MECP2 Is a Frequently Amplified Oncogene with a Novel Epigenetic Mechanism That Mimics the Role of Activated RAS in Malignancy

Several lines of evidence suggest that there are undiscovered oncogenic drivers in human cancers. For example, many recurrent amplifications found in human cancers do not contain known oncogenes, although these amplifications are clearly advantageous to tumor growth, as indicated by their presence in multiple tumor types (1). Furthermore, the recent comprehensive TCGA (The Cancer Genome Atlas) analysis of adenocarcinoma of the lung found significant subsets of tumors with activated MAPK or PI3K pathways without identifiable genomic alterations that explain these cancer-promoting signaling events (2), suggesting the existence of unrecognized oncogenes. In high-grade serous ovarian cancer, 45% of the tumors in the TCGA dataset have known alterations activating PI3K/RAS signaling, but it is not apparent what alterations play this or an analogous role in the remaining 55% (3).

To search for novel oncogenes that contribute to malignant transformation when overexpressed in their wild-type (WT) form, we performed an unbiased genome-scale screen. In this screen, we used a well-studied model system in which the combination of the SV40 early region (encoding SV40 large and small T antigens), hTERT, and activated RAS has been shown to be sufficient to transform a number of primary human epithelial cell types (4). Many investigations have defined the role of various oncogenes and tumor suppressors by leaving out one or more of these elements and substituting a genetic alteration present in human cancers. In our screen, rather than substituting an individual alteration for one of these elements, we added an expression library and selected cells transformed as a result.

We generated primary breast epithelial cells [human mammary epithelial cells (HMEC)] that expressed hTERT and the SV40 early region (“N minus RAS” cells). These cells are unable to grow in an anchorage-independent fashion in soft agar, a surrogate endpoint for transformation; the addition of activated RAS to these cells allows robust soft-agar growth as well as their growth as tumors in xenografts in immunocompromised mice (5). For our screen, we substituted a lentivirus-based expression library (6) for activated RAS and identified library members that allowed these cells to grow in soft agar.

We found that one of the 15 screen hits, methyl cytosine-guanine dinucleotide (CpG) binding protein 2 (MECP2), is amplified in a significant fraction of human malignancies and selected it for further study. MECP2 is an X-linked gene that when mutated causes the autism spectrum disorder, Rett Syndrome (7); MECP2 has no well-described role in malignancy. MECP2 encodes a protein known to exhibit epigenetic control functions. It binds methylated CpG DNA sequences (8) and attracts the SIN3A repressor and histone deacetylase (HDAC) complexes (9), thus maintaining local chromatin surrounding a methylated CpG island in a less active state. Recent studies show that it also acts as a transcriptional activator (10, 11), likely through binding to another epigenetic modification of DNA, 5-hydroxymethylcytosine (5hmC; ref. 12).

Epigenetic processes play important roles in human cancers (13). Tumors often exhibit global DNA hypomethylation, but at the same time reveal promoter hypermethylation (14). These alterations are thought to reflect gene expression changes important for tumorigenesis, including the repression of tumor suppressors. Recent genomic analyses of human tumors have revealed cancer-specific alterations in genes encoding epigenetic modifiers, including genes whose products function in DNA methylation, histone modifications, and chromatin remodeling (15). Given the importance of epigenetic processes in human cancer, attempts have been made to alter various epigenetic processes for therapeutic purposes. In general, epigenetic therapeutics such as DNA methylation inhibitors or HDAC inhibitors, alone or in combination, are thought to have beneficial effect by re-establishing the expression of tumor suppressors repressed by hypermethylation or repressive chromatin marks. In most circumstances, the identities of the tumor suppressors that may be targeted by such therapy are not known, nor are the factors that ultimately caused their repression during tumorigenesis. The observations reported below suggest another possible mechanism of action for epigenetic therapy, which is based upon targeting of the epigenetic mode of action of a dominant oncogene.

Here, we show that the two splicing isoforms of MECP2 activate distinct growth-factor pathways in a manner that recapitulates the major oncogenic functions of activated RAS. The ability of MECP2 to bind the epigenetic mark 5hmC is important for MECP2 to confer anchorage-independent growth. MECP2-overexpressing cell lines derived from human cancers are addicted to MECP2 expression, suggesting MECP2 may be useful as a therapeutic target. In sum, these experiments identify MECP2 as a previously unrecognized oncogene with an unusual epigenetic mode of action that has potential therapeutic implications.

