Interstitial lung disease (ILD) is characterized by interstitial inflammation of the lung and frequently associated with collagen diseases including rheumatoid arthritis (RA), systemic lupus erythematosus, systemic sclerosis (SSc), polymyositis/dermatomyositis (PM/DM), and polyarteritis nodosa. ILD associated with collagen diseases is then designated collagen vascular disease-associated ILD (CVD-ILD). CVD-ILD is one of the major manifestations of collagen disease and influences the prognosis of collagen disease [1]. A recent study reported that median survival following a diagnosis of ILD associated with RA was only three years [2]. It was also reported that the presence of underlying chronic CVD-ILD is a risk factor for acuteonset drug-induced ILD in RA, when treated with disease-modifying anti-rheumatic drugs. [3]. Acute-onset diffuse ILD (AoDILD) occurs in patients with collagen diseases with or without underlying CVDILD [4,5]. AoDILD includes acute exacerbation of ILD, drug-induced ILD, and Pneumocystis pneumonia. The prognosis of AoDILD is quite poor. Thus, it is important to be able to predict CVD-ILD and AoDILD.

Krebs von den lungen-6 (KL-6) and surfactant protein-D (SPD) are used as surrogate markers for ILD screening. KL-6 is MUC1, a high-molecular-weight glycoprotein, and SP-D belongs to the collectin subgroup of the C-type lectin superfamily. Standard cutoffs for KL-6 and SP-D have been set as the levels that resulted in the optimal diagnosis of idiopathic pulmonary fibrosis (IPF) and CVD-ILD in healthy controls and bacterial pneumonia, and are 500 and 110, respectively [6,7]. Many studies searching for ILD markers have been reported, but few have been validated for CVD-ILD in patients with collagen diseases, including RA. Since levels of KL-6 and SP-D in IPF were considerably higher than in CVD-ILD [6], standard cutoffs for KL-6 and SP-D were too low for the diagnosis of IPF and too high for that of CVD-ILD. Thus, these markers have a lower specificity for the diagnosis of IPF, and a lower sensitivity for the discrimination of CVD-ILD.

It was previously reported that changes in the plasma amino acid profiles of RA and chronic obstructive pulmonary disease (COPD) patients are due to an altered metabolism. A significant decrease of histidine (His) in RA patient plasma was reported [8,9]. Decreased levels of glutamic acid (Glu), asparagine (Asn), glutamine (Gln), and alanine (Ala) in the plasma of COPD patients were also reported [10]. The amino acid profiles of several types of cancer patients were investigated in large sample size and in replicated studies and were applied for screening of cancers. The amino acid metabolism of cancer bearing patients was thought to be altered [11-13]. These extensive studies suggest that the detection of metabolic changes using amino acid profiling may be useful indicators of the presence of CVD-ILD and AoDILD. Here, we investigated plasma amino acid profiles of CVDILD and AoDILD in collagen disease.

Patients and controls

Sixty four Japanese patients with RA were recruited at Sagamihara Hospital. ILD was diagnosed from computed tomography findings. Images were reviewed by two physicians specializing in CVD-ILD, and categorized from A to Z, according to the Sagamihara Criteria [14]. RA cases in categories A to D were diagnosed as “RA with ILD” and those in G and H were diagnosed as “RA without ILD”. RA cases with administration of corticosteroid ≥ 15 mg/day as prednisolone, or in categories E, F, X, Y, or Z were excluded.

Fifteen patients with collagen diseases were admitted to Sagamihara Hospital between 2001 and 2010, because of AoDILD requiring corticosteroid pulse therapy. AoDILD was defined as acute onset and progression within a month, the presence of clinical symptoms (fever, dry cough, or dyspnea), hypoxia, and computed tomography findings of ILD [4,5]. Patients with evidence of apparent bacterial infection or heart disease were excluded. These 15 collagen disease patients with AoDILD include 7 acute exacerbation of ILD and 8 drug-induced ILD. Plasma samples were collected on admission with AoDILD, and in the stable state, at least three months before admission. The patients were classified according to the American College of Rheumatology criteria for RA [15], SSc [16], and Bohan’s criteria for PM/DM [17]. Diagnoses of the patients included 11 RA, 2 SSc, and 2 PM/DM.

