2 hit hypothesis cancer

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

• View all journals
• My Account Login
• Explore content
• About the journal
• Publish with us
• Sign up for alerts
• Open access
• Published: 03 December 2021

Higher order genetic interactions switch cancer genes from two-hit to one-hit drivers

• Solip Park   ORCID: orcid.org/0000-0002-5988-5339 1 ,
• Fran Supek   ORCID: orcid.org/0000-0002-7811-6711 2 , 3 &
• Ben Lehner   ORCID: orcid.org/0000-0002-8817-1124 3 , 4 , 5  

Nature Communications volume  12 , Article number:  7051 ( 2021 ) Cite this article

10k Accesses

4 Citations

64 Altmetric

Metrics details

• Cancer genetics
• Cancer genomics
• Computational biology and bioinformatics

The classic two-hit model posits that both alleles of a tumor suppressor gene (TSG) must be inactivated to cause cancer. In contrast, for some oncogenes and haploinsufficient TSGs, a single genetic alteration can suffice to increase tumor fitness. Here, by quantifying the interactions between mutations and copy number alterations (CNAs) across 10,000 tumors, we show that many cancer genes actually switch between acting as one-hit or two-hit drivers. Third order genetic interactions identify the causes of some of these switches in dominance and dosage sensitivity as mutations in other genes in the same biological pathway. The correct genetic model for a gene thus depends on the other mutations in a genome, with a second hit in the same gene or an alteration in a different gene in the same pathway sometimes representing alternative evolutionary paths to cancer.

Similar content being viewed by others

Multicenter integrated analysis of noncoding CRISPRi screens

David Yao, Josh Tycko, … Steven K. Reilly

High-throughput evaluation of genetic variants with prime editing sensor libraries

Samuel I. Gould, Alexandra N. Wuest, … Francisco J. Sánchez Rivera

A single-cell atlas enables mapping of homeostatic cellular shifts in the adult human breast

Austin D. Reed, Sara Pensa, … Walid T. Khaled

Introduction

Cancer driver genes are classified as oncogenes (OGs) or tumor suppressor genes (TSGs) depending upon whether their activation or inactivation contributes to cancer. Whereas a single mutation in an OG can be sufficient to increase tumor fitness, inactivation of both copies of a TSG is often required, as envisaged by the classic two-hit hypothesis 1 , 2 . However, >500 genes have now been causally implicated in cancer 3 and exceptions to these models are known 4 , 5 . For example, some OGs are dosage-sensitive, with amplification of a mutated copy further increasing tumor fitness 6 , 7 , 8 and some TSGs are haploinsufficient, with inactivation of a single allele promoting cancer 4 , 9 , 10 , 11 .

Large-scale cancer genome sequencing presents an opportunity to systematically investigate the dosage sensitivity of OGs and the dominance of TSGs and the extent to which these are fixed or variable. In model organisms such as yeast, the activity-fitness functions of genes have been systematically experimentally quantified. For growth rate, activity-fitness functions are typically non-linear, are frequently ‘peaked’ (i.e., non-monotonic with maximal fitness at an intermediate activity level), and can change across conditions 12 . Consistent with this, whether some TSGs behave as haploinsufficient one-hit drivers or recessive two-hit drivers has been reported to vary across cancer types and patients 4 , 13 , 14 , 15 and mutated OGs can be highly amplified in one cancer type but not in others 16 .

Using data from ~10,000 tumors we show here that both OGs and TSGs quite often vary in whether they behave as one-hit or two-hit drivers. These changes in dosage sensitivity and dominance are examples of the interactions between mutations being contingent upon the context. In model organisms, such changes are often caused by higher-order epistasis with, for example, a third mutation modifying the interaction between two alterations 17 , 18 , 19 , 20 . We find that higher-order interactions are also important in human tumors and use third-order genetic interactions to identify mutations that switch cancer genes between behaving as two-hit and one-hit drivers. Taken together, our results suggest that the second hit in one driver and a hit in another driver in the same biological pathway can sometimes have similar consequences and be alternative evolutionary paths to cancer.

Interactions between mutations and CNA in 10,000 tumors

We employed a statistical test based on log-linear regression - a generalization of the chi-square test to more than two dimensions (see Methods; Fig. 1a, b )—to identify interactions between somatic mutations and copy number alterations (CNAs) in 201 cancer driver genes (see Methods) across ~10,000 tumors representing 33 types of cancer characterized as part of the TCGA project 21 , 22 , 23 . We tested for co-occurrence between mutations and either CNA gain (or amplification) or CNA loss for each gene in the cancer types in which it is significantly mutated (>2% mutation frequency in a single cancer type; a mean of 2.3 (median = 1) cancer types per gene, 454 gene-cancer type pairs in total. The 201 genes include 117 TSGs, 77 OGs, and 7 dual-functional genes (DFGs) (genes classed as both TSGs and OGs in different cancer types) (see Methods) 21 .

a Classical one-hit and two-hit models. b Log-linear regression for identifying interactions between mutations and CNAs. c Interaction coefficient (i.e., effect size) and FDR-values in each cancer type for 73 cancer genes tested in at least two cancer types. Cancer types with >150 samples and at least 1 significant interaction are shown (FDR = 10%). The full data set is shown in Supplementary Fig.  3 . Cancer-type abbreviations are listed in Supplementary Table  1 . Source data underlying c are provided as a Source Data file.

In total, interactions between mutations and CNAs were detected for 40 genes (19.9% of the tested genes) in at least one cancer type and for 17.4% of all tested gene-cancer type pairs (false discovery rate, FDR = 10%; Supplementary Figs.  1 and 2a ). The 40 genes had interactions between mutations and CNAs in a mean of 2.2 cancer types (median = 1, range: 1–17, with TP53 having the most interactions). Interactions were detected for 26 TSGs (65.0% of the detected genes), 12 OGs (30.0 %) with a total of 63 interactions between mutations and CNA loss, and 24 interactions between mutations and CNA gain (8 interactions were detected in both the loss and gain models; FDR = 10%, Fig. 1c ; the results for all tested pairs are shown in Supplementary Fig.  3 and Supplementary Data  1 ). Consistent with the two-hit model, 56/63 interactions between mutations and CNA loss (88.9%) were for TSGs and 15/24 interactions between mutations and CNA gain (62.5%) were for OGs. At FDR 20%, interactions between mutations and CNAs were detected for 56 genes (27.9% of the tested genes) in at least one cancer type and for 23.6% of all tested gene-cancer type pairs.

Four classes of driver genes

Clustering suggested the 73 genes tested for interactions between mutation and CNAs in at least two types of cancer fall into four major classes (Fig. 1c ;  Supplementary Fig.  2b ).

First, 23 TSGs, 18 OGs, and 3 other genes showed patterns consistent with them primarily functioning as one-hit drivers with no significant co-occurrence between mutations and CNAs in any cancer types in which they are significantly mutated (FDR = 10%). Examples of class I (‘one-hit’) drivers include the genes SF3B1 , NOTCH1 , and IDH1 .

Second, 15 TSGs, 1 OG, and 2 DFGs only had interactions between mutations and CNA loss in at least one cancer type, consistent with them acting in at least some cancers as two-hit drivers. These class 2 (‘two-hit loss’) drivers include the genes RB1 (3/13 cancer types), NF1 (5/8), NF2 (1/2), PTEN (7/15), and BAP1 (2/6).

Third, 2 TSGs and 6 OGs only had interactions between mutations and CNA gain. These class 3 (‘two-hit gain’) drivers include EGFR (2/3 cancer types) , KRAS (4/16), and BRAF (1/6). Changes in VAFs suggest it is the mutant allele that is normally amplified for class 3 genes (Supplementary Fig.  5 ).

Fourth, a set of 3 cancer genes had interactions between mutations and both CNA loss and gain. These class 4 (‘two-hit loss and gain’) drivers include 1 TSG where mutations interact with CNA gain in one cancer type but with CNA loss in a different type of cancer and 1 TSG and 1 OG where mutations interact with both CNA gain and loss in the same cancer type. For example, mutations in CUL3 interact with CNA loss in one type of cancer (kidney renal papillary cell carcinoma) but with CNA gain in head and neck squamous cell carcinomas and with no interactions detected in one additional cancer type in with the gene is significantly mutated. An additional striking example is TP53 which has interactions between mutations and CNA loss in 16 cancer types, between mutations and CNA gain in 7 cancer types, and between mutations and both CNA loss and gain in 6 cancer types.

We also investigated the alternative possibilities of two-way interactions through (i) promoter DNA hypermethylation (silencing) and somatic mutation or (ii) promoter DNA hypermethylation and CNA loss. Using 31 gene-tissue pairs in which a cancer gene is epigenetically silenced in >1% of samples, only one significant two-way interaction between promoter DNA hypermethylation and CNA loss– ZNF133 in ovarian cancer—was identified (FDR 10%; Supplementary Fig.  7b ; Supplementary Data  2 ).

Changes in the interactions between mutations and CNAs

Although the class 4 drivers are extreme examples of the interactions between mutations and CNAs changing across contexts (cancer types), the data suggest that this is also true for many of the other drivers (Fig. 1c ). For example, whereas mutations in the classic TSG NF1 interact with CNA loss in most cancer types (62.5%), the driver mutations in BRAF only interact with CNA gain in one of the four cancer types in which it is significantly mutated (skin cutaneous melanoma; SKCM) (Fig. 2a , Supplementary Fig.  6 ). To further explore changes in these interactions, we tested whether the strength of interaction between mutations and CNAs differs between detected cancer type (cancer type-specific two-way interaction; FDR = 10%) and other cancers in which the gene is mutated (>2%) using log-linear regression (Fig. 2b ; Supplementary Data  3 ). We tested whether 48 interactions for 26 genes (36 interactions for 18 genes in class 2 and 12 interactions for 8 genes in class 3, FDR = 10% in one cancer type) changed in strength in other significantly mutated cancers together (median = 3 cancer types; mean = 4.5) (Fig. 2c, d and Supplementary Fig.  6 ). This analysis revealed that 41.6% (20/48) of the interactions differ in strength between cancer types (FDR = 10%; 16 interactions from 10 genes in class 2, 4 interactions from 4 genes in class 3). For example, the interaction between BRAF mutation and CNA gain in SKCM is stronger than in the three other types in which BRAF is significantly mutated (FDR = 10%).

a Tested hypothesis. b Test for whether interactions change between cancer type A (detected; FDR 10%) and other cancers in which the gene is significantly mutated (compared). Ga Mut denotes the number of samples with somatic mutation of gene A in cancer type with a significant interaction, Ga CNAs indicates a number of samples with CNAs of gene A in cancer type with a significant interaction, and Tissue denotes the number of samples with genomic alterations of gene A in other significantly mutated cancer types. c Volcano plot comparing differences of the log of the odds ratios for the co-occurrences between mutation and CNA in two cancer types (i.e., detected cancer type and other significantly mutated cancer types together). A total of 48 detected interactions for 26 genes were tested. Color coding is for the cancer type in which the two-way interaction was detected (FDR = 10%). 0.5 was added to each frequency when calculating odds ratios to avoid division by zero frequencies. d Effect sizes (interaction coefficients) and FDR values for tissue-specific interactions were estimated as the ratio between the number of detected interactions in the permutated matrix and the number of detected with the real data with 100 permutations. Source data underlying c and d are provided as a Source Data file.

Third-order interactions switch genes from two-hit to one-hit drivers

Why do the interactions between mutations and CNAs change in different cancer types and even within the same type of cancer? One cause of changes in the pairwise interactions between mutations in model organisms is higher-order genetic interactions (also called higher-order epistasis) 18 , 24 , 25 . In the simplest examples—third-order interactions—the interaction between two mutations changes depending upon whether a third mutation is present or not. Third-order interactions can occur among mutations in the same 19 , 26 , 27 or different 17 , 28 genes and make important contributions to diverse phenotypic traits such as growth and drug resistance 18 , 29 . We hypothesized that conceptually similar higher-order interactions might be occurring in cancer genomes, with mutations in a second gene (between gene interaction) altering the interaction between a mutation and CNA in a cancer gene (within gene interaction) (Fig. 3a ). Specifically, we tested for third-order interactions involving two genetic alterations in one gene (somatic mutation and CNA) and the third alteration in a second gene (somatic mutation). To identify third-order genetic interactions, we only considered somatic mutations in the second gene, not including copy-number changes in the second gene to avoid possible confounding by the overall level of copy-number variation. Since many tumors carry more than two driver mutations 30 , there is plenty of opportunity for higher-order interactions amongst driver mutations.

a A third-order genetic interaction means the strength of a pairwise interaction (between mutation and CNA in the first gene) changes when a second gene is mutated. b Log-linear regression for identifying third-order interactions between mutations and CNAs in one gene and mutations in a second gene. The three-way model quantifies the strength of interactions of two genomic alterations within a single gene with a background alteration of the second gene (gene B) mutation in a cancer type. Ga Mut denotes the number of samples with somatic mutation of gene A, Ga CNAs indicates a number of samples with CNAs of gene A, and Gb Mut denotes the number of samples with somatic mutation of gene B. c – f Effect sizes (interaction coefficients) and two-sided P -values for 17 third-order interactions. Interactions between mutation and CNA in all samples ( c ), samples with gene B mutations ( d ), and samples without gene B mutations ( e ), and third-order interactions ( f ). g Frequencies of gene A CNAs in samples with gene A mutations (top panel) or gene A mutations in samples with gene A CNAs (bottom panel) depending on whether gene B is mutated or not. h Gene pairs involved in third-order interactions share functions and pathways more than random pairs of cancer genes ( P -values from two-sided Mann–Whitney U test). The median value of each gene set is displayed as a band inside each box. The length of each whisker is 1.5 times the interquartile range (shown as the height of each box). Source data underlying c – h are provided as a Source Data file.

