Doi:10.1016/s0140-6736(05)17947-

Articles
Gene-expression profiles to predict distant metastasis of
lymph-node-negative primary breast cancer
Yixin Wang, Jan G M Klijn, Yi Zhang, Anieta M Sieuwerts, Maxime P Look, Fei Yang, Dmitri Talantov, Mieke Timmermans, Lancet 2005; 365: 671–79
Marion E Meijer-van Gelder, Jack Yu, Tim Jatkoe, Els M J J Berns, David Atkins, John A Foekens See Comment
Veridex LLC, a Johnson &
Johnson Company, San Diego,
CA, USA (Y Wang PhD, Y Zhang
Background Genome-wide measures of gene expression can identify patterns of gene activity that subclassify PhD, F Yang MSc, D Talantov MD,
tumours and might provide a better means than is currently available for individual risk assessment in patients
J Yu PhD, T Jatkoe BSc); Veridex
with lymph-node-negative breast cancer.
LLC, a Johnson & Johnson
Company, Warren, NY, USA
(D Atkins PhD); and Department
Methods We analysed, with Affymetrix Human U133a GeneChips, the expression of 22 000 transcripts from total of Medical Oncology,
RNA of frozen tumour samples from 286 lymph-node-negative patients who had not received adjuvant systemic Erasmus MC–Daniel den Hoed,
treatment.
Rotterdam, Netherlands
(Prof J G M Klijn MD,
A M Sieuwerts BSc,
Findings In a training set of 115 tumours, we identified a 76-gene signature consisting of 60 genes for patients
positive for oestrogen receptors (ER) and 16 genes for ER-negative patients. This signature showed 93% sensitivity M Timmermans BSc,
and 48% specificity in a subsequent independent testing set of 171 lymph-node-negative patients. The gene profile
M E Meijer-van Gelder MD, E M J J Berns PhD, was highly informative in identifying patients who developed distant metastases within 5 years (hazard ratio 5·67
[95% CI 2·59–12·4]), even when corrected for traditional prognostic factors in multivariate analysis (5·55
[2·46–12·5]). The 76-gene profile also represented a strong prognostic factor for the development of metastasis in Dr John Foekens, Erasmus MC,
the subgroups of 84 premenopausal patients (9·60 [2·28–40·5]), 87 postmenopausal patients (4·04 [1·57–10·4]),
and 79 patients with tumours of 10–20 mm (14·1 [3·34–59·2]), a group of patients for whom prediction of
BE-426, Dr Molewaterplein 50,3015 GE Rotterdam, Netherlands prognosis is especially difficult.
[email protected]
Interpretation The identified signature provides a powerful tool for identification of patients at high risk of distant
recurrence. The ability to identify patients who have a favourable prognosis could, after independent
confirmation, allow clinicians to avoid adjuvant systemic therapy or to choose less aggressive therapeutic options.
Introduction
1980–95, but who did not receive systemic neoadjuvant About 60–70% of patients with lymph-node-negative or adjuvant therapy. Tumour samples were submitted breast cancer are cured by local or regional treatment to our reference laboratory from 25 regional hospitals alone.1,2 The most widely used treatment guidelines are for measurements of steroid-hormone receptors.
the St Gallen3 and the US National Institutes of Health4 Guidelines for primary treatment were similar for all consensus criteria. These guidelines recommend hospitals. Selection of tumours aimed to avoid bias. On adjuvant systemic therapy for 85–90% of lymph-node- the assumption of a relapse rate of 25–30% in 5 years, negative patients. There is a need for specific definition and a substantial loss of tumours for quality-control of an individual patient’s risk of disease recurrence to reasons, 436 samples of invasive tumours were ensure that she receives appropriate therapy. Currently, processed. Patients with poor, intermediate, and good few diagnostic tools are available to identify at-risk clinical outcome were included. Samples were rejected patients. To date, gene-expression patterns have been on the basis of insufficient tumour content (53), poor used to classify breast tumours into clinically relevant RNA quality (77), or poor chip quality (20); thus, subtypes.5–21 We report a comprehensive genome-wide 286 samples were eligible for further analysis. The assessment of gene expression to identify broadly study was approved by institutional medical ethics applicable prognostic markers.5,6 In this study, we committee (number 02·953). The median age of the aimed to develop a gene-expression-based algorithm patients at surgery was 52 years (range 26–83). 219 had and to use it to provide quantitative predictions on undergone breast-conserving surgery and 67 modified disease outcome for patients with lymph-node-negative radical mastectomy. Radiotherapy was given to 248 patients (87%) according to our institutionalprotocol. The proportions of patients who underwent breast-conserving therapy and radiotherapy are normal Patients’ samples
for lymph-node-negative disease. Patients were We selected from our tumour bank at the Erasmus included irrespective of radiotherapy status because Medical Center (Rotterdam, Netherlands) frozen this study did not aim to investigate the effects of a tumour samples from patients with lymph-node- specific type of surgery or adjuvant radiotherapy.
negative breast cancer who were treated during Furthermore, other studies have shown that www.thelancet.com Vol 365 February 19, 2005
Articles
mg protein or 10% positive tumour cells. Postoperative Study profile
follow-up involved examinations every 3 months for2 years, every 6 months for years 3–5, and every 12 months from year 5. The date of diagnosis ofmetastasis was defined as that at confirmation ofmetastasis after symptoms reported by the patient, detection of clinical signs, or at regular follow-up.
