Trinucleotide Repeat Expansion Diseases, RNAi and Cancer

Summary

Many neurodegenerative diseases are caused by unstable trinucleotide repeat (TNR) expansions located in disease-associated genes. siRNAs based on CAG repeat expansions effectively kill cancer cell lines in vitro through RNA interference (RNAi). They also cause significant reduction in tumor growth in a human ovarian cancer mouse model with no toxicity to the treated mice. This suggests that cancer cells are particularly sensitive to CAG TNR-derived siRNAs and explains a reported inverse correlation between the length of CAG TNRs and reduced global cancer incidences in some CAG TNR diseases. This review discusses both mutant proteins and mutant RNAs as a cause of TNR diseases with a focus on RNAi and its role in contributing to disease pathology and in suppressing cancer.

The TNR Diseases

Trinucleotide repeat (TNR) expansions in a number of genes are the cause of many degenerative diseases characterized by amplification of DNA triplet motifs [1]. The following triplets (when present on the coding strand) have been associated with neurodegenerative syndromes: CAG, CTG, GAA, CCG, and CGG. Most of them exhibit a negative correlation between the repeat length and the onset of the diseases and/or the severity of disease progression [2]. However, the mechanism of how the TNRs cause disease is quite diverse and is linked to the location of the TNRs within the affected genes. Traditionally, the diseases have been viewed as either caused by extended toxic protein stretches, toxic repeat RNAs, or by deregulated gene expression (Table 1). The most frequently amplified triplet is CAG. It has been implicated in Dentatorubral-pallidoluysian atrophy (DRPLA) [3]. Spinal and Bulbar Muscular Atrophy (SBMA) [4], and in Huntington’s disease (HD) [5]. In addition, a number of Spinocerebellar Ataxias (SCAs) exhibit elongated CAG repeats. They include SCA1 [6], SCA2 [7], SCA3 [8], SCA6 [9], SCA7 [10], SCA12 [11], and SCA17 [12]. Diseases caused by CTG-based TNRs overall are less frequent and include spinocerebellar ataxias type 8 (SCA8) [13]. Huntington disease-like 2 (HDL2) [14], and myotonic dystrophy type 1 (DM1). Even fewer diseases are caused by the expansion of the repeats GAA, CGG, and CCG. GAA-repeat expansion is associated with Friedreich’s ataxia (FRDA) [15]. CGG repeats underlie fragile X syndrome (FXS), fragile X-associated primary ovarian insufficiency (FXPOI) [16], and fragile X-associated tremor (FXTAS) [17], and CCG repeats cause fragile XE syndrome [18]. While this review focuses on TNRs, there are additional repeat diseases that have been identified with other repeat sizes: tetra-, penta-, or hexamers and even longer repeats [19]. They are present in myotonic dystrophy type 2 (DM2) (CCTG) [20, 21], SCA31 (TGGAA) [22], amyotrophic laterals sclerosis (ALS) and/or frontotemporal dementia (GGGGCC) [23, 24]. Detailed descriptions of disease etiology and neurological symptoms have been extensively reviewed elsewhere [2, 25]. In this review we will discuss the molecular causes of disease pathology as well as the connections between TNR diseases and cancer with a focus on RNAi.

Table 1:

