group and other groups. As a result, 27, 9, and 9 peptides showed a ≥ 1.2-fold difference in ion intensity between the RP and HC groups, between the RP and RA groups, and between the RP and HC + RA (nonRP) groups, respectively (p < 0.05) (Table 2). The serum peptide profile of RP differed from those of HC and RA. Based on these results, we generated biomarker models for RP based on the peptide ion intensity.

Table 2.

Comparison of peptide ion intensity.

Peptide ion		Differences (folds)		Number of peptides
intensity		(p < 0.05)		RP/HC	RP/RA	RP/(HC + RA)
Increased		2.0≤x		1	0	0
		1.5≤x＜2.0		1	0	0
		1.2≤x＜1.5		11	4	2
		-1.2＜x＜1.2		2	2	4
Decreased		-1.5＜x≤-1.2		10	5	7
		-2.0＜x≤-1.5		4	0	0
		x≤-2.0		0	0	0

The numbers of peptides for which the ion intensity was changed in the RP group in comparison to the HC group, the RA group, and the HC + RA group in the training set (p < 0.05).

First, we attempted to discriminate the RP group from the other groups using the ion intensity of single peptides. Nine peptides discriminated the RP group from the HC group with ≥ 65.0 % sensitivity and specificity both in the training and testing sets; however, no combinations of peptides were found that discriminated between the RP and RA groups and between the RP and nonRP groups with ≥ 65.0 % sensitivity and specificity in the testing set (Supplementary Fig. 2AB). It was difficult to discriminate the RP group from the nonRP group using ion intensity of single peptides.

Next, we attempted to discriminate between the RP group and the other groups using a multivariate analysis of peptide ion intensity. An unsupervised multivariate analysis of PCA was applied for discrimination. The ion intensity of all 160 peptides were used for discrimination between the RP and HC groups, the RP and RA groups, and the RP and nonRP groups in the training set. However, the PCA did not discriminate the RP group from any of the other three groups (Fig. 1D).

Generation of models to discriminate between the RP and HC groups and between the RP and RA groups by a multivariate analysis of peptide ion intensity

We subsequently performed an OPLS-DA as a supervised multivariate analysis to analyze the discriminative ability for RP. The ion intensity of all 160 peptides successfully discriminated between the RP and HC groups (RP/HC-160P model, Supplementary Fig. 3A–D). The R²Y value of this model (0.829) indicated its high discriminative ability in the present cohort, whereas the Q² value (0.173) indicated a relatively low predictive ability. To improve the predictive ability, we attempted to generate a new discriminant model by selecting the minimum number of peptides needed for complete discrimination. To achieve this, S-plot parameters (the x- and y-axes showed the magnitude and reliability of the peptides in the model, respectively, Supplementary Fig. 3C) and the variable importance in projection (VIP, x-axis indicated the contribution of each peptide to the generation of the model [average value is 1.0], Supplementary Fig. 3D) were used for selection. As a result, we found that only 11 peptides were necessary for complete discrimination between the RP and HC groups (RP/HC-11P model, Supplementary Fig. 3E–H). This RP/HC-11P model showed a higher Q² value (0.431); thus, the predictive ability was improved by directed selection of discriminatory peptides (Supplementary Fig. 3E).

Similarly, to discriminate the RP and RA groups in the training set, the ion intensity of all 160 peptides was subjected to an OPLS-DA (RP/RA-160P model, Supplementary Fig. 4A–D). The RP/RA-160P model completely discriminated between the two groups (Supplementary Fig. 4A). This model provided a high R²Y value (0.843), but a low Q² (0.191), indicating low predictive ability. Selecting peptides with high magnitude, reliability (Supplementary Fig. 4C), and VIP (Supplementary Fig. 4D) in the RP/RA-160P model, we generated a new model in which the minimum number of peptides for complete discrimination was nine (RP/RA-9P model, Supplementary Fig. 4E–H). The Q² value of this model improved (0.471), again showing improved predictive ability through directed peptide selection (Supplementary Fig. 4E).