Gene-expression analysis
Total RNA was isolated from 20–40 cryostat sections of
30 µm thickness (50–100 mg) with RNAzol B (CamproScientific, Veenendaal, Netherlands). Biotinylated targets were prepared by published methods (Affymetrix, Santa Clara, CA, USA)25 and hybridised to the Affymetrix oligonucleotide microarray U133aGeneChip. Arrays were scanned by standard Unsupervised clustering analysis
Affymetrix protocols. Each probe set was treated as aseparate gene. Expression values were calculated by useof Affymetrix GeneChip analysis software MAS 5.0.
Chips with average intensity of less than 40 orbackground signal of more than 100 were rejected. Forchip normalisation, probe sets were scaled to a targetintensity of 600, and scale mask files were not selected.
Statistical methods
17 819 genes were “present” in two or more samples and
were eligible for hierarchical clustering. Before
clustering, the expression level of each gene was divided
by its median expression level in the patients. This
ER-negative
ER-positive
standardisation step limited the effect of the magnitude ER-negative
ER-positive
of expression of genes, and grouped together genes withsimilar patterns of expression in the clustering analysis.
Figure 1: Profile for selection of samples for analysis and unsupervised
To identify subgroups of patients, we carried out average clustering analysis of gene-expression data for 286 patients with lymph-
linkage hierarchical clustering on both the genes and the node-negative breast cancer
samples using GeneSpring 6.0. To identify genes that ER status was used to identify subgroups. Each subgroup was then analysed separately for selection of markers. The patients in a subgroup were assigned toa training set or a testing set. The markers selected from each subgroup were metastases from those remaining metastasis-free for combined to form a single signature to predict tumour recurrence for all patients 5 years, we used two supervised class prediction in the testing set as a whole. The left panel of the clustering analysis is a view of approaches. In the first approach, 286 patients were the 17 819 informative genes. Red indicates high relative expression, green randomly assigned to training and testing sets of 80 and relative low expression. Each column is a sample and each row is a gene. Theright panel shows enlarged views of two dominant gene clusters that had drastic 206 patients, respectively. Kaplan-Meier survival curves26 differential expression between the two subgroups of patients. The upper gene for the two sets were examined to ensure that there was cluster has a group of 282 downregulated genes in the ER-positive subgroup, no significant difference and that no bias was introduced and the lower gene cluster is represented by a group of 339 upregulated genesin the ER-positive subgroup. The label bar at the foot of each dendrogramindicates the patient’s ER status measured by routine assays.
Panel: Calculation of relapse scores
radiotherapy has no clear effect on distant disease recurrence.22 Lymph-node negativity was based on ϩ⌺I · wx + B · (1–I)ϩ⌺(1–I) · wx pathological examination by regional pathologists.23 All 286 tumour samples were confirmed to have sufficient I=1 if ER is more than 10 fmol per mg protein; I=0 if ER is 10 (Ͼ70%) tumour and uniform involvement of tumour fmol per mg protein or less; w is the standardised Cox’s in 5 µm frozen sections stained with haematoxylin and regression coefficient for an ER-positive marker; x is the eosin. Amounts of oestrogen receptors (ER) and expression value of the ER-positive marker on a log scale; w progesterone receptors (PR) were measured by ligand- is the standardised Cox’s regression coefficient for an ER- binding assay, EIA,24 or immunohistochemistry (nine negative marker; x is the expression value of the ER-negative tumours). The cut-off value for classification of patients marker on a log scale; A and B are constants.
as positive or negative for ER and PR was 10 fmol per www.thelancet.com Vol 365 February 19, 2005
Articles
by the random selection of the training and testing sets.
highest specificity. Values of constants A of 313·5 and In the second approach, patients were allocated to one of B of 280 were chosen to centre the threshold of relapse two subgroups stratified by ER status (figure 1). Each score to zero for both ER-positive and ER-negative subgroup was analysed separately for selection of patients. Patients with positive or negative relapse markers. Patients in the ER-positive subgroup were scores were classified as those with poor or good randomly allocated into training and testing sets of prognosis, respectively. The gene signature and the 80 and 129 patients, respectively. The ER-negative cut-off were validated in the testing set. Kaplan-Meier subgroup was randomly divided into training and survival plots and log-rank tests were used to assess the testing sets of 35 and 42 patients, respectively. Markers differences in time to distant metastasis of the selected from each subgroup training set were combined predicted high-risk and low-risk groups. Odds ratios to form a single signature to predict tumour metastasis were calculated as the ratio of the odds of for both ER-positive and ER-negative patients in a distant metastasis between the patients predicted to experience relapse and those predicted to remain The sample size of the training set was determined by a resampling method to ensure its statistical confidence Univariate and multivariate analyses with Cox’s level. Briefly, the number of patients in the training set proportional-hazards regression were done on the started at 15 patients and was increased in steps of five.
individual clinical variables with and without the gene For a given sample size, ten training sets with randomly signature. The hazard ratio and its 95% CI were derived selected patients were made. A gene signature was from these results. Statistical analyses used S-Plus constructed from each of the training sets and tested in a designated testing set of patients by analysis of thereceiver operating characteristic (ROC) curve with Pathway analysis
distant metastasis within 5 years as the defining point.