This study was reviewed and approved by Sagamihara Hospital Research Ethics Committee. Written informed consent was obtained from all study participants except those already deceased before starting this study. The plasma samples collected before this study were anonymized in a fashion preventing any link with the patients´ identification and their analysis approved on that condition by Sagamihara Hospital Research Ethics Committee. This study was conducted in accordance with the principles expressed in the Declaration of Helsinki.

Plasma amino acid analysis

Blood samples were taken in tubes containing ethylenediaminetetraacetic acid dipotassium salt (Terumo, Tokyo, Japan). Plasma was separated by centrifugation at 1500 x g for 10 min, and then stored at –80°C until use. The plasma amino acid levels were analyzed as previously described [18]. Briefly, plasma samples were deproteinized in a final concentration of 80% acetonitrile and amino acid concentrations were measured by high-performance liquid chromatography/electrospray ionization tandem mass spectrometry after derivatization with 3-aminopyridyl-N-hydroxysuccinimidyl carbamate. An Agilent 1100 series liquid chromatography system (Agilent Technologies, Waldbrunn, Germany) was used. The system was coupled to an Applied Biosystems Sciex API 4000 triple quadrupole mass spectrometer (Applied Biosystems-MDS Sciex, Concord, Canada) equipped with a TurboIonSpray interface. Applied Biosystems-MDS Sciex Analyst 1.3.1 software was used to control these instruments. The following amino acids and related molecules (23 compounds) were measured and used in the analysis: Ala, alpha-aminobutyric acid (a-ABA), arginine (Arg), Asn, citruline (Cit), Glu, Gln, glycine (Gly), His, hydroxyproline (HyPro), isoleucine (Ile), leucine (Leu), lysine (Lys), methionine (Met), ornithine (Orn), phenylalanine (Phe), proline (Pro), serine (Ser), taurine (Tau), threonine (Thr), tryptophan (Trp), tyrosine (Tyr), and valine (Val). Ile, Leu, and Val are branchedchain amino acids (BCAA) and Phe, Trp, and Tyr are aromatic amino acids (AAA). Fischer’s ratio is the ratio between BCAAs (Ile, Leu, and Val) and AAAs (Phe and Tyr) and is used to track the progression of liver fibrosis [19]. His, Ile, Leu, Lys, Met, Phe, Thr, Trp, and Val are designated essential amino acids (EAAs) and the others (Ala, Arg, Asn, Glu, Gln, Gly, Pro, Ser, and Tyr) non-essential amino acids (NEAAs). Total amino acids are the sum of EAAs plus NEAAs. Plasma levels are given as μmol/l.

Statistical analysis

Differences in patient characteristics were analyzed by Mann- Whitney’s U test or chi-square analysis using 2x2 contingency tables. Mann-Whitney’s U test or Wilcoxon signed-rank test were performed in the comparison of laboratory findings and plasma amino acid assay results. GraphPad prism version 5 (Graph Pad Software, San Diego, CA) was used for statistical analyses. To develop support vector machine (svm) indices for ILD, Libsvm version 2.88 [20,21] and Matlab version 7.7.0 (MathWorks, Natick, MA) packages were applied for classification using svms (radial basis function kernel). For leave-one-out cross-validation, training sets consisting of all but the single test samples were used to identify the best index with the highest area under the curve (AUC) value of receiver operating characteristic (ROC) curves.

