Oncostatin M receptor β and cysteine/histidine-rich 1 are biomarkers of the response to erythropoietin in hemodialysis patients

Abstract

Biomarkers that evaluate the response to erythropoietic-stimulating agents largely measure inflammation and iron availability. While these are important factors in modifying an individual’s response to these agents, they do not address all aspects of a poor response. To clarify this, we isolated peptides in the serum of good and poor responders to erythropoietin in order to identify biomarkers of stimulating agent response. Ninety-one candidate biomarker targets were identified and characterized using mass spectrometry, of which tandem mass spectroscopy provided partial amino-acid sequence information of 17 different peptides for 16 peptide masses whose abundance significantly differed between poor and good responders. The analysis concluded that three peptides associated with a poor response were derived from oncostatin M receptor β (OSMRβ). The 13 serum peptides associated with a good response were derived from fibrinogen α and β, coagulation factor XIII, complement C3, and cysteine/histidine rich 1(CYHR1). Poor response was most strongly associated with the OSMRβ fragment with the largest molecular weight, while a good response was most strongly associated with CYHR1. Immunoblots found the abundance of intact OSMRβ and CYHR1 significantly differed between good and poor responders. Thus, two measurable biomarkers of the response to erythropoietic-stimulating agents are present in the serum of treated patients.

The administration of an exogenous erythropoietic-stimulating agent (ESA) is required in many patient populations in order to maintain hemoglobin concentrations >10 g/dl. Central to the current discussion of ESAs is the identification of patients who respond normally to pharmacologic concentrations of erythropoietin (EPO) (good response) and those patients who do not (poor response). Patients who are poor responders to EPO may be at risk for increased morbidity and mortality,¹ and the inability to reach a higher target hemoglobin than previously recommended in combination with an increased epotin-α dose is associated with an increased risk of death, myocardial infarction, congestive heart failure, or stroke.²

In general, the causes of nonresponse to an ESA are well documented and appear to affect ~10% of the population.³ The three most common reasons for nonresponse are noncompliance, absolute or functional iron deficiency, and inflammation,⁴ with intercurrent infection probably the most common reason for temporary resistance. Other reasons for non or poor response include inadequate dose, secondary hyperparathyroidism, aluminum toxicity, hemolysis, malignancies, thalassemia, sickle cell disease, AIDS, pregnancy, vitamin deficiency,^3,5 and rarely antibodies to EPO.⁶ A recent review of the literature identifies cytokines⁷ (interleukin (IL)-1, IL-6, interferon-γ, tumor necrosis factor), hepcidin, EPO receptor (EpoR), and subsequent intracellular signaling as regulators of EPO responsiveness.⁸ Furthermore, there is an interaction between IL-6 and hepcidin that is responsible for hypoferremia that may limit ESA response.^9,10 Soluble EpoR concentrations are also associated with EPO resistance in end-stage renal disease.^11,12

We addressed the hypothesis that there are qualitative and quantitative differences in serum proteins in a group of idiopathic poor responders to EPO that are predictive of poor EPO response. We undertook an examination of a population of patients without obvious signs of inflammation or iron deficiency and other overt reasons for poor response in an attempt to identify biomarkers of EPO response. To address this hypothesis we performed a top-down liquid chromatography-matrix-assisted laser desorption-ionization time-of-flight tandem mass spectrometry (LC-MALDI-TOF MS/MS) peptidomic analyses to compare serum peptides from patients with either a good or a poor response to EPO.

RESULTS

The demographics of the enrolled subjects are shown in Table 1. There were significant differences in ESA dose and average EPO response index, defined as the EPO dose divided by the resulting hemoglobin 1 month later, between the two study groups. The distribution of gender between groups was different but did not reach statistical significance. The reasons for exclusion in the study population are shown in Table 2. One sample in the good responder group yielded insufficient bound peptide solution to proceed with LC-MALDI.

Table 1.

Demographics of the subject population at the time of sample collection

	Good responder	Poor responder	P-value
Gender (M/F)	10/5	7/13	0.06a
Hemoglobin (g/dl)	11.5±0.6	11.2±1.1	0.4
ESA dose (U/treatment)	1467±673	8392±5833	<0.001
Total iron dose (5 months) (mg)	953±1187	2035±1604	0.035
Average ERI	0.11±0.036	0.87±0.51	<0.001
Kt/V	1.53±0.25	1.58±0.27	0.5
Albumin (g/dl)	3.9±0.3	3.8±0.3	0.3
Ferritin	767±353	781±416	0.9
Tsat (%)	30.5±7.9	26.4±16.5	0.4

Abbreviations: ERI, erythropoietin response index; ESA, erythropoietic-stimulating agent; F, female; Kt/V, adequacy of dialysis; M, male; Tsat, transferrin saturation.