We used log-linear regression to test for third-order interactions between mutations and CNAs in one gene (gene A) and mutations in a second gene (gene B, Fig. 3b ). Across cancer types, we were able to test for third-order interactions for 40 genes with 79 pairwise interactions between mutations and CNAs (63 interactions for 30 genes with CNA loss and 24 for 14 genes with CNA gain; 4 genes with both). Each pairwise interaction was tested for interactions with mutations in a mean of 19.1 other driver genes (median = 19) mutated in at least 2% of samples, with a total of 1511 third-order interactions tested (Supplementary Data  4 ).

In total, we identified 17 third-order interactions (FDR = 10%, this likely underestimates the true number of higher-order interactions because of the low statistical power to detect them). To illustrate how the presence of mutations in a second driver gene (gene B) alters the interaction between mutations and CNAs in a first driver (gene A), in Fig. 3c–f , we divide samples according to whether they do (Fig. 3d ) or do not (Fig. 3e ) carry mutations in gene B and then plot the frequency of gene A CNA in samples carrying gene A mutations. Most of the third-order interactions (76.5%, 13/17) are examples where the presence of a mutation in a second driver gene decreases the strength of the interaction between mutations and CNAs in the first driver (Fig. 3c–f ). For example, there is a strong interaction between mutations and CNA loss in KEAP1 in lung squamous cell carcinoma (LUSC) but not in samples that also carry a PTEN mutation. This suggests that mutations in PTEN sensitize lung cells to the effects of reduced KEAP1 activity. Similarly, in Fig. 3g we show how the frequency of gene A mutations in samples carrying gene A CNAs varies depending upon the presence of mutations in gene B. For example, there is a strong third-order interaction between BRAF mutations, BRAF CNA gain, and NRAS mutations in SKCM. Only 2.7% of samples with BRAF CNA gain and NRAS mutations have BRAF mutations whereas 81.9% of samples with BRAF CNA gain without NRAS mutations have BRAF mutations. This is consistent with mutations in both genes activating the same pathway.

Second hits in the same pathway switch genes from two-hit to one-hit drivers

Strikingly, most of the third-order interactions involve two genes from the same canonical cancer signaling pathway 31 . This includes the PI3K pathway (BRCA and UCEC), RTK/RAS pathway (SKCM), Nrf2 pathway (LUSC), TBF-β/SMAD4 pathway (COADREAD), cell growth pathway (LGG and LIHC), and p53 pathway (LUAD and LGG). Genes participating in third-order interactions are indeed enriched for shared functions and pathway membership (Fig. 3h ; Fig. 4 ).

Changes in the strength of interaction between mutations and CNAs in driver genes (‘gene A’) in the absence or presence of mutations in a second cancer gene (‘gene B’). A summary of the pathway in which the function of the gene is also shown.

This suggests a simple principle for why these third-order interactions occur and why cancer genes switch between being one-hit and two-hit drivers: two hits in one gene in a pathway or two hits in two different genes in a pathway can have similar functional consequences and so act as an alternative (partially redundant) evolutionary paths during tumor progression. Using this principle, one additional third-order interaction ( RB1-ASXL2 in BLCA) could be identified when gene pairs were restricted to genes sharing at least one Gene Ontology molecular process or pathway annotation (FDR = 10%; Supplementary Data  5 ).

We have shown here in an analysis of ~10,000 tumors that whether a cancer gene requires only one or two genetic alterations to contribute to cancer typically varies across different types of cancer and across individuals. We have also shown that higher-order genetic interactions are important in human tumors, i.e., that in order to understand the genetics of cancer, not only do the effects of individual mutations and their pairwise interactions need to be considered, but also what happens when three or more alterations are combined. Indeed, we have identified multiple examples where the pairwise interaction between two alterations changes when a third alteration is present in a cancer genome.

Quantifying these third-order interactions allows us to propose a simple principle for tumor evolution (Fig. 5 ): that for some cancer types (e.g., SKCM) tumor evolution can occur via two alternative evolutionary paths—either via a cell obtaining two hits in a single driver gene (e.g., BRAF ) or via a cell obtaining single hits in two different genes in the same pathway (e.g., BRAF and NRAS ). Put another way, a pathway needs two genetic alterations to be (in)activated, but these alterations can either both be in the same gene or they can be in two different genes in the pathway. This mutual exclusivity is observed in multiple cancer pathways and for multiple types of cancer (Fig. 3h ; Fig. 4 ). Indeed, this might reflect a more general principle of genetic architecture that a strong perturbation in one gene can have a similar functional consequence as the combination of two weaker perturbations in two different genes in a pathway.

Once a driver gene is mutated, either a second hit in the same gene or an alteration in a different gene in the same biological pathway can be alternative evolutionary paths to cancer. That is, for some pathways, two hits may be required to (in)activate it but these hits can either be in one gene or in two different genes. The ordering of events may differ from that illustrated here.

In contrast to the changing activity-fitness functions of many drivers, a subset of cancer genes (class 1 drivers) appears to nearly always behave as one-hit drivers, consistent with second alterations conferring either no benefit or actually a fitness cost to a tumor. Consistent with this second possibility, many class 1 TSGs have been previously identified as essential genes (Supplementary Table  2 ), suggesting these TSGs may have ‘peaked’ (non-monotonic) activity-fitness functions whereby reduced activity promotes cancer but complete inactivation is lethal to a cell. Two-hits in these drivers may never (or only very rarely) be compatible with cell viability. This suggests the intriguing translational implication that either re-activating or further inactivating class 1 TSGs may be viable therapeutic strategies to pursue.

Finally, we note that, although the results presented here apply to human cancers, it is likely that the principles about genetic architecture will also apply to other diseases and traits. The demonstration of the importance of higher-order interactions in cancer suggests that they will also contribute to the genetic architecture of other complex diseases. Moreover, it seems likely that additional diseases beyond cancer will also have alternative ‘within’ and ‘between’ locus causes, with disease resulting from the combination of mutations in one gene in some individuals and from the interaction between mutations in two different genes in others. More generally, similar principles may apply during evolution, with multiple mutations in a single gene and the interactions between mutations in different genes in a pathway representing parallel paths to the evolution of new phenotypic traits.

Sample preparation

We used comprehensive molecular datasets collected across 33 cancer types from 11,276 patients by the TCGA project 23 . We only considered samples that had available data across two genomic platforms: somatic mutations and CNAs. To obtain high-quality data set, we discarded samples that have been flagged with quality control issues or during pathology review (merged_sample_quality_annotations.tsv). After applying these filtered, hypermutated samples were also excluded (more variant than third quartile + interquartile range × 1.5). A total of 9175 patients' data were used in this work.

Somatic genomic alteration

Genomic data from TCGA Data were obtained from TCGA data Portal (Pan-Can Atlas). We downloaded version 2.8 of the mutation annotation format (MAF) file provided by the ‘Multi-Center Mutation Calling in Multiple Cancers’ (MC3) project, as a part of the TCGA Pan-Cancer Atlas effort 22 . These unified MC3 somatic mutations were called from seven software packages using a single-standardized protocol across many different individual studies (mc3.v0.2.8.PUBLIC.maf.gz). It provides high-quality variant calls after applying rigorous filtering steps to discard low-quality variants and remove possible sequencing artifacts. Somatic mutation calls were assigned to all premature truncation mutations (encompassing splicing variants, frameshift indels, and nonsense variants) and to non-synonymous (missense mutation), single-residue substitutions (in-frame indels). Predicted deleterious missense was assigned at least one of two tools (SIFT and PolyPhen2) predicted as deleterious/damaging variant 32 , 33 . Genomic regions with significant levels of CN arrangements and their target genes of these somatic CNAs were determined using GISTIC 2.0 with their q -values 34 . Gene-level CN data were obtained from Synapse (syn5049520). High-level deletion for a gene was defined as GISTIC threshold CN value of −2, whereas high-level amplification for a gene was assigned with threshold CN value +2. Broad-level deletion (loss) and broad-level amplification (gain) were estimated by GISTIC threshold CN values having less than −1 or greater than +1, respectively.

Somatic driver set

To characterize interactions between mutations and CNAs, we compiled a high-confidence list of 201 somatic cancer driver genes. First, 235 cancer genes were collected by the union of genes predicted by the eight driver gene predictors (20/20+, ActiveDriver, CompositeDriver, MuSiC, MutSig2CV, OncodriveCLUST, OncodriveFML, and e-Driver), manually reference search, and individual TCGA studies from a Pan-Cancer Atlas 21 . The entire process is described in more detail in 21 . Next, cancer genes were selected if they were: (1) listed as ‘oncogene’ or ‘tumor suppressor’ at least one single cancer type (not in ‘PANCAN’) and (2) mutation frequencies in a certain cancer type >2%.

Collected somatic drivers were categorized either as TSG or OG according to 20/20+ predictors across each cancer types 35 . DFGs are genes classified as both TSGs and OGs in different cancer types. This method is based on an improved version of the 20/20 rule (gene is considered to be a TSG when a gene has >20% truncating mutations, whereas OG, will be defined with >20% missense mutations) 36 using Random Forest machine learning algorithm for classifying TSG and OG from somatic mutations. It applies five different features: capturing mutational clustering, evolutionary conservation, predicted functional impact of variants, mutation consequence types, gene interaction network connectivity, and other relevant covariates. In addition, genes in the cancer types in which it is significantly mutated (>2% mutation frequency in a single cancer type) were only considered. Finally, 281 TSGs-single tissue pairs and 173 OGs-single tissue pairs were assigned from 201 genes (55 genes were not included they were not assigned any category either TSG or OG). TSGs and their cancer types were defined when they were annotated as ‘tsg’ or ‘possible tsg’, whereas OGs and their types were assigned with the annotation as ‘oncogene’ or ‘possible oncogene’.

Alternative epi/genomic alterations

Multiple driver mutations (MMs) in cancer genes were obtained from Saito et al. using DNVChecker in five cancer cohorts, of which 9230 TCGA samples 37 . It is designed to detect multiple single-nucleotide variants observed in the same codon and also check their allelic frequencies with the corresponding BAM files to define cis or trans-MMs.

Epigenetic silencing (promoter DNA hypermethylation) events of cancer genes in TCGA samples were obtained from Saghafinia et al. using the RESET method 38 . After collecting only probes mapping to a gene promoter region, high DNA methylation probes were defined with mean β -values (0, minimal level of DNA methylation; 1, the maximal level of DNA methylation) higher than 0.8 and standard deviation lower than 0.005 in normal samples. Next, the association between DNA hypermethylation (probe p ) and a significant decrease of mRNA expression (gene g that corresponds to p ) was analyzed to evaluate the effect of aberrant DNA methylation states. After 100 randomizations to test the significance, a hypermethylation call with FDR < 0.1 was selected. For running a log-linear regression model for two-way interaction between hypermethylation and CNA loss (or mutation), we converted hypermethylation events from probes to genes when >50% of corresponding probes have hypermethylation events.

Allelic imbalance (AI) was determined by testing for a change in variant allele frequencies (VAFs) in the tumor sample compared to in the matched non-tumor sample (mainly from blood) for each patient from our previous study 39 . To define AI in each sample across genes, we first collected all germline variants in the coding and noncoding regions within each gene (expanding the analyzed region bidirectionally to 100 kb for genes shorter than this size to reduce the gene length bias). Next, a two-tailed Fisher’s test that compares the sequencing read counts collecting variants and the reference alleles in the tumor were performed with each collected variant in a gene. The P -values from all collected variants in a gene were then pooled by Fisher’s method for combing P -values. Finally, the AI event in the gene was assigned if the pooled P -value was ≤0.05.

