Dual-color proteomic profiling of complex samples with a microarray of 810 cancer-related antibodies

Abstract

Antibody microarrays have the potential to enable comprehensive proteomic analysis of small amounts of sample material. Here, protocols are presented for the production, quality assessment, and reproducible application of antibody microarrays in a two-color mode with an array of 1,800 features, representing 810 antibodies that were directed at 741 cancer-related proteins. In addition to measures of array quality, we implemented indicators for the accuracy and significance of dual-color detection. Dual-color measurements outperform a single-color approach concerning assay reproducibility and discriminative power. In the analysis of serum samples, depletion of high-abundance proteins did not improve technical assay quality. On the contrary, depletion introduced a strong bias in protein representation. In an initial study, we demonstrated the applicability of the protocols to proteins derived from urine samples. We identified differences between urine samples from pancreatic cancer patients and healthy subjects and between sexes. This study demonstrates that biomedically relevant data can be produced. As demonstrated by the thorough quality analysis, the dual-color antibody array approach proved to be competitive with other proteomic techniques and comparable in performance to transcriptional microarray analyses.

Fig. 1.

Quality control measures for an antibody microarray of 1,800 features. a, two-color positional controls facilitate grid tracking, slide orientation and support the verification of spot segmentation. They also serve as color standards during the imaging process. b, Sypro Ruby staining of all proteins acts as a quality measure of antibody immobilization. Negative control spots show no immobilized protein. c, incubation with 5 nm concentrations of each fluorescently labeled antibody against rabbit IgG (green signals) and mouse IgG (red) confirmed that the majority of antibodies on the microarray were produced in rabbits. Some antibodies, which were stained by Sypro Ruby but not in the incubation with secondary antibodies, were produced in goat or hamster. d, fluorescence image of a microarray incubated with two plasma samples of healthy donors for array quality control. One sample was labeled with the dye Dy-549, the other with dye Dy-649 before a competitive incubation in a dual-color mode. e, fluorescence image of a representative array of the analysis of urine samples from patients with pancreatic cancer. A urine sample and a reference consisting of a pool of samples from diseased and healthy subjects were labeled with different fluorescent dyes and incubated on the array in a competitive dual-color assay.

Fig. 2.

Comparison of the assay robustness and differentiation power of one-color and dual-color mode experiments. a, two plasma samples (Sample A, Sample B) were labeled with the fluorescent dye Dy-549 and incubated in one-color and dual-color assays. CVs were calculated for 6 (Sample A), 4 (Sample B), or 10 (Sample A + B) replicate measurements derived from two, three, and five microarrays and their intra-array duplicates. Each box plot represents 872 data points that correspond to all array features besides positional and negative controls. CV were significantly lower for dual-color mode measurements of the two samples. The combination of both samples leads to increased CV for the dual-color but not for the one-color measurements. The increase (expected because of biological variability) indicates a better discriminative power of the dual-color approach. b, discriminative power of the two measurement modes was assessed by hierarchical clustering of the nonnormalized and non-log-transformed data. For dual-color measurements, repeated measurements of the samples on different microarrays clustered better, leading to a superior differentiation of the two samples.

Fig. 3.

Assay robustness of dual-color experiments on antibody microarrays. The same plasma sample was labeled with two different fluorescent dyes and incubated on the same slide. a, in the scatter plots, correlation of the red and green signal intensities of all array features (n = 1,744, tracking controls and negative controls excluded) are shown for three representative arrays. For the whole set of 20 microarrays, the average Pearson's correlation coefficient was 0.93. Only one protein, ACTB, captured by 18 antibody replicates (blue dots), exhibited significant differences. b, distribution of the CVs of the 1,800 array features. CVs were calculated for the ratios of red and green signal intensities. Each colored line represents the CVs for four arrays of the same production batch. The mean values ranged from 9 to 14% within production batches and 13% (black line) for all production batches combined. c, measurement accuracy of complex samples: a plasma sample was labeled with two fluorescent dyes in separate reactions and mixed in different ratios. After incubation, the ratios of the signal intensities show good correlation with the defined ratios of the plasma mixtures.

Fig. 4.

Influence of serum depletion. a, SNRs obtained for depleted and nondepleted serum samples: the data are derived from five array measurements each, including dye swapping. Serum depletion improved the overall SNR for neither all the array features (left panel) nor a set of 15 low-abundance cytokines (right panel). b, heat map of the proteins that showed differential abundance between the two samples; overrepresented proteins are shown in red and under-represented proteins, in green. c, a volcano plot of LIMMA analysis represents the distribution of log-fold change and adjusted p values for all proteins. Besides albumin, a large set of 110 proteins was codepleted, whereas others were enriched. d, biplot of proteins and samples according to their expression profiles resulting from correspondence analysis with M-CHiPS: samples are depicted as colored squares, differentially abundant proteins as black dots. The closer the colocalization of two spots (both samples and proteins), the higher is the degree of association between them. Samples located in the same direction from the plot centroid exhibit a similar expression pattern. The further a protein is located in the same direction as a set of samples, the more specific it is.

Fig. 5.

Profiling the urine proteome of patients with pancreatic adenocarcinoma and healthy controls. a, correspondence analysis resulted in a biplot of differentially abundant proteins and the samples. Samples are depicted as squares that are colored according to disease state and sex; black spots represent differentially expressed proteins. Each sample is represented by four measurements representing the incubations on two arrays as well as two intra-array replicates. Measurements located in the same direction from the centroid of the plot exhibit a similar expression pattern. The smaller the distance between two samples, the higher is the concordance of their expression profiles. In addition to the more gradual variations, proteins were found that are particularly associated with the different sample groups. This association is indicated in the correspondence analysis plot by localization in the same direction off the centroid as the respective sample type; the further the distance to the centroid, the better is the correlation. One protein (KLK3) could not be displayed in the graph because it is located outside of the plotting region to the left, very strongly differentiating the male from the female samples. b–e, volcano plots summarize the results of LIMMA analyses. The log-fold changes and adjusted p values are shown for the sex-specific comparisons of the healthy (b) and diseased (c) subgroups as well as the disease-specific comparisons of the female (d) and male (e) subgroups. The red line marks a significance level of p = 0.05.