Queen Mary University of London, Centre for Neuroscience and Trauma, Blizard Institute of Cell and Molecular Science, Barts and The London School of Medicine and Dentistry, London, UK
Queen Mary University of London, Centre for Neuroscience and Trauma, Blizard Institute of Cell and Molecular Science, Barts and The London School of Medicine and Dentistry, 4 Newark Street, London E1 2AT, UK.
The cancer phenotype can be described in terms of classic features, such as limitless replicative potential, evasion of apoptosis, tissue invasion and metastasis, self-sufficiency from growth signals and sustained angiogenesis 1. These features arise from the accumulation of genetic and epigenetic aberrations that lead to deregulation of normal regulatory cellular pathways. Recent advances in molecular biology have enabled the detailed analysis of these aberrations, leading to an improved understanding of malignant transformation and the identification of novel therapeutic targets.
Nuclear organization represents the dynamic three-dimensional architecture of the genome and its various regulatory components. It is now recognized as playing a fundamental role in regulating the activities of the genome and also as being abnormal in cancer 2–4. Changes in nuclear shape, nucleoplasmic texture and number of nucleoli are traditionally used to distinguish cancer cells from their normal counterparts at the cytological level 4, 5. However, the meaning of these cellular changes for the cancer phenotype is not understood. In this review, we evaluate the significance of the spatial organization of the genome as well as transcriptional regulation by nuclear matrix proteins in cancer cells.
Nuclear organization in normal and cancer cells
Chromosomes are packaged by hierarchical folding of chromatin into chromosome territories, which occupy discrete regions of the nucleus 6 (Figure 1). Interspersed between the chromosomes are a variety of discrete regulatory domains, such as PML bodies, Cajal bodies, splicing speckles and transcription factories (reviewed in 7). Chromosome territories occupy preferred radial positions that differ according to nuclear shape 6, 8–10. In round nuclei, such as those in lymphocytes, the radial organization of chromosome territories correlates with gene density. This is seen most clearly from chromosomes 18 and 19, which are of similar size, but chromosome 19 has double the number of genes compared to chromosome 18. In lymphocytes, chromosome 19 is positioned in the nuclear interior while chromosome 18 is at the periphery 11. This arrangement has been evolutionally conserved over 30 million years 12. In contrast, in oval-shaped nuclei, such as those in cultured fibroblasts, the radial organization correlates with chromosome size, so that small chromosomes are positioned in the nuclear interior and large ones at the periphery 13.
There are numerous reports of active genes being preferentially located at the nuclear interior and inactive genes at the periphery 14–16. These have led to the suggestion that the nuclear interior is generally per- missive for transcription, while the nuclear periphery is transcriptionally repressive, possibly due to the presence of heterochromatin at the nuclear lamina. More recent evidence indicates that the nuclear interior and periphery are not exclusively transcriptionally permissive or repressive, and that the overall determinant of radial chromatin positioning is local gene density 17–19.
The architecture of the chromosome territory has also been scrutinized in relation to transcriptional activity, since several reports suggest that expressed genes are preferentially located at the chromosome periphery, while repressed genes are in the dense interior 6, 20–22. According to the chromosome territory–interchromatin compartment model, chromosome territories are composed of a complex configuration of compact chromatin domains with a network of spaces between the compact chromatin and between adjacent chromosomes 23. In this model, active genes are positioned at the periphery of the compact chromatin subdomains and between adjacent chromosomes where they are exposed to regulatory molecules and transcription factories. A recent electron microscopy study that found significant intermingling of chromatin has led to an alternative model, the interchromosomal network model, in which there is no chromatin-free space 24. The authors propose that intermingling may facilitate interchromosomal translocations, which is described further in the next section.
