Professional Documents
Culture Documents
Background: The complex pathology of disease has TNF-␣, IL-8, and IL-6 (Linco, Biosource, Upstate, and
sparked the development of novel protein expression R&D) with correlation coefficients of ⴚ0.107 to 0.318.
profiling techniques that require validation in clinical Within- and between-run imprecision values for the
settings. This study focuses on multiplexed analyses of multiplex method were generally <15%. Relative
adipocytokines and biomarkers linked to the metabolic changes in plasma leptin and insulin concentrations
syndrome, diabetes, and cardiovascular disease. after diet-induced weight loss were similar whether
Methods: Multiplexed immunoassays using fluorescent assessed by multiplex assay or ELISA.
microspheres and the Luminex-100 system were per- Conclusion: Although this technology appears useful in
formed on plasma from 80 obese patients (40 with the clinical research studies, low assay sensitivity and poor
metabolic syndrome) before and after 6 – 8 weeks of correlations with conventional ELISA methods for some
analytes with very low plasma concentrations should be
diet-induced weight loss. Leptin, insulin, C-peptide,
considered when using the Luminex platform in clinical
monocyte chemoattractant protein-1 (MCP-1), eotaxin,
studies.
interleukin-8 (IL-8), tumor necrosis factor-␣ (TNF-␣), © 2005 American Association for Clinical Chemistry
and IL-6 concentrations measured with multiplex panels
from 3 different manufacturers were compared with Obesity has reached epidemic proportions in the United
results from commercial ELISAs. Detection limits and States; ⬃30% of US adults are obese, and the percentage is
between- and within-run imprecision were determined increasing (1 ). Obesity increases the risk for the metabolic
for each analyte. Bland–Altman analysis was used to syndrome, diabetes, hypertension, atherosclerosis, and
determine agreement between multiplexed immunoas- thrombosis (2– 4 ). The National Cholesterol Education
says and ELISAs. Program Adult Treatment Panel III (ATP III)5 defined the
Results: Correlation between the Luminex multiplexed metabolic syndrome using clinical criteria for abdominal
assays and ELISAs was good for leptin (Linco), insulin obesity, triglyceride and HDL-cholesterol concentrations,
(Linco), MCP-1 (Biosource and Upstate), and eotaxin blood pressure, and fasting glucose (5 ), and ⬎20% of
(Biosource) with correlation coefficients of 0.711– 0.895; adults in the United States meet these clinical criteria (6 ).
fair for eotaxin (Upstate) and C-peptide (Linco) with Novel technologies such as gene expression arrays have
correlation coefficients of 0.496 – 0.582; and poor for identified a large number of molecules that have altered
expression in adipose tissue and may be relevant to the
development of the metabolic syndrome, diabetes, and
cardiovascular disease with obesity. In contrast to the
ability to simultaneously examine the concentrations of
1
Section of Atherosclerosis, Department of Medicine, 2 Division of Endo- thousands of mRNA transcripts, plasma concentrations of
crinology, Diabetes and Metabolism, and 3 Section of Nutrition, Department of
Pediatrics, Baylor College of Medicine, Houston, TX.
4
Methodist Wellness Services, The Methodist Hospital, Houston, TX.
*Address correspondence to this author at: Baylor College of Medicine,
5
6565 Fannin, M.S. A-601, Houston, TX 77030. Fax 713-798-3057; e-mail Nonstandard abbreviations: ATP III, Adult Treatment Panel III; MCP-1,
cmb@bcm.tmc.edu. monocyte chemoattractant protein-1; TNF-␣, tumor necrosis factor-␣; IL,
Received December 17, 2004; accepted April 6, 2005. interleukin; and NIBSC, National Institute for Biological Standards and Con-
Previously published online at DOI: 10.1373/clinchem.2004.047084 trol.