Finally, from 201 tested cancer genes, 64 genes with MMs in their canonical cancer types, 19 epigenetically silenced genes (>1% in their canonical cancer types; 30 pairs), and 159 genes with AI events, were analyzed. In total, 32 interactions between mutation and AI were detected (FDR 10%), including significantly overlapping 2-way interactions from CNA loss model (20 interactions; 62.5%, odds ratio = 10.23, P  = 2.2E−16) (Supplementary Fig.  7a and Supplementary Data  6 ). From 64 cancer genes with MM events in TCGA, three genes ( APC , PTEN , and PIK3CA ) presented very high MMs frequencies in Colon/Rectum adenocarcinoma (COADREAD) and Uterine Corpus Endometrial Carcinoma (UCEC) that is assigned to the one-hit driver from CNA loss model (Supplementary Fig.  7c ). In particular, APC in COADREAD has 38.1% of MMs compared to 67.9% of mutated samples (100% of MMs were in cis, i.e., multiple mutations to the same alleles), PTEN in UCEC has 22.3% of MMs compared to 57.6% of mutated samples (majority of MMs, 68.8% were in trans, i.e., multiple mutations to the different alleles), and PIK3CA in UCEC has 6.1% of MMs from 43.7% of mutated samples (majority of MMs, 61.5% were in cis). Although we tested a diverse range of alternative possibilities that could contribute to two-way interactions, most one-hit drivers from the mutation-CNAs model still did not have alternative two-way interactions beyond the co-occurrences between mutations and CNAs.

Statistical evaluation of mutation—CNAs association

To determine the significance of the co-occurrence of a pair of somatic mutation (all kinds of non-synonymous) and CNAs within a gene across cancer types, we generated a model using a log-linear regression using the MASS package (version 7.3.53.1) in R 40 . Two separate models depending on CNAs (either gain or loss) have been performed in each individual gene-tissue pair as follows:

where: mut  = number of samples with somatic mutation of gene A; CNA-gain (or loss) = number of samples with CNA gain or amplified (or loss) of gene A; mut :CNA-gain (or loss) = number of samples with both mutation and CNA gain or amplified (or loss) of gene A. Therefore, two-way interactions between mutation and CNA gain (or loss) were measured by comparing between CNA wild-type (no copy-number changes) and CNA loss (only one-copy loss) or CNA gain (one-copy gain or more than two-copy gain) samples. The regression coefficient and P -value were computed for individual gene-cancer type pairs and derived from the mut:CNA gain (or loss) using the summary function in R. More negative values from CNA loss model refers to stronger co-occurrence between mutation and CNA loss within a gene-tissue pair, whereas more positive value from the CNA gain model represents stronger co-occurrence between mutation and CNA gain within a gene-tissue pair. To determine the significance of the co-occurrence of a pair of two genomic events in the same gene, we applied a permutation strategy 41 that controls for the heterogeneity in genomic alterations within and across samples. Using the permatswap function in the R package vegan (version 2.5.7) ( http://vegan.r-forge.r-project.org ), we produced permuted genomic alteration matrices that maintain the total number of genomic alterations for each alteration across samples as well as the total number of alterations per sample. Somatic mutation, CNAs loss, and gain events were considered as separate classes and the permutation was performed for each cancer type separately. With 100 permutations, FDR is estimated as the ratio between the number of detected interactions in the permuted matrix (i.e., false interaction) and the number detected with the real data (i.e., true interaction) for each P -value cut-off (Supplementary Fig.  1 ).

The three-way model quantifies the strength of interactions of three genomic alterations from two genes with a background alteration of the second gene (gene B) mutation in a cancer type in a similar manner as the two-way interaction model within a gene (gene A). High-order three-way interactions were identified using the following models:

where: mut  = number of samples with somatic mutation of gene A; CNA-gain (or loss)  = number of samples with CNA gain or amplified (or loss) of gene A; target  = number of samples with somatic mutation of gene B; CNAs-gain (or loss):Target  = number of samples with both somatic mutation of gene B and CNAs of gene A; mut:Target  = number of samples with both somatic mutation of gene A and gene B; mut: gene A CNAs:Target  = number of samples with somatic mutation of gene B when samples with both mutation and CNAs of gene A. 1 was added to each frequency when running the regression model to avoid division by zero frequencies. From the equation, high-order three-way interaction regression coefficient and P -values are derived from the mut:CNA gain (or loss):Target with summary statistics in R.

To check the effect of different mutation types on the two-way interactions between mutation and CNAs, we tested three additional two-way models across (i) only premature truncation variants (PTVs), (ii) only predicted deleterious missense mutations (DelMis), and (iii) only non-deleterious missense mutations (ND_Mis). While several mutation type-specific 2-way interactions were identified (FDR 10%; four interactions from PTVs–CNA loss and two interactions from DelMis–CNA loss), the majority of two-way interactions across different mutations types were also detected when testing for interactions with all types of somatic mutations (original design). In detail, 90.7% of PTVs, 93.3% of DelMis, and 100% of NonMis were overlapped with all types of somatic mutations (FDR 10%) (Supplementary Fig.  8 ; Supplementary Data  7 ).

Next, we restricted our analysis to functional non-synonymous mutations and PTVs after removing putative neutral non-synonymous mutations. We first survey the frequencies of functional non-synonymous mutations (Func-NSY) versus putative neutral non-synonymous mutation (Neutral-NSY) by following the definition of Mina et al. 42 : Func-NSY when recurrently detected at the same amino acid position (i.e., hotspot mutations) or having evidence of their functional role and Neutral-NSY is considered all other non-synonymous mutations. We found that median frequencies for Func-NSY in our collected non-synonymous mutations are 66.7% across cancer types (median for TSGs = 42.6% and for OGs = 87.5%; Supplementary Fig.  9a ). Next, we repeated a statistical test for two-way interactions between mutations and CNAs for all somatic mutations except for Neutral-NSY (that is, all PTVs + only Func-NSY) to evaluate the robustness of 2-way interactions when including putative Neutral-NSY or not. Overall, the Func-NSY only analysis presented very similar effect sizes to the original model (Pearson correlation coefficient (PCC) = 0.85 between coefficient values in an additional model and the original model in CNAs loss; PCC = 0.82 in CNAs gain). Also, the interactions identified using this Func-NSY only definition of somatic mutations strongly overlap with those identified in the original model at FDR 10%: 94.6% (53 over 56 interactions) in the CNA loss model and 85.7% (12 over 14 interactions) in the CNA gain model (Supplementary Fig.  9b ). Furthermore, detected three-way interactions (FDR 10%) in the Func-NSY only model also strongly overlap with the original model: 55.5% (5 out of 9) overlapping in the CNA loss model and 60.0% (6 out of 10) in the CNA gain model (Supplementary Fig.  9c ).

Functional annotation

Functional similarity between two genes with third-order interactions was tested for sharing functional annotation of GO biological process terms, molecular function terms were collected from DAVID 6.8 ( https://david.ncifcrf.gov/ ) 43 and Reactome biological pathway ( https://reactome.org/ ) 44 .

Essential genes from CRISPR–Cas9 and shRNA screening

The 2134 common essential genes identified in a CRISPR–Cas9 screen, which show strong dependencies in >90% of pan-cancer cell lines, were downloaded from DepMap ( https://depmap.org/portal/download/ ). Next, 297 conserved essential genes across three cancer types with 72 cell lines were collected from Marcotte et al. 45 .

Software for statistical analyses

Regression models were performed in the R statistical programming environment (v.3.6.2). Libraries that are required include vegan v.2.5.7, MASS v.7.3.53.1, stringr v.1.4.0, reshape2 v.1.4.4, data.table v.1.14.0, dplyr v.1.0.5, viridis v.0.5.1. Figures were generated using ggplot2 v.3.3.3 and pheatmap v.1.0.12. All statistical analyses were carried out using Python 2.7(packages Stats v.1.2.1 and NumPy v.1.16.5).

Reporting summary

Further information on research design is available in the  Nature Research Reporting Summary linked to this article.

Data availability

This study re-analyzed published data sets, including tumor data sets from TCGA Pan-Cancer Atlas. The TCGA somatic mutation data (mc3.v0.2.8.PUBLIC.maf.gz) was downloaded from https://gdc.cancer.gov/about-data/publications/pancanatlas and copy-number alteration data were downloaded from Synapse ( syn5049520 ). Gene Ontology (GO molecular function and biological process) was downloaded from the DAVID 6.8 ( https://david.ncifcrf.gov/ ) and list of Reactome pathways was downloaded from the https://reactome.org/ . A list of the somatic driver genes was compiled in Supplementary Table  1 from Bailey et al. ( https://doi.org/10.1016/j.cell.2018.02.060 ). Epigenetic silencing data can be obtained by contacting the corresponding author of the original publication ( https://doi.org/10.1016/j.celrep.2018.09.082 ). Multiple driver mutations (MMs) in cancer genes were obtained in Supplementary Table  3 from Saito et al. ( https://doi.org/10.1038/s41586-020-2175-2 ), and allelic imbalance data can be obtained by contacting the corresponding author of the original publication ( https://doi.org/10.1038/s41467-018-04900-7 ). A list of functional non-synonymous mutations was collected in Supplementary Table  1 from Mina et al. ( https://doi.org/10.1038/s41588-020-0703-5 ).  Source data are provided with this paper.

Code availability

Source code for log-linear regression test is available at https://github.com/SolipParkLab/CancerFitness .

Knudson, A. G. Jr. Mutation and cancer: statistical study of retinoblastoma. Proc. Natl Acad. Sci. USA 68 , 820–823 (1971).

Article   ADS   PubMed   PubMed Central   Google Scholar  

Kern, S. E. Whose hypothesis? Ciphering, sectorials, D lesions, freckles and the operation of Stigler’s Law. Cancer Biol. Ther. 1 , 571–581 (2002).

Article   PubMed   Google Scholar  

Sondka, Z. et al. The COSMIC Cancer Gene Census: describing genetic dysfunction across all human cancers. Nat. Rev. Cancer 18 , 696–705 (2018).

Article   CAS   PubMed   PubMed Central   Google Scholar  

Berger, A. H., Knudson, A. G. & Pandolfi, P. P. A continuum model for tumour suppression. Nature 476 , 163–169 (2011).

Article   ADS   CAS   PubMed   PubMed Central   Google Scholar  

Sherr, C. J. Principles of tumor suppression. Cell 116 , 235–246 (2004).

Article   CAS   PubMed   Google Scholar  

Nikolaev, S. et al. Extrachromosomal driver mutations in glioblastoma and low-grade glioma. Nat. Commun. 5 , 5690 (2014).

Article   ADS   CAS   PubMed   Google Scholar  

Takano, T. et al. Epidermal growth factor receptor gene mutations and increased copy numbers predict gefitinib sensitivity in patients with recurrent non-small-cell lung cancer. J. Clin. Oncol. 23 , 6829–6837 (2005).

Ding, L. et al. Somatic mutations affect key pathways in lung adenocarcinoma. Nature 455 , 1069–1075 (2008).

Davoli, T. et al. Cumulative haploinsufficiency and triplosensitivity drive aneuploidy patterns and shape the cancer genome. Cell 155 , 948–962 (2013).

Lindeboom, R. G., Supek, F. & Lehner, B. The rules and impact of nonsense-mediated mRNA decay in human cancers. Nat. Genet 48 , 1112–1118 (2016).

Solimini, N. L. et al. Recurrent hemizygous deletions in cancers may optimize proliferative potential. Science 337 , 104–109 (2012).

Keren, L. et al. Massively parallel interrogation of the effects of gene expression levels on fitness. Cell 166 , 1282–1294 e18 (2016).

Alimonti, A. et al. Subtle variations in Pten dose determine cancer susceptibility. Nat. Genet. 42 , 454–458 (2010).

Varley, J. M., Evans, D. G. & Birch, J. M. Li-Fraumeni syndrome-a molecular and clinical review. Br. J. Cancer 76 , 1–14 (1997).

Goss, K. H. et al. Enhanced tumor formation in mice heterozygous for Blm mutation. Science 297 , 2051–2053 (2002).

Article   ADS   PubMed   Google Scholar  

Bielski, C. M. et al. Widespread selection for oncogenic mutant allele imbalance in cancer. Cancer Cell 34 , 852–862 e4 (2018).

Kuzmin, E. et al. Systematic analysis of complex genetic interactions. Science 360 https://www.science.org/doi/10.1126/science.aao1729 (2018).

Taylor, M. B. & Ehrenreich, I. M. Higher-order genetic interactions and their contribution to complex traits. Trends Genet. 31 , 34–40 (2015).

Domingo, J., Diss, G. & Lehner, B. Pairwise and higher-order genetic interactions during the evolution of a tRNA. Nature 558 , 117–121 (2018).

Mullis, M. N., Matsui, T., Schell, R., Foree, R. & Ehrenreich, I. M. The complex underpinnings of genetic background effects. Nat. Commun. 9 , 3548 (2018).

Bailey, M. H. et al. Comprehensive characterization of cancer driver genes and mutations. Cell 173 , 371–385 e18 (2018).

Ellrott, K. et al. Scalable open science approach for mutation calling of tumor exomes using multiple genomic pipelines. Cell Syst. 6 , 271–281 e7 (2018).