Many reports indicate that the radial organization of chromosomes and genes can be altered in cancer. For example, while chromosomes in cancer cell lines with near-normal karyotypes are generally positioned according to their gene density, up to 31% of nuclei in some of these lines contain an inverted pattern of chromosomes 18 and 19 25. A similar inversion of chromosomes 18 and 19 has been found in about one-third of papillary thyroid carcinoma cells but not in normal thyroid cells 26. Furthermore, while activated oncogenes are found in some cancers to be preferentially located in the nuclear interior when compared to control cells 27–29, others appear at the nuclear periphery 30. In an extensive study of a cultured model of breast cancer, genes were identified which become repositioned during early tumourigenesis. However, there was no correlation between altered gene position and transcription for many of the genes tested 31. Although the significance of these findings for gene regulation is unclear, such studies might in time lead to the identification of nuclear markers for early transformation. The spatial organization of chromosomes will be discussed below, as it is important for our understanding of how genetic alterations occur in cancer cells.
Generation of chromosome rearrangements
Genetic alterations, consisting of mutations and chromosome rearrangements, are crucial to malignant transformation, as they give rise to abnormal gene expression and to genes with novel functions 32–35. The most common chromosome rearrangements are translocations, duplications, amplifications and deletions. Translocations often generate gene fusions with oncogenic activity; duplications and amplifications generate additional copies of genes; and deletions remove tumour suppressor genes and can also generate gene fusions. Many structural rearrangements are specific and recurrent for certain malignancies, enabling the development of molecular classification systems. Most importantly, identification of the affected genes is leading to advances in targeted therapeutic strategies. Here, we discuss features of nuclear organization that are important in the generation of gene fusions.
The genome endures double-strand breaks (DSBs) at every cell cycle, arising from exogenous agents such as ionizing radiation and intrinsic factors such as defects in DNA replication 36. DSBs are detected by the Mre11–Rad50–Nbs1 complex, which then recruits the ataxia–telangiectasia mutated (ATM) kinase, and post-translational modifications of the histone variant H2A.X occur in the surrounding chromatin 37, 38. These modifications act as a switch, either to promote recruitment of the repair machinery or to induce apoptosis. Sites of DNA repair are apparent as stable discrete foci where repair factors gather and constantly diffuse to and from the surrounding nuclear space 39. If there is a malfunction in the repair process and apoptosis does not occur, illegitimate joining of DSBs from different genomic regions may produce genetic rearrangements that give the cell a selective growth advantage.
Two models have been proposed for the formation of translocations in cancer 40, 41. In the ‘breakage-first’ model, genomic regions containing DSBs move in the nucleus and can illegitimately recombine with other genomic regions that contain other DSBs they may encounter. In the ‘contact-first’ model, local recombination events arise at DSBs in chromosomes that are already proximally positioned. Evidence from various approaches points to the ‘contact-first’ model being correct. For example, after a DSB has formed, it remains stably positioned in the nucleus by the DNA-end binding protein Ku-80, although the surrounding chromatin becomes decondensed 42. Furthermore, the frequency of observed recurrent gene fusions arising from translocations in cancer cells correlates with the frequency of side-by-side pairing of the relevant genes in normal control cells. These include BCR and ABL (fused in chronic myeloid leukaemia); PML and RARA (acute promyelocytic leukaemia); MYC and IGH, IGK or IGL (B cell leukaemia/lymphoma); IGH and CCND1, BCL2 or BCL6 (B cell leukaemia/lymphoma) 43–48.
Recombination events involving adjacent chromosomes must take place at the chromosome territory boundaries (Figure 2). There are several lines of evidence that transcriptional activity is important in the generation of recurrent translocations. In B lymphocytes, MYC and IGH frequently occupy the same transcription factory, thus leading to their close juxtaposition and facilitating recombination after the formation of DSBs 49. In the study mentioned above from which the interchromosomal network model was derived, the amount of interchromosome intermingling correlated with both transcriptional activity and the frequency of translocations between the chromosomes 24. As transcriptionally active loci appear to be located at the boundaries, these correlations also support suggestions that active loci have a greater propensity for translocation due to their ‘open’ or accessible chromatin architecture.