Diagnosis of ILD in RA

Twenty six RA cases were identified as also suffering from ILD [ILD(+)RA group] and 38 were diagnosed as RA without ILD [ILD(-) RA group] (Table 1). Mean age [ILD(+)RA: ILD(-)RA, 67.6 ± 8.9: 59.7 ± 11.1, year, P=0.0111], percentage of males (38.5: 15.8, P=0.0397), white blood cell count (9.5 ± 3.3: 6.9 ± 1.9, X1000/μl, P=0.0007), c-reactive protein (1.5 ± 1.6: 0.8 ± 1.3, mg/dl, P=0.0148), and rheumatoid factor (757.9 ± 1857.4: 211.6 ± 310.0, IU/ml, P=0.0482) were higher in the ILD(+)RA than ILD(-)RA group. KL-6 (556.3 ± 320.2: 245.1 ± 127.3, U/ml, P<0.0001) and SP-D (78.3 ± 71.6: 40.8 ± 32.4, ng/ml, P=0.0002) were both greatly elevated in the ILD(+)RA group compared with ILD(-)RA group. Sodium ion concentrations (139.5 ± 6.2: 142.9 ± 4.4, mEq/l, P=0.0111) were lower in the ILD(+)RA group.

Table 1: Characteristics of the RA patients.

Plasma amino acid profiles in RA patients

Plasma amino acid profiles in RA patients were analyzed (Table 2). The total amount of all plasma amino acids was comparable in both groups. However, the plasma Lys level (216 ± 45: 193 ± 36, μmol/l, P=0.0154) was significantly greater in the ILD(+)RA group. In addition, the plasma Ser (109 ± 24: 113 ± 29, μmol/l, P=0.8438) and Gly (200 ± 37: 231 ± 71, μmol/l, P=0.2124) levels in ILD(+)RA tended to be lower than in the ILD(-)RA group, but this difference did not reach statistical significance. Thus, the plasma amino acid profiles in ILD(+) RA and ILD(-)RA groups are different.

Table 2: Amino acid profiles of the RA patients.

Validation of diagnostic markers for CVD-ILD in RA patients

KL-6 and SP-D were both greatly elevated in the ILD(+)RA group compared with ILD(-)RA group (Figure 1A, B). The AUC of the KL-6 ROC curve is higher than that of SP-D (Figure 1C). Svm indices were generated from amino acid profiles of RA patients and validated with leave-one-out cross-validation to identify the best index, which was found to be an svm index (AA) comprising Ile, Thr, and Lys. Svm indices were also generated from amino acid profiles together with KL-6 and validated to identify the best index, which in this case was the svm index (AA, KL6) comprising Gly, Gln, and KL-6. We then compared these indices in the ILD(+)RA and ILD(-)RA groups (Figure1D and E). Both the svm indices (AA) and (AA, KL6) were much higher in the ILD(+)RA group. The AUC of the svm (AA, KL6) ROC curve is higher than that of the svm (AA) (Figure 1F).

Specificities and sensitivities of KL-6 and SP-D were estimated from the ROC curves conditional on the highest Youden’s index (Table 3). The optimized cut-off level of KL-6 for CVD-ILD in RA was 331.5 with a higher sensitivity and lower negative likelihood ratio (LR-) compared with its standard cut-off level of 500. The former cut-off level is better for lowering post-test probability of CVD-ILD in marker-negative RA patients. The optimized cut-off level of SP-D was 55.55 for CVDILD in RA with higher sensitivity and lower LR-. Thus, the optimized cut-off levels of KL-6 and SP-D for CVD-ILD in RA were lower than previously set for all ILD, including IPF and CVD-ILD [6].

Table 3: Diagnostic values of markers for ILD in RA patients.

Specificity and sensitivity of svm index (AA) and svm index (AA, KL6) were also estimated from the ROC curves in the same fashion (Table 3). The cut-off level of svm index (AA) was 0.1914 with a better specificity and greater sensitivity than SP-D, but worse than KL-6. The cut-off level of svm index (AA, KL6) was -0.99896, comparable in specificity and higher in sensitivity than KL-6 alone. The positive likelihood ratio (LR+) using svm index (AA, KL6) was also comparable and the LR- was lower than that of KL-6 with a cut-off level of 331.5. Negative predictive values were calculated with 10.7% of pretest probability for CVD-ILD in RA patients who visited Sagamihara Hospital [14]. The negative predictive value of svm index (AA, KL6) was higher than for other markers [96.8% for KL-6 with a cut-off level of 331.5, 95.4% for KL-6 with a cut-off level of 500, and 98.3% for svm index (AA, KL6)]. These data indicate that svm index (AA, KL6) could be a better screening marker for CVD-ILD in RA than KL-6 per se, though svm index (AA) is not.

