Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
, 11 (1), 167

Machine-learned Analysis of Global and Glial/Opioid Intersection-Related DNA Methylation in Patients With Persistent Pain After Breast Cancer Surgery

Affiliations

Machine-learned Analysis of Global and Glial/Opioid Intersection-Related DNA Methylation in Patients With Persistent Pain After Breast Cancer Surgery

Dario Kringel et al. Clin Epigenetics.

Abstract

Background: Glial cells in the central nervous system play a key role in neuroinflammation and subsequent central sensitization to pain. They are therefore involved in the development of persistent pain. One of the main sites of interaction of the immune system with persistent pain has been identified as neuro-immune crosstalk at the glial-opioid interface. The present study examined a potential association between the DNA methylation of two key players of glial/opioid intersection and persistent postoperative pain.

Methods: In a cohort of 140 women who had undergone breast cancer surgery, and were assigned based on a 3-year follow-up to either a persistent or non-persistent pain phenotype, the role of epigenetic regulation of key players in the glial-opioid interface was assessed. The methylation of genes coding for the Toll-like receptor 4 (TLR4) as a major mediator of glial contributions to persistent pain or for the μ-opioid receptor (OPRM1) was analyzed and its association with the pain phenotype was compared with that conferred by global genome-wide DNA methylation assessed via quantification of the methylation in the retrotransposon LINE1.

Results: Training of machine learning algorithms indicated that the global DNA methylation provided a similar diagnostic accuracy for persistent pain as previously established non-genetic predictors. However, the diagnosis can be based on a single DNA based marker. By contrast, the methylation of TLR4 or OPRM1 genes could not contribute further to the allocation of the patients to the pain-related phenotype groups.

Conclusions: While clearly supporting a predictive utility of epigenetic testing, the present analysis cannot provide support for specific epigenetic modulation of persistent postoperative pain via methylation of two key genes of the glial-opioid interface.

Conflict of interest statement

The authors declare that they have no competing interests.

