The study investigated the serum-to-plasma ratio for 86 proteins using matched serum and plasma samples from 12 patients. Six proteins had more than 75% of the measurements below the LOD in serum or in both serum and EDTA plasma (IL1 alpha, IL2, IL5, IL13, IL33 and fibroblast growth factor 2 (FGF2)). The temporal reproducibility and the effect of storage were assessed using data from 6 studies. In total, we had data from 36 plates where the same serum bridging samples had been analyzed. On 31 plates, all 8 serum samples had been included and on the last 5 between 6 and 7 serum samples were used (Table 1). The temporal reproducibility was assessed for 81 proteins across all 6 studies after removing proteins where the assay was changed between versions, not included in both versions, or failed quality control in at least one study. For 88 proteins, it was possible to compare results across the five studies using the newer version of the Olink I-O. Table 2 lists all proteins included in each sub study.

For most proteins, the serum and EDTA plasma concentrations were similar (Additional file 1: Figure S1). Of the 86 proteins included in the sub-study, the mean serum-to-EDTA plasma ratio ranged from 0.41 to 3.01 (Table 3). Fifty-two proteins had a minor variation, with a mean ratio from 0.80 to 1.20, and the variation seemed to be randomly distributed between samples (Fig. 1). For the 12 samples, bilirubin ranged from 5 to 812 with two patients having a higher bilirubin than 200. There was no significant association between serum-to-EDTA plasma ratio and bilirubin after adjusting for multiple comparisons. Next, we evaluated the reproducibility of the ratio using the above-mentioned criteria. Figure 2 illustrates the consort diagram showing which proteins fulfilled each criterion.

Boxplot of serum-to-plasma ratio on linear normalized protein expression (NPX) scale for the 86 proteins with more than 75% of samples over limit of detection. Proteins sorted according to median serum-to-plasma ratio. Samples from each patient connected with dashed lines

Criterion 1 (CV of the serum – EDTA plasma ratio < 20%): The geometric CV of the ratio ranged from 5.58% to 123.45%. Thirty proteins had a CV of more than 20%. Of note, most of these proteins had either a high or low median NPX in both serum and EDTA plasma (Additional file 1: Figure S2), and the highest CVs were observed for proteins with high or low ratios.

Criterion 2 (The ratio between serum and EDTA plasma reported by Olink Proteomics [13] is within range (Q1—1.5 × IQR and Q3 + 1.5 × IQR) of the observed in our study: Of the 56 proteins with a reasonable CV, there were four proteins where Olink reported a ratio that was markedly different from the range observed in our study [T-cell surface glycoprotein CD4 (CD4), CD40 ligand receptor (CD40), C–C motif chemokine 2 (MCP-1), matrix metalloproteinase-7 (MMP7), tumor necrosis factor receptor superfamily member 12A (TNFRSF12A)]. Seven proteins had both a high CV in our study and a ratio reported by Olink that was outside the range in study [pro-epidermal growth factor (EGF), CD40 ligand (CD40-L), C-X-C motif chemokine (CXCL) 5, CXCL11, C–C motif chemokine (CCL) 17, caspase-8 (CASP-8), latency-associated peptide transforming growth factor beta-1 (LAP TGF-beta-1)] (Fig. 3). The ratio was not reported for six proteins (KIR3DL1, MUC-16, TNF, IL15, LAG3, IFN-gamma) in the data provided by Olink, and it is therefore not possible to compare the results for these proteins. Of note, all six passed the other three criteria.

Serum-to-plasma ratio in our study and as reported by Olink on linear normalized protein expression (NPX) scale. Red lines indicate boundaries of acceptance (25% quartile (Q) + 1.5 interquartile range (IQR) and 75% Q + 1.5IQR). Blue points indicate serum-to-plasma ratio reported by Olink. Data were not reported by Olink for 6 proteins: KIR3DL1, MUC-16, TNF, IL15, LAG3, IFN-gamma

Criterion 3 (The correlation coefficient found in regression analysis was above 0.7): Seventeen proteins had a poor correlation between results obtained in serum and EDTA plasma, with a correlation coefficient (r) ≤ 0.7. However, 16 of these also had a poor CV or Olink data were not within range, and therefore only one protein—decorin (DCN), was removed due to this criterion. Correlation diagrams for all proteins are shown in Additional file 1: Figure S3.