Trinucleotide Repeat Expansion Diseases

Disease	Full disease name	Gene name	TNR	AA coded for	Functional disease group	Alternative mechanism	Chr band	Chr location (Homo sapiens Annotation Release 108, GRCh38.p7)	Location within gene	Normal repeat length	Intermediate repeat length	Mutant repeat length	Inheritance	Estimated frequency / Prevalence	Ref
SCA1	Spinocerebellar ataxia type 1; olivopontocerebellar atrophy I	ATXN1/ATX1/SCA1	CAG	Glu	Toxic protein		6p22.3	chr6:16,299,112 to 16,761,490	Exon-7	4 to 39	36 to 38	39 to 44	Autosomal dominant	1 to 2:100,000	6
SCA2	Spinocerebellar ataxia type 2	ATXN2	CAG	Glu	Toxic protein		12q24.12	chr12:111,452,214 to 111,599,676	Exon-1	≤ 31	32	≥ 33-200	Autosomal dominant	40:100,000	7
SCA3/MJD/MJD1	Spinocerebellar ataxia type 3; Machado-Joseph disease; Azorean disease	ATXN3/AT3	CAG	Glu	Toxic protein	Toxic RNA	14q32.12	chr14:92,058,552 to 92,106,621	Exon-10	12 to 43	45 to 60	60 to 87	Autosomal dominant	unknown	8
SCA6	Spinocerebellar ataxia type 6	CACNA1A/APCA	CAG	Glu	Toxic protein		19p13.13	chr19:13,206,442 to 13,506,460	Exon-47	4 to 18	19	20 to 33	Autosomal dominant	1:100,000	9
SCA7/ADCAII/OPCA3	Spinocerebellar ataxia type 7	ATXN7	CAG	Glu	Toxic protein		3p13.1	chr3:63,864,557 to 64,003,462	Exon-1	4 to 19	28-36	> 36-460	Autosomal dominant	<1:100,000	10
SCA17	Spinocerebellar ataxia type 17	TBP	CAG	Glu	Toxic protein		6q27	chr6:170,554,333-170,572,869	Exon-3	25 to 42	41 to 48	43 to 66	Autosomal dominant	0.47:1,000,000. to 0.16:100,000	12
DRPLA	Dentatorubral-pallidoluysian atrophy; Naito Oyanagi disease	ATN1/DRLPA	CAG	Glu	Toxic protein		12p13.31	chr12:6,924,463 to 6,942,321	Exon-5	6 to 35	unknown	≥ 48-93	Autosomal dominant	0.48:100,000 (in Japan)	147
SBMA	X-linked Spinal and bulbar muscular atrophy; Kennedy’s disease	AR	CAG	Glu	Toxic protein	Toxic RNA	Xq12	chrX:67,544,032 to 67,730,619	Exon-1	7 to 34	36 to 37	38 to 68	X-Linked Recessive	1:150,000	4
HD	Huntington’s disease	HTT/IT15	CAG	Glu	Toxic protein	Toxic RNA; siRNAs, RAN translation/poly Ala (S), poly Ser (S), poly Leu (AS), poly Cys (AS)	4p16.3	chr4:3,074,681-3,243,959	Exon-1	≤ 26	27 to 35	>36	Autosomal dominant	3 to 7:100,000	5
SCA8	Spinocerebellar ataxia type 8	ATXN8/ATXN8OS/(overlapping genes)	CTG (CAG repeat in ATX8)	n.a.	Toxic RNA	RAN translation/poly Ala (AS)	13q21/13q21.33	chr13:70,107,213-70,139,429 / ?	3’UTR	15 to 50	50 to 70 maybe	80 to 1300	Autosomal dominant	<1:100,000	13
SCA12	Spinocerebellar ataxia type 12	PPP2R2B	CAG	n.a.	Toxic RNA suspected	In vitro data: Increased gene expression	5q32	chr5:146,589,504 to 147,081,520	5’UTR	4 to 32	unknown	55 to 78	Autosomal dominant	rare, mostly found in Northern India	11
DM1/MD1	Myotonic dystrophy type 1; Steinert’s Disease	DMPK	CTG	n.a.	Toxic RNA	siRNAs; RAN translation/poly Glu (AS)	19q13.32	chr19:45,769,709 to 45,782,557	3’UTR	5 to 34	35 to 49	> 50-1000	Autosomal dominant	1:10,000 - 1:100,000	21
HDL2	Huntington disease-like 2	JPH3	CTG	n.a.	Toxic RNA		16q24.2	chr16:87,602,905-87,698,156	3’UTR; alternatively spliced exon-2A	6 to 28	29 to 39	41-58	Autosomal dominant	rare	14
FXTAS	Fragile X-associated tremor/ataxia syndrome	FMR1	CGG	n.a.	Toxic RNA	RAN translation/poly Gly (S)	Xq27.3	chrX:147,912,052-147,951,125	5’UTR	5 to 55	NA	55 to 200	X-Linked	1:4000 (males); 1:7800 (females)	84
FRAXA/FXS	Fragile X syndrome	FMR1	CGG	n.a.	Loss of function/gene silencing		Xq27.3	chrX:147,912,052-147,951,125	5’UTR	5 to 55	NA	> 200	X-Linked	1:3600-4000 (males); 1:40006000 (females)	85
FRDA	Friedreich’s ataxia	FXN/X25	GAA	n.a.	Loss of function		9p	chr9:69,035,259-69,074,213	Intron-1	5 to 33	34 to 65	66 to ~1300	Autosomal recessive	1:50,000	15
FRAXE	Fragile XE syndrome	FMR2/AFF2	CCG	n.a.	Loss of function/gene silencing		Xq28	chrX:148,500,619-149,000,663	5’UTR	<50	5-200	>200	X-Linked	1:50,000 to 1:100,000	81, 148

References: (1), [6]; (2), [7]; (3), [8]; (4), [9]; (5), [10]; (6), [12]; (7), [147]; (8), [4]; (9), [5]; (10), [13]; (11), [11]; (12), [21]; (13), [14]; (14), [84]; (15), [85]; (16), [15]; (17), [81, 148].

I need these refs here here for them to format correctly.

The Mechanisms of how TNRs Cause Disease

TNR diseases have been classified according to the mechanism causing pathology, which is directly connected to the location of the TNRs in the affected gene [25]. Originally, most TNRs located in open reading frames (ORFs) were believed to cause disease through translation of stretches of poly amino acids to produce toxic proteins. In contrast, TNRs located in UTRs may cause pathology through toxic RNAs. In addition to these gain-of-function mechanisms, loss-of-function TNR diseases are believed to result from insertion of TNRs in either the 5’UTRs or introns of genes. We present the TNRs sorted according to these three groups and discuss, in each case, examples of multiple mechanisms that contribute to individual diseases as well as highlight alternative mechanisms.

The Contribution of Toxic Proteins to Disease Pathology

Two of the first diseases identified associated with a TNR expansion had CAG expansions: SBMA [4] and HD [5].

SBMA is caused by an extended CAG repeat present in exon 1b of the androgen receptor (AR). It is an X-linked sex-limited recessive adult-onset neurodegenerative disorder that involves the death of the spinal and bulbar motor neurons and dorsal root ganglia [26, 27]. The disease is likely caused by the fact that the normal AR gene is expressed in the brain, as shown for mouse embryonic development [28]. Patients’ age at the time of disease onset correlates negatively with the length of the CAG repeat [29]. The mean length of the CAG repeat in the AR gene in different SBMA patient populations around the world is between 44 and 48 with repeat lengths of up to 68, whereas the normal length is between 9 and 34 [27, 30].