HMECs (16) were transduced with retroviruses expressing SV40 large T, small T, and hTERT (referred to here as “N minus RAS” cells). N minus RAS cells become fully transformed and capable of anchorage-independent growth with the addition of activated RAS (4, 5). To screen for new oncogenes that could substitute for activated RAS, we started with a genome-scale lentiviral expression library (6) containing over 16,000 unique open reading frames (ORF) in an arrayed format. Twenty-one pools of approximately 800 clones each were packaged as vesicular stomatitis virus G pseudotyped lentiviruses. Cells infected with the 21 pools, GFP-infected negative control cells, and HRAS^V12-infected positive control cells were plated in soft agar. After 3 weeks of growth, the positive control HRAS^V12-infected N minus RAS HMEC plates had numerous visible soft-agar colonies, the negative control GFP-infected N minus RAS HMECs had no visible colonies, and plates from 13 of the 21 pools from the ORFeome expression library also contained visible colonies. These colonies were picked and expanded, and DNA was extracted. Library members present in these colonies were identified by PCR using primers flanking the ORF and sequencing. Despite infection at a multiplicity of infection of 0.3 as judged by drug resistance, many of these colonies had multiple proviruses, perhaps because of frequent silencing or low expression of many proviruses (17). To validate the ability of ORFs recovered in this screen to transform N minus RAS HMECs, lentiviruses corresponding to each ORF identified by PCR and sequencing were packaged individually and used to infect N minus RAS HMECs, and growth in soft agar was assessed.

In this manner, this screen identified 15 genes capable of conferring anchorage-independent growth upon N minus RAS cells (Supplementary Table S1). Of these 15 validated hits, 13 were present in only a single colony, 1 was identified in 2 colonies, and 1 was identified in 6 colonies; based on these data, the screen is far from achieving saturation. We scrutinized these 15 genes for evidence of recurrent overexpression or amplification in human cancers by interrogating publicly available databases. One screen hit, OTX2, has been previously identified as an oncogene amplified in childhood medulloblastoma (18–21), and thus served as a proof of concept that this screen could identify oncogenes that participate in transformation when overexpressed in their WT state. Another of the screen hits, MECP2, was selected for further study because its high rate of amplification across the TCGA collection of tumors strongly suggested relevance for human cancer. It resides in an amplicon on the long arm of the X-chromosome (Xq28) that does not contain any known oncogene.

To investigate the prevalence of MECP2 amplification in human cancers, we analyzed 9,221 human tumor samples from the TCGA. Figure 1A shows the overall frequency of amplification of MECP2 across a number of human cancer types in the TCGA collection determined by Genomic Identification of Significant Targets in Cancer (GISTIC; ref. 22). Q-values (FDR-corrected significance of amplification frequency) below 0.25 suggest that MECP2 is significantly amplified above the background rate and that the presence of the amplified locus is enriched by selective pressure. The q-value of the amplicon containing MECP2 across the entire TCGA collection of cancers is 2.4 × 10⁻²⁴, demonstrating clear selective pressure favoring tumor cells containing this amplicon. The significance of amplification across Xq in the entire TCGA dataset of 9,221 cancers and heat maps of amplification for selected cancer types in the TCGA collection are shown in Fig. 1B. The q-value for amplification is most significant in the chromosomal region that contains MECP2 (green dotted line). The region containing MECP2 is the only amplicon on the X chromosome to have a significant q-value across all human cancers (Broad Institute TCGA Copy Number Portal; Supplementary Table S2).

The effect of MECP2 amplification on MECP2 mRNA levels is assessed in Fig. 1C for some cancer types with high rates of MECP2 amplification and in Supplementary Fig. S1A for other cancer types. In cancer types with significant frequencies of MECP2 amplification, copy-number gain of MECP2 is correlated in a highly significant manner with increased MECP2 mRNA level.

MECP2 expression is silenced on the inactive X chromosome in females (23); if the MECP2 amplicon in tumors confers selective advantage because of higher MECP2 expression, the MECP2 allele on the active X chromosome would be amplified in preference to the inactive allele in tumors arising in women. We observed a pattern of amplification that confirmed preferential amplification of the active allele in triple-negative breast cancer (TNBC) and ovarian cancer (Supplementary Fig. S1B).

Based upon the data above, TNBC, lung, and ovarian cancers were chosen for further study. Figure 1D shows Western blot analysis of whole-cell lysates obtained from the cell lines with and without MECP2 amplification representing these cancer types (Supplementary Table S3). We have also assayed patient-derived xenografts (PDX; ref. 24) established from TNBCs for MECP2 overexpression by Western blot; 4 of 13 (31%) consecutive TNBC PDXs overexpressed MECP2 (Fig. 1D), in accordance with the 33% overall frequency of MECP2 amplification in this breast cancer subset in the TCGA collection (Fig. 1A).

The MECP2 gene expresses two splicing isoforms that differ by inclusion of the second exon, resulting in a long isoform, termed e1, that consists of 21 unique amino acids at the amino terminus followed by a 477-amino acid shared region, and a short isoform, e2, that has 9 unique amino acids at the amino terminus attached to the same 477-amino acid shared region. In most cell types, the expression of e1 is higher than that of e2 at the protein level.

