A systematic review and meta-analysis of somatic and germline DNA sequence biomarkers of esophageal cancer survival, therapy response and stage

Recent advances in next generation sequencing reinforce the potential for DNA sequence markers to guide esophageal cancer management. We report the first systematic review and meta-analysis, identifying 94 markers of outcome and 41 of stage. Overall, evidence was poor. Meta-analyses demonstrated outcome associations for 6 tumor and 9 germline variants: priorities for prospective evaluation.

Abstract

Introduction

There is an urgent need for biomarkers to help predict prognosis and guide management of esophageal cancer. This review identifies, evaluates and meta-analyses the evidence for reported somatic and germline DNA sequence biomarkers of outcome and stage.

Methods

A systematic review was carried out of the PubMed, EMBASE and Cochrane databases (20 August 2014), in conjunction with the ASCO Level of Evidence scale for biomarker research. Meta-analyses were carried out for all reported markers associated with outcome measures by more than one study.

Results

Four thousand and four articles were identified, 762 retrieved and 182 studies included. There were 65 reported markers of survival or recurrence 12 (18.5%) were excluded due to multiple comparisons. Following meta-analysis, significant associations were seen for six tumor variants (mutant TP53 and PIK3CA, copy number gain of ERBB2/HER2, CCND1 and FGF3, and chromosomal instability/ploidy) and seven germline polymorphisms: ERCC1 rs3212986, ERCC2 rs1799793, TP53 rs1042522, MDM2 rs2279744, TYMS rs34743033, ABCB1 rs1045642 and MTHFR rs1801133. Twelve germline markers of treatment complications were reported; 10 were excluded. Two tumor and 15 germline markers (11 excluded) of chemo (radio)therapy response were reported. Following meta-analysis, associations were demonstrated for mutant TP53, ERCC1 rs11615 and XRCC1 rs25487. There were 41 tumor/germline reported markers of stage; 27 (65.9%) were excluded.

Conclusions

Numerous DNA markers of outcome and stage have been reported, yet few are backed by high-quality evidence. Despite this, a small number of variants appear reliable. These merit evaluation in prospective trials, within the context of high-throughput sequencing and gene expression.

introduction

Esophageal and gastroesophageal junctional (GEJ) carcinoma account for 3.9% of cancer diagnoses yet 5.9% of cancer deaths [1]. Worldwide, squamous cell carcinoma (SCC) predominates but, in Western countries, incidence of adenocarcinoma is increasing rapidly [2, 3]. Treatment with curative intent involves either resection with or without neoadjuvant therapy, or definitive chemoradiotherapy with or without salvage resection. More than 5000 patients undergo esophagectomy in the United States and the UK every year, with 85% receiving neoadjuvant therapy [4, 5]. However, the majority experience complications, operative mortality remains relatively high and quality of life may be significantly impaired [6–8]. Neoadjuvant, adjuvant and definitive chemo- and/or radiotherapy also carry risk [9], and while the absolute survival benefit of neoadjuvant therapy ranges from 7% to 13% at 2 years [9], 50%–60% of tumors are resistant [10].

Prognosis overall remains bleak; even following ostensibly curative treatment 5-year survival is just 35%–45% [11–13]. This highlights limitations in our biological understanding, and our urgent need for biomarkers to predict prognosis, recurrence and sensitivity to therapy, and ultimately better personalize care. Most clinical experience with esophageal biomarkers to date has largely involved protein expression with or without sequence changes; while such markers are used to select patients for early phase trials, the sole tumor marker in routine use is ERBB2/HER2 status [14, 15]. However, rapid advances in high-throughput next-generation sequencing (NGS) have highlighted the potential role of somatic DNA sequence markers. These may function as independent markers, serve to refine or explore existing expression markers, or constitute novel therapeutic targets [16–18]. Similarly, advances in custom and genome-wide single nucleotide polymorphism (SNP) arrays have emphasized the role of germline variants in modulating cancer and treatment outcome [19, 20].

We therefore undertook the first systematic review of DNA sequence biomarkers of esophageal cancer, to systematically identify and evaluate all candidate somatic and germline DNA sequence markers of outcome (survival, recurrence, therapy response and treatment complications) and stage. We then performed meta-analysis for all markers with a nominally statistical association in at least one study.

methods

inclusion criteria

Studies eligible were those testing association between a DNA sequence marker (germline or somatic) and outcome (clinical, radiological or pathological) or stage of esophageal/GEJ cancer. Markers included germline SNPs, tumor mutations, copy number variants (CNVs), loss of heterozygosity (LOH), microsatellite instability (MSI) and chromosomal instability (CIN; alterations in ploidy). Clinical outcomes comprised survival (any measure), recurrence, disease progression and treatment complications. Radiological outcomes comprised response to therapy. Histopathological outcomes comprised tumor response and incomplete resection. Stage comprised radiological TNM staging and pathological tumor grading [21].

exclusion criteria

Studies using cell lines or expression data were excluded unless discrete tumor or DNA-specific data could be extracted. Non-English articles were excluded.

literature search

A search was performed on 20 August 2014 of the PubMed, EMBASE and Cochrane databases, in accordance with MOOSE (Meta-analysis Of Observational Studies in Epidemiology) and PRISMA guidelines [22]. The following term was used: (esophageal OR esophagus OR gastroesophageal) AND (cancer OR carcinoma or adenocarcinoma OR SCC) AND (genomic OR genetic OR genome OR pharmacogenetic OR pharmacogenomic OR amplification OR copy OR mutation OR polymorphism OR polymorphic OR variant OR deletion OR insertion OR locus OR loci OR allele) AND (outcome OR prognosis OR survival OR response OR stage OR surgery OR chemotherapy OR radiotherapy OR marker OR biomarker OR complication). The references cited by retrieved articles were also assessed for relevant articles.

study data

Data extracted were: methodology; the variant(s) and gene(s) assessed; outcome measures and population. Extraction was carried out independently by two authors (JMF and IT). Gene names were standardized (HUGO Gene Nomenclature Committee) [23]. Variants were mapped to reference SNP (rs) identification numbers (US National Library of Medicine dbSNP database; http://www.ncbi.nlm.nih.gov/snp) when not provided by searching referenced methodology, in vitro polymerase chain reaction (http://genome.ucsc.edu) with specialized SNP flank BLAST^® (Basic Local Alignment Search Tool; http://blast.ncbi.nlm.nih.gov), or New England Biocutter v2.0 (NEBcutter; http://tools.neb.com/NEBcutter2). Gene function was classified using the US National Library of Medicine Gene Database (http://www.ncbi.nlm.nih.gov/gene). For all reported associations, it was determined whether statistical significance persisted following correction for multiple comparisons (Bonferroni or false discovery rate correction), or multivariate analysis of all variables including genotypes. Were none made, post hoc Bonferroni correction was carried out [24]. For genome-wide association studies, significance was assumed at P < 5 × 10⁻⁸. For reported markers assessed by a single study, for which P was <0.05 but >corrected α, effect metrics were calculated but the marker excluded and presented in supplementary Tables S1 and S2, available at Annals of Oncology online. Those assessed by more than one study underwent meta-analysis irrespective.

evidence quality

Quality was appraised using the revised American Society of Clinical Oncology Level of Evidence (LOE) scale for biomarker research [25].

meta-analysis

Meta-analysis was carried out for all markers with a statistically significant association (uncorrected P < 0.05) reported by at least one study. For SNPs, analysis was carried out using the major common allele as reference, using genotype permutations shared by all studies. In the case of A/T and C/G substitutions, the minor variant was confirmed from study allele frequencies. Where possible, separate analyses were carried out for major methodological differences such as adjusted/unadjusted hazard ratios (HRs), genotyping methods, treatment, cell type and ethnicity (as determined by the International HapMap Consortium) [26]. Natural logarithms of HR, odds ratios (ORs) and standard errors (SEs) were extracted. In studies not presenting these, these were estimated using the methods of Parmar, or extracted from magnified Kaplan–Meier survival curves: HR and SE were estimated at constant time points; censoring was assumed to be constant and starting from the minimal follow-up period, with censored patients allocated to the appropriate time interval [27]. In six meta-analyzed studies (all nonsignificant results) [28–33], it was not possible to extrapolate statistics for all variants; an lnHR of 0 (a HR of 1) and SE of the most closely matched study (regarding cell type, size and methodology) were used to minimize selection bias. When not presented, ORs were calculated from available data. Meta-analysis was carried out using RevMan v5.2 (Copenhagen: the Nordic Cochrane Centre, The Cochrane Collaboration).