^aPearson’s χ².

Table 2.

Description of subjects who were reviewed for inclusion in the study and a list of the reasons for subjects who were excluded

	Study population
Active in-center patients	187 Subjects
Deceased	13
Transferred or new modality	21
Incomplete data	5
Subjects evaluated for inclusion	148 Subjects
<3 Months EPO out of 5 months	23
	Calculated ERI
	35 Lowest ERI (good responder)	35 Highest ERI (poor responder)
Hepatitis C	9	6
Active lupus	1	1
Active infection	2	0
Malignancy	0	1
Valve disease	0	1
Refused consent	7	6
Final subjects	15	20

Abbreviations: EPO, erythropoietin; ERI, erythropoietin response index.

Each subject’s serum was assayed for C-reactive protein, serum hepcidin, IL-6, IL-7, IL-8, IL-10, and tumor necrosis factor-α. These data are shown in Figures 1 and 2. The results of the statistical analyses are shown in Table 3. The only difference detected was an interaction between group and gender in hepcidin, IL-6, and IL-8 where there appeared to be increased serum levels of hepcidin, IL-6, and IL-8 in male poor responders.

Figure 1. Plot of serum C-reactive protein (CRP) and hepcidin from subjects displayed according to group and gender.

Mean values are displayed where the vertical line and middle horizontal line intersect. The upper and lower horizontal lines represent the s.d.

Figure 2. Plot of serum interleukin (IL)-6, IL-7, IL-8, IL-10, and tumor necrosis factor-α (TNF-α) from subjects displayed according to group and gender.

Mean values are displayed where the vertical line and middle horizontal line intersect. The upper and lower horizontal lines represent the s.d.

Table 3.

Results of the statistical analysis of the CRP, hepcidin, and cytokine data

	P-value
	Group	Gender	Interaction
CRP	0.11	0.98	0.83
Hepcidin	0.068	0.11	0.043
IL-6	0.71	0.18	0.018
IL-7	0.37	0.26	0.59
IL-8	0.57	0.58	0.046
IL-10	0.11	0.19	0.16
TNF-α	0.51	0.97	0.76

Abbreviations: CRP, C-reactive protein; IL, interleukin; TNF-α, tumor necrosis factor-α.

Analysis of MS data

In general, most peptides were of low abundance and infrequently observed across all samples. A total of 939 freely soluble serum peptide masses were observed in ≥28 (82% of samples) serum samples. A total of 130 masses were observed in all 34 samples. Using t-test, 40 peptides were observed to be differentially abundant with significance at the P≤0.05 level and 3 peptides at the P≤0.001 level (Supplementary Table S1 online). A total of 558 protein-bound serum peptide masses were observed in ≥28 serum samples. A total of 90 masses were observed in all 34 samples. Using t-test, 51 peptides were observed to be differentially abundant with significance at the P≤0.05 level and 9 peptides at the P≤0.001 level (Supplementary Table S1 online). To graphically illustrate these data, plots of significance (P-value) versus serum abundance differences in good responder to poor responder (serum abundance ratio) for freely soluble (Figure 3a) and protein-bound serum peptides were constructed (Figure 3b) and annotated to indicate assignment of protein identities to respective data points. Mann and Kelleher have suggested fold expression changes of 1.3 to ≥2.0 to be meaningful for MS-based proteomics experiments.¹³ Our criteria for selecting peptides for tandem MS analysis to gain amino-acid sequence information (peptides with P-values ≤0.05) and differential abundance (abundance changes of ≥133 or ≤75% in good and poor responders) identified 38 freely soluble serum peptides (95% of all observed free peptides with P-value <0.05) and 50 protein-bound serum peptides (98% of all observed bound peptides with a P-value <0.05) and demonstrated fold-abundance changes of ≥1.3. For purposes of emphasis, peptides estimated to have t-test P-values ≤0.001 are highlighted in Figure 3 within the inset boxes.