Generation of models to discriminate between the RP and nonRP groups by a multivariate analysis of peptide ion intensity

We finally attempted to discriminate between the RP and nonRP groups in the training set by an OPLS-DA of the ion intensity of all 160 peptides (RP/nonRP-160P model, Fig. 3A–D). This RP/nonRP-160P model completely discriminated the two groups and provided a high R²Y (0.781), but a low Q² (0.151) (Fig. 3A). Therefore, we again selected peptides with high magnitude, reliability (Fig. 3C), and VIP (Fig. 3D) in the RP/nonRP-160P model to generate a better model with the minimum number of peptides. As a result, a combination of 14 peptides completely discriminated the two groups (RP/nonRP-14P model, Fig. 3E–H). The RP/nonRP-14P model provided a relatively high R²Y value (0.639); however, the Q² value was still low (0.297), suggesting a low predictive ability (Fig. 3E).

Fig. 3.

Models to discriminate between the RP and nonRP groups.

Two models were generated using the peptide ion intensity of the 19 RP patients and the non-RP subjects (HC subjects, n = 17; and RA patients, n = 21) in the training set. First, the ion intensity values of the 160 peptides were subjected to an OPLS-DA (RP/nonRP-160P model, A-D). A, T he score scatter plot of the RP/nonRP-160P model. The x axis indicates the first principal component for the discrimination. B, The loading scatter plot of the RP/nonRP-160P model. Parameters localized further away from the center of the x-axis contribute more to the discrimination. C, The S-plot of the RP/nonRP-160P model. The magnitude (x axis) and reliability (y axis) of the peptides are visualized. D, VIP of the RP/nonRP-160P model. Thirty-six of the 160 peptides relatively highly contributed to the generation of this model (VIP >1.0). The second model between the RP and nonRP groups was generated using a minimum of 14 peptides (RP/nonRP-14P model, E-H). The score scatter plot (E), loading scatter plot (F), S-plot (G), and VIP (H) of the RP/nonRP-14P model are shown. The numbers in brackets in H indicate the order of the VIP scores of the peptides in the RP/nonRP-160P model (D).

Generation of models to discriminate between the RP and nonRP groups by a multivariate analysis of the ion intensity of a small number of peptides

To improve the predictive ability of the models to discriminate between the RP and nonRP groups, we selected peptides from the 14 peptides of the RP/nonRP-14P model. To make this selection, we subjected the 14 peptides to tandem mass spectrometry (MS/MS) analysis, and were able to conclusively identify 10 peptides (Table 3). We generated a discriminant model for RP by an OPLS-DA using the ion intensity of all 10 peptides (RP/nonRP-10P model). The RP/nonRP-10P model did not completely discriminate between the RP and nonRP groups, however, it provided a sensitivity of 100 %, an adequate specificity of 65.8 %, and a AUROC of 0.892 in the training set (Table 4). To further select the peptides, we pared down the number of peptides in the models, and attempted to discriminate between the RP and nonRP groups by an OPLS-DA, retaining ≥ 65.0 % sensitivity and specificity.

Table 3.

Identification of characteristic serum peptides in RP.