A functional class was assigned to each prognostic The mean and the coefficient of variation of the area signature gene. Pathway analysis was done with under the curve (AUC) for a given sample size were Ingenuity software (version 1.0). Affymetrix probes calculated. A minimum number of patients required for were used as input to search for biological networks the training set was chosen at the point at which the built by the software. Biological networks identified by average AUC reached a plateau and the coefficient ofvariation of the ten AUC was less than 5%.
Characteristics
All patients
ER-positive
ER-negative
Validation set
Genes were selected as follows. First, univariate Cox’s training set (n=80) training set (n=35)
proportional-hazards regression was used to identify Age, years
genes for which expression (on a log scale) was correlated with the length of distant-metastasis-free survival. To reduce the effect of multiple testing and to test the robustness of the selected genes, the Cox’s model was constructed with bootstrapping of the Menopausal status
Premenopausal
patients in the training set.27 Briefly, 400 bootstrap samples of the training set were constructed, each with 80 patients randomly chosen with replacement. A Cox’s model was run on each of the bootstrap samples. A bootstrap score was created for each gene by removing the top and bottom 5% p values and averaging the inverses of the remaining bootstrap p values. This score was used to rank the genes. To construct a multiple gene signature, combinations of gene markers were ER status*
tested by adding one gene at a time according to the rank order. ROC analysis with distant metastasis within 5 years as the defining point was done to calculate the PR status*
Positive
AUC for each signature with increasing number of genes until a maximum AUC value was reached.
The relapse score was used to calculate each patient’s Metastases within 5 years
risk of distant metastasis (panel). The score was defined as the linear combination of weighted expression signals with the standardised Cox’sregression coefficient as the weight.
Data are number of patients unless otherwise stated. *Positive=Ͼ10 fmol per mg protein or Ͼ10% positive tumour cells.
The threshold was determined from the ROC curve Table 1: Clinical and pathological characteristics of patients and their tumours
of the training set to ensure 100% sensitivity and the www.thelancet.com Vol 365 February 19, 2005
Articles
Selection of genes
ROC curve of 76-gene signature
Distant-metastasis-free survival in validation set
Overall survival in validation set
Hazard ratio=5·67 (95% CI 2·59–12·4) Hazard ratio=8·62 (95% CI 2·57–27·9) Patients at risk
Figure 2: Establishment of the 76-gene profile and Kaplan-Meier analysis for distant-metastasis-free and overall survival
the program were assessed in the context of general 286 patients included, 93 (33%) showed evidence of functional classes by GO ontology classification.
distant metastasis within 5 years and were counted as Pathways with two or more genes in the prognostic failures in analysis of distant-metastasis-free survival.
signature were selected and investigated.
Five (2%) patients died without evidence of disease andwere censored at last follow-up. 83 (29%) died after Role of the funding sources
previous relapse. Therefore, 88 patients (31%) were This study was supported partly by the Dutch Cancer failures in the analysis of overall survival.
Society and the Netherlands Genomics Initiative/ Clinical and pathological features of 286 patients are Netherlands Organisation for Scientific Research.
summarised in table 1. There were no differences These organisations had no role in study design; the among the groups in age or menopausal status. The collection, analysis, or interpretation of data; writing of ER-negative training group had a slightly higher the paper; or in decisions relating to publication. The proportion of larger tumours and, as expected, more Erasmus Medical Centre was financially supported by poor-grade tumours than the ER-positive training Veridex LLC, a Johnson & Johnson Company, for tissue group. The validation group of 171 patients (129 ER- processing and isolating RNA for Affymetrix chip positive, 42 ER-negative) did not differ from the total analysis. The corresponding author had full access to group of 286 patients in any of the characteristics of all the data in the study and took final responsibility for the decision to submit the paper for publication.