Figure 3. Graphic comparison of the change in peptide abundance in good- to poor-responder serum samples with the corresponding statistical significance P-value for each peptide.

The t-test and ratio data are plotted using log–log graphing for each significant peptide from the freely soluble serum peptide (a) and protein-bound serum peptide (b) fractions. The shaded region denotes masses with a P-value >0.001. When possible, the parent protein for each protein fragment is annotated with the appropriate protein accession number. In some cases, multiple peptides originated from the sample protein and result in the multiple annotations of the same gene accession number. CO3, complement C3; CYHR1, cysteine and histidine-rich protein 1; FGA, fibrinogen α chain; FGB, fibrinogen/fibrinopeptide B; F13A coagulation factor XIII chain A.

Identification of peptide amino-acid sequences using MS/MS methods

Amino-acid sequences were tentatively assigned to a total of 16 peptides (Table 4) corresponding to six parent proteins. These peptide fragments were derived from a fibronectin III domain of oncostatin M receptor β chain (OSMRβ) (Figure 4a–c), fibrinogen α chain (FGA), fibrinogen/fibrinopeptide B (FGB), a fragment of the signal peptide region of the cysteine and histidine-rich protein 1 (CYHR1) (Figure 5a), coagulation factor XIII chain A (F13A), and complement C3 (CO3). One peptide (1534.689 m/z) was assigned slightly different but overlapping amino-acid sequences by the Mascot and the Paragon algorithms. Seven peptide masses required post-translational modifications to explain MS/MS fragmentation spectra. Four peptides were N-terminally modified, including two FGA modifications that were assigned a mass value only. One FGA modification was assigned as a post-translationally hydroxylated phenylalanine, given that the F27Y polymorphism has not been reported. One FGB peptide was assigned with an N-terminal pyro-glutamate residue resulting from an N-terminal glutamine rearrangement. Two OSMR peptides were assigned (Supplementary Methods) mass values consistent with modifications including di-sodiation (+45.99 m/z) and O-glucosamine modification.

Table 4.

Assignment of peptide sequence to differentially expressed serum peptides

Observed mass (m/z)	Parent protein name	Amino acid Amino-acid sequencea	Post-translational modification	Mascot MOWSE	Paragon unused scoreb
Freely soluble	Increased serum abundance with good response
1194.521	Fibrinogen α chain	D.SGEGDFLAEGGGV.R		46	14
1399.623	Fibrinogen α chain	S.GEGDFLAEGGGVR.G	(N-term +136.16)	116c
1463.656	Fibrinogen α chain	A.DSGEGDFLAEGGGVR.G			14.0
1481.653	Fibrinogen α chain	A.DSGEGDFLAEGGGVR.G	Phe->Tyr@7	132	14
1534.689	Fibrinogen α chain	D.ADSGEGDFLAEGGGVR.G			14.0
1534.689	Fibrinogen α chain	D.SGEGDFLAEGGGVR.G	(N-term +183.98)	93c
2466.054	Fibrinogen α chain	S.SSYSKQFTSSTSYNRGDSTFES.K		55	14
2553.101	Fibrinogen α chain	K.SSSYSKQFTSSTSYNRGDSTFES.K		75	14
2768.252	Fibrinogen α chain	K.SSSYSKQFTSSTSYNRGDSTFESKS.Y		115	14
2931.292	Fibrinogen α chain	K.SSSYSKQFTSSTSYNRGDSTFESKSY.K		63	14
1552.673	Fibrinogen β chain	S.QGVNDNEEGFFSAR.G	Gln-> pyro-Glu@N-term		2.7
Serum protein	Increased serum abundance with good response
bound
1210.582	Coagulation factor XIII A chain	M.SETSRTAFGGR.R	Acetyl@N-term		2.0
1488.800	Cysteine and histidine-rich protein 1	L.SHLVLGVVSLHAAVS.T			1.3d
1504.82	Complement C3	G.SPMYSIITPNILR.L		46	2
Freely soluble	Increased serum abundance with poor response
1273.633	Oncostatin-M-specific receptor subunit β	E.NKEVEEERIAG.T		50	2.1
1549.774	Oncostatin-M-specific receptor subunit β	E.NKEVEEERIAGTE.G	(C-term + 45.99)	48c
1664.801	Oncostatin-M-specific receptor subunit β	E.NKEVEEERIAGTE.G	(T(12) +161.02)	49c

^aAmino-acid sequence is presented using Paris Convention guidelines for presenting proteomics data. The proteolytic excision sites are offset with periods. Amino acids within periods comprise the amino-acid sequence for the observed peptide.