The expression library used for our screen contained only the shorter MECP2 splicing isoform e2; infection of N minus RAS cells with a single lentivirus encoding MECP2 e2 induced anchorage-independent growth to an extent similar to that caused by infection with activated HRAS, whereas the longer isoform, e1, was inactive in this regard (Fig. 2A, left). To rule out the possibility that the observed transformation of N minus RAS HMEC cells by MECP2 was attributable to some unique property of N minus RAS HMECs, we constructed N minus RAS cells from an entirely different human primary breast cell type, breast primary epithelial cells (BPEC; ref. 25). MECP2 e2 conferred anchorage-independent growth efficiently upon N minus RAS BPECs as well, and in this cell type, MECP2 e1 was capable of conferring growth in soft agar, albeit at considerably lower efficiency than the short isoform (Fig. 2A, right). Soft-agar colony size in N minus RAS HMECs that express MECP2 e2 was approximately equal to that induced by activated HRAS (Supplementary Fig. S2A), whereas in N minus RAS BPECs, colonies induced by MECP2 e2 overexpression were larger than those induced by activated HRAS (Supplementary Fig. S2B). Lastly, in N minus RAS BPECs, colonies induced by MECP2 e1 overexpression were much smaller than those induced by MECP2 e2 or activated RAS in these cells (Supplementary Fig. S2B).

To determine whether the expression of MECP2 was sufficient to allow growth of N minus RAS cells as tumors in nude mice, a series of N minus RAS cells infected with either of the two splicing isoforms of MECP2, both isoforms together, activated RAS, or a GFP control, were injected into the flanks of nude mice. In accordance with previous results (5, 25), N minus RAS HMECs infected with a retrovirus expressing activated RAS were able to form tumors in nude mice, but N minus RAS HMECs infected with a virus encoding GFP were not. N minus RAS HMECs infected with both the long and short forms (e1 and e2) formed tumors in nude mice (Fig. 2B and 2C), whereas N minus RAS HMECs infected with either isoform alone were unable to form tumors. To investigate the tumor-forming abilities of the MECP2 isoforms in a different cellular context, N minus RAS BPECs were used. N minus RAS BPECs infected with a vector expressing activated RAS are known to be about four orders of magnitude more tumorigenic in nude mice than are N minus RAS HMECs expressing activated RAS (25). In the context of N minus RAS BPECs, each isoform of MECP2 allowed the growth of tumors in nude mice, although N minus RAS BPECs expressing both isoforms had a higher take rate and faster tumor growth (Fig. 2B and D).

The histology of tumors induced by MECP2 isoforms was similar to those induced by HRAS (Supplementary Fig. S3A and S3B). In accordance with previously published results (25), N minus RAS HMECs expressing HRAS gave rise to invasive ductal carcinomas with significant areas of squamous differentiation (Supplementary Fig. S3B), whereas N minus RAS BPECs gave rise to tumors that were morphologically similar, but the latter contained more limited areas of squamous differentiation (Supplementary Fig. S3A). N minus RAS HMECs and BPECs expressing MECP2 displayed histologies similar to those seen with HRAS in the two cell types (Fig. 2E and F); further, in the case of BPECs, there was no histologic difference seen in tumors arising after infection with each MECP2 isoform (Supplementary Fig. S3A).

Several functional domains of MECP2 have been defined. In both MECP2 splicing isoforms, the methyl DNA-binding domain (MBD) lies near the amino-terminal end, followed by a transcription repression domain responsible for binding HDAC complexes (Fig. 2G). Three separate mutations in the MBD region of the MECP2 gene, R106W, R111G, and F155S, completely eliminate DNA binding and cause Rett Syndrome (26–28). These mutations were tested for their transforming ability in the context of the MECP2 e2 isoform. They all prevent the MECP2 e2 isoform from conferring anchorage-independent growth upon N minus RAS HMECs despite expression levels similar to WT MECP2 e2, as does a truncating mutation located in the transcription repression domain (Fig. 2H and I).

The MECP2 mutation, R133C, causes a less severe form of Rett Syndrome than other mutations in the DNA-binding region of MECP2 (29). R133C prevents MECP2 binding to 5hmC, an epigenetic mark associated with actively transcribed genes; however, it largely preserves binding to 5-methylcytosine (5mC; ref. 12). R133C in the context of the MECP2 e2 isoform was able to confer anchorage-independent growth upon N minus RAS HMECs at only about 10% of the efficiency of WT MECP2 e2 despite expression levels equivalent to WT MECP2 e2 (Fig. 2J). This observation suggests that binding to 5hmC, and perhaps the activation of gene expression, is important for the transforming ability of MECP2.