Gene fusions can also arise from intrachromosomal rearrangements such as inversions or DNA copy number gains. For example, the BRAF oncogene fuses with the AKAP9 and KIAA1549 genes as a result of inversions and copy number gains within chromosome 7 in papillary thyroid carcinoma and pilocytic astrocytoma, respectively 50, 51. Papillary thyroid carcinoma has similar gene fusions, such as NTRK1–TPR or RET–H4 and RET–NCO4 arising from rearrangements within chromosomes 1 and 10, respectively. Interestingly, these partner genes have been found to be proximally located in normal thyroid control cells, indicating that chromatin folding within individual chromosomes is also likely to be a factor in the formation of recurrent oncogenic rearrangements 52, 53. Some of the proteins that are involved in chromatin folding are discussed below.
Nuclear matrix proteins and cancer
Chromosome architecture is derived from successive looping and folding of chromatin, as described above. This process is mediated by proteins that are part of the nuclear matrix, which is the residual collection of proteins formed after the removal of soluble proteins with hypertonic buffers. In addition to their structural role, nuclear matrix proteins are involved in regulating gene expression, DNA replication and repair (reviewed in 54–57). The folding of chromatin facilitates interactions between remote genomic regions to enable or repress transcription, and also enables individual genomic regions to be insulated from the surrounding regions 22, 57, 58. Here, we describe nuclear matrix proteins that are themselves aberrantly expressed in cancer, and we evaluate their possible roles (Table 1, Figure 3).
Table 1. Nuclear matrix proteins and cancer. The table indicates the malignancies that have abnormal expression of these proteins
U, up-regulated compared to matched normal tissue; D, downregulated.
SATB1 (special AT-rich binding protein 1) is a homeobox protein that binds to AT-rich DNA sequences called ‘base unpairing regions’ to mediate chromatin looping 59. During T cell development, SATB1 plays a major role in gene expression by acting as a docking platform to recruit activators and repressors 60, 61. Its function as either activator or repressor depends on specific post-translational modifications of the protein 62. SATB1 is also required for transcriptional regulation of other cellular processes, such as adaptive immune responses 63.
Transfection of SATB1 was recently shown to reprogramme expression of multiple genes in a non-metastatic breast cancer cell line, leading to an aggressive cellular phenotype. Genes regulated by SATB1 include those associated with metastasis and poor prognosis, as well as genes involved in cell adhesion, apoptosis and the MAPK pathway. In primary breast tumours, high SATB1 expression was found to correlate with aggressiveness and poor clinical outcome, irrespective of lymph node status, enabling it to be considered as an independent breast cancer marker. As breast tumour growth and metastasis in mice could be prevented using RNA interference against SATB1, the protein is a potential novel target for therapy 64.
SATB2 similarly binds to AT-rich DNA sequences to regulate gene expression, and is a key regulator of neuronal and skeletal development 65, 66. High SATB2 expression is also found in aggressive breast cancer with poor clinical outcome 67. Conversely, in colorectal cancer patients, low SATB2 expression is linked to tumour invasion and distant metastases and poor prognosis and may also be suitable as an independent prognostic marker 68.
SAFB1 and SAFB2
The related homeodomain proteins scaffold attachment factor B1 and B2 (SAFB1 and SAFB2) bind to DNA at AT-rich genomic regions and mediate transcriptional regulation, RNA processing and stress responses (reviewed in 69). There are several indications that these proteins act as tumour suppressors. The SAFB1 and SAFB2 genes are located in chromosome band 19p13, where they are coordinately regulated by a bidirectional promoter. This is one of the most frequently deleted genomic regions in breast cancer, and poor prognosis tumours have low expression of SAFB1 and SAFB2 70–72. Experimental over-expression of SAFB1 and SAFB2 in breast cancer cells causes growth arrest and aneuploidy 71.