Peptides*	Ion int diff	Models#	Proteins	Accession ID	MW (Da)*		Confirmed sequences	Modification$
	(folds)				Observed	Theoretical
p1519	–	RP/RA-9P	Fibrinogen alpha chain	FIBA_HUMAN	1519.36	1517.67	20ADSGEGDFLAEGGGVR35	Dehydration (Ser)
		RP/nonRP-14P		(gi|1706799)
p1546	–	RP/HC-11P	Fibrinogen alpha chain	FIBA_HUMAN	1545.50	1544.61	21DSGEGDFLAEGGGVR35	Phosphorylation (Ser)
				(gi|1706799)
p1617	–	RP/HC-11P	Fibrinogen alpha chain	FIBA_HUMAN	1617.02	1615.65	20ADSGEGDFLAEGGGVR35	Phosphorylation (Ser)
		RP/RA-9P		(gi|1706799)
		RP/nonRP-14P
p2555	–	RP/nonRP-14P	Fibrinogen alpha chain	FIBA_HUMAN	2555.41	2555.09	576SSSYSKQFTSSTSYNRGDSTFES598	–
				(gi|1706799)		2553.08	576SSSYSKQFTSSTSYNRGDSTFES598	Deamidation (Asn)
p2661	–	RP/HC-11P	Fibrinogen alpha chain	FIBA_HUMAN	2660.93	2658.25	605DEAGSEADHEGTHSTKRGHAKSRPV629	–
				(gi|1706799)
p4091	–	RP/RA-9P	Fibrinogen alpha chain	FIBA_HUMAN	4090.90	4089.88	592GDSTFESKSYKMADEAGSEADHEGTHS	–
		RP/nonRP-14P		(gi|1706799)			TKRGHAKSRPV629
p4210	–	RP/HC-11P	Prothrombin	THRB_HUMAN	4209.71	4207.15	329FGSGEADCGLRPLFEKKSLEDKTERELL	Oxidation (Phe),
		RP/RA-9P		(gi|135807)			ESYIDGR363	Farnesylation (Cys)
		RP/nonRP-14P
p6628	–	RP/RA-9P	Apolipoprotein C-I,	APOC1_HUMAN	6628.00	6626.51	27TPDVSSALDKLKEFGNTLEDKARELISRI	–
			isoform CRA_a	(gi|119577712)			KQSELSAKMREWFSETFQKVKEKLKIDS83
p1466	2.08	RP/HC-11P	Fibrinogen alpha chain	FIBA_HUMAN	1465.81	1464.65	21DSGEGDFLAEGGGVR35	–
	(/HC)	RP/RA-9P		(gi|1706799)
		RP/nonRP-14P
p2770	1.44	RP/HC-11P	Fibrinogen alpha chain	FIBA_HUMAN	2770.02	2767.22	576SSSYSKQFTSSTSYNRGDSTFESKS600	–
	(/HC)			(gi|1706799)		2768.20	576SSSYSKQFTSSTSYNRGDSTFESKS600	Deamidation (Asn)
p3193	-1.49	RP/HC-11P	Fibrinogen alpha chain	FIBA_HUMAN	3192.50	3189.42	576SSSYSKQFTSSTSYNRGDSTFESKSYKM603	–
	(/HC)			(gi|1706799)		3190.40	576SSSYSKQFTSSTSYNRGDSTFESKSYKM603	Deamidation (Asn)
	-1.34
	(/HC+RA)
p3264	-1.72	RP/nonRP-14P	Fibrinogen alpha chain	FIBA_HUMAN	3263.69	3260.46	576SSSYSKQFTSSTSYNRGDSTFESKSYKMA604	–
	(/HC)			(gi|1706799)		3261.44	576SSSYSKQFTSSTSYNRGDSTFESKSYKMA604	Deamidation (Asn)
p3952	1.35	RP/HC-11P	DNA, FLJ93141, highly	B2R6V9_HUMAN	3951.63	3948.97	2SETSRTAFGGRRAVPPNNSNAAEDDLPTV	–
	(/HC)	RP/nonRP-14P	similar to Homo sapiens	(gi|164690537)			ELQGVVPR38
			coagulation factor XIII,
			A1 polypeptide (F13A1)
p5335	-1.43	RP/HC-11P	Fibrinogen alpha chain	FIBA_HUMAN	5335.32	5333.35	576SSSYSKQFTSSTSYNRGDSTFESKSYKMA	–
	(/HC)	RP/nonRP-14P		(gi|1706799)			DEAGSEADHEGTHSTKRGHA624
p5902	-1.64	RP/HC-11P	Fibrinogen alpha chain	FIBA_HUMAN	5902.29	5900.70	576SSSYSKQFTSSTSYNRGDSTFESKSYKMA	–
	(/HC)	RP/nonRP-14P		(gi|1706799)			DEAGSEADHEGTHSTKRGHAKSRPV629
p1206	1.60	–	Fibrinogen alpha chain	FIBA_HUMAN	1206.00	1204.62	25GDFLAEGGGVR35	Lys-add at N-term
	(/HC)			(gi|1706799)		1205.57	24EGDFLAEGGGVR35	–
p2933	-1.23	–	Fibrinogen alpha chain	FIBA_HUMAN	2930.28	2933.00	576SSSYSKQFTSSTSYNRGDSTFESKSY601	–
	(/RA)			(gi|1706799)
p3443	1.34	–	Alpha-2-antiplasmin	SERPINF2_HUMAN	3442.61	3440.65	462GFPRGDKLFGPDLKLVPPMEEDY	Sulfonylation (Tyr)
	(/HC)			(gi|31340775)			PQFGSPK491
	1.27
	(/HC+RA)
p3936	1.26	–	Fibrinogen alpha chain	FIBA_HUMAN	3935.54	3933.80	594STFESKSYKMADEAGSEADHEGT	Oxidation (Met)
	(/HC)			(gi|1706799)			HSTKRGHAKSRPV629