Because MECP2 was isolated in a screen in which it was able to substitute for the transformation function of activated RAS, we tested to what extent MECP2 could substitute for activated RAS in other situations. Certain cultured human cancer cell lines that contain activating KRAS mutations are dependent upon the continued presence of KRAS for growth (“RAS addiction”; refs. 30, 31). We tested whether complementation with exogenous MECP2 could rescue growth and survival defects in such a cell line after downregulation of KRAS expression. As shown in Fig. 3A, the expression of the MECP2 isoforms rescued growth substantially in a dose-dependent fashion after downregulation of KRAS in H358, a KRAS^G12C-addicted lung cancer cell line, using an shRNA targeting the KRAS-3′UTR (untranslated region). This shRNA had been previously validated as not having significant off-target effects in this cell line (designated as “K-RasC” shRNA in ref. 30).

If MECP2 amplification and RAS activating mutation confer similar functions during tumorigenesis, there would be little or no selective advantage conferred by the presence of both alterations in a given tumor. Therefore, there may be fewer tumors that contain both alterations than would be predicted by chance, i.e., there may be a mutual exclusivity relationship between the two alterations. We investigated the relationship of MECP2 amplification to the activation of RAS family members across human tumor types and within individual tumor types. In tumor types with the highest rates of RAS mutation, pancreatic (virtually all KRAS mutated) and colon adenocarcinoma (roughly 50% mutated RAS), there are few tumors with amplification of MECP2 (ref. 22; GISTIC detects no MECP2-containing amplicons in those tumor types). In contrast, the cancer types with the lowest rates of RAS mutation tend to have significant numbers of cancers with MECP2 amplification; for example, ovarian cancer and TNBC both have RAS mutations in less than 1% of tumors, but have high rates of MECP2 amplification, 38% and 33% respectively (Fig. 1A).

Uterine carcinoma demonstrates a different type of mutual exclusivity relationship. In the TCGA set of uterine carcinoma samples, there are high rates of MECP2 copy-number gain (15%) and high rates of KRAS mutation (21%), but statistically significantly fewer cases with both alterations than would be expected by chance (log OR, −1.215; P = 0.027; Fig. 3B). Uterine cancers can be subcategorized by the presence or absence of microsatellite instability (MSI), POLE mutation, and the level of copy-number abnormalities (CNA); in general, tumors with microsatellite instability or POLE mutation have high numbers of point mutations and few CNAs, whereas POLE WT, microsatellite-stable tumors have higher CNAs (32). These subtypes of uterine cancers in the TCGA dataset were shown to vary significantly by MECP2 status (P < 0.0001), such that tumors with MECP2 copy-number gain or amplification were more likely to be in the CN-high subtype and less likely to be in the MSI-positive subtype. Conversely, tumors with KRAS mutation were significantly less likely to be in the CN-high subtype and significantly more likely to be in the MSI-positive or POLE-mutated subtype (P < 0.0001; see Supplementary Data: Analysis of Uterine Cancer Subtypes). Therefore, the subtypes of uterine cancer drive the mutual exclusivity of MECP2 copy-number gain or amplification and KRAS mutation in uterine cancer.

There are also mutual exclusivity relationships between MECP2 copy-number gain and RAS alteration in some cancer types with relatively high rates of both alterations within the same subtype. Unlike uterine carcinoma, cervical cancer does not have readily identifiable subtypes (33). There is no overlap between cases with MECP2 gain or amplification and those with KRAS mutation in the TCGA collection of cervical carcinoma, a statistically significant mutual exclusivity result (log OR < −3; Fisher exact test, P = 0.041; Fig. 3B). In this tumor type, the MAPK pathway is sometimes activated by amplification or mutation of MAPK1 (ERK2; ref. 33). These alterations of MAPK1 are also statistically significantly mutually exclusive with MECP2 amplification (log OR < −3; P = 0.041), and there is no overlap between cases with MAPK1 amplification or mutational activation and cases of MECP2 gain or amplification, or KRAS mutation (Fig. 3B), strongly suggesting that these alterations are functionally redundant with one another. Similarly, head and neck cancer, a group of squamous cell carcinomas, has a high percentage of cases with MECP2 copy-number gain (30%) and some cases with RAS mutation or amplification. In this cancer type, there is a statistically significant mutual exclusivity between MECP2 copy-number gain and the union of all RAS gene mutations and amplifications (log OR, −0.845; Fisher exact test, P = 0.020; Fig. 3B), again suggesting that MECP2 copy-number gain and RAS activity are functionally redundant.