Two approaches were used to identify markers predictive of disease relapse. First, we randomly divided all the 286 patients (ER-positive and ER- The median follow-up for the 198 patients who negative combined) into a training set and a testing set.
survived was 101 months (range 20–171). Of the 35 genes were selected from 80 patients in the training www.thelancet.com Vol 365 February 19, 2005
Articles
Distant-metastasis-free survival
Overall survival
Premenopausal
Premenopausal
Hazard ratio=9·60 (2·28–40·5) Log-rank p=0·00015 Hazard ratio=14·8 (1·99–100) Log-rank p=0·00045 Patients at risk
Patients at risk
Postmenopausal
Postmenopausal
Hazard ratio=4·04 (1·57–10·4) Log-rank p=0·0017 Hazard ratio=3·49 (1·21–10·1) Log-rank p=0·0134 Patients at risk
Patients at risk
Tumours 10–20 mm
Tumours 10–20 mm
Hazard ratio=14·1 (3·34–59·2) Log-rank p<0·0001 Hazard ratio=20·2 (2·72–149) Log-rank p<0·0001 Patients at risk
Patients at risk
Figure 3: Analysis of distant-metastasis-free and overall survival in subgroups of patients with lymph-node-negative breast cancer
set and a Cox’s model to predict the occurrence of model to predict recurrence of cancer was built for distant metastasis was built. Moderate prognostic value all lymph-node-negative patients. Validation of the was observed (data not shown). Unsupervised 76-gene predictor in the testing set of 171 patients clustering analysis showed two distinct subgroups produced an ROC with an AUC of 0·694, sensitivity of highly correlated with the tumour ER status (␹2 test, 93% (52/56), and specificity of 48% (55/115; figure 2).
pϽ0·0001; figure 1), which supported our second Patients with a relapse score above the threshold of the approach in which patients were first grouped on the prognostic signature have an odds ratio of 11·9 basis of ER status. Each subgroup was analysed for (95% CI 4·04–35·1; pϽ0·0001) to develop distant selection of markers. 76 genes were selected from metastasis within 5 years. As the control, randomly patients in the training sets (60 for the ER-positive selected 76-gene sets were generated. These produced group, 16 for the ER-negative group; figure 2). With the ROC with an average AUC value of 0·515, sensitivity of selected genes and ER status taken together, a Cox’s 91%, and specificity of 12% in the testing group.
www.thelancet.com Vol 365 February 19, 2005
Articles
activities were identified that were well represented, Univariate analysis
Multivariate analysis*
such as cell death, cell cycle and proliferation, DNA replication and repair, and immune response (table 4).
Genes implicated in disease progression were found, including calpain2, origin recognition protein, dual- inhibitor, tumour necrosis factor (TNF) superfamily protein, complement component 3, microtubule- associated protein, protein phosphatase 1, and apoptosis regulator BCL-G. Furthermore, previously characterised prognostic genes such as cyclin E228 andCD4429 were in the gene signature.
*The multivariate model included 162 patients, owing to missing values in nine. †Grade: moderate/good vs poor; unknowngrade was included as a separate group.
The dataset has been submitted to the NCBI/ Genbank GEO database (series entry GSE2034).
Table 2: Univariate and multivariate analyses for distant-metastasis-free survival in the testing set of
171 patients
Discussion
We provide results of an analysis of primary tumours
Patients stratified by such a gene set would have an from 286 patients with lymph-node-negative breast odds ratio of 1·3 (0·50–3·90; p=0·8) for development cancer of all age-groups and tumour sizes. The patients of metastases, indicating a random classification. In had not received adjuvant systemic therapy, so the addition, the Kaplan-Meier analyses for distant- multigene assessment of prognosis was not subject to metastasis-free and overall survival as a function of the potentially confounding contributions by predictive factors related to systemic treatment.
differences in time to metastasis between the groups The study revealed a 76-gene signature that accurately predicted to have good and poor prognosis (figure 2). At predicts distant tumour recurrence. This signature could 60 months and 80 months, the respective absolute be applied to all lymph-node-negative patients differences in distant-metastasis-free survival between independently of age, tumour size and grade, and ER the groups with predicited good and poor prognosis status. In Cox’s multivariate analysis for distant- were 40% (93% vs 53%) and 39% (88% vs 49%), and metastasis-free survival, the 76-gene signature was the those in overall survival were 27% (97% vs 70%) and only significant variable, superseding clinical variables, 32% (95% vs 63%) respectively.
including grade. After 5 years, absolute differences in The 76-gene profile also represented a strong distant-metastasis-free and overall survival between the prognostic factor for the development of distant patients with the good and poor 76-gene signatures were metastasis in the subgroups of 84 premenopausal 40% and 27%, respectively. Of the patients with good- patients (hazard ratio 9·60), 87 postmenopausal prognosis signatures, 7% developed distant metastases patients (4·04), and 79 patients with tumour sizes of and 3% died within 5 years. If further validated, this signature will yield a positive predictive value of 37% and Univariate and multivariate Cox’s regression analyses a negative predictive value of 95%, on the assumption of are summarised in table 2. Other than the 76-gene a 25% rate of disease recurrence in lymph-node-negative signature, only grade was significant in univariate patients. In particular, this signature could be valuable analyses and moderate/good differentiation was for defining the risk of recurrence for the increasing associated with favourable distant-metastasis-free proportion of T1 tumours (Ͻ2 cm). Comparison with survival. Multivariate regression estimation of hazard the St Gallen and National Institutes of Health ratio for the occurrence of tumour metastasis within guidelines was instructive. Although ensuring that the 5 years was 5·55 (pϽ0·0001), indicating that the same number of high-risk patients would receive the76-gene set represents an independent prognostic necessary treatment, our 76-gene signature would signature strongly associated with a higher risk of recommend systemic adjuvant chemotherapy to only tumour metastasis. Univariate and multivariate 52% of low-risk patients, compared with 90% and 89% analyses were also done separately for ER-positive by the St Gallen and National Institutes of Healthand ER-negative patients; the 76-gene signature guidelines (table 5). Our gene signature, if furtherwas also an independent prognostic variable in confirmed, could result in a reduction of the number ofthe subgroups stratified by ER status (data not low-risk lymph-node-negative patients who would beshown).
recommended to have unnecessary adjuvant systemic The function of the 76 genes (table 3) in the prognostic signature was analysed to relate the genes to The 76 genes in our prognostic signature belong to biological pathways. Although 18 of the 76 genes have many functional classes, which suggests that different unknown function, several pathways or biochemical paths could lead to disease progression. The signature www.thelancet.com Vol 365 February 19, 2005
Articles
Standard Cox coefficient Cox p value Gene description
For ER-positive group
219340_s_at
gb:AF123759.1 /DEF=Homo sapiens putative transmembrane protein (CLN8) mRNA, complete cds gb:NM_016548.1 /DEF=Homo sapiens golgi membrane protein GP73 (LOC51280) gb:NM_020470.1 /DEF=Homo sapiens putative transmembrane protein; homologue of yeast Golgi membrane protein Yif1p gb:NM_001562.1 /DEF=Homo sapiens interleukin 18 (interferon-␥-inducing factor) (IL18) Consensus includes gb:BE748755 /heterochromatin-like protein 1 gb:BC002671.1 /DEF=Homo sapiens, dual specificity phosphatase 4 gb:NM_002710.1 /DEF=Homo sapiens protein phosphatase 1, catalytic subunit, ␥ isoform (PPP1CC) gb:NM_006720.1 /DEF=Homo sapiens actin binding LIM protein 1 (ABLIM), transcript variant ABLIM-s gb:AF114013.1 /DEF=Homo sapiens TNF-related death ligand-1␥ Consensus includes gb:AI636233 five-span transmembrane protein M83 gb:NM_000064.1 /DEF=Homo sapiens complement component 3 (C3) gb:NM_017760.1 /DEF=Homo sapiens hypothetical protein FLJ20311 gb:NM_013279.1 /DEF=Homo sapiens chromosome 11open reading frame 9 (C11ORF9) Consensus includes gb:AL523310 putative translation initiation factor gb:AF220152.2 /DEF=Homo sapiens TACC2 mRNA gb:NM_005496.1 /DEF=Homo sapiens chromosome-associated polypeptide C (CAP-C) gb:NM_013936.1 /DEF=Homo sapiens olfactory receptor, family 12, subfamily D, member 2 (OR12D2) gb:AF125507.1 /DEF=Homo sapiens origin recognition complex subunit 3 (ORC3) gb:NM_014109.1 /DEF=Homo sapiens PRO2000 protein (PRO2000) gb:AL136877.1 /SMC4 (structural maintenance of chromosomes 4, yeast)-like 1 /FL=gb:AB019987.1 gb:NM_005496.1 gb:AL136877.1 gb:NM_014796.1 /DEF=Homo sapiens KIAA0748 gene product (KIAA0748) gb:NM_001394.2 /DEF=Homo sapiens dual specificity phosphatase 4 (DUSP4) Consensus includes gb:AI493245 /CD44 antigen (homing function and Indian blood group system) gb:NM_005030.1 /DEF=Homo sapiens polo (Drosophila)-like kinase (PLK) gb:NM_006314.1 /DEF=Homo sapiens connector enhancer of KSR-like (Drosophila kinase suppressor of ras) (CNK1) gb:NM_003543.2 /DEF=Homo sapiens H4 histone family, member H (H4FH) gb:NM_004111.3 /DEF=Homo sapiens flap structure-specific endonuclease 1 (FEN1) gb:NM_004470.1 /DEF=Homo sapiens FK506-binding protein 2 (13kD) (FKBP2) gb:BC005978.1 /DEF=Homo sapiens, karyopherin ␣ 2 (RAG cohort 1, importin ␣ 1) gb:NM_015997.1 /DEF=Homo sapiens CGI-41 protein (LOC51093) gb:NM_030819.1 /DEF=Homo sapiens hypothetical protein MGC11335 (MGC11335) gb:BC006155.1 /DEF=Homo sapiens, clone MGC:13188 gb:NM_024629.1 /DEF=Homo sapiens hypothetical protein FLJ23468 (FLJ23468) Consensus includes gb:AA772093 /neuralised (Drosophila)-like /FL=gb:U87864.1 gb:AF029729.1 gb:NM_004210.1 gb:NM_007192.1 /DEF=Homo sapiens chromatin-specific transcription elongation factor, 140 kDa subunit (FACTP140) Consensus includes gb:U07802 /DEF=Human Tis11d gene gb:NM_001175.1 /DEF=Homo sapiens Rho GDP dissociation inhibitor (GDI) ␤ (ARHGDIB) gb:NM_002803.1 /DEF=Homo sapiens proteasome (prosome, macropain) 26S subunit, ATPase, 2 (PSMC2) gb:NM_017612.1 /DEF=Homo sapiens hypothetical protein DKFZp434E2220 (DKFZp434E2220) Consensus includes gb:R39094 /KIAA1085 protein gb:BC004372.1 /DEF=Homo sapiens, Similar to CD44 antigen (homing function and Indian blood group system) Consensus includes gb:AL117652.1 /DEF=Homo sapiens mRNA gb:NM_006416.1 /DEF=Homo sapiens solute carrier family 35 (CMP-sialic acid transporter), member 1 (SLC35A1) gb:NM_004702.1 /DEF=Homo sapiens cyclin E2 (CCNE2) Consensus includes gb:BF055474 / putative zinc finger protein NY-REN-34 antigen gb:NM_006596.1 /DEF=Homo sapiens polymerase (DNA directed), ␪ (POLQ) Consensus includes gb:AF041410.1 /DEF=Homo sapiens malignancy-associated protein gb:M23254.1 /DEF=Human Ca2-activated neutral protease large subunit (CANP) Consensus includes gb:AV693985 /ets variant gene 2 gb:NM_017859.1 /DEF=Homo sapiens hypothetical protein FLJ20517 (FLJ20517) Consensus includes gb:AV713720 /Homo sapiens mRNA for LST-1N protein Consensus includes gb:AI057637 /Hs.234898 ESTs, Weakly similar to 2109260A B cell growth factor Homo sapiens Consensus includes gb:U90030.1 /DEF=Homo sapiens bicaudal-D (BICD) mRNA, alternatively spliced, partial cds gb:NM_001958.1 /DEF=Homo sapiens eukaryotic translation elongation factor 1 ␣ 2 (EEF1A2) Consensus includes gb:BF055311 / hypothetical protein Consensus includes gb:AL133102.1 /DEF=Homo sapiens mRNA; cDNA DKFZp434C1722 gb:AF114012.1 /DEF=Homo sapiens TNF-related death ligand-1␤ mRNA Homo sapiens cDNA FLJ10418 fis, clone NT2RP1000130, moderately similar to hepatoma-derived growth factor gb:NM_004659.1 /DEF=Homo sapiens matrix metalloproteinase 23A (MMP23A) gb:BC006325.1 /DEF=Homo sapiens, G-2 and S-phase expressed 1 For ER-negative group
218430_s_at
gb:NM_022841.1 /DEF=Homo sapiens hypothetical protein FLJ12994 (FLJ12994) Consensus includes gb:X16468.1 /DEF=Human mRNA for ␣-1 type II collagen.
gb:NM_005256.1 /DEF=Homo sapiens growth arrest-specific 2 (GAS2) Homo sapiens cDNA FLJ11780 fis, clone HEMBA1005931, weakly similar to zinc finger protein 83 Consensus includes gb:D89324 /DEF=Homo sapiens DNA for alpha (1,31,4) fucosyltransferase gb:NM_017534.1 /DEF=Homo sapiens myosin, heavy polypeptide 2, skeletal muscle, adult (MYH2) gb:U57059.1 /DEF=Homo sapiens Apo-2 ligand mRNA Continued
www.thelancet.com Vol 365 February 19, 2005
Articles
Continued
Gene
Standard Cox coefficient Cox p value Gene description
gb:BC000596.1 /DEF=Homo sapiens, Similar to ribosomal protein L23a, clone MGC:2597 gb:NM_018558.1 /DEF=Homo sapiens GABA receptor, ␪ (GABRQ) gb:NM_006437.2 /DEF=Homo sapiens ADP-ribosyltransferase (NAD+; poly (ADP-ribose) polymerase)-like 1 (ADPRTL1) gb:NM_014042.1 /DEF=Homo sapiens DKFZP564M082 protein (DKFZP564M082) gb:NM_030766.1 /DEF=Homo sapiens apoptosis regulator BCL-G (BCLG) gb:BC001233.1 /DEF=Homo sapiens, Similar to KIAA0092 gene product, clone MGC:4896 Contains a novel gene and the 5 part of a gene for a novel protein similar to X-linked ribosomal protein 4 (RPS4X) gb:M55580.1 /DEF=Human spermidinespermine N1-acetyltransferase Consensus includes gb:AB014607.1 /DEF=Homo sapiens mRNA for KIAA0707 protein Table 3: 76 genes from the prognostic signature
included well-characterised genes and 18 unknown Patients guided to receive adjuvant
genes. This finding could explain the superior chemotherapy in the testing set
performance of this signature compared with other prognostic factors. Although genes involved in cell death, cell proliferation, and transcriptional regulation were found in both groups of patients stratified by ER National Institutes of Health 52/55 (95%) status, the 60 genes selected for the ER-positive group and the 16 selected for the ER-negative group had no St Gallen consensus criteria: tumour ୑2 cm, ER negative, grade 2–3, patient Ͻ35 years overlap. This result supports the idea that the extent of (any one of these criteria). National Institutes of Health: tumour Ͼ1 cm.
heterogeneity and the underlying mechanisms for Table 5: Comparison of the 76-gene signature and the current
disease progression could differ for the two ER-based conventional consensus on treatment of breast cancer
subgroups of breast-cancer patients.