1102
Clinical Chemistry 51, No. 7, 2005 1103
cytokines, chemokines, or other biomarkers are usually were not started on any medication that would influence
measured one at a time by ELISA. Measurement of glucose tolerance, insulin secretion, or insulin sensitivity.
plasma concentrations of each putative biomarker with
individual ELISAs incurs considerable time, cost, and collection and storage of blood samples
sample volume, which thus limits the ability to systemat- Blood samples were collected by venipuncture from the
ically examine the effects of a clinical intervention. participants into EDTA-containing Vacutainer Tubes after
Multiplexed proteomics research may offer a major an overnight fast. The freshly drawn blood was centri-
advance for clinical research. Protein microarrays are the fuged at 3000g for 20 min at 4 °C. Plasma was then
most commonly used for simultaneous determination of separated and subsequently stored at ⫺70 °C in small
multiple proteins in a biological fluid. The technique uses aliquots. All analyses were performed on frozen plasma
primary antibodies as the immobilized probe on a solid samples.
surface, and protein antigens labeled with fluorophores
calibrator replicates, and the respective analyte concentra- panel was not yet calibrated against the WHO standard
tions were calculated from the calibration curves. The and no WHO standard for eotaxin is currently available, it
recombinant protein MCP-1 and IL-6 calibrators provided was not possible to compare the detection limits of the
with the ELISA and Luminex reagents were added into Biosource panel with ELISAs. Unlike the Biosource panel,
normal healthy human plasma. The plasma was manu- the Upstate multiplex panel has a significantly higher
facturer pooled with specific concentrations of lipids detection limit for MCP-1 than the ELISA from R&D.
requested (lot no. BC0208A; Pacific Biometrics, Inc.). Cytokines TNF-␣, IL-8, and IL-6 in 3 different multi-
plex panels from different manufacturers (Linco, Upstate,
statistical analysis and R&D) were compared with ELISAs. In general, the
Statistical analyses were performed with SPSS data anal- ELISAs tended to have lower detection limits.
ysis software (Ver. 11.5; SPSS Inc.). Analyze-it software
(Ver. 1.71; Analyze-it Software Ltd.) was used for Deming percentage of samples with undetectable
Table 1. Comparison of detection limits for the Luminex multiplexed assays and the individual ELISAs.
Multiplex panels Analytes Luminex ELISA ELISA manufacturer
Linco Endocrine Leptin 0.13 g/L 1.39 g/L Linco
Insulin 0.31 mIU/L 0.14 mIU/L Alpco
C-peptide 0.037 g/L 0.132 g/L Linco
Biosource Chemokine MCP-1 2.35 ng/L 2.57 ng/L R&D
Eotaxin 4.37 ng/L 3.80 ng/L R&D
Upstate Chemokine MCP-1 20.5 ng/L (0.023 kIU/L) 2.57 ng/L (0.002 kIU/L) R&D
Eotaxin 27.72 ng/L 3.80 ng/L R&D
Linco Cytokine TNF-␣ 1.61 ng/L (0.249 kIU/L) 0.092 ng/L (0.0028 kIU/L) R&D
IL-8 2.52 ng/L (0.026 kIU/L) 0.26 ng/L (0.0018 kIU/L) Biosource
Upstate Cytokine TNF-␣ 0.065 ng/L (0.0019 kIU/L) 0.092 ng/L (0.0028 kIU/L) R&D
IL-8 0.70 ng/L (0.0003 kIU/L) 0.26 ng/L (0.0018 kIU/L) Biosource
IL-6 0.75 ng/L (0.1029 kIU/L) 0.068 ng/L (0.0084 kIU/L) R&D
R&D Cytokine TNF-␣ 1.53 ng/L (0.052 kIU/L) 0.092 ng/L (0.0028 kIU/L) R&D
IL-6 0.977 ng/L (0.1280 kIU/L) 0.068 ng/L (0.00843 kIU/L) R&D
Clinical Chemistry 51, No. 7, 2005 1105
Table 2. Samples with undetectable concentrations of MCP-1, TNF-␣, IL-6, and IL-8 in the ELISAs and the Luminex
multiplexed assays.
IL-6 TNF-␣ MCP-1 IL-8
assays with each sample analyzed in duplicate, was validation of the luminex method in human
⬎15%. The within-run precision represents the mean plasma
values of 3 samples with low, medium, and high concen- We chose ELISAs for comparison with the newer Lumi-
trations of the analyte that were analyzed 20 times in 1 nex technique because the ELISA is a generally accepted
assay. The within-run CVs for all analytes ranged from method for analyzing biomarkers in human serum or
6.6% to 12% and were within the acceptable limit. plasma. Two methods of comparison between Luminex
The between- and within-run precision of the ELISAs assays and ELISAs were carried out. We first used linear
are shown in Tables 1 and 3 of the online Data Supple- correlation to evaluate the relationship between Luminex
ment. The between-run CV was determined from 2 assays and ELISAs. We then applied the Bland–Altman
plasma control samples (QA and QC) because they were method (13 ) to assess the agreement between the 2
detectable by the high-sensitivity ELISAs. For each sam- measurement methods.
