We used regression models to directly test whether our aging studies were affected by six medical factors: renal cell carcinoma (RCC), transitional cell carcinoma, size of tumor, hypertension, systolic blood pressure or diastolic blood pressure. For renal cell carcinoma, we used a regression model predicting expression from Age, Sex, Tissue type, and a 0/1 variable indicating whether the sample came from a patient with renal cell carcinoma or not. The result gave a p-value for whether renal cell carcinoma affected each of the 44928 genes present on the Affymetrix DNA chip. The smallest p-value we saw was 0.00013. We would expect to see almost 6 such p-values by chance alone. This result indicates that the presence of renal cell carcinoma does not significantly affect the expression of any gene in the normal tissue from the same kidney, compared to normal tissues taken from kidneys without renal cell carcinoma.


Next, we plotted the results using only the age-regulated genes, to see if the presence of renal cell carcinoma might confound our results. We selected genes that show statistically significant (p< 0.001) age-regulation using either a model with renal cell carcinoma (RCC) term or without an RCC term. Supplemental figure 2 shows the age slopes for these genes. The vertical and horizontal axis has the slope from a model with and without the RCC term, respectively. The slopes change very little with and without the RCC term. As one might expect, many of the genes that are significant at the 0.001 level are just barely so. There were 866 genes significant in both models, 119 significant only when RCC was not in the model, and 86 significant only when RCC was in the model. The overall picture of age relationship changes very little whether a term for RCC is included or not in the model.


We repeated the analysis for other possible factors that might confound our results, such as transitional cell carcinoma, size of tumor, hypertension etc. Transitional cell carcinoma (TCC) was present in 13 patients, all of whom were old (Supplemental Figure 3). Thus if TCC affected gene expression from adjacent normal tissue, then it might bias our results on aging. Supplemental figure z shows data for presence or absence of TCC in the model. The smallest p-value for TCC was 8.8 x 10-6. The expected number of p-values this small in 44928 trials is 0.4, so this gene is not particularly compelling. The histogram of p-values looks uniform, as we would expect if TCC were very weakly related, or not related, to expression changes with age (data not shown). We have not used false discovery rate techniques for this problem, because the age coefficients for different genes are far from independent.

We also analyzed whether size of tumor, hypertension, systolic blood pressure or diastolic blood pressure biased our results on age-regulation (Supplemental figures 4-7). The regression slopes change very little with and without these factors, indicating that these factors do not strongly affect our analysis of age regulation.