Characteristics of the collagen disease patients with AoDILD

Characteristics of 15 collagen disease patients with AoDILD are given in Table 4. KL-6 [Stable: AoDILD, 882.2 ± 874.9: 1088.1 ± 865.5, U/ml, P=0.0262] was more increased in AoDILD than in the stable state. Albumin (4.0 ± 0.4: 3.6 ± 0.5, g/dl, P=0.0253) and sodium ion concentrations (143.0 ± 1.8: 139.3 ± 2.5, mEq/l, P=0.0431) were decreased in the AoDILD state compared to these patients in the stable state.

Table 4: Characteristics of the collagen disease patients.

Plasma amino acid profiles in the collagen disease patients with AoDILD

Plasma amino acid profiles in collagen disease patients in each state, i.e. stable and AoDILD, were analyzed (Table 5). The plasma Met (7 ± 6: 13 ± 2, μmol/l, P=0.0175) and Phe (79 ± 19: 95 ± 31, μmol/l, P=0.0035) levels were significantly increased in the AoDILD state, whereas Fischer’s ratio (3 ± 1: 2 ± 1, P=0.0159) was decreased. Thus, the plasma amino acid profiles were different in different states. These data suggest that plasma amino acid profiles could be a better marker for AoDILD in collagen disease.

Table 5: Amino acid profiles of the collagen disease patients.

Several studies have shown that plasma amino acid profiles are altered in various diseases and have explored the possibility of using these to detect disease [8-13]. Few studies have focused on plasma amino acid profiles in CVD-ILD. We attempted to diagnose the presence of CVD-ILD, one of the potentially life-threatening manifestations of collagen diseases, using plasma amino acid profiles. We found that plasma amino acid profiles are no better than KL-6 (Tables 1-3 and Figure 1), as the results of our validation study indicated. However, plasma amino acid profiles could be a better marker for AoDILD in collagen disease patients than KL-6 or SP-D (Tables 4 and 5).

Figure 1: Validation of diagnostic markers for ILD in RA patients. Distribution of KL-6 (A) and SP-D (B). Horizontal bars denote medians. Each horizontal dotted line represents a cut-off level. (C) ROC curves using KL-6 and SP-D as markers for ILD. Svm index (AA) (D) and svm index (AA, KL-6) (E). Horizontal bars denote medians. Each horizontal dotted line represents a cutoff level. (F) ROC curves using svm index (AA) and svm index (AA, KL-6).

To the best of our knowledge, this is the first report of plasma amino acid profiles in CVD-ILD. Plasma concentrations of Lys were significantly higher in the ILD(+)RA group (Table 2, P=0.0154). Plasma concentrations of Met and Phe were significantly higher and Fischer’s ratio was lower in the AoDILD state (Table 5, P=0.0175, 0.0035, and 0.0159, respectively). Liver dysfunction causes glucose starvation in the liver and the muscle, accelerating metabolism of Ala to produce glucose in the liver, and metabolism of BCAAs in the muscle to produce Ala, Gln and Glu. On the other hand, Phe and Tyr were not metabolized in the muscle. As a result, plasma levels of BCAAs were decreased and those of Ala, Gln, Glu, Phe, and Tyr were increased. Thus, Fischer’s ratio is decreased in liver dysfunction [19,22]. Hypoxia results in accelerated metabolism of Glu and BCAAs in the muscle of COPD patients and induced decreased levels of plasma Glu and BCAAs [10]. Increased levels of Phe and Tyr and decreased levels of BCAAs in the plasma of sepsis patients occurred due to inflammation [23]. The reduction of plasma BCAA levels were reported in the rat model of acute ischemic stroke [24]. Thus, similar, but not same, amino acid profiles could be observed in different pathological states. Although the implications of our findings are not clear, this might suggest that a different modulation of amino acid metabolism, such as hypoxia-induced catabolic change or/and hypercatabolism due to inflammation, might be occurring in the pathogenesis of CVD-ILD and AoDILD in collagen disease.