Ding, L. et al. Perspective on oncogenic processes at the end of the beginning of cancer genomics. Cell 173 , 305–320 e10 (2018).

Domingo, J., Baeza-Centurion, P. & Lehner, B. The causes and consequences of genetic interactions (epistasis). Annu Rev. Genomics Hum. Genet. 20 , 433–460 (2019).

Costanzo, M. et al. Global genetic networks and the genotype-to-phenotype relationship. Cell 177 , 85–100 (2019).

Palmer, A. C. et al. Delayed commitment to evolutionary fate in antibiotic resistance fitness landscapes. Nat. Commun. 6 , 7385 (2015).

Poelwijk, F. J., Socolich, M. & Ranganathan, R. Learning the pattern of epistasis linking genotype and phenotype in a protein. Nat. Commun. 10 , 4213 (2019).

New, A. M. & Lehner, B. Harmonious genetic combinations rewire regulatory networks and flip gene essentiality. Nat. Commun. 10 , 3657 (2019).

Lozovsky, E. R., Daniels, R. F., Heffernan, G. D., Jacobus, D. P. & Hartl, D. L. Relevance of higher-order epistasis in drug resistance. Mol. Biol. Evol. 38 , 142–151 (2021).

Martincorena, I. et al. Universal patterns of selection in cancer and somatic tissues. Cell 171 , 1029–1041 e21 (2017).

Sanchez-Vega, F. et al. Oncogenic signaling pathways in the cancer genome atlas. Cell 173 , 321–337 e10 (2018).

Ng, P. C. & Henikoff, S. Accounting for human polymorphisms predicted to affect protein function. Genome Res. 12 , 436–446 (2002).

Adzhubei, I., Jordan, D.M. & Sunyaev, S.R. Predicting functional effect of human missense mutations using PolyPhen-2. Curr. Protoc. Hum. Genet. Chapter 7 , Unit 7 20 (2013).

Mermel, C. H. et al. GISTIC2.0 facilitates sensitive and confident localization of the targets of focal somatic copy-number alteration in human cancers. Genome Biol. 12 , R41 (2011).

Article   PubMed   PubMed Central   Google Scholar  

Tokheim, C. J., Papadopoulos, N., Kinzler, K. W., Vogelstein, B. & Karchin, R. Evaluating the evaluation of cancer driver genes. Proc. Natl Acad. Sci. USA 113 , 14330–14335 (2016).

Vogelstein, B. et al. Cancer genome landscapes. Science 339 , 1546–1558 (2013).

Saito, Y. et al. Landscape and function of multiple mutations within individual oncogenes. Nature 582 , 95–99 (2020).

Saghafinia, S., Mina, M., Riggi, N., Hanahan, D. & Ciriello, G. Pan-cancer landscape of aberrant DNA methylation across human tumors. Cell Rep. 25 , 1066–1080 e8 (2018).

Park, S., Supek, F. & Lehner, B. Systematic discovery of germline cancer predisposition genes through the identification of somatic second hits. Nat. Commun. 9 , 2601 (2018).

Ripley, B.e.a. Package ‘mass’. Cran R (2013).

Park, S. & Lehner, B. Cancer type-dependent genetic interactions between cancer driver alterations indicate plasticity of epistasis across cell types. Mol. Syst. Biol. 11 , 824 (2015).

Mina, M., Iyer, A., Tavernari, D., Raynaud, F. & Ciriello, G. Discovering functional evolutionary dependencies in human cancers. Nat. Genet. 52 , 1198–1207 (2020).

Huang da, W., Sherman, B. T. & Lempicki, R. A. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat. Protoc. 4 , 44–57 (2009).

Jassal, B. et al. The reactome pathway knowledgebase. Nucleic Acids Res. 48 , D498–D503 (2020).

CAS   PubMed   Google Scholar  

Marcotte, R. et al. Essential gene profiles in breast, pancreatic, and ovarian cancer cells. Cancer Discov. 2 , 172–189 (2012).

Download references

Acknowledgements

We thank Luis Garcia-Jimeno for assistance with permutation. S.P. is supported by the Agencia Estatal de Investigación, Ministerio de Ciencia e Innovación (MCIN/AEI/10.13039/501100011033) through the RETOS project PID2019-109571RA-I00. This work was funded by the European Research Council (ERC) Starting grant (HYPER-INSIGHT, 757700) to F.S. and ERC Consolidator (IR-DC, 616434) and Advanced (MUTANOMICS, 883742) grants to B.L. F.S. and B.L. are funded by the ICREA Research Professor program. S.P., F.S., and B.L. acknowledge the support of the Severo Ochoa Centres of Excellence program to the CNIO, IRB Barcelona, and to the CRG (MCIN/AEI/10.13039/50110001103), respectively. B.L. and F.S. Work is funded with the grants BFU2017-89488-P and RegioMut BFU2017-89833-P (MCIN/AEI /10.13039/501100011033/FEDER “A way to make Europe”), respectively. B.L. is further supported by the Bettencourt Schueller Foundation, the Agencia de Gestio d’Ajuts Universitaris i de Recerca (2017 SGR 1322), and the Centres de Recerca de Catalunya (CERCA) program/Generalitat de Catalunya. B.L. also acknowledges the support of the Spanish Ministry of Economy, Industry, and Competitiveness to the European Molecular Biology Laboratory (EMBL) partnership. The results shown here are in whole or part based upon data generated by the TCGA Research Network.

Author information

Authors and affiliations.

Centro Nacional de Investigaciones Oncológicas (CNIO), Madrid, Spain

Institute for Research in Biomedicine (IRB Barcelona), The Barcelona Institute of Science and Technology (BIST), Barcelona, Spain

Catalan Institution for Research and Advanced Studies (ICREA), Barcelona, Spain

Fran Supek & Ben Lehner

Centre for Genomic Regulation (CRG), The Barcelona Institute of Science and Technology (BIST), Barcelona, Spain

Universitat Pompeu Fabra (UPF), Barcelona, Spain

You can also search for this author in PubMed   Google Scholar

Contributions

S.P. performed all analyses. F.S. designed methods for testing interactions between mutation and CNAs. S.P., F.S., and B.L. designed analyses, evaluated the results and wrote the paper.

Corresponding authors

Correspondence to Solip Park , Fran Supek or Ben Lehner .

Ethics declarations

Competing interests.

The authors declare no competing interests.

Additional information

Peer review information Nature Communications thanks Elena Kuzmin and the other anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Peer review file, supplementary dataset 1, supplementary dataset 2, supplementary dataset 4, supplementary dataset 3, supplementary dataset 5, supplementary dataset 6, supplementary dataset 7, reporting summary, source data, rights and permissions.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Cite this article.

Park, S., Supek, F. & Lehner, B. Higher order genetic interactions switch cancer genes from two-hit to one-hit drivers. Nat Commun 12 , 7051 (2021). https://doi.org/10.1038/s41467-021-27242-3

Download citation

Received : 17 May 2021

Accepted : 09 November 2021

Published : 03 December 2021

DOI : https://doi.org/10.1038/s41467-021-27242-3

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

This article is cited by

Co-amplification of cbx3 with egfr or rac1 in human cancers corroborated by a conserved genetic interaction among the genes.

• Giuseppe Bosso
• Francesca Cipressa
• Giovanni Cenci

Cell Death Discovery (2023)

Nonsense-mediated RNA decay: an emerging modulator of malignancy

• Dwayne G. Stupack
• Miles F. Wilkinson

Nature Reviews Cancer (2022)

In silico validation of RNA-Seq results can identify gene fusions with oncogenic potential in glioblastoma

• Ainhoa Hernandez
• Ana Maria Muñoz-Mármol
• Carmen Balana

Scientific Reports (2022)

By submitting a comment you agree to abide by our Terms and Community Guidelines . If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.

Quick links

• Explore articles by subject
• Guide to authors
• Editorial policies

Sign up for the Nature Briefing: Cancer newsletter — what matters in cancer research, free to your inbox weekly.

• November 1999

'Two-Hit' Hypothesis

Much of what scientists know about the origins of cancer and the role of tumor suppressors can be traced back 28 years to the elegant theory of cancer researcher alfred g. knudson. widely thought to be one of the most significant theories in modern biology, knudson's "two-hit" hypothesis was recognized nov. 19 at the john scott awards in philadelphia, along with the revolutionary research of benoit mandelbrot, the discoverer of the powerful mathematical laws governing fractal geometry and self-s.

Much of what scientists know about the origins of cancer and the role of tumor suppressors can be traced back 28 years to the elegant theory of cancer researcher Alfred G. Knudson . Widely thought to be one of the most significant theories in modern biology, Knudson's "two-hit" hypothesis was recognized Nov. 19 at the John Scott Awards in Philadelphia, along with the revolutionary research of Benoit Mandelbrot , the discoverer of the powerful mathematical laws governing fractal geometry and self-similarity. 1

Even amid ongoing rumors...

Interested in reading more?

Become a member of.

• public health

Advertisement

• Next Article

Introduction

Clinical management of retinoblastoma, the first knockout mouse models of retinoblastoma, the role of the rb family in retinal development, distinct roles for p107 and p130 in murine retinoblastoma, species-specific differences in retinoblastoma susceptibility, secondary genetic lesions in retinoblastoma and targeted chemotherapy, acknowledgments, retinoblastoma: from the two-hit hypothesis to targeted chemotherapy.

• Split-Screen
• Article contents
• Figures & tables
• Supplementary Data
• Peer Review
• Open the PDF for in another window
• Get Permissions
• Cite Icon Cite
• Search Site
• Version of Record August 15 2007

David MacPherson , Michael A. Dyer; Retinoblastoma: From the Two-Hit Hypothesis to Targeted Chemotherapy. Cancer Res 15 August 2007; 67 (16): 7547–7550. https://doi.org/10.1158/0008-5472.CAN-07-0276

Download citation file:

• Ris (Zotero)
• Reference Manager

Studies on retinoblastoma have been at the heart of many of the landmark discoveries in cancer genetics over the past 35 years. However, these advances in the laboratory have had little effect on the treatment of children with retinoblastoma. One of the reasons for this has been the lack of preclinical models that recapitulated the genetic and histopathologic features of human retinoblastoma. In the past three years, a series of new animal models of retinoblastoma has been developed and characterized from several different laboratories using a variety of experimental approaches. It is encouraging that there is broad agreement about the consequences of inactivation of the Rb family in retinal development from these studies. More importantly, these new mouse models of retinoblastoma have contributed to clinical trials and novel therapeutic approaches for treating this debilitating childhood cancer. [Cancer Res 2007;67(16):7547–50]

In 1971 Knudson proposed that retinoblastoma was initiated by inactivation of a putative tumor suppressor gene ( 1 ), and this hypothesis was subsequently confirmed by demonstration of loss of heterozygosity at 13q14 in retinoblastomas ( 2 ) and the cloning of the first tumor suppressor gene RB1 ( 3 ). A few years later, Harbour extended these findings to small cell lung cancer showing that the RB1 locus was disrupted in tumors other than retinoblastoma and osteosarcoma (reviewed in ref. 4 ). Since then, it has been found that most, if not all, tumors have defects in their Rb pathway through genetic lesions in the RB1 gene itself or other genes in the pathway. The history of retinoblastoma research highlights how basic research on a rare childhood cancer can have a much broader effect on a disease that affects millions of people each year worldwide.

Although we have learned a great deal about the Rb pathway and its contribution to cancer over the past 35 years, our knowledge should be better applied to treating retinoblastoma. Despite advances in the clinical management of retinoblastoma in the United States, there are still many children with advanced bilateral retinoblastoma that lose one or both of their eyes. In developing countries, 40% to 50% of children still lose their life to metastatic retinoblastoma.

Retinoblastoma is virtually always fatal if left untreated due to tumor metastasis. The objective of retinoblastoma treatment is to save vision without risking the child's life. Children with unilateral retinoblastoma typically undergo surgical enucleation, whereas children with bilateral retinoblastoma often receive a combination of chemotherapy and focal therapy ( 5 ). Today the focal therapies that are most widely used include laser therapy, cryotherapy, and brachytherapy ( 5 ). Every effort is made to avoid or delay radiation therapy because of the side effects associated with irradiating the surrounding tissue. In particular, children with RB1 mutations are prone to secondary cancers later in life, and DNA damage induced by ionizing radiation accelerates this process.

One of the limitations with current clinical management of retinoblastoma is that broad spectrum chemotherapy is given systemically. This leads to significant dose-limiting side effects that can be costly and difficult to manage. Current efforts are focused on local delivery of chemotherapy, and early studies using subconjunctival carboplatin have proved effective in the clinic ( 6 ). Ideally, local delivery of targeted chemotherapy may provide improved tumor response with minimal toxicity. The recent development of animal models and identification of molecular targets for chemotherapy have accelerated progress in this area.