HD is marked by progressive degeneration of neurons [1, 31]. It is caused by expansion of a CAG repeat in the huntingtin (HTT) gene. HTT is expressed in all mammalian cells, is highest in brain and testes, and deletion of HTT is embryonically lethal [32–34]. Anyone who inherits the expanded CAG TNR in the mutant (m)HTT gene in the full penetrance range (>39 repeats) will develop the disease. Similar to SBMA the length of the CAG amplifications determines the severity and age of onset of the disease [31]. Symptoms most often start between 30 to 50 years, and worsen over a 10 to 25-year period. Patients develop symptoms when their HTT gene contains more than 27 CAG repeats. Gene silencing experiments in mouse models have shown that when the expression of mutant (m)HTT is reduced, symptoms improve [35].

It was shown that the polyQ expansions in AR and HTT are toxic to cells [2, 4, 36]. The discovery of additional genes with expanded CAG repeats also supported the assumption that repeats present in exons are eventually translated into polyQ stretches, which subsequently cause toxicity and neuronal tissue degradation. This was shown for SCA1 [6]; SCA2 [7], SCA3 [8], and DRPLA [3].

A connection between mutant protein and disease progression may be best supported by data on mutant ATXN1—the CAG containing gene that causes SCA1. SCA1 was described as a polyQ disorder with a gain-of-function mechanism that involves nuclear aggregation of ATXN1 [2]. Resultant phenotypes in different transgenic mice suggested that the nuclear localization of the ATXN1 protein was required to initiate pathogenesis, whereas aggregation of ATXN1 was not [37].

Different mechanisms were proposed to explain the toxicity of mutant proteins containing extended polyQ stretches. It was suggested that mHTT contributes to HD pathogenesis by promoting the fibrillogenesis of wild-type (wt) HTT, which causes its aggregation [38]. In HD, considerable toxic activity seems to come from a proteolytic polyQ fragment of HTT [39]. Recently, disruption of the nuclear pore complex inhibiting nucleocytoplasmic transport was identified as a central component of HD [40, 41]. Most recently it was shown for SCA1 that an ATXN1-cognate partner capicua (CIC) complex is a primary driver of disease pathology through a gain-of-function mechanism [42]. In addition, in SCA1, phosphorylation of ATXN1 at Ser776 plays a role in the pathogenesis. Replacing Ser776 with a phospho-mimicking Asp converted ATXN1 with a wt glutamine tract into a pathogenic protein when expressed as a transgene in mice [43]. Subsequently, however, it was determined that while the Ser776 phosphorylation is important for neuronal dysfunction, it is not involved in neuronal death [44]. This suggests that while the protein context matters in the case of SCA1, it appears that different mechanistic aspects of the disease can have different causes. In summary, while a large number of data suggest that disease pathology in CAG TNR diseases might be caused by aggregation of mutant protein [45], there are conflicting data with this hypothesis [37, 46–49]. In most studies, involvement of mutant RNA could not completely be ruled out unless CAG TNRs were replaced with glutamine-coding CAA repeats and this has rarely been done.

The Contribution of Toxic RNA to Disease Pathology

While there is clear evidence for protein toxicity in TNR diseases, a number of observations suggest a role for RNA in the disease process [1, 50]. A shift from studying protein to RNA began with the recognition that many of the diseases-causing TNRs were located, not in the coding region of a gene, but in regions not expected to be translated [2].

The most widely studied disease caused by toxic repeat RNAs is DM1 [21], the most common muscular dystrophy presenting in adults. Distal skeletal muscles are typically involved and muscle atrophy occurs preferentially in type 1 fibers. In DM1, CTG TNRs are located in the 3’UTR of the myotonic dystrophy protein kinase (DMPK). Unaffected individuals carry less than 50 CTG TNRs in this gene and expansions of up to more than 1000 CTGs were reported in DM1 patients. A major breakthrough on how CUG repeats located in the 3’UTR of a gene could be toxic came in 1995 with the observation of nuclear foci containing the DMPK transcript in mutant cells [51]. The subsequent identification that RNA-binding proteins are present in nuclear foci [52, 53] and interact with the expanded repeat spurred a new hypothesis that repeat expansion leads to a toxic gain-of-function mediated through RNA. Such toxic RNA was found to sequester muscleblind like splicing regulator 1 (MBNL1) [54–56], which regulates alternative splicing, antagonizing the CUG binding protein 1 (CUG-BP1/CELF1) [57, 58]. The mechanism of sequestering RNA binding proteins has been shown for CUG in DM1 [56]. CAG in HD [59], and SCA3 [60]. A new mechanistic twist to RNA-induced toxicity in TNR diseases, including HD, was demonstrated in a recent report that such RNAs can undergo a sol-gel phase transition that contributes to nuclear RNA aggregate formation [61]. This suggests that TNR diseases with toxic polyQ proteins can also have a RNA component.