*Mass/charge ratios used for the designation of peptides were obtained by MALDI-TOF/TOF MS, whereas those for “observed MW” were obtained by LC-MS.

^#Peptides for which showed a ≥ 1.2-fold difference in ion intensity was observed between the RP and HC groups (/HC), between the RP and RA groups (/RA), or between the RP and HC + RA groups (/HC + RA) in the training set (p < 0.05), or that were used in the RP/HC-11P model, RP/RA-9P model, or RP/nonRP-14P model, were selected for identification. (The HC + RA group is the nonRP group.).

^$Possible modifications deduced from the results of the LC-MS analysis. Letters in parentheses show the targeted amino acids.

HC, healthy control; MW, molecular weights; MALDI-TOF/TOF MS, matrix-assisted laser desorption/ionization time-of-flight/time-of-flight mass spectrometry; LC–MS, liquid chromatography–mass spectrometry; RA, rheumatoid arthritis; RP, relapsing polychondritis.

Table 4.

Generation and validation of the models to discriminate between the RP and nonRP groups.

RP/nonRP-160P		Training set (RP 19, HC 17, RA 21)					Testing set (RP 18, HC 18, RA 21)					Validation set (RP 18, HC 18, RA 21, GPA7)
				Sens.	Spec.	AUROC			Sens.	Spec.	AUROC			Sens.	Spec.	AUROC
				100	100	1.000			55.6	56.4	0.662			55.6	58.7	0.669
		RP	nonRP				RP	nonRP				RP	nonRP, GPA
	RP	19	0				10	17				10	19
Cutoff (AU)	nonRP	0	38				8	22
>=0.627752	nonRP, GPA											8	27
RP/nonRP-14P		Training set (RP 19, HC 17, RA 21)					Testing set (RP 18, HC 18, RA 21)					Validation set (RP 18, HC 18, RA 21, GPA7)
1466, 1519, 1617, 1755,				Sens.	Spec.	AUROC			Sens.	Spec.	AUROC			Sens.	Spec.	AUROC
2555, 3264, 3366, 3952,				100	100	1.000			77.8	35.9	0.609			77.8	39.1	0.650
4091, 4210, 5335, 5859, 5902, 7761		RP	nonRP				RP	nonRP				RP	nonRP, GPA
	RP	19	0				14	25				14	28
Cutoff (AU)	nonRP	0	38				4	14
>=0.4802435	nonRP, GPA											4	18
RP/nonRP-10P		Training set (RP 19, HC 17, RA 21)					Testing set (RP 18, HC 18, RA 21)					Validation set (RP 18, HC 18, RA 21, GPA7)
1466, 1519, 1617, 2555,				Sens.	Spec.	AUROC			Sens.	Spec.	AUROC			Sens.	Spec.	AUROC
3264, 3952, 4091, 4210,				100	65.8	0.892			77.8	30.8	0.672			77.8	30.4	0.655
5335, 5902		RP	nonRP				RP	nonRP				RP	nonRP, GPA
	RP	19	13				14	27				14	32
Cutoff (AU)	nonRP	0	25				4	12
>=0.25268	nonRP, GPA											4	12
RP/nonRP-4P-2		Training set (RP 19, HC 17, RA 21)					Testing set (RP 18, HC 18, RA 21)					Validation set (RP 18, HC 18, RA 21, GPA7)
1466, 1617, 3264, 4091				Sens.	Spec.	AUROC			Sens.	Spec.	AUROC			Sens.	Spec.	AUROC
				79.0	65.8	0.789			83.3	74.4	0.823			83.3	71.7	0.802
		RP	nonRP				RP	nonRP				RP	nonRP, GPA