One of the reasons that research on retinoblastoma and Rb pathway has not had a greater effect on retinoblastoma treatment has been the lack of preclinical models that faithfully recapitulate genetic and histopathologic features of childhood retinoblastoma. In 1992, three groups published papers describing the phenotype of mice with a targeted deletion in their Rb1 gene (reviewed in ref. 4 ). Unlike children who inherit a defective copy of the RB1 gene, heterozygous mice failed to develop retinoblastoma. Chimeric mouse studies carried out by Berns and colleagues several years later showed that p107 suppressed retinoblastoma formation in mice (reviewed in ref. 4 ). In 2004, three research groups generated the first knockout mouse models of retinoblastoma by conditionally inactivating Rb Lox in the developing retinae of p107 -deficient mice using Nestin-Cre, Pax6-Cre , and Chx10-Cre ( 7 – 9 ). Inactivation of Rb and p130 also led to retinoblastoma ( 7 ) showing for the first time that p130 can suppress retinoblastoma in mice.

Interestingly, a previous attempt to generate a knockout mouse model of retinoblastoma in IRBP-Cre;Rb Lox/Lox ;p107 −/− mice was unsuccessful ( 10 ). The fundamental difference between IRBP-Cre and the other Cre transgenic lines mentioned above is the timing of Cre expression during retinal development. IRBP-Cre is expressed in cells committed to become photoreceptors just as they exit the cell cycle. In contrast, Nestin-Cre, Pax6-Cre , and Chx10-Cre are all expressed in retinal progenitor cells during development, suggesting that a retinal progenitor cell or a newly postmitotic cell distinct from the photoreceptor lineage is the cell of origin for retinoblastoma (discussed in ref. 11 ). Additional evidence for the identification of retinal progenitor cell as the cell of origin for retinoblastoma is the expression of retinal progenitor cell specific markers in the tumors ( 12 ) and the strong bias toward deregulated proliferation of Rb;p107 -deficient retinal progenitor cells compared with the newly postmitotic cells ( 12 ). These results highlight the importance of combining studies on pediatric cancer with studies on the normal development of that tissue.

M. A. D., unpublished.

In both Pax6-Cre;Rb Lox/Lox and Chx10-Cre Lox/Lox retinae, normal rods fail to form and cells in the outer nuclear layer (ONL) express retinal progenitor cell markers and exhibit morphologic features of retinal progenitor cells in electron micrographs ( 12, 13 ). 4 Photoreceptors are exceedingly sensitive to changes in their local microenvironment such that the cell autonomous requirement for Rb in rod maturation leads to noncell autonomous ONL degeneration in the developing retina, and this varies depending on the extent of Rb inactivation in the ONL in a particular retina. Indeed, there is some increased cell death during early rod development and several days to weeks after the normal period of rod maturation, degeneration in the ONL is observed in Nestin-Cre;Rb Lox/Lox , Pax6-Cre;Rb Lox/Lox , and Chx10-Cre;Rb Lox/Lox mice ( 7, 9, 12, 14 ). Studies using a point mutant knock-in of Rb have shown that E2F1 and E2F3 are likely candidates for regulating the rod-specific genes ( 15 ) and preliminary studies have shown that simultaneous inactivation of Rb and E2F1 rescues the rod developmental defect. 4 Rb-deficient immature cells in the ONL are largely quiescent and only occasionally reenter the cell cycle.

In addition to defective rod development, Pax-6 Cre;Rb Lox/Lox retinas exhibit failure to exit the cell cycle despite the expression of cell type–specific differentiation markers ( 9 ). Rb −/− retinal cells ultimately exited the cell cycle or underwent cell death ( 7, 9, 12, 14 ). It is not yet known if loss of cell types, such as bipolar and ganglion cells in the developing Rb -deficient retina, is due to cell autonomous effects, noncell autonomous effects, or a combination of the two.

The Rb family of proteins is made up of three members (Rb, p107, and p130) that share some functional protein domains and have overlapping roles in regulating growth control during development ( 16 ). With additional mutation of p107 or p130 , levels of developmental proliferation and apoptosis increase and animals become retinoblastoma prone ( Fig. 1 ; refs. 7 – 9, 12, 17, 18 ). Although these studies have revealed similarities in the effects of Rb deletion together with p130 or p107 loss, there are also clear differences. Retinal disorganization upon Rb and p107 mutation starts at embryonic stages, whereas disorganization occurs later, at postnatal stages in the absence of Rb and p130 . This is consistent with differences in the expression of p107 and p130 throughout development, as p107 is more highly expressed at embryonic stages, whereas p130 exhibits increased expression at postnatal stages of retinal development ( 12 ). Chx10-Cre;Rb Lox/Lox ;p130 −/− and Pax6-Cre;Rb Lox/Lox ;p130 −/− mice developed rapid bilateral retinoblastoma with 100% penetrance ( 18 ), 4 whereas Chx10-Cre;Rb Lox/Lox ;p107 −/− and Pax6-Cre;Rb Lox/Lox ;p107 −/− animals developed predominantly unilateral retinoblastoma with partial penetrance and delayed onset of tumorigenesis. Tumor progression, with invasion of the optical nerve, lymph nodes, and brain also occurred more rapidly in Pax6-Cre;Rb Lox/Lox ;p130 −/− animals compared with Pax6-Cre;Rb Lox/Lox ;p107 −/− mice. These data suggest that p130 is a stronger suppressor of retinoblastoma than p107, and the Pax6-Cre;Rb Lox/Lox ;p130 −/− animals may be particularly useful as a preclinical model for advanced bilateral retinoblastoma which presents the greatest challenge clinically.

Kinetics of murine retinoblastoma development. Plotted is the time to the first observation of unilateral and bilateral retinoblastoma in Pax6-Cre;Rb Lox/Lox with additional deletion of p107 or p130. Inset, Pax6-Cre;Rb Lox/Lox ;p130 −/− mouse at age of 18 wks with bilateral retinoblastoma.

A close examination of tumor emergence in the Pax6-Cre;Rb Lox/Lox ;p130 −/− model revealed that early lesions histologically resembling retinoblastoma at P21 and P31 were located at the far retina periphery ( 18 ). The peripheral tumor localization may reflect a specific cell of origin in this region, an aspect of the regional environment permissive for tumor cell survival and/or proliferation, or a property of Pax6-Cre expression. Tumors emergence was not restricted to the peripheral retina in the Chx10-Cre;Rb Lox/Lox ;p130 −/− mice. 4 Although it is possible that the tumors result from a normally postmitotic cell in this region, heterogeneity in the resulting tumors suggests that a multipotent cell type may be responsible. A critical area for future work will be to better define the nature of the cells giving rise to retinoblastoma and the similarities and differences between human retinoblastoma and the mouse models.

It has been well established that inactivation of the RB1 gene is the initiating genetic lesion for retinoblastoma. However, as mentioned above, Rb -deficient mouse retinoblasts do not form retinoblastoma. Detailed characterization of the expression of the Rb family (Rb, p107, and p130) in the developing mouse and human retina and analysis of intrinsic genetic compensation in Rb -deficient mouse and human retinoblasts has suggested a model to account for this species-specific difference in retinoblastoma susceptibility ( 12 ). In the developing mouse retina, p107 is the major family member expressed in embryonic retinal progenitor cells and Rb is the major family member expressed in postnatal retinal progenitor cells ( 12 ). There is some overlapping expression between Rb and p130 in the postnatal and adult mouse retina. When Rb is inactivated in the developing mouse retina, p107 is increased in a compensatory manner in many cells ( 12, 14 ). We hypothesize that the inactivation of Rb is not sufficient for retinoblastoma formation in the mouse due to intrinsic genetic compensation by p107 or intrinsic genetic redundancy by p130. As a result, retinoblastoma only forms in Rb;p107 -deficient or Rb;p130 -deficient retinoblasts in the mouse retina. In the developing human retina, RB1 is the major family member expressed. When RB1 is inactivated, there is little, if any, compensation by p107 or p130. This may explain why inactivation of RB1 is sufficient for retinoblastoma formation in humans but not in mice.

Refinement of this model will require a fuller understanding of the specific retinoblastoma-originating cells and the response of such cells to Rb deletion. Whereas p107 increases occur in postnatal stages of murine retinal development after Rb deletion, this is not seen at embryonic stages ( 7, 12 ). In embryonic retinas, the levels of a critical regulator of p107 activity, cyclin D1, decreases in response to Rb deletion, which may also restrain the effects of Rb activity at the level of pocket protein activity. The developmental stage at which the cells in the retina are prone to tumor initiation upon Rb family mutation is not known. Although these studies clearly show that the genetics of Rb family inactivation is different in humans and mice with respect to retinoblastoma susceptibility, there is a common theme across species. In both mice and humans, Rb family function must be inactivated to initiate retinoblastoma; therefore, the preclinical models recapitulate these early events in retinoblastoma tumorigenesis.

Whereas it has been well established that RB1 inactivation is the initiating genetic lesion in retinoblastoma, only a handful of secondary genetic lesions have been carefully analyzed. The newly developed knockout mouse models of retinoblastoma provided a critical tool to begin to identify secondary genetic lesions in retinoblastoma. Using the Pax6-Cre;Rb Lox/Lox ;p107 −/− and Pax6-Cre;Rb Lox/Lox ;p130 −/− models, it was recently found that N-myc also undergoes gene amplification in murine retinoblastoma ( 18 ). The N-myc oncogene has also been found to undergo gene amplification in a subset of human retinoblastomas. The frequency of N-myc amplification (found in 5 of 61 tumors overall) is similar to the frequency in human retinoblastomas ( 18 ). In tumors lacking Rb and p130, N-myc amplification was associated with metastasis, suggestive of a role for N-myc in tumor progression. Common amplification of N-myc in murine and human retinoblastoma provides further evidence that the murine models faithfully recapitulate human disease.

It was also recently shown that Chx10-Cre;Rb Lox/− ;p107 −/− ;p53 Lox/− mice develop bilateral invasive aggressive retinoblastoma with 100% penetrance. This is in contrast to Chx10-Cre;Rb Lox/− ;p107 −/− mice that develop unilateral minimally invasive retinoblastoma with ∼50% penetrance. These data suggested that p53 suppressed retinoblastoma in mice. However, previous studies have shown that the p53 gene was intact in human retinoblastomas. Molecular genetic analyses revealed that the MDMX gene was amplified in 65% of retinoblastoma cases and MDM2 was amplified in an additional 10% ( 19 ). Using mouse models of retinoblastoma, cultured retinoblastoma cell lines, human fetal retinae, and primary retinoblastomas from enucleated eyes, it was confirmed that amplification of MDMX suppresses p53-mediated cell death in RB1 -deficient retinoblasts and promotes clonal expansion of the tumors. Whereas these findings do not rule out the possibility of other nonapoptosis-related functions of p53 inactivation in tumorigenesis, there is a clear connection between suppression of p53-mediated apoptosis and retinoblastoma progression.

Not only do these data argue that retinoblastoma does not arise from an intrinsically death resistant cell as previously proposed ( 9 ), but it provided the first specific target for chemotherapy. Nutlin-3, a small molecule inhibitor of MDM2 was found to bind to MDMX and prevent it from binding to p53. Using some of the newly developed preclinical models of retinoblastoma ( 20 ), Nutlin-3 was found to have potent antitumor effects, especially when combined with another drug that also induces a p53 response, topotecan ( 19 ). This is the first example of a locally delivered targeted chemotherapy for any pediatric cancer. Importantly, preclinical models were essential for these studies on the basic biology of retinoblastoma and for testing new drug combinations and methods of delivery.

The past three years have witnessed a remarkable advancement in our understanding of retinoblastoma biology and the development of new therapeutic interventions. The field is rapidly moving toward implementing locally delivered targeted therapies which may improve outcomes while reducing toxicity associated with radiotherapy or systemic broad-spectrum chemotherapy. These advances have been built on the strong foundation of genetic studies of retinoblastoma over the past several decades and have been accelerated by the development of new mouse models of retinoblastoma. It is encouraging that there is broad agreement from independent laboratories using different approaches about the role of Rb family in retinal development and retinoblastoma. More importantly, these studies have raised exciting new questions that will advance our understanding of tumorigenesis in the developing central nervous system and the role or tumor suppressor genes in normal developmental processes in the years to come.

Citing articles via

Email alerts.

• Online First
• Collections
• Online ISSN 1538-7445
• Print ISSN 0008-5472

AACR Journals

• Blood Cancer Discovery
• Cancer Discovery
• Cancer Epidemiology, Biomarkers & Prevention
• Cancer Immunology Research
• Cancer Prevention Research
• Cancer Research
• Cancer Research Communications
• Clinical Cancer Research
• Molecular Cancer Research
• Molecular Cancer Therapeutics
• Info for Advertisers
• Information for Institutions/Librarians

• Privacy Policy
• Copyright © 2023 by the American Association for Cancer Research.

