p53 cooperates with DNA methylation and a suicidalinterferon response to maintain epigenetic silencingof repeats and noncoding RNAsKaterina I. Leonovaa, Leonid Brodskyb, Brittany Lipchicka, Mahadeb Pala, Liliya Novototskayaa, Alex A. Chenchikc,Ganes C. Send, Elena A. Komarovaa, and Andrei V. Gudkova,e,1

aDepartment of Cell Stress Biology, Roswell Park Cancer Institute, Buffalo, NY 14263; bTauber Bioinformatics Research Center, University of Haifa, Haifa31905, Israel; cCellecta, Inc., Mountain View, CA 94043; dDepartment of Molecular Genetics, Lerner Research Institute, Cleveland Clinic, Cleveland, OH 44195;and eTartis, Inc., Buffalo, NY 14203

Edited* by George R. Stark, Lerner Research Institute, Cleveland, OH, and approved November 15, 2012 (received for review September 28, 2012)

Large parts of mammalian genomes are transcriptionally inactiveand enriched with various classes of interspersed and tandemrepeats. Here we show that the tumor suppressor protein p53cooperates with DNA methylation to maintain silencing of a largeportion of the mouse genome. Massive transcription of majorclasses of short, interspersed nuclear elements (SINEs) B1 and B2,both strands of near-centromeric satellite DNAs consisting of tandemrepeats, and multiple species of noncoding RNAs was observed inp53-deficient but not in p53 wild-type mouse fibroblasts treatedwith the DNA demethylating agent 5-aza-2’-deoxycytidine. Theabundance of these transcripts exceeded the level of β-actin mRNAby more than 150-fold. Accumulation of these transcripts, whichare capable of forming double-stranded RNA (dsRNA), was accom-panied by a strong, endogenous, apoptosis-inducing type I IFNresponse. This phenomenon, which we named “TRAIN” (for “tran-scription of repeats activates interferon”), was observed in sponta-neous tumors in two models of cancer-prone mice, presumablyreflecting naturally occurring DNA hypomethylation and p53 in-activation in cancer. These observations suggest that p53 and IFNcooperate to prevent accumulation of cells with activated repeatsand provide a plausible explanation for the deregulation of IFNfunction frequently seen in tumors. Overall, this work reveals rolesfor p53 and IFN that are key for genetic stability and thereforerelevant to both tumorigenesis and the evolution of species.

interferon alpha-beta receptor | RNA sequencing | epigenetic repression |microarray hybridization

Mammalian genomes contain an abundance of noncodingDNA sequences as well as multiple classes of interspersed

repetitive elements such as DNA transposons and retrotransposons.They vary in length and copy number, are evolutionarily youngerthan structural genes, and are generally considered “genomicjunk” with uncertain physiological roles (1). Among these, themost abundant are short, interspersed nuclear elements (SINEs)that originate from reverse-transcription products of certain RNApolymerase III-driven transcripts (e.g., 7SL RNA and tRNAs) (2)and have been shown to retain some ability to retrotranspose (3,4). In addition, the mammalian genome also contains tandemlyorganized repeats known as “satellite DNAs” (5). These highlyreiterated sequences are the main component of the centromericand pericentromeric heterochromatin and are considered im-portant structural elements of the chromosome (6). Together,these classes of noncoding DNA sequence elements occupymore than 50% of the human and mouse genomes (1).Analysis of the phylogeny of SINEs suggests that they accu-

mulated through multiple “explosions,” or bursts of amplification,which started about 65 million years ago and were intermitted byperiods of dormancy (7, 8). The explosive nature of multiplicationof SINEs and the fact that they all are located predominantly inintergenic areas, pseudogenes, and noncoding regions of genessuggest that their bursts of amplification were genetic catastrophes

associated with massive loss of those genomes in which SINEsinserted within essential genes (2, 9, 10) and presumably con-tributed to the diversity of mammalian species. In fact, integrationof new SINE copies into coding or regulatory sequences has thepotential to disturb gene expression (11).Therefore, it is logical to assume that maintenance of the

genetic stability of mammalian cells and organisms depends ontheir ability to prevent the expression and amplification of SINEs.In fact, SINEs normally are transcribed not as distinct autono-mous elements but rather as integral parts of noncoding regionsof larger protein-coding transcripts (2). The great majority ofSINEs are located in heavily methylated regions of DNA and arebelieved to be epigenetically silenced through the mechanism ofDNA methylation (12, 13).However, under certain circumstances both retroelements and

some classes of satellite DNA can be transcribed. The eventstriggering such activation are not well understood but have beenshown to include a variety of stresses such as heat shock (14, 15),viral infections (16), various DNA-damaging treatments (17),and inhibition of translation (14). In addition, transcription ofboth retroelements and satellite DNA was observed in tumors ina variety of models of mouse and human origin (18). Both themechanisms underlying the transcriptional activation of normallysilent noncoding elements and repeats in stressed and trans-formed cells and the physiological consequences of this activa-tion remain poorly understood.Here we show that the tumor suppressor p53, known as a

positive and negative regulator of transcription, plays a majorrole, along with DNA methylation, in epigenetic silencing ofmajor classes of retroelements and satellite DNA in mice. In theabsence of functional p53, DNA hypomethylation results in amassive transcription of several classes of normally silent repeatsand in the accumulation of new, largely double-stranded RNAspecies, which are followed by a cytotoxic type I IFN response.These observations suggest that p53 and IFN cooperate throughregulation of transcription and cell death, respectively, to pre-vent the accumulation of cells with “unleashed” repeats. Thisphenomenon, which we have named “TRAIN” (for “transcrip-tion of repeats activates interferon”), provides a plausible expla-

Author contributions: K.I.L., G.C.S., and A.V.G. designed research; K.I.L., L.B., B.L., M.P.,L.N., and A.A.C. performed research; G.C.S. contributed new reagents/analytic tools; K.I.L.,L.B., M.P., A.A.C., E.A.K., and A.V.G. analyzed data; and K.I.L. and A.V.G. wrote the paper.

Conflict of interest statement: A.V.G. is a consultant and shareholder of Tartis, Inc., a bio-tech company that provided funding for this work.

*This Direct Submission article had a prearranged editor.

Freely available online through the PNAS open access option.1To whom correspondence should be addressed. E-mail: andrei.gudkov@roswellpark.org.

See Author Summary on page 22 (volume 110, number 1).

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1216922110/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1216922110 PNAS | Published online December 10, 2012 | E89–E98

MED

ICALSC

IENCE

SPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 26

, 202

1

nation for a number of well-known but never previously connectedproperties of tumor cells, including transcription of repeats andfrequent deregulation of IFN function.

ResultsDNA Demethylation Can Be Lethal for p53-Null, But Not for p53-WTCells. In 2004, Nieto et al. (19) reported that in vitro treatmentwith the DNA-demethylating agent 5-aza-2’-deoxycytidine (5-aza-dC) induced apoptosis in mouse embryonic fibroblasts(MEFs) lacking p53 but not in those with WT p53. We confirmedand extended this observation (Fig. 1 A and C) by showing thatMEFs, irrespective of the sex of the embryo (Fig. S1), becomehighly sensitive to 5-aza-dC following either shRNA-mediatedknockdown of p53 or ectopic expression of a dominant-negativep53 mutant, GSE56 (20) (Fig. 1 B and C). The latter result in-dicates that sensitization to 5-aza-dC does not require thephysical absence of p53 but only its functional inactivation. Highsensitivity to a DNA-demethylating agent is not limited to p53-deficient fibroblasts and also is the property of fibroblasts andepithelial cells isolated from different tissues of adult mice (Fig.1D). This phenomenon was observed only in proliferating p53-null cells (Fig. S2) and required at least 72 h of 5-aza-dC treat-ment, reflecting the time needed for the compound to exert itsDNA-demethylating activity fully (19). Death of 5-aza-dC–trea-ted p53-null cells was associated with activation of caspases 3 and7, indicating involvement of apoptosis (Fig. 1E).A potential trivial explanation for the observed phenomenon

would be that 5-aza-dC is inefficient as a demethylating agent incells with functional p53 because of their ability to undergogrowth arrest following DNA damage [known to be caused by5-aza-dC (19)]. This possibility was ruled out by our findings of (i)complete depletion of free DNA methyltransferase I (DNMT1)protein in both p53-WT and p53-null cells after 5-aza-dC treat-ment (Fig. 1F), and (ii) similar degrees of DNA demethylation inthe genomic DNA of p53-WT and p53-null cells following 5-aza-dC treatment as evidenced by similar patterns of DNA digestionby the methylation-sensitive restriction enzyme McrBC (Fig. 1G).In summary, sensitivity to 5-aza-dC–induced apoptosis is a com-

mon property of p53-deficient mouse cells regardless of theirtissue of origin or the mechanism of p53 inactivation (knockout,knockdown, or functional inactivation). The selective toxicity top53-null cells differentiates 5-aza-dC from other DNA damagingagents, which generally are more toxic to p53-WT cells (21).Indeed, doxorubicin treatment induced higher levels of caspases3 and 7 in p53-WT than in p53-null MEFs (Fig. 1E). Therefore,the selective toxicity of 5-aza-dC to p53-null cells is likely asso-ciated with its demethylating activity rather than with its DNAdamaging activity.