	RP	15	13				15	10				15	13
Cutoff (AU)	nonRP	4	25				3	29
>=0.327539	nonRP, GPA											3	33
RP/nonRP-4P-10		Training set (RP 19, HC 17, RA 21)					Testing set (RP 18, HC 18, RA 21)					Validation set (RP 18, HC 18, RA 21, GPA7)
1466, 3264, 4091, 5335				Sens.	Spec.	AUROC			Sens.	Spec.	AUROC			Sens.	Spec.	AUROC
				100	65.8	0.823			72.2	74.4	0.819			72.2	71.7	0.808
		RP	nonRP				RP	nonRP				RP	nonRP, GPA
	RP	19	13				13	10				13	13
Cutoff (AU)	nonRP	0	25				5	29
>=0.314752	nonRP, GPA											5	33
RP/nonRP-4P-11		Training set (RP 19, HC 17, RA 21)					Testing set (RP 18, HC 18, RA 21)					Validation set (RP 18, HC 18, RA 21, GPA7)
1466, 3264, 4091, 5902				Sens.	Spec.	AUROC			Sens.	Spec.	AUROC			Sens.	Spec.	AUROC
				79.0	65.8	0.795			77.8	71.8	0.782			77.8	71.7	0.779
		RP	nonRP				RP	nonRP				RP	nonRP, GPA
	RP	15	13				14	11				14	13
Cutoff (AU)	nonRP	4	25				4	28
>=0.36134	nonRP, GPA											4	33
RP/nonRP-4P-38		Training set (RP 19, HC 17, RA 21)					Testing set (RP 18, HC 18, RA 21)					Validation set (RP 18, HC 18, RA 21, GPA7)
1617, 3264, 4091, 5335				Sens.	Spec.	AUROC			Sens.	Spec.	AUROC			Sens.	Spec.	AUROC
				84.2	65.8	0.794			77.8	74.4	0.836			77.8	71.7	0.815
		RP	nonRP				RP	nonRP				RP	nonRP, GPA
	RP	16	13				14	10				14	13
Cutoff (AU)	nonRP	3	25				4	29
>=0.30244	nonRP, GPA											4	33

Models to discriminate between the RP and nonRP groups were generated using the training set and validated using the testing set and the validation set (the testing set + the GPA group).

Sens, sensitivity (%); Spec, specificity (%); AUROC, area under the receiver operating characteristic curve.

First, 10 models, consisting of nine peptides each, were generated by excluding one of each of the 10 peptides. Among these, the model with the highest “sensitivity + specificity” value was selected and designated as the RP/nonRP-9P model (Supplementary Table 2). Subsequently, using the highest ranking model for each peptide number, models comprising eight peptides down to one peptide (i.e., RP/nonRP-8P to 1P) were sequentially generated by excluding one peptide from the previous highest performing model. As a result of this analysis, the RP/nonRP-9P to -3P models were identified to all provide ≥ 84.0 % sensitivity, ≥81.0 % specificity, and an AUROC of ≥ 0.895 for the training set (Supplementary Table 2). However, the RP/nonRP-2P model provided lower discriminative ability with 79.0 % sensitivity, 68.3 % specificity, and AUROC of 0.769. Based on these results, we considered the optimal peptide number for providing high discriminative ability for RP to be three or four peptides out of the initial 10.

