Comparison of serum, EDTA plasma and P100 plasma for luminex-based biomarker multiplex assays in patients with chronic obstructive pulmonary disease in the SPIROMICS study

Blood is being collected from SPIROMICS participants as part of their baseline (initial) visit and additionally 1 and 3 years after the baseline visit. SPIROMICS subjects are requested to fast after midnight, and blood is drawn early in the day of the study visit. For the entire SPIROMICS, eight tubes are collected in the following order: Two 8.5 mL red-stoppered serum tubes [Vacutainer® Plus plastic serum tube; Becton-Dickinson (BD) Diagnostics, Franklin Lakes, NJ; product number 367888]; one 10 ml yellow-stoppered tube containing 1.5 mL ACD anticoagulant (BD product number 364606); two 10 mL and one 4 mL lavender-stoppered tubes containing a sprayed on K₂EDTA anticoagulant (BD product numbers 366643 and 367861); one 8.5 mL P100 red-stoppered plasma collection tube with a mechanical separator and sprayed on K₂EDTA anti-coagulant and proprietary protease inhibitor additives (antiproteases; BD product number 366448); one 2.5 mL red-stoppered tube with RNA preservation solution (Paxgene™ RNA, BD; product number 762165). All samples are processed within one hour of collection, aliquoted, and frozen at −80°C, shipped to the SPIROMICS Biospecimen Processing Center and kept frozen at −80°C for future use. Mean processing times for the samples used in this study were (in minutes) 39, 32, and 49 for serum, EDTA plasma, and P100 plasma, respectively. Processing involves immediate inversion of tubes several times after sample draw and centrifugation at room temperature at 1100–1300 relative centrifugal force (RCF) for 10 minutes in a swinging bucket rotor of 15 minutes in a fixed angle centrifuge for serum and EDTA plasma, and 2500 RCF for 15–20 minutes or 1100–1600 RCF for 30 minutes for P100 plasma. SPIROMICS protocols require dividing each blood collection tube into aliquots of 150 μl to minimize freeze-thaw cycles. The 13-plexes run in this pilot study required 3 aliquots each of serum, P100 plasma and EDTA plasma from each patient. The aliquots were sent frozen to Myriad-RBM, where they were thawed, pooled, diluted and immediately utilized for analyte determination according to standard practices. Each pooled sample was run in duplicate, providing 2 replicates from the same blood draw for each blood sample type.

We first identified priority biomarker candidates based on known COPD pathophysiology and previously published literature, then selected from the assays available at Myriad-RBM, which were primarily multiplexes (Luminex xMap technology, Myriad-RBM Inc., Austin TX). Each multiplex measured a number of analytes in addition to the priority biomarkers. The number of analytes per plex varied from 1–14 (see Additional file 1: Table S1 and Additional file 2: Table S2). In total, 105 analytes were evaluated on the 13 plexes, 12 of which were multiplexes.

We next selected samples from 24 SPIROMICS participants chosen to represent individuals with a range of disease severity assessed by Global Initiative for Chronic Obstructive Lung Disease (GOLD) spirometric stage classification (http://​www.​goldcopd.​org/​) at the time of the blood draw [mean/median age 64/65 years; 12 females, 12 males; six non-smokers, three at risk smokers, 5 GOLD stage 1 (mild), 4 GOLD stage 2 (moderate), 6 GOLD stage 3 (severe)]. Due to the small sample size, there was no intent to utilize the data from this study to correlate analyte levels to clinical phenotype; however, the range of GOLD spirometric stage provided an opportunity to assess some analytes that may have varying blood levels based on certain disease conditions associated with COPD.

The measured concentration of each analyte, as well as the lower limit of quantification (LLOQ), least detectible concentration (LDD; concentration three standard deviations above diluent blank reading), and the low to high normal range were provided by Myriad-RBM. The LLOQ was used as the lowest reliable value. It is defined as the lowest concentration of analyte reliably detected and at which the total error meets the laboratory’s requirements for precision. In this case, the laboratory’s requirements for precision is the concentration of an analyte at which the coefficient of variation of replicate standard (Myriad-RBM defined) samples is 30%. If a sample value was below the LLOQ, it was reported as < LLOQ for that analyte. Rarely, a sample could not be measured due to technical problems during processing and these were reported as ND (not determined).