This Feature Is Available To Subscribers Only

Sign In or Create an Account

When you choose to publish with PLOS, your research makes an impact. Make your work accessible to all, without restrictions, and accelerate scientific discovery with options like preprints and published peer review that make your work more Open.

• PLOS Biology
• PLOS Climate
• PLOS Complex Systems
• PLOS Computational Biology
• PLOS Digital Health
• PLOS Genetics
• PLOS Global Public Health
• PLOS Medicine
• PLOS Mental Health
• PLOS Neglected Tropical Diseases
• PLOS Pathogens
• PLOS Sustainability and Transformation
• PLOS Collections
• About This Blog
• Official PLOS Blog
• EveryONE Blog
• Speaking of Medicine
• PLOS Biologue
• Absolutely Maybe
• DNA Science
• PLOS ECR Community
• All Models Are Wrong
• About PLOS Blogs

Prostate and Colon Cancer News: The 2-Hit Hypothesis Revisited

A report and a case published in two major medical journals this week suggest that relatives of certain individuals with cancer may be at higher risk, due to inherited (germline) mutations in DNA repair genes.

Only 5 to 10 percent  of cancers are inherited. Such individuals inherit a cancer-predisposing mutation in all their cells, and then a somatic (body) cell undergoes a second mutation that initiates the disease. The second mutation may be spontaneous or in response to an environmental factor such as smoking.

The one-two punch of inherited cancers, called the 2-hit hypothesis, was first described in 1971.

An article published yesterday in The New England Journal of Medicine  from researchers at half a dozen top cancer centers implicates germinal DNA repair gene mutations in some men who have metastatic prostate cancer, which places their relatives at elevated risk for several types of cancer.

Of 692 men with metastatic prostate cancer but not selected for family history of cancer, 84 had germline mutations in at least one of 20 DNA repair genes, detected in blood, saliva, or cheek lining cells. That’s 11.8%. Among 499 men with localized prostate cancer it was 4.6%, and among 53,105 people without cancer  it was 2.7%. The researchers found other cancers in the families of the men with metastatic cancer who had germline repair mutations, compared to men who did not have the mutations. The repair genes found, including BRCA1 and 2 , ATM , CHEK2 , RAD51D , and PALB2 , are all well-studied and tests for them have been around for years.

Meanwhile, researchers from Rush University Medical Center published a Quiz in The Journal of the American Medical Association  based on the case of a woman with endometrial cancer who had a father and grandfather with a form of colorectal cancer called ( Lynch Syndrome ) (aka hereditary nonpolyposis colorectal cancer). Lynch syndrome is caused by mutations in at least four genes that confer up to 80% lifetime risk of colorectal cancer, 60% risk of endometrial cancer, and increased risk of stomach, ovarian, pancreatic, kidney, and prostate cancers.

Lynch syndrome is rare — only 1 in 35 patients diagnosed with colorectal or endometrial cancers has it. But accompanying the quiz is an editorial whose title says it all:  “ Lynch Syndrome Testing: A Missed Opportunity in the Era of Precision Medicine” . Identifying the mutations behind cancers can inform targeted treatment decisions.

GEE-WHIZ NEWS COVERAGE

As usual, I just about shut down after perusing the initial flurry of news reports on the NEJM paper. Time  heralded a “new genetic test” (no, DNA repair gene tests have been around quite awhile) and The Telegraph  headline “Genes which raise risk of prostate cancer discovered by scientists” is also ignorant of cancer gene history. The Washington Post ‘s  “defects in genes that are designed to fix damage to DNA” stopped me in my tracks. Genes aren’t designed.

My search wasn’t exhaustive, but I couldn’t find anyone linking the prostate and Lynch papers, nor recognizing their shared ancestry in the 2-hit hypothesis. These days news aggregation leaves no room for history and context, so that’s my contribution for this week.

HALFWAY TO CANCER

The two new reports trace their roots back to a 1971 paper famous for describing the first tumor suppressor gene, RB1 . (A mutation that removes or inactivates a tumor suppressor gene causes cancer. Or, a mutation that overexpresses a proto-oncogene creates an oncogene, which causes cancer.)

Alfred Knudson, from the Fox Chase Cancer Center, proposed the “2-hit” hypothesis  about the inherited eye cancer retinoblastoma (RB). Work over the previous two decades  had attributed cancer origin to a series of mutations, or “hits.”

Knudson examined the medical records of 48 children with RB admitted to M.D. Anderson Hospital between 1944 and 1969, recording whether one or both eyes were affected, the age of the child at the time of diagnosis, gender, number of tumors per eye, and whether relatives had RB. The fact that boys and girls were affected indicated autosomal inheritance, and about 50% of the kids of affected parents were also affected. That indicated dominant inheritance, which means the trait affects every generation.

But something puzzling arose.

In a few families, a child with tumors in both eyes had an affected grandparent, but parents with healthy eyes. Sometimes dominant genes do this, and we call it “incomplete penetrance.” But Knudson saw beyond the label to the mechanism. Perhaps an initial, inherited recessive mutation had to be followed by a second, somatic mutation in a cell in an eye to trigger a tumor. When occasionally this second mutation doesn’t happen, the result is unaffected parents nestled between two affected generations.

Knudson’s 2-hit hypothesis explained another, subtle finding. Children with tumors in both eyes were much younger at diagnosis, some even born with the opaque patches in the eye that reflect light strangely, often the first hint of a problem. Early onset makes sense if a child is dealt a mutation in one copy of the RB gene at conception and then acquires a somatic mutation in the second copy of the gene in a cell in a retina. Two somatic mutations happening in the same cell would take longer. In other words, in inherited RB, a child is born halfway to cancer – one somatic mutation in the eye is all that’s needed for a tumor to form. The risk is so high that tumors often develop in both eyes. In contrast, the one-eye, noninherited form appears later in childhood because it takes longer for somatic mutations to occur in both copies of the RB gene in the same cell.

“THREE STRIKES TO CANCER”

Notably, “metastasis” doesn’t have its own stage because the mutations that drive it are probably already in place early on – and that’s exactly why the two new reports are so important. Checking the relatives of patients likely to have germline mutations in DNA repair genes may prove to be life-extending or even life-saving.

[…] Source: Prostate and Colon Cancer News: The 2-Hit Hypothesis Revisited […]

Interestingly, my thesis on autovirulence can explain these findings. The crucial question will be to disambiguate the role(s) of stress-activated autovirons in contributing to germline mutations.

Which findings? Can you explain?

Leave a Reply Cancel reply

Your email address will not be published. Required fields are marked *

Save my name and email for the next time I comment.

Advertisement

Heterogenic Loss of BRCA in Breast Cancer: The “Two-Hit” Hypothesis Takes a Hit

• Published: 04 April 2007
• Volume 14 , pages 2428–2429, ( 2007 )

Cite this article

• Funda Meric-Bernstam MD 1  

1226 Accesses

14 Citations

3 Altmetric

Explore all metrics

This is a preview of subscription content, log in via an institution to check access.

Access this article

Price includes VAT (Russian Federation)

Instant access to the full article PDF.

Rent this article via DeepDyve

Institutional subscriptions

Knudson AG. Two genetic hits (more or less) to cancer. Nat Rev Cancer 2001; 1:157–62

Article   PubMed   CAS   Google Scholar  

Collins N, McManus R, Wooster R, et al. Consistent loss of the wild type allele in breast cancers from a family linked to the BRCA2 gene on chromosome 13q12–13. Oncogene 1995; 10:1673–5

PubMed   CAS   Google Scholar  

Gudmundsson J, Johannesdottir G, Bergthorsson JT, et al. Different tumor types from BRCA2 carriers show wild-type chromosome deletions on 13q12–q13. Cancer Res 1995; 55:4830–2

Neuhausen SL, Marshall CJ. Loss of heterozygosity in familial tumors from three BRCA1-linked kindreds. Cancer Res 1994; 54:6069–72

Osorio A, de la Hoya M, Rodriguez-Lopez R, et al. Loss of heterozygosity analysis at the BRCA loci in tumor samples from patients with familial breast cancer. Int J Cancer 2002; 99:305–9

Smith SA, Easton DF, Evans DG, Ponder BA. Allele losses in the region 17q12–21 in familial breast and ovarian cancer involve the wild-type chromosome. Nat Genet 1992; 2:128–31

Chan KY, Ozcelik H, Cheung AN, et al. Epigenetic factors controlling the BRCA1 and BRCA2 genes in sporadic ovarian cancer. Cancer Res 2002; 62:4151–6

Evers B, Jonkers J. Mouse models of BRCA1 and BRCA2 deficiency: past lessons, current understanding and future prospects. Oncogene 2006; 25:5885–97

Mote PA, Leary JA, Avery KA, et al. Germ-line mutations in BRCA1 or BRCA2 in the normal breast are associated with altered expression of estrogen-responsive proteins and the predominance of progesterone receptor A. Genes Chromosomes Cancer 2004; 39:236–48

Kote-Jarai Z, Matthews L, Osorio A, et al. Accurate prediction of BRCA1 and BRCA2 heterozygous genotype using expression profiling after induced DNA damage. Clin Cancer Res 2006; 12:3896–901

Download references

Author information

Authors and affiliations.

Department of Surgical Oncology, The University of Texas M. D. Anderson Cancer Center, 1400 Holcombe Boulevard, P.O. Box 301402, Houston, Texas, 77030, USA

Funda Meric-Bernstam MD

You can also search for this author in PubMed   Google Scholar

Corresponding author

Correspondence to Funda Meric-Bernstam MD .

Rights and permissions

Reprints and permissions

About this article

Meric-Bernstam, F. Heterogenic Loss of BRCA in Breast Cancer: The “Two-Hit” Hypothesis Takes a Hit. Ann Surg Oncol 14 , 2428–2429 (2007). https://doi.org/10.1245/s10434-007-9379-7

Download citation

Received : 08 December 2006

Accepted : 12 December 2006

Published : 04 April 2007

Issue Date : September 2007

DOI : https://doi.org/10.1245/s10434-007-9379-7

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Find a journal
• Publish with us
• Track your research

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

• Publications
• Account settings

Preview improvements coming to the PMC website in October 2024. Learn More or Try it out now .

• Advanced Search
• Journal List
• v.5(1); 2014 Jan

NF2/Merlin in hereditary neurofibromatosis 2 versus cancer: biologic mechanisms and clinical associations

Rebecca dunbar schroeder.

1 Department of Investigational Cancer Therapeutics (Phase I Program), The University of Texas MD Anderson Cancer Center, Houston, TX,

2 Program in Experimental Therapeutics, The University of Texas Graduate School of Biomedical Sciences, Houston, TX

Laura S. Angelo

Razelle kurzrock.

Inactivating germline mutations in the tumor suppressor gene NF2 cause the hereditary syndrome neurofibromatosis 2, which is characterized by the development of neoplasms of the nervous system, most notably bilateral vestibular schwannoma. Somatic NF2 mutations have also been reported in a variety of cancers, but interestingly these mutations do not cause the same tumors that are common in hereditary neurofibromatosis 2, even though the same gene is involved and there is overlap in the site of mutations. This review highlights cancers in which somatic NF2 mutations have been found, the cell signaling pathways involving NF2/merlin, current clinical trials treating neurofibromatosis 2 patients, and preclinical findings that promise to lead to new targeted therapies for both cancers harboring NF2 mutations and neurofibromatosis 2 patients.

INTRODUCTION

Neurofibromatosis type 2 (NF2) is a tumor suppressor gene on chromosome 22q12 that encodes a protein product named “merlin” (or schwannomin) affecting multiple cell signaling pathways (Figure ​ (Figure1 1 )[ 1 , 2 ]. Constitutional mutations in the NF2 gene cause an autosomal-dominant disorder (neurofibromatosis type 2) affecting about 1 in 33,000 people, and characterized by the development of tumors primarily affecting the nervous system[ 3 ]. Approximately half of the cases are due to de novo mutations not inherited from a family member [ 4 ]. NF2 mutations in neurofibromatosis 2 can either be germline (70%) or somatic mosaic (30%), the latter referring to mutations present in only a subset of cells[ 5 , 6 ].

Merlin inhibits PI3K signaling cascade at PIKE-L[ 19 , 30 ] and mTORC1[ 32 , 33 ]. The RAS and Src pathways are inhibited by merlin via the Src/FAK complex[ 34 , 36 ]. Transcription is inhibited by merlin via Hippo kinase cascade[ 81 , 82 ] and CRL4-DCAF[ 83 ]. Merlin can also down-regulates cell surface growth factor receptors. Dotted line indicates nuclear membrane. Targeted therapies are indicated in red.