Thus, we next generated models to discriminate between the RP and nonRP groups by using all combinations of three or four peptides out of the 10 peptides in the training set. A total of 330 models (RP/nonRP-4P-1 to -210 models and RP/nonRP-3P-1 to -120 models) were generated (Supplementary Table 3). We set the criteria for the selection of clinically useful models to be ≥ 65.0 % sensitivity and specificity and an AUROC of ≥ 0.770 (the latter being better than the AUROC of the RP/nonRP-2P model [0.769]). We found that 57 models consisting of four peptides and 12 models consisting of three peptides met the criteria (Supplementary Table 4) and subjected the 69 models to validation on the testing and the validation sets.

Validation of the models to discriminate between the RP and nonRP groups

Finally, we tried to find useful RP/nonRP-discriminating models regardless of the number of component peptides. For this purpose, we validated all the generated RP/nonRP-discriminating models, that is, the RP/nonRP-160P, -14P, -10P to -1P models, the 57 models derived from the RP/nonRP-4P model, and the 12 models derived from the RP/nonRP-3P model using the testing set and the validation set (Table 4; Supplementary Table 2, 4). As a result, four models generated with four peptides discriminated the RP and nonRP groups in both the testing set and the validation set, providing ≥ 70.0 % sensitivity and specificity and an AUROC of ≥ 0.770 (Table 4). These were the RP/nonRP-4P-2, -10, -11, and -38 models. The results of these models are also shown as patient flow diagrams according to the international standard methods [24] (Supplementary Figure 5).

Among the above four models, the RP/nonRP-4P-2 model provided the highest sensitivity of 83.3 %, with the same specificity of 71.7 % as the other three models, and an AUROC of 0.802 in the validation set. Interestingly, all peptides constituting the RP/nonRP-4P-2, -10, -11, and -38 models were the fragments of fibrinogen alpha chain (FIBA) (Table 3). The four models were made of different combinations of four peptides from p1466, p1617, p3264, p4091, p5335, and p5902.

In addition, the validation of the RP/HC-11P model (Supplementary Fig. 3E-H) using the testing set showed relatively high sensitivity (66.7 %) and specificity (72.2 %; AUROC, 0.762), whereas validation of the RP/RA-9P model (Supplementary Fig. 4E-H) revealed high sensitivity (88.9 %) but low specificity (23.8 %; AUROC, 0.585) (Supplementary Table 5).

Identification of the peptides

To examine the difference in the pathophysiology of RP from that of RA and from a healthy physiology, we attempted to identify the peptides that composed the RP/HC-11P model and the RP/RA-9P model in addition to the RP/nonRP-14P model. Peptides for which there was a ≥ 1.2-fold difference in ion intensity between the RP group and the other groups were also subjected to identification (Table 2).

Nineteen peptides were identified (Table 3). Interestingly, 15 out of the 19 peptides were fragments of FIBA. For the other peptides, fragments of prothrombin (p4210), apolipoprotein C-1 (p6628), and α2-antiplasmin (p3443) were identified. Almost all of the identified peptides were derived from proteins related to coagulation and fibrinolysis.

Discussion

In this study, we attempted to generate biomarker models for the diagnosis of RP. Since the RP group exhibited a different peptide profile from the other groups (p < 0.05, Table 2), we were able to generate serum peptide models for the discrimination of RP. A supervised multivariate analysis, OPLS-DA, using the ion intensity of multiple peptides, unlike a univariate analysis that uses the ion intensity of a single peptide and an unsupervised multivariate PCA employing the ion intensity of multiple peptides, completely discriminated the RP group from the HC, RA, and nonRP groups (Supplementary Figs. 2, 3, 4; Fig. 1D, 3). Approximately 10 peptides were found to contribute to the complete discrimination. However, the RP/nonRP-160P and -14P models, which showed complete discrimination in the training set, exhibited low sensitivity and/or specificity (30–60 %) in the testing and validation sets (Table 4), suggesting that useful peptides must be selected from the 14 core peptides that would provide high sensitivity and specificity in the validations.

