Therapeutic implications of cancer gene amplifications without mRNA overexpression: silence may not be golden

Targeted therapy resistance affects a subset of cancer patients [1, 2]. Indeed, ~ 13% of somatic mutations are not expressed at the RNA level [3]. Little is known regarding gene amplifications. We examined The Cancer Genome Atlas (TCGA) to determine RNA expression of high-level amplifications, including in cancer-related genes. We observed that a minority of high-level amplifications are silenced. Silencing is, unexpectedly, more frequent in focal than non-focal amplifications. However, most focal and non-focal amplifications are over-expressed. Therefore, our observations suggest important points, which require further clinical validation: (1) high-level amplifications can be silenced and therefore may not be amenable to therapeutic targeting: and (2) non-focal amplifications are sometimes not considered druggable; however, they are frequently overexpressed, suggesting that they could be pharmacologically tractable.

Data from 675 TCGA samples (23 tumor types) with differential RNA expression was retrieved (Additional file 1: Methods ; Additional file 2: Table S1). A total of 166,707 amplifications were reviewed; there was an average of 304 [268–340] (mean [95% confidence interval (CI)]) high-level amplifications (≥ 6 copies) per sample. Amplifications were categorized as “non-focal” when ≥ 1 other amplification was observed within a 0.1 megabase (Mb) genomic window. Non-focal amplifications result from large genome rearrangements encompassing > 1 gene. Using the aforementioned threshold, 137,819 (83%) amplifications were considered non-focal; 28,888 (17%) amplifications, focal; respectively, 5612 non-focal (68%) and 2599 focal (32%) amplifications in cancer-related genes (Table 1).

High-level amplifications correlated with a + 548 [+ 500 to + 595] % (mean, 95% CI) increase of expression for the corresponding mRNA in the tumor compared to the adjacent tissue. When considering only cancer-related genes, the overall expression increase was + 1084 [+ 742 to + 1426]% (mean, 95% CI). There was no difference between the focal and the non-focal groups (Table 1).

A subset of 9534 (6%) amplifications were silenced (i.e., presenting a decrease of expression > 80% in the tumor compared to normal adjacent tissue). This proportion was similar when only considering cancer-related genes (N = 485, 6%) (Table 1).

A Chi-square test was performed to examine the relationship between gene silencing and amplification type. Interestingly, gene amplifications were either consistently focal or non-focal. Amongst the 832 amplified cancer-related genes, 243 were always focally amplified and 589 were uniformly non-focally amplified. Focal amplifications were more likely to be silenced (9% vs 5% silenced amplifications; odds-ratio (OR) [95% CI] = 1.83 [1.74–1.91]; X² (1, N = 166,707) = 641.4, p < 0.00001), and this held true when considering only cancer-related genes (10% versus 4% silenced amplifications; OR = 2.48 [2.06–2.98]; X² (1, N = 8211) = 98.2, p < 0.00001) (Table 1). Interestingly, many cancer-related genes that are druggable/activate druggable pathways were less frequently silenced than cancer-related genes without active therapies (Fig. 1).

The mechanism of amplification silencing was not elucidated. Prior studies suggest potential mechanisms such as epigenetic modifications [4], miRNA regulation [5], and/or factors that influence RNA-decay such as RNA-binding proteins [6].

Overall, our findings indicate that high-level amplifications are prevalent in tumors. Further, as expected, gene copy-number amplifications correlated with an overall increase in mRNA expression. Even so, high-level amplifications may be associated with gene-specific silenced RNA expression. Surprisingly, focal amplifications were more likely to be silenced than non-focal amplifications, and this held true when considering only cancer-related genes. Furthermore, most non-focal amplifications were not silenced, suggesting that such amplifications may still be actionable. Of interest, cancer genes that were more likely to be considered druggable and/or have established prognostic or predictive attributes were usually not silenced, while cancer-related genes that are often considered therapeutically intractable were more often silenced. Since silencing of amplifications would nullify tumorigenic impact, and also lead to resistance (if the gene product was the treatment target), it is conceivable that more frequently silenced amplifications would be less likely to be associated with therapeutic or prognostic/predictive impact. These findings echo those previously published in gliomas where the authors suggested that, even when amplified, genes that are normally silent in a given cell type may play no role in tumor progression [7].

In conclusion, our study indicates that the consequences of silencing on response versus resistance after targeted therapies matched to oncogenic amplifications requires in vitro verification and prospective clinical studies. Taken together with the existing literature [3, 8, 9], we suggest that gene silencing may be an important mechanism of therapeutic resistance, and that optimal pharmacologic intervention in cancer may demand transcriptomic in addition to genomic interrogation and considerations for epigenetic modulation.