Figures

Fig. 1
Fig. 1
Methylation at d = 14 CpG sites located in the OPRM1 or TLR4 genes or in the retrotransposon LINE1 (raw data, for numerical results, see also Table 1). a Raw data are shown separately for group membership to the persistent pain or non-persistent pain phenotype groups. The widths of the boxes are proportional to the respective numbers of subjects per group. The quartiles and medians (solid horizontal line within the box) are used to construct a “box and whisker” plot. The whiskers add 1.5 times the interquartile range (IQR) to the 75th percentile or subtract 1.5 times the IQR from the 25th percentile and are expected to include 99.3% of the data if normally distributed. The notches indicate the confidence interval around the median based on median ± 1.57 ∙ IQR/n0.5. b Results of Wilcoxon tests for group differences in the methylation status at each CpG sites. The bars indicate the obtained p values, rescaled as –log10(p). Uncorrected and corrected significance thresholds are shown as horizontal lines. A significant difference is found when the bar exceeds the line. The figure has been created using the R software package (version 3.4.4 for Linux; http://CRAN.R-project.org/ [40])
Fig. 2
Fig. 2
Explorative analysis of the correlations between the methylation status at d = 14 CpG sites in OPRM1, TLR4, or LINE1 and with the pain ratings acquired between 1 and 36 months after breast cancer surgery. At the lower left part, the correlations are shown as ellipses. The narrower the ellipse is drawn, the higher is the correlation coefficient. Positive correlations are indicated by ellipses directed from the lower left corner to the upper right corner of each cell. Negative correlations are indicated by ellipses drawn in the opposite direction from the upper left to the lower right corner of each cell. Ellipses are colored according to the color code of Spearman’s ρ [45] shown at the bottom of the panels. At the upper right parts, the correlations are provided numerically as values of Spearman’s ρ (colored). The corresponding p values are shown in black numbers below the correlation coefficients; “0” indicates p < 1 × 10−5. The figure has been created using the R software package (version 3.4.4 for Linux; http://CRAN.R-project.org/ [40]) and the library “corrplot” (https://cran.r-project.org/package=corrplot [63])
Fig. 3
Fig. 3
Clustering of subjects based on DNA methylation at CpG sites in OPRM1, TLR4, and LINE1, obtained using unsupervised machine learning. U-matrix visualization of the data structure found via a projection onto a toroid neuronal grid using a parameter-free polar swarm, Pswarm consisting of so-called DataBots, which are self-organizing artificial “life forms” that carry vectors of the DNA methylation. a The U-matrix visualization was colored as a top view of a topographic map with brown (up to snow-covered) heights and green valleys with blue lakes. Watersheds indicate borderlines between two different clusters. b Superimposing the pain phenotype group structure indicated considerable coincidence with the cluster separation, which was supported by a significant χ2 test of the cross table of clusters versus pain phenotype groups. Please note the different meaning of the coloring of the data points in the two panels, cluster in panel a but pain phenotype groups in panel b. The figure has been created using the R software package (version 3.4.4 for Linux; http://CRAN.R-project.org/ [40]) and the library “DatabionicSwarm”, https://cran.r-project.org/package=DatabionicSwarm [64])
Fig. 4
Fig. 4
Data structure found in the input space of d = 14 CpG methylations acquired from patients with either persistent (n = 70) or non-persistent (n = 70) pain after breast cancer surgery. The data structure has been obtained by means of data projection principal component analysis on the non-normalized data as suggested by the results of the PC-corr analysis [50]. The PCA plot associated to this analysis shows the sample separation in the first and second component (PC1 versus PC2) yielded the best explained variance for non-normalized, non-centered PCA. The marginal distribution plots show the segregation of the pain phenotype groups along the first principal component. The figure has been created using the R software package (version 3.4.4 for Linux; http://CRAN.R-project.org/ [40]) and the library “ggplot2” (https://cran.r-project.org/package=ggplot2 [65])
Fig. 5
Fig. 5
Analysis of the drop in the classification accuracy (Table 3) of five different algorithms (classification and regression trees (CART), k-nearest neighbors (kNN), support vector machines (SVM), multinomial regression (“regression”), and naïve Bayes adaptive classification) when the methylation information, originally comprising a total of d = 14 CpG sites located in OPRM1, TLR4, or LINE1, was reduced to two or one genes. The numbers indicate the difference in classification accuracy, obtained in several training scenarios of reduced sets of gene-specific CpG islands, to that obtained with the respective algorithm when trained with the full data set. Subsequently, applying hierarchical clustering (Ward [67]) to these differences, a pattern of two groups of the tested scenarios emerged. In the first cluster (top), the accuracy did not change when using a reduced data set for training. By contrast, the accuracy dropped in scenarios included in the second cluster (bottom). The figure has been created using the R software package (version 3.4.4; http://CRAN.R-project.org/ [40]) and the “heatmap.2” function of the R package “gplots” (G.R. Warnes; https://cran.r-project.org/package=gplots)

Similar articles

See all similar articles

References

    1. Liu T, Gao YJ, Ji RR. Emerging role of Toll-like receptors in the control of pain and itch. Neurosci Bull. 2012;28(2):131–144. doi: 10.1007/s12264-012-1219-5. - DOI - PMC - PubMed
    1. Landmark T, Dale O, Romundstad P, Woodhouse A, Kaasa S, Borchgrevink PC. Development and course of chronic pain over 4 years in the general population: The HUNT pain study. Eur J Pain. 2018;22(9):1606–1616. doi: 10.1002/ejp.1243. - DOI - PubMed
    1. Woolf CJ, Salter MW. Neuronal plasticity: increasing the gain in pain. Science. 2000;288(5472):1765–1769. doi: 10.1126/science.288.5472.1765. - DOI - PubMed
    1. Stucky CL, Gold MS, Zhang X. Mechanisms of pain. Proc Natl Acad Sci U S A. 2001;98(21):11845–11846. doi: 10.1073/pnas.211373398. - DOI - PMC - PubMed
    1. Ji RR, Kohno T, Moore KA, Woolf CJ. Central sensitization and LTP: do pain and memory share similar mechanisms? Trends Neurosci. 2003;26(12):696–705. doi: 10.1016/j.tins.2003.09.017. - DOI - PubMed

Publication types

Feedback