Ribosomal DNA copy number amplification and loss in human cancers is linked to tumor genetic context, nucleolus activity, and proliferation

Abstract

Ribosomal RNAs (rRNAs) are transcribed from two multicopy DNA arrays: the 5S ribosomal DNA (rDNA) array residing in a single human autosome and the 45S rDNA array residing in five human autosomes. The arrays are among the most variable segments of the genome, exhibit concerted copy number variation (cCNV), encode essential components of the ribosome, and modulate global gene expression. Here we combined whole genome data from >700 tumors and paired normal tissues to provide a portrait of rDNA variation in human tissues and cancers of diverse mutational signatures, including stomach and lung adenocarcinomas, ovarian cancers, and others of the TCGA panel. We show that cancers undergo coupled 5S rDNA array expansion and 45S rDNA loss that is accompanied by increased estimates of proliferation rate and nucleolar activity. These somatic changes in rDNA CN occur in a background of over 10-fold naturally occurring rDNA CN variation across individuals and cCNV of 5S-45S arrays in some but not all tissues. Analysis of genetic context revealed associations between cancer rDNA CN amplification or loss and the presence of specific somatic alterations, including somatic SNPs and copy number gain/losses in protein coding genes across the cancer genome. For instance, somatic inactivation of the tumor suppressor gene TP53 emerged with a strong association with coupled 5S expansion / 45S loss in several cancers. Our results uncover frequent and contrasting changes in the 5S and 45S rDNA along rapidly proliferating cell lineages with high nucleolar activity. We suggest that 5S rDNA amplification facilitates increased proliferation, nucleolar activity, and ribosomal synthesis in cancer, whereas 45S rDNA loss emerges as a byproduct of transcription-replication conflict in rapidly replicating tumor cells. The observations raise the prospects of using the rDNA arrays as re-emerging targets for the design of novel strategies in cancer therapy.

Fig 1. Estimating rDNA copy number.

(A) For each 150bps window along the rDNA array reference, we calculated the coefficient of variation (CV) of the average depth among samples. The x-axes indicate the starting coordinate for each window on 18S (901–1871) and 28S (whole length). LUAD tumor and adjacent tissues have largely consistent CV across 18S and 28S sequences. The grey regions highlight the windows selected to assess CN in the 18S and 28S components. (B) The differences in copy number among the three 45S components in each sample were calculated as the mean of their absolute pairwise difference. Such differences are significantly smaller when using the selected segments than using the full length of the rDNA shown in (A) for 18S and 28S. (C) Nearly perfect pairwise Pearson correlations between 45S components. BLCA is displayed as an example. Copy number estimates are corrected for batch and ploidy.

Fig 2. Variable manifestation of concerted copy number variation (cCNV) across tissues.

Pearson correlation between the 5S and 45S rDNA arrays is variable among tissues. The correlation is significant and similarly manifested in normal and cancer lineages of LUAD and BLCA.

Fig 3. rDNA CN amplification and loss between cancer tissue and paired adjacent normal tissue.

Yellow and blue denote significant gain or loss in tumors compared with paired adjacent controls (one-sided paired Wilcoxon rank sum test P < 0.05 for OV and < 0.01 for others). There are 74, 32, 22, 24, 35, 38, 9, 10 and 19 informative patients for LUAD, LUSC, HNSC, BLCA, KIRC, STAD, OV, BRCA and LIHC respectively. OV, BRCA and LIHC were used as validation sets.

Fig 4. Association between genetic context and rDNA CN alterations.

(A) Associations between copy number alterations (SCNA) and 5S CN, 45S CN and 5S/45S ratio. (B) Significantly associated rDNA-SCNA pairs (P < 0.05) are preferentially implicated in greater 45S loss and greater 5S rDNA amplification (P < 0.05, binomial test). (C) Association between somatic mutations and 5S CN, 45S CN and 5S/45S ratio. (D) Significantly associated mutation-rDNA pairs (P < 0.05) are almost exclusively implicated in greater 45S loss and greater 5S rDNA amplification (P < 0.001, binomial test) in LUAD. For (B, C) Y-axis show the P-values for the associations between the SCNA or gene mutation event and 45S CN, 5S CN and 5S/45S ratio. rDNA associations were colored according to cancer type (P < 0.05). The up/down direction of triangles indicates that the somatic alteration is associated with increased or decreased CN or 5S/45S ratio. The X-axis shows the fraction of patients with the non-silent gene mutation or focal SCNAs (ploidy cutoff > 2.1 for amplification and < 1.9 for deletion).

Fig 5. Increased 1q42.13 ploidy partially explains increased 5S rDNA CN in cancers.

In each cancer type, all available tumors were included. Spearman’s rank correlation was used.

Fig 6. The 5S rDNA is increased through 1q42 segmental duplications and 5S array expansion.

(A) Most cancers displayed significantly increased 1q42.13 ploidy. (B) Significant 5S CN amplification was still observed in STAD when only considering patients that are closest to diploidy at 1q42.13 (P-value from one-tailed Wilcoxon rank sum test). Only patients with 5S rDNA CN estimated for both tumor and adjacent tissues are shown in B.

Fig 7. Nucleolar gene expression and proliferation in cancer-adjacent tissue pairs.

(A-C) Changes in the level of expression of (A) cytoplasmic ribosomal protein genes (cRPGs), (B) nucleolar genes as well as (C) PRI in tumors relative to their normal adjacent tissue within each individual. Seventeen cancer types with RNA-seq data in ≥ 5 tumor-adjacent pairs were shown, with sample sizes in brackets. Yellow and blue indicate significant up- and down- regulation in tumors compared with paired adjacent control tissue (Wilcoxon rank sum test P < 0.01), respectively. Non-significant changes are shown in grey. (D) PRI is more strongly correlated with the expression of nucleolar genes than with the expression of cRPGs. The upper two panels show the correlations of cRPGs and nucleolar genes with PRI in BLCA as an example, while the lower panel summarizes the Spearman’s correlation coefficients in all 17 cancer types (P from paired Wilcoxon rank sum test). COAD, colon adenocarcinoma; KIRP, kidney renal papillary cell carcinoma; THCA, thyroid carcinoma; READ, rectum adenocarcinoma; KICH, kidney chromophobe; PRAD, prostate adenocarcinoma; CHOL, cholangiocarcinoma; UCEC, uterine corpus endometrial carcinoma; ESCA, esophageal carcinoma. Other abbreviations are as in Table 1.

Fig 8. Associations between tumor proliferation and rDNA copy number.

(A-B) Spearman rank correlations between PRI (YW gene set) and 45S CN, 5S CN and 5S/45S ratio in LUAD tumor samples. (C) Changes in tumor PRI relative to normal adjacent tissue are positively correlated with changes in the 5S/45S ratio between tumor and normal adjacent tissue. The 31 LUAD patients with paired adjacent-tumor data (DNAseq and RNAseq) were used in C.