Mitochondrial transporter expression patterns distinguish tumor from normal tissue and identify cancer subtypes with different survival and metabolism

Abstract

Transporters of the inner mitochondrial membrane are essential to metabolism. We demonstrate that metabolism as represented by expression of genes encoding SLC25 transporters differentiates human cancers. Tumor to normal tissue expression ratios for clear cell renal cell carcinoma, colon adenocarcinoma, lung adenocarcinoma and breast invasive carcinoma were found to be highly significant. Affinity propagation trained on SLC25 gene expression patterns from 19 human cancer types (6825 TCGA samples) and normal tissues (2322 GTEx samples) was used to generate clusters. They differentiate cancers from normal tissues. They also indicate cancer subtypes with survivals distinct from the total patient population of the cancer type. Probing the kidney, colon, lung, and breast cancer clusters, subtype pairs of cancers were identified with distinct prognoses and differing in expression of protein coding genes from among 2080 metabolic enzymes assayed. We demonstrate that SLC25 expression clusters facilitate the identification of the tissue-of-origin, essential to efficacy of most cancer therapies, of CUPs (cancer-unknown-primary) known to have poor prognoses. Different cancer types within a single cluster have similar metabolic patterns and this raises the possibility that such cancers may respond similarly to existing and new anti-cancer therapies.

Figure 1

Mitochondrial SLC25 transporters. (A) Some mitochondrial transporters and their substrates. Transporters and enzymes assayed by RT-qPCR are shown in red (Supplementary Figs. S3–S5). (B) Gene expression of tumor and normal tissue. ‘A3’ denotes the SLC25A3 phosphate transport protein (PTP). ‘SLC25’ is the sum of FKPMs of all SLC25 genes. ‘mDNA’ is the sum of FPKMs of the 13 mitochondrial DNA protein coding genes. ‘Tissue’ refers to the sum of FKPMs of all protein coding genes on nuclear DNA. (C) Ratio of sum of FPKMs of MitoCarta over tissue genes in tumor and normal tissue from KIRC, COAD, LUAD and BRCA. (D) FPKM ratio of SLC25 over tissue genes from KIRC, COAD, LUAD, and BRCA and normal tissue. Error bar standard deviation. Only transporters with smallest coefficient of variation (standard deviation/mean) of transporter FPKM/sum of MitoCarta gene FPKMs from tumor and normal tissue for kidney [(E) KIRC)], colon [(F) COAD)], lung [(G) LUAD], and breast [(H) BRCA] are shown. Those of all SLC25 transporters are shown in Supplementary Fig. S1.

Figure 2

Expression of SLC25 genes in tumor and uninvolved tissue. (A) Heat map of uninvolved tissue FPKM ratios of single SLC25 over the sum of MitoCarta genes (FPKM SLC25Ax/ΣFPKM MitoCarta) from same person. ‘k’ (KIRC), ‘c’ (COAD), ‘l’ (LUAD), and ‘b’ (BRCA) with values shown × 10³. (B) SLC25Ax/ΣFPKM MitoCarta ratios for tumor over those of uninvolved tissue (same individual). The L10 transporters [(C) KIRC], [(D) COAD], [(E) LUAD], [(F) BRCA] and the S10 transporters [(G) KIRC], [(H) COAD], [(I) LUAD], [(J) BRCA]. The percentage of patients with the indicated L10 and S10 of the total number of patients probed is shown in (C–J). The transporters are arranged numerically as indicated in Table 1. Those transporters with sequence similarity are grouped together between the red lines.

Figure 3

SLC25 gene expression in the three major kidney cancers. KIRC (n = 475), KIRP (n = 236), KICH (60), and normal kidney (n = 138). (A–C) Distribution of cancer and normal patients in SLC25 expression clusters. (D) Expression of L26-S24 genes ≥ twofold or (E) less than < twofold larger in KIRC (cluster 24), KIRP (clusters 6), and KICH (cluster 13) cancers (c) compared to normal (n) tissue (cluster 34). Expression is median of the ratio of tumor (Supplementary Fig. S9) and normal (Supplementary Fig. S10) box plot. (F) Pie charts for the principal kidney tumor expression clusters. (G–I) Kaplan–Meier plots. ‘all’ is total population of cancer type. The number of patients is noted in parentheses. Statistical analyses compared survival of patients of a single cluster or defined group of clusters to the total patient cancer type population. α = p < 0.05; β = p < 0.025; γ = p < 0.02; δ = p < 0.01; ε = p < 0.001.

Figure 4

SLC25 expression of cancer types and matched normal tissue. COAD (n = 285); LUAD (n = 503); BRCA (n = 982). (A) Heat maps of gene expression of the indicated clusters of tumor and normal tissue L26-S24 SLC25 transporters. (B–D) Distribution of cancer and normal patients among SLC25 expression clusters. (E–G) Kaplan–Meier plots of patients grouped by SLC25 expression cluster. ‘all’ refers to total number of patients with the cancer type. Number in parenthesis is the number of patients within cluster or group of clusters used. Statistical analysis between survival of patients of SLC25 expression cluster(s) and the total patient population of the cancer type. α = p < 0.05; β = p < 0.025; γ = p < 0.02; δ = p < 0.01; ε = p < 0.001.

Figure 5

Metabolic enzymes with expressions that differentiate worse from better prognosis cancer subtypes. (A,B) Only those enzymes are shown with expression ratio of worse prognosis (red) over better prognosis (blue) of ≥ 2.0 and ≤ 0.5. (C) Only SLC16A14 falls into the range of ratios of (A,B). All others fall into smaller (2.0 ≥ 1.5) and larger (0.7 ≥ 0.5) ratios. (D) Same ratio ranges as (A,B) with left column ≥ 2.0 and right column ≤ 0.5. The left squares of both columns are of cancer subtypes with worse prognoses (clone 9) while those on the right are of cancer subtypes with better prognosis (clone 7). p < 0.05 (*), p < 0.01 (**), p < 0.001 (***), p < 0.0001 (****).

Figure 6

Diagrammatic presentation of the mitochondrial SLC25 transporter algorithm used to identify and characterize human cancer types and subtypes.