Different Subsets of Genes Are Activated by 5-aza-dC in p53-Null andp53-WT MEFs. It is believed that many potentially functional ge-netic elements in mammalian genomes are transcriptionally re-pressed by methylation of CpG dinucleotides in the vicinity oftheir promoters (22). Derepression of some genes regulated inthis way (e.g., tumor suppressors p16, PTEN, and Arf, amongothers) can have a detrimental effect on cell growth or even becytotoxic (23). Therefore, we hypothesized that the hypersensi-tivity of p53-null cells to 5-aza-dC might be caused by the acti-vation of transcription of some “killer” genes that normally arerepressed by the combined action of a DNA methylation-medi-ated mechanism and the transrepressor function of p53. Toidentify these putative killer genes, we compared sets of tran-scripts activated by 5-aza-dC treatment in p53-WT and p53-nullMEFs shortly before the onset of the toxic effect of the drug (i.e.,at 48 h) using microarray (Illumina MouseWG-6 v2.0) hybrid-ization-based global gene-expression profiling. The results of thiscomparison are summarized in Fig. 2.

Treatment of p53-WT and p53-null MEFs with 5-aza-dC ledto activation (by at least fivefold) of 55 and 124 genes, re-spectively (Tables S1 and S2). Surprisingly, there were no genesshared between these two lists (Venn diagram, Fig. 2A). To in-vestigate the reason underlying this difference, we compared thebasal expression levels of genes identified as 5-aza-dC–induciblein either cell line in untreated p53-WT and p53-null MEFs. Wefound that for the majority of genes induced by 5-aza-dC in p53-

ac�n

genotype

p53+ + + +

A

B

C

D

0 1 5 10 15 20

WT

0 1 5 10 15 20

WT

0

20

40

60

80

100

120

140

0 1 5 10 15 20

WT

WT

0

20

40

60

80

100

120

0 0.5 1 2 4

WT

shp53GSE56

genotype

DNMT1ac�n

G

E F

500

10001500

4000

12000bp

Fig. 1. Primary cells deficient in functional p53 are hypersensitive to 5-aza-dC treatment. (A) p53-WT and p53-null MEFs were treated with 0, 5, or10 μM 5-aza-dC for 120 h. Viable cells were visualized by methylene bluestaining. (B) Western blot detection of p53 and β-actin (loading control)proteins in p53-WT MEFs (lanes 1 and 2), p53-null MEFs (lanes 3 and 4), andp53-WT MEFs expressing shRNA against p53 (shp53, lanes 5 and 6) or thep53-inactivating genetic suppressor element-56 (GSE56, lanes 7 and 8) leftuntreated (−) or treated (+) with 500 nM doxorubicin (Dox) for 16 h. (C)Cytotoxicity of 5-aza-dC in cells described in B. Cells were treated with theindicated concentrations of 5-aza-dC for 5 d. Viability was determined bymethylene blue staining and extraction, followed by spectrophotometricquantification. Percent viability is shown relative to control cells treated with0.1% DMSO. In all relevant figures, error bars show SDs for assays performedin triplicate. (D) Cytotoxicity of 5-aza-dC (5 d in vitro treatment) in primarycells from different tissues of adult p53-WT and p53-null mice was de-termined as in C. (E) Caspase-3,7 activity (cleavage of the fluorescent sub-strate AC-Devd-AMC) in p53-WT and p53-null MEFs treated with theindicated concentrations of 5-aza-dC for 48 h or 1 μM doxorubicin (Dox) for16 h. (F) Western blot analysis of DNMTI and β-actin (loading control) proteinlevels in p53-WT and p53-null MEFs treated with the indicated concen-trations of 5-aza-dC for 48 h. (G) The overall extent of genomic DNAmethylation in p53-WT and p53-null MEFs left untreated (U) or treated (T)with 10 μM 5-aza-dC for 48 h was determined by digestion of DNA with themethylation-sensitive restriction enzyme McrBC, which cuts only its sites thatare methylated.

E90 | www.pnas.org/cgi/doi/10.1073/pnas.1216922110 Leonova et al.

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 26

, 202

1

WT MEFs, the induced level of RNA expression reached onlythe basal level observed in untreated p53-null cells (Fig. 2A, Left).Therefore, in the absence of p53, these genes are not suppressedby DNA methylation, and thus p53 has a role as a driver of DNAmethylation-mediated gene silencing.Transcripts up-regulated by 5-aza-dC treatment in p53-null

cells demonstrated a completely different pattern of behavior.With a few exceptions, they all remained silent (expressed at low/undetectable levels) in p53-WT MEFs regardless of 5-aza-dCtreatment but were strongly up-regulated by 5-aza-dC in p53-nullcells (Fig. 2A, Right). The literature indicated that most of thesegenes are known transcriptional targets lying downstream of typeI IFNs (IFN-α and IFN-β) (Table S2). Moreover, one gene in thegroup with the highest degree of induction encoded mouse IFN-β1 (Fig. 2B and Table S2). The microarray results were validatedusing semiquantitative RT-PCR analysis of several representa-tive genes from the identified subsets (Fig. 2C). Taken together,these data confirm association of p53 deficiency with (i) lack ofsilencing (high basal expression) of genes that are transcribed inp53-WT cells only after 5-aza-dC treatment and (ii) induction ofa strong IFN response by 5-aza-dC.

DNA Demethylation in p53-Null Cells Triggers a Lethal IFN Response.Production of type I IFNs is a major antiviral response that limitsthe infectivity of a wide range of DNA and RNA viruses (24). Inaddition to this major function, type I IFNs also are involved incomplex interactions with a wide range of biological outcomes,including cytostatic and cytotoxic effects (24). Therefore, because5-aza-dC induced a strong IFN response in p53-null (but not inp53-WT) cells, we set out to test the functionality of the inducedIFN response and determine whether it is involved in the hy-persensitivity of p53-null cells to the drug.

As a measure of the functionality of the IFN response, weassessed the ability of MEFs to support infection with an IFN-sensitive virus, vesicular stomatitis virus (VSV) (25). Treatmentwith 5-aza-dC strongly repressed virus replication only in p53-null cells (Fig. S3), thus confirming the functionality of the IFNresponse in these cells.To determine whether the 5-aza-dC–induced IFN response

plays a role in the toxicity of the drug toward p53-null cells, weused MEFs from IFN receptor 1 (ifnar) knockout mice (26),which are incapable of developing a type I IFN response. Weintroduced into these cells the constructs described above thateither knock down (shRNA) or inactivate (GSE56) p53 (Fig.3A). Although these methods of eliminating p53 activity madep53-WT MEFs with intact IFNAR sensitive to 5-aza-dC–medi-ated killing, they did not sensitize p53-WT, ifnar−/− MEFs (Fig. 3B and C). No signs of caspase activation were detected in ifnar−/−