There are other cancer types with trends toward mutual exclusivity of RAS mutation and MECP2 amplification, for example, adenocarcinoma of the lung and bladder carcinoma, but the relatively small percentage of one or the other alteration requires that more cases be subject to genomic analysis to have enough power to determine if these relationships are statistically significant. Taken as a whole, across all human tumor types, within a cancer type with multiple subtypes, and within monomorphic tumor types with relatively high rates of both RAS activation and MECP2 amplification, there is a tendency for tumors to have either RAS family activating mutations or MECP2 amplification, but not both, suggesting functional redundancy.

Given that MECP2 is able to substitute for activated RAS in a transformation assay as well as in a RAS-addicted cell line, and the mutual exclusivity analysis above between MECP2 copy-number gain and RAS alteration in several cancer types, we investigated the possibility that MECP2 isoforms activate the major RAS-induced growth factor pathways, the MAPK pathway and PI3K pathway. To investigate the state of MAPK pathway activation, the persistence of MAPK signaling after growth factor deprivation was assessed, a standard technique employed to determine if there is cell-intrinsic activation of this pathway (34, 35). The MECP2 short isoform e2, but not the long isoform e1, prolonged the presence of phosphorylated ERK1 and ERK2 in N minus RAS HMECs after growth factor deprivation (Fig. 4A; Supplementary Fig. S4). This effect required an intact MECP2 DNA-binding domain (Fig. 4B). To determine at what level along the signaling cascade the MECP2 e2 isoform activates the MAPK pathway, a CRAF affinity precipitation assay (36) was used to investigate whether MECP2 increases the amount of active, GTP-bound RAS. Figure 4C, top, shows that MECP2 does not increase the amount of activated, GTP-bound RAS, suggesting that the MAPK pathway is activated by MECP2 at a level below RAS in the signaling pathway. Increased MEK phosphorylation was detected in MECP2 e2–transformed cells, as was increased phosphorylation of the downstream MAPK pathway proteins p90RSK and ELK1 (Fig. 4C, bottom). Together, these results indicate that the short isoform of MECP2 activates the MAPK signaling pathway below RAS, but at or above the level of the MEK proteins.

To assess PI3K pathway activation, cells were deprived of growth factors and persistence of PI3K signaling monitored (Fig. 4D, top); or deprived of growth factors for 4 hours, and then stimulated with recombinant EGF for 15 minutes, and re-induction monitored (Fig. 4D, bottom), a standard assay of PI3K pathway activation (refs. 37–42; a timecourse of EGF treatment is shown in Supplementary Fig. S5). PI3K signaling as assessed by pAKT^S473 is sustained after 5 minutes of growth factor deprivation in N minus RAS cells infected with lentiviruses expressing either MECP2 isoform or activated RAS, but not in the same cells infected with a lentivirus expressing GFP. In addition, EGF treatment of MECP2- or RAS-expressing N minus RAS HMECs compared with the same cells transduced with a vector expressing GFP shows increased levels of pAKT and activated downstream PI3K pathway proteins p4EBP1 and pS6. These observations are consistent with prior reports of experiments with neuronal cells, demonstrating that the MECP2 long isoform e1 stimulated the PI3K pathway (43, 44). In neurons, PI3K pathway induction by MECP2 has been shown to involve increased brain-derived neurotrophic factor (BDNF) expression (44, 45) and/or the regulation of IGF1 signaling (46, 47).

Oncogene addiction, the need for continued expression of an oncogene for maintenance of the malignant phenotype (48), is necessary for therapy directed at that oncogene to be effective. The protumorigenic function of MECP2 could be necessary only during the early steps of transformation, or MECP2 could be required on an ongoing basis to exert its pro-oncogenic function. To distinguish between these possibilities, three non–small cell lung cancer (NSCLC) cell lines and three breast cancer cell lines were selected for further study. The NSCLC cell lines, NCI-H1755 and NCI-H23, both have copy-number gain of MECP2 and exhibited appreciable MECP2 expression, whereas NCI-H1437 did not express detectable MECP2 as assessed by Western blot (Fig. 1D). The two NSCLCs expressing MECP2 demonstrated significant growth inhibition in response to infection, with lentiviruses bearing either of two MECP2-directed shRNAs. However, the cell line that did not express MECP2, NCI-H1437, did not reveal significant growth inhibition (Fig. 4E). Similarly, two human breast cancer cell lines that have MECP2 copy-number gain and expressed significant amounts of MECP2, MDAMB468 and BT549 (Fig. 1D), revealed considerable growth inhibition when infected with either of two lentiviruses expressing shRNA directed at MECP2, whereas the low-expressing breast cancer cell line ZR75-1 showed little growth inhibition (Fig. 4E). A number of other cell lines with high MECP2 protein expression (Fig. 1D), including the breast cancer cell line BT20 and the lung cancer cell lines NCI-H2228 and NCI-H522, also show growth inhibition when MECP2 expression is inhibited (Supplementary Fig. S6).