ple, the CV was determined from the means of 6, 8, and 5
assays for MCP-1, TNF-␣, and IL-6, respectively, and were Correlation. The results of the Deming regression analysis
analyzed in duplicate. The within-run precision was cal- comparing the Luminex and ELISA methods are shown in
culated from the mean CVs of n samples analyzed in Table 4. The correlation coefficient indicates scatter of the
duplicate. Overall, the within-run CVs were within rea- measured data about the Deming regression line, and it is
sonable limits, ranging from 5.0% to 15%, except for IL-6 determined by the precision of the 2 assay methods. Table
(21%). For the high-sensitivity ELISAs, the between-run 4 shows the results for the Linco endocrine panel. The
imprecision generally indicated a ⬎15% CV of random correlation coefficients (R) were 0.812, 0.895, and 0.582 for
errors, and the within-run imprecision showed a ⬎15% leptin, insulin, and C-peptide, respectively. The correla-
CV of random errors. tion coefficient for C-peptide was 0.67 after omission of
the 2 outliers (defined as values outside 3 SD). In the
Deming regression analysis, the slope indicates a propor-
tional error that might have originated from incomplete
Table 3. Within-run assay imprecision. recovery in one of the methods or from different accura-
Analytes Within-run CV, % n
cies of the calibrators used. The slopes were 1.28, 2.38, and
Multiplexed panels
0.68 for leptin, insulin, and C-peptide, respectively. The
Linco Endocrine Leptin 11 120 smallest proportional error was for leptin, i.e., the values
Insulin 6.6 120 obtained from the 2 assay methods were the least differ-
C-Peptide 10 80 ent. The intercept indicates a systematic error that might
Biosource Chemokine MCP-1 12 40 be caused by interference, such as insufficient blank
Eotaxin 10 40 compensation. The intercepts were 2.26 g/L, ⫺3.03
Upstate Chemokine MCP-1 8.1 40 kIU/L, and ⫺1.83 g/L for leptin, insulin, and C-peptide,
Eotaxin 11 40 respectively. The smallest systematic error was for C-
ELISAs peptide.
Linco Leptin 5.9 120 The correlation coefficients were 0.263 and 0.899 for
Alpco Insulin 15 40 MCP-1 in the Biosource panel with and without the 2
Linco C-Peptide 10 40 outliers, respectively; outliers were defined as values
R&D MCP-1 5.4 120
outside 3 SD. For eotaxin, the correlation coefficient was
R&D Eotaxin 5.0 120
0.711 for the Biosource panel. The slopes for MCP-1
1106 Liu et al.: Luminex Multiplexed Analysis of Human Plasma
Table 4. Deming regression results for comparison of the multiplexed assays performed on the Luminex systems (y) and
individual ELISAs (x).
Slope Intercept
a b
Analyte Deming Sy円x R Coefficient SE 95% CI Mean SE 95% CI
Leptin (g/L) 9.90 0.812 1.28 1.28 1.11–1.44 2.26 3.19 ⫺4.05 to 8.57
Insulin (mIU/L) 5.62 0.895 2.38 0.19 1.99–2.77 ⫺3.03 3.47 ⫺10.06 to 4.00
C-Peptide (g/L) 2.32 0.582 0.68 0.15 0.37–0.99 ⫺1.83 1.20 ⫺4.25 to 0.59
Eotaxin (ng/L) 20.44 0.711 7.71 1.29 5.10–10.32 ⫺49.85 80.64 ⫺213.55 to 113.86
MCP-1c (ng/L)
With outliers 110.14 0.263 40.44 25.09 ⫺10.49 to 91.36 ⫺7829.64 6421.28 ⫺20 865.54 to 5206.25
Without outliers 49.99 0.899 4.97 0.42 4.11–5.82 79.60 109.36 ⫺142.90 to 302.09
without the outliers and for eotaxin were 4.97 and 7.71 for gave higher values than the ELISA, although only 2.5% of
the Biosource panel, respectively, and the intercepts were the values differed by ⬎2 SD. For C-peptide, most of the
79.60 and ⫺49.85 ng/L, respectively. Passing–Bablok Luminex measurements gave lower values than the
analysis revealed both proportional and constant bias for ELISA (mean difference, 3.9 g/L). At concentrations ⬍4
MCP-1 and eotaxin (data not shown). g/L, the 2 methods agreed very well, and only 2.5% of
We also analyzed the correlations between the Lumi- the values differed by ⬎2 SD. For MCP-1 in the Biosource
nex assays and ELISAs for the cytokines TNF-␣, IL-8, and panel, the 2 methods agreed well at concentrations ⬍0.5
IL-6. The correlation coefficients were 0.104 and ⫺0.107 kIU/L, and 5.5% of patients had values that differed by
for TNF-␣ and 0.266 and 0.318 for IL-8 in the Linco and ⬎2 SD. For insulin, C-peptide, and MCP-1, the only
Upstate panels, respectively. The correlation coefficient measurements that differed by ⬎2 SD were the extreme
was 0.298 for IL-6 in the Upstate panel. Overall, the correla- highest values of the analytes for patients in this study.