The oestrogen receptor (ER) binds oestrogen receptor elements (ERE) to regulate the expression of hundreds of target genes through the recruitment of co-activators or co-repressors 73. SAFB1 and SAFB2 function as co-repressors of the ER to create a repressed chromatin structure through indirect recruitment of histone deacetylases and other regulatory factors, including cJun, PPARα, PPARβ, PPARγ and VDR 69, 74, 75. Patients with ER-positive breast cancer have a better prognosis than those that are ER-negative, as they can be treated with oestrogen antagonists such as tamoxifen. SAFB1 and SAFB2 have increased binding affinity to EREs in the presence of tamoxifen, which in turn enhances recruitment of co-repressors 70, 72, 76. SAFB1 and SAFB2 are thus crucially involved in inhibiting transcription from ER-responsive genes, and play a central role in mediating the therapeutic effect of tamoxifen.
Polycomb group proteins
Polycomb group (PcG) proteins induce transcriptional silencing of specific sets of genes by mediating chromatin modifications 77. They are crucial in preservation of embryonic and adult stem cell phenotypes, as they down-regulate developmental genes 78–80. In embryonic stem cells, PcG proteins maintain a repressed but poised chromatin state characterized by the dual histone modifications H3K27me3 and H3K4me3 [correction made here after initial online publication], to enable subsequent activation in adult tissues 81.
PcG proteins are often over-expressed in cancer (Table 1), where it is believed they contribute to malignant transformation by repressing key developmental genes (reviewed in 80). One of the PcG proteins most strongly associated with cancer is BMI1, which was first identified as an inducer of B and T cell lymphomas through its cooperation with MYC 82. BMI1 has since been found associated with many malignancies, including medulloblastoma and glioma 83, 84 and is essential for the proliferation of cancer stem cells 85.
The mechanisms of PcG-mediated gene repression in cancer are not fully understood. In prostate cancer, downregulated genes, such as the tumour suppressor genes RARB, GAS2 and PIK3CG, are bound by PcGs and carry the dual histone marks described above 86. These genes are silenced in a promoter methylation-independent manner. However, there is evidence that methylation does play a part, as the PcG protein EZH2 directly controls promoter hypermethylation at target genes and PcGs fold hypermethylated DNA into loops, which appear to enforce epigenetic silencing 87–89. PcG proteins thus mediate their repressive effect by inducing classical epigenetic modifications and architectural changes of the chromatin.
CTCF (CTCCC-binding factor) is a ubiquitously expressed, highly conserved protein that binds DNA via an 11-zinc finger domain 90. The protein was first discovered as a transcription factor for MYC91, but is now known to insulate neighbouring genomic domains and regulate imprinting through the formation of chromatin loops (reviewed in 92). The CTCF gene maps to 16q, which is frequently deleted in breast and prostate cancer and other malignancies 93.
CTCF acts as an insulator in preventing the spread of methylation from adjacent genomic regions. In sporadic breast cancer, loss of CTCF binding upstream of the BRCA1 promoter leads to aberrant DNA methylation 94, 95. The same phenomenon occurs at the RB, RASSF1A and CDH1 tumour suppressor genes in other malignancies 96, 97. Post-translational poly-ADP-ribosylation of CTCF is essential for its insulating function 98. Defective ADP-ribosylation of CTCF destablizes binding at the methylation boundary of the p16INK4a locus in breast cancer cell lines and primary tumours, leading to epigenetic silencing through the spreading of repressive histone marks and aberrant promoter methylation 97, 99.
Loss of imprinting is found in many malignancies, including colorectal carcinoma and chronic myeloid leukaemia (reviewed in 100). CTCF appears to function in imprinting by preventing methylation at the expressed allele (reviewed in 101). In Wilms' tumour, loss of CTCF binding at the IGF2/H19 locus results in loss of imprinting through biallelic methylation 102. As this correlates with loss of one copy of CTCF in Wilms' tumour, without mutation of the remaining allele, haploinsufficiency (presence of only one functioning allele) of CTCF may be sufficient to induce defective imprinting at the IGF2/H19 locus.