To determine if the shRNA effects seen in Fig. 4E are attributable to off-target effects, the lung cancer line NCI-1755 was infected with lentiviruses encoding GFP, the cDNA of the MECP2 short isoform e2, the cDNA of the MECP2 long isoform e1, or viruses encoding the cDNA of both isoforms. These cell lines were then infected with a lentivirus expressing an shRNA targeting the 3′UTR of MECP2, which downregulates both endogenous isoforms of MECP2 (Fig. 4E), but would not be expected to downregulate expression of the cDNAs. Figure 4F shows that the long isoform of MECP2 did not rescue the inhibition of proliferation of NCI-H1755 caused by shMECP2-3′UTR, the short isoform partially rescued proliferation, but together, both isoforms almost completely rescued the proliferation inhibition caused by shRNA-mediated downregulation of both endogenous MECP2 isoforms, showing that the proliferative defects caused by this shRNA in Fig. 4E are not a result of off-target effects.

We sought to determine if activated RAS could substitute for MECP2 in an MECP2-addicted cell line. We found that expression of activated KRAS^G12V in the MECP2-addicted TNBC cell line MDAMB468 substantially rescued the growth inhibition seen upon downregulation of MECP2 (Fig. 4G). Thus, MECP2 can rescue the growth of a cell line addicted to activated RAS after RAS downregulation (Fig. 3A), and activated RAS can rescue the growth of a cell line addicted to MECP2 after MECP2 downregulation (Fig. 4G), demonstrating the complementary relationship between RAS activity and MECP2 function.

A genome-scale screen identified MECP2 as a gene that substitutes for activated RAS to allow anchorage-independent growth. MECP2 is frequently amplified in a number of human tumor types, and many cell lines derived from human tumors have MECP2 amplification and overexpression. Expression of MECP2 rescues growth of a human tumor line addicted to activated RAS after downregulation of RAS expression, and activated RAS rescues an MECP2-addicted cell line after MECP2 downregulation. MECP2 induces the MAPK and PI3K growth factor signaling pathways in common with activated RAS. MECP2 requires DNA binding for growth factor signaling and transformation, and is heavily dependent on binding to the epigenetic modification 5-hydroxymethylcytosine for these activities. In one cellular context, the two splicing isoforms of MECP2 have quite distinct activities despite substantial sequence identity, with the shorter e2 splicing isoform allowing anchorage-independent growth and activating both the MAPK and PI3K pathways, and the longer e1 isoform activating the PI3K pathway but not enabling anchorage-independent growth on its own. Together, the two splicing isoforms allowed efficient growth of xenografts in nude mice. These findings indicate that MECP2 may function as an oncogene with an unusual epigenetic mechanism of action across a substantial number of human tumors.

MECP2 is not an escape gene; it is silenced on the inactive X chromosome in females (23). The amplification of this class of X-linked genes affects gene expression more potently than amplification of other genes on a per-copy basis. This effect results from the fact that in males, and after X-inactivation in females, there is only one active copy of this class of genes. For example, on average, one extra active copy of such an X-linked gene results in 200% of the normal level of expression, whereas in comparison, a biallelically expressed autosomal gene with one extra copy is expressed at 150% of normal levels. Three extra active copies of an X-linked gene would be expected to result in 4× the normal level of expression, whereas an autosomal gene with three extra copies has only 2.5× the normal level of expression. Thus, relatively modest amplification of MECP2 that affects the active allele may have a relatively large effect on expression levels.

Our screen linked WT MECP2 overexpression to transformation. Inherited loss-of-function mutations in MECP2 are known to be responsible for the autism spectrum disorder Rett Syndrome (7). Though this relationship may seem surprising, the explanation may lie in connections between alterations in growth factor pathways and autism that are becoming increasingly apparent. A number of germline alterations that either increase or decrease growth factor pathway activity predispose to autism (reviewed in ref. 49). For example, there is a significant overrepresentation of autism among patients with Beckwith–Wiedemann Syndrome (in which patients overexpress IGF2 or cyclin-dependent kinase inhibitor 1C; ref. 50), in neurofibromatosis (NF1; refs. 51, 52), in the PTEN hamartoma tumor syndromes Cowden Syndrome and Bannayan–Riley–Ruvalcaba (53–57), and in tuberous sclerosis (58), and SNPs for three PI3K signaling pathway genes, INPP1, PIK3CG, and TSC2, are in linkage disequilibrium with autism (59). A cMET promoter SNP that decreases cMET expression is associated with a 2-fold increase in autism risk (60). Mutations in the RAS pathway scaffolding protein CNKSR2 are associated with an autism-like disorder (61). Finally, treatment with recombinant IGF1, which crosses the blood–brain barrier and activates growth factor signaling, has been shown to have therapeutic efficacy in a mouse model of Rett Syndrome (62) and in a phase I clinical trial treating girls with MECP2 mutations (63).