MEFs treated with 5-aza-dC (Fig. 3D). These observations sug-gest that the IFN response is a major mediator of death inducedby 5-aza-dC in p53-deficient cells.Because DNA demethylation is expected to induce new tran-

scripts, and the IFN response can be triggered by virus-like (i.e.,double-stranded) RNA species, we tested the IFN-inducing ca-pacity of RNA isolated from p53-WT and p53-null cells, un-treated or treated with 5-aza-dC. After removal of rRNA (toreduce high background levels of abundant and largely double-stranded and abundant RNA species), RNA samples were trans-fected into SCCVII cells (a mouse head-and-neck cancer–derivedcell line known to retain normal type I IFN response), and ex-pression of the IFN-responsive protein p49 (27) was evaluated byWestern blotting. Transfection of RNA isolated from 5-aza-dC–treated p53-null cells resulted in dramatically higher induction ofp49 expression than seen in RNA from 5-aza-dC–treated p53-

0.1

1

10

100

1000

IFN

B1CX

CL10

CXCL

9M

X1IR

GB1

0KR

T16

POU

3F1

CCL4

MX2

OAS

L1CD

274

P2RX

7IF

I47

CCL5

PRF1

RHBD

L2IS

G20

CCRL

2TY

KIIF

I205

DPEP

3CA

SP1

GBP

6IL

15 SCT

HAM

P2SE

CT…

IRF7

0100020003000400050006000700080009000

CXCL

10

MX2

IRF7

OAS

L1

IFI4

7

IRG

B10

0500

1000150020002500300035004000

MRG

PRF

ZCCH

C5

AQP1

NRN

1

SCX

ENO

3

WT55 124

A

B C

Fig. 2. Illumina microarray-based analysis of gene expression in p53-WT and p53-null MEFs left untreated or treated with 10 μM 5-aza-dC for 48 h. (A)(Center) Venn diagram showing no overlap between the 55 and 124 genes up-regulated (≥fivefold) by 5-aza-dC in p53-WT (Left) and p53-null (Right) cells. Bargraphs show mRNA expression levels (signal intensity on Illumina microarray) of genes identified as 5-aza-dC–induced in p53-WT MEFs (Left) or in p53-nullMEFs (Right) in all four samples (p53-WT and p53-null MEFs left untreated or treated with 5-aza-dC). (B) Fold-induction (log scale; 5-aza-dC–treated relative tountreated) of a subset of genes identified as 5-aza-dC–induced in either p53-WT MEFs or p53-null MEFs. Note the scale of induction of IFN-β1 in p53-null MEFs.(C) Validation of Illumina microarray gene expression analysis was performed by RT-PCR using independently isolated RNA from p53-WT and p53-null MEFsleft untreated or treated with 10 μM 5-aza-dC for 48 h and specific primers for mouse IFN-β1, H2-Q6, IRF7, and β-actin (control).

Leonova et al. PNAS | Published online December 10, 2012 | E91

MED

ICALSC

IENCE

SPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 26

, 202

1

WT cells or untreated cells of either genotype (Fig. 3E). Thisresult suggests the presence of an IFN-inducing RNA speciesspecifically in p53-null cells with reduced DNA methylation.

Massive Transcription of Repetitive Elements in p53-Null Cells Treatedwith 5-aza-dC. IFN responses normally are triggered by dsRNAinterpreted by the cell as an indication of viral infection (24).The gene-expression profiling that we performed using the Illu-mina MouseWG-6 v2.0 microarray assayed only protein-codingmRNA transcripts and did not reveal any candidate triggers forthe suicidal IFN response observed in 5-aza-dC–treated p53-nullcells. Therefore, we used a high-throughput RNA sequencingtechnique to build a comprehensive, unbiased picture of RNAspecies induced by 5-aza-dC treatment. Total RNA was isolatedfrom p53-WT and p53-null MEFs left untreated or treated for48 h with 5-aza-dC. After depletion of rRNA, the RNA sampleswere fragmented, converted to double-stranded cDNA usingrandom primers, ligated with adaptors, and sequenced using aHiSeq2000 instrument (Illumina) with ∼80 × 106 reads persample (50 nt per read).The sequencing results were analyzed by assigning individual

cDNA sequences to specific entries in mammalian genome data-bases that include, besides protein-coding transcripts, functionalRNA species (rRNA, small nuclear RNA, tRNA, and so forth),RNAs transcribed from various classes of repeats, and non-coding RNAs (28). The majority of database entries were rep-resented equally among the four RNA samples; however, threetypes of RNA transcripts were specifically and significantly moreabundant in the RNA sample from 5-aza-dC–treated p53-nullMEFs. These included (i) both major classes of mouse SINEs,namely B1 and B2 (Fig. 4 B and C), (ii) near-centromeric major(γ) satellite repeats (GSAT) (Fig. 4A), and (iii) a large numberof different noncoding RNA species (ncRNAs) (Fig. 4D andTable S3). Importantly, we also identified repeats and ncRNAsthat were induced in both p53-WT and p53-null MEFs upontreatment with 5-aza-dC, albeit to a greater extent in p53-nullcells. These transcripts apparently represent cases of methyla-tion-based silencing that depends only partially on p53 and in-cluded, for example, transcripts of endogenous retrotransposonssuch as intracisternal A particles (IAP) (Fig. S4), “primitive” re-latives of retroviruses present in mouse genomes at about 10,000copies (29). All these observations were confirmed using Northernblot hybridization with independently isolated RNA samples fromuntreated and 5-aza-dC–treated p53-WT and p53-null MEFs (Fig.4 A–D and Fig. S4).Northern blotting demonstrated that both 5-aza-dC–treated

and untreated p53-WT cells, as well as untreated p53-null cells,expressed sequences corresponding to B1 SINE predominantlywithin large RNA transcripts, whereas RNA corresponding toindividually transcribed B1 elements (116-bp-long RNA poly-merase III-driven transcript) was barely detectable (Fig. 4C).However, in the sample from 5-aza-dC–treated p53-null MEFs,the most intense signal was concentrated in the size range ofsingle-element transcripts, thus indicating that DNA demethy-lation in the absence of p53 results in strong activation of tran-scription of B1 and B2 elements. Annealing of these inducedmonomeric-positive strands of B1 and B2 transcripts to anti-

p53

WT shp53 GSE56

ifnargenotype

Dox

ac�n

A

B

C

0

20

40

60

80

100

120

0 0.5 1 2 4

WT

shp53GSE56

0

2000

4000

6000

8000

10000

12000

5 10 20

WT

ifnar / /

D

E

p49

Fig. 3. Treatment of p53-null MEFs with 5-aza-dC induces a lethal IFN re-sponse. (A) Western blot detection of p53 and β-actin (loading control)proteins in MEFs from ifnar−/− mice expressing endogenous WT p53 (lanes1 and 2) or shRNA against p53 (shp53, lanes 3 and 4) or the p53-inactivatinggenetic suppressor element-56 (GSE56, lanes 7 and 8) left untreated (−) ortreated (+) with 500 nM doxorubicin (Dox) for 16 h. (B) Cytotoxicity of 5-aza-dC in cells differing in p53 and IFNAR status. Cells were treated with theindicated concentrations of 5-aza-dC for 5 d. Viability was determined bymethylene blue staining and extraction, followed by spectrophotometricquantification. Percent viability is shown relative to control cells treated with0.1% DMSO. (C) Cytotoxicity of 5-aza-dC in MEFs from ifnar+/+ or ifnar−/−

mice expressing endogenous WT p53 (transduced with a nonspecific controlshRNA construct) or shRNA against p53 (shp53). Cells were left untreated ortreated with 10 μM 5-aza-dC for 120 h before detection of viable cells bymethylene blue staining. (D) Caspase-3,7 activity (cleavage of the fluorescentsubstrate AC-Devd-AMC) in MEFs from p53-WT, p53-null, and ifnar−/− miceand MEFs from ifnar−/− mice expressing shRNA against p53 (shp53). Cellswere treated with the indicated concentrations of 5-aza-dC for 48 h beforethe caspase assay. (E) Western blot analysis of expression of the IFN-in-

ducible protein p49 (and β-actin as a loading control) in SCCVII cells: intact(lane 1); mock transfected (lane 2); transfected with the GFP-expressionconstruct plv-CMV-GFP (250 ng per well of a six-well plate; used here andbelow for monitoring transfection efficiency) (lane 3); transfected with plv-CMV-GFP as above together with 500 ng of RNA (rRNA-depleted fraction)from 5-aza-dC–treated p53-WT (lane 4) and p53-null MEFs (lane 5) and un-treated p53-WT (lane 6) and p53-null MEFs (lane 7). Transfection of plv-CMV-GFP and double-stranded poly(I:C) RNA (1μg, an efficient IFN-inducingagent) was used as a positive control (lane 8).

