Our study design had three samples in each set; the samples consisted of DNA from blood, tumor, and adjacent normal tissue from the same individual. We analyzed 90 samples (30 sets, 3 samples in each set). Global DNA methylation was determined using methylation-sensitive Hpa II digestion followed by hybridization to Affymetrix 500K SNP arrays 
. Because our goal was to assess methylation in a quantitative manner, it was necessary to factor the underlying variation inherent in the DNA among tested individuals into the methylation score. Conventional genotyping without Hpa II digestion on the 90 samples supplied the baseline DNA variation information. To evaluate the quality of genotype experiments, we compared genotype calls that were generated using the Affymetrix Gtype 4.0 software between blood and normal experiments. The genotype call rates generally exceeded 99% for blood and normal esophageal DNA, and the concordances between the genotype calls of the two tissues were in the range of 98.8%–99.8%, demonstrating high quality of the genotype data. For quantitative evaluation of DNA methylation data, we applied a method that we previously developed to analyze chromatin immunoprecipitation (ChIP) data generated using the Affymetrix SNP array experiments 
. This method is briefly summarized in the Materials and Methods
To explore the variation of DNA methylation patterns in relation to tissue types and genetic background, we performed principal components analysis (PCA) to visualize DNA methylation patterns among samples in reduced dimension space (see Materials and Methods
for details about PCA analysis). We projected the samples using the first two principal components (PC1 and PC2). Each principal component (PC) is a linear combination of DNA methylation scores measured from SNPs across the whole genome with certain attributes. The Sty I Affymetrix SNP chip contains 238,304 SNPs. Of these SNPs, 62,765 were contained within Affymetrix probes homologous to regions in the sample DNA with the attributes required for our methylation analysis: (1) the Sty I fragment has at least one Hpa II site; (2) the SNP within the fragment does not overlap an Hpa II site; and (3) the SNP is located on an autosome. The results of PCA using data from these 62,765 SNPs are shown in (note: a similar result was observed using the data from the Nsp I chip). In this study PC1 and PC2 provided an efficient way to visualize relationships among the samples in two-dimension space. Two sample clusters were evident, which corresponded to the blood and normal samples, supporting the idea that DNA methylation is dependent on tissue types. Although PC1 and PC2 captured only 24% of the total variation, they grasped the biological variation due to different tissues and different individuals. Analyses involving additional principal components didn't yield any new insight into the nature of DNA methylation.
Analysis of DNA methylation patterns in blood and esophageal tissue from 30 ESCC patients.
A unique feature of our study design was the comparison of DNA methylation among multiple tissues from a single individual. Of particular interest was the comparison of DNA methylation in blood versus normal tissue. We wanted to know whether DNA methylation from blood and esophageal tissues from the same individual shared a similar pattern. This central question addresses the potential role of genetics in determining tissue methylation 
. In the graph shown in , we noted that blood and normal esophageal samples from a single individual had similar scores in PC1 (paired samples are connected by dotted lines in ), indicating that these two samples from an individual had a similar DNA methylation level as measured by the PC1 score. To illustrate that the two tissues from the same individual had similar PC1 scores, we generated a plot using a unit length arrow that emanated from the normal sample and pointed to the paired blood sample in the direction corresponding the line shown in for each individual (). These arrows clearly point in the same direction, indicating similar PC1 scores for the paired samples from the same individual. Our data showed a greater similarity in methylation between the two tissues from a single individual, as demonstrated by similar methylation in PC1 score. One interpretation of this result is that genetic background can influence DNA methylation, as we have also previously demonstrated at the level of chromatin modifications 
. To evaluate whether DNA sequences alone could explain this pattern of PCA, we also performed similar analyses using the measurements of genomic DNA from the genotyping experiments (Sty experiment without Hpa II digestion). PCA was also performed using quantitative values of the hybridization signals 
rather than the conventional interpretative approach, i.e. using genotyping calls generated according to a calling algorithm. Comparing the results from DNA methylation and genomic DNA, we found two important differences: (i) The clusters in the projection of genomic DNA were not well separated (); and (ii) The arrows in and were more randomly distributed than those in and . In the DNA analysis, blood and normal esophageal samples from a single individual produced very different scores in PC1 (). Thus, the similarities of PC1 scores in DNA methylation from two different tissues of a single individual are specific. To understand the distribution of the arrows, we generated an angle plot (Figure S1
). The angles are more uniform and have smaller standard deviations in the methylation data than the DNA data (Figure S1A
and Figure S1B
, also see Table S1
). We conclude that global DNA methylation is affected by the genetic background. The two clusters of samples correspond to the two tissue types (), indicating that tissue type is an important determinant of DNA methylation.