^bProtein Pilot Paragon confidence interval scoring for the protein and the peptides listed were 99% (unused score ≥2.0) with simultaneous adjustment for decoy data base analysis and removal of false-positive identifications.

^cMatrix Science Mascot MOWSE score following post hoc error-tolerant analysis.

^dProtein Pilot Paragon confidence interval scoring was 95% (unused score=1.3) with simultaneous adjustment for decoy data base analysis and removal of false-positive identifications.

Figure 4. Distribution of serum oncostatin M receptor β (OSMRβ) fragments and protein levels associated with erythropoietic-stimulating agent (ESA) responsiveness.

Peptide abundance data were extracted from aligned mass spectrometry (MS) data sets and peptide spectral abundance was calculated from the MS ion cluster area.¹⁰ For purposes of illustration, peptides not observed in particular samples were assigned non-zero background abundance values arbitrarily equal to 10. Vertical scatter plots (mean±s.e.m.) for the differences in serum abundance for (a–c) three OSMRβ fragments and for (d, e) densitometry measurements of circulating OSMRβ illustrate significant differences in abundance between good-response (GR) and poor-response (PR) groups.

Figure 5. Distribution of serum cysteine/histidine rich 1 (CYHR1) fragment and protein levels associated with erythropoietic-stimulating agent (ESA) responsiveness.

Peptide abundance data were extracted from aligned mass spectrometry (MS) data sets and peptide spectral abundance was calculated from the MS ion cluster area. Vertical scatter plots (mean±s.e.m.) for the differences in serum abundance for (a) one CYHR1 fragment and for (b, c) densitometry measurements of circulating oncostatin M receptor β (OSMRβ) illustrate significant differences in abundance between good-response (GR) and poor-response (PR) groups. (d) The specificity of the CYHR1 antibody was confirmed using immunogen competition experiments where primary antibody was preincubated with tenfold excess of synthetic immunogen before applying to a freshly blotted and blocked membrane.

Differential expression of intact OSMRβ and CYHR1 in patient serum

An OSMR-positive band was identified in denaturing conditions migrating at 50 kDa (Figure 4e) and was significantly (P<0.05) increased (75% above poor responders) in the serum of good responders. The analysis of these samples using nonreducing and denaturing sodium dodecyl sulfate-polyacrylamide gel electrophoresis identified OSMR bands migrating at 100–110 kDa (data not shown) with a like expression trend. Similarly, CHYR1 was identified (Figure 5c) and validated using immunogen peptide blocking experiments (Figure 5d) migrating at ~72 kDa and was significantly (Figure 5b; P<0.05) increased in the serum of EPO poor responders (30% above good responders).

Summary statistical analysis

Summary statistics for all identified peptides are shown in Table 5. In the case of OSMRβ and FGA, more than one fragment of a parent protein was identified and a Bonferroni adjusted P-value is shown and an analysis of the sum of the fragment abundances was performed. In addition, the receiver operating characteristic value associated with the sensitivity and specificity of the peptide to predict either good or poor response was calculated.

Table 5.

Statistical analysis of the identified peptides

Peptide	m/z	Group	Gender	Interaction	ROC
			P-valuesa
	Increased abundance predicts poor response
OSMR	1273	<0.0001 (< 0.0001)	0.040 (1.00)	0.85 (1.00)	0.95
	1549	<0.0001 (< 0.0001)	0.026 (0.08)	0.21 (0.63)	0.96
	1664	<0.0001 (< 0.0001)	0.19 (0.56)	0.62 (1.00)	0.99
Sum of fragments		< 0.0001	0.20	0.65	0.98
	Increased abundance predicts good response
CYHR1	1488	<0.0001	0.60	0.60	0.91
FGA	1194	0.052 (0.47)	0.49 (1.00)	0.29 (1.00)	0.79
	1399	0.081 (0.73)	0.67 (1.00)	0.92 (1.00)	0.62
	1463	0.021 (0.19)	0.58 (1.00)	0.33 (1.00)	0.69
	1481	0.045 (0.041)	0.46 (1.00)	0.80 (1.00)	0.72
	1534	0.063 (0.57)	0.78 (1.00)	0.85 (1.00)	0.62
	2466	0.03 (0.27)	0.067 (0.60)	0.85 (1.00)	0.82
	2553	0.019 (0.17)	0.27 (1.00)	0.59 (1.00)	0.81
	2768	0.053 (0.48)	0.16 (1.00)	0.58 (1.00)	0.77
	2931	0.004 (0.036)	0.17 (1.00)	0.33 (1.00)	0.81
Sum of fragments		0.005	0.17	0.59	0.88
FGB	1552	0.0010	0.89	0.72	0.80
Factor XIII	1210	0.015	0.51	0.71	0.75
CO3	1504	0.07	0.64	0.20	0.74