For SBMA, the contribution of the mutant AR protein to disease progression is well established [62]. Increasing the CAG repeat length in AR results in reduced transactivation activity of AR and promotes its degradation [63, 64]. However, disease progression seems to be regulated by androgens. Removal of testosterone from young mutant transgenic male mice blocked development of disease [65], while in older mutant males castration restored normal function [66]. Correspondingly, female mice carrying a mutant AR gene developed disease upon administration of testosterone [65]. Recently, expression of polyglutamine expanded AR in a mouse model of SBMA was shown to upregulate the gene encoding calcitonin gene-related peptide (CGRP1) [67]. CGRP1 expression was found to be high in motor neurons of patients with SBMA providing an explanation for how AR protein contributes to this disease. However, even though these data provide strong support for the contribution of the AR protein to the disease pathology, there is also room for a RNA contribution to the disease. The first evidence arose during generation of a mouse model for SBMA. After a number of unsuccessful attempts to model the disease in mice [68, 69] and the generation of transgenic mice that showed the expected pathology [70–72], a mouse model mimicking the pathology seen in SBMA patients was created by introducing a transgene of 239 CAG repeats driven by the AR promoter with no other AR sequences [73]. The mice showed disease pathology mimicking aspects of SBMA suggesting that androgen-regulated AR protein may not be the only factor involved in the etiology of SBMA. It needs to be noted that for unknown reasons in this and many other mouse models of TNR diseases disease causing repeat lengths are much larger than the alleles typically observed in affected humans. There is an alternative explanation for how AR stimulation could be driving disease without necessarily requiring AR protein. Androgen has been shown to induce upregulation of AR in a number of normal tissues such as bone and neuronal cells [74, 75]. While in many situations androgen represses its own expression, in certain disease relevant tissues androgen may drive upregulation of AR mRNA thus promoting the disease both at the protein and the RNA level. While there is strong evidence that mutant AR protein contributes to the neurodegenerative disease, in particular the last finding suggests that RNA-based toxicity can contribute to disease pathology.

Evidence to support a role for RNA in TNR expansion toxicity also comes from other model organisms. It was shown in Drosophila that the toxicity of the CAG repeat disease gene for SCA3, ATXN3, is in large part due to the TNR RNA and not the polyQ protein [76]. Replacing some of the glutamine coding CAG repeats with the other codon coding for glutamine, CAA, mitigated the toxicity despite similar polyQ protein expression levels. A direct toxicity of mRNAs with extended CAG repeats was also demonstrated in mice [77].

While the location of TNR expansions in untranslated gene regions points at the RNA being toxic, one has to recognize that repeat-associated, non-ATG translation (RAN translation) was discovered as another translation-level pathogenic mechanism of CAG repeat-containing mRNAs in a number of TNR diseases giving rise to multiple poly amino acid stretches [78] (Table 1). These peptides are the result of bidirectional transcription and translation from all three reading frames of a repeat without the need of an AUG start codon.

Loss-of-function mechanisms

In addition to localizing to the ORF or the 3’UTR of mutant genes, TNRs are also found in the 5’UTRs or in introns. In 1996, a disease-associated gene was identified containing an intronic GAA repeat expansion in Friedreich’s ataxia patients [15]. This TNR results in loss-of-function as the affected gene FXN is no longer expressed due to formation of a stable triple helix at the gene locus that blocks transcription [79]. FRAXA (CGG TNR), and FRAXE (CCG TNR) are both caused by TNRs present in their 5’UTRs. In both cases this insertion results in epigenetic silencing of the promoter reducing protein expression [80, 81]. TNR pathology can also be caused by increasing gene expression. This is found in SCA12 [11]. In this disease a CAG TNR present in the 5’UTR of the PPP2RE2B phosphatase resulting in increased gene expression, at least in vitro [82].

One of the most interesting cases, mechanistically, is the situation found with the CGG TNR inserted into the 5’UTR of the FMR1 gene. Depending on the repeat length within this gene, three different diseases can arise [83]. The normal repeat length is <55 CGGs. At a repeat length between 55 and 200 (the premutation) elderly carriers can develop the progressive neurodegenerative disorder fragile X-associated tremor/ataxia syndrome (FXTAS) caused by toxic RNA. [84]. Females who carry the premutation are also at risk for FXPOI [16]. However, a repeat size above 200 CGGs results in fragile X syndrome (FXS/FRAXA) [85], and this is caused by silencing of the FMR1 gene through methylation of the repeat and the promoter region. Taken together, the data suggests that just because a TNR is located in specific sites in a gene does not allow one to predict whether pathology arises from protein or RNA components.

Involvement of RNAi in TNR Diseases

RNA toxicity in TNR diseases is thought to be mostly mediated through the formation of hairpin structures [86]. While such hairpins have been posited to be toxic through the recruitment of RNA-binding proteins to nuclear RNA foci [54], they can also be processed into small interfering RNAs [87]. The RNA interference (RNAi) machinery converts double-stranded RNAs into small RNAs that negatively regulate gene expression [88, 89]. Endogenous small RNAi-active RNAs are micro(mi)RNAs, which are small 19–22 nucleotide long noncoding RNAs generated in the nucleus from longer precursor primary miRNAs. These primary miRNAs are processed by the microprocessor complex Drosha/DGCR8 and exported into the cytosol where they are further trimmed by Dicer/TRBP. Mature miRNAs are then loaded into the RNA-induced silencing complex (RISC) [90, 91]. Once in the RISC, the single-stranded miRNA guide strand then regulates expression of certain genes by targeting in most cases the 3’UTR, resulting in either degradation of the mRNA or translational silencing through various well-studied mechanisms [92, 93]. The extent of reverse complementarity determines the level of interaction between a miRNA and its mRNA targets. Small interfering RNAs (siRNAs) act through a similar mechanism as miRNAs. However, they are usually designed by investigators to be fully complementary to the targeted gene and result in degradation of the targeted mRNA.