Analysis of DNA methylation patterns in blood, normal esophageal tissue, and tumors.
To further characterize the relationship among different individuals and tissues, we performed pair wise correlation analyses and displayed the results of methylation in and DNA in . The results from these analyses support our interpretation of PCA. Specifically, correlations between 3 tissues are apparent in (DNA) revealed by the 3 yellow diagonal lines. The correlations between blood and tumor samples are least because the presence of genomic instability in tumors and some cross-contamination between normal and tumor samples. Correlations in methylation analyses are generally less compared to the DNA analysis. However, the 3 yellow diagonal lines remain, suggesting that DNA methylation is similar among different tissues from the same individual. The main conclusion that genetic background influences methylation is supported by correlation analysis.
Heat map displays pair wise correlation results for methylation and genomic DNA data.
We extended the DNA methylation analysis to include tumors from the same individuals. As shown in , three clusters are evident, which correspond to the blood, normal, and tumor samples. PC1 and PC2 capture 23% of the variance in these data. Furthermore, all three tissues from the same individual shared similar DNA methyation angle signature, as demonstrated by a similar direction of the arrows shown in (green arrows point from normal to blood; red arrows point from tumor to blood). As a control, PCA projection using data from DNA analysis showed random distribution of samples (). Similarly, the angle plots showed narrower distribution and smaller standard deviation from methylation data than DNA data (Figure S1c
vs. Figure S1d
and Figure S1e
vs. Figure S1f
, also see Table S1
). Thus, global DNA methylation angle signatures in different tissues, including both normal and tumor tissues, are similar in the same individual, indicating a strong influence of genetic background on DNA methylation.
To further analyze genetic influence on DNA methylation, the differences between blood and normal tissue for the same individual and different individuals are shown in for ten selected SNP-marked fragments within the indicated genes. Each circle represents the difference in methylation measurements between blood and normal tissue (blue, from the same individual, labeled as Hpa2.paired; red, from two different individuals, labeled as Hpa2.unpaired) or the difference in DNA analysis (green, from the same individual labeled as gDNA.paired; pink, from two different individuals, labeled as gDNA.unpaired). The absolute difference in methylation is smaller in the two tissues from the same individual in these graphs (blue circles, labeled as Hpa2.paired) than in the two tissues from two different individuals (red circles, labeled as Hpa2.unpaired). To understand the overall distribution of the quantitative effect of genetic influence on DNA methylation across SNPs, we applied the Ansari-Bradley two-sample test, a non-parametric method, to compare the scale parameters for two sets of differences between blood and normal assays: the first set contains 30 differences with one for each individual; the second set contains 870 differences with one for each pair of two different individuals. The method tests the ratio of the scales for the two sets of the differences. The alternative hypothesis is that a ratio less than one means that the scale for the paired differences is less than the scale for unpaired differences. As a control, we performed similar analyses using data from DNA measurements. The ratio of the scales is significantly less than one for many of the tested SNPs (, red bars, labeled as Hpa2, indicate methylation data), showing greater similarity of methylation in two different tissues from the same individual. Some SNPs also showed smaller p-values for DNA measurements (, green bars, labeled as gDNA), reflecting DNA differences in the genetic background of different individuals. Nevertheless, the distribution is clearly shifted to the right (smaller p-value, indicating a smaller ratio of the scales) for the methylation data (, red bars). A non-parametric method was used instead of an F-test because the Shapiro-Wilk normality test showed that the differences were not normal for 65% of SNPs in the methylation data and 70% of SNPs in the genomic DNA data.
Analyses of DNA methylation difference between blood and normal esophageal tissue for single individuals.
In conclusion, we found that DNA methylation characteristics in normal esophageal tissue and blood from the same individual were remarkably similar. Our results indicate that genetic background as well as tissue environment can influence global DNA methyation patterns. This conclusion is consistent with previous studies that showed genetic background affected global chromatin modifications