E92 | www.pnas.org/cgi/doi/10.1073/pnas.1216922110 Leonova et al.

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 26

, 202

1

sense-oriented B1 and B2 sequences present in the untranslatedregions of numerous protein-coding mRNAs can be predicted tocreate a large pool of dsRNA potentially capable of triggering anIFN response (Fig. 4F).In contrast to SINEs, transcripts of GSAT sequences vary in

length from 240 bp to more than 10 kb (Northern blot in Fig.4A). Detection of GSAT transcripts can be revealed by radio-labeled probes representing both positive and negative strands ofGSAT DNA as demonstrated by dot-blot hybridization with theprobes specific for different RNA strands, indicative of bi-directional transcription. This result is consistent with previous

reports in which temporary activation of GSAT transcription andthe appearance of dsRNA of various lengths was observed duringspecific stages of embryonic development (30) and in the heartsof old adult mice (31). dsRNA formed by annealed comple-mentary GSAT RNA strands could trigger IFN induction.To gain an appreciation of the potential biological impact of

the identified 5-aza-dC–induced transcripts in p53-null MEFs,we estimated their overall abundance vis-à-vis the abundance ofmRNAs for β-actin, a highly expressed transcript commonly usedas a reference in gene-expression studies. We calculated the re-lative representation of RNAs transcribed from SINEs (B1 and

020406080

100120140160

p53K

O-a

zaW

T-az

ap5

3KO

WT

sat DNASINEsncRNAERVs

A

FE

RNA

abun

danc

e(β

-ac�

nun

its)

RNA

abun

danc

e(β

-ac�

nun

its)

D

Pol IIItranscripts

gene

Pol II transcripts

dsRNA

exonintronnon-coding regionB1, senseB1, an�sense

RNA

abun

danc

e(β

-ac�

nun

its)

IIIIIbIaI

B C

Fig. 4. Massive transcriptional up-regulation of repetitive elements in p53-null MEFs treated with 5-aza-dC. The abundance of GSAT (A), B2 (B), B1 (C), andncRNA (D) transcripts in RNA samples isolated from untreated p53-WT MEFs, 5-aza-dC–treated p53-WT MEFs, untreated p53-null MEFs, and 5-aza-dC–treatedMEFs relative to the abundance of β-actin mRNA is shown in bar graphs (calculations are based on the results of total RNA sequencing) and Northern blots. ForNorthern blots the positions of 18S and 28S rRNAs are indicated by arrowheads, and ethidium bromide staining of the gel before transfer is shown in A asa common control for RNA loading and quality. (E) The overall abundance of RNA transcripts representing GSAT DNA (sat DNA), SINEs B1 and B2, ncRNAs, andIAPs in the p53-null (KO) or p53-WT MEFs, untreated or treated with 5-aza-dC, is shown in β-actin units (y axis). Pie diagrams show the proportion of each ofthe above-listed classes of RNAs in the pool of new transcripts induced by 5-aza-dC treatment in p53-null MEFs (Upper) and in the pool of transcripts presentin untreated p53-null cells versus untreated p53-WTcells (Lower). (F) Hypothetical scheme of formation of dsRNA by annealing of RNA-polymerase III-driventranscripts of B1 or B2 SINEs with B1 and B2 sequences present in antisense orientation in polymerase II-driven mRNAs. Introns within mRNA are spliced outbefore nuclear export, so it is unlikely that a dsRNA will be formed by the annealing of a polymerase III-transcribed SINE sequence to SINE sequences withinmRNA introns for further detection by pattern recognition receptors outside the nucleus, such as PKR.

Leonova et al. PNAS | Published online December 10, 2012 | E93

MED

ICALSC

IENCE

SPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 26

, 202

1

A

DC

0

20

40

60

80

100

120

0 1:101:201:40Compe�ng DNA, fold

0 10 20 40

Contr.

B1

p53 cons

Rela

�ve

band

inte

nsity

,%

B1 SINE 1-GCCGGGCATGGTGGCGCACGCCTTTAATCCCAGCACTTGGGAGGCAGAGGCAGGCGGATTTCTGAGTTCGAGGCCAGCCTGGTCTACANA-90 91-GTGAGTTCCAGGACAGCCAGGGCTACACAGAGAAACCCTGTCTCG-135

B2 SINE 1-GGGCTGGAGAGATGGCTCAGTGGTTAAGAGCACCTGACTGCTCTTCCAGAGGTCCTGAGTTCAATTCCCAGCAACCACATGGTGGCTCAC-90

91-AACCATCTGTAATGAGATCTGATGCCCTCTTCTGGTGTGTCTGAAGACAGCTACAGTGTACTTACATATAATAAATAAATAAATAAATAA-180181-ATCTTAAAAAAAAAAAAAAGAAAGAAAAA-209

Major (gamma) satellite DNA (GSAT) 1-GACCTGTAATATGGCGAGAAAACAGAAAATCACGGAAAATGAGAAATACACACTTTAGGACGTGAAATATGGCGAGGAAAACTGAAAAAG-90 91-GTGGAAAATTTAGAAATGTCCACTGTAGGACGTGGAATATGGCAAGAAAACTGAAAATCATGGAAAATGATAAACATCCACTTGACGACT-180181-TGAAAAATGACGAAATCACTAAAAGACGTCAAAAATGAGAAATGCACACTGAAA-254

B

FE

GGG A T G T GGCG AC CC T

324GT T GA GCC AGC T G T C T

5866G T T C A GAC AGC AG GC T

41159

-20

-10

0

10

20

30

40

Max

diff

eren

ce o

f mut

atio

n p-

valu

es (

in S

D u

nits

)

0

1

2

3

4

5

6

5+6+

15+1

6

4+7+

14+1

7

3+8+

13+1

8

2+9+

12+1

9

1+10

+11+

20

0.004*

0.51 0.52

0.094

0.68

Min

us L

ogF

test

for u

p vs

. dow

n m

uta�

on

rate

in 5

aza-

dC-t

reat

ed p

53-K

O c

ells

Fig. 5. Structural features of major repetitive sequences that are transcriptionally repressed by p53. (A and B) Comparison of putative p53-binding sites fromB1, B2, and GSAT repeats with “activating” and “repressing” p53-binding consensus sequences (32). (C and D) p53 binding to its consensus sequence withina 32P-labeled oligonucleotide was detected by EMSA (C, lane 1), and the specificity of the observed p53-probe complex was confirmed by supershift with theanti-p53 antibody Ab421 (lane 2). Unlabeled competitor oligonucleotides were added at 10:1, 20:1, and 40:1 molar excess over the labeled probe in lanes 3–5(oligonucleotide containing the p53-binding consensus sequence, positive control), lanes 6–8 (an 84-bp oligonucleotide containing the first two B1-derivedputative p53-binding sequences), and lanes 9–11 (λpL promoter-derived oligonucleotides of similar length, negative control). Bands corresponding to p53-binding complexes were quantified by densitometry and normalized to the free radio-labeled probe. (E) Relative frequency of deviations from consensus atspecific nucleotides positions within putative p53-binding sites of B1 elements activated by 5-aza-dC in p53-null MEFs. Mutation rates (calculated as P values)were determined for the indicated sets of nucleotides of the three putative p53-binding sites found in B1 consensus sequence. The directed minimal dif-ference was calculated between P values in the 5-aza-dC–treated p53-KO sample versus the first three samples. Analysis of each key position (highlighted inred) of the p53-binding sites within the SINE B1 revealed that the positive minimal differences (black bars) are indicative of lower mutation rates found in thetreated p53-KO samples as compared with any other sample set comprising other nucleotides from p53 recognition element (blue bars; shown for only one offour sets analyzed). (F) Comparison of the mutation rate in the key nucleotide positions with mutation rates in other nucleotides within putative p53-bindingsites in all four series was performed by the voting method. For each series, the sum of squares of positive differences is χ2 distributed; the sum of squares ofthe negative differences also is χ2 distributed. The first χ2 value indicates the overall voting of series’ positions for lower rate of mutation in 5-aza-dC–treatedp53-KO. Conversely, the second χ2 value is voting for higher overall rate of mutations in the series. The ratio of these two values normalized by the cor-responding degrees of freedom (numbers of positive and negative differences) is F-distributed. Results are shown as minus log P values of the F-test in key-position series (red) and four control series (blue). The F-test P values of a lower mutation rate in the 5aza+p53ko sample across five series of positions arelisted. *Only one P value (0.004 of the key positions) is statistically significant.