To establish a clinical laboratory examination using serum peptides, the number of peptides should be reduced to one or, at most, a few, and the amino acid sequences of the peptides should be identified. Accordingly, we attempted to select useful peptides from the 10 identified peptides of the 14 core peptides (Table 4, Supplementary Tables 2-4), and succeeded in creating four biomarker models for RP with ≥ 70.0 % sensitivity and specificity, and an AUROC of ≥ 0.770 in the testing set, using four differently selected peptides (RP/nonRP-4P-2, -10, -11, -38 models) (Table 4). Since the models generated with two or three peptides did not provide ≥ 65.0 % sensitivity and specificity in the testing set (Supplementary Tables 2-4), at least four peptides may be required to discriminate RP from nonRP in the OPLS-DA.

These four models also provided ≥ 70.0 % sensitivity and specificity in the validation set, suggesting their usefulness for the discrimination of RP not only from RA, but also from GPA (Table 4, Supplementary Table 4, Supplementary Fig. 5). In particular, the RP/nonRP-4P-2 model provided 83.3 % sensitivity, 71.7 % specificity, and an AUROC of 0.802 in the validation set, suggesting its high applicability for clinical use. Since no diagnostic biomarkers for RP are currently available, these four RP/nonRP-4P models may become the first clinically useful RP biomarkers. We can simultaneously measure the serum concentration of four peptides by establishing an ELISA for each peptide. Each of the RP/nonRP-4P models demonstrated its own potential usefulness. For example, the RP/nonRP-4P-2 model showed a high discriminative ability for the RP group (83.3 %) and the RA group (61.9 %), while the RP/nonRP-4P-11 model exhibited a high discriminative ability for the GPA group (71.4 %) and the HC group (94.4 %) (Fig. 4). Specifically, the high discriminative ability for HC subjects (94.4 %) would be a significant advantage of the model because healthy individuals constitute the majority of an actual cohort. Since the four RP/nonRP-4P models consist of only six peptides (p1466, p1617, p3264, p4091, p5335, and p5902), quantification of the six peptides and the use of all four RP/nonRP-4P models may increase the accuracy of the discrimination of RP.

Fig. 4.

Validation of the RP/nonRP-4P models.

In each model, the prediction values of individual patients and subjects were calculated as the value of an ideal RP patient as 1.0 and as that of an ideal non-RP patient or subject as 0.0 in the training set. The prediction values of the patients and subjects in the validation set were calculated using the formula obtained by the above calculation of the training set. Dotted lines show the cutoff points (See Table 4). In the RP group, ratios of the cases discriminated as RP are shown. In the RA, GPA, and HC groups, ratios of the patients and subjects discriminated as non-RP are shown. The significance of the differences between the RP group and the other groups is shown (*p < 0.05).

Some biomarkers for inflammatory diseases fluctuate due to disease activity or other conditions in the clinical course [13,16]. Furthermore, biomarkers may be affected by the presence or absence of a particular symptom or the treatment of the disease. To examine the effects of such factors on the four RP/nonRP-4P models, we analyzed correlations between the models and each of the RP-specific symptoms (auricular chondritis, polyarthritis, nasal chondritis, eye lesions, and airway involvement), disease duration, disease activity (CRP values), and the dosage of a drug (prednisolone) in the RP patients (Supplementary Tables 6 and 7). The correlations were examined in the testing set using the prediction values of the four models (shown in Fig. 4), because the prediction values of the RP patients in the training set were calculated as nearly 1.0 to generate the OPLS-DA models in the SIMCA-P⁺ software program. As a result, no correlation was found between the models and any of the aforementioned clinical factors (Supplementary Figs. 6–8 and Supplementary Table 8). It was suggested that the four RP/nonRP-4P models were not affected by any of the disease-related factors, but may reflect more basic, underlying mechanisms of RP. Thus, the four RP/nonRP-4P models may be feasible for the diagnosis of RP in the early and late phases of the disease, in which nonspecific symptoms dominate and the disease has already been treated, respectively.