tions between Luminex assays and ELISAs for the 3 cyto-
kines analyzed were poor in all the multiplex panels. recovery study
Recovery experiments detect inaccuracy of a method
Bland–Altman analysis. The difference between the Lumi- resulting from systematic errors. The results of the recov-
nex assay and the ELISA was plotted against the mean of ery study for MCP-1 in both the ELISA and Luminex
the Luminex assay and ELISA for each plasma sample assay are shown in Table 2 of the online Data Supplement.
measured. The Bland–Altman plots for several analytes The recombinant protein calibrators provided with the
are shown in Fig. 1. The Bland–Altman analysis was ELISA and the Luminex assay were added at 3 different
carried out only for (a) analytes that showed good corre- concentrations in pooled normal healthy human serum
lation and used the same calibrators for calibration curves for measurement of recovery with 10 replicates or more.
in both the Luminex assays and ELISAs (leptin, insulin, The 3 concentrations (32, 128, and 512 ng/L) of MCP-1
and C-peptide) or (b) assays already calibrated against the were selected because they were commonly seen in our
WHO standards (MCP-1 and insulin). study in healthy individuals and in patients with meta-
The Bland–Altman plots for leptin, insulin, and C- bolic syndrome. The pools containing added Luminex
peptide in the Linco endocrine panel and MCP-1 in the calibrator were measured by the Luminex method, and
Upstate panel are shown in Fig. 1. The systematic differ- the pools containing the added ELISA calibrator were
ences between the 2 assay methods as calculated by the measured by ELISA. As shown in Table 2 of the online
Bland–Altman analysis were 11.0 g/L, 13.8 kIU/L, 3.9 Data Supplement, for the pools containing 128 and 512
g/L, and 0.21 kIU/L for leptin, insulin, C-peptide, and ng/L MCP-1, the ELISA gave mean recoveries of 98.8%
MCP-1, respectively. The 2 methods agreed fairly well for and 102.1%, respectively. However, in the Luminex assay,
leptin values ⬍50 g/L. Above 50 g/L, however, the the mean recoveries of MCP-1 were 58.2%, 49.1%, and
Luminex assays gave both higher and lower values than 54.6%, respectively, for the 3 concentrations.
the ELISA; the discrepancy for some measurements was
as large as 75%. For leptin, 7.5% of the patients had values effects of active weight loss on leptin and
that differed by ⬎2 SD. Agreement between 2 methods is insulin in individuals with the metabolic
generally acceptable if ⬍5% of values differ by ⬎2 SD. For syndrome
insulin, the 2 assays agreed very well at concentrations The patients with metabolic syndrome lost 7% of their
⬍30 mIU/L, but above 30 mIU/L, the Luminex assay initial weight [8.0 (0.55) kg, or 17.6 (1.2) pounds] after 4 – 6
Clinical Chemistry 51, No. 7, 2005 1107
weeks of the diet-induced weight loss program. Plasma and the median percentage changes in insulin were ⫺41%
concentrations of leptin and insulin at baseline and after and ⫺56%, respectively.
weight loss were measured by both the Luminex and
ELISA methods, and the results were compared. Mean comparison of MCP-1 between healthy
(SD) leptin concentrations measured by ELISA were sig- controls and metabolic syndrome patients
nificantly higher before weight loss compared with after with 4 or more risk factors
weight loss [36.3 (21.5) vs 19.8 (18.7) g/L; n ⫽ 40; P We compared plasma MCP-1 concentrations in healthy
⬍0.001, Wilcoxon signed-rank test]. The Luminex assay controls (n ⫽ 14) and metabolic syndrome patients with 4
also showed a significant mean (SD) difference after or more risk factors (n ⫽ 20). The MCP-1 concentrations
weight loss [47.4 (25.6) vs 27.6 (22.1) g/L; n ⫽ 40; P measured by ELISA and the Luminex assay are shown in
⬍0.001]. Insulin concentrations were also significantly Fig. 2 of the online Data Supplement. The mean (SD)
higher before weight loss than after weight loss whether MCP-1 concentrations measured by ELISA were signifi-
measured by ELISA [19.2 (17.3) vs 9.4 (7.8) mIU/L; n ⫽ cantly higher in the metabolic syndrome patients than in
13; P ⫽ 0.039] or the Luminex assay [37.4 (41.1) vs healthy controls [536.5 (316.1) vs 263.3 (96.5) ng/L; P
14.7 (16.0) mIU/L; n ⫽ 13; P ⫽ 0.028]. The percentage ⬍0.001, Wilcoxon signed-rank test]. The median MCP-1
changes in leptin and insulin after weight loss are shown values measured by ELISA were 414.6 and 231.4 ng/L for
in Fig. 1 of the online Data Supplement. The median patients and controls, respectively. The results obtained
percentage changes in leptin were ⫺54% and ⫺46% as with the Luminex assay also indicated that the mean (SD)
measured by the ELISA and Luminex assay, respectively, MCP-1 concentrations in the metabolic syndrome patients
1108 Liu et al.: Luminex Multiplexed Analysis of Human Plasma
were significantly higher than those in healthy controls (Fig. 1 of the online Data Supplement). The relative
[83.5 (101.1) vs 20.4 (8.6) ng/L; P ⬍0.001, Wilcoxon difference in plasma concentrations of MCP-1 between
signed-rank test]. In the Luminex assay, the median controls and metabolic syndrome patients also were sim-
MCP-1 values were 63.9 and 18.4 ng/L for patients and ilar in the 2 methods (Fig. 2 of the online Data Supple-
controls, respectively. ment). The total error acceptable for an assay depends on
the clinical setting in which the assay is being used. For
Discussion example, in clinical research, studies are frequently de-
Multiplex immunoassays using a small sample volume signed to examine the association of plasma concentra-
could facilitate clinical research in complex disorders such tions of analytes with the disease phenotype (in this case,
as the metabolic syndrome and could be extremely help- obese patients with the metabolic syndrome vs lean
ful in the evaluation of changes in biomarkers with controls) and the impact of a therapeutic intervention on
therapeutic interventions. The detection limits shown in the disease state (such as weight loss). In the present
and Thorpe (20 ). However, our data, as well as results 7. Tarnok A, Hambsch J, Chen R, Varro R. Cytometric bead array to
from a growing number of research studies, demonstrate measure six cytokines in twenty-five microliters of serum. Clin
that multiplex technology may be useful in clinical re- Chem 2003;49:1000 –2.
search to measure a large number of analytes to examine 8. Prabhakar U, Eirikis E, Davis HM. Simultaneous quantification of
proinflammatory cytokines in human plasma using the LabMAP
the association with a clinical phenotype and the effects of
assay. J Immunol Methods 2002;260:207–18.
therapeutic interventions, and that this technology may 9. Kellar KL, Kalwar RR, Dubois KA, Crouse D, Chafin WD, Kane BE.
be particularly useful when sample volume is limited, Multiplexed fluorescent bead-based immunoassays for quantita-
such as in large epidemiologic studies and clinical trials. tion of human cytokines in serum and culture supernatants.
Cytometry 2001;45:27–36.
10. Vignali DAA. Multiplexed particle-based flow cytometric assays.
This study was supported by Applied Technology Grant J Immunol Methods 2000;243:243–55.
No. 004949-0093-2001 from the state of Texas and an 11. Oliver KG, Kettman JR, Fulton RJ. Multiplexed analysis of human