For screening CVD-ILD in RA, negative predictive values for CVD-ILD markers should be higher, because the prognosis of CVDILD is extremely poor. From this point of view, KL-6 with a standard cut-off value of 500 is not suitable for screening for CVD-ILD in RA. The negative predictive value of svm index (AA, KL6) was higher than for the other markers. In the case of KL-6 with a cut-off level of 331.5, there was still 3.2% post-test probability of ILD in marker-negative RA patients. When using the svm index (AA, KL6), however, this was reduced to only 1.7% ILD in marker-negative patients. These cases would not to be evaluable by chest CT. The data therefore indicate that svm index (AA, KL6) could be a better marker to screen for CVD-ILD than KL-6 per se, despite significant differences in amino acid profiles being detected only for Lys.

Disease-modifying anti-rheumatic drug, immunosuppressant, or corticosteroid administration [25,26] might change the plasma amino acid profiles in patients. This possibility could be addressed by plasma amino acid profiling of collagen disease patients before starting treatment. Because of the limited sample size of this study, the evaluation of plasma amino acid profiles should be confirmed in future independent studies. Indices comprising amino-acid profiles were used for indicators of the presence of various diseases, including cancer [11-13]. Further large-scale studies would provide a possibility of generating better markers for AoDILD in collagen disease patients using indices comprising plasma amino-acid profiles.

In conclusion, this is the first report of plasma amino acid profiling of CDV-ILD. The svm index (AA, KL6) will be a better marker for diagnosing CVD-ILD in RA than KL-6 alone, though svm index (AA) is not. The plasma amino acid profiles could be a better marker for AoDILD in collagen disease patients.

We thank Ms. Hiromi Hayakawa (Sagamihara Hospital) for providing technical assistance for the study at Sagamihara Hospital; and Ms. Mayumi Yokoyama (Sagamihara Hospital) for secretarial assistance. The work was supported by Grants-in-Aid for Scientific Research from the Japan Society for the Promotion of Science, Health and Labour Science Research Grants from the Ministry of Health, Labour, and Welfare of Japan, Grants-in-Aid for Clinical Research from National Hospital Organization, Research Grants from Japan Research Foundation for Clinical Pharmacology, Research Grants from Takeda Science Foundation, Research Grants from Daiwa Securities Health Foundation, Research Grants from The Nakatomi Foundation.

HF has the following conflicts. The following funders are supported in whole or in part by the subsequent companies. The Japan Research Foundation for Clinical Pharmacology is run by Daiichi Sankyo, the Takeda Science Foundation is supported by an endowment from Takeda Pharmaceutical Company and the Nakatomi Foundation was established by Hisamitsu Pharmaceutical Co., Inc. The Daiwa Securities Health Foundation was established by Daiwa Securities Group Inc. ST was supported by research grants from pharmaceutical companies: Abbott Japan Co., Ltd., Astellas Pharma Inc., AstraZeneca K.K., Bristol-Myers Squibb Co Ltd., Chugai Pharmaceutical Co., Ltd., Eisai Co., Ltd., Medical & Biological Laboratories Co., Ltd, Mitsubishi Tanabe Pharma Corporation, Merck Sharp and Dohme Inc., Pfizer Japan Inc., Takeda Pharmaceutical Company Limited, and Teijin Pharma Limited. KT and TM are employees of Ajinomoto Pharmaceuticals and Ajinomoto Companies, respectively. The other authors declare no financial or commercial conflict of interest.