It is not clear how defects in MECP2 lead to Rett syndrome. Given the connection between autism spectrum disorders and growth factor pathway alterations, and our demonstration that the two splicing isoforms of MECP2 each contribute to the activation of growth factor pathways, the work described here may indirectly lead to further hypotheses regarding the role of abnormal growth factor pathway activation in patients with autism spectrum disorders. Further, it is tempting to speculate that the MECP2 isoform–specific differences in growth factor pathway induction demonstrated here may have arisen to titrate and balance the activity of different growth factor pathways by the process of alternative splicing in neurons. Because the growth factor pathways induced by MECP2 activity, the MAPK and PI3K pathways, are key pathways in human tumors, it seems likely that cells overexpressing MECP2 by amplification or other means would have a selective advantage during tumorigenesis.

Experiments presented here demonstrate that several human tumor cell lines with overexpressed MECP2 show significant growth impairment after MECP2 downregulation by shRNA. These observations suggest that patients with tumors overexpressing MECP2 might benefit from therapy targeting MECP2 function. Further, the necessity of DNA-binding activity of MECP2 for transformation and growth factor pathway induction suggests that interfering with the ability of MECP2 to bind DNA might have therapeutic effect. Because MECP2 binds specifically to methylated or hydroxymethylated cytosine, blocking the formation of these modified cytosines might specifically inhibit MECP2-transformed cells. The FDA-approved drugs 5-azacytidine and decitabine inhibit DNA methylation and therefore inhibit the formation of 5mC and 5hmC, and so may be therapeutic in tumors with overexpressed MECP2. Experiments are ongoing to test this hypothesis.

Data presented here implicate the binding of MECP2 to 5hmC as important for the cancer-related functions of MECP2. 5hmC is formed from 5mC by the action of the TET enzymes (64, 65). TET2 is a tumor suppressor in some settings; TET2-inactivating mutations occur commonly in myelodysplasia and myeloid malignancies. Further, neomorphic oncogenic mutations have been found in the enzymes IDH1 or IDH2 in glioblastoma, acute myeloid leukemia, chondrosarcoma, and cholangiocarcinoma that result in the accumulation of (R)-2-hydroxyglutarate (66), an abnormal metabolite that inhibits the TET enzymes, as well as other 2-oxoglutarate–dependent dioxygenases. However, TET1 also possesses oncogenic activity (67). Given the dependence of the MECP2 transformation activity on 5hmC, it is possible that inhibition of the TET enzymes may inhibit the growth of MECP2-related tumors; this inhibition may be accomplished by compounds similar to (R)-2-hydroxyglutarate (68). Further insight into the mechanisms MECP2 uses to drive transformation may reveal additional therapeutic targets.

MECP2 has several properties that make it an unusual oncogene. First, its two splicing isoforms cooperate in tumor formation, with MAPK pathway induction contributed uniquely by the short isoform. Second, its mechanism of action requires the epigenetic modification of cytosine, including the recently discovered modification, 5hmC. This property may provide unusual therapeutic opportunities. Lastly, MECP2 is a member of a small class of genes that are involved in two entirely different human diseases. When mutated, MECP2 causes Rett Syndrome, and when amplified/overexpressed, it is involved in cancer.

Details of lentiviral and retroviral expression vectors, shRNA vectors (with target sequences), construction of mutants (including primer sequences) and cloning are provided in the Supplementary Methods section.

Virus particles were produced by transient cotransfection of 293T cells using a 3-plasmid system (for retrovirus) or 5-plasmid system (for lentivirus) as detailed in the Supplementary Methods section. Further information on transduction of target cells with cDNA viruses for overexpression studies and shRNA viruses for knockdown studies is described in related subsections of Supplementary Methods.