study heterogeneity and bias

Heterogeneity was quantified using I² and χ² estimates; for moderate heterogeneity (I² ≥ 50%) random rather than fixed-effects models were used. Heterogeneity and bias were also assessed by funnel plot asymmetry; [34] visually for all analyses, and statistically for analyses involving at least 10 studies [34, 35] using Begg's and Egger's tests. Statistical significance was assumed at P < 0.05. Following consideration of alternative causes, probable publication bias was corrected using the ‘trim and fill’ method [36]. All other analysis was carried out using R (v3.0.2) [37]. Sensitivity analyses were carried out for all analyses including five studies or more, whereby studies were omitted one by one.

results

study characteristics

Four thousand and four articles were identified, 762 retrieved for evaluation, 580 excluded (Figure 1) and 184 included (supplementary Tables S3–S25, available at Annals of Oncology online), published between 1989 and 2014. Seventy-three assessed markers of clinical outcome, 80 clinical outcome and stage and 29 stage alone. Survival measures were overall survival (OS; n = 133), disease-free survival (DFS; n = 20), recurrence (n = 19), progression-free survival (PFS; n = 4) and disease-specific survival (DSS; n = 4). Twenty nine studies assessed response to therapy (chemo ± radiotherapy, or biological). Eleven assessed treatment complications. Treatment intent was curative (n = 156), palliative (n = 5), mixed (n = 21) and unspecified in 1. Curative modalities were resection alone (n = 111), resection ± neoadjuvant (n = 33) or adjuvant (n = 2) therapy, or definitive chemoradiotherapy (n = 9). All chemotherapy regimens involved platinum agents with or without 5-fluorouracil, except three (bleomycin, gefitinib, irinotecan). One hundred and seventy-six studies were candidate based, and 6 genome-wide (1 SNP, 5 CNV). One hundred and seventeen studies assessed tumor variants, 64 germline and 1 both. Cell types assessed were SCC (n = 117), adenocarcinoma (AC) (n = 40), both (n = 3) and unspecified (n = 22).

Figure 1.

PRISMA flow diagram.

methodological quality

LOE was B for five studies (2.75%), C for 104 (57.1%) and D for 73 (40.1%). Median number of subjects was 90 (range 10–2932), although 48 studies included fewer than 50. Forty-six (25.3%) studies were prospective; 135 (74.7%) were retrospective; 1 (0.55%) had both components. Multivariate adjustment of effect sizes was carried out by 57 studies (31.3%).

molecular quality

Just 37 (56.3%) of 65 studies assessing germline variants assessed Hardy–Weinberg equilibrium; 59 (90.8%) reported genotyping success rate (or provided data allowing its calculation).

markers of survival and recurrence

There were 65 reported markers of survival or recurrence: 24 tumor (Table 1; 3 mutations, 16 CNV, 2 LOH regions, 1 telomere length ratio, CIN, heterogeneous ploidy) and 40 germline polymorphisms (Table 2).

Table 1.

Reported tumor markers (mutations, copy number variants, and chromosomal instability) associated with survival and recurrence following treatment of esophageal cancer

LOE	Variant	Gene / function	Association – minor variant	Association – wild type	No association	Cell type	Population	LOE	Meta-HR [effect variant]	Chi	I2	N	P
Mutations
III	Exon mutant	TP53 Apoptotic / DNA repair regulator	OS and DFS Casson 2003 [38] OSA – Schneider 2000 [39] OS and DFS–Madani 2010 [40] OSA – Yamasaki 2010 [41] OSA – Kunisaki 2006 OS – Kobayashi 1999 [42] OS – Uchino 1996 [29] OS and DFS – Ribeiro 1998 [43] OS and DSS – Kandioler 2014 [44]		OS – Puhringer 2006 [45] OS – Soontrapornchai 1999 [46] OS – Makino 2010 [47] OS – Shimada 1997 [48] OS – Egashira 2011 [49] OS – Ito 2001 [50] OS – Lam 1997 [51] OS – Shibagaki 1995 [52] OS: Goan 2005 [53] OS: Cao 2004 [54] OSA – Gibson 2003 [55] OSA – Coggi 1997 [56]	AC AC AC AC AC SCC SCC SCC SCC SCC SCC SCC SCC SCC SCC SCC SCC US US US US	Res Res Res Res + /−NAC(RT) (CF) Res Mixed Res + NAC Res+ NAC Res + NACRT (CF) Res Res Res DCRT Res Res Res Res Res + NACRT Res Res + NACRT (CF + IFN) Res + NAC (CF)	C C C B D D C D D D C C C D C D C C D D C	OS(A)PB: 1.27 (1.01–1.59) OSA PB: 1.19 (0.63–2.27) AC(A): 1.99 (1.44–2.81) SCC(APB: 1.47 (1.24–1.73) US(A)PB: 0.77 (0.41–1.47) DFS(A): 2.67 (1.38–5.15) Res OS(A)PB: 1.35 (1.04–1.76) NAC/RA(PB): 1.23 (0.81–1.87) SSCP analysis OS(A)PB: 1.56 (1.33–1.82) Direct sequencing only OS(A)PB: 0.96 (0.62–1.46) SCCP band only	68.1 34.5 2.68 22.7 21.3 7.44 27.0 32.5 27.8 28.3	62 80 0 43 72 73 56 66 44 65	21 6 5 11 5 3 13 7 13 8	0.04 0.590 < 0.001 < 0.001 0.440 0.003 0.030 0.320 < 0.001 0.830
IV	Exon 9/20 mutation	PIK3CA Cell signalling kinase		OSA and DFSA – Shigaki 2013 [58]	Rec – Shigaki 2013 OSA and RecEx – Wang 2014 OSEx and DFSEx – Hou 2014	SCC SCC SCC	Res + /−NACRT (CF + /−tax) Res + /−NACRT Res	D D C	OS(A): 0.63 (0.26–1.56) DFS(A): 0.42 (0.21–0.85) RecEx: 0.64 (0.23–1.75)	6.13 0.26 2.81	67 0 64	3 2 2	0.320 0.020 0.390
IV	Exon mutation	NRF2/BIRC2 Transcription factor	OS and Rec – Shibata 2011			SCC	Resection + NACRT (F) (Japan)	D	OSEx: 3.54 (1.60–7.88) Rec—NP	NA	NA	1	0.005 0.046
Copy number variants
II	Gain	EGFR Epidermal growth factor receptor	OS – Marx 2010 [59] OS – Kitagawa 1996 [60] OS – Lennerz 2011 [61]	OSA – Luber 2011 [62]	OS – Miller 2003 [28] OS – Janmaat 2006 [63] OS Rec – Chikuba 1995 [30] OS –Itakura 1994 [31] OS – Sunpaweravong 2005 [64]	AC SCC AC SCC SCC SCC US AC SCC	Res Res Res Palliative gefitinib Res + ACRT (C) Res Mixed PC (OLF + cetuximab) Res	D D C B D D D B D	Gain assessed by FISH/CISH OS: 2.43 (0.75–7.84) Gain assessed by slot/Southern blot OS: 1.63 (0.63–4.22) Excluding Miller 2003 (qPCR)	23.9 18.1	92 89	3 3	0.140 0.320
III	Gain	ERBB2/HER2 Epidermal growth factor receptor	OS – Prins 2013 [65] DFSA – Rauser 2007 [66] OSA – Brien 2000 [67] OS – Zhan 2012 [68] OS – Sato-Kuwabara 2009 [69] OSA – Mimura 2005 [70]		DSSA, OSA – Yoon 2012 [71] OSA – Rauser 2007 [66] OS – Thompson 2011 [72] OS – Miller 2003 [28] OS – König 2013 [73] OS – Sunpawerayong 2005 [64] OS – Suzuki 1997 [74] OS and DSS Ikeda 1996 [75] OS – Lennerz 2011 [61]	AC AC AC AC AC AC Both SCC SCC SCC SCC SCC SCC US	Res Res Res Res Res Res Res Res Res Res Ress Mixed	D C D D D C D D D D C D D D	OS(A): 1.63 (1.20–2.21) OSA: 2.31 (1.64–3.24) OS(A) (AC): 1.59 (0.99–2.56) OS(A) (SCC): 1.92 (1.12–3.29) DFSA: 2.1 (1.06–4.26) Gain assessed via FISH/SISH/IHC; Miller 2003, Ideka 1996, Suzuki 1997 excluded	31.2 0.72 15.4 27.7 NA	62 0 68 82 NA	11 3 6 5 1	0.002 < 0.001 0.060 0.020 0.033
III	Gain	ERBB2/HER2 Epidermal growth factor receptor	DSSA, OSA – Yoon 2012 [71] (heterogeneous amplification)			AC	Res	C	OS: 2.02 (1.09–3.74) DSS: 2.04 (1.09–3.79)	NA NA	NA NA	1 1	0.026 0.025
III	Gain	CCND1 Cell cycle kinase	OSA: Wang 2012b [76] OSA – Miller 2003 [28] OSA – Takeshita 2010 [77] OS – Shimada 1997 [48] OS – Shinozaki 1996 [78] RecA – Komatsu 2014 [79] (ctDNA)		OS–Sunpawerayong 2005 [64] OS – Gramlich 1994 [80]	SCC AC SCC SCC SCC SCC SCC SCC	Res Res Res Res Res Res Res Res	C C C D C D D C	Gain assessed by qPCR only OSA: 2.09 (1.27–3.42) Gain assessed by FISH/IHC OSA: 1.54 (0.93–2.57) Gain assessed by slot blot OS: 4.29 (2.47–7.45)	3.17 3.94 NA	5 75 NA	4 2 1	0.004 0.100 < 0.001
III	Gain	1p36.32	OSA – Carneiro 2008 [81]			SCC	Res	C	OSA: HR 19.6 (2.5–153.9)	NA	NA	1	0.005
III	Gain	19p13.3	OSA – Carneiro 2008 [81]			SCC	Res	C	OSA: HR 7.0 (1.5–31.9)	NA	NA	1	0.011
III	Gain	MDM2 Ubiquitin ligase	OS – Shibagaki 1995 [52]			SCC	Res	C	OS: HREx 3.82 (1.81–8.07)	NA	NA	1	5.3x10−3
IV	Gain	FGF3/INT2 Fibroblast growth factor	OS – Ikeda 1996 [75]		OS – Mori 1992 [82] OS – Shimada 1997 [48] OS – Suzuki 1997 [74]	AC SCC SCC SCC	Res Res Res Res	D D D D	OS: PBHR 1.83 (1.18–2.83) PB – Corrected for Ikeda 1996	5.65	29	4	0.006
IV	Gain	FGF4/HST1 Fibroblast growth factor	OS – Chikuba 1995 [30]		Rec – Chikuba 1995 [30]	SCC SCC	Res + ACRT (C) Res	D D	Median survival different but: OS: HREx 1.4 (0.86–2.30) Rec: HREx 1.09 (0.659–1.80)	NA	NA	1	>0.05
IV	Gain	TERC Telomerase	OS – Wang 2013 [83]			SCC	Res	D	OS: HREx 7.87 (3.32–18.7)	NA	NA	1	0.010
IV	Gain	MET Growth factor	OS – Lennerz 2011 [61]			US	Mixed	D	OS: HREx 3.72 (2.56–5.39)	NA	NA	1	< 0.001
IV	Gain	CPT1A Mitochondrial oxidation	OSA – Shi 2011 [84]			SCC	Res	D	OSA: 4.39 (1.34–14.14)	NA	NA	1	0.015
Telomere length
III		Telomere length ratio (>1.17)	OSA – Gertler 2008 [85]			AC	Res	C	OS: HRA 3.40 (1.3–8.9)	NA	NA	1	< 0.02
LOH
III		LOH at one of 2p, 3p, 17p	OS – Ikeguchi 1999 [86]			SCC	Res	C	2 loci: OS (3yr) 48% versus 75% HREx: 1.81 (0.53–6.25)	NA	NA	1	0.048 > 0.05
IV		LOH 1q21-23	OS – Maru 2009 [87]			AC	Res	D	HREx: 3.90 (1.13–13.5)	NA	NA	1	0.030
Chromosomal instability
III		Aneuploid / polyploidy	OS – Doki 1993 [88] Rec – Tsutsui 1992 [89] Rec – Kaketani 1989 [90] OS – Ohno 1989 [91] OS – Kuwano 1995 [92]		OSA – Wang 1999 [32] OS Edwards 1989 [93]	SCC SCC SCC SCC SCC SCC SCC	Res Res Res Res + NACRT Res Res Res	D D D C C D D	PBOS(A)*: 1.63 (1.25–2.11) Rec: 5.41 (0.87–33.8)	2.99 2.11	25 53	4 2	2x10−4 0.070
IV		Ploidy heterogeneity		(Homogeneity) OS – Deguchi 1993 [94]	DFS- Deguchi 1993 [94]	SCC	Res	D	OS: 0.10 (0.03–0.36) DFS: 0.34 (0.08–1.42)	NA NA	NA	1	< 0.05

OS = overall survival; DFS = disease-free survival; rec = recurrence; A = adjusted; (A) = including adjusted; PB = adjusted for publication bias; Ex = extrapolated; NP = not presented; AC = adenocarcinoma; SCC = squamous cell carcinoma; US = unspecified; NAC = neoadjuvant chemotherapy; NACRT = neoadjuvant chemoradiotherapy; DCRT = definitive chemoradiotherapy; PC = palliative chemotherapy; Res = resection; C = cisplatin; F = 5FU; CFL = cisplatin-5FU-leucovorin; OFL = cisplatin-5FU-leucovorin;; tax = taxane; SSCP = single strand conformation polymorphism; ctDNA = circulating tumor DNA; NA = not applicable

Table 2.

Reported germline markers (polymorphisms) associated with survival and recurrence following treatment of esophageal cancer

LOE	SNP (major/minor allele)	Gene / function	Association– minor variant	Association– wild type	Non-significant	Cell type	Population		Meta-HR [variant allele / genotype] versus wild-type' [Ethnicity]	Chi	I2	N	P
II	rs3212986 (C/A)	ERCC1 DNA NER		OSA–Bradbury 2009b [95] DFSA–Wang 2011 [96] OSA–Rumiato 2013 [97]	OS–Warnecke 2009 [33]	US SCC US US	Res + NAC (CF) PC (CF) Res +/− NAC (CF) Res + NACRT (CF)	B C C C	OS(A/Ex): 0.63 (0.42–0.93) [TT/CT, cisplatin, Caucasian] DFS(A): 1.98 (1.19–3.03) [AA/CA + cis,,Chinese]	4.12 NA	51 NA	3 1	0.02 0.001
II	rs1799793 (G/A)	ERCC2 DNA NER		OSA–Bradbury 2009b [95]	OS and Rec – Ott 2011 [98] OS–Rumiato 2013 [97]	US AC US	Res + NAC (CF) Res + NAC (C/OF +/−tax) Res +/−NAC (CF)	B B C	OS(A/Ex): 0.71 (0.54–0.94) [GA/AA; cisplatin; Caucasian] DFS: 0.33 (0.20–0.07) [AA]	3.88 NA	48 NA	3 1	0.020 0.002
II	rs13181 (T/G)	ERCC2 DNA NER		OSA and DFSA–Bradbury 2009b [95]	OS and Rec - Ott 2011 [98] OS–Rumiato 2013 [97] OSA, RecA–Wu 2006	US AC US US	Res + NAC (CF) Res + NAC (C/OF +/−tax) Res +/−NAC (CF) Res + NACRT (CF +/−tax)	B B C D	OS(A/Ex): 0.82 (0.65–1.05) [TG/GG, Caucasian] DFSA: 0.32 (0.20–0.60) RecA: 0.94 (0.30–2.81)	5.14 NA 0.03	42 NA 0	4 1 2	0.110 0.002 0.91
III	rs11614913	MIR196A2 Micro RNA		OSA–Wu 2014 [99]	OSA–Yang 2014b [100]	SCC SCC	PC Mixed	C C	OSA: 1.26 (0.72–2.21) [TT; Chinese/Taiwanese] PFSA: 1.01 (0.54–1.88)	2.17 NA	54 NA	2 1	0.420 0.972
III	CA-SSR-1 (DNR)	EGFR Epidermal growth factor receptor	OS and Rec–Vashist 2014 [101]			Both	Res	C	OS (AC): 1.70 (1.20–2.80) [LL; Caucasian] Rec (AC): 2.70 (1.70–4.30) OS (SCC): 3.50 (2.10–6.00) Rec (SCC): 2.50 (1.30–4.80)	NA	NA	1	0.010 < 0.001 < 0.001 0.005
III	rs1800796 (C/G)	IL6 Interleukin	OS–Motoyama 2012b [102]		DSS–Motoyama 2012b [102]	SCC	Res +/− AC	C	OS: HRA 3.40 (CI NP) [GG/GC; Japanese] OS: HRAEx 3.49 (1.28–9.45)	NA	NA	1	5.9x10−3 < 0.05
III	rs238406 (G/T)	ERCC2 NER repair		OSA and DFSA Lee 2011 [103]		SCC	Res + NACRT (CF/C + tax)	C	OSA: 0.61 (0.40–0.93) [CC; Taiwanese] DFSA: 0.57 (0.38–0.85)	NA	NA	1	0.02 0.007
III	rs1800975 (G/A)	XPA DNA NER repair	OSA–Yang 2013 [104]		DFS–Yang 2013 [104]	SCC	Mixed	C	OSA: 1.36 (1.06–1.76) [AG; [Taiwanese] DFS: 1.20 (0.95–1.51) [AG]	NA	NA	1	0.014 0.126
III	rs34743033 (STR 2/3/4)	TYMS DNA repair/ replication	OSA–Kaneko 2011 [105]		OS–Okuno 2007 [106] OS–Sarbia 2006 [107] OS–Rumiato 2013	SCC SCC SCC US	DCRT (CF) Res + NACRT (CF) Res + NACRT (CF + E) Res + NAC (CF)	C C C C	OSA: 1.54 (1.00–2.38) [≥2/3; Caucasian/Japanese] OS(A): 2.47(1.20, 5.06) [Japanese] OSA: 1.18 (0.69, 2.03) [Caucasian]	5.78 0.45 0.18	65 0 0	3 2 2	0.61 0.010 0.550
III	rs2279744 (T/G)	MDM2 Ubiquitin ligase	OSA–Renouf 2013 [108]	DFSA Boonstra 2011 [109] OSA and DFSA Cescon 2009 [110]		Both Both AC	Res Mixed	D C	(SCC) DFSA: 2.78 (0.24–31.9) [GG; Caucasian] (AC) DFSA: 0.92 (0.65–1.29) [GG] (AC) OSA: 2.01 (1.38–2.95) (SCC) OSA: 7.89 (2.40–26.0)	9.36 0.04 0.48 NA	89 0 0 NA	2 2 2 1	0.410 0.620 < 0.001 < 0.001
III	rs2273535 (A/T)	AURKA Cell cycle kinase	OS and DFSA–Pan 2012 [111]		DFSA–Boonstra 2011 [109]	US Both	Res + NACRT (CF + tax) Res	C D	OSEx: 0.30 (0.10–0.92) [TT; Caucasian] DFSA: 0.55 (0.17–1.74)	NA 2.92	NA 66	1 2	< 0.05 0.310
III	rs2010963 (G/C)	VEGFA Epithelial mitogen		OS–Tamura 2012 [112] OS - Yang 2014 [113]		SCC SCC	DCRT (CF) Res +/−NACRT (CF)	C	OS HR(Ex): 0.68 (0.50–0.92) [CC]	0.16	0	2	0.01
III	rs3025039 (C/T)	VEGFA Epithelial mitogen	DFSA–Lorenzen 2011 [114]	OSA–Bradbury 2009 [115]	OS–Tamura 2012 [112]	AC AC SCC	Res + NAC (CF) Mixed DCRT (CF)	C C C	OS(A): 0.75 (0.54–1.03) [CT; Caucasian/Japanese] DFSA: 1.8 (1.04–3.09) [CT/TT; Caucasian]	0.79 NA	0 NA	2 1	0.080 0.040
III	rs1042522 (C/G)	TP53 Apoptotic / DNA repair regulator	OSA and PFSA–Renouf 2013 [108] OSA, DFSA–Cescon 2009 [110]		OSA, RecA–Wu 2006	AC AC US	Mixed Mixed Res + NACRT (CF +/−tax)	C C D	OSA: 1,84 (1.34–2.53) [GG; Caucasian] DFSA: 2.03 (1.29–3.18) [GG] RecA: 1.29 (0.24–7.14) [GG]	0.44 NA NA	0 NA NA	3 1 1	< 0.001 0.002 > 0.05
III	rs2069762 (A/C)	IL2 Cytokine	DSSA–Motoyama 2011 [116]			SCC	Res	C	DSS: 3.54 1(1.69–7.39) [C; Japanese] DSSA: 3.36 (NA)	NA	NA	1	0.0231 0.0136
III	rs1800471 (C/G)	TGFB1 Growth factor regulator	OS–Tang 2013 [117]			SCC	Mixed	C	OS: 3.51 (2.18–5.67) [CG/GG; Chinese]	NA	NA	1	< 0.001
III	rs1050631 (G/A)	SLC39A6 Zinc transporter	OSA–Wu 2013 [118]			SCC	Mixed	C	OSA: 1.3 (1.19–1.43) [AA; Chinese]	NA	NA	1	3.77x10−8
III	rs41458645 (C/T)	Mitochondrial D loop	OSA–Zhang 2010 [119]			SCC	Mixed	C	OSA: 3.00 (1.03–8.76) [CT; Chinese]	NA	NA	1	0.044
III	rs139001869 (A/G)	Mitochondrial D loop	OSA–Zhang 2010 [119]			SCC	Mixed	C	OSA: 3.48 (1.07–11.36) [AG; Chinese]	NA	NA	1	0.039
III	rs3769818 (G/A)	CASP8 Caspase	OSA –Umar 2011 [120]			SCC	Mixed	C	OSA: 3.36 (1.07–10.61) [AA; Indian]	NA	NA	1	0.039
III	rs1695 (A/G)	GSTP1 Detoxification enxyme	OSA–Lee 2005 [121]		OS–Okuno 2007 [106] OS–Warnecke 2009 [33] OSA and RecA – Wu 2006 [122] OSA – Rumiato 2013	SCC SCC US US US	Res + NAC (C + F/tax) Res + NACRT (CF) Res + NACRT (CF) Res + NACRT (CF +/−tax) Res + NAC (CF)	C C C D C	OSA: 1.29 (1.03–1.61) [TG/GG; Caucasian/Taiwanese/Japanese] OSA 1.15 (0.78–1.70) [TG/GG; Caucasian] RecA: 0.50 (0.16–1.58) [Caucasian]	2.36 1.84 NA	0 0 NA	5 3 1	0.030 0.490 > 0.05
III	rs72214039 (CA ins)	EGFR Epidermal growth factor receptor	OSA–Lee 2011b [123]			SCC	Res NACRT (CF/C + tax)	C	OSA: 1.88 (1.02–3.49) [Short/Short; Taiwanese]	NA	NA	1	0.045
III	rs7121 (T/C)	GNAS G protein subunit		OSA – Alakus 2014 [124] OSA and DFSA–Vashist 2011 [125]	OS – Alakus 2009 [126]	US US US	Res Res Res + NACRT (CF)	D C B	OSA: 0.73 (0.46–1.16) [CC; Caucasian] DFSA: 0.55 (0.34–0.89	9.08 NA	67 NA	3 1	0.180 2.50x10−3
III	rs111509018 (STR)	ECRG2/SPINK7 Serpin-inhibitor	OSA and DFSA Kaifi 2007 [127]			US	Res	C	OSA: 2.56 (1.53–4.29) [TCA4/TCA4; Caucasian] DFSA: 2.30 (1.37–3.87)	NA	NA	1 1	< 0.001 < 0.001
III	rs9344 (G/A)	CCND1 Cell cycle kinase	OSA–Izzo 2007 [57]			AC	Res	D	OSA: 3.48 (1.94–6.23) [AA]	NA	NA	1	< 0.001
IV	rs1801133 (G/A)	MTHFR Folate metabolism		RecA–Wu 2006 [122]	OSA–Wu 2006 [122] OSA and RecA Ott 2011 [98] OS – Lu 2011 [128] OS–Umar 2010 [129] OS–Sarbia 2006 [107] OS–Warnecke 2009 [33]	US AC SCC SCC SCC SCC	Res + NACRT (CF +/−tax) Res + NAC +(C/OF +/−tax) Res only Mixed Res + NACRT (CF, E) Res + NACRT (CF)	D B C C C C	OSA: 0.93 (0.67–1.29)[AA; Caucasian/Chinese/Indian] OSA: 0.92 (0.63–1.33) [Caucasian] RecA: 0.40 (0.21–0.78) [Caucasian]	3.06 0.69 1.54	0 0 35	6 4 2	0.660 0.640 0.007
IV	rs1801131 (A/C)	MTHFR Folate metabolism		OSA–Wu 2006 [122]	RecA–Wu 2006 [122] OS and rec–Ott 2011 [98] OS–Warnecke 2009 [33]	US AC SCC	Res + NACRT (CF +/−tax) Res + NAC (C/OF +/−tax) Res + NACRT (CF)	D B C	OSA: 0.99 (0.64–1.55) [AA; Caucasian] RecA: 0.80 (0.22–2.95)	0.67 2.38	0 54	3 2	0.970 0.740
IV	rs1045642 (C/T)	ABCB1 Drug efflux		OSA and RecA – Wu 2006 [122]	OS – Narumiya 2011 [130] OS – Okuno 2007 [106] OS – Warnecke 2009 [33]	US SCC US US	Res + NACRT (CF) Res + NACRT (CF) Res + NACRT (CF) Res + NACRT (CF +/−tax)	D C C D	OSA: 0.57 (0.37–0.87) [TT; Caucasian/Japanese] OSA: 0.51 (0.32–0.81) [Caucasian] Rec: 0.26 (0.09–0.81)	1.87 0.68 NA	0 0 NA	4 3 1	0.009 0.004 < 0.05
IV	rs11267092 (DEL/INS)	F2R Angiogensis	OSA and RecA–Lurje 2011 [131]			AC	Res	D	OSEx: 1.70 (1.16–2.48) [INS/INS / INS/DEL; Caucasian] Rec: 2.41 (1.25–4.65) [INS/INS]	NA	NA	1 1	< 0.001 0.003

OS = overall survival; DFS = disease-free survival; rec = recurrence; A = adjusted; (A) = including adjusted; PB = adjusted for publication bias; Ex = extrapolated; NP = not presented; AC = adenocarcinoma; SCC = squamous cell carcinoma; US = unspecified; Res = resection; NAC = neoadjuvant chemotherapy; NACRT = neoadjuvantchemoradiotherapy; DCRT = definitive chemoradiotherapy; C/Cis = cisplatin; O = oxaliplatin; tax = taxane; F = 5FU; E = etoposide; Res = resection; Ex = Extraplotated; INS = insertion; DEL = deletion; NA = not applicable

tumor mutations

Three mutant genes were reported to be associated with outcome: TP53, PIK3CA and NRF2 (LOE IV). TP53 (n = 21) and PIK3CA (n = 3, SCC) underwent meta-analysis.

TP53 status was variably defined and genotyped, although all studies assessed exons 5–8 as a minimum, by single-strand conformation polymorphism (SSCP) analysis with or without sequencing (n = 13), or Sanger sequencing alone (n = 8). Following correction for likely publication bias, a significant negative survival association was demonstrated for mutant TP53 tumors: HR 1.27 (1.01–1.59; P = 0.04; n = 21 studies; supplementary Figures S1 and S2, available at Annals of Oncology online). Significant associations were demonstrated on subgroup meta-analysis of: genotyping technique (SCCP), AC and SCC cell types, and treatment (resection only). Directions of effect were consistent but nonsignificant for adjusted HR alone (n = 6) and use of neoadjuvant chemoradiotherapy (n = 7).

An association with DFS was demonstrated for mutant PIK3CA SCC tumors [HR 0.42 (0.21–0.85); n = 2; P = 0.02], but not OS.

tumor copy number variants

Sixteen tumor CNVs had previously reported associations with prognosis; three were excluded due multiple comparisons (supplementary Table S1, available at Annals of Oncology online). For the remaining 13, LOE was II (1), III (7) and IV (5). Four markers underwent meta-analysis: associations with worse survival were demonstrated for gains in ERBB2 (HER2; LOE III), CCND1 (LOE III) and FGF3 (LOE IV), but not EGFR (LOE II). For all four, there was heterogeneity regarding definition of CNV (absolute copy numbers, or ratio to normal), and genotyping technique [fluorescent/silver in situ hybridization (F/SISH), quantitative (q)PCR and slot/southern blot].

ERBB2/HER2 analysis was restricted to 11 studies performing ISH; 3 (using qPCR [28], or slot [75]/southern [74] blot) were excluded. Worse OS was demonstrated for ERBB2/HER2 gain overall: HR 1.63 (1.20–2.21; P = 2 × 10⁻⁴, n = 11). Significance persisted for adjusted HR [2.32 (1.64–2.58); P < 1 × 10⁻⁵; n = 3], and cell type (SCC; AC P = 0.06). Treatment regimens were resection alone for all studies, except one [61] including mixed regimens involving the c-MET-GFR inhibitor Crizotinib.

Two meta-analyses were carried out for CCND1: studies using qPCR (n = 4), and slot blot/FISH (n = 2). Worse OS was demonstrated for qPCR [HR = 2.09 (1.27–3.42); P = 0.004], with a concordant nonsignificant trend for FISH/blot. All four studies assessing FGF3 used slot/southern blot; an association with worse OS was demonstrated [HR 1.83 (1.18–2.83); P = 0.006]. EGFR meta-analysis was carried out for FISH and blot techniques (excluding two studies using anti-EGFR therapy [62, 63], and one performing qPCR) [28]. No associations were demonstrated.

loss of heterozygosity

LOH (six markers in total) was associated with outcome by four studies; one study (6p and 13q) was excluded due to multiple comparisons; [132] for another (2p, 3p and 12p), while 3-year survival rates were significant the extrapolated HR was not [86]. 1q22–23 LOH was associated with worse OS in one study (LOE IV).

telomere length ratio

One study reported worse OS with a tumor : normal telomere length >1.17 (LOE III).

genomic instability

CIN was assessed by six studies. Following exclusion of one study including intratumoral heterogeneous ploidy [93] an association with worse OS was demonstrated [HR 1.63 (1.25–2.11); n = 4; P = 0.2 × 10⁻⁴]. One study assessed intratumoral heterogeneity alone, reporting better survival than with homogeneity [94]. There were no associations between MSI and survival.

germline polymorphisms

Twenty-nine reported associations were identified (following exclusion of 12 due to multiple comparisons) [107, 109, 112, 133]. Cumulative LOE was II (n = 3), III (n = 22) and IV (n = 4). Fifteen variants underwent meta-analysis (Table 2). Significant associations were demonstrated for six SNPs: ERCC1 rs3212986 (cisplatin treatment; LOE II; Caucasian ethnicity), ERCC2 rs1799793 (cisplatin; Caucasian) TP53 rs1042522 (Caucasian), MDM2 rs2279744 (Caucasian), TYMS rs34743033 (Japanese; LOE III) ABCB1 rs1045642 (Caucasian and Japanese; LOE IV). An association was demonstrated for VEGFA rs2010963, but combining two studies with East Asian ethnicities (Taiwanese and Japanese). One association with recurrence was demonstrated: MTHFR rs1801133 (Caucasian; LOE III).

markers of treatment complications

Twelve reported germline associations (8 studies) were identified; 10 were excluded due to multiple comparisons. One marker (TNFA rs1800629) [134] underwent meta-analysis (nonsignificant). The remaining variant, ACE rs4646994 (LOE III), was associated with postoperative pulmonary complications by one study (Table 3).

Table 3.

Reported tumor variants and germline polymorphisms associated with treatment complications and response to chemo(radio)therapy

LOE	Variant	Gene	Association – mutant	Association – wild type	No association	Cell type	Population	LOE	Meta OR [effect allele / genotype / haplotype]	Chi	I2	N	P
Treatment complications
III	rs4646994 (INS/DEL)	ACE Vasodilator	Post-op pulmonaryA Lee 2005b [135]			US	Res	C	ORA: 3.12 (1.01–9.65) [DEL/DEL; Taiwanese]	NA	NA	1	0.049
III	rs1800629 (G/A)	TNFA Cytokine	Post-op infectionA – Azim 2007 [134]		Motoyama 2009 [136]	US SCC	Res +/− NAC(RT) Res	C D	ORA: 4.02 (0.00–18347) [GG; Caucasian/Japanese]	0	0	2	0.750
Response to chemo(radio)therapy
Tumor
III	High DNA ploidy	-	mPR – HCR versus CR – Ohno 1989 [91]			SCC	Res + NACRT versus HNACRT (bleomycin)	C	mPR 13.18 (5.30–32.7) [High ploidy]	NA	NA	1	< 0.001
IV	Mutant (exon)	TP53 Apoptosis / DNA repair regulator		mPR – Ribeiro 1998 [43] mCR –Yamasaki 2010 [41] cPR – Makino 2010 [47] mCR – Kunisaki 2006	cCR – Ito 2001 [50] cCR – Gibson 2003 [55]	US SCC SCC SCC SCC US	Res, NACRT (CF + IFN) Res + NAC Res + NACRT / DCRT (CF) DCRT Res + NACRT (CF) Res + NACRT (CF)	D D D C C C	mPR 0.24 (0.06–0.95) [mutant] mCR 0.43 (0.08–2.24)	2.87 14.2	30 86	3 3	0.040 0.320
Germline
II	rs7121 (T/C)	GNAS1 G protein subunit	mPRA – Alakus 2009 [126]			US	Res + NACRT (CF)	B	mPRA 7.25 (1.30–40.62) [CC; Caucasian]	NA	NA	1	< 0.05
III	rs3212986 (C/A)	ERCC1 DNA NER	mCRA–Wang 2011 [96]		mPR–Warnecke 2009 [33] mPR–Rumiato 2013 [97]	SCC US US	PC (CF) Res + NAC (CF) Res + NAC (CF)	C B C	mCR ORA: 2.62 (1.11–6.23) [AA/CA; Chinese] mPR OR: 1.52 (0.73–3.20) [CT/CC; Caucasian]	NA 0.76	NA 0	1 2	< 0.05 0.260
III	rs11615 (A/G)	ERCC1 DNA NER	mPR–Metzger 2012 [137] mPR–Warnecke 2009 [33]		mPR-Rumiato 2013 [97]	AC US US	Res + NACRT (CF) Res + NACRT (CF) Res + NAC (CF)	D C C	mPR OR: 4.57 (3.01–6.94) [TT/CT; Caucasian]	3.48	43	3	< 1x10−5
IV	rs25487 (G/C)	XRCC1 DNA repair	cPR–Wu 2006 [122]		cPR - Ott 2011 [98] mPR–Warnecke 2009 [33]	US AC US	Res + NACRT (CF +/−tax) Res + NAC (C/OF +/−tax)) Res + NACRT(CF)	D B C	m/cPR 1.91 (1.30–2.81) [GG; Caucasian]	3.13	36	3	0.001

OS = overall survival; DFS = disease-free survival; rec = recurrence; A = adjusted; AC = adenocarcinoma; SCC = squamous cell carcinoma; US = unspecified; NAC = neoadjuvant chemotherapy; NACRT = neoadjuvantchemoradiotherapy; DCRT = definitive chemoradiotherapy; HNACRT = hyperthermicneoadjuvant chemoradiotherapy; PC = palliative chemotherapy; CF = cisplatin-5FU; DEL = deletion; OF = cisplatin-5FU; Res = resection mPR = major pathological response; cPR = complete pathological response; mCR = major clinical response; cCR = complete clinical response; OR = odds ratio; IFN = interferon; LOE = level of evidence; NA = not applicable

markers of response to chemo(radio)therapy

Two tumor variants (mutant TP53 and CIN) and 15 germline polymorphisms were reported to be associated with clinical or pathological response to chemo ± radiotherapy (Table 3); 11 polymorphisms were excluded due to multiple comparisons).

Mutant TP53 was assessed by six studies; three for pathological and three for clinical response. A lower OR of pathological response was demonstrated [OR 0.24 (0.06–0.95), n = 3, P = 0.04]; effect direction for clinical response was concordant but nonsignificant [OR 0.43 (0.08–2.24); P = 0.32].

Following meta-analysis, two polymorphisms were associated with a major pathological response to platinum-based chemo/radiotherapy in Caucasians: wild-type XRCC1 rs25487 [GG genotype, LOE III; OR 1.91 (1.30–2.81), n = 3, P = 0.001], and variant ERCC1 rs11615 [TT/CT; OR 4.57 (3.01–6.94); n = 3; P < 1 × 10⁻⁵]. The AA variant of ERCC1 rs3212986 (LOE III) was associated with radiological response to palliative cisplatin-based chemotherapy in one study (Chinese ethnicity), but not major pathological response to neoadjuvant chemotherapy in two (Caucasian).

markers of stage

Twenty-four tumor markers were reported: 2 mutations, 12 CNV, 7 LOH, 2 MSI and CIN; 15 were excluded.

tumor mutations

Following exclusion of PIK3CA, the sole tumor mutation with a reported association was TP53 (n = 19; LOE III; Table 4). Following meta-analysis, mutant TP53 tumors were associated with more advanced T (T3/T4) and N (≥N1) stages, but not overall TNM stage (III/IV), grade (G3/4) or positive resection margin (R1).

Table 5.

Reported germline markers (polymorphisms) associated with stage of esophageal cancer

LOE	Variant	Gene	Association—variant	Association—wild type	No association	Cell type	Population	LOE	Meta OR [effect allele/genotype/haplotype]	Chi	I2	N	P
II	rs6573 (C/A)	RAP1A RAS oncogene		OA—Wang 2012 [149]		SCC	Mixed	B	OA: 1.89 (1.06–3.36) [CA/AA; Chinese]	NA	NA	1	0.030
II	rs1800471 (G/C)	TGFB1 Growth factor regulator	O, G—Tang 2013 [117]			SCC	Mixed	B	O: ORA 2.71 (1.44–5.09) [GC/CC; Chinese] G: ORA 2.65 (1.44–4.87)	NA NA	NA NA	1 1	<0.001 0.002
III	rs353163 (T/C)	TMPRSS11A Serine peptidase	N—Umar 2013b [150]			SCC	RT/DCRT (CF)	C	N: 3.27 (1.68–6.39) [CC; Indian]	NA	NA	1	<0.001
III	rs2273535 (A/T)	AURKA Cell cycle kinase	O—Miao 2004 [151]			SCC	Res	C	O: 2.13 (1.04–4.39 [TT; Chinese]	NA	NA	1	<0.05
III	rs7121 (C/T)	GNAS1 G protein subunit	O, N—Vashist 2011 [125]		T, M, G—Vashist 2011 T, N, R—Alakus 2009 [126]	US US	Res Res + NACRT (CF)	C B	O: 2.10 (1.17–3.76) [T; Caucasian] N: 1.16 (0.76–1.77) [T]	NA 3.94	NA 49	1 3	0.013 0.500

AC, adenocarcinoma; SCC, squamous cell carcinoma; US, unspecified carcinoma; NACRT, neoadjuvant chemoradiotherapy; RT, radiotherapy; DCRT, definitive chemoradiotherapy; T, T stage (III/IV versus I/II); N, nodal stage (≥N1 versus N0); M, metastatic stage (M0 versus M1); G, cell grade (III/IV versus I/II); O, overall stage (III/IV versus I/II); R, resection stage; L, L stage; CF, cisplatin–5FU; Res, resection; LOE, level of evidence; NA, not applicable.

Table 4.

Reported tumor markers (mutations, copy number variants, genomic and chromosomal instability) associated with stage of esophageal cancer