There is convincing evidence that CAG/CUG TNRs can give rise to RNAi-active small RNAs. In human neuronal cells, expression of the CAG expanded exon 1 of HTT (above the threshold for complete penetrance which is >40) [77] caused an increase in small CAG repeat-derived RNAs (sCAG) of about 21 nt in length. Above a certain length, CAG/CUG repeats were found to be cleaved by Dicer and to have RNAi activity. For a TNR sequence to be cleaved by Dicer it must form a hairpin stem structure that resembles a miRNA. TNRs of the CNG type (CAG, CUG, CCG and CGG) have been shown to be able to form stable hairpins that can be cleaved by Dicer [87]. CNGs of 17 repeats in length were shown not to be cleaved. However, CNG repeats as short as 25 repeats were Dicer substrates [87]. Interestingly, both CAG and CUG repeat structures were found to be much more susceptible to Dicer cleavage than those composed of CCG or CGG repeats. In the context of entire mRNAs, 21 to 26 nt long Dicer cleavage products were produced from SCA1 transcripts with 53 CAG repeats, or HTT transcripts with 21 to 44 CAG repeats. In contrast, DMPK transcripts with 15 CUG repeats (normal) were not processed by Dicer, yet transcripts with 80 CUG repeats (disease) were [87]. Hence, Dicer activity controls the levels of the mutant transcripts, and at the same time produces small RNAi-active CAG/CUG-derived double stranded RNAs. One study identified sCAGs as a disease-causing agent. sCAGs, isolated from HD human brains, when transfected into neurons reduced viability [50], suggesting that these sequences affect cell viability through RNAi by targeting genes that regulate cell survival. RNAi-active sCAG RNAs were also isolated from a R6/2 HD mouse model that expresses human HTT with 90 CAG repeats and were shown to kill neuronal cell lines after transfection [94].

Compelling evidence of sCAG RNAs contributing substantially to disease pathology came from studies that used locked nucleic acid–modified antisense oligonucleotides complementary to the CAG repeat (LNA-CTG), an inhibitor that unlike siRNAs or conventional antisense oligonucleotides, blocks the activity of sCAGs without reducing the level of either HTT mRNA or protein [94]. In the R6/2 HD mouse model, treatment with LNA-CTG produced a rapid and sustained improvement of motor deficits.

Direct evidence for involvement of RNAi in the toxicity of TNR-based small RNAs was provided in a study that demonstrated that pathogenic ATXN3 with amplified CAG repeats showed strongly enhanced toxicity in HeLa cells after knockdown of Dicer [95]. In addition, it was previously shown in Drosophila that impairing miRNA processing dramatically enhanced neurodegeneration caused by the CAG repeat gene ATXN3 [96]. This data can be interpreted as sCAGs getting better access to the RISC in the absence of most miRNAs. Indeed, these authors concluded that miRNA pathways normally play a protective role on TNR induced pathology.

We recently screened the toxicity of all possible TNR-derived siRNAs on two cancer cell lines (one human and one mouse) [97]. We found that siRNA based on the members of one family of TNRs (three frames: CAG, AGC, GCA and their reverse complements, CUG, UGC, and GCU) effectively killed all transfected cancer cells in quantities as low as 10 pM [97]. A hybrid duplex between the CAG and CUG based disease molecules was also previously tested in a well-established Drosophila model of DM1 [98]. Expression of the two transcripts led to the generation of Dicer-2 (Dcr-2) and Argonaute-2-dependent 21-nt TNR-derived siRNAs resulting in high toxicity to the cells. In a separate study, it was shown that expression of these complementary repeat RNAs leads to Dcr-2-dependent neurodegeneration [99]. These results suggest that co-expression of CAG and CUG repeat-derived sequences may dramatically enhance toxicity in human repeat expansion diseases through RNAi in which anti-sense transcription occurs. Antisense transcription that has the potential to create CAG/CUG repeat duplexes occurs in SCA8 in two genes that overlap on these TNRs: ATXN8 (CAG repeat), on the sense strand and ATXN8OS (CTG repeat) on the antisense strand [100].

TNR Diseases and Cancer

The majority of the TNR diseases belong to the CAG/CUG family [101] (Table 1). For most of the diseases, a connection to cancer has not been established, likely due to the fact that patient numbers in many of these diseases are too small to correlate disease genotype with cancer incidences. However, for a number of the CAG/CUG TNR diseases such clinical analyses could be performed.

Patient populations in Sweden with CAG-based TNR diseases had an overall reduced risk of cancer (1510 patients with HD, 471 with SBMA, and 3425 with HA; standardized incidence ratios (SIR) were 0.47 (95% confidence interval (CI) 0.38–0.58), 0.65 (0.45–0.91), and 0.77 (0.70–0.85), respectively). This affected cancer of the digestive tract (HD, SBMA, hereditary ataxia [HA] - encompassing spinocerebellar ataxia types 1, 2, 3, 6, 7, and 17), the lung (HD, HA), the breast (HD), female genital organs (HD, HA), prostate (HD, HA), hematopoietic, lymphoid tissues (HD), and other sites (HD) [102]. Patients in France had a generally reduced risk of cancer (372 patients with HD and 134 with SCA; SIRs were 0.21, 95% CI 0.13–0.31) in HD and 0.23, 95% CI 0.08–0.42 in SCA patients) [103], and patients in Denmark (all HD) were reported to have reduced cancer (694 HD patients; SIR 0.6 with a 95% CI 0.5– 0.8). This affected cancers of the digestive organs, lung, breast, female genital organs, male genital organs, urinary system, skin, lymphatic and hematopoietic tissue, and other sites [104]. An HD patient cohort in England was reported to have reduced cancer (4865 HD patients; SIR 0.71, 95% CI 0.61–0.83) of the breast, colon, kidney, lung, esophagus, pancreas, rectum, stomach, and the upper gastrointestinal tract. Malignant melanoma, non-melanoma skin cancer, and Non-Hodgkin’s lymphoma was also reduced. In addition, secondary cancers were also less frequent [105]. Finally, a strong inverted relationship between CAG repeat length and cancer incidence was also reported for SCA3 (154 patients; SIR 0.27, 95% CI 0.13 to 0.58) [106].