Interestingly, 15 of the 19 identified peptides that included the six peptides of the four RP/nonRP-4P models were fragments of FIBA (Table 3). All of the 15 FIBA peptides were derived from either fibrinopeptide A (FPA) or the C-terminal subdomain of the αC-domain (αCDC) (Supplementary Fig. 6), which is similar to the results of our previous studies [17,18,19,25]. Thrombin, plasmin, and neutrophil elastase have been reported to cleave FPA and αCDC (Supplementary Fig. 9) [26,27]. The change in the activity of these enzymes may be involved in a complicated manner in the formation of serum peptide profiles of RP. Since not only FIBA-derived but also prothrombin- and α2-antiplasmin-derived peptides were identified (Table 3), it was suggested that dysregulation of blood coagulation and the fibrinolytic system may be involved in the pathophysiology of RP. Indeed, fibrinogen deposits have been detected in RP lesions, including auricular chondritis [28,29]. Based on these present and previous findings, we deduced that: 1) the coagulation system in RP is upregulated to repair micro-wounds caused by inflammation, 2) the production of fibrinogen is increased due to the upregulation of the coagulation system, 3) aberrant fragmentation of fibrinogen occurs in the blood of RP due to the aberrant production and/or activity of the cleavage enzymes, and 4) finally, the excess whole molecule or fragments of fibrinogen deposit onto the inflamed sites of RP, the destination of repair. Further studies are required to confirm this hypothesis. In addition, none of the RP-specific inflammatory symptoms correlated with the RP/nonRP-4P models, suggesting that the involvement of the fibrinolytic system is not specific to a particular type of RP.

One of the limitations of this study is the limited control disease types and the small overall sample size. To demonstrate the usefulness of the RP/nonRP-4P-2, -10, -11, -38 models, it is necessary to validate the models using a large number of cases and controls that include diseases other than RA and GPA. Another limitation is that the comprehensive analysis system of ClinProTools used in this study is a method employed to screen for useful peptides and is not fully optimized for reproducible quantification. Thus, an ELISA for the respective peptides and/or a method for mass spectrometry quantification with internal standard control peptides should be established for the accurate quantification of each peptide in the near future.

Conclusions

We generated four RP/nonRP-4P models as diagnostic candidate biomarkers for RP. Notably, the RP/nonRP-4P-2 model provided 83.3 % sensitivity and 71.7 % specificity in validation. Most of the identified peptides important for the discrimination of RP were fragments of FIBA. Dysregulation of the coagulation system may be involved in the pathophysiology of RP. Further validation of the four RP/nonRP-4P models through the quantification of the peptides using ELISA or mass spectrometry may lead to the establishment of the first useful biomarker for RP.

Ethics Statement

This research was approved by the Ethics Committee of St. Marianna University School of Medicine (approval number 3864) and registered with University Hospital Medical Information Network Clinical Trials Registry (UMIN 000037212). The study protocol conformed to the ethical guidelines of the Declaration of Helsinki revised in 2013. Informed consent was taken from all the participants included in this study.

Funding Support

This work was financially supported by JSPS KAKENHI Grant Number JP 19 K05721 from Japanese Ministry of Education, Culture, Sports, Science and Technology, and St. Marianna University School of Medicine Research Grant.

CRediT authorship contribution statement

Toshiyuki Sato: Writing – review & editing, Writing – original draft, Validation, Investigation, Funding acquisition, Formal analysis. Masaaki Sato: Writing – review & editing, Investigation, Formal analysis. Kouhei Nagai: Writing – review & editing, Investigation, Data curation. Masahiko Fukasawa: Writing – review & editing, Resources, Conceptualization. Yoshiaki Nagashima: Writing – review & editing, Visualization. Teisuke Uchida: Writing – review & editing, Resources. Atsuhiro Tsutiya: Writing – review & editing, Investigation. Kazuki Omoteyama: Writing – review & editing, Investigation. Mitsumi Arito: Writing – review & editing, Investigation. Yukiko Takakuwa: Writing – review & editing, Resources. Seido Ooka: Writing – review & editing, Resources, Data curation. Naoya Suematsu: Writing – review & editing, Investigation. Kimito Kawahata: Writing – review & editing, Resources. Yoshihisa Yamano: Writing – review & editing, Resources. Tomohiro Kato: Writing – review & editing, Supervision, Project administration, Methodology. Manae S. Kurokawa: Writing – review & editing, Writing – original draft, Validation, Supervision, Project administration, Methodology, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article: