Challenges in Identifying Candidate Amplification Targets in Human Cancers

Abstract

Detailed genomic characterization of cancer specimens is required to identify all genes whose dysregulation contributes to tumorigenesis and/or tumor progression. These include amplification target genes, whose oncogenic functions derive from their overexpression in response to increased gene copy number, and which increasingly serve as therapeutic targets and predictive markers. We propose that identifying novel amplification target genes is becoming more challenging, and may require the comparative analysis of multiple studies mapping gene copy number changes and/or defining associations between gene copy number and expression. We therefore reviewed the array comparative genomic hybridization and single nucleotide polymorphism profiling literature to identify copy number increases that were restricted to chromosome 8q21 in human cancers, which were reported most frequently in breast cancer. We determined the minimal regions of overlap between gained regions and then examined which chromosome 8q21 genes were most frequently overexpressed, or otherwise supported, in individual studies. As these combined approaches supported the previously proposed amplification targets TCEB1, TPD52, and WWP1, the comparison of multiple genomic studies may therefore effectively predict candidate gene amplification targets, and prioritize these for further study.

Introduction

Gene amplification targets increasingly serve as cancer therapy targets, or otherwise highlight signaling pathways for therapeutic intervention.¹ Amplification target genes are by definition increased in copy number in cancer tissue compared with normal somatic tissues, and are additionally expected to map within a minimal region of genomic gain, where they may be the sole gene within the region or amplicon.² Amplification target genes are also expected to display significant positive associations between gene copy number and transcript and/or protein levels. Additional criteria further strengthen a gene’s candidature, such as associations between gene expression and clinical variables such as patient outcome, and phenotypes from candidate gene overexpression and/or knockdown studies in cultured cells or model organisms. ² While fulfilling all these criteria may require years of investigation, confirmation that a gene maps within a minimal region of genomic gain, and is frequently overexpressed in response to increased gene copy number, is sufficient to justify further experiments to either support or refute that gene’s candidature.

Given the therapeutic targeting of defined amplification targets such as ERBB2, there is strong interest in identifying additional amplification target genes. It is therefore important to consider the likely attributes of amplification target genes that await discovery or recognition. Mutational studies first highlighted the “mountain and hill” concept, namely that few genes are frequently mutated in most specimens of a given cancer type and in multiple cancers (“mountains”), whereas many individual genes are mutated at lower frequencies (“hills”).³ An analogous situation is likely to pertain to gene amplification targets,⁴ where comparatively few genes are amplified frequently, and/or at high amplitude, in most cases of a given cancer type and/or across multiple cancers. In contrast, a greater number of genes may be less frequently gained and/or gained in only specific tumor types or within patient subpopulations. Such genes could also be gained at comparatively lower amplitudes, and/or show lesser associations with copy number. As for the case of gene mutations, it seems likely that most amplification “mountains” have been identified, and the “hills” remain to be discovered. It would be easy to dismiss “hills” as being of lesser importance, but their large numbers alone argue against this.^3,4 Furthermore, “mountains” and “hills” may in fact map to common signaling pathways, where therapeutic targeting of more than one pathway member may be required for extinction of function. Identifying “hills” should therefore improve our understanding of defined signaling pathways, as well as provide new and unexpected insights into tumor initiation and progression.

It would be logical to predict that “hills” will be more difficult to identify than “mountains.” Gene copy number and expression studies frequently identify known amplification targets because these are readily identifiable within a single study and quickly recognized by researchers. It is also possible that some high-resolution and high-throughput genomic studies may be unwittingly biased towards identifying known, as opposed to novel, amplification targets. The availability of information regarding the copy number and/or expression of many individual genes encourages the use of high levels of statistical significance to prioritize candidates. However, as unknown amplification target genes may be less frequently gained, gained at lower amplitude, and/or show lesser associations between copy number and gene expression, such genes may fall below thresholds imposed for candidate gene identification.

We propose that the identification of novel amplification target genes may instead require the detailed comparison of the results of multiple studies. This could involve comparing the extents of genomic gain in order to identify minimal common regions or minimal regions of overlap, and then examining relationships between copy number and expression for genes within regions of interest. As a case study, we considered that it would be valuable to perform such an analysis for a chromosomal region where candidate amplification targets have been proposed, but are not universally recognized. We therefore reviewed the array comparative genomic hybridization (aCGH) and single nucleotide polymorphism (SNP) profiling literature reporting genomic gains in human cancers that were either largely or entirely restricted to human chromosome 8q21.

Chromosome 8q21 is a relatively gene-dense region encompassing 19.5 Mb (74,000,001-93,500,000 bp) and up to 91 genes. As will be outlined, chromosome 8q21 is frequently gained in different cancer types, and its gain has been ascribed prognostic significance. A number of amplification targets within this region have been proposed,^5-8 but genomic studies can still refer to chromosome 8q21 as lacking a clear oncogene candidate.⁹ There has never been a systematic analysis of the genomic copy number literature to determine whether unbiased genomic screening studies support already proposed chromosome 8q21 amplification target genes, or whether other genes emerge as superior candidates. It is also not known whether the same or different chromosome 8q21 genes are likely to be targeted in different cancer types.

Early Reports of Chromosome 8q21 Gain in Cancer

The advent of CGH highlighted the frequent gain of the chromosome 8q arm in breast^10-14 and prostate cancer,^15,16 where this could be the most frequently gained genomic region.^16-18 Gain of entire chromosomal arms may imply the existence of more than one target gene, and this was supported by chromosome 8q24, 8q22-q23, and 8q21 regions being found to be gained separately in breast cancers or cell lines.¹⁰ Similarly, minimal regions of overlap including chromosome 8q21 were reported in prostate cancer,¹⁵ osteosarcoma,¹⁹ ovarian clear cell carcinoma,²⁰ and serous ovarian carcinoma.²¹ Furthermore, chromosome 8q21 was the most frequently gained genomic region in prostate cancer metastases,¹⁶ and chromosome 8q21-q23 was the most frequently gained region in familial breast cancers.²² These authors also associated chromosome 8q21-q23 gain with higher tumor grade, higher mitosis number, and increased Ki67 expression.²² Chromosome 8q21 gain was significantly associated with increased risk of death in a large breast cancer cohort,²³ and was also significantly more frequent in estrogen receptor–positive breast tumors, which recurred after adjuvant tamoxifen and chemotherapy, and associated with reduced metastasis-free survival.²⁴ There is thus substantial evidence that chromosome 8q21 may be gained independently of more distal chromosome 8q regions in multiple cancer types, and that chromosome 8q21 therefore harbors one or more genes with oncogenic function(s).

Does the Literature Support Existing or Novel Chromosome 8q21 Amplification Targets?

Based on cytogenetic evidence of chromosome 8q21 gain in breast and other cancers, 3 separate 8q21 amplification target genes have been proposed: elongin C or TCEB1 transcription elongation factor B (SIII), polypeptide 1 (TCEB1) at chromosome 8q21.11,⁶ tumor protein D52 (TPD52) at chromosome 8q21.13,⁵ and WW domain containing E3 ubiquitin protein ligase 1 (WWP1) at chromosome 8q21.3.⁷ In all cases, the reported effects of gene overexpression and/or knockdown obtained in targeted studies^7,8,25,26 have supported the positive associations reported between gene copy number and expression.^5-8,26

We wished to examine whether these or other genes were supported by the results of aCGH or SNP profiling analyses, particularly when these approaches were coupled with expression profiling analyses of the same samples. We therefore conducted literature searches to identify studies identifying chromosome 8q21 gain in cancer, using HighWire and Google search engines (search term “8q21 AND cancer”) performed until March 2012. This identified 22 studies that reported genomic gains in different cancers that were largely or entirely restricted to chromosome 8q21 (Table 1). The reported gained regions typically represented minimal regions of overlap derived from comparing multiple independent samples or cell lines. The reviewed studies predominantly examined breast tumors (n = 9), followed by prostate cancer (n = 3), collections of diverse cancer types and/or cell lines (n = 2), or hepatocellular carcinoma (n = 2), with single studies examining colorectal cancer, epithelial ovarian cancer, lung cancer, esophageal squamous cell carcinoma, small bowel carcinoid tumors, or osteosarcoma (Table 1). Eight of the reviewed studies presented the extents of genomic gains in the absence of gene expression data, whereas the remaining 14 studies also reported associations between gene expression and copy number. Gene expression and copy number associations reported by reviewed studies (Table 1) were compared with those of additional unbiased genomic studies, which reported associations between copy number and gene expression or other parameters for chromosome 8q21 genes, but did not define the extent of chromosome 8q21 gains (Tables 2 -4).

Table 1.

Genomic Studies Reporting Chromosomal Gains Largely or Entirely Restricted to Chromosome 8q21 in Cancer Specimens or Cell Lines

Reference	Samples examined	Genomic platform(s) employed	Definition of genomic gains/ minimal regions of overlap	Copy number associations with gene expression presented?	Transcriptomic platform(s) employed
Man et al.29	Osteosarcomas (n = 48, from 42 patients)	Spectral Genomics BAC	Statistics	No	N/A
van Duin et al.37	Prostate cancers (n = 22), prostate cancer xenografts (n = 9), prostate cancer cell lines (n = 3)	Institutional BAC	Threshold	Yesa	N/A
Yao et al.30	Ductal carcinoma in situ (n = 9), invasive ductal carcinomas (n = 18), lymph node metastases (n = 2)	Agilent cDNA	Threshold	Yesb	N/A
Jönsson et al.45	Breast (cancer) cell lines (n = 11)	Institutional BAC	Threshold	Yes	Institutional oligonucleotide
Chin et al.33	Primary operable invasive breast cancers (n = 171), breast cancer cell lines (n = 49)	Institutional oligonucleotide	Threshold	Yes	Institutional oligonucleotide
Rodriguez et al.32	Breast cancer cell lines (n = 8)	Institutional BAC	Threshold	Yes	Invitrogen HEEBO oligonucleotide
Kim et al.34	Prostate cancer specimens, localized (n = 18) or metastatic (n = 17)	Institutional cDNA	Threshold	Yes	Institutional cDNA
Weir et al.9	Primary lung adenocarcinomas (n = 371)	Affymetrix SNP Sty I	Statistics (GISTIC)	No	N/A
Kulke et al.31	Small bowel carcinoid primary tumors (n = 14) or metastases (n = 10), from 18 patients	Affymetrix 100K SNP	Threshold	No	N/A
Marchiò et al.42	Micropapillary breast carcinomas (n = 12), invasive breast carcinomas of no special type (n = 24)	Institutional BAC	Threshold	No	N/A
Natrajan et al.43	Grade III invasive breast carcinoma (n = 95)	Institutional BAC	Threshold	No	N/A
Woo et al.38	Hepatocellular carcinoma (n = 15)	NimbleGen oligonucleotide	Statistics	Yes	NimbleGen oligonucleotide
Hu et al.56	Esophageal squamous cell carcinoma (n = 30)	Affymetrix SNP Nsp I or Sty I	Threshold	Yes	Affymetrix GeneChip HG-U133A 2.0
Kao et al.40	Breast cancer cell lines (n = 52)	Institutional cDNA	Statistics	Yes	HEEBO oligonucleotide
Holcomb et al.39	Castration-resistant prostate cancers (n = 54, from 14 patients), localized prostate cancers (n = 9)	Institutional BAC	Threshold	Yes	Institutional cDNA
Sayagués et al.44	Primary colorectal carcinomas from patients who developed liver metastases (n = 23)	Affymetrix SNP Nsp I or Sty I	Threshold	No	N/A
Staaf et al.54	HER2-positive breast cancers (n = 200)	BAC arrays, SCIBLU Genomics Centre	Statistics (GISTIC)	Yes	Published oligonucleotide and cDNA datasets
Abkevich et al.27	Cancer cell lines (n = 178, 18 tissues of origin)	Affymetrix SNP Nsp I or Sty I, Agilent 244K CGH	Threshold	No	N/A
Ramakrishna et al.35	Epithelial ovarian tumors: serous (n = 37), endometrioid (n = 14), mucinous (n = 7), clear cell (n = 9)	Affymetrix Genome-Wide Human SNP 6.0	Threshold	Yes	Affymetrix Human Gene1.0 ST
Beroukhim et al.28	Cancer tissue specimens (n = 2,520, representing 26 histological subtypes), cell lines (n = 541), melanoma short-term cultures (n = 70)	Affymetrix SNP mapping Sty I	Statistics (GISTIC)	No	N/A
Jia et al.41	Hepatocellular carcinoma (n = 58 with paired normal samples)	Affymetrix Genome-Wide Human SNP 6.0	Threshold	Yes	Affymetrix GeneChip HG-U133 Plus 2.0
Guedj et al.46	Breast cancers (n = 488)	CIT-CGH array (V6), BAC and PAC clones	Statistics	Yes	Affymetrix HG-U133 Plus 2.0, other published datasets

Note: BAC = Bacterial artificial chromosome; SNP = single nucleotide polymorphism; CGH = comparative genomic hybridization; CIT = Cartes d’identité des tumeurs; PAC = P1-derived artificial chromosome; GISTIC = Genomic identification of significant targets in cancer; N/A = not applicable; HEEBO = Human exonic evidence based oligonucleotide.

^aRT-PCR for candidate genes.

^bStatistical analysis of breast cancer SAGE libraries.

Table 2.

Reported Associations between Gene Expression and Copy Number for Chromosome 8q21.11 Genes

		Reviewed studies reporting gene overexpression associated with defined increased copy number		Other genomic studies supporting amplification target gene status
Gene symbol	Chromosome 8 location (Mb)	Tumor	Reference	Tumor	Reference
RPESP/C8orf84	74.141-74.168	Breast (t)	Chin et al.33	Breast (c)	Hyman et al.53
		Prostate (t)	Kim et al.34
STAU2	74.624-74.821	Breast (t)	Chin et al.33	Breast (t)	Adélaïde et al.48
		Ovarian (t)	Ramakrishna et al.35	Breast (t)	Staaf et al.54
UBE2W/FLJ11011	74.865-74.953	Breast (t)	Chin et al.33	Breast (c)	Hyman et al.53
		Ovarian (t)	Ramakrishna et al.35	Breast (t)	André et al.50
				Breast (t)	Natrajan et al.51
				Prostate (t)	Kim et al.34
TCEB1	75.021-75.046	Breast (t)	Chin et al.33	Breast (t)	Zhang et al.36
		Breast (c)	Rodriguez et al.32	Breast DCIS (t)	Vincent-Salomon et al.67
		Prostate (t)	Kim et al.34	Breast (t)	Natrajan et al.51
		Ovarian (t)	Ramakrishna et al.35	Prostate (t)	Holcomb et al.39
TMEM70	75.050-75.057	Breast (t)	Chin et al.33	Breast (t)	André et al.50
		Ovarian (t)	Ramakrishna et al.35	Breast (t)	Natrajan et al.51
				Breast (t)	Staaf et al.54
GDAP1	75.425-75.441	Breast (t)	Chin et al.33	Breast (t)	Adélaïde et al.48
		Prostate (t)	Kim et al.34	Breast (t)	Natrajan et al.51
PXMP3	78.055-78.075	Breast (c)	Jönsson et al.45	Breast (c)	Hyman et al.53
		Ovarian (t)	Ramakrishna et al.35	Breast (c)	Rodriguez et al.32
				Breast (t)	André et al.50
				Breast (t)	Natrajan et al.51

Note: Genes contained within minimal regions of overlap (Fig. 2) are shown in bold. t = tissue; c = cell lines; x = xenografts; DCIS = Ductal carcinoma in situ.

Table 3.

Reported Associations between Gene Expression and Copy Number for Chromosome 8q21.13 Genes

		Reviewed studies reporting gene overexpression associated with defined increased copy number		Other genomic studies supporting amplification target gene status
Gene symbol	Chromosome 8 location (Mb)	Tumor	Reference	Tumor	Reference
HEY1	80.838-80.842	Breast (c)	Kao et al.40	Breast (c)	Rodriguez et al.32
		Liver (t)	Jia et al.41
MRPS28	80.993-81.105	Breast (t)	Chin et al.33	Breast (c)	Rodriguez et al.32
		Breast (c)	Jönsson et al.45	Breast (t)	André et al.50
		Liver (t)	Woo et al.38	Breast (t)	Natrajan et al.51
		Breast (t)	Guedj et al.46	Breast (t)	Staaf et al.54
TPD52	81.109-81.155	Prostate (t, c, x)	van Duin et al.37	Breast (c, t)	Pollack et al.47
		Breast (c)	Jönsson et al.45	Prostate (t)	Paris et al.49
		Breast (t)	Chin et al.33	Breast (t)	Adélaïde et al.48
		Breast (c)	Rodriguez et al.32	Breast (t)	André et al.50
		Breast (c)	Kao et al.40	Breast (t)	Zhang et al.36
		Breast (t)	Guedj et al.46	Hematopoietic (c)	Mattison et al.52
				Breast (t)	Natrajan et al.51
ZBTB10/RINZF	81.561-81.597	Prostate (t, c, x)	van Duin et al.37	Breast (c)	Hyman et al.53
		Breast (c)	Kao et al.40	Breast (c, t)	Pollack et al.47
		Ovarian (t)	Ramakrishna et al.35	Breast (c)	Rodriguez et al.32
				Breast (t)	Zhang et al.36
				Breast (t)	Staaf et al.54
FABP5	82.355-82.359	Prostate (t)	Kim et al.34	Prostate (t)	Paris et al.49
				Liver (t)	Jia et al.41
ZFAND1	82.776-82.796	Breast (t)	Chin et al.33	Breast (t)	André et al.50
		Ovarian (t)	Ramakrishna et al.35	Breast (t)	Natrajan et al.51
		Breast (t)	Guedj et al.46	Breast (t)	Staaf et al.54
CHMP4C	82.807-82.834	Breast (t)	Chin et al.33	Breast (t)	Natrajan et al.51
		Ovarian (t)	Ramakrishna et al.35
		Breast (t)	Guedj et al.46
SNX16	82.874-82.916	Ovarian (t)	Ramakrishna et al.35	Breast (t)	André et al.50
				Breast (c)	Kao et al.40
				Breast (t)	Natrajan et al.51

Note: Genes contained within minimal regions of overlap (Fig. 3) are shown in bold. t = tissue; c = cell lines; x = xenografts.

Table 4.

Reported Associations between Gene Expression and Copy Number for Chromosome 8q21.2 and 8q21.3 Genes

			Reviewed studies reporting gene overexpression associated with defined increased copy number		Other genomic studies supporting amplification target gene status
Gene symbol	Chromosome 8q band	Chromosome 8 location (Mb)	Tumor	Reference	Tumor	Reference
E2F5	8q21.2	86.276-86.314	Breast (t)	Chin et al.33	Breast (t)	André et al.50
			Prostate (t)	Kim et al.34	Breast (t)	Natrajan et al.51
			Ovarian (t)	Ramakrishna et al.35	Breast (t)	Staaf et al.54
C8orf59	8q21.2	86.313-86.319	Breast (t)	Chin et al.33	Breast (t)	Natrajan et al.51
			Ovarian (t)	Ramakrishna et al.35
WWP1	8q21.3	87.424-87.549	Breast (t)	Yao et al.30	Breast (t)	Adélaïde et al.48
			Breast (t)	Chin et al.33	Breast (t)	André et al.50
			Ovarian (t)	Ramakrishna et al.35	Breast (c)	Kao et al.40
					Prostate (t)	Holcomb et al.39
					Breast (t)	Staaf et al.54
FAM82B	8q21.3	87.553-87.590	Breast (t)	Chin et al.33	Breast (t)	Adélaïde et al.48
			Ovarian (t)	Ramakrishna et al.35	Breast (t)	Staaf et al.54
CPNE3	8q21.3	87.595-87.642	Breast (t)	Chin et al.33	Breast (t)	André et al.50
			Ovarian (t)	Ramakrishna et al.35	Breast (t)	Natrajan et al.51
					Breast (t)	Staaf et al.54
NBN/NBS	8q21.3	91.014-91.066			Breast (c)	Hyman et al.53
					Breast (c)	Rodriguez et al.32
					Breast (t)	André et al.50
					Breast (t)	Natrajan et al.51
DECR1	8q21.3	91.082-91.133	Breast (t)	Chin et al.33	Breast (t)	André et al.50
					Breast (t)	Natrajan et al.51
			Breast (c)	Rodriguez et al.32	Breast (t)	Staaf et al.54

Note: Genes contained within minimal regions of overlap (Fig. 4) are shown in bold. t = tissue; c = cell lines; x = xenografts.

There are several issues to consider when comparing the results of genomic studies. The reviewed analyses used a variety of genomic platforms differing in resolution (Table 1), and data analysis (Table 1) and reporting varied both between laboratories and over time. Methods of data analysis could be broadly categorized as employing variable threshold criteria (in terms of amplitude, length, and/or frequency of genomic gain), or statistical methods, to identify genomic regions to be compared and/or reported (Table 1). In all cases, we compared genomic gains or regions of overlap as they were defined and reported by each individual study. However, while technical and analytical differences could affect the extent of gained genomic regions reported by individual studies, these should not impede the derivation of minimal regions of overlap across multiple studies. We also considered that it was valid to compare gains reported in different tumor types, as performed in some of the reviewed studies,^27,28 because many gene amplification targets are implicated in more than one type of cancer.² The possibility that reported gains may represent copy number variants needs to be considered, particularly as many of the reviewed studies were performed before the prevalence of copy number variants was appreciated. Genomic coordinates have also changed over time, so we identified the NCBI/hg build used to produce genomic coordinates for each reviewed study. The majority of these used the NCBI35/hg17 or NCBI36/hg18 builds, which showed identical coordinates for all compared regions of chromosome 8q21. In the case of 3 studies that used the earlier NCBI34/hg16 build,^29-31 genomic coordinates were converted to NCBI36/hg18 coordinates using the convert function of the UCSC Genome Browser. All quoted genomic coordinates are according to the NCBI36/hg18 build. Finally, where necessary, we updated the gene nomenclature used to that accepted by HUGO.

Overview of Chromosome 8q21 Regions Gained in Cancer

To first assess whether gains of different regions of chromosome 8q21 have been reported, we compared gains largely or entirely restricted to chromosome 8q21, which were at least 1 Mb in length (Fig. 1). Several regions of overlap emerged from the different studies and tumor types compared, at approximately 75 to 76 Mb, 81 to 82 Mb, 82.6 to 83.7 Mb, and 87.42 to 87.89 Mb, and these were predicted by the results of 4, 10, 9, and 4 studies, respectively (Fig. 1). Notably, 3 of these 4 regions of overlap include previously proposed amplification target genes, namely TCEB1 at chromosome 8q21.11 (75.021-75.046 Mb), TPD52 at chromosome 8q21.13 (81.110-81.246 Mb), and WWP1 at chromosome 8q21.3 (87.424-87.549 Mb).

Figure 1.

Summary of chromosome 8q21 copy number gains more than 1 Mb in length in cancer tissues or cell lines. Gained regions are indicated using horizontal lines according to the chromosome 8 coordinates below (in Mb), with the corresponding cytogenetic bands indicated on the lower ideogram. Arrows within cytogenetic bands indicate that these extend beyond the region shown. The study reporting each gained region is indicated to the left or right, or as space permitted, with colored lines and text highlighting studies examining particular cancer types. Regions of overlap between gained regions (shaded in light blue) were identified at 75 to 76 Mb (supported by 3 studies), 81 to 82 Mb (9 studies), 82.6 to 83.7 Mb (8 studies), and 87.42 to 87.89 Mb (4 studies). Three of these regions include the proposed amplification target genes TCEB1, TPD52, or WWP1, the 5′ positions of which are indicated using vertical arrows.

In order to refine these regions of overlap, we next compared gains of 1 Mb or greater with gains less than 1 Mb in length, typically reported in more recent studies. Gains further informing these regions of overlap will be described in the following sections.

Candidate Amplification Target Genes at Chromosome 8q21.11

The addition of gains less than 1 Mb in length indicated that the region of overlap at 75 to 76 Mb (Fig. 1) could be refined to 75.01 to 75.22 Mb (Fig. 2) through a gain reported in SK-BR-3 cells.³² This region includes TCEB1 (75.021-75.046 Mb), a gene that was also supported by additional breast,³³ prostate,³⁴ and ovarian cancer studies,³⁵ where TCEB1 was included within mapped amplicons and reported to be overexpressed (Table 2). While other genes within these amplicons were also overexpressed,^33,34 no other chromosome 8q21.11 gene was supported by as many individual studies as TCEB1 (Table 2). Notably, Zhang et al.³⁶ reported increased TCEB1 copy number and expression to be associated with poor breast cancer patient outcome. Thus, TCEB1 mapped within a minimal region of overlap supported by 4 independent studies and showed associations between copy number and expression in these and other studies (Table 2).

Figure 2.

Summary of chromosome 8q21.11 copy number gains in cancer tissues or cell lines. Gained regions are indicated using horizontal lines according to the chromosome 8 coordinates below (in Mb), with the corresponding cytogenetic bands indicated on the lower ideogram. Arrows within cytogenetic bands indicate that these extend beyond the region shown. The study reporting each gained region is indicated to the left or right, with colored lines and text highlighting studies examining particular cancer types. The minimal region of overlap (supported by 4 studies) is shaded in light blue and includes the proposed amplification target gene TCEB1, the 5′ position of which is shown using a vertical arrow. The position of the minimally gained region identified through consideration of gains of at least 1 Mb in length (Fig. 1) is shown as a horizontal bar below the chromosome 8 coordinate scale.

The minimal region of overlap including TCEB1 contrasted with a chromosome 8q21.12 region of overlap defined at 78.84 to 79.59 Mb (Fig. 1). The consideration of gains less than 1 Mb in length did not identify additional gains which contributed to or refined this region (Fig. 2), and the single gene that partially maps within this region (PKIA) was reported to show associations been copy number and gene expression in only one study.³⁴

Candidate Amplification Target Genes at Chromosome 8q21.13

Some 13 studies reported gains largely restricted to chromosome 8q21.13 from 79 to 84 Mb (Fig. 3A). This region includes 2 of the regions of overlap shown in Figure 1. The more proximal of these was located at 81 to 82 Mb (Fig. 1) and was defined by gains in prostate cancer specimens, xenografts, and cell lines reported by van Duin et al.³⁷ The addition of gains less than 1 Mb in length highlighted 2 smaller minimal regions of overlap (Fig. 3A). The most proximal of these was at 80.83 to 80.92 Mb, and was defined by the proximal extent of gain reported for hepatocellular carcinoma³⁸ and breast cancer³³ and the distal extent of gain reported by Holcomb et al.³⁹ in prostate cancer (Fig. 3A). This region is included in overlapping gains reported by 8 studies (Fig. 3A). The single gene within this region is HEY1 (80.838-80.842 Mb) (Fig. 3B), which mapped to gained regions and was overexpressed in breast cancer cell lines⁴⁰ and hepatocellular carcinoma⁴¹ (Table 3). Other studies reporting HEY1 gain did not examine gene expression,^9,39,42-44 and only one other study reported an association between HEY1 expression and copy number³² (Table 3). From these combined results, HEY1 might be specifically targeted in hepatocellular carcinoma, but its role in other cancers is less clear.

Figure 3.

(A) Summary of chromosome 8q21.13 copy number gains in cancer tissues or cell lines. Gained regions are indicated using horizontal lines according to the chromosome 8 coordinates below (in Mb), with the corresponding cytogenetic bands indicated on the lower ideogram. Arrows within cytogenetic bands indicate that these extend beyond the region shown. The study reporting each gained region is indicated to the left or right, with colored lines and text highlighting studies examining particular cancer types. Minimal regions of overlap between gained regions (shaded in light blue) were identified at 80.83 to 80.92 Mb (supported by 8 studies) and 81.11 to 81.29 Mb (9 studies) and include HEY1 and TPD52, respectively. The ZBTB10 and ZNF704 genes included in the amplicons reported by Beroukhim et al.²⁸ and Abkevich et al.²⁷ are also shown, with the 5′ position of each gene being indicated using a vertical arrow. The position of a minimally gained region identified through the consideration of gains of at least 1 Mb in length (Fig. 1) was defined by the gain reported by van Duin et al.,³⁷ whereas the second region is shown as a horizontal bar below the chromosome 8 coordinate scale. (B) Larger scale diagram of the minimal regions of overlap shown in A, highlighting the positions of all genes (as horizontal black boxes, with horizontal arrows to show the direction of transcription). The STMN2 and ZNF704 genes are incompletely contained within the region shown.

A second minimal region of overlap was identified at 81.11 to 81.29 Mb by comparing the results of 9 studies (Fig. 3A) and was defined by the proximal extent of gain reported by Kim et al.³⁴ in prostate cancer and the distal extent of gain in SK-BR-3 cells³² and micropapillary⁴² or luminal breast cancer.⁴³ Six of the 9 gained regions also included HEY1, but 3 others did not^32,34,37 (Fig. 3A). The minimal region of overlap includes TPD52 (81.110-81.246 Mb) (Fig. 3B), first proposed as a chromosome 8q21 amplification target in breast cancer.⁵ While several studies reporting TPD52 gain did not examine gene expression,^9,43,44 TPD52 was included in amplicons and overexpressed in breast^{32,33,40,45,46} and prostate cancer studies³⁷ (Table 3). Other breast or prostate cancer studies also reported associations between increased TPD52 copy number and gene expression^47-51 or poor patient prognosis.³⁶ Increased TPD52 copy number in cancer cell lines was supported by the identification of viral integration sites upstream of Tpd52 in a mouse lymphoma model.⁵² However, while many studies supported TPD52 as an amplification target, others reporting TPD52 gain did not associate this with increased TPD52 expression.^30,35,39,53

From the comparison of gains of 1 Mb or greater, a third region of overlap at chromosome 8q21.13 could be defined between 82.6 to 83.7 Mb (Fig. 1). However, the inclusion of gained regions of less than 1 Mb did not highlight a clear minimal region of overlap between these genomic coordinates (Fig. 3A). Similarly, no gene within this region was consistently amplified and overexpressed (Table 3).

Candidate Amplification Target Genes at Chromosome 8q21.2-q21.3

The comparison of gains of at least 1 Mb in length highlighted a fourth region of overlap at 87.42 to 87.89 Mb, which was common to gains reported in breast,^30,43 prostate,³⁴ or ovarian cancers.³⁵ This region was defined by the distal boundary of gain reported in luminal breast cancer,⁴³ and the proximal boundary reported in prostate cancer³⁴ (Fig. 1). When gains of less than 1 Mb were also considered, this region was supported by additional breast^33,42 and colorectal cancer studies,⁴⁴ and could be reduced to 87.42 to 87.53 Mb by the distal extent of gain reported by Chin et al.³³ (Fig. 4).

Figure 4.

Summary of chromosome 8q21.2-q21.3 copy number gains in cancer tissues or cell lines. Gained regions are indicated using horizontal lines according to the chromosome 8 coordinates below (in Mb), with the corresponding cytogenetic bands indicated on the lower ideogram. Arrows within cytogenetic bands indicate that these extend beyond the region shown. The study reporting each gained region is indicated to the left or right, with colored lines and text highlighting studies examining particular cancer types. The minimal region of overlap (supported by 6 studies) is shaded in light blue and includes the proposed amplification target gene WWP1, the 5′ position of which is shown using a vertical arrow. The position of the minimally gained region identified through the consideration of gains of at least 1 Mb in length (Fig. 1) is shown as a horizontal bar below the chromosome 8 coordinate scale.

The 87.42- to 87.53-Mb region includes WWP1 (87.424-87.549 Mb), previously proposed as an amplification target in prostate and breast cancer.^7,8 Overall, WWP1 mapped within defined amplicons and showed associated overexpression in breast^30,33 and ovarian cancer studies³⁵ (Table 4) and showed associations between expression and copy number in additional breast^46,48,50,54 and prostate cancer studies³⁹ (Table 4). WWP1 was also included within a 420-kb region of gain defined in a colorectal cancer cohort, where gene expression was not investigated⁴⁴ (Fig. 4). It may be significant that increased WWP1 levels have been reported in colorectal cancer biopsies versus nontumor mucosa using antibody microarrays.⁵⁵

Other regions of overlap at chromosome 8q21.2 and 8q21.3 were also considered, but these did not conclusively support additional target genes. A region of overlap at 86.62 to 86.89 Mb could be defined by a gained region reported in ERBB2-positive breast cancer ⁴³ and was supported by 4 other studies (Fig. 4). However the REXO1L1 gene within this region was overexpressed in only one study that supported the region of overlap.³⁵ Another region between 88.73 to 89.24 Mb was defined by the proximal extent of gain reported by Natrajan et al.⁴³ and the distal extent of gain reported by Ramakrishna et al.³⁵ and supported by 4 studies. However, the inclusion of gains less than 1 Mb in length did not refine this region further, and the single gene MMP16 that partially maps within this region was not supported by expression or other criteria.

Other Chromosome 8q21 Target Genes Predicted by Individual Studies

The study of Beroukhim et al.²⁸ represented a landmark in terms of the numbers of cancer samples analyzed at a high level of genomic resolution, and from diverse tumor types. We therefore considered whether the findings of this single large study supported the combined predictions of all studies reviewed. Beroukhim et al.²⁸ identified a peak region of amplification at chromosome 8q21.13 at 81.242 to 81.979 Mb (Fig. 3A). This was strikingly similar to that reported by Abkevich et al.²⁷ at 81.240 to 81.974 Mb, which occurred in approximately 3% of the 178 cancer cell lines analyzed (Fig. 3A). These regions overlap with gains reported in breast cancer,^43,46 breast cell lines,⁴⁰ colorectal carcinoma,⁴⁴ lung cancer,⁹ prostate carcinoma,³⁷ epithelial ovarian cancer,³⁵ esophageal squamous cell carcinoma,⁵⁶ and small bowel carcinoid tumors³¹ (Figs. 1 and 3A). However, these regions did not contribute to the minimal regions of overlap predicted at chromosome 8q21.13 (Fig. 3A and 3B).

While Abkevich et al.²⁷ proposed TPD52 as the relevant target gene, only the first alternatively spliced TPD52 exon was included within the defined regions of gain.^27,28 Instead, these include the full gene sequences of ZBTB10 (81.561-81.597 Mb) and ZNF704 (81.713-81.949 Mb). We therefore considered whether ZBTB10 and/or ZNF704 genes could be supported as candidate gene amplification targets. ZNF704 overexpression in association with increased copy number was reported infrequently,^35,51 and to date, this gene has not been functionally characterized. In contrast, ZBTB10 was overexpressed when increased in copy number in 3 studies,^35,37,40 and other breast cancer studies have reported associations between ZBTB10 copy number and gene expression^32,47,53,54 or patient prognosis³⁶ (Table 3). Oncogenic functions are yet to be reported for ZBTB10, whose overexpression in MCF-7 cells arrested cell cycle progression and reduced estrogen receptor-a expression.⁵⁷ ZBTB10 and ZNF704 may therefore represent passenger genes, and the inclusion of a TPD52 exon (and possibly upstream regulatory sequences) within the amplicons described by Abkevich et al.²⁷ and Beroukhim et al.²⁸ could indicate that these target TPD52. Passenger genes adjacent to target genes may alter the positions of minimal common regions such that these do not include the true target,⁵⁸ so the comparison of results obtained by multiple studies may help to distinguish passengers from targets. Ultimately, direct experimentation will be required to determine whether ZBTB10 and/or ZNF704 present chromosome 8q21.13 passengers, or drivers of genomic gain.

Summary

We performed a literature analysis in order to predict candidate gene amplification targets at chromosome 8q21, using several approaches. Firstly, we performed 2-stage comparisons of chromosome 8q21 gains reported by different studies (Table 1) to highlight a number of regions of overlap between these (Figs. 1 -4). Where gene expression analyses were performed in parallel with amplicon mapping (Table 1), we then considered which genes included in defined 8q21 gains were overexpressed (Tables 2 -4). Finally, we considered which chromosome 8q21 genes showed associations between gene copy number and expression, or other parameters, in additional unbiased genomic studies (Tables 2 -4). These approaches highlighted 3 previously proposed amplification target genes TCEB1, TPD52, and WWP1 as mapping within minimal regions of overlap (Figs. 1 -3) and showing associations between copy number and expression in both the reviewed and additional studies (Tables 2 -4). Based upon human genomic copy number variants reported to date, it seems unlikely that increased copy number at these genes reflects germline copy number variations,^59-62 although this possibility needs to be considered in future studies. Our findings therefore suggest that the detailed analysis of multiple studies may reliably predict candidate amplification targets, and that for common tumor types, the combined existing literature is likely to be helpfully predictive.

It is important to note that multiple forms of evidence were required to confidently predict candidate amplification targets. Some reviewed studies reported more than 10 discrete amplicons within chromosome 8q21, and many genes whose expression was upregulated.^33,34 This led to some 22 chromosome 8q21 genes being supported by at least 3 of the studies considered (Tables 2 -4). Similarly, equal numbers of studies supported the chromosome 8q21.11 genes UBE2W, PXMP3, and TCEB1 if associations between gene copy number and expression or patient outcome were considered (Table 2). However when gene expression results obtained in association with amplicon mapping were considered, TCEB1 emerged as the stronger candidate (Table 2). This was consistent with this gene being located within the minimal region of overlap for gains at chromosome 8q21.11 (Fig. 2).

It was also notable that no chromosome 8q21 gene was unanimously identified as a putative amplification target by all relevant studies, and even ERBB2 was not strongly supported by associations between gene copy number and expression on occasions.⁴⁰ Thus, the identities of amplification target genes may only emerge when the results of multiple studies are combined. Such an approach is clearly not infallible in that one region of overlap (Fig. 1) was not reproduced when gains of less than 1 Mb were considered (Fig. 3A and Table 3) and others (Figs. 2 and 4) did not target overexpressed genes. The latter could indicate the targeting of regulatory genomic regions as opposed to a coding gene, which could be further explored. However, our analyses suggest that comparative approaches may help prioritize genes for further study, and lead to more productive use of downstream resources. For example, careful predictions based upon genomic data may improve subsequent high-throughput gene inactivation assays, by allowing genes to be designated as either candidate drivers or passengers at the time of experimental design.

Future Perspectives

Biology has been recently transformed from a data-poor to a data-rich science.⁶³ The possibilities allowed by this transition are enormous, but history suggests that our capacity for data generation is not immediately matched by our capacity for data analysis, particularly when these activities require different skill sets and tools. This gap translates to a failure to fully realize the opportunities allowed by technological advances, which becomes more significant when data generation comes at a high cost.

This gap may be widening through the use of next- generation sequencing techniques, where some 25,000 sequenced cancer genomes will be available for analysis in the coming 5 years.⁶⁴ While next-generation sequencing has primarily been used to detect small sequence variations such as point mutations,⁶⁴ this can also provide exact measures of copy number for individual genes. Unlike array-based approaches, genomic sequencing can also reveal the physical structure of DNA amplicons, and whether these include inversions, translocations, and/or insertions.⁶⁵ For example, sequencing has recently revealed that SK-BR-3 cells contain a large 9.1-Mb tandem duplication at chromosome 8q21, which was also involved in highly amplified regions, and an interchromosomal translocation connecting chromosomes 8 and 17.⁶⁵ However, the amplified chromosome 8q21 regions reported through sequencing⁶⁵ were broader than those identified using array-based approaches.³² Thus, while array-based techniques may become less widely used for discovery, genomic profiling data should continue to invaluably complement sequencing data and contribute to its more rapid and accurate interpretation.

To facilitate the analysis of valuable genomic data by as many researchers as possible, data need to be very broadly accessible. Our analyses suggest that the aCGH and SNP profiling literature represents a rich and accessible source of information regarding the identities of critical genes. However, the potential of this literature may not have been fully exploited. While comparative literature analyses are available to all researchers, they are time intensive and can be made more so through variable and/or incomplete reporting of results, particularly for the genomic extent of gain or loss. Literature analyses are also facilitated by specialist knowledge of particular genomic regions, and not all researchers equipped with this knowledge will have the time or motivation to undertake these.

Electronic databases greatly help to distill information from the literature, and render this more accessible to a broader range of researchers. The particular databases that warehouse genomic data have been recently reviewed.⁶⁶ However, databases compiling genomic aberrations reported by different studies may only provide information regarding whether a particular gene is gained or lost in cancer, without corresponding information regarding the genomic extent of gain or loss.⁶⁶ While this approach may serve the needs of researchers interested in particular genes, this does not allow the merits of adjacent or neighboring genes to be compared, and may not allow driver and passenger genes to be distinguished. The Tumorscape database²⁸ provides a comparative visual representation of the genomic extent of gains and losses reported for individual cases, and a statistical estimate of the significance of the gain or loss of individual genes. This database is also easily accessed by untrained users. However, differences between amplicons predicted by Beroukhim et al.²⁸ and other studies reviewed here (Fig. 3A) highlight the value of comparing results obtained by different studies. There is currently no broadly accessible database that allows the comparison of amplicons and deletions reported by different investigators. Such a database would provide an invaluable resource for a broad range of cancer researchers, particularly if this also allowed cross-referencing to gene expression, mutation, and other data. Compared with the time required for literature-based analyses, such a database could greatly accelerate the identification of amplification target “hills” in genomic regions that lack broadly accepted candidates, and assist in the interpretation of next-generation sequencing data.