For each analyte, identical analyses were carried out for the three sample types. In the first step, the percentage of samples below the LLOQ was calculated, and the number of subjects with both replicates ≥ LLOQ was determined. No further analysis was done on a particular analyte if one or both replicates were less than LLOQ in more than 50% of subjects. For the remaining analytes, descriptive statistics were calculated for the subset of subjects with both replicate values ≥ LLOQ including the mean, within-subject standard deviation, reliability coefficient, and within-subject coefficient of variation (CV). All statistical analyses were conducted using SAS version 9.2.

The within-subject and between-subject variance were calculated from the overall variance using a simple random effects linear model with a random subject intercept. The within-subject standard deviation is the square root of the within-subject variance. The reliability coefficient is the ratio of the between-subject variance and the total variance in the samples. Reliability is dependent on the overall variability of the samples. Thus, for analytes with very little variability between subjects, small variation in replicate samples within a subject may result in a reduced reliability coefficient. Conversely, for analytes with large variability between subjects, a relatively large variation in replicate samples within a subject may still result in a favorable reliability coefficient.

The within-subject coefficient of variation (CV) was calculated as the ratio of the within-subject standard deviation and the overall mean, multiplied by 100 (expressed as percentage). Smaller values of the CV indicate that the within-subject variation is small compared to the mean. In general, a CV of less than 10% is considered acceptable [10]. Alternatively, an analytic CV one half of the biologic CV may be useful in identifying change in analyte value [11]. Due to the large number of analytes and the relatively small sample size, no tests of statistical significance were performed. The following descriptive comparisons between sample types were identified as notable: differences in reliability >15%, ratios of coefficient of variation < 0.667 or > 1.5, and ratios of means < 0.667 or > 1.5.

The majority of analytes produced similar results in all sample types, both based on detection and on measurements of reliability and CV between duplicate samples (Figures 1, 2, and 3; Additional file 2: Table S2, Additional file 3: Figure S1). Exceptions are listed in Table 1 and highlighted in Figures 2 and 3.

Reliability and CV were not determined for twenty-three of the 105 analytes (22%) because they were not detectable in both replicates for at least 50% of subjects within each of the three sample types (Table 1). Four analytes, IL1A, IL1RN, IL12B, and OLR1 were consistently detected in serum but not in plasma; conversely, three analytes, fibrinogen [FGA_FGB_FGG], IFNG, and MMP2 were consistently detected in plasma samples but not serum. An additional 7 analytes (7%) were consistently detected in at least one of the 3 sample types, but had low reliability measurements (< 60%) or high CV (> 20%) in the sample types that were consistently detected.

Twelve analytes (11%) performed notably better in serum versus EDTA plasma and another 11 better in EDTA plasma than serum, based on descriptive cut-points for differences of reliability and ratios of coefficient of variation values (see Table 1, Figures 1 and 2). The majority of the analytes performed similarly in EDTA plasma compared to P100 plasma (Figure 1). Only one analyte (CCL4) performed better in P100 plasma relative to EDTA plasma, yet the difference was modest (Additional file 2: Table S2). Eight analytes showed notably better performance (lower CV) in EDTA plasma compared to P100 (Table 1). Reliability of EDTA and P100 plasma was within 15% for all consistently detectable analytes.

Except for analytes that were only detected in either serum or plasma, the mean expression values differed by > 1.5 fold between plasma and serum for 20 analytes (19%) (analytes listed in Table 1; Additional file 2: Table S2; Supplemental Figure 1 displays the mean and a dot plot for each analyte and sample type). Of these, serum produced higher values for 17 analytes (16%), while plasma produced higher values for three (3%). We identified no notable differences in mean levels between EDTA plasma and P100 plasma samples (data not shown).

Each multiplex had its own unique features related to analytes detection and sample-type performance, which are summarized in Table 2. Within each multiplex, a majority of the analytes were detectable in either serum or plasma, with the exception of Myriad-RBM multiplexes HMPC49 and HMPCORE1 where 3/5 and 12/17 analytes were not detectable or not reliably measured, respectively. Multiplex HMPCORE4 performed with noticeably worse CV for most analytes compared to other multiplexes, with only one analyte (GC) having a CV <10%. Serum had noticeably better reliability and/or CV for detected analytes in HCVD4, HMPC62 and HMPC83; while plasma generally performed better for detected analytes in HMP8, HMPC35, and HMPCORE4. For multiplexes HMPC19, HMPCORE1, and HMPCORE2, certain represented analytes performed better in serum and others in plasma. For HMPC42 and HMPC84, all sample types performed similarly.