Comparison of our results with those of Van de Vijver and colleagues12 is difficult because of differences in overlap between the two signatures (cyclin E2, origin patients, techniques, and materials used. Their study recognition complex, and TNF superfamily protein).
included node-negative and node-positive patients, who Despite the apparent difference, both signatures included had or had not received adjuvant systemic therapy, and genes that identified several common pathways that might only women younger than 53 years. Furthermore, the be involved in tumour recurrence. This finding supports microarray platforms used in the studies differ— the idea that although there might be redundancy in gene Affymetrix and Agilent. Of the 70 genes in the study by members, effective signatures could be required to van‘t Veer and co-workers,11 48 are present on the include representation of specific pathways.
Affymetrix U133a array, whereas only 38 of our 76 genes The strengths of our study compared with the study of are present on the Agilent array. There is a three-gene Van de Vijver and colleagues12 are the larger number ofuntreated lymph-node-negative patients (286 vs 141), and Functional class
76-gene signature
the independence of our 76-gene signature with respect TNFSF10, TNFSF13, MAP4, CD44, IL18, GAS2, NEFL, EEF1A2, BCLG, C3 to age, menopausal status, and tumour size. The CCNE2, CD44, MAP4, SMC4L1, TNFSF10, AP2A2, FEN1, KPNA2, ORC3L, PLK1 validation set of patients is completely without overlap CD44, IL18, TNFSF10, TNFSF13, PPP1CC, CAPN2, PLK1, SAT with the training set, in contrast to 90% of other reports.30 DNA replication, recombination, and repair TNFSF10, SMC4L1, FEN1, ORC3L, KPNA2, SUPT16H, POLQ, ADPRTL1 In conclusion, since only 30–40% of untreated lymph- TNFSF10, CD44, IL18, TNFSF13, ARHGDIB, C3 PPP1CC, CD44, IL18, TNFSF10, SAT, HDGFRP3 node-negative patients develop tumour recurrence, our prognostic signature could provide a powerful tool to KPNA2, DUSP4, SUPT16H, DKFZP434E2220, PHF11, ETV2 identify those patients at low risk preventing overtreatment in substantial numbers of patients. If confirmed in subsequent studies, the recommendation of adjuvant systemic therapy in patients with lymph- node-negative primary breast cancer could be guided by this prognostic signature. The predictive value of our gene signature with respect to the efficacy of different modes of systemic therapy could be tested in the adjuvant setting or in patients with metastatic disease.
Contributors
Y Wang, J G M Klijn, E M J J Berns, D Atkins, and J A Foekensdesigned the study, interpreted the data, and wrote the report. Table 4: Pathway analysis of the 76 genes from the prognostic signature
Y Zhang, J Yu, and T Jatkoe analysed the data and developed theprognostic signature. M Timmermans and D Talantov were www.thelancet.com Vol 365 February 19, 2005
Articles
responsible for laboratory experiments and pathological assessment of Van de Vijver MJ, Yudong HE, Van‘t Veer L, et al. A gene the tissue samples. F Yang and A M Sieuwerts did laboratory expression signature as a predictor of survival in breast cancer. experiments on the isolation of RNA and quality assessment. N Engl J Med 2002; 347: 1999–2009.
M P Look and M E Meijer-van Gelder collected and handled the Ahr A, Kam T, Solbach C, et al. Identification of high risk breast- patients’ data and contributed to the survival analyses.
cancer patients by gene-expression profiling. Lancet 2002; 359:
131–32.
Conflict of interest statement
Huang E, Cheng SH, Dressman H, et al. Gene expression YW, YZ, FY, DT, JY, TJ, and DA are employed by Veridex LLC, a predictors of breast cancer outcomes. Lancet 2003; 361: 1590–96.
Johnson & Johnson Company, which is in the business of Sotiriou C, Neo S-Y, McShane LM, et al. Breast cancer commercialising diagnostic products. The other authors declare no classification and prognosis based on gene expression profiles from a population-based study. Proc Natl Acad Sci USA 2003; 100:
10393–98.