RUNX (runt-transcription factor) transcription factors regulate genes involved in growth, survival and differentiation 103. RUNX1 is involved in haematopoesis, RUNX2 in skeletal development, and RUNX3 in neurogenesis and gut development (reviewed in 104–106). All three RUNX proteins contain a highly conserved 128 amino acid ‘runt’ domain that binds to DNA, as well as transcriptional regulatory domains and subcellular localization signals. They either activate or repress transcription, depending on their post-translational modifications and the availability of co-factors 107. RUNX proteins localize to discrete domains in the nucleus, emphasizing the compartmentalization of regulatory proteins to specific nuclear microenvironments 3.
Aberrant expression of RUNX genes has been reported in numerous malignancies, including breast, colon, pituitary and prostate (Table 1)(reviewed in 108). Furthermore, specific mutations of RUNX1 are found in haematological malignancies and RUNX3 in gastric cancer. These result in defective nuclear localization, with the proteins being retained in the cytoplasm or showing aberrant subnuclear targeting, and a dysregulation of hundreds of target genes 3.
The RUNX1/AML gene is one of the most common gene fusion partners in acute leukaemia 108. In childhood acute lymphoblastic leukaemia, 20–25% of cases have a TEL–RUNX1 gene fusion from a 12;21 chromosome translocation that correlates with good clinical prognosis 109. Examples of other translocations are shown in Table 1. The TEL–RUNX1 and RUNX1–ETO fusion proteins have an intact runt domain but their regulatory domains are replaced by sequences from TEL and ETO, giving rise to aberrant recruitment of co-repressors to RUNX target genes.
Nuclear matrix proteins as cancer markers
Proteins that are aberrantly expressed in specific forms of cancer can be used to improve diagnosis, predict prognosis and devise and monitor new therapeutic approaches. The serum markers prostate-specific antigen (PSA), carcinoembryonic antigen (CEA), α-feto protein (αFP) and human chorionic gonadotrophin (HCG) have been used for many years as diagnostic markers, yet they perform poorly in terms of sensitivity and specificity for normal tissue type and cancer 110. In order to find gold standard cancer markers, nuclear matrix preparations from normal and cancer cells have been examined to identify differences in protein composition. Many of the proteins that are differentially expressed have not yet been matched to known genes and the molecular mechanisms that bring about their specific expression are not understood. However, several of these proteins appear specific and are promising candidates as cancer markers (Table 2) (reviewed in 4, 111–114).
Table 2. Diagnostic nuclear matrix proteins
ECPA 1–2, early prostate cancer antigen 1–2; HMG-I(Y), high mobility group protein I(Y); CCSA 3–4, colon cancer-specific antigen 3–4; BLCA 1–4, bladder cancer-specific antigen 1–4; BLNL 1–3, bladder normal protein 1–3; NMP-22, nuclear matrix protein-22; NMBC 1–6, nuclear matrix breast cancer; P114, protein 114; CvC 1–5, cervical cancer protein; NMP-179, nuclear matrix protein-179; RCCA 1–2, renal cell carcinoma antigen 1–2; U, up-regulated compared to matched normal tissue.
This review has approached cancer biology from a novel viewpoint. As well as informing our understanding of malignant transformation and maintenance of the cancer phenotype, integration of nuclear organization with classical cancer genetics is likely to have implications for diagnostics and the development of future therapeutics. For example, targeting nuclear matrix proteins using small molecular inhibitors or activators could be highly effective against cancer cells. Treatments directed against particular post-translationally modified forms of these proteins would provide further specificity for downstream gene networks. Finally, such a comprehensive analysis is likely to throw light on the significance of the abnormal nuclear structures seen in cancer.
We thank all members of the Sheer laboratory for critical reading of the manuscript, in particular Diego Ottaviani and Tania Jones. We also thank Tania Jones for Figure 1. This study was funded by Cancer Research UK Programme Grant No. C5321/A8318.
PowerPoint slides of the figures from this review are supplied as supporting information in the online version of this article.