There are a few studies that did not find an inverse correlation between CAG length and cancer incidence. One study reported that mHTT accelerates breast tumor development and metastasis through ErbB2/HER2 signaling, both in patients and in a HD mouse model [107]. Another study found an increased risk of melanoma in HD patients [103]. The reason for these discrepancies is unclear at this moment but it could be that certain cancers benefit from having either the mutant RNA or protein expressed. In addition, the tumor suppressive effect of TNR sequences seems to be restricted to CAG TNRs and is not found with CUG TNRs. Patients with DM1 were reported to have increased risk of skin cancer (1061 DM1 patients; hazard ratio 55.44, 95% CI 3.33–8.89) [108] or cancer of various anatomic sites (1658 myotonic muscular dystrophy patients; SIR 2.0; 95% CI 1.6-2.4) [109]. More recently, it was suggested that this increased risk may be caused by downregulation of miR-200c/141, through an unknown mechanism particularly in women and for gynecologic, brain, and thyroid cancer [110]. In contrast the miR-200 target BMI1, an oncogenic polycomb protein, was upregulated in these patients possibly outweighing the effects of a CUG derived toxic siRNA.

Perhaps the most convincing inverse relationship between CAG repeat length and the risk/severity of a human cancer was reported between the CAG extension in AR and prostate cancer (PCa). Of all TNR containing genes, AR is the gene with the longest continuous CAG repeat in unaffected individuals. It is located in the amino-terminal domain. The number of CAG repeats in AR is polymorphic; the average number is 22 in Caucasian males [111]. The CAG repeat length in the AR has been inversely linked to cancer. Carriers of short polymorphic CAG repeats have increased susceptibility to prostate cancer [112]. A shorter CAG repeat has been shown to result in a two-fold increased risk of PCa [113], a more aggressive disease, a high risk of distant metastases [114–116], and a younger age of disease onset [111]. The odds of having <17 CAG TNRs is 8-times greater in men with lymph-node positive disease [117]. In contrast, a longer repeat (>20 CAGs) confers a protective effect among the PCa patients 45 years or older [118]. Most significantly, shortening of CAG repeat length was found in in situ lesions of PCa and its possible precursors [119], suggesting that PCa selects for shorter repeats. Interestingly, not only prostate cancer is affected by the TNR expansion in the AR gene. Shorter AR CAG repeats length and a resulting increased AR activity are associated with increased ovarian cancer risk [120].

These last two observations are consistent with our recent report on siRNAs based on CAG/CUG TNRs (siCAG/CUG) that are super toxic to all cancer cells [97]. We did not see any toxicity to normal cells in mice treated with siCAG/CUG coupled to nanoparticles [97]. This suggests that cancer cells are more sensitive to this form of cell death than normal cells, possibly because normal cells are protected from RNAi-mediated toxicity by miRNAs which are more highly expressed in normal cells than in cancer cells [121]. This interpretation is also supported by our data demonstrating that toxic siRNA derived from the gene FASLG when delivered in vivo affected cancer cells without toxicity to normal cells [122]. Mechanistically we showed that titering in a non toxic siRNA ameliorated toxicity of such a toxic siRNA to cancer cells [123] and as a reverse we demonstrated that when most miRNAs were eliminated by knocking out Drosha or Dicer, cells became hypersensitive to any toxic siRNA [97, 124].

Even though the toxicity in the TNR diseases is ultimately detrimental to the affected patients, genes carrying the TNR amplifications (i.e. AR or HTT) are often abundantly expressed in multiple tissues (Figure 1). However, the disease only affects specific tissues and only after years of exposure suggesting that most tissues are relatively resistant to the toxic effects of either the mutant proteins or the extended RNAs, assuming that most tissues have the machinery to process long repeat containing transcripts and utilize them in the RNAi pathway. An evolutionary trade-off between reduced cancer and HD has been discussed before [105]. In fact, HD has been recognized as one of a number of diseases that may persist in the human population because the disease-causing mutation may provide an evolutionary advantage, a phenomenon called antagonistic pleiotropy [125]. Based on these data we recently proposed to use siRNAs based on the CAG TNR to treat cancer (Figure 1).

Figure 1: CAG-based TNR diseases point at a new form of cancer treatment.

Left: CAG TNR disease patients have reduced rates of multiple cancers. Symptoms in neurodegenerative diseases usually do not manifest in the first few decades of life and the disease often affects selected tissues, usually the brain. This suggests that the disease-causing agent is not generally highly toxic to most cells. Right: Recently, it was found that CAG-derived siRNAs are highly toxic to all cancers and can be administered to mice to reduce tumor growth without signs of toxicity [97]. This suggests an evolutionary trade-off (disease outbreak but reduced cancer rate) and a possible use of TNR-based siRNAs as a novel form of cancer therapy.

A Model to Explain Reduced Cancer Incidences in Certain CAG-based TNR Diseases

In our recent analysis the toxicity of all 60 TNR-based siRNAs to cancer cells correlated well with the number of reverse complementary TNR repeats in the ORFs of genes [97]. Interestingly, siRNAs based on either CAG or CUG triplets were the most toxic TNRs out of all TNRs tested. In addition, both CAG and GUG repeats (>18 nt in length) are highly abundant in both human and mouse coding genes - with CUG being the most abundant of all 60 TNRs in both humans and mice [97]. Based on these findings we are now proposing a model in which TNR expansions are part of an anti-cancer mechanism (Figure 2). In order to understand the rationale behind the model, one needs to understand how TNR expansions occur.