Abbreviations: CO3, complement C3; CYHR1, cysteine and histidine-rich protein 1; FGA, fibrinogen α chain; FGB, fibrinogen/fibrinopeptide B; OSMR, oncostatin M receptor; ROC, receiver operating characteristic.

^aValues in parentheses are P-values following Bonferroni correction for multiple comparisons.

DISCUSSION

Our goal was to identify serum peptides associated with a poor response to EPO. We studied patients who routinely attended their dialysis session, received adequate EPO and iron dosing, and did not appear to have risk factors for EPO failure, such as chronic inflammation or infection. Using a peptidomic approach to generate MS data and develop a list of differentially abundant peptides and ranked on P-values, we assigned amino-acid sequences to 16 peptides whose serum abundance significantly differed between poor and good responders for further analysis. Three of the serum peptides associated with poor EPO response through sequence alignment of the peptides to the parent protein were found to be derived from the fibronectin III domain of the OSMRβ chain. The 13 serum peptides associated with good EPO response were concluded to be derived from FGA, FGB, FXIIIA, CO3, and CYHR1.

We performed an immunoblot analysis of the serum for the presence of OSMRβ (reducing and nonreducing conditions) and CYHR1 (reducing conditions) in a subset of the total population. The results of the OSMRβ immunoblot experiments suggest that receptor is present in the serum as a dimer. The molecular weight of the receptor observed in reducing conditions is consistent with molecular weights for shed OSMR, leukemia inhibitory factor, and IL-6R ecto-domains.^14,15 We were able to demonstrate the presence of both intact proteins in serum, with OSMR increased in the serum of good responders and CYHR1 increased in the serum of poor responders.

OSMR intact protein is high in the serum of good responders, whereas peptide fragments are high in poor responders. For CYHR1, intact protein is high in the serum of poor responders, whereas peptide fragments are high in the serum of good responders. We speculate that the difference is because of altered catabolism of OSMR and CYHR1 over the range of EPO response index or increased receptor turnover in the case of OSMR. Animal data appear to support this speculation for OSMR, where OSMR knockout mice have low hematocrit and decreased red blood cells.¹⁶

These findings were not confounded with other measured markers of EPO response. We did not observe differences between poor and good responders for C-reactive protein, hepcidin, IL-6, IL-7, IL-8, IL-10, tumor necrosis factor-α, transferrin saturation, ferritin, and albumin. There was a significant interaction between responder type and gender for hepcidin, IL-6, and IL-8. We feel that the gender interaction is most likely related to several male poor responders who showed signs of inflammation in their measured laboratory values.

We have two novel peptidomic findings in this study. Three fragments of OSMRβ are strongly associated with poor EPO response and one fragment of CYHR1 is strongly associated with good EPO response. Oncostatin-M (OSM) is proposed to be an important EpoR-phosphoY343-Stat5-induced gene product that participates in erythroblast survival.¹⁷ OSM is secreted from cytokine-activated T cells and monocytes and is involved in inflammation.^18,19 OSM binds to two different OSM receptors in humans: the type 1 receptor is identical to leukemia inhibitory factor receptor that consists of gp130, also found in the IL-6 receptor, and the type 2 receptor that consists of gp130 and OSM-specific receptor β subunit (OSMR);^20,21 the OSMR fragments identified in the serum of our patients are from the type 2 receptor. Animal studies indicate that OSMR has an important role in erythropoiesis, and OSMR knockout mice have decreased number of circulating red blood cells and a decreased hematocrit compared with wild type.¹³ OSMR knockout mice also have decreased numbers of erythroid colony-forming units and erythrocyte-producing colonies in the bone marrow. Work in human fibroblast or epithelial cells show that OSM ligand binding to OSMR induces receptor degradation and then increases the level of receptor synthesis.²² In hepatocytes and hepatoma cells, OSM induces hypoxia-inducible factor 1α gene transcription via a Janus kinase/signal transducer.²³ Considered in the context of the literature that indicates that OSM helps regulate erythropoiesis, the fact that we observed increased abundance of OSMR fragments in patients with poor EPO response might be expected.