LOE	Variant	Gene	Association – mutant	No association	Cell type	Population	LOE	Meta OR [effect allele / genotype / haplotype]	Chi	I2	N	P
Mutations
III	Mutant	TP53 Apoptosis / DNA repair regulator	T, N – Madani 2010 [40] N – Cao 2004 [54] N – Hattori 2003 [138] O Ribeiro 1998 O, T, N, G-Casson 2003 [38] N - Makino 2010 [47]	R – Madani 2010 [40] T – Cao 2004 [54] O, T, N – Schneider 2000 [39] T, N, G – Soontrapornchai 1999 T, N, G – Egashira 2011 [49] T, N, M, O – Yamasaki 2010 [41] T, M, G – Ito 2001 [50] O – Kobayashi 1999 [42] T, N, M, G – Uchino 1996 [29] O – Coggi 1997 [56] O, T, N – Goan 2005 [53] O, G – Lam 1997 [51] O, T, N – Shibagaki 1995 [52] T, N – Puhringer 2006 [45] T, M, G – Makino 2010 [47]	AC SCC SCC US AC AC SCC SCC SCC SCC SCC US AC SCC SCC SCC AC SCC	Res Res Res Res + NACRT Res Mixed Res NAC (CF) +/− Res Res + NACRT (CF) Res Res Res Res Res Res Res Res +/− NACRT (CF) Res +/− NACRT /DCRT (CF)	C D C D C D C D C D D D C D D C B D	OPB: 1.28 (0.71–2.31) T: 1.40 (1.12–1.74) N: 1.39 (1.07–1.81) M: 1.21 (0.72–2.03) G: 1.46 (0.83–2.58) R: 2.10 (0.470–9.35) O: B p = 0.540; E p = 0.275 T: B p = 0.393; E p = 0.071 N: B p = 0.765; E p = 0.443 M: B p = 0.207; E = 0.492 G: B p = 0.719; p = 0.543 PB Corrected for Ribeiro 1998	32.4 17.5 18.1 2.99 19.1 3.45	63 20 17 0 53 71	12 15 15 5 10 2	0.410 0.003 0.010 0.480 0.190 0.330
Copy number variants
III	Gain	SPK2 Protein kinase	N,O – Wang 2009 [139]		SCC	Res	C	O: 8.00 (2.25–28.5) N: 8.10 (2.28–28.8)	NA	NA	1 1	1.30x10−3 1.20x10−2
III	Gain	PRKC1 Serpin	T – Yang 2008 [140]	N, O, G – NS – Yang 2008 [140]	SCC	Res	C	O: 4.64 (1.71–12.4) N: 3.12 (1.21–8.02) T: 2.63 (0.78–8.81)	NA NA NA	NA NA NA	1 1 1	0.002 0.019 0.118
III	Gain	HER2 (ERBB2) Epidermal growth factor receptor	T, G – Yoon 2012 [71] G,O – Zhan 2012 [68] O, G – Lennerz 2011 [61] N – Ikeda 1996 [75]	T, M, L – Zhan 2012 [68] O, T, N – Suzuki 1997 [74] O, T, N, M, R, G – Reichelt 2007 [141] O, N - Brien 2000 [67] O, N, G - Mimura 2005 [70] O, T, N, G, R – Sato-Kuwabara 2009 [69] O, G – Sunpaweravong 2005 [64] O, T, N, G, M, R – Thompson 2011 [72] O, T, N, G, M – Al-Kasspooles [142] T, N, G – Prins 2013 [65]	AC SCC US SCC SCC US AC SCC SCC SCC AC AC AC	Res Res Mixed Res Res Res Res Res Res Res Res Res Res	C D D C D C D D D D D C D	O: 1.13 (0.83–1.54) T: 0.84 (0.55–1.27) NPB: 0.96 (0.69–1.35) M: 1.77 (0.69–4.56) GPB: 0.61 (0.34–1.09) PB – Corrected for Ikeda 1996, Al-Kasspooles 1993, Suzuki 1997, Minmura 2005 PB – Corrected for Sunpaweravong 2005, Zhan 2012	14.02 17.5 20.2 3.85 40.0	43 48 31 22 72	10 10 11 4 10	0.510 0.400 0.510 0.240 0.100
IV	Gain	EGFR Epidermal growth factor receptor	T,N – Marx 2010 [59] G – Lennerz 2011 [61] N – Kitagawa 1996 [60] N – Yang 2012 [143]	M, G – Marx 2010 [59] O – Lennerz 2011 [61] O, T – Kitagawa [60] O,G – Yang 2012 [143] O, T,N,M – Miller 2003 [28] O, T,N,M,G – Al-Kasspooles 1993 [142] O,T,N,G – Itakura 1994 [31]	AC US SCC SCC AC AC SCC	Res Mixed Res Res Res Res Res	D D D D C C D	OPB: 0.96 (0.6–1.53) T: 0.51 (0.37–0.71) N: 0.95 (0.41–2.16) M: 0.91 (0.29–2.87) G: 1.24 (0.70–2.20) PB – Corrected for Kitagawa 1996	9.27 2.31 8.44 1.38 3.36	35 0 41 0 8	6 5 6 3 3	0.850 < 0.001 0.890 0.870 0.460
Loss of heterozygosity
III	LOH	3p14.2	TA – Qin 2008 [144]	N, M, G – Qin 2008 [144]	SCC	Res	C	T: 5.67 (1.77–18.2)	NA	NA	1	0.003
III	LOH	13q	T, G – Huang 2002 [145] N – Harada 1999 [146] N – Shibagaki 1994 [132]	N – Huang 2002 [145] T,M,H,O – Harada 1999 [146]	SCC SCC SCC	Res Res Res	C C C	T: 3.08 (0.17–60.8) N: 4.17 (1.84–9.47)	5.04 3.06	80 35	2 3	0.440 6x10−4
Chromosomal instability
III	CIN		T,N,G Yu 1989 [147] G – Doki 1993 [88] T,N - Kuwano 1995 [92] N, Ohno 1989 [91]	T,/N – Doki 1993 [88] N,V,G – Kuwano 1995 [92] O,T - Ohno 1989 [91] T,N,G – Tsutsui 1992 [89] T,L,G – Edwards 1989 [93] O, T, N G – Wang 1999 T – Kaketani 1989 [90]	SCC SCC SCC SCC SCC SCC SCC SCC SCC	Res Res Res Res +/− NACRT Res Res Res +/− AR Res Res	C D C C D D D D D	O: 2.68 (1.10–6.54) T: 1.41 (0.97–2.05) N: 2.18 (1.06–4.47) G: 1.51 (0.99–2.31)	2.01 7.21 18.2 5.29	0 0 57 5	2 9 7 6	0.030 0.070 0.030 0.060
Microsatellite instability
IV	MSI	17q24-25 + Bethesda markers	T – Matsumoto 2007 [148]	N,G,O – Matsumoto 2007 [148]	SCC	Res	D	T: 0.325 (0.11–0.96)	NA	NA	1	0.043

AC = adenocarcinoma; SCC = squamous cell carcinoma; US = unspecified carcinoma; NAC = neoadjuvant chemotherapy; NACRT = neoadjuvant chemoradiotherapy; CF = cisplatin and 5-fluoruracil; T = T stage (III/IV versus I/II); N = nodal stage(N0 versus ≥N1); M = metastatic stage (M0 versus M1); G = cell grade (III/IV versus I/II); O = overall stage (III/IV versus I/II); R = resection stage (R1 versus R0); L = L stage (L1 versus L0); V = venous invasion (V1 versus V0); Res = resection; PB = corrected for publication bias; LOE = level of evidence; LOH = Loss of Heterozygosity; CIN = chromosomal instability; NA = not applicable

copy number variants

Twelve tumor CNVs were identified; 8 were excluded due to multiple comparisons (Table 4). Meta-analysis was possible for two markers, with one significant association demonstrated: EGFR (T1/2 stage; LOE IV).

loss of heterozygosity

Seven LOH variants were identified; five were excluded (Table 4). Meta-analysis was possible for one marker. A significant association was demonstrated for LOH 13p and ≥N1 stage (Table 4).

genomic instability

Following meta-analysis CIN (LOE III; Table 4) was associated with overall stage [III/IV: OR 2.68 (1.10–6.54); P = 0.03; n = 2 studies] and nodal stage [OR 2.18 (1.06–4.47); P = 0.03; n = 7 studies], but not T stage or grade. MSI (using the 5 Bethesda markers, and 10 at 17q24–25) was associated with more advanced stage; another measure was excluded.

germline polymorphisms

Seventeen polymorphisms were identified; 12 were excluded due to multiple comparisons. One marker (GNAS1 rs7172) underwent meta-analysis, without significance.

funnel plot asymmetry, heterogeneity and publication bias

Begg's and Egger's tests were nonsignificant for all meta-analyses (supplementary Table S26, available at Annals of Oncology online). Visual inspection of plots identified asymmetry for nine outcome analyses: mutant TP53 (OS overall, adjusted HR, SCC and unspecified cell types, neoadjuvant therapy and SSCP/direct sequencing analyses; supplementary Tables S2 and S3, available at Annals of Oncology online), ERRBB2/HER2 (OS) and FGF3 (OS), and three stage analyses: EGFR (overall) and ERBB2/HER2 (N and grade). These were interpreted as likely publication bias and corrected (without affecting any conclusions). All sensitivity analyses were negative.

conclusions

We identified 182 studies, which assessed a total of 165 candidate genomic markers. Overall, 91 markers were reported to have significant associations with esophageal cancer outcome, and 41 with stage. Overall study quality was poor: most studies were retrospective with small sample sizes, and all except 5 (2.75%) were of level C or D quality. There was considerable heterogeneity in patient selection, treatment approach, genotyping techniques and definitions used. Common areas of weakness were failure to perform subgroup analysis for AC and SCC; failure of quality control such as reporting call rates and Hardy–Weinberg equilibrium; failure to perform/report multivariate adjustment of HRs; and failure to adjust for multiple comparisons. Furthermore, just 30.2% of reported markers subsequently had attempted validation data published.

Despite these limitations, sufficient data were available for appropriate meta-analyses. These demonstrated a small number of associations of DNA sequence markers with worse survival (mutant TP53, HER2, CCND1 and FGF3 copy number gain and CIN) and resistance to chemo–radio ± therapy (TP53).

As far as we are aware, this is the first attempt to collate and evaluate all evidence of DNA sequence markers and esophageal cancer, and to demonstrate the above associations by meta-analysis. As such it has a number of generic and specific strengths and weaknesses. A comprehensive search strategy was used to minimize identification and selection bias (requiring detailed appraisal of studies including gastric cancer, cell lines and expression data), it is possible that studies were not identified. For those included, methodological heterogeneity and small sample sizes introduce potential for bias. Although there was no statistical evidence of funnel plot asymmetry using Begg's and Egger's tests, these are underpowered in meta-analyses of fewer than 25 studies [152]; we therefore inspected all funnel plots, explored the reasons for any apparent asymmetry, and corrected eight analyses for likely publication bias (without altering overall effects). While the small number of studies involved in each analysis precluded meaningful meta-regression to explore additional potential confounding factors [153], we sought to address potential bias by performing subgroup analyses, including cell type genotyping techniques and ethnicity. There are also limitations to the revised American Society of Clinical Oncology guidelines in this context; firstly, regarding capture of the complexity inherent in data quality, and secondly determination of LOE: evidence can be upgraded by validation studies, yet disagreement of effect size and direction between studies is not always reflected in the ultimate LOE.