Neurofibromatosis 2 follows the Knudson 2-hit hypothesis, with constitutional (germline) mutations occurring initially (first hit) and additional somatic mutations (second hit) being required for loss of heterozygosity and tumor suppressor inactivation followed by tumor formation[ 7 - 9 ]. The majority of tumors associated with neurofibromatosis type 2 are schwannomas, meningiomas, and ependymomas[ 7 ]. Other than these specific tumor types, there is no published evidence to support an increased incidence of cancer in individuals with neurofibromatosis 2[ 7 ]. In general, in neurofibromatosis 2, constitutional nonsense or frameshift NF2 mutations are associated with more severe disease, while missense mutations, large deletions, or somatic mosaicism results in milder disease (fewer tumors and older age of onset)

NF2 somatic mutations have also been found in multiple cancer types, including but not limited to mesothelioma, anaplastic thyroid cancer, breast cancers, endometrial and liver cancers, in patients not having constitutional NF2 mutations (Table ​ (Table1 1 )[ 10 ]. Missense mutations are more frequent in cancer than in neurofibromatosis 2; they occur in only a small subset of patients with latter condition. This may in part explain the lack of predisposition to developing cancer in patients with neurofibromatosis 2, though the explanation is not complete, since there remains considerable overlap between neurofibromatosis 2 and cancer-related NF2 aberrations (Figure ​ (Figure2 2 )[ 11 , 12 ].

NF2 mutations have been found in neurofibromatosis 2 and also in numerous cancers[ 15 ]. Shown are the non-truncating constitutional and somatic mutations found in neurofibromatosis 2 [blue boxes] and non-truncating (missense) and truncating (nonsense or frameshift) mutations found in human cancer [black boxes][ 84 ]. Gray boxes indicate constitutional mutations while white boxes indicate somatic mutations. Because there are numerous truncating lesions found in neurofibromatosis 2[ 69 ] and in non-hereditary meningiomas, ependymomas, and schwannomas,[ 69 , 85 ] these could not be depicted in figure ​ figure2. 2 . Truncating lesions that are found in both neurofibromatosis 2 and cancer are shown in red. Missense mutations that are found in both neurofibromatosis 2 and cancer are also shown in red followed by [mis]. Termination sites of translation for frameshift mutations is indicated with an (*) and for nonsense mutations with an (X).

Clinical Manifestations of Neurofibromatosis 2

The most common tumor associated with neurofibromatosis 2 is bilateral vestibular schwannoma (acoustic neuromas)[ 12 , 13 ]. Initial symptoms of neurofibromatosis 2 include hearing loss, facial nerve impairment, visual disruptions, and skin tumors; with the majority of patients presenting before the age of 20[ 12 ]. Many patients with multiple tumor sites live only into their early thirties, and overall survival rate for patients 20 years after diagnosis is approximately 40%[ 12 , 13 ]. Treatment plans for patients with NF2 revolve around monitoring, surgery, and radiation therapy[ 12 ]. Due to recent advancements in our understanding of the signaling pathways affected by merlin, several targeted therapies are now being tested[ 14 ]. However, there remains a paucity of clinical trials, and a pressing need to accelerate the pace of testing in the clinical setting.

The NF2 gene contains 17 exons and its product— merlin—is a tumor suppressor that controls protean cell signaling pathways implicated in cell growth, proliferation, and morphology[ 15 , 16 ]. Merlin acts as a membrane-cytoskeleton scaffolding (Ezrin–Radixin– Moesin (ERM)) protein that localizes underneath the plasma membrane at cell–cell junctions and other actinrich sites[ 17 , 18 ]. It links the cytoskeleton to the cell membrane. Two isoforms of merlin have been described that differ by the presence (type 2 merlin) or absence (type 1 merlin; includes exons 1-15 and 17) of exon 16 sequences inserted into the extreme carboxyl terminus of the protein.

Myosin phosphatase MYPT1–PP1δ directly activates merlin by dephosphorylating it. Inactivation (phosphorylation) of merlin contributes to malignant conversion in multiple cell types, as does loss of merlin expression because of mutation [ 19 ]. Importantly, phosphorylation not only causes the conformation of merlin to change to the open/inactive state, but also targets merlin for polyubiquitination and degradation by the proteasome [ 19 - 21 ]. The decreased ability of mutant merlin to effectively act as a tumor suppressor is at least in part a direct result of the decreased half-life of mutant compared to wild-type merlin[ 22 , 23 ].

Functions of NF2/merlin

Several different signaling pathways crucial to cell proliferation are inhibited by merlin: PIKE-L/PI3K, mTORC1, Src/Fak, Mst1/2, Ras/Rac/PAK, ERK1/2, AKT and CRL4-DCAF (Figure ​ (Figure1 1 )[ 1 , 24 , 25 ]. Merlin also acts as an unconventional cell cycle regulator, and it links receptors at the plasma membrane to their cytoplasmic kinases to facilitate contact inhibition [ 26 , 27 ].

PI3K/PIKE-L

The phosphatidy linositol 3-kinase (PI3K) pathway is activated in multiple human cancers, and contributes to increased cellular proliferation and metabolism, and decreased apoptosis when activated[ 28 , 29 ]. Merlin suppresses the PI3K pathway by binding with the PI3K-enhancer-isoform-L (PIKE-L), hence preventing PIKE-L from binding PI3K[ 30 , 31 ].

Inhibition of mammalian-target-of-rapamycin-complex-1 (mTORC1) due to merlin activation leads to the inhibition of mRNA translation, which causes an increase in apoptosis limiting cell survival and blocking tumor initiation[ 32 ]. mTORC1 inhibition is accomplished through multiple pathways including AKT/ERK inhibition and integrin specifc adhesion via p21-activated kinase (PAK)[ 32 , 33 ]. mTORC1 has also been identified as an important pathway in meningioma and schwannoma cell growth[ 33 ].

ErbB2/Src/FAK/Paxillin

Merlin binds the receptor ErbB2 and also to Src, inhibiting Src from binding to and being phosphorylated by ErbB2 in a competitive manner [ 34 ]. As a result, Src cannot activate focal adhesion kinase (FAK) and paxillin, which inhibits cellular proliferation and growth[ 34 ]. Interestingly, binding of merlin to ErbB2 is initiated by paxillin binding to merlin on exon 2 at residues 50-70, causing merlin to move to the plasma membrane where it binds ErbB2[ 35 ]. Merlin also inhibits FAK from binding to Src and PI3K [ 36 ].

ERK1/2, AKT, and PDGFR

Schwannoma cell lines lacking the tumor suppressor activity of merlin have high basal levels of phosphorylated extracellular signal-regulated kinase (ERK1/2) and AKT, which are activated by the Src/FAK/Ras and PI3K/Raf signaling cascades and platelet-derived growth factor receptor beta (PDGFRβ) (Figure ​ (Figure1 1 )[ 37 ]. Merlin inhibits the activity of ERK and MAPK via the upstream effector Raf-1[ 38 , 39 ]. In normal cells, merlin promotes PDGFRβ degradation as well, thereby inhibiting cellular proliferation [ 37 , 40 ]. In a positive feedback loop, activated AKT also binds and phosphorylates merlin on residues Thr230 and Ser315, which causes merlin to be ubiquitinated and targeted for degradation[ 19 , 20 , 41 ].

Due to merlin's sequence homology with ERM proteins, merlin binds the same proteins, but unlike ERM proteins merlin acts in an inhibitory manner by suppressing cell growth and proliferation [ 42 , 43 ]. Merlin may suppress the oncogenic signaling of the small GTPases Ras and Rac by interfering with guanine nucleotide-exchange factor (GEF) activity, which is required for Ras and Rac activation[ 43 , 44 ]. Merlin inhibits signaling of the Ras pathway downstream by inhibiting Rac activation, which results in the inhibition of phosphorylation/activation of the serine/threonine kinase PAK [ 43 ]. When PAK is not phosphorylated it is unable to phosphorylate RAF and MEK, which are both necessary to activate the Ras signaling pathway[ 43 , 45 ].

NF2 mutations in cancer

Data collected from COSMIC and ICGC databases and published studies show that diverse cancers harbor somatic NF2 mutations (Table ​ (Table1 1 ).

Thoracic tumors (mesothelioma and lung cancer)

The most common cancer linked to NF2 aberrations is mesothelioma, with approximately 30-50% of tumors having mutations in the NF2 coding regio[ 22 , 46 - 48 ]. The NF2 mutations found in neurofbromatosis 2 differ from the mutations found in mesotheliomas in that mutations in hereditary NF2 are not usually missense mutations. Only ~5% of neurofibromatosis 2 patients have constitutional missense mutations and these types of mutations are typically associated with a milder version of NF2 [ 49 , 50 ]. This dichotomy may partially explain why patients with neurofibromatosis 2 do not develop mesothelioma[ 51 ]. A variety of lung cancers have also been found to harbor NF2 mutations, but at a much lower rate (1-2%)[ 11 ].

Sporadic schwannomas, meningiomas, and ependymomas

Sporadic schwannomas, meningiomas, and ependymomas fund in patients who do not suffer from constitutional neurofibromatosis 2 have somatic NF2 mutations at rates of 42%, 27%, and 4%, respectively[ 46 ]. These are also the most prevalent tumor types found in individuals with constitutional NF2 mutations[ 7 ]. Mutations found in these sporadic tumors tend to include more frameshift mutations, while those found in neurofibromatosis 2 are often nonsense mutations[ 52 , 53 ].

Thyroid cancer

NF2 mutations have also been discerned in 18% (two of 11) anaplastic thyroid cancers. These are uncommon (accounting for only 2% of all thyroid cancers) but very aggressive cancers, with a median survival of only a few months[ 54 ]. Of interest, other pathways (PIK3/AKT/MTOR, RAS/RAK and ERK) activated via mutation in this cancer type are negatively regulated by active merlin [ 54 - 59 ].

Breast cancer

One to two percent of breast cancers have NF2 mutations [ 15 , 60 - 62 ]. A decrease in merlin expression correlates with increase in tumor grade [ 47 ]. When merlin expression is reestablished in breast xenograft models, tumorigenesis is reduced [ 47 ]. Since NF2 mutations activate mTor, we treated a patient with NF2-mutant metaplastic breast cancer (a highly refractory form of breast cancer) with a temsirolimus-based regimen, which resulted in a complete remission [ 63 ].

Glioblastoma multiforme

Decreased expression of merlin RNA and protein levels have been observed in several human glioblastoma multiforme tumors; 27% of grade 4 gliomas have loss of merlin expression [ 46 ]. Merlin decreases proliferation and invasiveness while increasing apoptosis-induced cell death when its tumor suppressor capabilities are reestablished in the corresponding glioma cell lines [ 46 ].

Endometrial cancer

Ten percent of endometrial carcinomas harbor NF2 mutations. This observation is of interest since NF2 mutations can activate mTor, the downstream effector of PIK3CA, and a high percentage of this cancer type (>80%) have PI3K pathway aberrations [ 30 , 64 ].

Hepatocellular tumors

Twenty-three percent of liver cancers have mutations in NF2. Mice with heterozygous NF2 mutations develop hepatocellular carcinoma and cholangiocarcinoma [ 65 , 66 ].

Comparing NF2 aberrations in hereditary neurofibromatosis and in cancer

Overlap in truncating and non-truncating mutations found in cancers and hereditary neurofibromatosis 2 are shown in Figure ​ Figure2 2 (red lettering). There have not been any mutations found in exons 16 and 17, which are the two exons affected by alternative splicing, and no known proteins have been found to bind to exon 16[ 33 , 67 ]. Tumor suppressor activity can be attenuated by truncating mutations in any of the other exons 1-15[ 68 ]. In neurofibromatosis 2, truncating mutations (frameshift and nonsense) are associated with an increase in disease severity when compared with missense mutations[ 7 ]. Nonsense mutations are more frequent in hereditary neurofibromatosis 2 than in sporadic schwannomas, ependymomas, and meningiomas where frameshift mutations are more common [ 53 , 69 ]. Missense mutations are more frequent in cancer, but occur in only about 5% of hereditary neurofibromatosis 2.

There are several binding sites in the merlin protein that are crucial for merlin's function, being critical to interactions with other proteins or for correct merlin folding and activation. Merlin must associate with itself in order to change conformation and become active. This is achieved first by the folding of the N-terminal domain, which involves the interaction of amino acids 8-121 with amino acids 200-320[ 67 , 70 ]. After the conformational change of the N-terminus is complete, merlin can transition to the active, closed conformation by the association of the N-terminal amino acids 302-308 with the C-terminal amino acids 580-595[ 67 , 70 ]. When examining Figure ​ Figure2, 2 , it is remarkable that no mutations are found in the region where the C-terminus and N-terminus interact, but multiple mutations are found in the regions where the N-terminus interacts with itself.

The phosphatase (MYPT1-PP1δ) responsible for the dephosphorylation and activation of merlin (at residue S518) interacts with merlin via its MYPT1 subunit at residues 312-341 of merlin (exons 10 and 11)[ 67 ]. As shown in figure ​ figure2, 2 , this region has multiple missense mutations in both cancer and NF2.

