Cell type-specific properties and environment shape tissue specificity of cancer genes

Abstract

One of the biggest mysteries in cancer research remains why mutations in certain genes cause cancer only at specific sites in the human body. The poor correlation between the expression level of a cancer gene and the tissues in which it causes malignant transformations raises the question of which factors determine the tissue-specific effects of a mutation. Here, we explore why some cancer genes are associated only with few different cancer types (i.e., are specific), while others are found mutated in a large number of different types of cancer (i.e., are general). We do so by contrasting cellular functions of specific-cancer genes with those of general ones to identify properties that determine where in the body a gene mutation is causing malignant transformations. We identified different groups of cancer genes that did not behave as expected (i.e., DNA repair genes being tissue specific, immune response genes showing a bimodal specificity function or strong association of generally expressed genes to particular cancers). Analysis of these three groups demonstrates the importance of environmental impact for understanding why certain cancer genes are only involved in the development of some cancer types but are rarely found mutated in other types of cancer.

Figure 1. Overview of the study.

(A) Some cancer genes (e.g. BRCA1) are only found mutated in certain types of cancers while general-cancer genes (e.g. p53) are associated with cancers from many different tissue origins. (B) A cancer gene specificity score was computed for each gene implementing the intuition that a high significance in a single tumor type and a low significance in the pan-cancer set are indicative of a specific-cancer gene. (C) The ranking of cancer genes by specificity was used to computationally detect functions that are predominantly associated with specific-cancer genes as compared to general ones (and vice versa) using a variant of the GSEA algorithm. (D) Proteins were grouped by the tissues in which they are involved in the development of cancer. Environmental chemicals and viral proteins frequently interacting with genes associated with one type of cancer were detected.

Figure 2. Expression promiscuity and agreement with tissue of pathology of cancer genes.

(A) The number of tissues in which a cancer gene is expressed is shown versus its specificity. The small correlation between these variables (0.12; Pearson correlation) is not significant (p > 0.05). (B) We determined the tissue with the highest expression level of 130 highly specific-cancer genes (i.e., only associated with one type of cancer) and found that only 4 are expressed highest in the tissue (or in a closely related tissue) from which the respective cancer originates.

Figure 3. Cellular functions associated with specific- or general-cancer genes.

(A) Enrichment of representative functional categories among specific- and general-cancer genes. Strongly associated (q < 0.05 or q < 0.001) categories are shown in darker colors, marginally significant categories (p < 0.05) in lighter colors. (B) Pie chart illustrating the amount of functional categories enriched among specific- (6 categories) and general- (52 categories) cancer genes. (C) The most cancer-general genes tend to be functionally more similar to each other (measured by the distribution of GO semantic similarities of all pairs of genes) than the most cancer-specific genes (p < 4.8e-15; Wilcoxon-Mann-Whitney test). To minimize effects due to curation differences, only genes studied 500 times or less were considered. Among those the 30 most specific and most general genes were compared with respect to semantic similarity and number of studies. The two resulting gene sets differed only with respect to functional heterogeneity and not in the number of studies (p = 0.51; Wilcoxon-Mann-Whitney test). (D) Immune response-related genes show a bimodal specificity value distribution. (E) This bimodality is caused by innate immune response genes showing the tendency to be more cancer-general and all other immune response genes to be more specific (p = 0.001; Wilcoxon-Mann-Whitney test).

Figure 4. Specificity of cancer hallmark classes.

(A) DNA repair genes (n = 8) are significantly (*p < 0.05; **p < 0.01; Wilcoxon-Mann-Whitney test) more specific than genes from the other cancer hallmark classes of oncogenes (n = 25) and tumor suppressors (n = 18). (B) The distribution of cancer-related DNA repair pathway genes differs between tissues. (C) Principal component projection of cancer tissues in DNA repair pathway space. The red arrows show the original axes in the new space.

Figure 5. Interactions between viral and cancer proteins.

(A) Cancer genes targeted by maximally one virus tend to be more specific (***p < 0.001). (B) Cancer proteins of different cancer types tend to interact with proteins from different virus types. Cancer types are grouped by similar tissue origin. Cancer proteins from hematological origin tend to interact with EBV proteins and proteins from cancer types of urogenital origin tend to interact with HPV proteins.

Figure 6. Interactions between lung cancer genes and air pollutants.

(A) Out of the eight specific lung adenocarcinoma proteins, four proteins (blue nodes) have significant associations with environmental chemicals (red nodes). Air pollutants are shown in dark red, and other chemicals, in light red. (B) A high fraction of air pollutants are among the environmental chemicals that interact with specific lung adenocarcinoma genes.