The HMECs were purchased from Lonza, and the BPECs were a gift from Dr. Tan Ince (University of Miami, Coral Gables, FL) in January 2010; no authentication of these cells has been done by the authors. The BT549, MDAMB468, ZR75-1, BT20, HCC38, MCF7, MDAMB231, MDAMB453, and T47D cell lines were purchased from the ATCC, and no authentication has been done by the authors. The NCI-H1755, NCI-H23, NCI-H1437, NCI-H1395, and NCI-H2347 cell lines from the Minna and Gazdar laboratories (The University of Texas Southwestern Medical Center, Dallas, TX) were provided to the Belfer Institute/Dana-Farber through the Meyerson Lab at the Broad Institute (with permission from the originators) and were obtained by the authors in November 2012. The NCI-H2228, NCI-H522, NCI-H1563, NCI-H2170, and NCI-H358 cell lines were purchased from the ATCC, and the NCI-H2009 and NCI-H1993 cell lines were obtained from the Belfer Institute in November 2012. All of these lung cancer cell lines have been short-tandem repeat (STR)–profiled and Mycoplasma-tested at the Belfer Institute. The EFO21, IGROV1, SKOV3, CAOV3, OVCAR3, OVCAR4, OVCAR5, and OVCAR8 cell lines were gifts from Dr. Ronny Drapkin (University of Pennsylvania, Philadelphia, PA) in March 2013 and have been STR-fingerprinted at the Drapkin laboratory. All the cell lines used for the experiments described in this article were used from early passages (for less than 6 months) of cell stocks frozen at the time of receipt. All cell lines were maintained at 37°C with 10% CO₂. Details about culture conditions, media composition, and generation and maintenance of stable cell lines are described in the Supplementary Methods section.

Anchorage-independent growth of cells in soft agar was determined by plating 3 × 10⁴ cells (HMEC or BPEC derivatives) in 0.3% Noble agar (Difco). Colonies were counted after 3 weeks of growth. Xenograft study was carried out in NCr nude mice (Taconic), and the study protocol was approved by the Dana-Farber Institutional Animal Care and Use Committee. For further details on both of these assays, see Supplementary Methods.

Cell proliferation was quantified by either measuring cellular ATP content by CellTiter-Glo Luminescent Cell Viability Assay (Promega) or by extracting the cell-associated crystal violet dye with 10% acetic acid.

Cells were lysed in RIPA buffer, and Western blot was carried out using standard protocol. For assays involving MAPK pathway activity, HMEC derivatives were washed three times with PBS, reset in growth factor–deprived medium (MEBM) at 37°C for 5 minutes or longer, and then lysed (see Supplementary Methods for details). For assays involving PI3K pathway activity, HMEC derivatives were washed three times with PBS, starved in MEBM for 5 minutes and analyzed, or starved for 4 hours, treated with EGF (2.5 ng/mL; Sigma) at 37°C for 15 minutes, and then lysed. Primary antibodies used for immunoblotting are listed in Supplementary Methods.

SNP6 array data generated by TCGA from 9,221 cancers across 29 cancer types were analyzed for significantly recurrent amplifications using GISTIC 2.0 (69) and according to the methods described in ref. 70. Further details about the samples, analysis parameters, and detailed results can be found in the Broad Institute TCGA Copy Number Portal, under the analysis version “2014-07-08 stddata_2014_06_14.” For details of the expression versus copy number and allelic imbalance analyses, please see Supplementary Methods.

M. Neupane is an inventor on a Dana-Farber Cancer Institute patent application on the uses of MECP2. R. Beroukhim reports receiving commercial research support from Novartis and is a consultant/advisory board member for the same. D.E. Hill is an inventor on a Dana-Farber Cancer Institute patent application on the uses of MECP2. D.P. Silver is an inventor on a Dana-Farber Cancer Institute patent application on the uses of MECP2. No potential conflicts of interest were disclosed by the other authors.

Conception and design: M. Neupane, D.P. Silver

Development of methodology: M. Neupane, D.E. Hill, D.P. Silver

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Neupane, A.P. Clark, N.J. Birkbak, E. Lim, R. Beroukhim, M. Vidal, D.E. Hill

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Neupane, A.P. Clark, N.J. Birkbak, A.C. Eklund, A.C. Culhane, W.T. Barry, S.E. Schumacher, R. Beroukhim, Z. Szallasi, D.E. Hill, D.P. Silver

Writing, review, and/or revision of the manuscript: M. Neupane, A.P. Clark, S. Landini, E. Lim, R. Beroukhim, Z. Szallasi, M. Vidal, D.E. Hill, D.P. Silver

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Landini

Study supervision: D.P. Silver

The authors thank A. Richardson for evaluating the xenograft pathology specimens, A. MacWilliams and T. Hao for help with the lentivirus orfeome library, T. Joshi for bioinformatics analysis, and D. Livingston and members of the Livingston lab for discussions and advice. They also thank all of the investigators, institutions, and patients who contributed to TCGA research network; the results published here are in part based upon data generated by the TCGA research network.

This work was supported by a grant from the Cogan Family Foundation (to D.P. Silver), Dana-Farber Cancer Institute (DFCI) startup funds from philanthropic sources (to D.P. Silver), the DF/HCC SPORE in breast cancer NIH P50 CA168504-01A1, a U01HG001715 (to M. Vidal and D.E. Hill), DFCI Institute Sponsored Research (to M. Vidal and D.E. Hill), and the Ellison Foundation, Boston, MA (to M. Vidal and D.E. Hill).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.