P21-activating kinases (PAK) are regulated by merlin and also regulate the activity of merlin by phosphorylating S518 and causing merlin to change to the open, inactive state[ 67 ]. PAKs interact with merlin at the N-terminus between amino acids 1-313, and there are many missense mutations in both cancer and neurofibromatosis 2 found within this region (Figure ​ (Figure2) 2 ) [ 67 ].

PIKE-L binds to merlin in the region containing amino acids 1-332, whereas mutant merlin harboring the L64P missense mutation does not bind PIKE-L[ 30 , 67 ]. Importantly, merlin cannot suppress tumorigenesis, including cellular proliferation, via inhibition of the PI3K pathway without binding to PIKE-L[ 30 , 67 ]. Merlin's interaction with PIKE-L could also regulate downstream effectors of PI3K such as AKT and mTOR[ 67 ]. The L64P mutation has not been found in any cancers but is reported in neurofibromatosis 2.

Clinical Trials

Novel targeted therapies are now being used in a small number of cases to treat hereditary neurofibromatosis 2 and cancers harboring NF2 mutations (Table ​ (Table2). 2 ). The vast majority of these trials are ongoing and results are not yet published. A search of clinicaltrials.gov was undertaken in order to find clinical trials treating NF2 patients.

Receptor tyrosine kinase (RTK) Inhibitors

Lapatinib, which targets EGFR and ErbB-2, is being examined in both a phase 0 study treating vestibular schwannomas ( {"type":"clinical-trial","attrs":{"text":"NCT00863122","term_id":"NCT00863122"}} NCT00863122 ) and a phase II ( {"type":"clinical-trial","attrs":{"text":"NCT00973739","term_id":"NCT00973739"}} NCT00973739 ) trial for its effectiveness in all NF2 related tumor types. Clinical analysis has shown that vestibular schwannomas overexpress ErbB2/3 and that EGFR and its ligand are up-regulated in the majority of NF2-related vestibular schwannomas[ 71 , 72 ].

Sporadic and neurofibromatosis 2-related vestibular schwannomas overexpress c-kit and PDGFR, which are targets of the receptor tyrosine kinase (RTK) inhibitor imatinib. In vitro studies using the vestibular schwannoma cell line HEI-193 show a decrease in proliferation and an increase in apoptosis in response to treatment with imatinib. Currently sunitinib and nilotinib, are being tested in patients with neurofibromatosis 2[ 13 ]. Sunitinib is an RTK inhibitor that targets multiple receptors such as PDGFR, VEGFR, and c-KIT. Two phase II clinical trials are currently evaluating the effectiveness of sunitinib in unresectable meningiomas (NCT00561665) and ( {"type":"clinical-trial","attrs":{"text":"NCT00589784","term_id":"NCT00589784"}} NCT00589784 ). Nilotinib is a RTK inhibitor that targets Bcr-Abl, PDGFR, and c-Kit, and is also being used in a phase II clinical trial ( {"type":"clinical-trial","attrs":{"text":"NCT01201538","term_id":"NCT01201538"}} NCT01201538 ) focusing on progressing vestibular schwannomas.

Vascular Endothelial Growth Factor (VEGF) Inhibitors

Bevacizumab is an anti-VEGF monoclonal antibody that inhibits angiogenesis, slowing tumor growth and formation. Schwannomas produce high levels of VEGF. In a study of ten NF2 patients treated with bevacizumab, nine had tumor shrinkage, and seven experienced hearing improvement. Two phase II clinical trials ( {"type":"clinical-trial","attrs":{"text":"NCT01207687","term_id":"NCT01207687"}} NCT01207687 , {"type":"clinical-trial","attrs":{"text":"NCT01125046","term_id":"NCT01125046"}} NCT01125046 ) are assessing the effectiveness of bevacizumab in treating neurofibromatosis 2 patients with symptomatic vestibular schwannomas and recurrent or progressive meningiomas. Another drug PTC299 targets VEGF by inhibiting its synthesis upstream and interfering with post-transcriptional processing [ 73 ]. PTC299 has been studied in neurofibromatosis 2 patients via a phase II clinical trial [ 73 ].

mTOR Inhibitors

Everolimus (RAD-001) is an mTOR inhibitor. Merlin is a negative regulator of mTOR. Therefore, in patients with deactivating NF2 mutations, one could hypothesize that an mTOR inhibitor would restore merlin's inhibition of mTOR and arrest tumor formation [ 32 ]. There are three phase II clinical trials using everolimus to treat NF2-related tumors: {"type":"clinical-trial","attrs":{"text":"NCT01490476","term_id":"NCT01490476"}} NCT01490476 , {"type":"clinical-trial","attrs":{"text":"NCT01345136","term_id":"NCT01345136"}} NCT01345136 and {"type":"clinical-trial","attrs":{"text":"NCT01419639","term_id":"NCT01419639"}} NCT01419639 . A fourth phase II clinical trial ( {"type":"clinical-trial","attrs":{"text":"NCT01024946","term_id":"NCT01024946"}} NCT01024946 ) is available for treatment of malignant pleural mesotheliomas using NF2/merlin loss as a biomarker to predict everolimus sensitivity. The mTOR inhibitor temsirolimus was tried in combination with the anti-VEGF antibody bezacizumab in a clinical case series focusing on patients with neurofibromatosis 2; of the two patients on this regimen, one achieved a 33% reduction in tumor size[ 74 ]. Of interest, one patient with metaplastic breast cancer who harbored an NF2 mutation achieved a complete remission on a temsirolimus-containing regimen [ 63 ].

RAS and CDK inhibitors

Merlin blocks RAS activation. When merlin activity is lost, RAS can then move into the active state and promote cell growth and proliferation [ 13 ]. S-trans, trans-farnesyl-thiosalicylic-acid (FTS) is a RAS inhibitor that was administered to two patients with hereditary neurofibromatosis 2, one of whom achieved stable disease for over 4.5 years[ 74 , 75 ].

Preclinical Studies

Treatment of human schwannoma cells in vitro with curcumin (diferuloylmethane), caused dephosphorylation of AKT and ERK1/2 and activation of the merlin phosphatase MYPT1-pp1δ, which is responsible for the dephosphorylation of S518 [ 76 ]. Interestingly, hsp70 was up-regulated following curcumin treatment, which could serve as a resistance mechanism. Hence, a heat shock protein inhibitor (KNK437) was used in combination with curcumin to block this potential resistance mechanism [ 76 ]. Merlin loss also causes constitutive activation of receptor tyrosine kinases (EGFR, ErbB2, and ErbB3), which can be targeted by EGFR and ErbB2 inhibitors such as erlotinib or lapatinib[ 13 ]. Finally, sorafnib, a PDGFR and c-RAF inhibitor has been shown to decrease proliferation in human schwannoma cell line [ 37 ].

CONCLUSIONS

NF2 is a complex gene with somatic mutations being associated with various cancers (e.g. 30 to 50 percent of mesotheliomas harbor NF2 mutations (Table ​ (Table1)). 1 )). Germline mutations cause the autosomal dominant disorder neurofibromatosis 2. Although the somatic mutations sometimes overlap with those in hereditary NF2 (Figure ​ (Figure2), 2 ), there are no published papers documenting an increased risk of cancer in neurofibromatosis patients; individuals with hereditary NF2 do develop schwannomas, ependymomas, and meningiomas. Though there is overlap in the types of NF2 mutations/aberrations between NF2-related conditions (Figure ​ (Figure2), 2 ), nonsense mutations are more frequent in neurofibromatosis 2, frameshift mutations in sporadic schwannomas, meningiomas, and ependymomas, and missense mutations in cancer[ 69 ].

Merlin inhibits PIK3CA and mTOR function along with several other pathways including Src/Fak, Mst1/2 (hippo), Ras/Rac/PAK, ERK1/2, AKT and CRL4-DCAF (Figure ​ (Figure1). 1 ). A small series of mostly anecdotal reports describe activity for anti-angiogenesis agents (bevacizumab), a RAS inhibitor, and the mTor inhibitor temsirolimus in patients with hereditary neurofibromatosis 2[ 74 , 77 ]. In patients with malignancy, an increasing number of studies have established that matching targeted therapy to even a single aberration in patients whose tumors harbor multiple genomic abnormalities can at times result in remarkable responses[ 78 - 80 ]. Since neurofibromatosis 2, unlike cancer, is a single gene disorder, it seems conceivable that proper targeting would result in tumor regressions. In order to explore this possibility, a variety of relevant agents should be explored in the clinical setting. This is especially important because of the morbidity and mortality associated with neurofibromatosis 2, and because many agents that impact NF2-related pathways are already available. Rare conditions are an accrual challenge for larger trials. Therefore, pilot trials whose aim is tumor response may be a mechanism to initially establish activity of an agent, and more of these trials are urgently warranted.

A polygenic two-hit hypothesis for prostate cancer

Affiliations.

• 1 Department of Human Genetics, University of California, Los Angeles, CA, USA.
• 2 Jonsson Comprehensive Cancer Center, University of California, Los Angeles, CA, USA.
• 3 Department of Medical Biophysics, University of Toronto, Toronto, Canada.
• 4 Institute for Precision Health, University of California, Los Angeles, CA, USA.
• 5 Ontario Institute for Cancer Research, Toronto, Canada.
• 6 Vector Institute, Toronto, Canada.
• 7 Department of Urology, University of California, Los Angeles, CA, USA.
• 8 Australian Prostate Cancer Research Centre Epworth, Richmond, VIC, Australia.
• 9 Department of Surgery, The University of Melbourne, Parkville, VIC, Australia.
• 10 Princess Margaret Cancer Centre, University Health Network, Toronto, Canada.
• 11 Bioinformatics Division, Walter and Eliza Hall Institute, Parkville, VIC, Australia.
• 12 Melbourne Bioinformatics, The University of Melbourne, Parkville, VIC, Australia.
• 13 Division of Urology, Royal Melbourne Hospital, Parkville, VIC, Australia.
• 14 Department of Clinical Pathology, The University of Melbourne, Parkville, VIC, Australia.
• 15 Department of Medicine, Central Clinical School, Faculty of Medicine Nursing and Health Sciences, Monash University, Melbourne, VIC, Australia.
• 16 Department of Radiation Oncology, University of California, Los Angeles, CA, USA.
• 17 Department of Urology, Peninsula Health, Frankston, VIC, Australia.
• 18 The Victorian Comprehensive Cancer Centre, Parkville, VIC, Australia.
• 19 Manchester Cancer Research Centre, Manchester, UK.
• 20 Centre for Molecular Medicine Norway (NCMM), Nordic EMBL Partnership, University of Oslo, and Oslo University Hospital, Oslo, Norway.
• 21 Department of Pediatric Research, Division of Paediatric and Adolescent Medicine, Rikshospitalet, Oslo University Hospital, Oslo, Norway.
• 22 Department of Neurology, University of California, San Francisco, San Francisco, CA, USA.
• 23 Biotech Research and Innovation Centre (BRIC), University of Copenhagen, Copenhagen, Denmark.
• 24 Finsen Laboratory, Rigshospitalet, Copenhagen, Denmark.
• 25 Department of Urology, Charité-Universitätsmedizin Berlin, Berlin, Germany.
• 26 Prosserman Centre for Population Health Research, Lunenfeld-Tanenbaum Research Institute, Sinai Health System, Toronto, Canada.
• 27 Epidemiology Division, Dalla Lana School of Public Health, University of Toronto, Toronto, Canada.
• 28 Department of Pharmacology and Toxicology, University of Toronto, Toronto, Canada.
• PMID: 36610996
• PMCID: PMC10086625
• DOI: 10.1093/jnci/djad001

Prostate cancer is one of the most heritable cancers. Hundreds of germline polymorphisms have been linked to prostate cancer diagnosis and prognosis. Polygenic risk scores can predict genetic risk of a prostate cancer diagnosis. Although these scores inform the probability of developing a tumor, it remains unknown how germline risk influences the tumor molecular evolution. We cultivated a cohort of 1250 localized European-descent patients with germline and somatic DNA profiling. Men of European descent with higher genetic risk were diagnosed earlier and had less genomic instability and fewer driver genes mutated. Higher genetic risk was associated with better outcome. These data imply a polygenic "two-hit" model where germline risk reduces the number of somatic alterations required for tumorigenesis. These findings support further clinical studies of polygenic risk scores as inexpensive and minimally invasive adjuncts to standard risk stratification. Further studies are required to interrogate generalizability to more ancestrally and clinically diverse populations.

© The Author(s) 2023. Published by Oxford University Press. All rights reserved. For permissions, please email: [email protected].

Publication types

• Research Support, N.I.H., Extramural
• Research Support, Non-U.S. Gov't
• Genetic Predisposition to Disease
• Prostatic Neoplasms* / genetics
• Prostatic Neoplasms* / pathology
• Risk Factors

Grants and funding

• P30 CA016042/CA/NCI NIH HHS/United States
• U24 CA248265/CA/NCI NIH HHS/United States
• P30CA016042/CA/NCI NIH HHS/United States