E94 | www.pnas.org/cgi/doi/10.1073/pnas.1216922110 Leonova et al.

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 26

, 202

1

B2), satellite DNA (GSAT and SATMIN), IAPs, and ncRNAs ineach sample based on the proportions of their correspondingsequences in RNA samples used for high-throughput sequenc-ing. The results shown in Fig. 4E demonstrate that the amount ofnew RNA (counted as the number of monomeric copies) syn-thesized specifically in p53-null cells following treatment with5-aza-dC was more than 150 times greater than the level of β-actinmRNA. Two-thirds of this new pool of RNA comprised tran-scripts of SINEs and satellite DNA in nearly equal proportions,and the remaining one-third consisted of equal proportions ofIAP transcripts and ncRNAs. Interestingly, however, IAPs werenot among the RNA species that showed differential expressionbetween untreated p53-null and untreated p53-WT MEFs, thussuggesting that their DNA methylation-based silencing is notstrictly p53 dependent.

Structural Properties of Major Classes of p53-Controlled Repeats.Reconstruction of the phylogenetic history of SINEs suggeststhat their amplification in mammalian genomes started about 65million years ago and involved a series of explosions that createdsubfamilies of repeats, each with shared mutations (9, 12, 32). Todetermine whether B1 and B2 transcription initiated by DNAhypomethylation in p53-null cells involved random or specificsubsets of repeats, we calculated the deviation from the con-sensus SINE sequence (built by analyzing the entire SINE familyin the mouse genome) for each nucleotide position of the B1 andB2 sequences identified in sequencing our four RNA samples(p53-WT and p53-null MEFs, untreated and 5-aza-dC–treated).No major differences were found in the profiles of nucleotidepolymorphisms along B1 and B2 transcripts among thefour sam-ples, thus indicating roughly random activation of transcription ofdifferent subsets of SINEs in p53-null cells treated with 5-aza-dC(correlation coefficients >0.96). However, more precise analysisrevealed significant shifts in frequency of mutations in a number ofpositions in B1 transcripts in the RNA sample from 5-aza-dC–treated p53-null MEFs as compared with the three other subsets.Interestingly, nucleotide substitutions were significantly less fre-quent in specific positions within putative p53-binding sequences(Fig. 5 E and F).p53 is known as a repressor of transcription acting via recog-

nition of specific sequences that are similar to those that serve asp53-binding sites in p53-induced genes (33, 34). A search forputative p53-binding–like sites revealed a series of candidatesequences in B1, B2, and GSAT elements (Fig. 5 A and B). GSATelements were found to be especially enriched with potentialp53-binding sequences. The functionality of predicted p53-bindingsites was tested using EMSA with synthetic oligonucleotides rep-resenting different B1 fragments incubated with nuclear extractsof mouse cells known to have WT and functional p53. Althoughthe oligonucleotides used did not show stable binding with p53under the applied conditions, they nevertheless were capable ofinhibiting p53 binding to a control oligonucleotide containing aconsensus p53-binding site (Fig. 5 C and D). This inhibition wasdependent on the presence of intact putative p53-binding sitesin B1-derived oligonucleotides, because point mutations at keynucleotides of the p53-binding sequence completely abrogatedthe ability of oligonucleotide to compete for p53 binding (Fig.S5). Remarkably, mutations were less frequent in these particularnucleotides than in the rest of the positions within p53-bindingsites of B1 transcripts induced by 5-aza-dC treatment (Fig. 5 E andF). These observations suggest that B1 repeats are capable ofspecific p53 binding, albeit with low affinity, and that this bindingis required for their transcriptional repression by p53.

Transcription of Repeats and ncRNAs Can Occur in Tumors. p53 is themost frequently mutated gene in tumors. However, even in tumorsthat retain WT p53 gene sequences, p53 commonly is inactivatedby other means [e.g., viral protein expression, overexpression of

the natural p53 inhibitor mdm2, or loss of Arf (35)]. Hence, wehypothesized that tumor cells might be prone to induction oftranscription of repeats and ncRNAs following 5-aza-dC treat-ment, as was observed in p53-null MEFs. To test this hypothesis,we chose three mouse tumor-derived cell lines (SCC-VII, CT26,and LLC) without significant levels of basal repetitive elementtranscription (as illustrated by low levels of GSAT transcripts;Fig. 6A). We found that SCC-VII, CT26, and LLC cells showedstrong, intermediate, and undetectable expression of GSAT RNA,respectively, upon treatment with 5-aza-dC (Fig. 6A). In addition,we detected strong up-regulation of the mRNAs encoding IFN-β1and its downstream responsive genes, IFN regulatory factor 7(IRF7) and CXCL10, in 5-aza-dC–treated SCC-VII cells and, toa somewhat lesser extent, in 5-aza-dC–treated CT26 cells (Fig.6B). In contrast, none of these IFN response-associated tran-scripts was detected in 5-aza-dC–treated LLC cells. Consistentwith these findings, SCC-VII cells were the most 5-aza-dC–sen-sitive cell line (LD50 = 0.125 μM), CT26 cells showed in-termediate sensitivity (LD50 = 2 μM), and LLC cells were theleast sensitive (LD50 = 14 μM) (Fig. 6C). The identified positivecorrelation between transcription of repeats, induction of an IFNresponse, and sensitivity to 5-aza-dC supports our model inwhich tumors with reduced p53 function (36–38) are prone toactivation of transcription of repeats under conditions of re-duced DNA methylation. These results suggest that the sensi-tivity of tumor cells with silent repeats to 5-aza-dC (i.e., the drugdecitabine approved for treatment of myelodysplastic syndromeand considered for anticancer use) may depend on whether theyhave functional p53.Decreased genome-wide DNA methylation is another com-

mon property of tumors acquired during in vivo growth (39).Together with inactivation of p53, this decreased methylationcould create conditions sufficient to induce transcription ofrepeats and ncRNAs normally suppressed by a combination ofp53 and DNA methylation. Therefore, we hypothesized thattranscription of repeats leading to the induction of an IFN re-sponse might occur spontaneously during tumor growth andprogression in vivo. We tested this hypothesis in two mouse tu-mor models. First, we showed that all tested spontaneous thymiclymphomas [the tumors that develop most frequently in cancer-prone p53-null mice (40)] contained much higher levels of re-petitive element transcripts than normal tissues of p53-null mice,including thymuses of p53-null mice before lymphoma de-velopment (Fig. 6D, shown for GSAT RNA transcribed fromboth DNA strands). As in p53-null MEFs treated with 5-aza-dC,the activation of transcription of repeats correlated with the in-duction of IFN response, as demonstrated by RT-PCR results(Fig. 6D).Similarly, repetitive element expression was observed in six of

the eight the mammary gland tumors that developed spontane-ously in untreated MMTV-her2/neu transgenic mice (41) (Fig.6E). Previous reports indicate acquisition of missense mutationsin at least 37% of these tumors (42). Note that although expres-sion of only one specific class of repeats is shown for each of theabove cases, activation of GSAT, B1, and B2 always occurred to-gether. Our finding that growing tumors express high levels ofrepeats capable of generating dsRNA and subsequently inducing alethal IFN response suggest that such tumors have passed success-fully through selection for resistance to endogenous IFN toxicity.