The finding of a fragment of CYHR1 in the serum of our study subjects was unexpected. Analysis of the estimated amino-acid sequence suggests the fragment is derived from the signal peptide sequence of CYHR1. This CYHR1 fragment is almost as good a predictor of good response (receiver operating characteristic=0.91) as OSMR is a predictor of poor response (receiver operating characteristic=0.98). The current state of knowledge on CYHR1 is limited regarding its predicted protein structure, protein–protein interactions, subcellular localization, and chromosome mapping. CYHR1 is proposed to contain four functional transmembrane helices and was first identified using a yeast two-hybrid system to search for cytoplasmic proteins that associate with galectin-3.²⁴ Further work was performed using recombinant hamster galectin-3 and murine CYHR1 and demonstrated that CYHR1 binds to the carbohydrate-recognition domain of galectin-3.²⁵

FGA, FGB, FXIIIA, and CO3 had increased abundances in good responders and are likely related to low levels of inflammation that are present in hemodialysis patients.²⁶ The prediction of EPO response has been related to the baseline fibrinogen, baseline transferrin receptor concentration, and the change in the transferrin receptor concentration after 2 weeks for EPO therapy.⁵ The observed increase in abundance of fragments of both fibrinogen and factor XIII may be related through thrombin and a result of inflammation.²⁷ In a study of 100 hemodialysis patients, the authors concluded that subclinical inflammation was an important determinant of anemia.²⁸ Furthermore, the authors of this manuscript were able to look at 51 subjects who had not received EPO and found that there was a negative relationship between hemoglobin and fibrinogen in ESA-treated subjects but not in non-ESA-treated subjects.

The current study looked at patients who were receiving EPO for a period of at least 6 months. As such, the results may be related to the dose of EPO those patients received and may not represent forward predicting markers. We will need to study additional patients entering dialysis not receiving any ESA to validate the utility of these markers. However, in an analysis not shown, we determined that OSMR better predicted membership in the two groups than dose, even though the two groups were defined as a function of dose.

In summary, we performed a peptidomic analysis of the serum of subjects without overt signs of inflammation or extraordinary blood loss who were good and poor responders to exogenous EPO. The analysis resulted in the identification of 16 peptide fragments that were differentially expressed in the two groups, with OSMRβ and CYHR1 showing the best association. OSMRβ has a demonstrated role in erythropoiesis following the production and binding of oncostatin M, whereas CYHR1 has no known biologic action. The other identified fragments, fibrinogen α and β, factor XIII, and compliment C3, were not as strongly associated with ESA response and may reflect an underlying inflammatory process. OSMRβ and CYHR1 have been identified as candidate biomarkers of the response to exogenous EPO. These proteins and their fragments will need to be validated in larger studies as biomarkers of the response to ESA therapy.

MATERIALS AND METHODS

Human subjects

The research protocol conformed to the Declaration of Helsinki and informed consent was obtained from each subject before participation in the study. The study was approved by the institutional review boards of both the University of Louisville and the Louisville Veterans Administration Medical Center. EPO dose and hemoglobin data were collected over a 5-month period of treatment for 187 hemodialysis patients in the Kidney Disease Program hemodialysis. We calculated an EPO response index as the EPO dose (per 1000 units) divided by the resulting hemoglobin 1 month later. These data were averaged over the 5 months of collected data. The mean data were sorted in ascending order and subjects for the good response group were selected from the lowest quintile group. Subjects for the poor response group were selected from the highest quintile group. EPO resistance was defined as those subjects with the largest EPO response index values averaged over 5 months.

Subjects were excluded if they changed dialysis modality, had incomplete data, had not received EPO in >3 of the 5 months evaluated, had hepatitis C, active lupus or infection, malignancy, valve disease, and for lack of consent.

Samples and sample handling

Study subjects were asked to donate 20 ml of blood for proteomic analysis following an informed consent process. Blood was collected in 2–10 ml red top vacutainer tubes before the dialysis session, processed immediately for serum separation, and stored in 0.5 ml aliquots at −80 °C until analyzed.

