Correlation analysis of two-dimensional gel electrophoretic protein patterns and biological variables

Abstract

Background: Two-dimensional gel electrophoresis (2DE) is a powerful technique to examine post-translational modifications of complexly modulated proteins. Currently, spot detection is a necessary step to assess relations between spots and biological variables. This often proves time consuming and difficult when working with non-perfect gels. We developed an analysis technique to measure correlation between 2DE images and biological variables on a pixel by pixel basis. After image alignment and normalization, the biological parameters and pixel values are replaced by their specific rank. These rank adjusted images and parameters are then put into a standard linear Pearson correlation and further tested for significance and variance.

Results: We validated this technique on a set of simulated 2DE images, which revealed also correct working under the presence of normalization factors. This was followed by an analysis of p53 2DE immunoblots from cancer cells, known to have unique signaling networks. Since p53 is altered through these signaling networks, we expected to find correlations between the cancer type (acute lymphoblastic leukemia and acute myeloid leukemia) and the p53 profiles. A second correlation analysis revealed a more complex relation between the differentiation stage in acute myeloid leukemia and p53 protein isoforms.

Conclusion: The presented analysis method measures relations between 2DE images and external variables without requiring spot detection, thereby enabling the exploration of biosignatures of complex signaling networks in biological systems.

Figure 1

2DE Image Correlation. 2DE image correlation relies on an aligned, normalized stack of 2DE images and a numerical label associated with every gel. Pixel per pixel correlation between gel intensities (red arrow) and the external variable (blue arrow) creates a new image, showing areas in the gel that relate to the external parameter. In comparison to standard gel analysis methods, spot detection is not necessary and therefore less bias is introduced into the analysis process. This technique also recognizes moving spots and spot shapes that change.

Figure 2

Correlation towards a simulated 2DE gel-stack. (A) Eight snapshots taken from a stack of 15 simulated gels generated using Gaussian bumps. Each image contains simulated spots with particular characteristics. See Material and Methods for formula and details. (B) Correlation between the gel-stack and the variable t. Upper gels (a-c) visualizes the correlation, lower gels (a'-c') are masked to visualizes only useful correlations. Correlation analysis was performed relying on different background removal methods. (a, a') without background removal; (b, b') with background subtraction, (c, c') using background division. (C) Correlation analysis under the presence of normal distributed noise: (a) 25%, (b) 50%, (c) 75% and (d) stepwise increasing noise from 0% to 75%. (D) Correlations towards two randomized sets of biological parameters. (E) Correlation towards the variable t polluted with (a) 6% outliers and (b) 13% outliers.

Figure 3

Correlation of p53 isoforms towards cancer type (AML or ALL). (A) Correlation analysis of p53 2DE immunoblots from AML and ALL patient samples. A total of 73 immunoblot images from AML and 16 images from ALL were analyzed (left, correlation; right, masked correlation). Green color indicates positive correlation for ALL (maximum positive correlation 0.5557), and brown color indicates negative correlation (maximum negative correlation 0.1464). (B) p53 protein expression in mature lymphocytes, neutrophile granulocytes, and monocytes from healthy donors, examined by 2DE (a-c) and one-dimensional immunoblot (d). For comparison, protein extract from the different normal cell types were analyzed by one-dimensional gel electrophoresis and immunoblotting (d). In the monocyte isolates, immunoglobulin subunits (heavy chain, light chain) from the isolation procedure were detected in addition to weak p53-δ spots (c). Parallel samples of lymphocytes, monocytes and neutrophile granulocytes were analyzed by one-dimensional gel electrophoresis for direct comparison of the protein level in the different cell types on the same immunoblot (d). All immunoblots shown are representative for three or more performed experiments. See Material and Methods for details on cell separation and immunoblotting technique.

Figure 4

Correlation of p53 protein isoforms towards AML differentiation (FAB). Green indicates correlation with more differentiated forms of AML. Such areas in 2DE images of M5 will have a higher intensity than 2DE images of M0. Brown indicates anti-correlation with the more mature forms of leukemia cells. Such areas in 2DE images of M0 will have a higher intensity than 2DE images of M5. (A) Correlation landscape of p53 in 73 AML images related to differentiation direction and stage (FAB, French-American-British classification). The vertical axis sets out the absolute correlation value. (B) Correlation image demonstrating statistical significant alterations in p53. Profile 1 shows the p53-δ region containing four correlating spots (r = 0.2). Profile 2 shows the sub-δ region anti-correlating at positions e and f. Profile 3 is the p63 region (p53 family member) correlating towards the more differentiated leukemia's. Profile 4, a p53 region anti-correlating with differentiated AML.

Figure 5

Intra-image testing for verification whether a combination of opposing or similar correlations relates to the external variable. When the correlation analysis reveals opposing or similar correlation in two areas, the relation between those two areas might correlate towards the external variable. Two examples are given. (A) shows that the α-area correlates negatively and the δ-area correlates positively. Does the intensity difference between the α-area and δ-area correlate with the external parameter ? To answer this, one first calculates for every image the total intensity in areas with the size of the bounding boxes of α and δ. (Their sizes are designated sx_α, sy_α, sx_δ and sy_δ). Thereafter, the images are slided over each other (the red arrow, translation dx, dy) and subtracted prior to correlation. (B) The result shows no correlation at observation point o₁, indicating that the difference between α and δ does not relate to the AML differentiation stage. (C) Given the positive correlation in the δ-region and negative correlation in the sub-δ region, we want to determine whether a mass change relates to the AML differentiation stage. Image preprocessing consists of shifting the image upwards (along the red arrow, which is parallel to the mass axis) and subtracting, it from the original prior to correlation. (D) The result at observation point o₂ indicates that a mass change of p53-δ strongly correlates to AML differentiation.