DiscussionThis study reveals a function of p53 that, in cooperation with DNAmethylation, keeps large families of interspersed and tandem re-peats transcriptionally dormant. Treatment of p53-deficient but notp53-WT cells with the DNA-demethylating agent 5-aza-dC wasshown to cause transcriptional derepression of several classes ofnormally silent interspersed repeats (e.g., SINEs such as B1 and B2repeats), tandem DNA satellites (e.g., GSAT), and ncRNAs. To-

Leonova et al. PNAS | Published online December 10, 2012 | E95

MED

ICALSC

IENCE

SPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 26

, 202

1

gether, these elements represent a significant proportion of themouse genome (>10%) and, when transcribed, give rise to newRNA species comparable in their abundance to the bulk ofcellular mRNA.Transcriptional derepression of repeats resulting from a com-

bined lack of p53 function and DNA methylation was accom-panied by induction of the classical type I IFN signaling pathway,which in MEFs is driven by IFN-β1 and leads to apoptotic celldeath. We have named this phenomenon TRAIN. The criticalrole played by the IFN response in death of MEFs under con-ditions of TRAIN was demonstrated using cells deficient in IFNreceptor expression. Although the precise mechanism underlyingthis toxicity remains to be determined, our findings are consistentwith a growing body of evidence indicating that activation of astrong IFN response by dsRNA can result in apoptosis mediatedeither by induction of the proapoptotic protein kinase R (PKR)and 2’-5′-oligoadenylate synthetase (OAS)/RNase L pathways oractivation of TRAIL/FAS death receptors (24).In mammalian cells, IFN signaling commonly is triggered as an

antiviral defense mechanism in response to dsRNA producedduring viral replication (24). Our results suggest that RNA speciesproduced following transcriptional derepression of endogenousrepeats in the absence of p53 and methylation also can trigger ofthe IFN response. Transcripts of repetitive elements, retro-transposons, and satellite DNA sequences can form dsRNAbecause of their intense secondary structure and the presence of

complementary transcripts in the cell. For SINEs, such com-plementary sequences exist in noncoding regions of mRNAs thatwere transcribed through SINEs integrated in their antisenseorientation (Fig. 4F). In the case of GSAT, both strands aretranscribed, thereby producing cRNA strands (30). Although wehave shown that total RNA from 2-aza-dC–treated p53-null cellshas stronger IFN-inducing capacity than RNA from untreatedp53-null cells or from both treated and untreated p53-WT cells(Fig. 3E), our current technical capabilities are not sufficient todetermine the relative impact of each of these classes of tran-scripts on TRAIN. In fact, the scale of expression of “new”transcripts that appear in p53−/− cells treated with 5-aza-dC is sostrong and their copy number in the genome is so high as topreclude the use of any gene knockout or knockdown techniqueto assess this impact directly. Nevertheless, because transcripts ofSINEs and GSAT, both of which are capable of forming dsRNAs,comprise the major proportion of new RNA synthesized in p53-null cells following DNA hypomethylation, it is highly probablythat these RNA species are responsible for IFN induction.Overall, our results support a model in which epigenetic si-

lencing of repeats (an essential condition for genomic stability andviability of currently existing species) is controlled by three factors:(i) p53-mediated transcriptional silencing, (ii) DNA methylation-mediated suppression of transcription, and (iii) a suicidal IFN re-sponse which eliminates cells that escape the first two lines of con-trol. This model presents a major role for p53 as a “guardian ofrepeats,” which is likely an important component of its function asa “guardian of the genome” (43). In fact, for at least interspersedrepeats (products of reverse transcription), transcription is an es-sential prerequisite for their amplification and subsequent possibleinsertional mutagenesis. Given the abundance of repeats in mam-malian genomes and their potential for producing large amountsof transcripts (which in the presence of reverse transcriptase maybe converted into insertional mutagens), it is reasonable to statethat silencing of repeats is likely an evolutionarily importantfunction of p53 that may have been critical for survival of prede-cessors of current species during times of active repeat amplifi-cation. Our work also reveals a function for the IFN response inaddition to its role in antiviral innate immunity: maintenance ofgenomic stability through the elimination of cells that have lostepigenetic silencing of interspersed and tandem repeats. Althoughall the results reported here were obtained in mice, the literatureprovides numerous indications that TRAIN may be universalamong mammals. For example, aberrant overexpression of satel-lite repeats was reported recently in pancreatic and other epi-thelial human cancers (18). In addition, hypomethylation andactivation of polymerase III-driven Alu (the only major SINE inthe human genome) transcription was found in human tumors(44, 45). TRAIN is likely to be the explanation for the previouslydescribed phenomenon of inhibition of growth of pancreatictumor cells by 5-aza-dC accompanied with the induction of IFNresponse signaling (46).Although the precise mechanism by which p53 controls si-

lencing of repeats remains to be defined, it is reasonable toconsider that it may involve the known transcriptional repressorfunction of p53 (34). The power of p53 as an epigenetic silenceris implied by the fact that none of the multiple, independentlygenerated lines of transgenic mice carrying p53-responsive re-porter genes (47–49) was able to maintain transgene inducibilityfor more than two generations (in contrast to mice with reportersof other transcriptional regulators). p53 exerts its transcriptionalrepressor function by binding to sites that are similar to thosethat mediate its transcriptional activation function but that havemore sequence flexibility (50). Thus, we hypothesized that p53-mediated silencing of repeats might occur through a similarmechanism involving direct binding of p53 to specific sites withinthe repetitive elements. Support for this hypothesis was providedby our identification of predicted p53-binding sites within ele-

CT26- + - + - +

LLC-1 SCC-VII5-aza-dC

100

120 LLC-1CT26

A C

GSAT-F

GSAT-R

0

20

40

60

80

100

Viab

ility

(%)

SCC-VII

B- + - + - +

IFNβ

β

β

1

SCC-VII CT26LLC-15-aza-dC (μM)

00.

125

0.25 0.

5 1 2 4 8 16 32

5-aza-dC (μM)

D

CXCL10

IRF7

E1 2 3 4 5 6 7 8

MMTV-her2/neu tumors

28S18SL1 L2 L3 L4 L5

Thymus Thymic lymphomasT4 T5

B1

28S18S

GSAT-F

GSAT-R

IFN 1

-

Fig. 6. Detection of transcripts of repetitive elements in mouse tumor celllines and spontaneous tumors. (A) Detection of mouse GSAT sequences intotal RNA frommouse tumor cell lines CT-26 (colon tumor), LLC-1 (Lewis lungcarcinoma), and SCC-VII (squamous cell carcinoma) left untreated or treatedwith 10 μM 5-aza-dC for 48 h. Dot blotting was performed with 500 ng totalRNA per dot and single-strand hybridization probes GSAT-F and GSAT-R. (B)Detection of IFN-β1, IRF-7, CXCL10, and β-actin (loading control) mRNA byRT-PCR in the cells described in A. (C) Cytotoxicity of 5-aza-dC in LLC-1, CT26,and SCC-VII cells. Cells were treated with the indicated concentrations of5-aza-dC for 5 d. Viability was determined by methylene blue staining andextraction, followed by spectrophotometric quantification. Percent viabilityis shown relative to control cells treated with 0.1% DMSO. (D) (Upper TwoPanels) GSAT sequences were detected in total RNA from thymic lymphomasof p53-null mice (five tumors, L1–L5, from five different mice assayed induplicate) and in two normal thymuses (T4 and T5) isolated from p53-nullmice using dot blotting as described in A. (Lower Two Panels) RT-PCR analysisof IFN-β1 and β-actin (loading control) mRNA expression. (E) Northern hy-bridization was used to detect SINE B1 sequences in total RNA from MMTV-her2/neu mammary tumors (eight tumors from eight different mice). 18Sand 28S rRNA levels detected by ethidium bromide staining confirmedequivalent RNA quality and loading for all tumors.

E96 | www.pnas.org/cgi/doi/10.1073/pnas.1216922110 Leonova et al.