Acknowledgments
We thank Anneke Goedheer, Anita Trapman-Jansen, Miranda Arnold,
Woelfle U, Cloos J, Sauter G, et al. Molecular signature associatedwith bone marrow micrometastasis in human breast cancer.
and Roberto Rodriguez-Garcia for technical assistance, and the Cancer Res 2003; 63: 5679–84.
surgeons, pathologists, and internists of the St Clara Hospital, Ikazia Ma X-J, Salunga R, Tuggle JT, et al. Gene expression profiles of Hospital, St Franciscus Gasthuis at Rotterdam, and Ruwaard van human breast cancer progression. Proc Natl Acad Sci USA 2003; Putten Hospital at Spijkenisse for the supply of tumour tissues, for 100: 5974–79.
their assistance in the collection of the clinical follow-up data, or both.
Ramaswamy S, Ross KN, Lander ES, Golub TR. A molecular References
signature of metastasis in primary solid tumors. Nat Genet 2003; Early Breast Cancer Trialists’ Collaborative Group.
33: 1–6.
Polychemotherapy for early breast cancer: an overview of the Chang JC, Wooten EC, Tsimelzon A, et al. Gene expression randomised trials. Lancet 1998; 352: 930–42.
profiling for the prediction of therapeutic response to docetaxel in Early Breast Cancer Trialists’ Collaborative Group. Tamoxifen for patients with breast cancer. Lancet 2003; 362: 362–69.
early breast cancer: an overview of randomised trials. Lancet 1998; Sotiriou C, Powles TJ, Dowsett M, et al. Gene expression profiles 351: 1451–67.
derived from fine needle aspiration correlate with response to Goldhirsch A, Wood C, Gelber RD, Coates AS, Thürlimann B, systemic chemotherapy in breast cancer. Breast Cancer Res 2003; Senn HJ. Meeting highlights: updated international expert consensus on the primary therapy of early breast cancer. Hedenfalk I, Duggan D, Chen Y, et al. Gene-expression profiles in J Clin Oncol 2003; 21: 3357–65.
hereditary breast cancer. N Engl J Med 2001; 344: 539–48.
Eifel P, Axelson JA, Costa J, et al. National Institutes of Health Early Breast Cancer Trialists’ Collaborative Group. Effects of Consensus Development Conference Statement: adjuvant therapy radiotherapy and surgery in early breast cancer: an overview of the for breast cancer, Nov 1–3, 2000. J Natl Cancer Inst 2001; 93:
randomized trials. N Engl J Med 1995; 333: 1444–55.
Foekens JA, Portengen H, van Putten WLJ, et al. Prognostic value Ntzani E, Ionnidis JPA. Predictive ability of DNA microarrays for of receptors for insulin-like growth factor 1, somatostatin, and cancer outcomes and correlates: an empirical assessment. Lancet epidermal growth factor in human breast cancer. Cancer Res 1989; 2003; 362: 1439–44.
49: 7002–09.
Wang Y, Jatkoe T, Zhang Y, et al. Gene expression profiles and Foekens JA, Portengen H, van Putten WLJ, et al. Prognostic value molecular markers to predict recurrence of Dukes’ B colon of estrogen and progesterone receptors measured by enzyme cancer. J Clin Oncol 2004; 22: 1564–71.
immunoassays in human breast tumor cytosols. Cancer Res 1989; Perou CM, Sørlie T, Eisen MB, et al. Molecular portraits of 49: 5823–28.
human breast tumours. Nature 2000; 406: 747–52.
Lipshutz RJ, Fodor SP, Gingeras TR, Lockhart DJ. High density Sørlie T, Perou CM, Tibshirani R, et al. Gene expression patterns synthetic oligonucleotide arrays. Nat Genet 1999; 21: 20–24.
of breast carcinomas distinguish tumor subclasses with clinical Kaplan EL, Maier P. Non-parametric estimation of incomplete implications. Proc Natl Acad Sci USA 2001; 98: 10869–74.
observations. J Am Stat Assoc 1958; 53: 457–81.
Sørlie T, Tibshirani R, Parker J, et al. Repeated observation of Efron B. Censored data and the bootstrap. J Am Stat Assoc 1981; breast tumor subtypes in independent gene expression data sets.
76: 312–19.
Proc Natl Acad Sci USA 2003; 100: 8418–23.
Keyomarsi K, Tucker SL, Buchholz TA, et al. Cyclin E and survival Gruvberger S, Ringnér M, Chen Y, et al. Estrogen receptor status in patients with breast cancer. N Engl J Med 2002; 347: 1566–75.
in breast cancer is associated with remarkably distinct gene Herrera-Gayol A, Jothy S. Adhesion proteins in the biology of
expression patterns. Cancer Res 2001; 61: 5979–84.
breast cancer: contribution of CD44. Exp Mol Pathol 1999; 66:
Van ‘t Veer L, Dai H, Van de Vijver MJ, et al. Gene expression profiling predicts clinical outcome of breast cancer. Nature 2002; Ransohoff DF. Rules of evidence for cancer molecular-marker 415: 530–36.
discovery and validation. Nat Rev Cancer 2004; 4: 309–14.
www.thelancet.com Vol 365 February 19, 2005

Source: http://www.bioinf.at/teaching/ss2010/irab/Wang/Wang2005.pdf