Figure 2: A model to explain the reduced cancer incidences in certain CAG-based TNR diseases.

Genotoxic or oxidative stress as well as hypoxia all often found in neoplastically transformed cells, cause an increase in the length of both CUG and CAG repeats in certain genes through replication slippage (I). Both TNRs are transcribed and these transcripts are subject to Dicer processing resulting in short RNAi-active CAG (sCAG) and CUG (sCUG) RNAs (II). sCAGs attack multiple CUG TNR containing genes through RNAi (III) that in combination are critical for cell survival, particularly of cancer cells (IV). sCUGs may also be RNAi active but the consequences of this activity are not clear (?). In addition, there are more genes in the human and mouse genomes that contain long CUG TNRs (>18nt) than genes containing long CAG TNRs that could be targeted [97]. Because both CAG and CUG repeats are unstable and may increase in size with progressing genomic instability, this system may function as a mutational/cancer sensor eliminating cancer cells, resulting in an apparent reduction of cancer incidences in affected patients.

Repetitive regions such as TNR expansions are prone to slippage during DNA replication and this can result in a small loop in either DNA strand. With impaired mismatch repair function, these loops may become permanent, and alleles of different sizes can be formed during the next round of replication [126, 127]. Inactivation of mismatch repair genes (i.e. hMLH1, hMSH2, hPMS1, hPMS2, GTBP/hMSH6) is found in cancer. It is responsible for the microsatellite instability seen in more than 90% of hereditary non-polyposis colorectal cancers (HNPCC), also known as Lynch syndrome [126]. Two components of mismatch repair machinery (MutSβ, a heterodimer of MSH2 and MSH3 and MutL, a heterodimer of MLH1 and PMS2) have been directly linked to CAG/CTG repeat expansions [128–130]. While MutSβ plays a major role in the repair of larger insertion/deletion mispairs [131], it was also shown to be essential for CAG/CTG repeat expansions in mice [132, 133]. Mechanistically, MutSβ recognizes (CAG)n or (CTG)n hairpins formed in the nascent DNA strand, and recruits Pô to the complex, which then utilizes the hairpin as a primer for extension, fixing the hairpin structure and leading to CAG/CTG repeat expansion. All these mechanisms make CAG/CUG repeats highly unstable. Consequently, at least 10% of HD cases are due to a new mutation [134, 135].

We are proposing that MutSβ is tumor suppressive in two ways: 1) Through its activities in p53−/− mice by being involved in double strand break repair [136] through mediating ATR activation [137]. 2) By driving expansion of CAG/CTG repeats by forming a complex with Pô binding to (CAG)n or (GTG)n hairpins.

In general, repeat expansions can be triggered by a number stresses such as cold, heat, hypoxic, and oxidative stress [14], some of which arise in cancerous cells. H₂O₂ exposure was shown to cause a significant increase in repeat length [138]. Direct evidence that oxidative stress causes TNR expansions in vivo was obtained for the R6/1 HD mouse model [138]. In addition, repeated exposure to genotoxic agents inducing oxidative DNA damage also gave a significant and dose dependent increase in somatic triplet expansion. Increased oxidative stress is also found in HD individuals, and oxidative DNA damage preferentially accumulates at CAG repeats in a length dependent manner in HD mouse models [139–141].

The pathway for stress-induced TNR mutagenesis shows intriguing parallels with pathways that lead to cancer. Cancer cells survive in hostile environments by increasing the expression of key stress response factors (reviewed in [142]). One of the consequences of stress response factor expression in hypoxic cancer cells is increased mutagenesis [143], similar to the stress induced mutagenesis of TNRs discussed here. Pathways for tumorigenesis and TNR mutagenesis may therefore intersect at TNR repeats. Interestingly, TNR repeats are five times more frequent in cancer-related genes, and may make those genes more prone to mutagenesis [144]. With increasing genomic instability and increasing stress, CAG/CTG expansions increase in size (Figure 2, I). We are proposing that the initial event that results in repeat expansion is random and affects precancerous cells. This may involve any gene that already has CAG/CUG repeats of a certain nonpathologic length, explaining the lack of species conservation. When CAG/CUG repeats are expanded, they give rise to hairpins that are cleaved by Dicer into double stranded RNAi-active RNAs 21 to 26 nt in length (Figure 2, II). The small CAG/CUG RNAs then target mRNAs of survival genes that contain fully complementary CUG/CAG repeat regions in genes present in the genome (Figure 2, III). It was shown that the silencing effect of short CAG/CUGs depends on the length of the targeted complementary CUG/CAG repeat. It was efficient in cells expressing an EGFP reporter attached to a 70 or 200 nt long targeted repeat but not a 30 nt long repeat [87]. We found that of all TNRs present in human and mouse genes the CUG TNRs are the longest and most abundant in both genomes [97] making them the best target for CAG TNR-derived siRNAs. As a result cancer cells and normal cells undergoing neoplastic transformation will die, reducing cancer load (Figure 2, IV).