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 26

, 202

1

ments representing all the major classes of repeats that becomeactivated in the absence of p53 and DNA methylation anddemonstration of p53 binding (albeit weak) to such sites inEMSA competition assays. After low-affinity binding to sites inrepeat elements, p53 may recruit other factors to establish tran-scriptional repression. Candidates for such factors include DNMT1(51) and DNA methyltransferase 3a (DNMT3a) (52), which havebeen shown to interact with p53 and execute p53-mediated generepression. It should be noted, nevertheless, that other p53-in-dependent mechanisms must exist to explain DNA methylation-mediated silencing of repeats observed in tissues of p53-nullmice. In this regard, it would be interesting to determine whetherother p53 family members (e.g., p63 or p73) might contribute tothe suppression of TRAIN.A striking observation related to the phenomenon of TRAIN

is the presence of p53-binding sites within major classes of SINEsof both primates [Alu repeats (53)] and rodents [B1, B2, andGSAT (this work)], even though primate and rodent repeatsevolved independently after the divergence of these two phylo-genetic branches. The presence of these sites suggests that p53may play a similar role in silencing SINEs in both genera andthat this role may be evolutionarily important for preventingSINE expansion and maintaining genetic stability.The contribution of the TRAIN-suppressor function of p53 to

its tumor-suppressor activity (if any) remains to be elucidated.Tumors developing spontaneously in p53-null mice demon-strated naturally occurring derepression of repetitive elementsinvolved in TRAIN, thus suggesting a potential link betweenTRAIN and carcinogenesis. Whether unleashed expression ofrepeats in tumors causes amplification of repeats and increasedgenomic instability remains to be determined. It is noteworthythat activation of transcription of reverse transcriptase-associ-ated repetitive elements was limited to SINEs and did not in-volve long interspersed nuclear elements, the likely source offunctional reverse transcriptase in mammalian genome (54).Our observation of repeat derepression in spontaneous breast

tumors from p53-WT MMTV-neu transgenic mice suggests thattumor progression provides a platform upon which such dere-pression occurs naturally. Tumors, including those that developin this particular model (42), frequently are characterized by p53inactivation, which enables unconstrained cell proliferation.There also is extensive evidence showing that the general degreeof DNA methylation declines during tumor progression (23).Hence, both conditions required to initiate TRAIN (loss of p53function and hypomethylation) occur naturally and with highfrequency in tumors. In fact, transcription of normally silent andheavily methylated sequences representing different types ofrepeats (both interspersed and tandem repeats) has been shownto occur in tumors (refs. 18, 45, and this work) as well as underother stress conditions (15). Nevertheless, tumor cells appar-ently escape TRAIN-induced cell death. This avoidance pre-sumably involves selection of tumor cells that have acquiredchanges that make them resistant to suicidal IFN induction, thusimplying an important tumor-suppressor role for the IFNpathway. In fact, the combination of p53 deficiency with theknockout of the irf1 gene, one of the major triggers of IFNsignaling by dsRNA (24), dramatically increases the frequencyand spontaneous tumor development (55).Tolerance of tumor cells to constitutive expression of repetitive

elements associated with TRAIN is likely caused by (i) acquired

defects in IFN induction or (ii) acquired or intrinsic resistance tothe suicidal IFN response. Consistent with this notion, there arenumerous indications of loss of IFN function in multiple tumortypes (56, 57), including, for example, homozygous loss of chro-mosome 9p21, which contains the genes for all type I IFNs (58,59). In this regard, it should be noted that tumor cells frequentlyare hypersensitive to lytic viruses (60), a property that could bewell explained by acquisition of defects in IFN responses drivenby the necessity to survive TRAIN.Thus, the phenomenon of TRAIN explains a series of previ-

ously unconnected but well-documented properties of tumorcells, including transcription of repeats, deregulation of IFNfunction, and increased sensitivity to lytic viruses. Identificationof specific derepressed sequences involved in TRAIN in humancells and development of assays for detection of TRAIN inclinical samples of human tumors will be an essential step inlinking this biological process to the diagnosis of specific tumortypes and stages of progression as well as the prediction of theirprognosis and sensitivity to treatment. For example, detection ofTRAIN-associated marker transcripts or signs of constitutivelyactive IFN signaling (without signs of toxicity) could be used toassess p53 functionality in tumors.

Materials and MethodsAnalysis of Results of Total RNA Sequencing. Mapping of Illumina reads ontorepetitive elements and ncRNA species was performed by the BOWTIE pro-gram (61) under default parameters. As to ndRNAs, the sequences from allof the RNAdb (http://research.imb.uq.edu.au/rnadb/) subbases (fantom 3,RNAz, and others) were used for mapping of reads from the samples. TheRepbase database (www.girinst.org/repbase/) was used as a set of uniquerepetitive elements for the mapping. The expression level of each ncRNAand repetitive element was calculated as median per-position coverage of itssequence by mapped reads, the per-position normalized measure of ex-pression that does not depend on the sequence length. To give a biologicaldimension to this measure of expression for the studied ncRNA species andrepetitive elements, each value was recalculated in “β-actin units.” Namely,first, the reads of four RNA-sequence samples were mapped onto β-actinmRNA, and the median per-position counts of this mRNA across sampleswere calculated. Next, the per-position measure of expression for a particu-lar ncRNA or a repetitive element in a sample was divided by the per-posi-tion value of expression for β-actin in the same sample, giving the expressionlevel of each studied sequence across the four RNA-seq samples in units ofthe β-actin expression in the corresponding sample.

Mapping of reads from the RNA samples onto B1 and B2 consensussequences was performed by a BLAST-like alignment program that detectsmutations more accurately than the programs based on Burrows–Wheelerindexing. For each position, a significance of the mutation in three treatedsamples regarding the WT sample was calculated via Poisson statistics.Namely, for each position of the consensus, the expected number of countsof the consensus nucleotide at this position in a treated sample was calcu-lated based on the frequency of this nucleotide in the WT sample. A distancein SD units between the real number of counts of the consensus nucleotidein a treated sample and expected number of nucleotide counts defines thepolymorphism of this particular position. A limit of 20 SDs was consideredthe threshold of mutation significance.

ACKNOWLEDGMENTS. We thank Patricia Stanhope-Baker for help in man-uscript preparation; Gus Frangou, Mikhail Makhanov, Jaime Wetzel, AnatoliGleiberman, and Natalia Fedtsova for technical help; and Katerina Gurova forcritical reading of the manuscript. This work fulfilled part of the requirementsfor the PhD earned by K.I.L. at State University of New York at Buffalo. Thiswork was funded by a grant from Tartis, Inc. (to A.V.G.) and a grant from theNational Institutes of Health AI073303 (to G.C.S.).

1. Makałowski W (2000) Genomic scrap yard: How genomes utilize all that junk. Gene259(1-2):61–67.

2. Kramerov DA, Vassetzky NS (2011) SINEs. Wiley Interdiscip Rev RNA 2(6):772–786.

3. Smit AF (1999) Interspersed repeats and other mementos of transposable elements inmammalian genomes. Curr Opin Genet Dev 9(6):657–663.

4. Ostertag EM, Kazazian HH, Jr. (2001) Biology of mammalian L1 retrotransposons.Annu Rev Genet 35:501–538.

5. Charlesworth B, Sniegowski P, Stephan W (1994) The evolutionary dynamics ofrepetitive DNA in eukaryotes. Nature 371(6494):215–220.

6. Plohl M, Luchetti A, Mestrovi�c N, Mantovani B (2008) Satellite DNAs betweenselfishness and functionality: Structure, genomics and evolution of tandem repeats incentromeric (hetero)chromatin. Gene 409(1-2):72–82.

7. Quentin Y (1989) Successive waves of fixation of B1 variants in rodent lineage history.J Mol Evol 28(4):299–305.

8. Deininger PL, Batzer MA (2002) Mammalian retroelements. Genome Res 12(10):1455–1465.

Leonova et al. PNAS | Published online December 10, 2012 | E97

MED

ICALSC

IENCE

SPN

ASPL

US

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 26

, 202

1

9. Amariglio N, Rechavi G (1993) Insertional mutagenesis by transposable elements inthe mammalian genome. Environ Mol Mutagen 21(3):212–218.

10. Ferrigno O, et al. (2001) Transposable B2 SINE elements can provide mobile RNApolymerase II promoters. Nat Genet 28(1):77–81.

11. Kramerov DA, Vassetzky NS (2011) Origin and evolution of SINEs in eukaryoticgenomes. Heredity (Edinb) 107(6):487–495.

12. Carnell AN, Goodman JI (2003) The long (LINEs) and the short (SINEs) of it: Alteredmethylation as a precursor to toxicity. Toxicol Sci 75(2):229–235.

13. Rhee I, et al. (2002) DNMT1 and DNMT3b cooperate to silence genes in human cancercells. Nature 416(6880):552–556.

14. Liu WM, Chu WM, Choudary PV, Schmid CW (1995) Cell stress and translationalinhibitors transiently increase the abundance of mammalian SINE transcripts. NucleicAcids Res 23(10):1758–1765.

15. Allen TA, Von Kaenel S, Goodrich JA, Kugel JF (2004) The SINE-encoded mouse B2RNA represses mRNA transcription in response to heat shock. Nat Struct Mol Biol11(9):816–821.