Peptides were isolated from serum using a Vivaspin 2 (Sartorius AG, Gottingen, Germany) spin filtration devices housing 5000 Dalton nominal molecular weight cutoff Hydrosart (cellulosic) membranes. Two peptide fractions, freely soluble and protein bound, were isolated and assayed using a microBCA method (Pierce, Rockford, IL). One good responder sample provided insufficient amounts of protein-bound peptide for study by LC-MALDI-TOF MS; both the freely soluble and the protein-bound fractions for that patient were excluded from LC-MALDI-TOF MS studies.

Peptidomic analyses

Isolated peptides were analyzed using a method we previously published.²⁹ Briefly, LC-MALDI-TOF MS ion chromatograms were constructed using Data Explorer software (Applied Biosystems, Foster City, CA) using signal-to-noise criteria of ≥10 to define a peptide peak, exported as peak list text files, and used for determination of differential peptide abundance. Only peptides found in >80% (n = 28) samples were included for further analysis. The integrated signal areas for the MALDI-TOF MS data for these peptides were compared using unpaired Student’s t-test. Peptides having P-values >0.05 (P>0.05) and fold abundance changes of ≥133 or ≤75% in good and poor responders were considered important and used to create a list of peptides for targeted tandem MS studies.

Computer-assisted tandem MS data analysis

We investigated selected peptides using a tandem MS method we have previously published.²³ The peptide fragmentation information was searched using Matrix Science Mascot software (version 2.1) (Matrix Science, Boston, MA) and the Paragon algorithm of Protein Pilot (Applied Biosystems). Analyses were conducted simultaneously against a reversed SwissProt database (uniprot.org) and automated filtering, retaining peptide scoring with q-values (false discovery rate filter) ≥0.10. We sought to identify the protein source for the discriminating serum peptides using tandem mass spectrometry. Unidentified high-quality spectra at this point were suspected to be post-translationally modified. To address this problem, the data were reanalyzed using two bioinformatic approaches. We first used a post hoc error-tolerant analysis with Mascot software and second, we employed the Paragon algorithm of Protein Pilot. Both methods attempt to analyze the data in an iterative approach, considering known post-translational modifications, polymorphisms, and point mutations.

Immunoblot analyses for serum protein abundance

We performed immunoblot analyses using a method we have previously described.³⁰ We determined the serum abundance of intact OSMRβ and CYHR1 using native (denaturing) and Laemmli (reducing/denaturing) sample buffers. These analyses were conducted using polyclonal antibodies raised to either full-length human OSMRβ (cat. no. ab67805; Abcam, Cambridge, MA) or an internal epitope of human CYHR1 (sc-87664; Santa Cruz Biotechnology, Santa Cruz, CA). The expression of OSMRβ and CYHR1 were examined using previously unthawed, contemporaneous aliquots of the serum sample set used for peptidomic analyses.

Analysis of serum markers of inflammation and iron status

High-sensitivity human serum cytokine measurements were made under contract services by Millipore (Billerica, MA) labs. High-sensitivity C-reactive protein measurements in patient serum samples were made using the Immulite 1000 (Siemens Healthcare Diagnostics, Deerfield, IL) High-Sensitivity C-reactive protein kit (Diagnostics Products, Los Angeles, CA) according to the manufacturer’s guidelines. Hepcidin-25 peptide measurements were made using the hepcidin-25 peptide enzyme immunoassay kit S-1337 (Bachem Group, Torrance, CA) and using the hepcidin-25 standard LEAP1 from Peptides International (Louisville, KY) as a positive control. The R² was 0.9967 for LEAP1-positive control peptide from 0 to 50 ng/ml using sigmoid regression, whereas it was 0.9969 for the hepcidin-25 standard provided in Bachem enzyme immunoassay kit. The coefficient of variation for a given hepcidin-25 concentration of 1.56 ng/ml was 3.49% intra-assay and 3.43% inter-assay.

Statistical analysis

Statistical analysis was performed using PASW Statistics 18 (SPSS, Chicago, IL). Comparisons of proportions, means, and means by gender were done using Pearson’s χ², t-test, and analysis of variance, respectively. When multiple fragments of the same parent protein were analyzed, a Bonferroni correction was applied to address the problem of multiple comparisons and the data were analyzed as the sum of all fragment abundances. The ability of the identified biomarkers to discriminate between groups was analyzed using receiver operating characteristic curve.