While longer CAG expansions are correlated with reduced cancer risk there is little evidence that this is the case for CUG expansions. On the contrary, it appears that longer CUG repeats correlate with an increase in cancer risk (Figure 2), even though both repeat siRNAs were highly toxic to cancer cells in our screen [97]. There are a number of possible explanations for this observation. Just like CAG TNRs, CUG TNRs may also give rise to small RNAi-active siRNAs, but since there are fewer long CAG TNR containing potential target genes [97], in vivo they may not be as toxic as CAG TNR-derived siRNAs. Alternatively, targeted CAG and CUG TNRs could be located in genes with different functions in tumorigenesis. Finally, we recently found that the 7mer seed sequence of the CAG-based siRNA is identical to one of the most tumor suppressive miRNAs, miR-15a-5p/16-1-5p (AGCAGCA). In contrast, a major oncogenic miRNA, miR-93-3p, shares a 6mer seed sequence with the CUG-containing siRNA (CUGCUG). It is therefore possible that in addition to their toxicity the CAG/CUG siRNAs may also target a set of oncogenic and tumor suppressive targets, respectively, providing a possible explanation for the apparent opposite correlations between CAG and CUG repeat expansions on cancer risk in affected patients. This can now experimentally be tested.

Our model provides an explanation for the observed reduced cancer incidence in multiple CAG TNR diseases and points at a path to develop a novel form of cancer treatment based on CAG/CUG repeats. We have recently shown that a siRNA based on CAG/CUG TNRs can be used to treat ovarian cancer in mice with no detectable toxicity to the animals [97].

Concluding Remarks

TNRs can give rise to toxic gain-of-function proteins, however, there is growing evidence that toxic RNAs are also involved in the disease etiology, irrespective of where the TNR is located in the mutated gene. Ever since the discovery of the genetic basis of TNR-induced degenerative diseases it has been a fundamental goal to target the repeat derived poly amino acids or the coding RNA as a means to treat these diseases. Promising strategies are being developed involving targeting the proteins by using translation blockers, antisense oligonucleotides, or siRNAs (reviewed in [145, 146]). It is now becoming apparent that RNAi may also be a disease-causing mechanism. Data on the connection between TNR diseases and reduced global cancer incidence and the discovery that TNR-derived siRNAs have the potential to kill all cancer cells, now points at a new field of study located at the interface between neurodegenerative diseases and cancer. This could be beneficial to the understanding of disease progression and the development of new treatments for TNR-based diseases, and for the development of a new form of cancer treatment. What needs to be considered is the fact that the mechanisms that cause neurodegenerative diseases and affect cancer could be separate. One could involve toxic proteins/RNAs and the other RNAi. Future studies (see outstanding questions) will have to address the precise contribution of protein, RNA and other mechanisms to the disease progression and cancer.

Box -. Outstanding Questions.

Why are all CAG/CUG derived siRNAs toxic to cancer cells? There are 64 different TNRs but most of them are not toxic when delivered as siRNAs.

What could protect normal cells from this toxicity? Cells lacking most miRNAs are hypersensitive to toxic siRNAs.

Why do patients with certain TNR diseases have reduced while others have increased cancer incidence? Do TNR-derived small RNAs differentially target tumor suppressors and oncogenes?

Are nonconserved CAG/CUG TNRs found throughout the genome part of an anti-cancer mechanism? The location of TNRs in the genome is poorly conserved - including in exons - suggesting that they may not be required for protein function. Are small CAG or CUG derived RNAs ever released in cells to kill cancerous cells?

Can si- or shRNAs targeting TNR repeats be used to treat cancer patients? Is there a therapeutic window?

Can the TNR-cancer connection be modeled in a disease mouse model? Will Huntington’s disease or SBMA mouse models have reduced cancer incidences?

Box 1: TNRs in the Genome of Higher Organisms.

While the CAG repeat amplifications in the human HTT and AR genes can cause neuropathology, their location and length are not well conserved between human and mouse (Figure I, Text Box 1). Despite this fact, overall both genes show a high degree of sequence similarity between these two species. The sequence of HTT outside of the CAG repeat region has more than 86% sequence identity at the nucleotide level, that of the AR gene is 83 to >90% identical outside three TNR regions in this gene (data not shown). Even in monkeys and great apes, closely related to humans, the CAG repeat length is only poorly conserved (Figure 2, bottom). This argues against polyQ stretches playing an important role in these otherwise highly conserved proteins. The observation of low conservation can be extended to other TNRs and genes. We recently analyzed the open reading frames (ORFs) and 3’UTRs of all human and mouse genes for the presence of CTG repeats (the targets of CAG repeat-based siRNAs) of >18 nucleotides in length [97]. We found between human and mouse, a total of 92 genes carrying such longer repeats in their ORF. Of these genes only 7.6% were shared between the two species. In addition, among the 31 mouse and human genes carrying CTG repeats longer than 18 nucleotides in their 3’UTRs, none of them were conserved between the two species. These data suggest that while different mammalian species all have TNRs, they may not play an important role in either normal protein function or gene regulation. This raises the question of why these sequences exist and what their role is in the genome. To begin to address this question one may have to ask how these sequences are being generated. One observation in mice expressing TNRs as transgenes and consistent with patient data of multiple diseases is that they are notoriously unstable, and the length can vary from one generation to the next [149, 150]. There are a few theories that explain how TNRs are created. According to one theory they are generated by replication slippage [150], and according to another by defects in the single nucleotide repair machinery [151].

Figure I, Text Box 1: The location and length of the CAG TNRs in the HTT and AR genes are poorly conserved among mammalian species.

DNA alignment of the regions in human HTT (top) and human AR (bottom), containing the CAG TNRs in exon-1 with other primates and murine orthologs. The degree of conservation is indicated. X = nucleotide position without consensus.