16. Jang KL, Latchman DS (1992) The herpes simplex virus immediate-early protein ICP27stimulates the transcription of cellular Alu repeated sequences by increasing theactivity of transcription factor TFIIIC. Biochem J 284(Pt 3):667–673.

17. Hagan CR, Rudin CM (2007) DNA cleavage and Trp53 differentially affect SINEtranscription. Genes Chromosomes Cancer 46(3):248–260.

18. Ting DT, et al. (2011) Aberrant overexpression of satellite repeats in pancreatic andother epithelial cancers. Science 331(6017):593–596.

19. Nieto M, et al. (2004) The absence of p53 is critical for the induction of apoptosis by5-aza-2′-deoxycytidine. Oncogene 23(3):735–743.

20. Ossovskaya VS, et al. (1996) Use of genetic suppressor elements to dissect distinctbiological effects of separate p53 domains. Proc Natl Acad Sci USA 93(19):10309–10314.

21. Bunz F, et al. (1999) Disruption of p53 in human cancer cells alters the responses totherapeutic agents. J Clin Invest 104(3):263–269.

22. Yoo CB, Jones PA (2006) Epigenetic therapy of cancer: Past, present and future. NatRev Drug Discov 5(1):37–50.

23. Jones PA, Baylin SB (2002) The fundamental role of epigenetic events in cancer. NatRev Genet 3(6):415–428.

24. Borden EC, et al. (2007) Interferons at age 50: Past, current and future impact onbiomedicine. Nat Rev Drug Discov 6(12):975–990.

25. Liu S-Y, Sanchez DJ, Aliyari R, Lu S, Cheng G (2012) Systematic identification of type Iand type II interferon-induced antiviral factors. Proc Natl Acad Sci USA 109(11):4239–4244.

26. Müller U, et al. (1994) Functional role of type I and type II interferons in antiviraldefense. Science 264(5167):1918–1921.

27. Fensterl V, White CL, Yamashita M, Sen GC (2008) Novel characteristics of the functionand induction of murine p56 family proteins. J Virol 82(22):11045–11053.

28. Wang Z, Gerstein M, Snyder M (2009) RNA-Seq: A revolutionary tool for transcriptomics.Nat Rev Genet 10(1):57–63.

29. Barbot W, Dupressoir A, Lazar V, Heidmann T (2002) Epigenetic regulation of an IAPretrotransposon in the aging mouse: Progressive demethylation and de-silencing ofthe element by its repetitive induction. Nucleic Acids Res 30(11):2365–2373.

30. Probst AV, et al. (2010) A strand-specific burst in transcription of pericentric satellitesis required for chromocenter formation and early mouse development. Dev Cell 19(4):625–638.

31. Gaubatz JW, Cutler RG (1990) Mouse satellite DNA is transcribed in senescent cardiacmuscle. J Biol Chem 265(29):17753–17758.

32. Kidwell MG, Lisch DR (2001) Perspective: Transposable elements, parasitic DNA, andgenome evolution. Evolution 55(1):1–24.

33. Ho J, Benchimol S (2003) Transcriptional repression mediated by the p53 tumoursuppressor. Cell Death Differ 10(4):404–408.

34. Wang B, Xiao Z, Ren EC (2009) Redefining the p53 response element. Proc Natl AcadSci USA 106(34):14373–14378.

35. Vogelstein B, Lane D, Levine AJ (2000) Surfing the p53 network. Nature 408(6810):307–310.

36. Hauser U, et al. (2002) Reliable detection of p53 aberrations in squamous cellcarcinomas of the head and neck requires transcript analysis of the entire codingregion. Head Neck 24(9):868–873.

37. Langlois MJ, et al. (2010) The PTEN phosphatase controls intestinal epithelial cellpolarity and barrier function: Role in colorectal cancer progression. PLoS ONE 5(12):e15742.

38. Burdelya LG, et al. (2006) Inhibition of p53 response in tumor stroma improvesefficacy of anticancer treatment by increasing antiangiogenic effects of chemotherapyand radiotherapy in mice. Cancer Res 66(19):9356–9361.

39. Smet CD, Loriot A (2010) Epigenetic scars of a neoplastic journey DNA hypomethylationin cancer. Epigenetics 5:206–213.

40. Donehower LA, et al. (1992) Mice deficient for p53 are developmentally normal butsusceptible to spontaneous tumours. Nature 356(6366):215–221.

41. Green JE, Hudson T (2005) The promise of genetically engineered mice for cancerprevention studies. Nat Rev Cancer 5(3):184–198.

42. Li B, Rosen JM, McMenamin-Balano J, Muller WJ, Perkins AS (1997) neu/ERBB2cooperates with p53-172H during mammary tumorigenesis in transgenic mice. MolCell Biol 17(6):3155–3163.

43. Lane DP (1992) Cancer. p53, guardian of the genome. Nature 358(6381):15–16.44. Lerat E, Sémon M (2007) Influence of the transposable element neighborhood on

human gene expression in normal and tumor tissues. Gene 396(2):303–311.45. Li TH, Kim C, Rubin CM, Schmid CW (2000) K562 cells implicate increased chromatin

accessibility in Alu transcriptional activation. Nucleic Acids Res 28(16):3031–3039.46. Missiaglia E, et al. (2005) Growth delay of human pancreatic cancer cells by methylase

inhibitor 5-aza-2′-deoxycytidine treatment is associated with activation of theinterferon signalling pathway. Oncogene 24(1):199–211.

47. Komarova EA, et al. (1997) Transgenic mice with p53-responsive lacZ: p53 activityvaries dramatically during normal development and determines radiation and drugsensitivity in vivo. EMBO J 16(6):1391–1400.

48. Gottlieb E, et al. (1997) Transgenic mouse model for studying the transcriptionalactivity of the p53 protein: Age- and tissue-dependent changes in radiation-inducedactivation during embryogenesis. EMBO J 16(6):1381–1390.

49. MacCallum DE, et al. (1996) The p53 response to ionising radiation in adult anddeveloping murine tissues. Oncogene 13(12):2575–2587.

50. Riley T, Sontag E, Chen P, Levine AJ (2008) Transcriptional control of human p53-regulated genes. Nat Rev Mol Cell Biol 9(5):402–412.

51. Estève PO, Chin HG, Pradhan S (2005) Human maintenance DNA (cytosine-5)-methyltransferase and p53 modulate expression of p53-repressed promoters. ProcNatl Acad Sci USA 102(4):1000–1005.

52. Wang YA, et al. (2005) DNA methyltransferase-3a interacts with p53 and repressesp53-mediated gene expression. Cancer Biol Ther 4(10):1138–1143.

53. Cui F, Sirotin MV, Zhurkin VB (2011) Impact of Alu repeats on the evolution of humanp53 binding sites. Biol Direct 6:2.

54. Beck CR, Garcia-Perez JL, Badge RM, Moran JV (2011) LINE-1 elements in structuralvariation and disease. Annu Rev Genomics Hum Genet 12:187–215.

55. Nozawa H, et al. (1999) Loss of transcription factor IRF-1 affects tumor susceptibility inmice carrying the Ha-ras transgene or nullizygosity for p53. Genes Dev 13(10):1240–1245.

56. Haus O (2000) The genes of interferons and interferon-related factors: Localizationand relationships with chromosome aberrations in cancer. Arch Immunol Ther Exp(Warsz) 48(2):95–100.

57. Heyman M, et al. (1994) Interferon system defects in malignant T-cells. Leukemia 8(3):425–434.

58. Fountain JW, et al. (1992) Homozygous deletions within human chromosome band9p21 in melanoma. Proc Natl Acad Sci USA 89(21):10557–10561.

59. Lydiatt WM, Davidson BJ, Schantz SP, Caruana S, Chaganti RS (1998) 9p21 deletioncorrelates with recurrence in head and neck cancer. Head Neck 20(2):113–118.

60. Ring CJA (2002) Cytolytic viruses as potential anti-cancer agents. J Gen Virol 83(Pt 3):491–502.

61. Langmead B, Trapnell C, Pop M, Salzberg SL (2009) Ultrafast and memory-efficientalignment of short DNA sequences to the human genome. Genome Biol 10(3):R25.

E98 | www.pnas.org/cgi/doi/10.1073/pnas.1216922110 Leonova et al.

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 26

, 202

1