The strongest evidence we found for an outcome marker was tumor TP53 mutation. Association with worse survival was demonstrated for both AC and SCC. Whether this is truly independent of the association demonstrated with T and N stage (independent pathological markers of outcome) [154] was not conclusively demonstrated, and indeed only assessed by six studies. Although four reported significant adjusted HR, the resultant meta-analyzed direction of effect was concordant but not significant due to the use of a random-effects model. We also found TP53 mutant tumors to be less chemo(radio)sensitive.

As other recent meta-analyses have reported similar findings in breast and colorectal carcinoma [155, 156], this is of particular translatable relevance. TP53 is one of the most frequently mutated and studied genes in human cancer [157], with resultant attempts to develop targeted therapies [158]. Ninety-five percent of functional mutations occur within exons 5–9, which encode the DNA binding domain, and typically cause loss of efficacy either directly by disrupting DNA contact, or indirectly by aberrant protein folding [159]. Subsequently, cell cycle, DNA repair and apoptotic regulation may fail [160], although oncogenic gains of function are occasionally seen [161]. The most characterized variant is the germline rs1042522 G > C substitution, itself conferring a worse HR for both OS and DFS in this meta-analysis.

TP53 as an esophageal tumor biomarker is often considered in terms of TP53 status: aberrant expression, with or without mutation. An association with expression alone and worse outcome has been demonstrated on meta-analysis for SCC [162], as has aberrant status [increased expression (28 studies) with or without mutation (3 studies)] and reduced likelihood of response to chemotherapy [163]. However, TP53 mutational and expression statuses may be discordant [164] particularly in the case of high-impact mutations precluding expression, or dramatically reducing half-life. The ability to predict this from sequencing data reinforces the need to explore the interaction of these aspects of status in parallel [165, 166].

Typically, resection specimens are used to assess associations between tumor markers and pathological response to chemotherapy. However, by definition these comprise clonal populations selected for chemo/radio-resistance. While such tumors appear to be disproportionately TP53 mutated, deep re-sequencing and clonal studies comparing the prevalence and associations of pre- and post-treatment tumor are required to establish the true pretreatment predictive utility of TP53 mutations in this regard.

Three associations between tumor copy number gain (albeit variably quantified) were demonstrated by meta-analysis: ERBB2/HER2, CCND1 and FGF3 gain. ERBB2/HER2 is particularly relevant; a proto-oncogene, it is the sole molecular marker in clinical use for gastroesophageal cancer, guiding the use of targeted therapies [14]. Our findings build on a recent meta-analysis of HER2 status, defining positivity by overexpression or amplification, including six of the studies included in this meta-analysis [15]. We found gain to confer a worse prognosis for both AC and SCC, independent of stage. Interestingly, all patients in 10 of the 11 studies underwent radical treatment with resection; while palliative monoclonal antibody therapy for HER2-positive gastroesophageal AC is effective in prolonging survival [167], an urgent unanswered question is therefore whether it has a role in curative treatment.

Similarly, regarding the cell cycle regulator CCND1, phase I and II data have suggested a possible role for cyclin-dependent kinase inhibitors in nongastroesophgageal cancer [168, 169]. Our findings therefore suggest the need to assess their effect in esophageal tumors with CCND1 gain. EGFR, a tyrosine kinase receptor, has also been extensively investigated within gastroesophageal cancer; phase II data support targeted therapy (antibodies and tyrosine kinase inhibitors) for metastatic disease [170, 171], although not yet neoadjuvant regimens [172, 173]. While we found no association with outcome using the requisite random-effects model, significant effects were evident with a fixed model; consequently, there may be an undetected association. We also found CIN to be associated with worse outcome, in keeping with a previous colorectal cancer meta-analysis [174], although whether it modulates chemo-sensitivity is unclear.

We also demonstrated survival associations for six common germline polymorphisms by meta-analysis: ERCC1 rs3212986 (for cisplatin treatment and Caucasian ethnicity), ERCC2 rs1799793 (cisplatin and Caucasian), TP53 rs1042522 (Caucasian), MDM2 rs2279744 (Caucasian), TYMS rs34743033 (Japanese) ABCB1 rs1045642 (both Caucasian and Japanese). The association of VEGFA rs2010963 was evident only on combining Taiwanese and Japanese study populations. MTHFR rs1801133 was associated with recurrence in Caucasians. XRCC1 rs25487 and ERCC1 rs11615 were associated with response to chemotherapy in Caucasians. These associations are likely to be due to aberrant protein expression or function.

rs3212986 modifies ERCC1 mRNA stability [175], a component of the nucleotide excision repair (NER) pathway, variants of which are associated with platinum sensitivity and survival in pancreatic, gastric, colorectal and lung cancers [95–177]. The missense rs1799793 SNP results in an aspartate–asparagine substitution at codon 312 of the ERCC2 component of the NER pathway, and has been similarly associated with survival in gastric and other cancers [178]. The rs10456402 SNP in exon 26 of ABCB1 (Multi Drug Resistance 1) reduces expression (and consequent platinum-analogue membrane transportation) [179], and is similarly associated with colorectal cancer prognosis [180]. rs2279744 increases mRNA expression of MDM2, which suppresses TP53 activity [181], and is associated with increased susceptibility to a number of cancers (including gastric) [182]. rs34743033 is a 28-bp variable number tandem repeat in TYMS (thymidylate synthase), with enhancer function correlating with increased TYMS expression [183], and survival in platinum-treatment nonsmall-cell lung carcinoma [184]. The rs1801133 missense SNP induces an alanine–valine substitution at codon 222, with reduced activity of methylenetetrahydrofolate reductase [185], and increased susceptibility to gastric cancer [186]. rs25487 induces a glutamine-arginine substation in codon 399, with resultant reduction in function of the DNA repair gene XRCC1 [187], and an association with survival of lung cancer [188]. rs11615 reduces ERCC1 expression [189], and increases likelihood of response to platinum chemotherapy in gastric and colorectal cancer [176].

Biomarkers themselves carry a number of limitations. Typically, they are classified as ‘prognostic’ or ‘predictive’; however, in reality, these are not mutually exclusive, and we therefore did not attempt classification. Biomarker development culminates in demonstration of clinical, requiring at least multicenter prospective validation for prognosis, and incorporation into interaction randomized controlled trials for prediction. These challenges reinforce the utility of retrospectively analyzing samples archived during prospective trials. Other challenges include the use of pretreatment biopsies. First, analysis may be impaired by inclusion of noncancerous tissue; while this can be mitigated by techniques such as laser-capture microdissection and higher depth sequencing, these are time and cost-intensive. More profound is the challenge of intratumoral heterogeneity and clonality: a single biopsy is representative of just 34% of the mutational burden of a ‘single’ cancer [190], and will not include metastatic subclones. How to surmount this is not yet clear.

Finally, while it may be pragmatic to consider DNA sequence variants in isolation, their effects (and therefore utility) are subject to complicated modulation by the other ‘omics', (epigenomics, transcriptomics, metabolomics and proteomics), genes and clinical and environmental covariates [191, 192]. While a discrete variable might provide useful complementary information of itself, this complexity at present precludes its use to dichotomize decision making. Consequently, a robust approach to personalized cancer medicine must incorporate parallel processing of DNA, RNA, proteins and metabolites.

In conclusion, numerous DNA sequence markers have been described for esophageal cancer. However, as with complementary fields within personalized cancer research, the underlying research is largely poor in quality and disparate in methodology, with a lack of robust validation of markers and incorporation into trials. While a number of promising candidates have been identified the data required to incorporate these into prognostic/predictive models do not yet exist; future validation will require larger studies, with improvements in the standardized collection of samples for analysis, parallel assessment of expression and the incorporation of parallel biomarkers within high-quality clinical trials, robust adjustment for confounding variables and sharing of resultant data with multicenter collaboration.

funding

JF is supported by the NIHR Oxford Biomedical Research Centre. Core funding is provided to the Wellcome Trust Centre for Human Genetics from the Wellcome Trust (090532/Z/09/Z).

disclosure

MRM has received payment for advisory/consulting roles within the last 2 years from Amgen, Bristol-Meyers Sqibb, GlaxoSmithKline, Merck, Millennium, and Roche, and institutional funding from Amgen, AstraZeneca, Bristol-Meyers Squibb, Clovis, Eisai, GlaxoSmithKline, Immunocore, Johnson & Johnson, Merck, Millennium, Novartis, Pfizer, Roche and Vertex. The authors have declared no conflicts of interest.