Criterion 4 (No association between concentration and ratio): For six proteins we identified an association between ratio and concentration. Five proteins had a slope of the Passing-Bablok regression which was significantly lower than 1 (platelet-derived growth factor subunit B (PDGF subunit B), angiopoietin-1 (ANGPT1), CXCL13, CCL23, MMP12), and for one protein the slope was significantly higher (CXCL1) (Additional file 1: Figure S3). A similar result was seen using Bland–Altman plots, where 5 of the 6 proteins (CXCL1, PDGF subunit B, ANGPT1, CXCL13, CCL23) had a significant association between ratio and protein concentration using linear regression (Additional file 1: Fig. S4). Five of the six proteins were already removed due to one of the prior criteria. Only CCL23 was removed due to an association between concentration and ratio.

Normalized protein expression for four bridging samples in BX0223 and BX0182. Boxplots showing results of all plates (chips) for each protein and bridging sample as measured in BX0223 (number of assays = 5) and BX0182 (number of assays = 9). Each graph represents results from one bridging sample. Results from each assay connected with colored lines

Of the 92 proteins included in the Olink I-O panel, 80 proteins were possible to evaluate on all four criteria for serum-to-EDTA plasma variation reproducibility. Of these, 44 proteins fulfilled all four and 36 failed on one or more of the criteria (Table 3 and Fig. 2).

In general, we observed a similar intra-study inter-assay variation based on the serum bridging samples in the three studies that used three or more plates and in those that used the newest Olink panel (BX0182, BX0188, and BX0223). For these studies, 88 proteins were in common after removing four (IL1 alpha, IL13, IL33, and ARG1) that did not pass quality control. Very few proteins had a variation higher than 1 NPX, and only a few random outliers were observed. Figure 4 shows NPX values for all assays in BX0223 and BX0182 for 4 of the bridging samples. The three studies had a similar intra-study inter-assay variation. The mean CV was 11.3% in BX0188, 16.4% in BX0182, and 12.3% in BX0223. For the most recent of the three studies (BX0223), where bridging samples had been stored the longest, only seven proteins had a CV higher than 20% (IL2, IL5, CD40-L, endothelial nitric oxide synthase (NOS3), natural killer cells antigen CD94 (KIR3DL), pleiotrophin (PTN), IL4), but except for CD40-L, all were characterized by low NPX values, with several being below LOD. For the oldest study (BX0144) that used an earlier version of the Olink I-O panel, the inter-assay variation was higher. When comparing all 91 proteins used in the panel and passed quality control in the study, the interstudy CV was 26.0%, and when only the 81 proteins in common between all studies were included, the CV was 22.6%.

In all studies, pooled serum samples had been included on all plates as recommended by the manufacturer. For all studies the inter-assay intra-study variation based on the serum bridging samples were similar or better than the inter-assay variation based on pooled serum samples (36% for BX0144, 17% for BX0182, 10% for BX0188, and 17% for BX0223).

The mean inter-study CV across all studies based on 81 proteins measured using the serum bridging samples was 41.3% (range: 17.6–109.9%). After normalization between studies the mean inter-study CV decreased to 26.2% (range 11.3–108.5%). The inter-study variation was significantly lower after removing BX0144, which used the oldest version of the panel (pre-normalization mean CV = 26.9%, range 13.6–73.8%, post-normalization CV = 18.9%, range 7.3–66.1%). The pre-normalized CV was further reduced in the studies that only used v. 3111 (CV = 19.7%, range: 11.8–59.1%), but not the post-normalization CV (mean = 19.05%, range 7.5–67.2%). For the five studies using v. 3111, 3 proteins had a CV higher than 40% [IL4 (CV = 58.9%), MHC class I polypeptide-related sequence A/B (MIC A/B) (CV = 51.7%), and IL5 (CV = 41.6%)], and all three had low median NPX values and/or several samples below LOD. Pearson correlation analysis (Fig. 5) showed a high correlation (r ≥ 0.97) between results obtained in studies that used the newer version of the panel (v. 3111 and v. 3112) and highest for studies using the same version. The correlation decreased (r = 0.92–0.93) when results from BX0144 were compared with the six other studies, including the 81 proteins in common. One protein, LAP.TGF.beta.1, had a large difference between BX0144 (NPX = 2.99) and the other studies (NPX: 9.76–10.56). When LAP.TGF.beta.1 was removed, the correlation between the studies improved (r = 0.95–0.97). The Bland–Altman plots showed a similar pattern, with a smaller variation between studies using the same panel but a larger variation between the BX0144 and the other studies. Of note, no trends were observed in the plots (Additional file 1 Figure S5).

Correlation diagram comparing bridging sample results between studies using median NPX for all proteins in common between BX0144 and BX0182 and the 5 other studies. In each plot, a dot denotes a given protein. Latency-associated peptide transforming growth factor beta-1 (LAP TGF-beta-1) highlighted as a significant outlier